diffusion, obsolescence and disposal of endof--life ...file/… · diffusion, obsolescence and...

Diffusion, Obsolescence and Disposal of End-of-Life Consumer Durables: Models for Forecasting Waste Flows

DISSERTATION of the University of St. Gallen,

School of Management, Economics, Law, Social Sciences

and International Affairs to obtain the title of Doctor Oeconomiae

submitted by

Deepali Sinha from

India

Approved on the application of

Prof. Dr. Markus Schwaninger

and

Prof. Dr. Lorenz Hilty

Dissertation no. 4113

Gutenberg AG, Schaan, 2013

The University of St.Gallen, School of Management, Economics, Law, Social Sciences and International Affairs hereby consents to the printing of the present dissertation, without hereby expressing any opinion on the views herein expressed. St. Gallen, October 23, 2012

The President: Prof. Dr. Thomas Bieger

This dissertation is dedicated to my mum Subhra Sinha

For her immense belief, encouragement and support in embarking on and completing my doctoral studies.

Acknowledgement This thesis was made possible only because of the support, encouragement and good wishes of many wonderful people who have been with me on this interesting journey. Firstly, I am very grateful to Prof. Markus Schwaninger for accepting me as a doctoral student and guiding me with his prompt and powerful insights. I also owe many thanks to Prof. Lorenz Hilty, who, as my co-supervisor, contributed greatly with his incisive comments which helped to not only bring focus to the thesis but to also make me a better researcher. I owe a special debt of gratitude to Rolf Widmer for hand-holding me through the journey – for his constant support, advice and encouragement and especially for taking time out from his busy schedule to chat, comment, correct and coach on the thesis as it evolved – his support has been instrumental and without it the thesis would not have been realised. I have my brother, Tanmoy Sinha to thank for his software wizardry which made the modelling much simpler. Furthermore, I would like to the many, many friends and colleagues from around the world for their unflagging support – for cheering me up when I was low, for calming me when I was in a panic and for being considerate when I was obnoxious! I highly appreciate the support provided by EMPA and the TSL team towards my doctoral studies, without which this research would never have been accomplished. Most of all, I’d like to thank my parents, who whole-heartedly supported my doctoral adventure and were always there for me, encouraging me and providing me with invaluable moral support. And last, but not the least, my husband Rupesh, who was patient when I was preoccupied, encouraging when I was unconfident and relieved when it was over! Deepali Sinha London, 2012

Table of Contents Table of Contents ............................................................................................................ 5

List of Figures ................................................................................................................. 9

List of Tables ................................................................................................................ 10

List of Abbreviations .................................................................................................... 11

CHAPTER I

The Diffusion, Obsolescence and Disposal of Consumer Durables .............................. 1

1. Introduction ................................................................................................... 1

1.1. Diffusion of Consumer Durables ............................................................................. 1 1.2. Obsolescence of Consumer Durables ....................................................................... 2 1.3. Disposal of Consumer Durables ............................................................................... 4

2. Existing Research .......................................................................................... 5

2.1. Research in Waste Management .............................................................................. 5 2.2. Research on Innovation and Diffusion of Consumer Durables ................................ 6 2.3. Research on Obsolescence, Replacement and Disposal of Consumer Durables ...... 7

3. Research Gap ................................................................................................ 8

4. Research Goal ............................................................................................... 9

5. Thesis Architecture ..................................................................................... 10

5.1. Chapter I ................................................................................................................. 10 5.2. Chapter II ............................................................................................................... 10 5.3. Chapter III .............................................................................................................. 11 5.4. Chapter IV .............................................................................................................. 13

6. Discussion and Conclusions ........................................................................ 13

6.1. Key Findings .......................................................................................................... 13 6.1.1. Applicability of Diffusion Modelling in Waste Management Research ............ 13 6.1.2. Improving Forecasting Models for Disposal of Consumer Durables ................. 14 6.1.3. Importance of Consumer Behaviour: ................................................................. 15

6.2. Practical Implications ............................................................................................. 16 6.3. Conclusion ............................................................................................................. 17

References .................................................................................................................... 19

CHAPTER II

From Introduction to Obsolescence: Estimating Societal Stocks and Flows of Consumer Durables ...................................................................................................... 24

1. Introduction ................................................................................................. 25

2. Literature Review ........................................................................................ 27

2.1. Delay Models or Market Supply Models ............................................................... 27 2.2. Material Flow Analysis (MFA) Models ................................................................. 27 2.3. Diffusion Models ................................................................................................... 28

3. Goal and Purpose ........................................................................................ 29

4. Conceptual Model & Mathematical Framework ........................................ 30

5. Model Parameters and Variables ................................................................ 30

5.1. Stocks ..................................................................................................................... 30 5.2. Inflows ................................................................................................................... 32 5.3. Outflows ................................................................................................................. 32 5.4. Disposal Distribution Function .............................................................................. 34 5.5. Technology Substitution ........................................................................................ 35

6. Experimental Frame .................................................................................... 36

6.1. Application to the Case of TVs in Switzerland ...................................................... 36 6.2. TVs in Switzerland –Background .......................................................................... 37

7. Data ............................................................................................................. 37

8. Extensions to the Specific Model ................................................................ 38

8.1. Estimating the Development of Household Population ......................................... 39 8.2. Estimating Devices per Household - Multi-unit Ownership Sub-model ............... 39

8.2.1. Number of Devices per Swiss Household .......................................................... 41

9. Results ......................................................................................................... 42

9.1. Swiss TVs – Installed Base .................................................................................... 42 9.2. Swiss TVs – Sales and Disposals ........................................................................... 42

10. Model Validation ........................................................................................ 44

11. Discussion and Conclusion ......................................................................... 45

References .................................................................................................................... 48

CHAPTER III

Reverse Diffusion: Estimating Disposal of Consumer Durables through Application of Diffusion Modelling ..................................................................................................... 52

1. Introduction ................................................................................................. 53

2. Literature Review ........................................................................................ 54

2.1. Forecasting Adoption of Consumer Durables ........................................................ 54 2.2. Consumer Disposition Behaviour .......................................................................... 56 2.3. Waste Forecasting Models ..................................................................................... 57

3. Goal and Purpose ........................................................................................ 58

4. Conceptual Model and Mathematical Framework ...................................... 58

4.1. Bass Diffusion Model ............................................................................................ 58

4.2. Reverse Diffusion .................................................................................................. 60 4.3. Model Parameters .................................................................................................. 62 4.4. Assumptions ........................................................................................................... 62

5. Application: Case Study – Consumer Durables in Switzerland ................. 63

5.1. Data ........................................................................................................................ 64 5.2. Case Study 1: CRT Monitors in Switzerland ......................................................... 65 5.3. Case Study 2: CRT TVs in Switzerland ................................................................. 67 5.4. Case Study 3: LCD Monitors in Switzerland ......................................................... 68

6. Model Validation ........................................................................................ 70

6.1. CRT Monitors ........................................................................................................ 70 6.2. CRT TVs ................................................................................................................ 71 6.3. LCD Monitors ........................................................................................................ 71

7. Discussion & Conclusion ............................................................................ 72

References .................................................................................................................... 75

CHAPTER IV

Forecasting Consumer Durable Disposals: A Review and Comparison of Modelling Approaches ................................................................................................................... 80

1. Introduction ................................................................................................. 81

2. Modelling Approaches to Estimate Consumer Durable Disposals ............. 83

2.1. Terminology ........................................................................................................... 84 2.1.1. Inflows ............................................................................................................... 84 2.1.2. Stocks ................................................................................................................. 84 2.1.3. Delay Distribution .............................................................................................. 85 2.1.4. Product Mass ...................................................................................................... 85

3. Goal and Purpose ........................................................................................ 86

4. Delay Model Approach ............................................................................... 86

4.1. Delay Model A: Example Reference – Oguchi et al., 2008 ................................... 87 4.2. Delay Model B: Example Reference – Gregory et al., 2009 .................................. 88 4.3. Delay Model C: Example Reference – Chapter II ................................................. 89

5. Reverse Diffusion Model Approach ........................................................... 90

6. Structural Comparison ................................................................................ 91

7. Experimental Frame .................................................................................... 93

8. Results ......................................................................................................... 93

8.1. Output Comparison ................................................................................................ 94 8.2. Predictive Validity ................................................................................................. 95 8.3. Fit Improvement ..................................................................................................... 96 8.4. Sensitivity Analysis ................................................................................................ 98

8.4.1. Sensitivity of Delay Model Parameters .............................................................. 98 8.4.2. Sensitivity of Reverse Diffusion Parameters ..................................................... 98

9. Discussion and Conclusion ......................................................................... 99

References .................................................................................................................. 103

Annexes ...................................................................................................................... 107

Annexe 1: CRT Monitor Sales in Switzerland ............................................................... 107 Annexe 2: CRT and FPD TV Sales in Switzerland ........................................................ 108 Annexe 3: CRT Glass Collection by SWICO Recycling ............................................... 109 Annexe 4: LCD Monitor Collection by SWICO Recycling ........................................... 109 Annexe 5: TV Permits in Switzerland ............................................................................ 110 Annexe 6: Household TV Ownership in Switzerland .................................................... 110 Annexe 7: Swiss population of households .................................................................... 110

List of Figures Figure 1: Annual sales of electronic consumer durables .......................................... 2 Figure 2: Technological evolution of the television … ............................................ 3 Figure 3: Annual Sales of Televisions in Switzerland .............................................. 4 Figure 4: Thesis architecture ................................................................................... 12 Figure 5: Generic stock-flow model ....................................................................... 31 Figure 6: Stock and flow diagram for the delay model .......................................... 33 Figure 7: Technology Substitution - CRT TV to non-CRT TV in Switzerland ..... 37 Figure 8: Multi-unit Devices per Household (DPH)............................................... 41 Figure 9: Societal Stock – TVs (CRT and Non-CRT TVs) in Switzerland ............ 42 Figure 10: Societal Flows - Sales and Disposals of CRT TVs in Switzerland ....... 43 Figure 11: Disposal and Collection of CRT TVs .................................................... 44 Figure 12: Stock and flow diagram for the reverse diffusion model ...................... 61 Figure 13: Cumulative Disposal R(t)- CRT Monitor Glass .................................... 66 Figure 14: Disposal curve r(t) – CRT Monitor Glass ............................................. 66 Figure 15: Cumulative Disposal R(t) - CRT TVs ................................................... 67 Figure 16: Disposal Curve r(t) - CRT TVs ............................................................. 68 Figure 17: Disposal Curve - LCD Monitors ........................................................... 69 Figure 18: Delay Model Structure .......................................................................... 91 Figure 19: Reverse Diffusion Model Structure ....................................................... 91 Figure 20: Model Comparison - CRT Glass Disposal Estimates vs Observed ...... 94 Figure 21: Disposal Forecasts ................................................................................. 96 Figure 22: Fit improvement by introducing product mass function ....................... 97 Figure 23: Sensitivity of delay model parameters .................................................. 98 Figure 24: Sensitivity of reverse diffusion model parameters ................................ 99

List of Tables

Table 1: Primary data gathered and sources ........................................................... 38 Table 2: Parameters of multi-unit adoption - devices per household ..................... 42 Table 3: Parameters of the disposal function for CRT TVs ................................... 43 Table 4: Data collected ........................................................................................... 64 Table 5: Parameter values for CRT Monitor reverse disposal ................................ 65 Table 6: Parameter values for CRT TVs reverse disposal ...................................... 68 Table 7: Parameter values and fit statistics ............................................................. 70 Table 8: Model forecast vs actual disposal – CRT PC Monitors ............................ 71 Table 9: Model forecast vs actual disposal – CRT Glass from CRT TVs .............. 71 Table 10: Model forecast vs actual disposal - LCD Monitors ................................ 71 Table 11: Summary table of model characteristics ................................................. 92 Table 12: Parameter values and fit statistics ........................................................... 95 Table 13: Comparison of Predictive Power ............................................................ 96 Table 14: Fit improvement statistics ....................................................................... 97 Table 15: CRT and FPD Monitor Sales in Switzerland. ....................................... 107 Table 16: CRT and FPD TV Sales. ....................................................................... 108 Table 17: CRT Glass Collection by SWICO. ....................................................... 109 Table 18: LCD Monitor Collection by SWICO. ................................................... 109 Table 19: TV licences issued by Billag. ............................................................... 110 Table 20: Household Ownership of TVs. ............................................................. 110 Table 21: Number of households. ......................................................................... 110

List of Abbreviations 2D – Two Dimensional 3D – Three Dimensional BDM – Bass Diffusion Model BfS – Swiss Federal Statistical Office (BfS) CRT – Cathode Ray Tube DPH – Devices per Household EMPA – Swiss Federal Laboratories for Material Science and Technology EOLE – End-of-Life Equipment FPD – Flat Panel Displays ICT – Information and Communications Technology ITU – International Telecommunications Union LCD – Liquid Crystal Display MAE – Mean Absolute Error MFA – Material Flow Analysis OLED – Organic Light Emitting Diode PC – Personal Computer SCEA – Swiss Consumer Electronics Association SFA – Substance Flow Analysis TSL – Technology and Society Lab TV – Television UK – United Kingdom UNEP – United Nations Environment Programme WEEE – Waste Electrical and Electronic Equipment

Summary Consumer durables pervade modern society. Technological innovation has brought a plethora of products to market that are ever more accessible and affordable. The past decades have seen the diffusion of consumer durables on a phenomenal scale globally. The corollary to this widespread diffusion of consumer durables is that there are ever greater numbers of products reaching the waste stream. The growing magnitude of this waste stream raises both theoretical and practical questions, including how many products will be disposed of and when. My dissertation tackles the question of estimation and forecasting of end-of-life consumer durables. Across three papers, I develop, apply and compare models to estimate and forecast such disposals. In my first paper, I develop a societal stock and flow model based on the delay modelling approach which incorporates elements from diffusion models commonly used in marketing. I propose a sub-model that extends the diffusion model to estimate multiple-unit adoptions of consumer durables. The stock and flow model also incorporates technological substitution, allowing the estimation and forecasting of disposals in the light of scant or patchy data. In my second paper, I propose a reverse diffusion model which extends the application of diffusion models to the waste forecasting domain. Arguing that the dynamics of disposal are not dissimilar to that of adoption of consumer durables, the model is empirically validated through three case studies. Finally, in the third paper, I critique the modelling approaches discussed in papers one and two, namely the delay model and the diffusion model. Applying the same data set to three variants of the delay model and the reverse diffusion model, the outputs and predictive validity of the models are compared, and sensitivity of their parameters discussed. Through the three papers, the research identifies the advantages and limitations of both modelling approaches and suggests improvements to enable better forecasts and provide insights into consumer disposal behaviour.

Zusammenfassung Die moderne Gesellschaft ist von langlebigen Gebrauchsgütern durchdrungen. Technologische Fortschritte haben eine umfangreiche Palette von Konsumgütern auf den Markt gebracht, die den Menschen immer zugänglicher und erschwinglicher werden. In den letzten Jahrzehnten haben sich Gebrauchsgüter auf phänomenale Weise weltweit verbreitet. Die logische Folge dieser umfassenden, globalen Ausbreitung ist, dass eine immer größere Anzahl von Altgeräten in den Abfallstrom gelangt. Das wachsende Ausmaß dieses Stroms wirft sowohl theoretische als auch praktische Fragen auf, etwa wie viele und zu welchem Zeitpunkt Altgeräte entsorgt werden. Diese Doktorarbeit geht die Frage der Mengenabschätzung und Prognostizierung von Altgeräten an. In drei Abhandlungen werde ich verschiedene Modelle erarbeiten, anwenden und vergleichen, um die Entsorgungen von Altgeräten zu bewerten und zu prognostizieren.

Im ersten Papier entwickle ich ein gesellschaftliches Stock-Flow-Modell, das auf dem Modellierungsansatz einer Zeitverzögerung basiert und auch Elemente von Diffusionsmodellen, die häufig im Marketing verwendet werden, berücksichtigt. Ich schlage ein Teilmodell vor, welches das übliche Diffusionsmodell erweitert, um den Mehrfachbesitz von Gebrauchsgütern zu bewerten. Das Stock-Flow-Modell bezieht auch technologischen Ersatz ein und ermöglicht somit die Bewertung und Voraussage von Entsorgungen angesichts karger oder lückenhafter Daten.

Im zweiten Papier schlage ich ein Umkehr-Diffusionsmodell vor, das die Anwendbarkeit von Diffusionsmodellen auf das Gebiet der Abfallprognose ausweitet. Mit der Argumentation, dass die Dynamik der Entsorgung nicht viel anders als die der Einführung von Gebrauchsgütern ist, wird dieses Modell anhand von drei Fallbeispielen empirisch bestätigt.

Im dritten Papier werden schliesslich die Modellierungsansätze, die in den ersten zwei Papieren behandelt wurden, nämlich das Zeitverzögerungs-Modell und das Umkehr-Diffusionsmodell, miteinander verglichen. Indem derselbe reale Input-Datensatz an drei Varianten des Zeitverzögerungs-Modells und des Umkehr-Diffusionsmodells angewendet wird, werden die Outputs und die Validität der Modelle verglichen und die Sensitivitäten bezüglich der Modelparameter besprochen. In den drei Abhandlungen identifiziert diese Forschungsarbeit die Vorteile und die Einschränkungen beider Modellierungsansätze und macht Verbesserungsvorschläge, um bessere Prognosen zu ermöglichen und Einsichten in das Entsorgungsverhalten der Konsumenten zu gewinnen.

CHAPTER I

The Diffusion, Obsolescence and Disposal of Consumer Durables

1. Introduction

1.1. Diffusion of Consumer Durables

Innovation in the consumer durables industry has brought new products with new and improved functions and at increasingly affordable prices. The spread of an innovation in a market is termed “diffusion”. Schumpeter (1942) distinguishes three stages in the process of adoption of a new technological innovation: invention - the development of a scientifically or technically new idea; innovation – the incorporation of the idea in a product made available on the market; and finally diffusion – the process by which the product is available widely. The diffusion of consumer durables in modern society has been not only rapid, but also widespread. Consumer durables such as mobile phones, music players, personal computers and televisions are, to name a few, some of the most commonly available, widely adopted and aspirational products globally. Since their introduction in the early 20th century, consumer durables have become pervasive in homes and offices and many are no longer considered a luxury, but rather a necessity. In the past decades, electronic products have multiplied and become more accessible, affordable and numerable. From simple calculators to the latest smartphones and tablet computers, the price of electronic products has kept falling, while their features and functionality has kept rising. Recent figures from the International Telecommunications Union (ITU) indicate that global penetration of mobile phone subscriptions in 2011 reached 87% of the world population (ITU, 2012). Additionally, consumer durables are no longer limited to single-unit ownership per household. Households often have multiple mobile phones, music players, laptop computers and televisions. Not surprisingly, the number of electronic consumer durables sold annually has nearly tripled in the past decade as shown in Figure 1 (Euromonitor, 2011). Moreover, the miniaturization of electronics, instead of reducing the physical mass flow of hardware

Chapter I – Introduction to Thesis 1

by providing the same function with less, has in fact not helped to reduce the demand for products or total mass flow of products. In some cases it has in fact led to lower per functional unit costs, thereby greater demand, overcompensating for the reduction in material inputs (Hilty, 2005; Hilty et al., 2006).

Figure 1: Annual sales of electronic consumer durables

1.2. Obsolescence of Consumer Durables

Technological change, and the rapid pace at which it is taking place, is making not only products, but entire technologies obsolete. The personal computer-printer combine made typewriters extinct; the compact disc player replaced the cassette player and was itself replaced by MP3 players. Sood and Tellis (2005) define technological change in terms of the intrinsic characteristics of the technology, suggesting three types of technological change, namely through platform innovation (eg. from magnetic tape cassettes to compact disks), component innovation (eg. from magnetic tape to floppy disks), and design innovation (eg. from 5.25 inch floppy disks to 2.5 inch floppy disks). However, what is common to all three types of technological change is the emergence of a ‘dominant design’, when the new entrants displace incumbent technologies (Christensen, Suárez & Utterback, 1998; Srinivasan et al., 2006). Thus, technological change is understood as a shift from the existing dominant design to a new dominant design. A dominant design is a synthesis of fragmented technological innovations which may have been


introduced independently in prior products for specific user requirements. Thus, a dominant design embodies the requirements of many classes of users of a particular product and has the effect of enforcing industry standardisation (Suárez & Utterback, 1995). The emergence of a dominant design is a key event in the evolution of an industry (Utterback and Abernathy, 1975; Anderson and Tushman, 1990). Such a dominant design represents a milestone or transition point in the life of an industry as the standardisation allows production economies and effective competition can then take place on the basis of cost and product performance. An example of such a technological change is the shift in the display industry’s dominant design – from the Cathode Ray Tube (CRT) based displays, to the Flat Panel Displays (FPD). As Utterback & Suárez (1993) have identified, the CRT emerged as the dominant design for televisions in 1956, after which there have been incremental improvements in performance, functionalities and production techniques. This dominance continued for a long time, until recently, when the CRT was challenged by various flat panel display technologies. The graphic below shows the evolution of the television, starting as a CRT, with incremental improvements for 45 years until new display technologies came to the market in the late 1990s.

Figure 2: Technological evolution of the television (Source: Ahonen, 2011)


Technological substitution from one dominant design to another follows a classic sigmoidal curve (Fisher and Pry, 1971; Norton and Bass, 1987). The seminal paper by Fisher and Pry (1971) proposed a simple substitution model of technological change and has since been referenced extensively in diffusion literature to forecast the sales of new innovations. Figure 3 below illustrates this technological substitution with the transition of the Swiss television market from CRT TVs to FPD TVs (Source of data: see Annexe 2). Following the introduction of FPD TVs in 1998, CRT TVs start to decline, and are completely substituted within ten years.

Figure 3: Annual Sales of Televisions in Switzerland

1.3. Disposal of Consumer Durables

Rapid technological obsolescence is leading to growing quantities of consumer durables reaching the waste stream. Ongondo et al., (2011), referencing a study by Greenpeace, suggest an indicative range between 20 – 50 million tonnes of electric and electronic consumer durables disposed of annually worldwide. According to Huisman et al. (2008), the 27 European Union Member States alone generated between 8.3 and 9.1 million tonnes in 2005, with the figure expected to rise annually between 2.5% - 2.7% to reach 12.3 million tonnes by 2020. While the advantages of consumer durable products are evident, there is growing recognition of their environmental and social impacts given their mass production and intensive use of increasingly scarce resources (Brett, 2009; UNEP, 2009), improper


disposal, especially in developing countries and overall greenhouse gas emissions during their lifecycle (Erdmann and Hilty, 2010; Van Nes and Cramer, 2006; UNEP, 2009). The intensive use of scare resources in their mass production has raised also raised worries about future disruptions to supply of scarce raw materials (Wäger et al., 2010). For example, by some estimates, the global material supply of some critical metals in the manufacture of electronics such as Indium might be exhausted soon (UNEP, 2009). The increasing quantities of consumer durables in the waste stream and their associated environmental and social effects have led to legislation in many countries specifically for the disposal of such products. It has also led to the establishment of both formal and informal systems for their collection and a recycling. All this has made it ever more important to estimate and forecast the existing stocks and future flows of consumer durables. Such appraisals are useful for early recognition of environmental problems, for investment planning in production and waste management infrastructures, as also for government policy formulation, such as environmental policy, R&D funding emphasis, or strategic stockpile objectives (Müller, 2006). Better understanding and forecasting of this waste stream is therefore critical for producers, waste managers and policy makers alike.

2. Existing Research This thesis straddles two domains, namely waste management and consumer behaviour. Waste management research on consumer durables has a fairly recent history, with the majority of literature less than ten years old. In comparison, researchers in the field of marketing have been studying consumer behaviour related to the purchase for durable goods for over five decades.

2.1. Research in Waste Management

Most waste management research has tended to focus on municipal solid waste, in particular packaging. Early waste management models paid attention to the problems in subsystems, e.g. routing of vehicles and location of treatment and disposal facilities, with a strong focus on reducing costs, etc. More recently, waste management models have incorporated demographic, social and economic dynamics (Beigl et al., 2008).


However, these models are for frequent, high volume municipal solid waste comprising largely of organic matter, plastics, papers and packaging. The most common modelling approach for estimating and forecasting post-consumer end-of-life product flows is the “delay model approach” (Van der Voet, 2002), also sometimes referred to as the market supply approach (Widmer et al., 2005), based on combining sales data with average lifetime of the consumer durable. In its simplest form, time-series of sales of consumer durables is time-shifted by the fixed average lifetime. More recently, several researchers have improved upon this traditional model by combining product sales with a lifetime distribution such as Weibull (Oguchi et al., 2008) or a derived lifetime (Gregory et al., 2009). Such a delay model approach is also used in Material and Substance Flow Analysis (MFA and SFA) models to estimate flows of materials or substances through society, for example, cement (Müller, 2006; Kapur et al., 2008), lead (Elshkaki et al., 2005), copper (Lifset et al., 2002; Spatari et al., 2005) and zinc (Gordon et al., 2004). Citing Lohse et al. (1998), Widmer et al. (2005) have also mentioned the “consumption and use” and “market saturation” approaches for estimating end-of-life consumer durables. The consumption and use method takes the average number of consumer durables of a typical household as the basis for a prediction of the potential amount of end-of-life products, while the market saturation approach is based on the assumption that private households are already saturated with consumer durables, and for each new product purchased, an old one reaches its end-of-life. However, in literature, no applications of or further research on these approaches were found.

2.2. Research on Innovation and Diffusion of Consumer Durables

The diffusion, or adoption, of new products in the market has been of much managerial interest, and there is a wealth of research especially on the adoption of consumer durables, popularised in the marketing literature with the seminal article by Bass (1969). Numerous researchers have since extended the diffusion model further incorporating consumer behaviour insights. Diffusion models traditionally have been used in the context of forecasting, though they may also have other objectives, being used for descriptive or normative purposes, as pointed out by Mahajan and Wind (1985) and Kalish and Lilien (1986). Diffusion


models have been particularly useful in providing frameworks for understanding the processes by which new products come into circulation and spread across populations of adopters. The literature indicates that the predominant application of such models has been for purposes of forecasting new product adoption to predict the trajectory and ultimate market potential of the product. Mahajan et al. (1990), Mahajan et al.(1995) and Lilien, Rangaswamy and Van den Bulte (2000) describe some generic uses of diffusion modelling in marketing which include pre-launch forecasting; business valuation and strategic decision analysis based on the product life cycle; and the determination of optimal prices, etc.

2.3. Research on Obsolescence, Replacement and Disposal of Consumer Durables

Obsolescence, replacement and disposal of consumer durables are connected but separate concepts. Replacement and disposal are consumer decision points with a time gap between them ranging from seconds to decades. As Khetriwal and First (2011) suggest, forced replacements (eg. broken appliances) are very likely to lead to immediate disposal. Would forced replacement be the only disposal trigger, disposal and replacement could, within waste management, be used interchangeably. However, as many products are replaced before they fail, and/or subsequently often stored or reused elsewhere, the disposal decision can be long after the replacement decision. Khetriwal and First (2011) also differentiate replacement from obsolescence, suggesting that while unforced replacement is a possible outcome of obsolescence, obsolescence is a more inclusive situational factor which leads to disposal, regardless of whether a replacement occurred or not. Once the consumer starts perceiving a product as obsolete, the product might be either directly disposed of without being replaced (e.g. once the lifestyle or trends changes), can be first replaced and then disposed of (immediately or after a period of storage), or disposed and only later replaced. Research on disposal, or disposition (a term more commonly used in the consumer research domain), of durable goods started in the late 70s as an offshoot of consumer behaviour research. In their seminal article, Jacoby et al. (1977) concluded that although consumption consists of three stages, namely acquisition, actual consumption and disposition, the research focus had been largely on acquisition


phase, and almost non-existent on the disposition stage. Looking deeper at disposition behaviour, Jacoby et al. (1977) showed that factors influencing disposal behaviour are psychological characteristics of a decision maker, factors intrinsic to the product, and situational factors extrinsic to the product. Disposition behaviour thus is a function of disposition intention, social factors and situational factors (Hanson, 1980). Antonides (1991) also notes that the lifetime of a durable good is determined by a consumer’s decision which is in turn determined by economic, psychological and product-technical factors. A more recent study by Cooper (2004) focussed on disposal of consumer durables in UK households found similar results, with respondents echoing similar reasons for disposing of their durables. Cooper states that the three main reasons people dispose of their products are changes in consumer needs, dissatisfaction with product functionality and product failure.

3. Research Gap In the relatively recent literature in waste management focussing on the end-of-life disposal of consumer durables, forecasting models most commonly used have tended to base their forecasts only on available sales or shipment time series data, and generally using the “delay modelling approach”, forecast the timing and quantity of disposals. In case sales data was available for only a partial time period since the introduction of the product, these models have tended to estimate disposals based on only the available sales time-series. This neglects products in the waste stream from sales that took place in time periods before the sales data was available for. Surprisingly, none of the models estimating and forecasting end-of-life flows of consumer durables until now have validated their forecasts against real-system data, making it difficult to judge the accuracy or predictive validity of these models. In comparison, there exists a rich body of knowledge on the diffusion of consumer durables that analyses the timing of durable goods purchases by consumers. The literature indicates that the predominant application of “diffusion models” has been for purposes of forecasting new product adoption to predict the trajectory and ultimate market potential of the product. As diffusion models have largely been used in the marketing domain, the few authors that have incorporated replacements of consumer durables in the model are more from a perspective of estimating replacement sales, rather than estimating disposals. Thus, insights from the diffusion of consumer


durables have as yet not been incorporated into models for forecasting the timing of disposal of consumer durables. Additionally, there is as yet no research which provides a critique and comparison of models for forecasting disposals of consumer durables proposed by different authors. Comparing the outputs of the models to each other and against real-system data provides useful information not only in terms of the predictive validity of the model, but also enables a better understanding of the similarities and differences between the models, and their applicability in a given case.

4. Research Goal The research gaps identified above serve as a platform for this thesis. The main goal of the research is to apply concepts and insights from consumer durable diffusion to improve our understanding of, and forecasting models for, the obsolescence and disposal of end-of-life consumer durables. The overall research questions the thesis answers is whether and how forecasting models for disposal of consumer durables can be improved by incorporating insights from the extensive research on the adoption of consumer durables. This overall goal is achieved through the cumulative contribution of three specific parts: In the first part, diffusion modelling concepts are used to recreate inflows in the case of missing or patchy data. A diffusion model is proposed for multiple-unit adoptions of consumer durables. Applied together with the technology substitution model, the societal stock-flow model enables estimation and forecasting of disposal of consumer durables by the “delay model” approach. In the second part, insights from literature on consumer durable adoption and diffusion models to forecast sales of such products provides the conceptual basis for the “reverse diffusion” model for forecasting disposal of consumer durables.


In the third part, the performance, strengths and limitations of the delay model and the reverse diffusion modelling approaches are compared and improvements to both models are discussed based on insights from literature on consumer durable diffusion.

5. Thesis Architecture Models are central to this thesis as they are simplified representations of real systems, and can be used to explain, anticipate or design a real system (Schwaninger and Groesser, 2009). This thesis presents two models for anticipating waste flows from consumer durables, validating their results against real system data and finally suggesting improvements. The thesis is structured into four chapters as illustrated in Figure 4.

5.1. Chapter I

This first part, the “Introduction”, provides the context for the research and establishes its overall research purpose. Outlining the theoretical underpinnings of the following parts, the introduction provides the red-thread running through the three independent, yet interlinked papers presented in the thesis. It offers an overview of the related research on the diffusion of consumer durable products and forecasting models developed for the same, as well as disposal of end-of-life consumer durables and their estimation models.

5.2. Chapter II

Titled “From Introduction to Obsolescence: Estimating Societal Stocks and Flows of Consumer Durables”, is the first of three papers. This paper presents a stock and flow model to estimate the post-consumer flow of end-of-life consumer durables through society. The model developed in the paper, hereafter referred to as a “delay model”, both recreates and anticipates the stocks and flows of consumer durable inflows and outflows as well as the delay between the stock and the outflow while also incorporating aspects of technology substitution and multiple-unit product ownership. The model is empirically validated using the diffusion and disposal of cathode ray tube televisions in Switzerland as a case study.


5.3. Chapter III

Titled “Reverse Diffusion: Estimating Disposal of Consumer Durables through Application of Diffusion Modelling”, is the second of three papers. This paper proposes that the dynamics of disposal of consumer durables are not dissimilar to the adoption of new products. It proposes a new approach to forecasting disposals, building on extant literature on demand forecasting of new products based on diffusion models. A model operationalizing this approach is developed and empirically tested using the diffusion and disposal of three consumer durables in Switzerland as case studies.


Figure 4: Thesis architecture

CHAPTER IIntroductionExisting ResearchResearch in waste managementResearch on innovation and diffusion of consumer durablesResearch on obsolescence, replacement and disposal of consumer durablesResearch GapResearch GoalThesis ArchitectureDiscussion and Conclusions

CHAPTER II CHAPTER IIIIntroduction IntroductionLiterature Review Literature ReviewDelay Models or Market supply models Forecasting Adoption of Consumer DurablesMaterial Flow Analysis (MFA) models Consumer Disposition BehaviourDiffusion models Waste Forecasting ModelsGoal and Purpose Goal and PurposeConceptual Model & Mathematical Framework Conceptual Model and Mathematical Framework

Stocks Bass Diffusion ModelInflows Reverse DiffusionOutflows Model ParametersDisposal Distribution Function AssumptionsTechnology substitutionExperimental FrameApplication: Case Study CRT TVs in Switzerland Application: Case Study – Consumer Durables in

SwitzerlandExtensions to the Specific Model Case Study 1: CRT Monitors in SwitzerlandEstimating the Development of Household Population Case Study 2: CRT TVs in Switzerland

Estimating Devices per household - Multi-unit ownership sub-model

Case Study 3: LCD Monitors in Switzerland

ResultsModel Validation Model ValidationDiscussion & Conclusion Discussion & Conclusion

CHAPTER IV:IntroductionModelling approaches to estimate consumer durable disposalsTerminologyGoal and PurposeDelay Model ApproachDelay Model A: Example Reference – Oguchi et al., 2008Delay Model B: Example Reference – Gregory et al., 2009Delay Model C: Example Reference – Chapter IIReverse Diffusion Model ApproachStructural ComparisonExperimental FrameworkResultsOutput comparisonPredictive ValidityFit ImprovementSensitivity AnalysisDiscussion and Conclusion


5.4. Chapter IV

Titled “Estimating Consumer Durable Disposals: A Review and Comparison of Modelling Approaches”, is the last of the three papers. This paper reviews two modelling approaches to forecasting disposal of consumer durables, namely the “delay model” approach and the “reverse diffusion model” approach. Applying the same dataset on the disposal of cathode ray tube monitors in Switzerland to both the approaches, the estimates and forecasts of the models are compared against real system data, sensitivity of the model parameters examined and the assumptions, strengths, limitations and applicability of both modelling approaches discussed. The comparison also provides an opportunity to discuss further improvements to both modelling approaches, especially the importance of developing models which incorporate consumer disposal behaviour.

6. Discussion and Conclusions

6.1. Key Findings

Key findings and their theoretical and practical implications of the thesis are summarised below.

6.1.1. Applicability of Diffusion Modelling in Waste Management Research

Diffusion models have been successfully applied in the forecasting of the adoption of consumer durables. This research has extended the applicability of diffusion modelling in three ways – by proposing a multiple-unit adoption model, by incorporating technology substitution into the stock-flow model and by developing the reverse diffusion model. Multiple-unit adoption model: Although multiple-unit adoptions for many durable products (eg. televisions, mobile phones, gaming consoles, laptop computers) are becoming common, they have remained largely ignored in diffusion models. In Chapter II of the thesis, a multiple-unit adoption model based on the diffusion modelling tradition is presented. The model indicates that the multiple-unit adoptions take place in a phased approach. The first phase is given by the traditional diffusion model which models the first unit adoption. The additional-unit adoptions thereafter take place in the second stage, once first-unit market potential has been achieved, and


can also be characterised by a sigmoidal curve. By combining both phases, it is possible to estimate multiple-unit adoptions. Such a model is not only useful for waste forecasting, but for sales forecasting as well. Technology substitution: One of the main reasons for the disposal of consumer durables is obsolescence as a result of technological developments. Incorporating technological substitution into the stock-flow model, it is possible to improve the estimation and forecasting of stocks and flows of consumer durables. The Fisher-Pry technology substitution model which has been well documented in diffusion literature and empirically validated to fit across a range of technologies provides a useful model to combine with stock-flow model. Reverse diffusion model: The diffusion modelling framework provides the theoretical underpinnings of the reverse diffusion model described in Chapter III. The empirical results of the reverse diffusion model indicate that diffusion modelling concepts can be applied to the de-adoption, or disposal, of consumer durables. It lays a basis for further extensions to the reverse diffusion model, just as the diffusion model has been extended and improved over time. The advantages of the reverse diffusion model as compared to other models of estimating disposal of consumer durables are two-fold. Firstly, the model is parsimonious as it is able to provide disposal estimates even in the event of relatively sparse data on disposal. Secondly, the model makes it possible to estimate disposal flows in the absence of average lifetime and sales data as required by delay models.

6.1.2. Improving Forecasting Models for Disposal of Consumer Durables

A limitation of forecasting models for disposals of durable goods has been that it has not been possible to assess their predictive validity, largely due to a lack of data on disposals. An output comparison against real system data therefore provides a glimpse into the performance and predictive power of the models. The results presented in Chapter IV of the thesis indicate that societal stock-flow model and the reverse diffusion model perform better than two other models compared. Thus, it is suggested that the application of diffusion models can be used to improve models of forecasting disposals of consumer durables. Improving delay models: For delay models, time series data on inflows of consumer durables is essential as is the lifetime distribution. The sensitivity analysis illustrates Chapter I – Introduction to Thesis 14

the importance of parameter estimates, in particular the influence of average lifetime. In the absence of accurate lifetime distribution data, it may be better to estimate the parameters by linear optimisation within the model, as proposed in Chapter II to get more accurate forecasts. Such an approach can be particularly useful in cases where fragmented or incomplete data are available. Improving reverse diffusion models: The early forecasting efficacy of the reverse diffusion model is highly dependent on the upper limit of potential adoption of the durable. Sales forecasting models have successfully based estimates of potential adoption on the diffusion of analogous products. With the formalisation of the take-back and recycling system for end-of-life consumer durables, more data on disposal is expected to become available, and with it the possibility to have a library of reverse diffusion parameter estimates across product categories.

6.1.3. Importance of Consumer Behaviour:

The thesis sheds light on the importance of consumer behaviour on the disposal of durables. Additionally, based on the parameter values of the reverse diffusion model, it may be inferred that significantly smaller value of the coefficient of technical disposal, as compared to the value of the coefficient of discretionary disposal, indicates that the large majority of disposal of consumer durable products is driven by consumer behaviour. Influence of new technology: In Chapter II, the model overestimates sales of CRT TVs as compared to actual sales soon after the introduction of new technology products. This indicates that it is likely that consumers held back new TV purchases immediately following the introduction of the new technology. This latent demand for new TVs is then reflected in the sharp rise in the disposals of old technology CRT TVs soon after. Cultural influences: The research demonstrates the importance of the lifetime distribution in generating accurate forecasts. Using lifetime distribution estimates from Japan to estimate disposals in Switzerland resulted in significantly different disposal estimates. The values suggested a more rapid disposal rate in Japan than in Switzerland, suggesting that the timing of disposal of consumer durables may be attributed to cultural influences. It is likely that the lifetimes of consumer durables in Chapter I – Introduction to Thesis 15

Japan and Switzerland are significantly different due to different consumer usage patterns, or that there was a difference in the storage time, or likely a combination of both factors.

6.2. Practical Implications

Multiple forecasting models: The research presents and compares several forecasting models for disposal of consumer durables, based on two modelling approaches. The comparison of the two approaches provides valuable insights for forecasting disposals of consumer durables. Both modelling approaches are equally valid, each with its strengths and weaknesses, and can provide reasonably accurate forecasts given sufficient data. The choice of model used would therefore largely be dependent on the quantity and quality of the data available. The societal stock-flow model offers insights into the societal stocks and flows, from its introduction to its obsolescence, including peaks of sales, disposals and installed base of a consumer durable, especially with only limited data availability. Such a model can provide collection and take-back systems as well as recycling companies with better forecasts to help plan their capacities. For policy makers, it provides a gauge of the collection efficiency of a formal take-back and collection system, and a basis to check against potentially environmentally harmful or materially significant leakages. The reverse diffusion model provides a simple and parsimonious model as it is able to provide disposal estimates even in the event of relatively sparse data on disposal. Additionally, the model makes it possible to estimate disposals in the absence of data on average lifetime and sales time series. Such a model can be particularly useful for established take-back and collection systems. Moreover, its parameters can be updated every year by fitting to the latest available data, thereby enabling ever more accurate forecasts for the years ahead. Estimating historic flows: In addition to providing estimates of future waste flows, the societal stock-flow model presented in the thesis makes it possible to estimate historic disposals. Such data can indicate the existence of anthropogenic stores of disposed


consumer durables which could pose a risk due to hazardous substances, but are also increasingly being considered as urban mines containing precious and rare materials.

6.3. Conclusion

This cumulative thesis on the disposal, obsolescence and disposal of consumer durables sheds new light on the disposal of consumer durables and builds interdisciplinary bridges between the sales forecasting and waste forecasting domains. The research is however not without limitations, which provide directions for future research. Models for forecasting consumer durable disposals currently provide only statistical estimates, without any explicit representation of the underlying consumer disposal behaviour and the socio-economic factors which may play a role in the timing of disposal of consumer durables. Future research can look at incorporating such aspects into both the delay as well as reverse diffusion modelling approaches. The reverse diffusion model could potentially be extended by disaggregating discretionary disposal into various aspects to get better understanding of drivers of disposal. Further research can look to incorporating variables such as competitive effects of new technology, advertising, product quality, price and income effects into the model. Another limitation of the models discussed in this research is that they do not explicitly account for storage time between consumers replacing their consumer durables and disposing them. In the delay model approach, the assumption is made that consumer durables are either in active stock or disposed of, not accounting for time spent in private storage. Consumer durables may be stored in attics or basements or garages for months or even years before finally being disposed of. The models presented in the thesis are, as yet, unable to provide insight into such behaviour. Incorporating such aspects in a model could help give estimates regarding “hibernating” stocks, which could be particularly helpful in understanding potential anthropogenic stocks available for recycling and recovery, especially in the light of material scarcity. The delay model is well suited to disaggregation into stages such as reuse and storage. Such a “nested-delay” model has been presented by Widmer et al. (2005) for disposals of personal computers, albeit using a fixed (dirac) lifetime distribution. Further research on consumer behaviour will not only be able to provide


better parameter estimates for the delay functions at every stage, but also inform model design in terms of stages an end-of-life consumer durable. Furthermore, neither of the models incorporates the time-varying nature of consumer behaviour. Given the fairly long societal existence of consumer durables compared to the rapidly changing consumer preferences and perceptions of obsolescence, it is likely that the residence time of a consumer durable changes between its introduction and decline. Anecdotal evidence for PCs and mobile phones has shown that average lifetime of these products has reduced over time, with more frequent replacements and disposals taking place. In the delay models, the disposal function is time-invariant, while in the reverse diffusion model, the coefficients p (coefficient of technical disposal) and q (coefficient of discretionary disposal) are constant over the time horizon of the model application. In both cases changing consumer behaviour due to newer products, lower prices of newer products, more convenient disposal opportunities, etc. (which all may lead to disposals), are not explicitly considered. Moreover, it would also provide an opportunity to examine whether the disposal of consumer durables accelerates between technology generations. Further research into consumer behaviour to get insights into why, when and how consumers dispose their durable products, will provide useful information that could lead not only to better, more accurate forecasting models, but also inform consumer education and awareness programs directed towards improving consumer attitudes towards disposal of durable products.


References Ahonen, Tomi. 2011 ‘Digital Divide: Global Household Penetration Rates for Technology’. Bright Side Of News*. http://www.brightsideofnews.com/news/2011/1/26/digital-divide-global-household-penetration-rates-for-technology.aspx. [Accessed 21st February, 2012]. Anderson, Philip, and Michael L. Tushman. 1990. ‘Technological Discontinuities and Dominant Designs: A Cyclical Model of Technological Change’. Administrative Science Quarterly 35 (4) (December 1): 604–633. doi:10.2307/2393511. Antonides, Gerrit. 1991. ‘An Economic-psychological Model of Scrapping Behavior’. Journal of Economic Psychology 12 (2) (June): 357–379. doi:16/0167-4870(91)90020-T. Bass, Frank M. 1969. ‘A New Product Growth for Model Consumer Durables’. Management Science 15 (5, Theory Series) (January): 215–227. Beigl, Peter, Sandra Lebersorger, and Stefan Salhofer. 2008. ‘Modelling Municipal Solid Waste Generation: A Review’. Waste Management 28 (1): 200–214. doi:10.1016/j.wasman.2006.12.011. Brett H., Robinson. 2009. ‘E-waste: An Assessment of Global Production and Environmental Impacts’. Science of The Total Environment 408 (2) (December 20): 183–191. doi:10.1016/j.scitotenv.2009.09.044. Cooper. 2004. ‘Inadequate Life? Evidence of Consumer Attitudes to Product Obsolescence’. Journal of Consumer Policy 27 (4) (December 1): 421–449. doi:10.1007/s10603-004-2284-6. Christensen, Clayton M., Fermando.F. Suárez, and James M. Utterback. 1998. ‘Strategies for Survival in Fast-changing Industries’. Management Science: 207–220. Elshkaki, Ayman, Ester van der Voet, Veerle Timmermans, and Mirja Van Holderbeke. 2005. ‘Dynamic Stock Modelling: A Method for the Identification and


Estimation of Future Waste Streams and Emissions Based on Past Production and Product Stock Characteristics’. Energy 30 (8) (June): 1353–1363. doi:10.1016/j.energy.2004.02.019. Erdmann, Lorenz, and Lorenz M Hilty. 2010. ‘Scenario Analysis’. Journal of Industrial Ecology 14 (5) (October 1): 826–843. doi:10.1111/j.1530-9290.2010.00277.x. Euromonitor. 2011. ‘Consumer Electronics 2010’. Fisher, J. C., and R. H. Pry. 1971. ‘A Simple Substitution Model for Technology Change’. Technological Forecasting and Social Change 3 (1): 75–88. Gordon, R. B., R. J. Lifset, M. Bertram, B. Reck, T. E. Graedel, and S. Spatari. 2004. ‘Where Is All the Zinc Going: The Stocks and Flows Project, Part 2’. JOM Journal of the Minerals, Metals and Materials Society 56 (1): 24–29. Gregory, Jeremy R., Marie-Claude Nadeau, and Randolph E. Kirchain. 2009. ‘Evaluating the Economic Viability of a Material Recovery System: The Case of Cathode Ray Tube Glass’. Environmental Science & Technology 43 (24) (December): 9245–9251. doi:10.1021/es901341n. Hanson, James W. 1980. ‘A Proposed Paradigm for Consumer Product Disposition Processes.’ Journal of Consumer Affairs 14 (1): 49.. Hilty, Lorenz M. 2005. ‘Electronic Waste? An Emerging Risk?’ Environmental Impact Assessment Review 25: 431? 435. Hilty, Lorenz M., Andreas Köhler, Fabian von Schéele, Rainer Zah, and Thomas Ruddy. 2006. “Rebound Effects of Progress in Information Technology.” Poiesis & Praxis. International Journal of Technology Assessment and Ethics of Science 1 (4): 19-38 Huisman, J., F. Magalini,, R. Kuehr, C. Maurer, J. Poll, C. Delgado, E. Artim, et al. 2008. ‘2008 Review of Directive 2002/96 on Waste Electrical and Electronic Equipment (WEEE), Final Report’. Chapter I – Introduction to Thesis 20

ITU. 2012. ‘The World in 2011 – ICT Facts and Figures’. http://www.itu.int/ITU-D/ict/facts/2011/material/ICTFactsFigures2011.pdf [Accessed 21st February 2012) Jacoby, Jacob, Carol K. Berning, and Thomas F. Dietvorst. 1977. ‘What About Disposition?’ The Journal of Marketing 41 (2) (April): 22–28. Kalish, Shlomo, and Gary L. Lilien. 1986. ‘A Market Entry Timing Model for New Technologies’. Management Science 32 (2) (February 1): 194–205. Kapur, Amit, Gregory Keoleian, Alissa Kendall, and Stephen E. Kesler. 2008. ‘Dynamic Modeling of In-Use Cement Stocks in the United States’. Journal of Industrial Ecology 12 (4) (August): 539–556. doi:10.1111/j.1530-9290.2008.00055.x. Khetriwal, Deepali S., and Ivana First. 2011. ‘Enabling Closed Resource Loops in Electronics: Understanding Consumer Disposal Behaviour Using Insights from Diffusion Models’. In Proceedings of CROMAR 2011. Pula. Lifset, RJ, RB Gordon, TE Graedel, S. Spatari, and M. Bertram. 2002. ‘Where Has All the Copper Gone: The Stocks and Flows Project, Part 1’. JOM Journal of the Minerals, Metals and Materials Society 54 (10): 21–26. Lilien, G. L., A. Rangaswamy, and C. Van den Bulte. 2000. ‘Diffusion Models: Managerial Applications and Software’. New Product Diffusion Models: 295–311. Lohse, J., S. Winteler, and J. Wulf-Schnabel. 1998. Collection Targets for Waste from Electrical and Electronic Equipment (WEEE) the Directorate General (DG XI) Environment. Mahajan, V., Y. Wind, and Wharton School. Marketing Dept. 1985. Innovation Diffusion Models of New Product Acceptance: A Reexamination. Wharton School, University of Pennsylvania, Marketing Dept. Mahajan, Vijay, Eitan Muller, and Frank M. Bass. 1990. ‘New Product Diffusion Models in Marketing: A Review and Directions for Research’. The Journal of Marketing 54 (1) (January): 1–26.


Mahajan, V., Eitan Muller, and Frank M. Bass. 1995. ‘Diffusion of New Products: Empirical Generalizations and Managerial Uses’. Marketing Science 14 (3): 79–88. Müller, Daniel B. 2006. ‘Stock Dynamics for Forecasting Material flows—Case Study for Housing in The Netherlands’. Ecological Economics 59 (1) (August): 142–156. doi:10.1016/j.ecolecon.2005.09.025. Norton, John A., and Frank M. Bass. 1987. ‘A Diffusion Theory Model of Adoption and Substitution for Successive Generations of High-Technology Products’. Management Science 33 (9) (September): 1069–1086. Oguchi, Masahiro, Takashi Kameya, Suguru Yagi, and Kohei Urano. 2008. ‘Product Flow Analysis of Various Consumer Durables in Japan’. Resources, Conservation and Recycling 52 (3) (January): 463–480. doi:10.1016/j.resconrec.2007.06.001. Ongondo, F.O., I.D. Williams, and T.J. Cherrett. 2011. ‘How Are WEEE Doing? A Global Review of the Management of Electrical and Electronic Wastes’. Waste Management 31 (4) (April): 714–730. doi:10.1016/j.wasman.2010.10.023. Schumpeter, J. A. 1942. ‘Capitalism’. Socialism and Democracy 3. Schwaninger, M., and S.N. Grösser. 2009. “System dynamics modeling: validation for quality assurance.” Encyclopedia of Complexity and System Science. Springer, Berlin. Spatari, S., M. Bertram, R. B Gordon, K. Henderson, and T. E. Graedel. 2005. ‘Twentieth Century Copper Stocks and Flows in North America: A Dynamic Analysis’. Ecological Economics 54 (1): 37–51. Sood, Ashish, and Gerard J. Tellis. 2005. ‘Technological Evolution and Radical Innovation’. The Journal of Marketing 69 (3) (July 1): 152–168. Srinivasan, Raji, Gary L. Lilien, and Arvind Rangaswamy. 2006. ‘The Emergence of Dominant Designs’. SSRN eLibrary. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=861304.


Suárez, Fernando F, and James M Utterback. 1995. ‘Dominant Designs and the Survival of Firms’. Strategic Management Journal 16 (6) (January 1): 415–430. doi:10.1002/smj.4250160602. UNEP. 2009. ‘Recycling – From E-waste to Resources’. Utterback, James M, and William J Abernathy. 1975. ‘A Dynamic Model of Process and Product Innovation’. Omega 3 (6) (December): 639–656. doi:10.1016/0305-0483(75)90068-7. Utterback, James M., and Fernando F. Suárez. 1993. ‘Innovation, Competition, and Industry Structure’. Research Policy 22 (1): 1–21. Van der Voet, Ester, René Kleijn, Ruben Huele, Masanobu Ishikawa, and Evert Verkuijlen. 2002. ‘Predicting Future Emissions Based on Characteristics of Stocks’. Ecological Economics 41 (2) (May): 223–234. doi:10.1016/S0921-8009(02)00028-9. Van Nes, Nicole, and Jacqueline Cramer. 2006. ‘Product Lifetime Optimization: a Challenging Strategy Towards More Sustainable Consumption Patterns’. Journal of Cleaner Production 14 (15-16): 1307–1318. doi:10.1016/j.jclepro.2005.04.006. Wäger, Patrick, Daniel Lang, Raimund Bleischwitz, Christian Hagelücken, Simon Meissner, Armin Reller, Dominic Wittmer. 2010. Raw Material for Technologies of the Future. http://www.worldresourcesforum.org/files/SelteneMetalle_kurz_EN.pdf. [Accessed 4th September, 2011] Widmer, R., H. Oswald-Krapf, D. Sinha-Khetriwal, M. Schnellmann, and H. Böni. 2005. ‘Global Perspectives on E-waste’. Environmental Impact Assessment Review 25 (5): 436–458.


CHAPTER II

From Introduction to Obsolescence: Estimating Societal Stocks and Flows of Consumer Durables

Deepali Sinha, University of St.Gallen Abstract The long residence time of consumer durables in households makes it important to estimate their entire anthropogenic stocks and flows, both from a waste management as well as a material management perspective. This paper presents a stock and flow model to estimate the post-consumer flow of end-of-life consumer durables through society. The descriptive model developed in the paper both recreates and anticipates the stocks and flows. The model incorporates the aspects of technology substitution as well as multiple-unit product ownership. The model is empirically validated using the diffusion and disposal of cathode ray tube televisions in Switzerland as a case study.

Chapter II – From Introduction to Obsolescence: 24 Estimating Societal Stocks and Flows of Consumer Durables

1. Introduction Electrical and electronic consumer durables have become entrenched in modern society since their first appearance in the early 19th century. The success of the electronics industry over the last several decades in developing a mass consumer market for personal electronic equipment has been phenomenal. Not only have electrical and electronic products proliferated, they have also been the focus of intense technological innovation that has been both incremental and disruptive. This technological progress is making not only products, but entire technologies extinct. The intense competition in the industry has meant an ever increasing pace of technological change, making the time interval between successive generations of products and technologies relatively small in comparison with the time interval between replacing technologies using historical norms (Norton and Bass, 1987). The personal computer-printer combine made typewriters extinct; the compact disc replaced cassette players and was itself replaced by even more compact solid state storage music players; cathode ray tube televisions are being replaced by liquid crystal displays which in turn will be made obsolete by Organic Light Emitting Diode (OLED) televisions in the near future, also taking a leap from two dimensional (2D) to three dimensional (3D). Additionally, the rise in the number of consumer durables is because many products are no longer limited to single-unit ownership per household. Households often have multiple mobile phones, music players, laptops and televisions. Multiple-unit adoptions are a major component of sales for many consumer durable product categories, and authors from the marketing domain such as Steffens (2003) have identified the importance of including multiple-unit adoptions in sales forecasting models. Moreover, as Hilty (2005) suggests, the continued miniaturisation of consumer durables, especially electronics, has resulted in the price per functional unit falling, triggering greater demand which compensates, or sometime even overcompensating for the miniaturisation effect in terms of mass flow. Based on data for Switzerland, Hilty et al. (2006) show the considerable reduction in the average physical mass of a mobile phone from over 350 g in 1990 to about 80 g in 2005, which corresponds to a reduction by a factor of 4.4, was accompanied by an increase in the number of


subscribers, which in turn led to a rise of the total mass flow by a factor of 8.0 in that period. Thus, the increasing number of consumer durables, combined with increasing affordability and multiple-unit ownership further compounded by shorter time periods between technological generations has meant an increasing volume of end-of-life consumer durables being disposed. The long residence time of consumer durables in households makes it important to estimate their entire anthropogenic stock and flows, both from a waste management as well as a material management perspective. Estimation and quantification of this growing volume of end-of-life equipment (EOLE) commonly known as e-waste, has drawn the attention of several scholars. Various models of forecasting EOLE flows have been presented, the most common being the delay modelling approach. A drawback of these models is that their forecasts are highly dependent on the sales data and average lifetime, and are unable to estimate or forecast the entire stock and flow of the consumer durable over its societal lifetime, from its introduction to its obsolescence. The goal of this paper is to present an approach to estimate the stocks and flows of a product through society, especially in light of incomplete and patchy data over the entire time period. The model developed in the paper both recreates and anticipates the stocks and flows (Schwaninger, 2010). The paper makes a contribution in two ways: firstly, the model incorporates multiple-unit ownership of consumer durables in a dynamic model to estimate inflows, outflows and stocks of a product. Secondly, the model provides an estimate of stocks and outflows of a consumer durable, given patchy data on sales, disposals and installed base. Societal stocks and flows of consumer durables, especially those in the midst of a technological shift, are of particular interest to producers, policy makers and waste managers in understanding the quantity and timing of the outflow of these durables into the waste stream. This model will therefore be useful as a tool for managerial decision making. The paper is organised as follows. In the next section, a review of relevant literature is presented followed by the purpose and objectives of the research. Section 4 describes


the conceptual and analytical bases of the model, followed by a description of the model parameters and variables in Section 5, the experimental frame in Section 6 and the data in Section 7. Section 8 presents the sub-model for multiple-unit adoption. In Section 9, the model formulation is empirically examined with data for televisions in Switzerland following which model validity is discussed. Section 11 concludes the paper with a summary of the contributions, limitations and suggests areas for future research.

2. Literature Review A number of studies have been conducted on the estimated product flow of consumer durables, both sales to households and disposals from households. In this section, a literature review of previous research on existing models of estimation and forecasting of end-of-life products is presented. Three main modelling approaches which have been applied to consumer durables are reviewed, namely market supply models, material flow analysis models and diffusion models.

2.1. Delay Models or Market Supply Models

Market supply models are perhaps the most common approach used for forecasting post-consumer end-of-life product flows. Waste estimates (outflows) can be made by modelling discards as a function of sales (inflows), distributed over time given by a lifetime distribution function or a derived lifetime distribution. The most basic market supply models combine sales data with a fixed average lifetime or residence time to forecast waste flows of end-of-life consumer durables (Widmer et al., 2005; Kang and Schoenung, 2006). More sophisticated market supply models combine product sales with a lifetime distribution such as Weibull (Oguchi et al., 2008), a derived lifetime distribution (Gregory et al., 2010; Yu et al., 2010) or a likelihood of failure in the shape of a bath-tub curve (Linton et al., 2002). While this method provides estimates of outflows, a pre-requisite for these models is data on inflows, i.e. sales or shipments data.

2.2. Material Flow Analysis (MFA) Models

Material and substance flow analysis models have been commonly used to estimate societal stocks and flows of materials such as cement (Mueller, 2006; Kapur et al.,


2008), lead (Elshkaki et al., 2005), copper (Lifset et al., 2002; Spatari et al., 2005) and zinc (Gordon et al., 2004). Most previous MFA models focussed only on flows, however, recently researchers have realised that stocks may be equally or sometimes even more important, especially in the prediction of future emissions and waste flows of products with a long life span (Kleijn et al., 2000). Recent material flow analysis models have been used to forecast waste material flows, especially for construction and demolition waste (Bergsdal et al., 2007; Hu et al., 2010) as well as e-waste (Streicher-Porte et al., 2005) who applied MFA to assess obsolete PC processing in the informal sector in Delhi, India. However, most material flow analysis models have system boundaries which include material flows from production to disposal, including mining, fabrication and manufacture before the consumption phase as they look at materials rather than products, which are a composite of many materials.

2.3. Diffusion Models

Forecasting the sales of consumer durables has been the focus of a significant body of research, ever since the 1960’s. The seminal article by Bass (1969) elaborating the application of a logistic-curve based model to forecast the diffusion of innovations into a market was the starting point for a wave of models which since further extended and improved upon the original Bass Diffusion Model (BDM). While BDMs have been applied to all sorts of products, one of the most common applications is the forecasting of consumer durables sales. A comprehensive review of the literature on diffusion of new products can be found in Mahajan et al. (1990), Meade and Islam (2006), and most recently Peres et al. (2010). Three especially relevant extensions of the BDM in the context of obsolescence of consumer durables are the inclusion of replacement sales (Kamakura and Balasubramanian, 1987; Bayus, 1991; Mahajan and Mueller, 1996), technology substitution (Fisher and Pry, 1971), and multi-generation product models (Norton and Bass, 1987). Walk (2009) is one of the very few authors to utilise diffusion modelling from the marketing domain for the estimation and forecasting of end-of-life consumer durables. He presents a three-step forecasting model for CRT appliances (television sets and


monitors) for a region in Germany which includes modelling product life time, extrapolation of stocks and modelling technology change based on the Fisher-Pry substitution model. More recently, Yu et al. (2010) have also used a diffusion model together with material flow analysis to forecast the global generation of obsolete Personal Computers (PCs). They apply the diffusion model to estimate historic and future sales of PCs, combining with an average lifetime distribution to provide scenarios of generation of obsolete PCs. This paper utilises approaches presented in papers by Walk (2009) and Yu et al. (2010), as well as that of material flow analysis, combining and extending their concepts to present a model of societal stocks and flows of consumer durables.

3. Goal and Purpose From the above discussion of the existing models of forecasting flows, it is clear that considerable work has been done in the estimation and forecasting of end-of-life consumer durables. However, the drawbacks of the models discussed above is that they can be used only given sufficient sales data, and are unable to estimate the flows of consumer durables with multiple-unit ownership, or are limited to only a part of the period since the introduction to the obsolescence of the product, rather than looking at the entire societal flow. Hence, the goal of this paper is to develop a model to estimate the stocks and flows of a consumer durable product through society from its introduction to its obsolescence, incorporating the substitution effect of changing technology, which includes the possibility of multiple-unit ownership of the durable good. The purpose of the model is to provide a forecast for the end-of-life waste flows of a consumer durable product, providing estimates to waste managers, policy makers and recyclers on existing stock and expected waste volumes. The contribution of the paper is twofold: Firstly, it provides a model to estimate the total societal stock and flow of a consumer durable product category over its lifetime in the market, from its introduction to withdrawal. Secondly, in the course of


developing this stock-flow model, a sub-model for estimation of stock for durables of which households may have multiple-unit ownership is also presented.

4. Conceptual Model & Mathematical Framework Societal models to estimate the stocks and flows of consumer durables can be thought to be similar to population models, with a stock or installed base equivalent to total population and sales (inflows) = births and disposals (outflows) = deaths. They can also be considered to be analogous to models of epidemics, with the rise and fall in the number of infected persons equivalent to the sales and disposals of a product (Sterman, 2000). Though population models have relatively long time horizons, typically decades or centuries and epidemics relatively shorter time horizons typically from a few weeks to a few years, with consumer durables somewhere in between, the underlying dynamics of stocks and flows are similar. Not surprisingly, the logistic model of diffusion of consumer durables has its roots in ecology in modelling population growth (Yu et al., 2010). In this paper, a stock-flow model is presented, combining concepts of MFA models, waste flow estimation based on distributed residence time as well as diffusion models of consumer durable adoption and substitution. In MFA, time step changes in stock are determined by tracking flows into and out from the stock. In the context of consumer durables, for this model, the installed base of a product in households is the stock, with flows in being the equivalent of sales or shipments and flows out being disposals of the product.

5. Model Parameters and Variables

5.1. Stocks

A stock is the integral of inflows and outflows over time – stocks grow when the inflows exceed the outflows of a system, and vice versa. Kleijn et al. (2000) distinguish between two types of relations between stocks and flows: stocks as a size buffer and stocks as a time buffer. A stock acts like a size buffer when the outflow is proportional to the magnitude of the stock and independent of the time of inflow into the system. In contrast, when outflow is dependent on the time of inflow, stocks act


as a time buffer. Consumer durable stocks are time buffers because the timing of disposal of a product is linked to when it entered into use, as it is more likely for older devices to be disposed of than newer devices. The stock-flow diagram below graphically illustrates the relationship between inflows, stocks and outflows.

Figure 5: Graphical illustration of generic stock-flow model The relationship between stocks and flows is represented by: In differential form: 𝑑𝑑𝑆(𝑡𝑡) = 𝐼𝐼𝐼𝐼(𝑡𝑡)𝑑𝑑𝑡𝑡 − 𝑂𝑂𝑂𝑂𝑡𝑡(𝑡𝑡)𝑑𝑑𝑡𝑡

(1)

In integral form:

𝑆(𝑡𝑡) = ��𝐼𝐼𝐼𝐼(𝜏) − 𝑂𝑂𝑂𝑂𝑡𝑡(𝜏)�𝑑𝑑𝜏𝑡

𝑡0

(2)

In difference form : ∆𝑆𝑡+1 = 𝑆𝑡+1 − 𝑆𝑡 = (𝐼𝐼𝐼𝐼𝑡 − 𝑂𝑂𝑂𝑂𝑡𝑡𝑡)∆𝑡𝑡

(3)

As a recurrence relation (to numerically integrate the differential equation (1)): 𝑆𝑡+1 = 𝑆𝑡 + (𝐼𝐼𝐼𝐼𝑡 − 𝑂𝑂𝑂𝑂𝑡𝑡𝑡)∆𝑡𝑡

(4)

where 𝑆(𝑡𝑡), 𝑆𝑡 represent the stock, 𝐼𝐼𝐼𝐼(𝑡𝑡), 𝐼𝐼𝐼𝐼(𝜏), 𝐼𝐼𝐼𝐼𝑡 represent the inflow and 𝑂𝑂𝑂𝑂𝑡𝑡(𝑡𝑡),𝑂𝑂𝑂𝑂𝑡𝑡(𝜏),𝑂𝑂𝑂𝑂𝑡𝑡𝑡 represent the outflow.

�[𝐼𝐼𝐼𝐼(𝑡𝑡) − 𝑂𝑂𝑂𝑂𝑡𝑡(𝑡𝑡)]𝑑𝑑𝑡𝑡

Stock (t) = Out (t) In (t)


The stock of a consumer durable in households at any time t can be considered as the number of all households 𝐻𝐻𝑡 multiplied by the number of devices of the consumer durable 𝐷𝑃𝐻𝑡 each household owns on average at the time t.

𝑆𝑡 = 𝐻𝐻𝑡 ∙ 𝐷𝑃𝐻𝑡 (5) In the event of technological change, several technologies compete to render the same service, thus, the total installed base is a sum of all stocks of all technologies.

𝑆𝑡𝑜𝑡𝑎𝑙 = �𝑆𝑘𝑘

(6)

where 𝑆𝑡𝑜𝑡𝑎𝑙 is the total stock of all devices of all technologies 𝑘 at any time. Initially, when there is only one type of technology to render the same service, change in the stock of technology type 1 is equal to the change in total stock, 𝑆𝑡𝑜𝑡𝑎𝑙. However, with the advent of new technology and saturated 𝑆𝑡𝑜𝑡𝑎𝑙 = 𝑐𝑜𝐼𝐼𝑠𝑡𝑡, as technology type 1 is substituted by type 2, stock of technology type 1, 𝑆1, decreases as stock of technology type 2, 𝑆2, increases. Therefore, ∆𝑆1 is positive initially, becoming negative following the introduction and diffusion of technology type 2. Technological substitution is discussed separately further in the paper

5.2. Inflows

Previous authors have used sales or shipment data from market research to get inflow values. However, in the absence of existing sales data, especially historic sales data going back to the introduction of the consumer durable, the inflow can be constructed by using data on change in stock and disposals. From equation (3),

𝐼𝐼𝐼𝐼𝑡 = Δ𝑆𝑡+1 + 𝑂𝑂𝑂𝑂𝑡𝑡𝑡 (7)

5.3. Outflows

Outflows, or disposals, of consumer durables are a function of inflow (sales) and of the residence time (given as a survival or reliability function R(t)) of the product in a


household. As this model assumes that all products are disposed of at end of life the lifetime distribution function 𝐹(𝑡𝑡) = 1 − 𝑅(𝑡𝑡) is also the disposal distribution D(t) and its derivative, the disposal density function d(t), describes the rate of disposals.

Figure 6: Stock and flow diagram for the delay model ('*' represents a convolution) The outflows (disposals) at time t are then expressed as a convolution of the inflows In and the disposal density d, In discrete form the convolution writes:

(𝑓 ∗ 𝑔)[𝑘] = � 𝑓[𝑖]∞

𝑖= −∞

⋅ 𝑔[𝑘 − 𝑖] (8)

Thus,

𝑂𝑂𝑂𝑂𝑡𝑡𝑡 = � 𝑑𝑑𝑖 ⋅ 𝐼𝐼𝐼𝐼𝑡 − 𝑖

∞

𝑖=−∞

(9)

where 𝑂𝑂𝑂𝑂𝑡𝑡𝑡 is the number of products disposed and 𝑑𝑑𝑡(𝑖) the disposal function.

�𝑑𝑑𝑡𝑡

disposal d(t)In(t)*d(t)

Out (t) In (t)

Stock (t)


If both 𝐼𝐼𝐼𝐼𝑡 and 𝑂𝑂𝑂𝑂𝑡𝑡𝑡 are zero outside a time window 𝑇𝑚𝑖𝑛 < 𝑖, 𝑡𝑡 < 𝑇𝑚𝑎𝑥 then the integration interval can be limited to [2𝑇𝑚𝑖𝑛, 2𝑇𝑚𝑎𝑥]. Combining equations (4) and (9) and taking care of the required integration interval e.g. 𝑇𝑚𝑖𝑛 = 0,𝑇𝑚𝑎𝑥 = 𝑇𝑙𝑎𝑠𝑡_𝑠𝑎𝑙𝑒 + 𝑇𝑙𝑎𝑠𝑡_𝑓𝑎𝑖𝑙𝑢𝑟𝑒:

𝑆𝑡+1 = 𝑆𝑡 + �𝐼𝐼𝐼𝐼𝑡 − �𝑑𝑑𝑖 ⋅ 𝐼𝐼𝐼𝐼𝑡 − 𝑖

𝑡

𝑖=0

�∆𝑡𝑡

(10)

5.4. Disposal Distribution Function

The outflow in the delay model is solely determined by the inflow and the residence time of the product in a household. The latter is defined as the probability that the time of obsolescence is later than some specified time t. 𝐹(𝑡𝑡) = 𝐷(𝑡𝑡) = 1 − 𝑅(𝑡𝑡) (11)

and

𝐷(𝑡𝑡) = � 𝑑𝑑(𝜏)𝑑𝑑𝜏𝑡

0

(12)

in discretized form:

𝐷𝑡 = �𝑑𝑑𝑖

𝑡

𝑖=0

(13)

where 𝑑𝑑𝑖 is the discrete disposal density of the consumer durable. As consumer durables are often reused, resold and stored by consumers before finally being disposed of (Widmer et al., 2005), it is important to specify the point at which the product is considered as waste and included in outflow. As Yu et al., (2010) note, different definitions of lifetime are possible which could be limited only to the length of time a household uses a device or could be inclusive of time in storage after the device has left the active stock. Spatari et. al., (2005) and Yu et. al., (2010) consider


lifetime as limited to only products in use, not including time spent “hibernating”, i.e. products that have been retired but remain in households unused. In contrast, Kang & Schoenung (2006) and Oguchi et al.,(2008) among other consider their stocks to include all products in households, whether being used or stored. They explicitly state that hibernating or out-of-use products stored in households are not considered as waste, and hence form part of the consumption stock. In this model, the definition of residence time used is similar to Spatari et. al., (2005) and Yu et. al., (2010) whereby the disposal function does not consider stored or hibernating products in stocks, but rather includes them in the outflows once they are not part of the active stock. The shape of the disposal density function is also important. While some models use a Dirac distribution, or a fixed average lifetime, distributed lifetimes are more common because some individual products will be discarded earlier than others. The rate of disposal of a consumer durable device can be described by many probability densities in an analogous way to product failure. This is a reasonable assumption given that disposals are either due to technical failure or discretionary obsolescence, both of which are correlated with the product’s age (Steffens, 2003). Following its acceptance in failure analysis, the Weibull distribution is the most commonly used distribution to model the lifetime distribution (Oguchi et al., 2008; Walk, 2009; Yu et al., 2010). Therefore, in this model, a Weibull distribution is used for the disposal distribution function; it's discrete density is given by the equation:

𝑑𝑑𝑖 = 𝛾𝛼

�𝑖𝛼�𝛾−1

𝑒−�𝑖𝛼�

𝛾

(14)

where 𝛾 is the shape parameter and 𝛼 the scale parameter of the disposal function.

5.5. Technology Substitution

Technological advances and the emergence of a new dominant design (Utterback and Abernathy, 1975) signal the substitution of an old technology with a new one. Research on technology substitution has shown that substitution follows a classic


logistic curve (Fisher and Pry, 1971). The Fisher-Pry substitution model of technological change is incorporated in the model to estimate the inflow curve of the new technology product, and its rate of substitution of the old technology product. The Fisher-Pry substitution model is given by:

𝑠(𝑡𝑡) =12

[1 + tanh𝛼(𝑡𝑡 − 𝑡𝑡ℎ)] (15)

where s(t) is the fraction of the sales substituted by the new technology, 𝑡𝑡ℎ is the time at which the new technology has substituted 50% of the market share, and 𝛼 is interpreted as a parameter indicating half the annual fractional growth rate in the early years of the new technology product. As sales of the new technology devices increases, their stock or installed base also increases, reducing the share of the stock of old technology devices and change in stock of old technology devices becomes negative as new technology devices gain 50% of market share.

6. Experimental Frame The generic model presented above is applicable to all types of consumer durable goods which can be characterised as high value purchases used over extended periods of time. As there is a significant time and monetary commitment involved in purchasing such a product (Steffens, 2003), the adopting and disposing unit is assumed to be the household rather than an individual. However, theoretically, it is possible to apply the model for individuals as well. An important assumption the model makes is that the disposal function remains the same over the entire time period of the technology.

6.1. Application to the Case of TVs in Switzerland

The generic model described above is now applied to a specific case of TVs, in particular the Cathode Ray Tube TV (CRT TV), in Switzerland. The data available on this case provides the opportunity to empirically test and validate the model. Not only is the television a widely adopted consumer durable, it is also one that has seen both incremental and disruptive technological innovation over its societal lifetime from the time it was introduced in the early 1950s. The dominant design in TV technology since then had been the CRT TV, until the late nineties when the first Flat Panel Display TVs (FPD TV) were introduced in the market. The rapid technological


substitution meant that in a span of 10 years, CRT TVs were made completely obsolete in Switzerland. Thus, the specific case of CRT TVs in Switzerland from 1950 – 2010 provides a suitable case to estimate stocks and flows of a consumer durable through society from introduction to obsolescence.

6.2. TVs in Switzerland –Background

The TV market in Switzerland has seen a rapid change, as shown below, in Figure 7, with the sharp drop in sales of CRT TVs matched by the steep rise in sales of non-CRT TVs, reflecting a classic substitution. In the space of 7 years since its introduction in 1998, the sale of non-CRT TVs has overtaken sale of CRT TVs in Switzerland, with CRT TVs completely phased out of the market from 2009.

Figure 7: Technology Substitution - CRT TV to non-CRT TV in Switzerland (Data source: see Table 1)

7. Data Data for the model was obtained from numerous sources, as it was fragmented and incomplete. Confidence in figures presented here was developed using cross-checks, parallel sources of information and expert discussions. Some data can be referenced to literature, with others collected through primary research. The data on number of households and household ownership of TVs in Switzerland was obtained from the Swiss Federal Statistical Office (BfS). The Swiss Consumer Electronics Association (SCEA) provided sales data for TVs in Switzerland from 1995 onwards. For the period before 1995 on early TV ownership, data from the TV


licencing authority in Switzerland, Billag, was gathered. The table below provides a summary of the data gathered and their sources (See also annexes for data values). Available data (for Switzerland)

Measurement unit

Time period

Comment Source

Household Population

No. of households

1950 – 2009 (intermittent)

Data for every 10 years as per census; linear interpolation for continuous series.

Swiss Federal Statistical Office (BfS)

Household TV Ownership

% of TV households


Relatively short and intermittent time series; Survey in 2005 gathered data on households with multiple-unit TV ownership.

Swiss Federal Statistical Office (BfS)

TV Licenses No. of licenses issued


Patchy data on early TV ownership, i.e. not for all years. For recent years, no. of TV licenses issued does not fully reflect no. of households with TVs as not all households pay license fees due to exemptions; also does not capture multiple TV ownership.

Billag AG (is the government organisation that collects TV license fees applicable to all Swiss households by law, unless exempt.

TV Sales No. of TVs 1995 – 2009 (continuous)

Industry sales data with categorization of type (i.e. CRT / non-CRT) and size of TV (in ranges).

SCEA (the Association of Swiss Consumer Electronics Manufacturers)

Table 1: Primary data gathered and sources

8. Extensions to the Specific Model Two elements have to be added to adapt the model to the empirical case:


• Due to the long time horizon, change in household population is included in the model.

• Because data from the Swiss Federal Statistical Office confirms multi-unit ownership of TVs by many households, a sub-model to estimate the number of devices per household is included.

8.1. Estimating the Development of Household Population

Census data is used to estimate the development of the number of households in Switzerland between 1950 and 2010. As the census data is only available for 10 year intervals, the household population in the intervening years is estimated by linear interpolation.

8.2. Estimating Devices per Household - Multi-unit Ownership Sub-model

Ownership or adoption of consumer durables by households is often also known as the market penetration of the consumer durable. In markets with all households owning the durable product, diffusion of the product is complete, and market is considered saturated, with market penetration considered 100%. However, for many consumer durable products such as televisions to mobile phones and even automobiles, households may often own more than one device. Therefore, it is important to establish the evolution of number of devices per household over time. Surprisingly little attention has been given to multiple-unit ownership in diffusion modelling literature. Of the few models developed for multiple-unit adoptions, the majority, with the exception of Steffens (2003) who considered automobiles, have been for fast moving consumer goods with short life cycles rather than durable consumer goods with longer-term ownership patterns. Multiple-unit adoption is a long-term process, and existing diffusion models do not include the saturation effect for multiple-unit adoptions. To overcome this, a solution is presented based on the Bass model which reconceptualises multiple-unit adoptions as a two-step diffusion-based process. The logistic function displays an S-shaped behaviour and has been found to empirically describe the diffusion of a range of technologies ranging from mobile phone, home electric appliances to computers.


Following Walk (2009), it is assumed that the number of devices owned per household follows an S-shaped logistic curve, given by the equation:

𝐷𝑃𝐻(𝑡𝑡) = 𝑠 ×exp (𝛼 × [𝑡𝑡 − 𝑡𝑡ℎ])

1 + exp (𝛼 × [𝑡𝑡 − 𝑡𝑡ℎ]) (16)

where DPH (t) is the number of devices per household at time t, s is market potential, i.e. the number of devices per household at which market saturation occurs, 𝛼 is the shape parameter and 𝑡𝑡ℎ is the time at which half the households own the device, or in other words, 50% of market potential is reached. For first-unit adoptions, the market saturation is assumed to be 100%, that is to say that the first-unit diffusion of the device complete when all households own the product. This is given by the following equation:

𝐷𝑃𝐻1(𝑡𝑡) = 𝑠1 ×exp (𝛼 × [ 𝑡𝑡 − 𝑡𝑡ℎ1])

1 + exp (𝛼 × [ 𝑡𝑡 − 𝑡𝑡ℎ1]) (17)

where 𝐷𝑃𝐻1(𝑡𝑡) is the number of first-unit devices, 𝑠1 is the market potential of the first-unit diffusion and 𝑡𝑡ℎ1 is the time at which 50% of the market potential of the first phase has been achieved. The second phase of additional-unit adoptions are given by a similar equation, with the second phase starting once first-unit market potential has been achieved. Therefore,

𝐷𝑃𝐻2(𝑡𝑡) = 𝑠1 + �𝑠2 ×exp (𝛽 × [ 𝑡𝑡 − 𝑡𝑡ℎ2])

1 + exp ( 𝛽 × [ 𝑡𝑡 − 𝑡𝑡ℎ2])� (18)

where 𝐷𝑃𝐻2(𝑡𝑡) is the number of additional devices, 𝑠2 is the market potential for additional unit adoptions and 𝑡𝑡ℎ2 is the time at which half of the additional-unit adoption is attained. The overall diffusion of multiple-unit ownership, 𝐷𝑃𝐻𝑀(𝑡𝑡), is given by combining both phases. Therefore,


𝐷𝑃𝐻𝑀(𝑡𝑡) = 𝐷𝑃𝐻1(𝑡𝑡) ∙ 𝐷𝑃𝐻2(𝑡𝑡) (19)

8.2.1. Number of Devices per Swiss Household

Applying the multi-unit ownership model to historical penetration rates on TV ownership available from Billag and BfS data results in the devices per household for the period 1950 – 2010. The first-unit adoption curve is fitted to the data on TV licences issued by Billag, with the market potential 𝑠1 set to 1. The total market potential for the Swiss TV market is estimated to be 2.24 TVs per household, based on one TV per person as per official statistics on the average number of persons per

household. Therefore, for the multi-unit adoption curve, the total market potential 𝑠2 is set at the difference between the average no. of persons per household and 𝑠1. The parameters are fitted to the data by minimising the sum of squared residuals.

Figure 8: Multi-unit Devices per Household (DPH) A good fit of the diffusion curve to the available data for both the first unit adoption and the multi-unit adoption was found using the parameters given in Table 2. The data show that the first TV adoption was quite rapid, and reached 50% market saturation in just 18 years from introduction, reaching 100% market saturation by 1988. In comparison, multi-unit adoption diffuses more slowly, with 50% of households having multiple units only in 2021, nearly 40 years after the start of multiple unit adoptions. This is reflected in the values of shape parameters α, which has a value of 0.4, and β which has a value of any 0.12.


First-unit adoption Multi-unit adoption 𝒔𝟏 1 𝒔𝟐 1.24 𝜶 0.4 𝜷 0.12 𝒕𝒉𝟏 1968 𝒕𝒉𝟐 2021

Table 2: Parameters of multi-unit adoption - devices per household

9. Results

9.1. Swiss TVs – Installed Base

Combining data on number of households per year and the discretised total TV devices per household per year results in the total installed base of TVs in Swiss households from 1950 – 2010. Figure 9 below shows the development of Swiss societal stocks of TVs through time. Following the introduction of FPD TVs, the fall in the stocks of CRT TVs is estimated by subtracting non-CRT TV stock, estimated using the Fisher-Pry model substitution (15), from total stocks. The stock of CRT TVs reached a peak of nearly 4.2 million TVs in 2003, after which substitution by non-CRT TVs rapidly depletes the stock, with only a little over a third remaining in households in 2010.

Figure 9: Societal Stock – TVs (CRT and Non-CRT TVs) in Switzerland

9.2. Swiss TVs – Sales and Disposals

Equation (10) above is used to estimate sales and disposals of CRT TVs over its societal existence. The parameters of the disposal function are iteratively determined by fitting the model data to the sales data from SCEA for the period 1995 – 2009 by


minimising the sum of squared residuals. The scale parameter is interpreted as the average lifetime of CRT TVs and the shape parameter the deviation from the mean. As it is not possible to solve the convolution in equation (8) analytically, it is discretised and integrated numerically, as given by (9). Parameter Description Value

𝜸 Shape parameter 2.79 𝜶 Scale parameter 10.27

Table 3: Parameters of the disposal function for CRT TVs Figure 4 below shows the societal flows – both sales and disposals – of CRT TVs through Swiss society from 1950, the time of introduction of the product, to 2020 by when it is almost fully disposed.

Figure 10: Societal Flows - Sales and Disposals of CRT TVs in Switzerland The results show that nearly 15 million CRT TVs were sold between 1950 and 2008. According to the model, CRT TV sales peaked in the year 2001 at 487,000 units, and then rapidly declined to be completely substituted by non-CRT TVs in 2008. In 2011, as per model estimates, over a million CRT TVs still remained in Swiss households. However, rapid disposal would see the large majority of CRT TVs disposed of from Swiss households by 2016, with only a fraction, just a little over 200,000 TVs, remaining to be disposed of. The model shows that the peak disposal of CRT TVs took place in 2010, with nearly 600,000 CRT TVs estimated to have been disposed in the year. The model results show that disposals of old technology devices


shoot up sharply following the take-off and mass adoption of new technology devices. This is also reflected in the data of actual disposals, which also show a large increase from 2009 to 2010.

10. Model Validation Validity, or the model’s property of adequately reflecting the system modelled, is the primary measure of model quality (Schwaninger and Groesser, 2009). From a model validation perspective, Switzerland provides an exceptional source of data on the disposal of CRT TVs through the organised collection and recycling operated by SWICO Recycling, allowing validation of the model against real system data which is entirely separate from the data used for parameter estimation (as described in the “Data” section above). SWICO Recycling started collecting CRT TVs only in 2002, and has over time built consumer awareness regarding the possibility of disposing their CRT TVs through its take-back channels. As consumer awareness has grown, the collection efficiency of the system has also increased, with few CRT TVs leaking out of the system or disposed of in other waste streams. The take-back and recycling system currently has a collection efficiency of 90%, as confirmed by SWICO Recycling experts. Data specifically on number of CRT TVs disposed is available from 2006 from SWICO Recycling annual reports. Although this dataset is only recent, with only a short time series available, it provides a basis to assess the predictive validity of the model, i.e. the extent to which the model actually predicts the outcome that it is intended to model.

Figure 11: Disposal and Collection of CRT TVs


It is seen that both model estimates and the data on actual collection of CRT TVs show the same underlying trend with the correlation coefficient between the two a high 0.89. However, in the period 2006 - 2009, there is a significant difference between the actual collection by SWICO Recycling and the model estimates. This may be attributed to two reasons:

• The assumption of collection efficiency of 90% in the early years of the take-back and collection system. It is plausible that the collection efficiency of the system has grown over time – it was likely not as high in the initial years as it is currently. In such a case, the actual disposals including both those collected and not collected by the system, would have been closer to the model estimated disposals.

• The balance discrepancy may be explained by the stored unused or “hibernating” CRT TVs in households which have not as yet reached the waste stream.

11. Discussion and Conclusion The model offers insights into the societal stocks and flows, from its introduction to its obsolescence, including peaks of sales, disposals and installed base of a consumer durable, especially with only limited data availability. Such a model can provide collection and take-back systems as well as recycling companies with better forecasts to help plan their capacities. For policy makers, it provides a gauge of the collection efficiency of a formal take-back and collection system, and a basis to check against potentially environmentally harmful or materially significant leakages. In addition to providing estimates of future waste flows, it also provides an insight into historic disposals which have taken place before the advent of formal and organised collection and take-back systems. Such data can indicate the existence of anthropogenic stores of disposed consumer durables which could pose a risk due to hazardous substances, but are also increasingly being considered as urban mines containing precious and rare materials. The overestimation of modelled sales as compared to actual sales soon after the introduction of new technology products indicates that it is likely that consumers held back new TV purchases immediately following the introduction of the new


technology. This latent demand for new TVs is possibly also contributing to the sharp rise in the disposals of old technology CRT TVs soon after. This insight could also be useful to predict similar phenomena occurring with other high-value consumer durables which are subject to radical technology changes. The multi-unit adoption sub-model developed to estimate the devices per household is an addition to the extensive literature on diffusion of consumer durable products, and provides a useful model especially as multi-unit ownership of consumer durables is increasing. The current model is not without limitations which can, however provide directions for future research. Firstly, the model provides only statistical estimates of stocks and flows; it incorporates no explicit representation of the underlying consumer disposal behaviour and the socio-economic factors which may play a role in the timing of disposal of consumer durables. Secondly, the assumption that consumer durables are either in active stock or disposed does not account for time spent in private storage. Consumer durables may be stored in attics or basements or garages for months or even years before finally being disposed of. The model presented above is so far unable to provide insight into such behaviour. Thirdly, the time-invariant disposal function has much scope for improvement. Given the fairly long societal existence of consumer durables compared to the rapidly changing consumer preferences and perceptions of obsolescence, it is likely that the residence time of a consumer durable changes between its introduction and decline. Anecdotal evidence for PCs and mobile phones has shown that average lifetime of these products has reduced over time, with more frequent replacement taking place. Fourthly, the model assumes there are only two competing technologies which render the same service – the existing dominant technology and the challenger. However, it is possible to have more than one technology challenging the dominant technology. Currently, the model is unable to incorporate such multiple technology substitution.


As the model is applied only to a single case study of a consumer durable product in a specific geography, further empirical work is required to substantiate the generalizability of the model.


References Bass, Frank M. 1969. “A New Product Growth for Model Consumer Durables.” Management Science 15 (5, Theory Series) (January): 215-227. Bayus, Barry L. 1991. “The Consumer Durable Replacement Buyer.” Journal of Marketing 55 (1) (January): 42-51. Bergsdal, Håvard, Helge Brattebø, Rolf A Bohne, and Daniel B Müller. 2007. “Dynamic material flow analysis for Norway’s dwelling stock.” Building Research and Information 35 (September): 557-570. doi:10.1080/09613210701287588. Elshkaki, Ayman, Ester van der Voet, Veerle Timmermans, and Mirja Van Holderbeke. 2005. “Dynamic stock modelling: A method for the identification and estimation of future waste streams and emissions based on past production and product stock characteristics*1.” Energy 30 (8) (June): 1353-1363. doi:10.1016/j.energy.2004.02.019. Fisher, J. C., and R. H. Pry. 1971. “A Simple Substitution Model for Technology Change.” Technological Forecasting and Social Change 3 (1): 75–88. Gordon, R., R. Lifset, M. Bertram, B. Reck, T. Graedel, and S. Spatari. 2004. “Where is all the zinc going: The stocks and flows project, Part 2.” JOM Journal of the Minerals, Metals and Materials Society 56 (1) (January 1): 24-29. doi:10.1007/s11837-004-0266-4. Gregory, Jeremy R., Marie-Claude Nadeau, and Randolph E. Kirchain. 2009. “Evaluating the Economic Viability of a Material Recovery System: The Case of Cathode Ray Tube Glass.” Environmental Science & Technology 43 (24) (December): 9245-9251. doi:10.1021/es901341n. Hilty, LM, A. Köhler, F. von Schéele, R. Zah, and T. Ruddy. 2006. ‘Rebound Effects of Progress in Information Technology. Poiesis & Praxis’. International Journal of Technology Assessment and Ethics of Science 1 (4).


Hu, Mingming, Ester Van Der Voet, and Gjalt Huppes. 2010. “Dynamic Material Flow Analysis for Strategic Construction and Demolition Waste Management in Beijing.” Journal of Industrial Ecology 14 (3) (May): 440-456. doi:10.1111/j.1530-9290.2010.00245.x. Kamakura, W. A, and S. K Balasubramanian. 1987. “Long-term forecasting with innovation diffusion models: The impact of replacement purchases.” Journal of Forecasting 6 (1): 1–19. Kang, H. Y, and J. M Schoenung. 2006. “Economic analysis of electronic waste recycling: Modeling the cost and revenue of a materials recovery facility in California.” Environmental Science & Technology 40 (5): 1672-1680. Kapur, Amit, Gregory Keoleian, Alissa Kendall, and Stephen E. Kesler. 2008. “Dynamic Modeling of In-Use Cement Stocks in the United States.” Journal of Industrial Ecology 12 (4) (August): 539-556. doi:10.1111/j.1530-9290.2008.00055.x. Kleijn, R., R. Huele, and E. van der Voet. 2000. “Dynamic substance flow analysis: the delaying mechanism of stocks, with the case of PVC in Sweden.” Ecological Economics 32 (2): 241–254. Lifset, Gordon, Graedel, Spatari, and Bertram. 2002. “Where has all the copper gone: The stocks and flows project, part 1.” JOM Journal of the Minerals, Metals and Materials Society 54 (10) (October 15): 21-26. doi:10.1007/BF02709216. Linton, J. D., J. S. Yeomans, and R. Yoogalingam. 2002. “Supply planning for industrial ecology and remanufacturing under uncertainty: a numerical study of leaded-waste recovery from television disposal.” Journal of the Operational Research Society 53 (11): 1185-1196. Mahajan, Vijay, Eitan Muller, and Frank M. Bass. 1990. “New Product Diffusion Models in Marketing: A Review and Directions for Research.” The Journal of Marketing 54 (1) (January): 1-26.


Mahajan, V., and E. Muller. 1996. “Timing, diffusion, and substitution of successive generations of technological innovations: The IBM mainframe case.” Technological Forecasting and Social Change 51 (2): 109–132. Meade, Nigel, and Towhidul Islam. 2006. “Modelling and forecasting the diffusion of innovation - A 25-year review.” International Journal of Forecasting 22 (3): 519-545. Müller, Daniel B. 2006. “Stock dynamics for forecasting material flows—Case study for housing in The Netherlands.” Ecological Economics 59 (1) (August): 142-156. doi:10.1016/j.ecolecon.2005.09.025. Norton, John A., and Frank M. Bass. 1987. “A Diffusion Theory Model of Adoption and Substitution for Successive Generations of High-Technology Products.” Management Science 33 (9) (September): 1069-1086. Oguchi, M., T. Kameya, S. Yagi, and K. Urano. 2008. “Product flow analysis of various consumer durables in Japan.” Resources, Conservation & Recycling 52 (3): 463–480. Peres, Renana, Eitan Muller, and Vijay Mahajan. 2010. “Innovation diffusion and new product growth models: A critical review and research directions.” International Journal of Research in Marketing 27 (2) (June): 91-106. doi:10.1016/j.ijresmar.2009.12.012. Schwaninger, M. 2010. “Model-based management (MBM): a vital prerequisite for organizational viability.” Kybernetes 39 (9/10): 1419–1428. Schwaninger, M., and S.N. Grösser. 2009. “System dynamics modeling: validation for quality assurance.” Encyclopedia of Complexity and System Science. Springer, Berlin. Spatari, S., M. Bertram, R. B Gordon, K. Henderson, and T. E Graedel. 2005. “Twentieth century copper stocks and flows in North America: A dynamic analysis.” Ecological Economics 54 (1): 37-51. Steffens, P. R. 2003. “A model of multiple-unit ownership as a diffusion process.” Technological Forecasting and Social Change 70 (9): 901–917.


Sterman, J. D. 2000. Business dynamics: Systems thinking and modeling for a complex world with CD-ROM. Irwin/McGraw-Hill. Streicher-Porte, M., R. Widmer, A. Jain, H. P. Bader, R. Scheidegger, and S. Kytzia. 2005. “Key drivers of the e-waste recycling system: Assessing and modelling e-waste processing in the informal sector in Delhi.” Environmental Impact Assessment Review 25 (5): 472-491. Utterback, James M, and William J Abernathy. 1975. “A dynamic model of process and product innovation.” Omega 3 (6) (December): 639-656. doi:10.1016/0305-0483(75)90068-7. Walk, W. 2009. “Forecasting quantities of disused household CRT appliances–A regional case study approach and its application to Baden-W\ürttemberg.” Waste Management 29 (2): 945–951. Widmer, R., H. Oswald-Krapf, D. Sinha-Khetriwal, M. Schnellmann, and H. Böni. 2005. “Global perspectives on e-waste.” Environmental Impact Assessment Review 25 (5): 436-458. Yu, J., E. Williams, M. Ju, and Y. Yang. 2010. “Forecasting global generation of obsolete personal computers.” Environmental science & technology 44 (9): 3232–3237.


CHAPTER III

Reverse Diffusion: Estimating Disposal of Consumer Durables through Application of Diffusion Modelling

Deepali Sinha, University of St.Gallen Abstract The diffusion of consumer durables has been extensively modelled, in particular by the Bass Diffusion Model which has been empirically validated across a range of products. This paper proposes that the dynamics of disposal of consumer durables are not dissimilar to the adoption of new products. Building on the extant literature on demand forecasting of new products in the diffusion modelling tradition, a model for forecasting the disposal, or “reverse diffusion”, of consumer durables is developed and validated using three case studies. The results show that the model provides a simple, yet effective method of estimation and forecasting of waste flows from end-of-life consumer durables.

Chapter III – Reverse Diffusion: 52 Estimating Disposal of Consumer Durables through Application of Diffusion Modelling

1. Introduction Consumer durables pervade modern lifestyles and their usage is growing rapidly around the world. Innovation and intense competition in the consumer durables industry has brought new products with new and improved functions and at increasingly affordable prices. In the past decades, electrical and electronic consumer durables have multiplied and become more accessible and numerable, becoming standard items for the vast majority of households. On the flip side, newer products and improved functionalities are resulting in greater obsolescence of consumer durables, with large quantities of these products being replaced and disposed. Given the rapid growth and affordability of consumer durables, the large majority of consumer durables are scrapped before they are technically broken (Bayus, 1988). Increasingly, the obsolescence of consumer durable products is often discretionary in nature rather than technical. The pervasiveness of gadgets in modern lifestyles and their rapid obsolescence makes end-of-life consumer durables ranging from televisions to computers to washing machines, commonly known as Waste Electrical and Electronic Equipment (WEEE), one of the fastest growing waste streams not only in developed countries but around the world. This waste stream has been recognised for its hazard potential and the need to ensure it is disposed of properly. Growing concern regarding the importance of managing this waste has led many countries to implement legislation designed to reduce the volume of this waste, remove its hazardous components, encourage recycling and minimize the environmental and health risks of unsound waste disposal. The WEEE Directive of the European Union, which came into force in August 2004, is the most prominent legislation focussing on the end-of-life management of disposed consumer durables. The Directive, which places the responsibility of the end-of-life management of WEEE on producers of consumer durable products, has led to the development of recycling and take-back systems in European Union Member States since it came into force. Similar legislation in many other countries is also in place or upcoming, and with it more formalised systems to take-back end-of-life consumer durables. Forecasts of waste flows are as essential to waste planners as estimated potential sales and the timing of sales are to marketers. While sales forecasting models of consumer Chapter III – Reverse Diffusion: 53 Estimating Disposal of Consumer Durables through Application of Diffusion Modelling

durables are dominated by diffusion models, waste forecasting models are more commonly based on delay modelling, with the emphasis on Material Flow Analysis (MFA). Consumer behaviour, for both purchase and disposal of durables, is a result of interacting social and market driven factors. However, there is little interdisciplinary research between the two, despite the fact that both domains share a significant overlap regarding consumer behaviour. This research builds on the extant literature on demand forecasting of new products based on well documented diffusion models which form the conceptual basis of this research. It proposes that the dynamics of disposal of consumer durables are not dissimilar to the diffusion of new products. A “reverse diffusion” model to forecast the disposal of consumer durables is developed and validated using three case studies, thereby extending the application of diffusion models from the classical questions of forecasting the timing and rate of adoption of consumer durables to the forecasting of timing and rate of disposal of these products. The paper is organised as follows: In the next section, a review of relevant literature is presented, followed by the purpose and objectives of the research. Section 4 describes the conceptual and analytical bases of the model. In the following sections, the model is empirically examined, applied to three case studies of disposal of consumer durable products in Switzerland, and validated against real system data. The final section concludes the paper with a summary of the contribution, limitations and suggests areas for future research.

2. Literature Review For this paper, two streams of literature are reviewed – firstly research from the marketing domain on consumer behaviour, including models for forecasting adoption of durable products and secondly research from the waste management domain regarding models of forecasting waste flows of consumer durables.

2.1. Forecasting Adoption of Consumer Durables

The adoption of consumer durables is extensively researched in diffusion models, popularised in the marketing literature with the seminal article by Bass (1969). The Bass Diffusion Model (BDM) is the most widely known model of diffusion of


consumer durables, with the diffusion framework developed by Bass providing the basis for the large body of diffusion research that has followed. Diffusion research seeks to understand the spread of innovations by modelling their entire life cycle from the perspective of communication and consumer interactions. Diffusion models have been particularly useful in providing frameworks for understanding the processes by which new products come into circulation and spread across populations of adopters. In his paper, Bass presented a growth model for the timing of initial purchases of new products, suggesting that new technologies are not adopted immediately by all the potential buyers, but rather a diffusion process is set in motion in which there are largely two groups of adopters – the innovators and the imitators. The innovators are uninfluenced by the other members of the social system in their adoption behaviour. The imitators, on the other hand, are those who are influenced in the timing of the adoption by the decisions of other members of the social system. This model assumes that the trajectory of cumulative adoptions of a new product follows a function whose growth rate depends on two parameters. One parameter captures a consumer’s intrinsic tendency to purchase and is independent of the number of previous adopters, and is called the coefficient of innovation. The other parameter captures the influence of previous adopters, being called the coefficient of imitation. It is analogous to the spread of an epidemic, which is spread quickly by contact between the infected and non-infected, and once the large majority of the population has been affected, the infection growth slows down. Essentially, the BDM attempts to predict how many customers will eventually adopt the new product, but most importantly, when they will adopt it. The great appeal of the BDM is that it is parsimonious, has been shown to fit data, and provides parameters that have an intuitive behavioural interpretation. Many variations and extensions to the BDM have since been proposed, extending the original model to include marketing mix variables such as effect of price (Jain and Rao, 1990; Kamakura and Balasubramanian, 1988), placement (Jones and Ritz, 1991), and advertising (Horsky and Simon, 1983; Simon and Sebastian, 1987). Others have proposed models to include multi-generation products (Norton and Bass, 1987), replacement sales (Bayus, 1991), the influence of technology substitution (Fisher and Pry, 1971) and multiple product ownership (Steffens, 2003).


Diffusion models are employed in two basic ways: to develop a better general understanding of diffusion phenomena (descriptive), and to predict diffusion paths for new technologies (predictive) before there is significant amount of data available (Peres et al., 2010). Mahajan et al. (1990), Mahajan et al.(1995) and Lilien, Rangaswamy and Van den Bulte (2000) describe some generic uses of diffusion modelling in marketing which include pre-launch forecasting, business valuation and strategic decision analysis based on the product life cycle, and the determination of optimal prices, etc. A comprehensive review of the literature on diffusion of new products can be found in Mahajan et al. (1990), Meade and Islam (2006), and most recently Peres et al. (2010).

2.2. Consumer Disposition Behaviour

Research on disposal of durable goods started in the late 70s as an offshoot of consumer behaviour research, following a broader research thrust on consumer behaviour. Jacoby et al. (1977) identified three stages of consumption of consumer durables – namely acquisition, consumption and disposition or disposal. Hanson (1980) further suggests that the evaluation process for disposition involves concepts similar to those in the acquisition evaluation process. More recent studies of consumer acquisition behaviour have looked at replacement of consumer durables, given that the large majority of sales of many durables, especially in industrialised economies, are replacement purchases. However, the focus of this research is more inclined towards consumer behaviour regarding purchase of new products to replace existing ones, rather than the disposition of existing products. Though both replacement and disposition are closely related, there is a distinction between replacement and disposition, especially with regards to the timing, and there are only a few studies dealing specifically with the topic of consumer disposition behaviour. Antonides (1991) notes that the lifetime of a durable good is determined by a consumer’s decision which is in turn determined by economic, psychological and product-technical factors. According to Hanson (1980), the decision to dispose of consumer durables is aroused by some triggering cues such as product damage or obsolescence which could be in terms of product function, psychological or style obsolescence. Disposition or disposal behaviour thus is a function of disposal intention, social factors and situational factors (Hanson, 1980).


Consumer motivation for replacement of consumer durables is in many ways similar to disposal. Consumers replace durable products for a variety of reasons including product failure, likely due to wear and tear or defects due to improper use or breakage (Cooper, 2004; Islam and Meade, 2000), change in consumer needs perhaps due to socio-economic reasons such as higher income (Pickering, 1981; Cooper, 2004; Bayus and Gupta, 1992), dissatisfaction with product functionality or style preferences, largely arising as a result of the availability of new technology (Cooper, 2004; Islam and Meade, 2000; DeBell and Dardis, 1979; Hoffer and Reilly 1984; Sherman and Hoffer 1971). Some authors (Bayus, 1991; Kamakura and Balasubramanian, 1987) include replacement sales in their forecasting models, however, there are surprisingly few sales forecasting models explicitly incorporating replacement sales.

2.3. Waste Forecasting Models

Models to estimate and forecast consumer durable disposals are part of a growing literature on waste forecasting. The most commonly used model for forecasting post-consumer end-of-life product flows is based on combining sales data with a fixed average lifetime or residence time to forecast waste flows of end-of-life consumer durables (Widmer et al., 2005; Kang and Schoenung, 2006). More recently, several researchers have improved upon this traditional model by combining product sales with a lifetime distribution such as Weibull (Oguchi et al., 2008) or a derived lifetime (Gregory et al., 2009, Yu et al., 2010). Material and substance flow analysis (MFA and SFA) have also been commonly used to estimate societal stocks and flows of materials often used in durable products such as lead in TVs (Elshkaki et al., 2005) and copper in electrical and electronic equipment (Lifset et al., 2002; Spatari et al., 2005). Hilty et al. (2006a) have used a macro-level System Dynamics model of ICT (Information and Communications Technology) use to predict e-waste flows and other ICT impacts. Dynamic material flow analysis models have been used to forecast waste material flows, especially for construction and demolition waste (Bergsdal et al., 2007; Hu et al., 2010) as well as electronic waste (Streicher-Porte, 2005). Common to existing models of waste forecasting is that they are based on sales and average lifetime data, rather than disposal data. The proposed model applies concepts


from diffusion models of consumer durable acquisition to estimate and forecast disposition, based instead on disposal data.

3. Goal and Purpose From the above discussion of the existing models of forecasting adoption and disposal of consumer durables, it is clear that there are distinctions between sales forecasting models, which focus on consumer behaviour, and waste forecasting models, which are focussed on material flows. The model put forth in this paper derives from prior work in the areas of diffusion modelling, building upon knowledge gained in those studies to inform and improve waste forecasting models. Hence, the goal of this paper is to develop a model to estimate the disposals of a consumer durable product, applying the diffusion modelling framework. The purpose of the model is to forecast disposals of a consumer durable product, providing estimates to waste managers, policy makers and recyclers on expected waste volumes. The contribution of the paper is twofold: Firstly, it makes a unique contribution by extending the applicability of the diffusion modelling framework, bringing the insights from the sales forecasting domain to the waste forecasting domain in a relatively simple and parsimonious form. Secondly, it provides an alternative model for estimation of the disposal of consumer durables without requiring estimates of average product lifetime and time series sales data.

4. Conceptual Model and Mathematical Framework

4.1. Bass Diffusion Model

Diffusion models have been particularly useful in providing frameworks for understanding the processes by which new products come into circulation and spread across populations of adopters. The literature indicates that the predominant application of diffusion models has been for purposes of forecasting the trajectory of new product adoption – for newly introduced products as well as for products to be introduced that are similar in some way to existing products whose diffusion history is known (Lilien, Rangaswamy and Van der Bulte, 1999).


The Bass Diffusion Model (BDM) has a behavioural rationale that is consistent with studies in social science literature on the adoption and diffusion of innovations (Norton and Bass, 1987). Over a large number of new products and technological innovations, the BDM describes the empirical adoption curve quite well. Therefore, because of its simplicity and generality, the BDM forms the conceptual basis for this research. In the BDM, the cumulative adoption of a consumer durable follows sigmoid function, as sales of a product are driven initially by innovative demand followed by imitative demand until the market potential, or saturation is reached. This behaviour can be described by the nonlinear Bass differential equation (a Riccati equation): �̇� = 𝑝 + (𝑞 − 𝑝)𝑆−𝑞𝑆2

(20)

Where �̇� is the change in stock, p is the coefficient of innovation and q is the coefficient of imitation. which can be rewritten as: �̇� = 𝑝 ∙ (1 − 𝑆) + 𝑞𝑆 ∙ (1 − 𝑆)

(21)

which for: q=0 becomes �̇� = Innovators = p ∙ (1 − S) which results in an exponential decay function for p=0 becomes �̇� = 𝐼𝐼𝑚𝑖𝑡𝑡𝑎𝑡𝑡𝑜𝑟𝑠 = 𝑞𝑆 ∙ (1 − 𝑆) which results in a logistic function The dynamics of the stock 𝑆 is thus solely dependent on the stock and is a superposition of an exponential decay and a logistic function. The equation is solved by the following cumulative distribution function S(t) and the respective probability density function s(t):

𝑆(𝑡𝑡) =1 − 𝑒−(𝑝+𝑞)𝑡

1 + 𝑞𝑝 𝑒

−(𝑝+𝑞)𝑡 (22)


𝑠(𝑡𝑡) = (𝑝 + 𝑞)2

𝑝

𝑒−(𝑝+𝑞)𝑡

�1 + 𝑞𝑝 𝑒−(𝑝+𝑞)𝑡�

2 (23)

where parameters p and q are interpreted as the ‘coefficient of innovation’ and the ‘coefficient of imitation’ respectively. The coefficient p captures the influence on potential adopters’ decisions that is independent of the existing number of adopters, i.e., the influence that is not obtained through interpersonal (word-of-mouth) communication with existing adopters. The coefficient q expresses the influence of existing number of adopters on purchase decisions of other people yet to adopt the new product. Sales are given by the equation: 𝑠𝑎𝑙𝑒𝑠(𝑡𝑡) = 𝑚 ∙ 𝑓(𝑡𝑡) (24) where 𝑚 is the total market potential. Total market potential in the BDM is specified as total number of adoptions of a product, thus equation (24) gives the adoption rate, or in other words, sales of a product.

4.2. Reverse Diffusion

With the BDM as the conceptual basis, the model is applied to forecast the disposal, rather than adoption, of a consumer durable – a ‘reverse diffusion’. We use the term ‘reverse diffusion’ to refer to the opposite of diffusion: while diffusion is about how products enter the market, reverse diffusion is about how they exit the market. Thus, reverse diffusion is defined as the process of de-adoption and outflow of a product from society which may be driven by technical and social influences. The outflow feeds a disposal stock which determines, together with the parameters p and q the dynamics of the outflow. The total stock in the market which is to be depleted to the disposal stock scales the outflow.


Figure 12: Stock and flow diagram for the reverse diffusion model In many ways, reverse diffusion is similar to innovation diffusion, in that it follows an S-curve, the shape and the trajectory of which are influenced by consumer behaviour, until market depletion is reached, with the decay in the number of potential consumers. For the reverse diffusion process, two types of ‘disposers’ are suggested - namely those who dispose of products for technical reasons, and those who choose to dispose of products for discretionary reasons. Technical disposals take place due to product not functioning, or reaching their end-of-technical life. Such disposals are analogous to the innovator adoptions in the BDM. On the other hand, consumer durables disposed of for discretionary reasons are much like the imitator adoptions in the BDM. These are influenced due to socio-economic factors, such as perception of obsolescence of existing products, network or complementarity effects, aesthetics, social status etc., much like word-of-mouth effects in product diffusion. Thus, the discretionary disposers, like the imitative adopters create a reinforcing loop – as more households dispose of a product, the more obsolete it is perceived, and the more households want to dispose of it. In the reverse diffusion model, we consider the total quantity sold over the product lifetime as the market potential. In other words, the reverse diffusion is complete when the cumulative sum of products sold over time (adoption) is equal to the cumulative disposal of products (deadoption). If product mass is known, reverse diffusion is complete with the cumulative disposal of the total material inflow.

Market (t) Disposal (t) In (t) Out (t)

p q


4.3. Model Parameters

Similar to innovation diffusion, the reverse diffusion distribution function can be characterised by the parameters p and q, where p is considered the coefficient of technical disposal, influenced by factors such as breakdown and wear and tear; and q is considered the coefficient of discretionary disposal, influenced by social and functional signals such as status, new product technologies, etc. Operating under assumptions as described below, the reverse diffusion equation results in:

𝑅(𝑡𝑡) =1 − 𝑒−(𝑝+𝑞)𝑡

1 + 𝑞𝑝 𝑒−(𝑝+𝑞)𝑡

(25)

𝑟(𝑡𝑡) = (𝑝 + 𝑞)2

𝑝


�1 + 𝑞𝑝 𝑒−(𝑝+𝑞)𝑡�

2 (26)

where 𝑅(𝑡𝑡) is the cumulative disposal function, 𝑟(𝑡𝑡) is the disposal density function, i.e. the specific disposal rate. The absolute disposal rate, d(t) is given by: 𝑑𝑑(𝑡𝑡) = 𝑚 ∙ 𝑟(𝑡𝑡) (27) where 𝑚 is the total number of the durable products in the market stock which is to be depleted to the disposal stock.

4.4. Assumptions

The following assumptions characterise the reverse diffusion model: 1. The reverse diffusion model is a model of disposal, accounting for products

that have entered into the waste stream. Similar to one of the limiting assumptions of the Bass model, it is assumed that consumers either use or dispose of a consumer durable, without accounting for an intermediate stage of storage.


2. The model is time invariant, in that parameters p and q do not change over time. This in turn implies that consumer disposal behaviour remains the same over time. (However, as consumer durable products tend to be used over longer periods of time, it is possible that consumer behaviour changes over time).

3. The average mass and composition of a product is also considered constant.

While this is done in the interest of model parsimony, it may provide inaccurate estimates of the mass flow which experience shows is not the same over time. For example, there was a considerable reduction in the average physical mass of a mobile phone from over 350 g in 1990 to about 80 g in 2005 (Hilty, et al., 2006b). By assuming an average mass, the model could underestimate the mass flow in the earlier years and overestimate it in the latter.

4. The model assumes there are no other factors that can influence disposal of

consumer durable products such as the awareness and convenience of disposal options. While the discretionary disposal coefficient captures the overall influence of all non-technical factors leading to disposal, it is unable to distinguish between different drivers of discretionary disposal.

5. Application: Case Study – Consumer Durables in Switzerland

Three empirical case studies applying the reverse diffusion model to disposed consumer durables in Switzerland are presented. Switzerland has an established collection system for consumer durables since 1994, giving time series data over 15 years. This not only provides us with disposal data to estimate parameters, but also importantly, from a model validation perspective, it enables the comparison of the model forecasts with real system data. In this paper, the reverse diffusion model is used to estimate the disposal paths of three consumer durable products namely Cathode Ray Tube (CRT) monitors, CRT TVs and Flat Panel Display (FPD) Monitors. In the FPD monitor category, of the two technologies, namely Liquid Crystal Display (LCD) and plasma display, plasma displays form a miniscule share of the market and are therefore not considered.


5.1. Data

Diffusion models are normally estimated with data from ownership surveys, or early sales data. For the reverse diffusion model, data on disposals was obtained from the SWICO Recycling annual reports and EMPA, who do the monitoring and control of material flows of the take-back system. Sales data on the three products were also collected. The table below shows the data collected, the time period over which data was available and the relevant sources. (See also the Annexes). Consumer Durable Product

Data Type Time Series Source

CRT Monitors Sales of CRT Monitors (in units)

1983 – 2005 Robert Weiss Consulting

CRT Monitors Disposal of CRT Glass (in tonnes)

1994 – 2010 SWICO Recycling Annual Reports; EMPA

CRT TVs Sales of CRT TVs (in units)

1998 – 2007 SCEA (the Association of Swiss Consumer Electronics Manufacturers)

CRT TVs Disposal of CRT Glass (in tonnes)


LCD Monitor Sales of LCD Monitors (in units)

1999 – 2009 Robert Weiss Consulting

LCD Monitors Disposal of LCD Monitors (in units)


Table 4: Data collected Data on units of CRT Monitors and CRT TVs collected by SWICO Recycling is available only from 2006 onwards, however, data on CRT glass collected by the system is available for the entire period, from 1994 – 2010. Where necessary, data is converted from tonnes of CRT glass to number of units, based on a fixed average mass of leaded CRT glass per TV or monitor.


5.2. Case Study 1: CRT Monitors in Switzerland

CRT monitors were sold in Switzerland between 1983 and 2005, reaching peak sales in 1998 before being made obsolete by LCD monitors. While initially limited to business users, as the personal computer became more affordable, CRT monitors diffused into households as well, becoming a consumer durable just as TVs and radios. With the advent of LCD monitors, however, their adoption rate declined, with no further CRT monitor sales after 2005. With data on both the annual sales and disposals of CRT monitors in Switzerland for over 15 years, the CRT monitor presents an excellent case study to validate the conceptual and predictive validity of the reverse diffusion model. Switzerland provides an exceptional source of data on the disposal of CRT monitors through the organised collection and recycling operated by SWICO Recycling, allowing validation of the model against real system data. SWICO Recycling started collecting Personal Computer (PC) monitors in 1994, and has over time built consumer awareness regarding the disposal of PC monitors through its take-back channels. As consumer awareness has grown, the collection efficiency of the system has also increased, with few monitors leaking out of the system or disposed of in other waste streams. The take-back and recycling system currently has a collection efficiency of 90%, as confirmed by SWICO Recycling experts. Although it has taken time for the SWICO Recycling system to achieve such high collection efficiency, with clearly lower collection efficiency in the early days of the system, for simplicity, 90% collection efficiency is assumed throughout the time period. From Figure 13Figure 13: Cumulative Disposal R(t)- CRT Monitor Glass below, it can be seen that the disposal of CRT monitors follows an S-shaped curve, terminating when maximum potential CRT monitor inflow (m) has been depleted. The fit statistics are given in Table 5.

p 0.0001 q 0.3400 m 61,735 [tonnes CRT Glass] R2 0.98

Table 5: Parameter values for CRT Monitor reverse disposal


Figure 13: Cumulative Disposal R(t)- CRT Monitor Glass Figure 14 below shows the timing of the peak disposals, which occurred in 2005, by when half the stock of CRT monitors were disposed of.

Figure 14: Disposal curve r(t) – CRT Monitor Glass The parameter values for p and q, the maximum potential ownership, m, and the coefficient of determination, R2, are given in the table below: Interpreting the parameters, given the low value of p and much higher value of q, it can be inferred that technical failures trigger only a small number of disposals, with the large majority of disposals driven by discretionary factors.

0

1000

2000

3000

4000

5000

6000

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

CR

T M

onito

r G

lass

[t/a

]

Actual CRT Monitor Glass Disposal [tonnes]

Model Estimated CRT Monitor Glass Disposal[tonnes]


5.3. Case Study 2: CRT TVs in Switzerland

CRT TVs have been sold in Switzerland since the 1950’s, and until 1998 when LCD TVs were introduced, all TVs sold were CRT TVs. Although the take-back of CRT monitors started in 1994, TVs were included in the SWICO Recycling system only in 2002. With SWICO Recycling collection data available from 2002 onwards, sales data for the period 1995-2009 are considered, on the basis of the assumption that TVs have a minimum lifetime of seven years. Cumulative sales of CRT TVs in Switzerland in the period are considered as m. Figure 15 below shows the cumulative disposal of CRT TV glass from CRT TVs sold since 1995. As per the model, it is estimated that 97% of CRT TVs will be disposed of by 2020. The parameter values and fit statistics are given in Table 6Table 7.

Figure 15: Cumulative Disposal R(t) - CRT TVs Figure 16 below shows that the peak disposal of CRT TVs is expected to take place in 2012.


Figure 16: Disposal Curve r(t) - CRT TVs The parameter values for p and q, the maximum potential ownership, m, and the coefficient of determination, R2, are given in the table below:

p 0.0004 q 0.3933 m 95,337 [tonnes CRT Glass]

R2 0.98 Table 6: Parameter values for CRT TVs reverse disposal Similar to parameters for CRT monitors, the low parameter value of p as compared to q suggests that the driver of disposal for CRT TVs is greater due to discretionary rather than technical factors.

5.4. Case Study 3: LCD Monitors in Switzerland

LCD monitors were introduced into the Swiss market in 1999, and within 6 years had made CRT monitors completely obsolete. LCD monitors were already being seen in the waste stream within 5-6 years, and SWICO data on LCD monitors disposed is available from 2006. Technology trends indicate that LCD monitors may themselves be made probably obsolete by newer OLED monitors in the coming years. However, currently, LCD monitors continue to be sold. This poses a challenge in terms of estimating the maximum cumulative sales, which is an important factor affecting the model’s predictive value (Yu et al., 2010).


This challenge is addressed through a bounding analysis, similar to Yu et al., (2010), who consider three scenarios for their model, namely upper, baseline, as well as lower value of maximum carrying capacity for diffusion of PCs. Similarly, for this analysis, three scenarios of the maximum potential inflow of LCD monitors in Switzerland are considered. Cumulative sales of monitors in Switzerland from 1999 to 2009 are taken as the basis for the three scenarios. From the example of CRT monitors for which sales declined once a new technology was introduced, a similar substitution is expected to take place for LCD monitors as they are replaced by newer technologies. Therefore, the continued sales of LCD monitors are to an extent contingent on the speed of diffusion of next generation monitors. In the first scenario, in the event of a rapid substitution of LCD monitors, it is assumed that 75% of total potential inflow of LCD monitor has been achieved by 2009, with only a quarter of the cumulative sales remaining to be achieved thereafter. In the second scenario, at a slower substitution rate of LCD monitors, it is assumed that 65% of maximum potential inflows of LCD monitors have been achieved until 2009. In the third scenario, the slowest substitution rate of the three scenarios is assumed, with only 55% of LCD monitors expected to have been sold until 2009. Fitting the parameter estimates with SWICO collection data for LCD monitors from 2006 -2010 for each of the years, the peak disposal of LCD monitors will be in 2013 in case of rapid substitution, and in 2014 in the medium and slow substitution scenarios.

Figure 17: Disposal Curve - LCD Monitors Chapter III – Reverse Diffusion: 69 Estimating Disposal of Consumer Durables through Application of Diffusion Modelling

For each of the three scenarios, the parameter values for p and q, the maximum potential ownership, m, and the coefficient of determination, R2, are given in the table below:

Scenario 1 Scenario 2 Scenario 3

p 0.0001 0.0001 0.0001 q 0.5791 0.5643 0.5494 m 5,510,013 6,357,708 7,513,655 R2 0.98 0.98 0.97

Table 7: Parameter values and fit statistics

6. Model Validation Validity is a model’s property of adequately reflecting the system modelled and a primary measure of model quality (Schwaninger and Groesser, 2009). As the reverse diffusion model attempts to predict when consumers will eventually dispose of their consumer durable product, the predictive validity of the model is demonstrated by comparing the model forecast to the actual disposals. Bass, Krishnan and Jain (1994), have compared predictive qualities of their two diffusion models using step-ahead forecasting. First fitting the model for n periods, a forecast is made for the n+1th period. Re-estimating the model for n+1 periods, a forecast is made for the n+2th period and so on. In this paper, following Bass et al., (1994) a one-year ahead forecast is made for the year 2011 by using data until 2010 for parameterising the model to test the forecasting performance of the reverse diffusion model. The Mean Absolute Error is calculated to test the forecasting efficacy of the model.

6.1. CRT Monitors

The results indicate the reverse diffusion model underestimates actual disposals of CRT glass from PC monitors by 16%. The table below shows the model forecast and actual values for CRT monitors for 2011 as well as the Mean Absolute Error (MAE). A potential reason for the under-estimation could be that the monitors coming into the waste stream are bigger, with greater CRT glass mass, as compared to the fixed average mass of glass per PC used for in the model.


CRT Glass Disposal from CRT PC Monitors – 2011 Model Forecast [tonnes]

Actual Disposal [tonnes]

MAE [tonnes]

% Difference

2,142 2,559 417 - 16% Table 8: Model forecast vs actual disposal – CRT PC Monitors

6.2. CRT TVs

The reverse diffusion model forecast for CRT glass from CRT TVs also underestimates the actual disposal by approximately 19%. The table below shows the model forecast and actual values for CRT glass from CRT TVs for 2011 as well as the Mean Absolute Error (MAE). As in the case of PC monitors, the difference may be explained by the underestimation of the mass of CRT glass per TV. As larger screen CRT TVs, containing greater mass of glass per TV, were more popular, it is likely that the average mass of glass of the disposed TVs is higher than the average mass used in the model.

CRT Glass Disposal from CRT TVs – 2011

Model Forecast [tonnes]

Actual Disposal [tonnes]

MAE [tonnes]

% Difference

9,805 12,031 2226 - 19% Table 9: Model forecast vs actual disposal – CRT Glass from CRT TVs

6.3. LCD Monitors

Contrary to the CRT monitor and CRT TV forecasts, the reverse diffusion model overestimates the disposal of LCD monitors between 55% - 61% in the three scenarios described above. Counterintuitively, the disposal of LCD monitors in 2011 was fractionally lower than in 2010. A possible reason is the influence of uncertainty and volatility in macro-economic conditions which make consumers more conservative in their replacement and disposal of consumer durables. LCD Monitors Disposal – 2011 Scenario 1 Scenario 2 Scenario 3 Actual Disposal [units] 375,519 375,519 375,519 Model Forecast [units] 587, 648 603,564 583,116 MAE 212,129 228,045 207,597 % Difference 56% 61% 55% Table 10: Model forecast vs actual disposal - LCD Monitors


7. Discussion & Conclusion The three case studies presented above validate the model based on three consumer durable products collected by the SWICO Recycling take-back system. In each case, the parameter estimates are realistic and provide a reasonable fit to each product’s disposal path. The p values (the coefficient of technical disposal) in all cases are significantly smaller than the q values (the coefficient of discretionary disposal), indicating that the large majority of disposal of consumer durable products is driven by consumer behaviour. This is in keeping with evidence from Cripps and Meyer (1994) and Grewal et al. (2004) which indicates that discretionary replacements are more likely to occur when the motivation is perceived technological obsolescence than in the case of technical deterioration. The advantages of the reverse diffusion model as compared to other models of estimating disposal of consumer durables are two-fold. Firstly, the proposed model is parsimonious as it is able to provide disposal estimates even in the event of relatively sparse data on disposal. Secondly, the model makes it possible to estimate disposal flows in the absence of any sales data as required by previous models. To forecast the disposal path of a consumer durable by means of the reverse diffusion model, it is necessary to have some initial values to estimate the model parameters. Lilien, Rangaswamy and Van der Bulte (2000) suggest that with data for usually four or more periods it is possible to obtain the p and q parameters of the Bass model. With the formalisation of the take-back and recycling system for end-of-life consumer durables, as more data on disposal becomes available, the reverse diffusion model can similarly be used for forecasting the disposal trajectory with initial data from four to five periods as shown in the LCD monitor case study. The early forecasting efficacy of the model in such a case is highly dependent upon generating accurate estimates of the upper limit of potential adoption which can be estimated based on analogy with similar products or bounding analyses of expert opinions. For durable products or regions for which disposal data is as yet unavailable, it is possible to gain insights from the reverse diffusion history of analogous products or regions. This may be specially relevant when projections regarding the disposal must be made during the early stages of the product penetration, and may be typically be based on using reverse diffusion parameters from previous generations as analogy. Chapter III – Reverse Diffusion: 72 Estimating Disposal of Consumer Durables through Application of Diffusion Modelling

Lilien, Rangaswamy and Van der Bulte (2000) have suggested that analogies based on similarities in expected consumer behaviour are preferable to analogies based on product similarities alone, as well as a weighted average of p and q values of multiple analogues. It would be interesting for future research to explore how applicable analogies are for reverse diffusion models. However, the model is not without limitations. These limitations, however, could provide directions for future research. The case studies presented in this paper are only three products among a vast array of consumer durables. Generally, empirical data indicates that disposal of consumer durables follows the classic logistic curve. A more robust test would be required across product categories and across geographies to test the model. Future research should test the generalizability of this model for other durable goods such as, laptop computers, refrigerators, automobiles, etc. Another limitation, similar to Bass Diffusion Models, is the assumption that the coefficients p and q are constant over the time horizon of the model application, which implies that time-varying factors, such as changing consumer behaviour due to newer products, lower prices of newer products, more convenient disposal opportunities, etc. (which all may lead to disposals), are not explicitly considered. Several authors have indicated that the major reasons for replacement of consumer durables are change in consumer needs, socio-economic reasons such as higher income (Cooper, 2004; Bayus and Gupta, 1992) or dissatisfaction with production functionality, largely arising as a result of the availability of new technology (Cooper, 2004; Islam and Meade, 2000). The underestimated model values as compared to the actual values for the disposal of CRT TVs in 2010 when there was a sharp jump in the disposals of CRT TVs is likely due to changing consumer disposal behaviour. The sudden rise in disposals may be largely attributed to discretionary disposals by consumers who replaced their old CRT TVs for newer technology TVs such as LCD and Plasma TVs became more affordable as well as promotional deals offered during the year. The reverse diffusion model also does not shed light on the reasons for discretionary disposal, or the number of products disposed for technical or discretionary reasons. Therefore, it would be interesting to extend the model by explicitly including consumer behaviour aspects which can provide greater understanding of the drivers for disposal, especially discretionary disposals. Further research can look to


incorporating variables such as competitive effects of new technology, advertising, product quality, price and income effects into the model. Several authors (Mahajan and Muller, 1996; Norton and Bass, 1987) have examined the issue of whether diffusion accelerates between technology generations. Similarly, this could also be true for disposals, and it would be an area for further research whether reverse diffusion accelerates between technology generations.


References Antonides, Gerrit. 1991. “An economic-psychological model of scrapping behavior.” Journal of Economic Psychology 12 (2) (June): 357-379. doi:16/0167-4870(91)90020-T. Bass, Frank M. 1969. “A New Product Growth for Model Consumer Durables.” Management Science 15 (5, Theory Series) (January): 215-227. Bass, Frank M., Trichy V. Krishnan, and Dipak C. Jain. 1994. ‘Why the Bass Model Fits Without Decision Variables’. Marketing Science 13 (3): 203–223. Bayus, Barry. 1988. “Accelerating the Durable Replacement Cycle with Marketing Mix Variables.” Journal of Product Innovation Management 5 (3): 216-226. doi:10.1111/1540-5885.530216. Bayus, Barry L. 1991. “The Consumer Durable Replacement Buyer.” Journal of Marketing 55 (1) (January): 42-51. Bayus, Barry L., and Sachin Gupta. 1992. “An empirical analysis of consumer durable replacement intentions.” International Journal of Research in Marketing 9 (3) (August): 257-267. doi:10.1016/0167-8116(92)90021-C. Bergsdal, Håvard, Brattebø, Helge, Rolf A. Bohne, Müller,Daniel B. 2007. “Dynamic material flow analysis for Norway’s dwelling stock.” Building Research and Information 35 (September): 557-570. doi:10.1080/09613210701287588. Cripps, John D., and Robert J. Meyer. 1994. “Heuristics and Biases in Timing the Replacement of Durable Products.” The Journal of Consumer Research 21 (2) (September): 304-318. Cooper. 2004. “Inadequate Life? Evidence of Consumer Attitudes to Product Obsolescence.” Journal of Consumer Policy 27 (4) (December 1): 421-449. doi:10.1007/s10603-004-2284-6.


DeBell, M., and R. Dardis. 1979. “Extending product life: Technology isn’t the only issue.” Advances in Consumer Research 6 (1): 381–385. Elshkaki, Ayman, Ester van der Voet, Veerle Timmermans, and Mirja Van Holderbeke. 2005. “Dynamic stock modelling: A method for the identification and estimation of future waste streams and emissions based on past production and product stock characteristics*1.” Energy 30 (8) (June): 1353-1363. doi:10.1016/j.energy.2004.02.019. Fisher, J. C., and R. H. Pry. 1971. “A Simple Substitution Model for Technology Change.” Technological Forecasting and Social Change 3 (1): 75–88. Gregory, Jeremy R., Marie-Claude Nadeau, and Randolph E. Kirchain. 2009. “Evaluating the Economic Viability of a Material Recovery System: The Case of Cathode Ray Tube Glass.” Environmental Science & Technology 43 (24) (December): 9245-9251. doi:10.1021/es901341n. Grewal, Rajdeep, Raj Mehta, and Frank R. Kardes. 2004. “The Timing of Repeat Purchases of Consumer Durable Goods: The Role of Functional Bases of Consumer Attitudes.” Journal of Marketing Research 41 (1) (February): 101-115. doi:10.1509/jmkr.41.1.101.25090. Hanson, James W. 1980. “A Proposed Paradigm for Consumer Product Disposition Processes.” Journal of Consumer Affairs 14 (1): 49. Hilty, Lorenz M., Peter Arnfalk, Lorenz Erdmann; James Goodman, Martin Lehmann, and Patrick Wäger. 2006a: “The Relevance of Information and Communication Technologies for Environmental Sustainability – A Prospective Simulation Study.” Environmental Modelling & Software, 11 (21): 1618-1629 Hilty, Lorenz M., Andreas Köhler, Fabian von Schéele, Rainer Zah, and Thomas Ruddy. 2006b. “Rebound Effects of Progress in Information Technology.” Poiesis & Praxis. International Journal of Technology Assessment and Ethics of Science 1 (4): 19-38


Hoffer, G., and R. J Reilly. 1984. “Automobile styling as a shift variable: An investigation by firm and by industry.” Applied Economics 16 (2): 291–97. Horsky, Dan, and Leonard S. Simon. 1983. “Advertising and the Diffusion of New Products.” Marketing Science 2 (1): 1-17. Hu, Mingming, Ester Van Der Voet, and Gjalt Huppes. 2010. “Dynamic Material Flow Analysis for Strategic Construction and Demolition Waste Management in Beijing.” Journal of Industrial Ecology 14 (3) (May): 440-456. doi:10.1111/j.1530-9290.2010.00245.x. Islam, Towhidul, and Nigel Meade. 2000. “Modelling diffusion and replacement.” European Journal of Operational Research 125 (3) (September): 551-570. doi:10.1016/S0377-2217(99)00225-8. Jacoby, Jacob, Carol K. Berning, and Thomas F. Dietvorst. 1977. “What about Disposition?” The Journal of Marketing 41 (2) (April): 22-28. Jain, Dipak C., and Ram C. Rao. 1990. “Effect of Price on the Demand for Durables: Modeling, Estimation, and Findings.” Journal of Business & Economic Statistics 8 (2) (April): 163-170. Jones, J. Morgan, and Christopher J. Ritz. 1991. “Incorporating distribution into new product diffusion models.” International Journal of Research in Marketing 8 (2) (June): 91-112. doi:16/0167-8116(91)90018-3. Kang, Hai-Yong, and Julie M. Schoenung. 2006. “Estimation of future outflows and infrastructure needed to recycle personal computer systems in California.” Journal of Hazardous Materials 137 (2) (September): 1165-1174. doi:10.1016/j.jhazmat.2006.03.062. Kamakura, Wagner A, and Siva K Balasubramanian. 1987. ‘Long‐term Forecasting with Innovation Diffusion Models: The Impact of Replacement Purchases’. Journal of Forecasting 6 (1) (January 1): 1–19. doi:10.1002/for.3980060102.


Kamakura, Wagner A., and Siva K. Balasubramanian. 1988. “Long-term view of the diffusion of durables A study of the role of price and adoption influence processes via tests of nested models.” International Journal of Research in Marketing 5 (1): 1-13. Lifset, Gordon, Graedel, Spatari, and Bertram. 2002. “Where has all the copper gone: The stocks and flows project, part 1.” JOM Journal of the Minerals, Metals and Materials Society 54 (10) (October 15): 21-26. doi:10.1007/BF02709216. Lilien, G. L., A. Rangaswamy, and C. Van den Bulte. 2000. “Diffusion Models: Managerial Applications and Software.” New Product Diffusion Models: 295-311. Oguchi, Masahiro, Takashi Kameya, Suguru Yagi, and Kohei Urano. 2008. “Product flow analysis of various consumer durables in Japan.” Resources, Conservation and Recycling 52 (3) (January): 463-480. doi:10.1016/j.resconrec.2007.06.001. Peres, Renana, Eitan Muller, and Vijay Mahajan. 2010. “Innovation diffusion and new product growth models: A critical review and research directions.” International Journal of Research in Marketing 27 (2) (June): 91-106. doi:10.1016/j.ijresmar.2009.12.012. Pickering, J. F. 1981. “A behavioral model of the demand for consumer durables.” Journal of Economic Psychology 1 (1): 59–77. Mahajan, Vijay, Eitan Muller, and Frank M. Bass. 1990. “New Product Diffusion Models in Marketing: A Review and Directions for Research.” The Journal of Marketing 54 (1) (January): 1-26. Mahajan, V., E. Muller, and F. M. Bass. 1995. “Diffusion of New Products: Empirical Generalizations and Managerial Uses.” Marketing Science 14 (3): 79-88. Mahajan, V., and E. Muller. 1996. “Timing, diffusion, and substitution of successive generations of technological innovations: The IBM mainframe case.” Technological Forecasting and Social Change 51 (2): 109–132. Meade, Nigel, and Towhidul Islam. 2006. “Modelling and forecasting the diffusion of innovation - A 25-year review.” International Journal of Forecasting 22 (3): 519-545.


Norton, John A., and Frank M. Bass. 1987. “A Diffusion Theory Model of Adoption and Substitution for Successive Generations of High-Technology Products.” Management Science 33 (9) (September): 1069-1086. Schwaninger, M., and S.N. Grösser. 2009. “System dynamics modeling: validation for quality assurance.” Encyclopedia of Complexity and System Science. Springer, Berlin.(9): 9000-9014. Sherman, R., and G. Hoffer. 1971. “Does Automobile Style Change Payoff?” Applied Economics 3 (3): 153–65. Simon, H., and K. H. Sebastian. 1987. “Diffusion and Advertising: The German Telephone Campaign.” Management Science 33 (4): 451-466. Spatari, S., M. Bertram, R. B Gordon, K. Henderson, and T. E. Graedel. 2005. “Twentieth century copper stocks and flows in North America: A dynamic analysis.” Ecological Economics 54 (1): 37–51. Steffens, P. R. 2003. “A model of multiple-unit ownership as a diffusion process.” Technological Forecasting and Social Change 70 (9): 901–917. Streicher-Porte, M., R. Widmer, A. Jain, H. P Bader, R. Scheidegger, and S. Kytzia. 2005. “Key drivers of the e-waste recycling system: Assessing and modelling e-waste processing in the informal sector in Delhi.” Environmental Impact Assessment Review 25 (5): 472-491. Widmer, R., H. Oswald-Krapf, D. Sinha-Khetriwal, M. Schnellmann, and H. Böni. 2005. “Global perspectives on e-waste.” Environmental Impact Assessment Review 25 (5): 436-458. Yu, J., E. Williams, M. Ju, and Y. Yang. 2010. “Forecasting global generation of obsolete personal computers.” Environmental science & technology 44 (9): 3232–3237.


CHAPTER IV

Forecasting Consumer Durable Disposals: A Review and Comparison of Modelling Approaches

Deepali Sinha, University of St.Gallen Abstract This paper reviews two modelling approaches to forecasting disposal of consumer durables, namely the “delay model” approach and the “reverse diffusion model” approach. Applying the same dataset on the disposal of cathode ray tube monitors in Switzerland to both the approaches, the estimates and forecasts of the models are compared against real system data, sensitivity of the model parameters examined and the assumptions, strengths, limitations and applicability of both modelling approaches discussed. The comparison also provides an opportunity to discuss further improvements to both modelling approaches, especially the importance of developing models which incorporate consumer disposal behaviour.

Chapter IV – Forecasting Consumer Durable Disposals: 80 A Review and Comparison of Modelling Approaches

1. Introduction Consumer durable goods – from household appliances such as refrigerators and vacuum cleaners to entertainment electronics such as televisions and music players to information technology products such as computers and printers – have proliferated over the last fifty years. Not only have the number and types of products multiplied, their increased affordability has seen sales of consumer durables skyrocket. Estimates by Euromonitor (2011) suggest that global sales of consumer electronics were to the tune of 3 billion units in 2010. The flip side to increasing consumption of durable goods is the growing volumes reaching their end-of-life. The disposal of these end-of-life products has gained importance over the past decades because not only do they contain toxins which can be harmful to human health and the environment, but they also contain valuable precious metals and rare earths which are critical for the manufacture of new generation consumer durables. For example, by some estimates, the global material supply of critical metals in the manufacture of electronics such as Indium might be exhausted before the end of the decade (UNEP, 2009). The reservoirs of end-of-life products are therefore increasingly seen as valuable “mines” to recover scarce elements from. Over the past decade, these social, environmental and economic drivers have necessitated better management of end-of-life consumer durables, spurring legislation such as the European Union’s Waste Electrical and Electronic Equipment (WEEE) Directive, the Japanese Specified Home Appliances Recycling Law (SHAR) and the Swiss Ordinance on the Return, Taking back and Disposal of Electrical and Electronic Equipment (ORDEE), among others, which are specifically aimed at end-of-life consumer durables. Essential for sustainable and efficient management of end-of-life consumer durables is an accurate estimation of the timing and quantity of disposed consumer durables (Kang and Schoenung, 2006). Models, as simplified representations of real systems, can both reproduce or recreate ( “portrait”) and anticipate (“paragon”) and are crucial in providing an understanding of the real system (Schwaninger, 2010). Modelling end-of-life consumer durables to quantify the waste stream has drawn the attention of Chapter IV – Forecasting Consumer Durable Disposals: 81 A Review and Comparison of Modelling Approaches

several scholars with the aim of improving decision making efforts for policy makers (for example in setting targets), recyclers (for estimating availability of supply) or take-back systems (to estimate logistical and cost implications). Research so far on modelling disposals of consumer durables has tended to focus more on the sales and marketing of consumer durables rather than their disposal. In part, this is likely due to the fairly recent emergence of the problems associated with disposal of consumer durables. The main challenges however of modelling end-of-life consumer durables are:

1. Insufficient, or at best fragmented, time-series data regarding sales, existing products in use, and average product mass.

2. Insufficient understanding of consumer disposal behaviour, especially how and

when end-of-life durables are disposed of by the consumer. Nevertheless, several authors have proposed models to forecast the waste flow of durable goods, with some focussing on a product, quantifying the number or weight of one or many consumer durable product(s) (Widmer et al., 2005; Oguchi, et al., 2008; Yang, et al., 2008), while others study flows of materials from consumer durables, quantifying flows at the material level, such as lead, plastics, copper and glass (Elshkaki et al., 2005; Macauley, et al., 2003; Spatari et al., 2005, Gregory et al., 2009; Krivtsov et al., 2004). However, as yet, there has been no validation of the disposal forecasts of the models against observed data from a real system. As Schwaninger (2010) advises, it is not enough to build insightful models; they must also be valid. Although validity is not the only criterion of model quality, with other criteria including parsimony, ease-of-use, and practicality also important, model validity is considered the primary measure of model quality (Schwaninger and Groesser, 2009). Bolstering the argument for validation is a recent global review of the management of electrical and electronic waste management by Ongondo et al. (2011) who conclude that reported global quantities of WEEE are grossly underestimated.


Additionally, there is as yet no study that compares the structural properties, assumptions, strengths and limitations of the various models which differ from one to another. This paper aims to provide such a comparison between models, focussing on two generic modelling approaches, namely the delay model approach and the diffusion model approach. The paper also compares their predictive validity by applying each model to the case of the disposal of Cathode Ray Tube monitors (CRT monitors) in Switzerland. The paper is organised as follows: The next section provides a brief overview of the two modelling approaches and a summary of the terminology and definitions, followed by the purpose and objectives of the research. In Sections 4 and 5, both modelling approaches are described in detail, including three variants of the delay model approach, with a structural comparison and a summary table comparing their main characteristics in Section 6. In the following section, the forecasting results of four models are compared against real system data on cathode ray tube monitors disposed of in Switzerland. A fit improvement and sensitivity analysis is also conducted to compare the performance of the models. The final section discusses the advantages and limitations and suggests areas for future research in improving the models.

2. Modelling Approaches to Estimate Consumer Durable Disposals

The most commonly used modelling approach to estimate disposals of consumer durable product is the “delay model” approach, also sometimes referred to as the market supply approach (Widmer et al., 2005). Citing Lohse et al. (1998), Widmer et al. (2005) have also mentioned the “consumption and use” and “market saturation” approaches for estimating end-of-life consumer durables. The consumption and use method takes the average consumer durables of a typical household as the basis for a prediction of the potential amount of end-of-life products, while the market saturation approach is based on the assumption that private households are already saturated with consumer durables, and for each new product purchased, an old one reaches its end-of-life. However, in literature, no applications of or further research on these approaches were found, and therefore are not considered in this paper.


The second modelling approach discussed and compared in this paper is the “reverse diffusion” approach. Diffusion models are extensively used in the forecasting of sales of new consumer durables. Chapter III of this thesis, suggests that similar to sales of new products which follow a sigmoid curve, the cumulative disposal of old consumer durables also tends to follow a sigmoid curve and have applied the diffusion modelling approach to disposal of consumer durables. Both the delay model and the reverse diffusion model approach are discussed further in detail in the next section.

2.1. Terminology

One of the challenges in comparing models is to find a common terminology, especially with the aim of understanding both implicit and explicit assumptions. The main variables, and various terms used for these in delay and diffusion models are discussed briefly in the section below.

2.1.1. Inflows

“Sales” and “shipments” are often synonymously used for inflows of consumer durables into the consumption stock, even though there can be significant time gaps between shipments of products from manufacturers to distributers before being finally sold to a consumer. However, some authors (e.g. Oguchi et al., 2008) argue that given lean inventory management, the delay between shipment and sales is insignificant, making it a close approximation of sales data and therefore used for inflow data. An implicit assumption that all models make regarding inflows is that all sales are of new products within geographic boundaries, not accounting for second-hand sales or imports of second-hand consumer durables.

2.1.2. Stocks

The stock of consumer durables has been conceptualised differently by different authors. As consumer durables are often reused, resold and stored by consumers before finally being disposed of (Widmer et al., 2005), differences in the definition of “stock” can have significant implications in the magnitude of the stock. Spatari et. al., (2005) consider stocks in their model to include only those products in use, not including “hibernating” products comprising those that have been retired and remain Chapter IV – Forecasting Consumer Durable Disposals: 84 A Review and Comparison of Modelling Approaches

in households unused. In contrast, Oguchi et al.,(2008) consider their stocks to include all products in households, whether being used or stored. They explicitly state that hibernating or out-of-use products stored in households are not considered as waste, and hence form part of the consumption stock. Kang & Schoenung (2006) take a similar approach, defining stock in the consumer phase as the sum of the amount being used by the first user, any following users and stored products that are not in active. Consequently, Spatari et al.(2005) calculate the discard rate of WEEE without distinguishing between discarded electronics, i.e. those which are disposed of, and ‘hibernating’ electronics, i.e. those which remain in households unused. In contrast, (Oguchi et al., 2008) define WEEE flows only as those out-of-use products that are disposed of.

2.1.3. Delay Distribution

The lifetime of a consumer durable is an essential piece of information in modelling waste flows at end-of-life in the delay model approach. Various terms are used in literature by authors to describe the time a durable product is considered as part of a stock, between inflow and outflow. Commonly used terms are “average lifetime”, “residence time”, “survival function”, “disposal distribution function”, “product lifetime function”, “disposal function”, “domestic service lifetime”, “total lifetime”, and “possession time”. Though very similar, the terms may likely include or exclude one or several intermediate stages (e.g. reuse or storage) between purchase and disposal by a consumer. Additionally, some authors consider lifetime as a fixed value (or Dirac distribution) while others use parametric or non-parametric distributions in their models. Oguchi et al., (2010) provide a comprehensive review of the different types of lifetime distribution and as well as distribution estimation methodologies.

2.1.4. Product Mass

Estimating units of disposed consumer durables, while useful, may often not be sufficient for decision making. In such cases, the total estimated physical mass or the composition of the waste stream in terms of specific materials is required. Most authors use a constant average product mass throughout the entire time horizon of the model. However some authors caution against a constant value, especially given that the size and material composition of products change over time. The reduction in the average physical mass of a mobile phone from over 350 g in 1990 to about 80 g in


2005 (Hilty, et al., 2006a) demonstrates the benefits of product mass function that reflects the changes to the product mass over time over a constant value.

3. Goal and Purpose The goal of the paper is to review and compare two modelling approaches used to estimate and forecast disposals of consumer durables. Applying the same dataset to the different models, the paper examines their differences and compares the performance of outputs of the models against a real system. Through such a comparison, the paper aims to provide a summary of strengths, limitations, applicability and acuity of two modelling approaches as well as discuss research gaps and areas for improvement in consumer durable disposal modelling. Thus, the paper will enable policy makers, recyclers and other stakeholders to make better and more informed forecasts of consumer durable disposals by using the most appropriate modelling approach in view of available data.

4. Delay Model Approach Modelling waste streams from durable products using the delay model is well established and has been used to forecast waste such as concrete from buildings (Müller, 2006), automobiles (Steffens, 2001) as well as consumer durables. In the delay model approach, the outflow of products (disposals) is dependent on the inflow of products (sales) and the time spent as stock, given by the delay function (lifetime). The outflow therefore is independent of the stock, which acts as time buffer, with the timing of disposal of a product linked to when the product entered into stock, as it is more likely for older devices to be disposed of than newer devices. The outflow is determined by the inflow and the average delay time. The outflows (disposals) at time t are expressed as a convolution of the inflows In and the disposal density d: Thus,


∞

𝑖=−∞

(28)


Most authors use the delay model approach with variations in the delay time distribution for average life. In its simplest form, the average delay time has a Dirac distribution (Streicher-Porte et al., 2005; Kang & Schoenung, 2006) with disposals following exactly the same curve as sales, only shifted in time by the fixed delay. Although authors have previously used a fixed delay distribution, it is increasingly common to use distributed delays which are more realistic, as products from a cohort are not all disposed at the same time, with some disposed of earlier than others. Three variants of the delay model approach are discussed below:

1. A delay model with the Weibull distribution, the most commonly used lifetime distribution for estimation of disposals of end-of-life consumer durables. The model by Oguchi et al. (2008) is used an example.

2. The second delay model uses a derived lifetime distribution, as well as a

product mass function. The model by Gregory et al. (2009) is used as an example.

3. The third delay model, presented in Chapter II of this thesis also uses a Weibull

lifetime distribution, albeit coupled with a technology substitution model, and estimates the parameters of the Weibull within the model, as opposed to externally.

4.1. Delay Model A: Example Reference – Oguchi et al., 2008

In this model, the mass of disposed consumer durables is expressed as a product sum of the shipments or sales, the lifetime distribution and a fixed average mass per unit. Oguchi et al. (2008), Elshkaki et al. (2005), Steffens (2001), Mueller et al. (2006) among others use a Weibull distribution function for the lifetime. The model specification is as follows:


∞

𝑖=−∞

⋅ 𝑚 (29)


where 𝐼𝐼𝐼𝐼 is the shipments or sales into the market, 𝑑𝑑𝑖 is the lifetime distribution and 𝑚 is the mass of the product which is considered to remain constant over time. The two parameters of the Weibull distribution are interpreted as the average lifetime and the deviation. These parameter values are estimated separately, in most cases either through a sample survey of the age of disposed products at the point of disposal (eg. recycling facilities) or through a questionnaire survey of consumers regarding age of owned products at time of disposal. Oguchi et al., (2008) have modelled waste flows for a diverse range of consumer durables ranging from large household appliances such as washing machines and refrigerators to consumer electronics equipment such as televisions and radios to information and communication technology products such as personal computers and printers.

4.2. Delay Model B: Example Reference – Gregory et al., 2009

Gregory et al. (2009), Yang et al. (2008), Yu et al. (2010) have modelled the disposal of consumer durables by specifying their delay models using non-parametric distributions. Oguchi et al. (2010) suggest that with sufficient data, a non-parametric approach can derive a more precise distribution. In the case of Gregory et al. (2009), the retirement probability or lifetime distribution was derived from information based on data from a collection trial. In addition, Gregory et al. (2009) have also included in their model a product mass function that incorporates the changing weight of TVs and PC Monitors as heavier, larger screen size products are sold. The model specification is as follows:


∞

𝑖=−∞

⋅ 𝑚𝑖 (30)

where 𝐼𝐼𝐼𝐼 is the shipments or sales into the market, 𝑑𝑑𝑖 is the derived lifetime distribution and 𝑚𝑖 is sales-weighted product mass.


4.3. Delay Model C: Example Reference – Chapter II

The model presented in Chapter II of this thesis adopts the delay model approach to estimate the societal stocks and flows of consumer durables. The model incorporates the Fisher-Pry technology substitution model (Fisher and Pry, 1971) to estimate sales of a durable product. The Fisher-Pry substitution model is given by:

𝑠(𝑡𝑡) =12

[1 + tanh𝛼(𝑡𝑡 − 𝑡𝑡ℎ)] (31)

where s(t) is the fraction of the sales substituted by the new technology, 𝑡𝑡ℎ is the time at which the new technology has substituted 50% of the market share, and 𝛼 is interpreted as a parameter indicating half the annual fractional growth rate in the early years of the new technology product. Disposals are expressed as a product sum of the sales, the lifetime (a Weibull distribution) and a fixed average mass per unit. The model specification is as follows:


∞

𝑖=−∞

⋅ 𝑚 (32)

where 𝐼𝐼𝐼𝐼 is the sales into the market, 𝑑𝑑𝑖 is the lifetime distribution and 𝑚 is the mass of the product which is considered to remain constant over time. The difference between the delay models A and C is in the estimation of the Weibull parameters. Whereas in delay model A the Weibull parameters values are externally estimated, in delay model C, the values are estimated iteratively by optimisation, setting the objective function to minimize the sum of squared residuals.


5. Reverse Diffusion Model Approach The diffusion model, which has its origins in ecology, was first applied as a model to forecast sales of consumer durables by Bass in 1969. It has since become one of the most widely researched models in marketing, with several extensions and modifications to the original Bass Diffusion Model. In the reverse diffusion model proposed in Chapter III, the market depletion of a durable is driven by technical disposals and discretionary disposals, characterised by the parameters p and q, where p is considered the coefficient of technical disposal, influenced by factors such as breakdown and wear and tear; and q is considered the coefficient of discretionary disposal, influenced by social and functional signals such as status, new product technologies, etc. The reverse diffusion differential equation is given as: �̇� = 𝑝 ∙ (1 − 𝑆) + 𝑞𝑆 ∙ (1 − 𝑆) (33) Where �̇� is the change in stock, p is the coefficient of innovation and q is the coefficient of imitation and has the following solutions.

𝑅(𝑡𝑡) =1 − 𝑒−(𝑝+𝑞)𝑡

1 + 𝑞𝑝 𝑒−(𝑝+𝑞)𝑡

(34)

𝑟(𝑡𝑡) = (𝑝 + 𝑞)2

𝑝


�1 + 𝑞𝑝 𝑒−(𝑝+𝑞)𝑡�

2 (35)

where R(t) is the cumulative disposal function, r(t) is the disposal density function, i.e. the specific disposal rate. The absolute disposal rate, d(t) is given by: d(t) = m ∙ r(t) (36)


where m is the total number of the durable products in the market stock which is to be depleted to the disposal stock.

6. Structural Comparison The underlying structure of the delay and the reverse diffusion models, as discussed in Chapters II and III respectively, are illustrated in the diagrams below.

Figure 18: Delay Model Structure

Figure 19: Reverse Diffusion Model Structure From the diagrams, it can be seen that both models are structurally different. In the delay model, the outflows (disposals) are expressed as a convolution of the inflows In and the disposal density d, and entirely independent of the stock. In comparison, in the case of the reverse diffusion model, the outflow is independent of the inflow. The outflow feeds a disposal stock which determines, together with the parameters p and q

�𝑑𝑑𝑡𝑡

disposal d(t)

Out (t) In (t)

Stock (t)

Market (t) Disposal (t) In (t) Out (t)

p q


the dynamics of the outflow. The total stock in the market which is to be depleted to the disposal stock scales the outflow. The table below provides a summary of the characteristics of the four models being compared.

Model 1 - Delay Model A

Model 2 - Delay Model B

Model 3 - Delay Model C

Model 4 - Diffusion Model

Example Reference

Oguchi et al. (2008)

Gregory et al. (2009)

Chapter II Chapter III

Model Independent Variables

Inflows of product

Inflows of product; Product mass function (time variant)

Stock of product; Inflow of new technology

Disposal time-series

Delay distribution

Weibull Derived non-parametric lifetime distribution

Weibull n/a

Model Parameters

Fixed average mass of product (time invariant) Weibull parameters alpha and beta

n/a

Fixed average mass of product (time invariant) Weibull parameters alpha and beta

p (coefficient of technical disposal) q (coefficient of discretionary disposal)

Lifetime distribution estimation methodology

Based on a user survey

Based on survey at disposal point

Parameters estimated within model

n/a

Terminology - Point of “disposal”

Does not consider storage time – if not part of active stock, considered as disposed



Storage time implicitly included – only counted if actually disposed

Table 11: Summary table of model characteristics [n/a = not applicable]


7. Experimental Frame The models discussed above are compared to each other by applying them to the case of the disposal of Cathode Ray Tube (CRT) monitors in Switzerland. CRTs had been the dominant display technology for both televisions and Personal Computer (PC) monitors until they were rendered obsolete with the advent of new flat panel display technologies. Personal computers were introduced to the Swiss market in the early 1980s. Disposals of these products followed soon after, and by 1994, there was a formal collection and take-back system to ensure sound disposal of end-of-life PCs. With reliable data on sales, stocks and specially disposals, estimating and forecasting flows of CRT monitors makes an excellent case study to compare and validate the models discussed above. Data on the stock and inflow (sales) of PC monitors was obtained from Robert Weiss Consulting (www.weissbuch.ch), while data on outflows (disposals) was acquired from SWICO Recycling, the Swiss producer responsibility organisation that manages the collection and take-back system for several consumer durables, including IT equipment such as PC monitors and EMPA, who provide the monitoring and control for the SWICO system.

8. Results Applying the data on CRT monitors to all four models discussed above, annual CRT glass disposal in Switzerland is estimated and compared with collection data from the SWICO Recycling system. Several assumptions are made in order to compare the model outputs.

1. Given the long experience and high consumer awareness of the SWICO Recycling take-back and collection system, currently the collection efficiency of the system is very high, at 90%, as confirmed by SWICO Recycling experts, with only a small fraction of end-of-life product disposals not captured by the system. For simplicity, a collection efficiency of 90% is assumed throughout


the time period modelled, although it is likely that collection efficiency in earlier years was not as high.

2. The value of the average mass of glass in CRT monitors, 9.5 kgs/monitor is taken as the average of 10.5 kgs/monitor from a study by Monchamp et al. (2001) and 8.5 from a study by Huisman et al. (2008).

3. It is assumed that the parameter values of the Weibull distribution in Delay

Model A are the same for Switzerland as they are for Japan as estimated by Oguchi et al. (2008), given that both are highly developed economies with likely similar consumption and disposal patterns.

4. The derived lifetime distribution as also the sales-weighted product mass function has been estimated for the United States of America by Gregory et al. (2009). It is assumed that the same distribution and product mass function are applicable to Switzerland, given that both are developed, industrialised economies with similar consumption and disposal patterns.

8.1. Output Comparison

In Figure 20 below, the disposal curve of the four models is plotted alongside the observed data from the SWICO Recycling collection system. The reverse diffusion model performs the best, followed by Delay Models C, A and B.

Figure 20: Model Comparison - CRT Glass Disposal Estimates vs Observed


Parameter and fit statistics for all four models is given in the table below. 𝛼 and 𝛾 are the scale and shape parameters of the Weibull distribution respectively, and p and q are parameters of the reverse diffusion model. 𝛼 is interpreted as the average lifetime of the product, with 𝛾 the deviation from average lifetime.

Delay Model A

Delay Model B

Delay Model C

Reverse Diffusion Model

𝜶 OR p* 6.70 n/a 8.72 0.0001 𝜸 OR q* 4.80 n/a 2.83 0.3490 m (crt glass kgs/monitor) OR m (total tonnes)* 9.5 7.7 – 11.3 9.5 61,735 R2 0.05 -0.01 0.81 0.99

Table 12: Parameter values and fit statistics (* indicates parameters of reverse diffusion model) From Figure 20, it is clear that the delay models A and B are predicting earlier disposals than actually observed. Delay model A forecasts the quickest disposal of CRT monitors, with the disposal nearly complete by 2011. Model C performs the best of the three delay models, with the highest R2 of 0.81 of the three delay models. This is likely due to the higher average age of Model C as compared to Model A, indicating that the average age of CRT PC monitors in Switzerland at the time of disposal was closer to 9 years than 7 years. Sampling by SWICO Recycling has shown that the average age of PCs collected by the take-back system is approximately 9 years (Widmer et al., 2005) which explains the better fit of Model C as compared to Model A and B. The influence of the Weibull parameters in forecasting disposals is investigated further as part of the sensitivity analysis of the parameters.

8.2. Predictive Validity

The predictive ability of a model is demonstrated by comparing the model forecast to the observed values of the real system. To validate and compare the predictive power of the four models, truncated data until 2005 is used for fitting the models and forecasts produced for the balance time periods (2005-2011) by extrapolating the fitted models (Figure 21). The mean absolute error (MAE) is used to measure how


close forecasts are to the eventual outcomes. The lower the MAE, the better the predictive performance. The mean absolute error is highest for delay model B, which is 6.89 times higher than the lowest mean absolute error for the reverse diffusion model. The results are reported in Table 13: Comparison of Predictive Power below.

Delay Model A

Delay Model B

Delay Model C

Reverse Diffusion Model

Mean Absolute Error 1872.68 2014.29 700.45 292.35 Table 13: Comparison of Predictive Power

Figure 21: Disposal Forecasts

8.3. Fit Improvement

To test if the fit of delay models A and C can be improved by changing the mass from a fixed value to a variable function, the sales-weighted product function from model B is applied to models A and C. The Modified Models A and C are recalculated with the weight of CRT Glass mass per monitor varying from 7.7 to 11.3 kgs/ monitor depending on the year of sale.


Figure 22: Fit improvement by introducing product mass function Both models show an improved fit as a result, as shown in Table 14. This reflects the increased sales over time of monitors with larger screen sizes, thereby containing more glass per monitor. For the Modified Delay Model A, the coefficient of determination, the R2, improves substantially by 0.17 to become 0.22. Delay Model C also sees an improvement in the fit, increasing marginally by 0.09. However, it must be noted that this improvement is at the cost of simplicity as it introduces an additional variable and data requirements.

Model Weibull Coefficients

CRT Glass Mass R2 𝚫R2

𝜶 𝜸 [kgs/monitor] Modified Delay Model A

6.70 4.80 Sales-weighted 7.7 – 11.3 0.22 0.17

Modified Delay Model C

8.72 2.83 Sales-weighted 7.7 – 11.3 0.90 0.09

Table 14: Fit improvement statistics


8.4. Sensitivity Analysis

A sensitivity analysis of the model parameters is performed to determine which parameters exert the most influence on model results. For delay models, sensitivity of Weibull parameters is tested, while for the reverse diffusion model, sensitivity of the coefficients, p and q, is analysed. Delay model A is used as a representative for the delay models as it is expected that the parameter sensitivity of delay models is similar.

8.4.1. Sensitivity of Delay Model Parameters

A sensitivity analysis is performed by testing the fit of model by a change of 30%, both positive and negative for both Weibull parameters (𝛼, 𝛾) of the delay model A. As seen in Figure 23, the model is far more sensitive to the change in average lifetime (𝛼) than the change in the deviation (𝛾). As lifetime parameter, 𝛼, is increased, the fit of the model, indicated by R2 plotted on the Y-axis, improves significantly. In comparison, even a large change in the deviation of the distribution parameter, 𝛾, affects only a very small change in the fit.

Figure 23: Sensitivity of delay model parameters

8.4.2. Sensitivity of Reverse Diffusion Parameters

A sensitivity analysis of the reverse diffusion model parameters shows that the fitted parameters were at the optima, maximising R2. The model is sensitive to changes in both p and q parameters, with the q parameter having a greater influence on the model


fit than the p parameter. Importantly, it indicates that any change in the coefficient of discretionary disposal can reduce the fit of the model significantly. In case of changing consumer disposal behaviour, with greater or lesser discretionary disposal, the model fit might be reduced.

Figure 24: Sensitivity of reverse diffusion model parameters

9. Discussion and Conclusion A comparison of the two modelling approaches namely the delay modelling approach and the reverse diffusion approach provides valuable insights for forecasting disposals of consumer durables. Both modelling approaches are equally valid, and can provide reasonably accurate forecasts given sufficient data. The choice of model would largely be dependent on the quantity and quality of the data available. For delay models, time series data on inflows of consumer durables is essential as is the lifetime distribution. The sensitivity analysis illustrates the importance of parameter estimates, in particular regarding the alpha parameter (interpreted as the average lifetime), as a 20% change in the parameter results in the fit improving by 330%. In the absence of accurate lifetime distribution data, it may be better to estimate the parameters within the model, as in Model C to get more accurate forecasts. Such


an approach can be particularly useful in cases where fragmented or incomplete data are available. As seen above, the lifetime distribution is crucial in providing accurate estimates of the timing of disposals of consumer durables. Parameters from Japan did not provide such a good fit to Swiss data, potentially indicating that either the lifetime in Japan and Switzerland are significantly different due to different consumer usage patterns, or that there was a long period of storage between the time the products were considered obsolete and their physical disposal, or likely a combination of both factors. In every case, it is important to understand consumer behaviour. In comparison, data requirements for the reverse diffusion model are very different – namely that it requires data on outflows, rather than inflows. This simple and parsimonious model can be particularly useful for established take-back and collection systems with data over several years. Additionally, its parameters can be updated every year by fitting to the latest available data, thereby enabling ever more accurate forecasts for the years ahead. However, a limitation of the reverse diffusion model is that it needs data for at least a minimum of 3 years, and preferably more, ideally until the reverse diffusion reaches the mid-way point, to be able to provide forecasts with greater accuracy. Additionally, it is likely that the collection efficiency of the system biases the disposal results, as collection data would not reflect the true disposals especially in the case of any leakages from the system. End-of-life consumer durables such as TVs and PCs are often shipped from developed countries to developing countries illegally, thereby are not accounted for in the formal collection system. A model fitted to only collection data from legal, formal systems risks underestimating actual disposals. Disposals of consumer durable products are governed by socio-economic factors that we are barely beginning to understand. Several authors (Elshkaki, 2004; Oguchi, 2008) have hinted that technological advancement and social acceptability may also be factors that influence disposal. However, as yet there is little or no research regarding the timing and fate of end-of-life consumer durables. In a pilot research exploring the consumer motivations behind disposal of consumer durables, Khetriwal and First (2011) propose that disposals are dependent on “triggers” and “influencers”.


In their study on the disposal of TVs, they found evidence that the price of new products, new product features and promotions of new products, as well as changes in the technological landscape trigger product obsolescence. Furthermore, they also found social, price and technology sensitivity of consumers influencing the effect of the aforementioned obsolescence triggers. The preliminary results of their research showed that consumer disposal decisions on why, when, and how to dispose of a durable product are not as much based on the product life as based on technical failure, but more so on subjective reasons and consumer perceptions which lead to discretionary obsolescence of consumer durables. One of the major drawbacks of both models is that neither incorporates the time-varying nature of consumer behaviour. Anecdotal evidence for PCs and mobile phones has shown that average lifetime of these products has reduced over time, with more frequent replacements and disposals taking place. Another drawback in all four models discussed above is that they do not account for storage time between consumers displacing their old products and disposing them. Though consumers may replace their old products, they rarely dispose of these immediately. For policy makers and waste managers, it is essential to bear in mind that even in saturated markets, with most sales being replacement sales, they may not find as many products disposed of due to extensive storage periods as well as the importance of second hand market, especially for some consumer durables. Incorporating such aspects in a model can help give estimates regarding “hibernating” stocks, which can be particularly helpful in understanding potential anthropogenic stocks available for recycling and recovery, especially in the light of material scarcity. The delay model is well suited to disaggregation into stages such as reuse and storage. Such a “nested-delay” model has been presented by Widmer et al. (2005) for PC disposals, albeit using a fixed (dirac) lifetime distribution. Further research on consumer behaviour will not only be able to provide better parameter estimates for the delay functions at every stage, but also inform model design in terms of stages an end-of-life consumer durable. For the reverse diffusion model, future research should be aimed at disaggregating discretionary disposal into various aspects to get better understanding of drivers of


disposal. Such insights will be especially relevant given that the large majority of consumer products are replaced not because they stop functioning, but for reasons motivated by consumer decisions. Additionally, this would also help overcome another assumption the model makes – in that all consumers are homogeneous. Given that several consumer durables are dual-use products, widely used in homes and in business and commercial environments, it would provide insights into the different drivers and disposal options preferred by different user groups. The models presented above do not shed any light on why and how consumers dispose of their durable goods, especially when they are still functional. Neither model provides any insight into the factors that influence disposal decisions or the alternatives to disposal that consumers may consider. This calls for interdisciplinary behavioural research efforts to understand consumer disposal behaviour which can be incorporated into forecasting models. Using system dynamics models is one possible approach in this direction. Hilty et al., (2006b) have previously used system dynamics modelling in their simulation of the environmental sustainability of information and communication technologies, which includes a sub-model on waste generated from electronics. Such models can be expanded to include consumer dynamics, as has been explored by Ulli-Beer (2006) in her model for solid waste management at the local level which incorporates consumer recycling behaviour. Further research into consumer behaviour to get insights into why, when and how consumers dispose their durable products, will provide useful information that could lead not only to better, more accurate forecasting models, but also inform consumer education and awareness programs directed towards improving consumer attitudes towards disposal of durable products.


References Bass, Frank M. 1969. ‘A New Product Growth for Model Consumer Durables’. Management Science 15 (5, Theory Series) (January): 215–227. Elshkaki, A., E. Van der Voet, M. Van Holderbeke, and V. Timmermans. 2004. ‘The Environmental and Economic Consequences of the Developments of Lead Stocks in the Dutch Economic System’. Resources, Conservation & Recycling 42 (2): 133–154. Elshkaki, Ayman, Ester van der Voet, Veerle Timmermans, and Mirja Van Holderbeke. 2005. ‘Dynamic Stock Modelling: A Method for the Identification and Estimation of Future Waste Streams and Emissions Based on Past Production and Product Stock Characteristics*1’. Energy 30 (8) (June): 1353–1363. doi:10.1016/j.energy.2004.02.019. Euromonitor. 2011. ‘Consumer Electronics 2010’. Fisher, J. C., and R. H. Pry. 1971. ‘A Simple Substitution Model for Technology Change’. Technological Forecasting and Social Change 3 (1): 75–88. Gregory, Jeremy R., Marie-Claude Nadeau, and Randolph E. Kirchain. 2009. ‘Evaluating the Economic Viability of a Material Recovery System: The Case of Cathode Ray Tube Glass’. Environmental Science & Technology 43 (24) (December): 9245–9251. doi:10.1021/es901341n. Hilty, Lorenz M., Andreas Köhler, Fabian von Schéele, Rainer Zah, and Thomas Ruddy. 2006a. “Rebound Effects of Progress in Information Technology.” Poiesis & Praxis. International Journal of Technology Assessment and Ethics of Science 1 (4): 19-38 Hilty, Lorenz M., Peter Arnfalk, Lorenz Erdmann; James Goodman, Martin Lehmann, and Patrick Wäger. 2006b: “The Relevance of Information and Communication Technologies for Environmental Sustainability – A Prospective Simulation Study.” Environmental Modelling & Software, 11 (21): 1618-1629 Chapter IV – Forecasting Consumer Durable Disposals: 103 A Review and Comparison of Modelling Approaches

Huisman, J., F. Magalini,, R. Kuehr, C. Maurer, J. Poll, C. Delgado, E. Artim, et al. 2008. ‘2008 Review of Directive 2002/96 on Waste Electrical and Electronic Equipment (WEEE), Final Report’. Kang, H. Y, and J. M Schoenung. 2006. ‘Estimation of Future Outflows and Infrastructure Needed to Recycle Personal Computer Systems in California’. Journal of Hazardous Materials 137 (2): 1165–1174. Khetriwal, Deepali S., and Ivana First. 2011. ‘Enabling Closed Resource Loops in Electronics: Understanding Consumer Disposal Behaviour Using Insights from Diffusion Models’. In Proceedings of CROMAR 2011. Pula. Krivtsov, V, P.A Wäger, P Dacombe, P.W Gilgen, S Heaven, L.M Hilty, and C.J Banks. 2004. ‘Analysis of Energy Footprints Associated with Recycling of Glass and Plastic—case Studies for Industrial Ecology’. Ecological Modelling 174 (1–2) (May 1): 175–189. doi:10.1016/j.ecolmodel.2004.01.007. Lohse, J., S. Winteler, and J. Wulf-Schnabel. 1998. Collection Targets for Waste from Electrical and Electronic Equipment (WEEE) the Directorate General (DG XI) Environment. Macauley, Molly, Karen Palmer, and Jhih-Shyang Shih. 2003. ‘Dealing with Electronic Waste: Modeling the Costs and Environmental Benefits of Computer Monitor Disposal’. Journal of Environmental Management 68: 13–22. Monchamp, A., H. Evans, J. Nardone, S. Wood, E. Proch, and T. Wagner. 2001. Cathode Ray Tube Manufacturing and Recycling: Analysis of Industry Survey. Washington DC (USA): Environmental Health Center - A Division of the National Safety Council. Müller, D. 2006. ‘Stock Dynamics for Forecasting Material flows—Case Study for Housing in the Netherlands’. Ecological Economics 59 (1): 142–156.


Oguchi, Masahiro, Takashi Kameya, Suguru Yagi, and Kohei Urano. 2008. ‘Product Flow Analysis of Various Consumer Durables in Japan’. Resources, Conservation and Recycling 52 (3) (January): 463–480. doi:10.1016/j.resconrec.2007.06.001. Oguchi, Masahiro, Shinsuke Murakami, Tomohiro Tasaki, Ichiro Daigo, and Seiji Hashimoto. 2010. ‘Lifespan of Commodities, Part II’. Journal of Industrial Ecology 14 (4) (August 1): 613–626. doi:10.1111/j.1530-9290.2010.00251.x. Ongondo, F.O., I.D. Williams, and T.J. Cherrett. 2011. ‘How Are WEEE Doing? A Global Review of the Management of Electrical and Electronic Wastes’. Waste Management 31 (4) (April): 714–730. doi:10.1016/j.wasman.2010.10.023. Schwaninger, M. 2010. ‘Model-based Management (MBM): a Vital Prerequisite for Organizational Viability’. Kybernetes 39 (9/10): 1419–1428. Schwaninger, M., and S.N. Grösser. 2009. ‘System Dynamics Modeling: Validation for Quality Assurance’. Encyclopedia of Complexity and System Science. Springer, Berlin. Spatari, S., M. Bertram, R. B Gordon, K. Henderson, and T. E. Graedel. 2005. ‘Twentieth Century Copper Stocks and Flows in North America: A Dynamic Analysis’. Ecological Economics 54 (1): 37–51. Steffens, P. R. 2001. ‘An Aggregate Sales Model for Consumer Durables Incorporating a Time-varying Mean Replacement Age’. Journal of Forecasting 20 (1): 63–77. Streicher-Porte, M., R. Widmer, A. Jain, H. P. Bader, R. Scheidegger, and S. Kytzia. 2005. ‘Key Drivers of the E-waste Recycling System: Assessing and Modelling E-waste Processing in the Informal Sector in Delhi’. Environmental Impact Assessment Review 25 (5): 472–491. Ulli-Beer, Silvia. 2006. Citizens’ Choice and Public Policy: A System Dynamics Model for Recycling Management at the Local Level. Shaker Verlag GmbH, Germany.


UNEP. 2009. ‘Recycling – From E-waste to Resources’. Widmer, R., H. Oswald-Krapf, D. Sinha-Khetriwal, M. Schnellmann, and H. Böni. 2005. ‘Global Perspectives on E-waste’. Environmental Impact Assessment Review 25 (5): 436–458. Yang, J., B. Lu, and C. Xu. 2008. ‘WEEE Flow and Mitigating Measures in China’. Waste Management 28 (9): 1589–1597. Yu, J., E. Williams, M. Ju, and Y. Yang. 2010. ‘Forecasting Global Generation of Obsolete Personal Computers’. Environmental Science & Technology 44 (9): 3232–3237.


Annexes

Annexe 1: CRT Monitor Sales in Switzerland

Year CRT Monitor Sales [units] FPD Monitor Sales [units] 1983 10,000 1984 18,500 1985 38,000 1986 67,000 1987 110,000 1988 160,000 1989 235,000 1990 274,000 1991 297,000 1992 320,000 1993 377,000 1994 449,000 1995 519,000 1996 538,000 1997 632,000 1998 717,250 1999 618,540 4,020 2000 584,000 13,250 2001 326,800 59,400 2002 124,380 128,800 2003 55,760 210,000 2004 15,780 310,800 2005 8,340 597,800 2006 755,440 2007 695,000 2008 726,000 2009 632,000 2010 564,000

Table 15: CRT and FPD Monitor Sales in Switzerland. Source: Robert Weiss Consulting

Annexe 107

Annexe 2: CRT and FPD TV Sales in Switzerland

Year Sale - CRT TV

[units] Sales - FPD TVs [units]

1995 439,000 1996 455,000 1997 455,000 1998 458,900 100 1999 467,000 1,000 2000 471,500 3,500 2001 415,000 5,200 2002 373,350 22,300 2003 307,950 65,500 2004 268,000 107,000 2005 174,000 262,500 2006 75,000 414,000 2007 17,500 553,000 2008 10,000 680,000 2009 3,000 747,000 2009 - 865,000 Table 16: CRT and FPD TV Sales. Source – SCEA

Annexe 108

Annexe 3: CRT Glass Collection by SWICO Recycling

Year CRT Glass [tonnes] CRT Monitors CRT TVs

1994 60 1995 655 1996 836 1997 910 1998 1,522 1999 1,927 2000 2,397 2001 2,840 2002 3,585 896 2003 4,395 1,465 2004 4,804 2,059 2005 5,032 3,354 2006 4,804 3,925 2007 4,633 4,436 2008 3,088 4,279 2009 3,453 6,440 2010 2,375 10,021 2011 2,304 10,937

Table 17: CRT Glass Collection by SWICO. Source: SWICO Annual Reports/ EMPA

Annexe 4: LCD Monitor Collection by SWICO Recycling

Year SWICO Collection - LCD Monitors [units]

2006 79,000 2007 85,283 2008 140,125 2009 312,844 2010 348,759 2011 375,520 Table 18: LCD Monitor Collection by SWICO. Source: SWICO Annual Reports/ EMPA

Annexe 109

Annexe 5: TV Permits in Switzerland

Year # TV Permits ['000] 1954 4 1968 1,011 1973 1,627 1987 2,289 1993 2,560 2000 2,650 2005 2,885 2006 2,885 2007 2,928 2008 2,972

Table 19: TV licences issued by Billag. Source: Billag

Annexe 6: Household TV Ownership in Switzerland

Year % Households with at least 1 TV

1990 85.77% 1998 91.72% 2000 92.52% 2001 93.33% 2002 93.60% 2003 94.30% 2004 93.70% 2005 94.20%

Table 20: Household Ownership of TVs. Source: BfS

Annexe 7: Swiss population of households

Year # Households ['000] 1960 1,594.0 1970 2,062.4 1980 2,459.3 1990 2,859.8 2000 3,181.6

Table 21: Number of households. Source: BfS

Annexe 110

Deepali Sinha Khetriwal 11, St.Peter’s Court, NW4 2HG,

London, United Kingdom

Academic Background

10/ 2005 –07/2012

University of St.Gallen, St.Gallen, Switzerland Doctoral Program in International Management Dissertation Diffusion, Obsolescence and Disposal of End-of-Life Consumer Durables: Models for Forecasting Waste Flows

10/ 2002 – 10/ 2004

University of St.Gallen, St.Gallen, Switzerland Master of International Management Master thesis: The Management of Electronic Waste: A Comparative Study on India And Switzerland

08/ 2003 – 12/ 2003

Indiana University, Bloomington, IN, USA Exchange Semester - Kelly School of Business

07/ 1998 – 05/ 2000

University of Pune, Pune, India Master of Arts in Economics - Department of Economics, University of Pune

07/ 1995 – 05/1998

University of Pune, Pune, India Bachelor of Arts in Economics - Nowrosjee Wadia College, Pune

Publications Journal Articles Khetriwal, D.S., (with R. Widmer, R. Kuehr, J. Huisman) (2011). “One WEEE, many species: Lessons from the European Experience” in Waste Management and Research in Waste Management and Research., Volume 29, Issue 9, Pages 954-962. Khetriwal, D.S., (with I. First) (2010) “Exploring the Relationship Between Environmental Orientation and Brand Value: Is There Fire or Only Smoke?” in Business Strategy and the Environment, Volume 19, Issue 2, Pages 90-103. Khetriwal, D.S., (with R. Widmer, P.Kraeuchi) (2009) “Producer Responsibility for E-Waste Management: Key Issues for Consideration - Learning from the Swiss Experience” in Journal of Environmental Management, Volume 90, Issue 1, Pages 153 – 165. Khetriwal, D.S., (with P.Krauechi, M. Schwaninger) (2005) “A comparison of electronic waste recycling in Switzerland and in India” in Environmental Impact Assessment Review, Volume 25, Issue 5, Pages 492-504.

Curriculum Vitae 111

Khetriwal, D.S., (with R. Widmer, H.Oswald-Krapf, M. Schnellmann, H. Boeni) (2005) “Global perspectives on e-waste” in Environmental Impact Assessment Review, Volume 25, Issue 5, Pages 436-458. Refereed Conference Papers Khetriwal, D.S (with R. Widmer, L. Hilty, M. Schwaninger) (2012) “Application of System Dynamics to Assess Mass Flows of Waste Electrical and Electronic Equipment (WEEE)” presented at the 30th International Conference of the System Dynamics Society, July 2012, St.Gallen, Switzerland. Khetriwal, D.S (with I. First) (2011) “Enabling Closed Resource Loops In Electronics: Understanding Consumer Disposal Behaviour Using Insights From Diffusion Models” presented at the 22nd Cromar Congress, October 2011, Pula, Croatia. Khetriwal, D.S (2011) “Consumption and Obsolescence: The Consumer Link to Sustainable Global Electronic Product Chains” presented at the 17th Annual International Sustainable Development Research Conference, May 2011, New York, USA. Khetriwal, D.S (with R. Widmer, R. Kuehr, J. Huisman) (2008) “One WEEE, Many Species” – Lessons from Europe” presented at ISWA Congress, September 2008, Singapore. Khetriwal, D.S “Technological Substitution and its End-of-Life Impact: A Case of the Television” presented at the Electronics Goes Green 2008+, September 2008, Berlin, Germany. Khetriwal, D.S (with I. First) (2007) “The Influence of Environmental Orientation on Brand Value: a Case of the Electronic & Electrical Equipment Industry presented at the 2nd IIMA Conference on Research in Marketing, Ahmedabad, India, 2007. Others Khetriwal, D.S (with K. Wankhede, S.Sinha) (2007) “Mumbai: Choking on E-Waste - A study on the status of e-waste in Mumbai” by Toxics Link, New Delhi


Professional Experience

Professional Qualifications and Memberships Certified EMS Auditor – IEMA Approved Advanced Environmental

Management AIEMA – Associate Member Institute of Environmental Management and

Assessment Systems Auditor

03/ 2009 – Ongoing

United Nations University, Bonn, Germany Research Associate, London Conceptual and strategic inputs to capacity building programs

under the aegis of the StEP Initiative Lead the development of the summer school concept, structure and

programme 09/ 2010 – Ongoing

SOFIES Consulting, Geneva, Switzerland Associate Consultant, London Overview of legislative and regulatory institutional setup and

stakeholder analysis for e-waste management in India Identification of relevant environmental legislation and impact

analysis of compliance requirement on producer Ongoing tracking and assessment of developments regarding E-

waste legislation and stakeholder actions 01/ 2006 – 06/ 2006

UNEP (United Nations Environment Programme), Paris, France Project Consultant, E-waste in India, Mumbai Liaised with international & local experts, government officials

and other stakeholders in e-waste area Identified potential local implementation partners and developed

Terms of Reference 02/ 2005 – 12/ 2008

EMPA (Swiss Federal Laboratories for Material Testing & Research), St.Gallen, Switzerland Consultant, Mumbai/ London Knowledge transfer and capacity building through web and

academic publishing Conducted rapid market assessment studies on e-waste generation

and recycling, especially in Mumbai Provided support for evidence based policy and supported

development of e-waste related legislation Initiated and maintained relationships with government, industry

and other stakeholders for assessments, pilot case study, trainings, factory visits etc.


diffusion, obsolescence and disposal of endof--life ...file/… · diffusion, obsolescence and...

Documents