Journal of Transportation Management

Volume 27 | Issue 2 Article 3

1-1-2017

Using an analytic network process model toincorporate qualitative factors into multi-criteriaglobal modal choice decisionsTyler T. PrevettUSTRANSCOM, ttprevett@gmail.com

Bradley E. AndersonBall State University, beanderson@bsu.edu

Follow this and additional works at: https://digitalcommons.wayne.edu/jotm

This Article is brought to you for free and open access by DigitalCommons@WayneState. It has been accepted for inclusion in Journal ofTransportation Management by an authorized editor of DigitalCommons@WayneState.

Recommended CitationPrevett, Tyler T., and Anderson, Bradley E. (2017). Using an analytic network process model to incorporate qualitative factors intomulti-criteria global modal choice decisions. Journal of Transportation Management, 27(1), 7-21. doi: 10.22237/jotm/1498867320

Winter 20177

USING AN ANALYTIC NETWORK PROCESS MODEL TOINCORPORATE QUALITATIVE FACTORS INTO MULTI-CRITERIA

GLOBAL MODAL CHOICE DECISIONS

Tyler T. PrevettUSTRANSCOM

Bradley E. AndersonBall State University

ABSTRACT

This research develops and evaluates an Analytic Network Process (ANP) model to choose the correctmode of global transportation in the presence of complicating qualitative influences. The ANP modeleffectively combines important qualitative and quantitative factors into a global modal choice model.Although there is a great deal of research in the area of modal choice, the research often focuses singularlyon cost or time factors. This research incorporates security, public opinion, and customer opinion intomodal choice. One of the most difficult choices a transportation planner faces is deciding when qualitativefactors outweigh the quantitative ones. A reliable tool to validate choice by including the importantqualitative factors with the quantitative is quite valuable in military operations, humanitarian support, anddisaster relief.

INTRODUCTION

Modal choice for global transportation requirementsis a complex decision. Generally, sealift is slower,but more cost-effective, while airlift is faster andmore expensive. The decision to move somethingby air or sea is influenced not only by cost andtechnical limitations, but also by qualitative factors.This seemingly simple decision is complicated by amyriad of influential factors.

Most transportation research has focused onmeasurable metrics that address such factors ashow much it costs to move something, whether thecargo arrives on time, the volume of cargo moved,or the number of items moved. This research isuseful, but sometimes falls far short of what’sneeded, as it neglects to consider the elements ofthe decision that aren’t dependent on numbers.Examples of ‘qualitative’ factors include pressurefrom public opinion, urgent need of materials forhuman survival and the political message conveyed.One only needs to look at the news, both historicaland modern day, to see how political, social,environmental, and public administration

considerations affect the execution of manyimportant endeavors.

Construction of the Trans-Alaskan pipeline is a vividhistorical example of how a failure to consider thepeople and the environment, despite a national needfor the oil, would cost billions of dollars and years ofprogress (Coates, 1991). A more recent exampleof choosing between 2 important alternatives in theface of many complicating factors is phase 4construction of the Keystone Pipeline, also knownas the Keystone XL Pipeline. The Keystone XLproposal faced criticism from environmentalists anda minority of the members of the United StatesCongress. In January 2012, President BarackObama rejected the application amid protests aboutthe pipeline’s impact on Nebraska’s environmentallysensitive Sand Hills region (BloombergBusinessWeek, 2012). The debate on building thepipeline rages on. These are just 2 examples of thedifficulty that looms when many qualitative factorsare involved. These factors are common in militarytransportation, humanitarian assistance and disasterrelief efforts, but decisions must be made muchfaster, as lives are at state. There has not been a

Journal of Transportation Management8

systematic examination of qualitative factors in thearea of global modal choice, yet these factors cangreatly influence mode choice.

In order to address the problem of choosing thecorrect mode of global transportation in thepresence of complicated qualitative influences, thisresearch formulates an Analytic Network Process(ANP) model that effectively and efficientlycombines the qualitative and quantitative factors intoa single global modal choice model. The goal of themodel is to get the required equipment to itsdestination at the required time, but to do it at theminimum total cost, while addressing the qualitativevariables as best as can be done. The importanceof incorporating qualitative factors into atransportation decision model is furtherdemonstrated through the details of the HighMobility Multipurpose Wheeled Vehicle(HMMWV) case used to validate the developedmodel.

We will first use a Literature Review to show thisresearch’s position within the existing body of work,

as well as to justify this research and themethodology. The development and use of theANP model follows the literature review in the orderpresented in Figure 1. This figure shows both theflow and order of the discussion as well as themethodology used to create and utilize the ANPmodel. All of the inputs to the ANP Model aredescribed in the Research Design (Building the ANPModel) section. ANP Model Implementation isdescribed in the following section, and ModelBuilding Results are shown in the next section. Thepaper concludes with the HMMWV Case andRecommendations for Future Research.

LITERATURE REVIEW

Perhaps the two most thorough reviews of modalchoice research were conducted by McGinnis(1989) and Meixell & Norbis (2008). However,none of the examined models consider qualitativeaspects, nor do they integrate experts’ opinion intothe decision.

Winter 2017 9

Some research efforts have attempted to identify thequalitative criteria that influence the freight modechoice like Brooks (1990), Lu (2003) and Jeffs andHills (1990). Jeffs and Hills (1990) formulated afactor analysis of the determinants applicable tofreight movement in the United Kingdom. Theyhighlight six categories of variables that influence thedecision, but they were unable to formulate anoverall decision model that incorporated thecomplex interactions among the variables, as isneeded for an accurate decision to be made inresearch such as this. Without modeling thoseinteractions, accuracy of the results is not assured.Jeffs and Hills (1990) recommended future researchfocus on the specific situation, characteristics andneeds of the organization. Their view of evaluatingdeterminants of the mode choice with anorganizational-specific model is shared by Young,Richardson, Ogden, and Rattay (1982).

In their examination of behavioral influence onfreight mode choice criteria, Bolis and Maggi(2003) conclude that the mode choice must be inalignment with the organization‘s overall logisticsstrategy. The authors interviewed 4 logisticsmanagers within an Adaptive Stated Preferenceexperiment. While this is a small sample size, Bolisand Maggi (2003) argue that because of theexpertise of these managers in the transportationsystem, their inputs can yield valid results. Thispaper utilizes a similar argument, but uses a greaternumber of interviews that better encompass theentire transportation system. Bolis and Maggi(2003) present the notion that an organization’soverall logistics strategy will impact how importantdifferent criteria are to the decision maker in anygiven situation, which is also key to this research.

While existing research focuses on rail vs. truckmodalities, Bergantino and Bolis (2004) consideredthe behavioral elements associated with a maritimemode choice. The study again shows that theparticular situation and strategy of the customerplays an important role in determining applicablecriteria for the mode choice. Their research relies oncomparisons between the criteria, but also does notaccount for interactions among the criteria in a

systematic way. Therefore, it is not particularlyuseful for a high visibility global mode choice.

The ability to consider both quantitative andqualitative data in a systematic way is a particularstrength of an Analytic Hierarchy Process (AHP).Liberatore and Miller (1995) considered the modechoice between sealift and airlift using the AHPmethodology with a focus on the entire logisticsstrategy of the focal firm. The authors note that theirresearch represents the first time that AHP had beenspecifically applied to mode choice (Liberatore andMiller, 1995). They conclude that AHP offers acomprehensive, yet flexible methodology foraddressing transport carrier and mode selectionproblems (Liberatore and Miller, 1995). Hundredsof researchers have used this methodology to modeland make real-world decisions, such as choosing asubway layout in Istanbul or determining anappropriate mix of advertising media (Saaty, 2000).

The AHP decision model approach has also beensubject to some criticism. Some authors havevoiced objections to the model’s mathematical andtheoretical base, arguing that relative comparisonscan be arbitrary, and that the higher and lowercriterion of the decision can have interdependencies(Dyer, 1990; Harker and Vargas, 1987). The basisof this criticism is that the AHP only allowsunidirectional influence along the hierarchicalrelationships (Cheng, Li, and Yu, 2005). Anotherlimitation of the AHP is that the criteria at any givenlevel of the hierarchy are considered to be mutuallyindependent (Büyükyazici and Sucu, 2003). Whilethis may be true in some cases and may be assumedin some cases to simplify the decision model, real-world problems are seldom so simple. Saaty (1999)points out that assuming independence unnecessarilylimits interactions among elements within the samelevel and at different levels of the hierarchy. Anothercriticism of the model is that the decision maker canintroduce inconsistencies into the model. Forexample, suppose in a car purchase example thebuyer said that speed was more important thanstyle, that style was more important than cost, butthat cost was more important than speed. This is thecircular inconsistent logic of A>B, B>C, but C>A.

Journal of Transportation Management10

To properly deal with these criticisms, the decisionmodel developed here has been generalized toinclude interdependencies between internal elementsand feedback between hierarchical criteria levels ofthe decision. This generalization of the AHP istermed an Analytic Network Process (ANP) model(Saaty, 2001). Feedback and inner dependence areincluded by way of another matrix transformation onrelative priorities. Inner dependence is observedwhen an element within a cluster influences anotherelement within the same cluster. A cluster is a groupof closely related model elements, like the sub-criteria of an overall criterion or the alternatives tobe considered. This model is also seeing wideusage in academic and professional arenas to assistin decision making (Coulter and Sarkis, 2005;Cheng and Li, 2005; Shang, Youxu, and Yizhong,2004; Lee and Soung, 2000).

The ability of ANP to model inner dependence andfeedback within a decision hierarchy makes it anappropriate model for mode choice. Ali Görenerprovides significant proof and further validation thatANP is the proper and most suitable method to use.His comparison of ANP and AHP in amanufacturing setting shows that there are significantdifferences between AHP and ANP outcomesderived from interdependencies, outerdependenciesand feedbacks (Görener, 2012). In Rozann Saaty’spaper “A Validation of the Effectiveness of InnerDependence in an ANP Model,” she shows at eachstep the results are nearer what we know occurs inthe real world. Using a direct comparison of theAHP and ANP models, this validation shows thatusing feedback and dependence in an ANP modelcan get us closer to reality (Saaty 2013).

However, a review of the literature revealed noprevious research of modal choice using the ANPmethodology.

RESEARCH DESIGN(BUILDING THE ANP MODEL)

To build an ANP model, we begin with the threedistinct steps of problem identification (Forman andSelly, 2001). These steps are - 1) defining theproblem, 2) identifying alternatives and 3)

researching the alternatives. The first two steps arestraightforward, as our problem is in choosing thecorrect mode of global transportation, and thechoices are airlift or sealift. Researching thealternatives for step 3 is quite challenging andreflects most of the effort and reason behind thisresearch. ANP was chosen as the modelingmethodology for the third step, due to its capacity toincorporate complex interactions among criteria andto capture qualitative and quantitative variables.After the three problem identification steps areproperly addressed, the ANP model is constructedfor determining global modal choice.

To build the model, data is incorporated from theliterature and personal interviews with transportationexperts in different regions. Existing literature andinterviews are used to determine appropriatecriterion to include in the decision model, as well astheir relative importance. Subjects for theinterviews were selected by contacting the U.S.Transportation Command for experts in thematching of sealift or airlift assets to a movementrequest. Ten subject matter experts, of thoserequested, agreed to be interviewed. Theinterviewees consist of both service providers andcustomers in different regions of the world and allfour service branches of the U.S. military (Army, AirForce, Navy and Marines). Experience rangesfrom a senior leader with 34 years of logisticsexperience to an interviewee with seven monthsexperience in coordinating movement requests. Allinterviewed subjects have detailed knowledge oftransportation functions and limitations and areexperienced in both movement requests and modeselections. The subjects collectively haveexperience in all major regions of the world.

Interviews were conducted with subject matterexperts during which each was asked seven primaryand very open ended questions as follows:

1. Where do you work? How long have youworked in the transportation system? Whatdo you do in the strategic Airlift/sealiftsystem?

2. How does a movement request happen?3. What do you perceive as the major criteria

to consider when requesting sealift or airliftas a mode of transportation?

Winter 2017 11

4. How often does the lift occur exactly asrequested? Can you recall any cases whereone mode was requested, but a differentmode was used?

5. From your perspective, what are a fewmajor Criteria that complicate the modalchoice?

6. What could be done to improve thedecision making process?

7. Is there someone else you couldrecommend I talk to about these kinds ofissues?

An affinity diagram procedure was used to organizethe ideas using the following steps (Brassard, 1989):

Step 1 - Generate Ideas. The survey questionswere used to generate a list of ideas. Interviewswere recorded and subsequently transcribed to text.All subjects were advised that the interview wouldbe recorded and that their identities would remainconfidential.

Step 2 - Display the Ideas. The ideas wereposted randomly on a table. Each interview wasdecomposed into discrete statements of the criteriathat influence the mode decision, resulting in 240+individual statements. Each statement represented aspecific criterion and element applicable to themode choice, and could also indicate interactionwith other elements.

Step 3 - Sort the Ideas into Related Groups.The researchers physically sorted the individualstatements and they were then reorganized intorelated concepts and groupings, using the followingprocess:

Start by looking for two ideas that seemrelated in some way. Place them together ina column off to one side.Look for ideas that are related to thoseyou’ve already set aside and add them tothat group.Look for other ideas that are related to eachother and establish new groups.

Step 4 - Create Header Cards for the Groups.A header is an idea that captures the essential link

among the ideas contained in a group of cards. Thisidea is written on a single card or post-it and mustconsist of a phrase or sentence that clearly conveysthe meaning. The researchers developed headers forthe groups by:

Finding already existing cards within thegroups that will serve well as headers andplacing them at the top of the group ofrelated cards.Alternatively, discussing and agreeing on thewording of cards created specifically to beheaders.Discovering a relationship among two ormore groups and arranging them in columnsunder a superheader. The same rules applyfor superheaders as for regular headercards.

Step 5 - Draw the Finished Affinity Diagram.Write a problem statement at the top of thediagram.Place header and superheader cards abovethe groups of ideas.Review and clarify the ideas and groupings.Document the finished Affinity Diagram.

Once the cards have been sorted into groups, largeclusters are sorted into subgroups for easiermanagement and analysis. The result of the affinitydiagramming procedure is a cause and effectdiagram. Once the statements are transformed intothe needed criterion, elements, and interactions thatrepresent policy, practice, and priorities, they are allinputted to the SuperDecisions 1.6.0 softwareselected for this task and utilized for themathematical formulation of the mode choice. Datafor needed mode decisions are inputted to themodel for recommendation as discussed later in theHMMWV case study.

ANP MODEL IMPLEMENTATION

Now that we have developed our ANP model, wediscuss the steps the model goes through inmodeling a decision and presenting arecommendation. SuperDecisions 1.6.0 softwareis utilized for the mathematical formulation of themodel in this research, where the pairwise

Journal of Transportation Management12

judgments and synthesis are performed via graphicaluser interface (Super Decisions, 2006). There arethree basic steps to modeling a decision using AHPthat are also applicable to ANP: 1) decomposition,2) comparative judgment, and 3) synthesis (Saaty,1990; Büyükyazici and Sucu, 2003).

DecompositionDecomposition of the decision problem is verysimilar between ANP and AHP and is the first phaseof AHP model development. Since ANP breaksdown the traditional one-way influence of thehierarchy, it is graphically represented differently asseen in the following figure as adapted fromBüyükyazici and Sucu (2003).

There are several important differences in this ANPnetwork model. The first is terminology. Thenumerous criteria and alternatives depicted in themodel are represented by the multiple clusters andthe sub-criterion are called elements. When anelement within a cluster influences another elementwithin the same cluster, this relationship is calledinner dependence and is represented by the arrowlooping back to the same cluster as shown above.Similarly, when an element from one clusterinfluences an element from another cluster, thisrelationship is called outer dependence and isrepresented by the arrows between clusters. Whenthe characteristics of one of the alternativesinfluences a criterion or sub-criterion, thisrelationship is called feedback and is represented

by an arrow moving from the alternatives cluster toa criterion cluster. In this way, the importance of thecriteria and their sub-criteria not only influence thepriority of the alternatives, but also influence thepriority of the various criteria and other elementswithin the model. Saaty (1990/1997) helps usunderstand these relationships with a couple ofexamples.

“In inner influence one comparesthe influence of elements in a groupon each one. For example if onetakes a family of father mother andchild, and then take them one at atime say the child first, one askswho contributes more to the child’ssurvival, its father or its mother,itself or its father, itself or itsmother. In this case the child is notso important in contributing to itssurvival as its parents are. But if wetake the mother and ask the samequestion on who contributes to hersurvival more, herself or herhusband, herself would be higher,or herself and the child, againherself. Another example of innerdependence is making electricity.To make electricity you need steelto make turbines, and you needfuel. So we have the electricindustry, the steel industry and thefuel industry. What does the

Winter 2017 13

electric industry depend on more tomake electricity, itself or the steelindustry, steel is more important,itself or fuel, fuel industry is muchmore important, steel or fuel, fuel ismore important. The electricindustry does not need its ownelectricity to make electricity. Itneeds fuel. Its electricity is onlyused to light the rooms, which itmay not even need. If we thinkabout it carefully everything can beseen to influence everythingincluding itself according to manycriteria. The world is far moreinterdependent than we know howto deal with using our existing waysof thinking and acting. The ANP isour logical way to deal withdependence.”

This powerful network model can also representextremely complex decision making by including asmany clusters and elements as required for theobjective, and each cluster can also entirely containsub networks of criteria. The depth and complexityof the network model is limited only by the needs ofthe decision maker. The network model can also beexpanded to incorporate different dimensions of thedecision, each of which corresponds to a givenpreference for an alternative based on an overallenvironment. For example, in estimating marketshare one might evaluate different alternatives basedon the risk, opportunity, costs, or benefits eachalternative presents (Saaty, 2003). These higherlevel dimensions are called control hierarchies, andrepresent a separate network of the same criteriafor each dimension being considered (Saaty, 1999).The control hierarchies can be viewed as importanthigher level aspects in choosing between the modelalternatives identified by the decision maker andintegrated via the questions asked to gather theneeded data. Saaty shows some example questionson bringing in data for a control hierarchy:

“For benefits and opportunities, askwhat gives the most benefits orpresents the greatest opportunity toinfluence fulfillment of that control

criterion. For costs and risks, ask whatincurs the most cost or faces thegreatest risk. Sometimes (very rarely),the comparisons are made simply interms of benefits, opportunities, costs,and risks in the aggregate without usingcontrol criteria and subcriteria.”

In this transportation model the control hierarchyencompasses the entire model represented by theoverall objective of “Sealift or Airlift for GlobalModal Choice?”

In this research, data is drawn from interviews withsubject experts in the military distribution system,and it is compiled to determine appropriate criteriato include in the model. Decomposition of thedecision through interviews has been used by manyresearchers such as Bolis and Maggi (2003). Afterdecomposition is complete, we begin the secondphase of AHP building, which is comparativejudgment.

Comparative JudgementThe comparative judgment phase of ANP isessentially the same as AHP. Each cluster iscompared in a pairwise fashion relative to itsimportance to the objective. Similarly, pairwisecomparisons of each element within a cluster arealso constructed using the same qualitative andquantitative methods as described with AHPmodels. The comparisons are done to establish theirrelative importance to weight the correspondingblocks of the supermatrix and make it columnstochastic. A cluster impacts another cluster when itis linked from it, that is, when atleast one node in the source cluster is linked tonodes in the target cluster. The clusters linked fromthe source cluster are pairwise compared for theimportance of their impact on it with respect tomode choice, resulting in the column of priorities forthat cluster in the cluster matrix. The process isrepeated for each cluster in the network.

Relative comparisons for all of the information in thedecision can be interpreted as pairwise judgmentsand included in the overall decision model. InSaaty’s SuperDecisions 1.6.0 software, this

Journal of Transportation Management14

pairwise comparison can be made by pie chart, barchart, questionnaire, or input directly into a pairwisecomparison matrix (Super Decisions, 2006).

A node is a sub-division of a cluster and typicallyrepresents the lowest level of input data that isexamined. If customers is a cluster, then age,gender, and salary could be used as nodes. After allthe nodes are created, nodes are chosen and linkingto the other nodes in the model that influence it. The“children” nodes will then be pairwise comparedwith respect to that node as a “parent” node. Anarrow will automatically appear going from thecluster the parent node cluster to the cluster with itschildren nodes. When a node is linked to nodes inits own cluster, the arrow becomes a loop on thatcluster and we say there is inner dependence (Saaty,1990/1997).

As an example, Figure 3 shows a notional modelnode that indicates that the type of cargo being

transported “very strongly” favors the use of sealift.The reason for the strength of the pairwisecomparison might be that the commodity isextremely dense or heavy and might not be able tobe transported by air. Within the software, a similarcomparison is accomplished for each sub-criterionwith respect to the alternatives.

SynthesisAfter the second phase of AHP development iscomplete, we can begin the third and final stage ofmodel development, known as Synthesis. Thesynthesis phase of ANP is much different than inAHP. In ANP, a supermatrix of the pairwisecomparisons is constructed. This supermatrixconsists of several partitions or sub-matrices whichtake into account the impact of elements on eachother (Büyükyazici and Sucu, 2003). Thesupermatrix is organized with each element of themodel occupying a column and a row. Thesecolumns and rows are grouped by their parent

Winter 2017 15

cluster. In this way, there is a representation forhow much each element of the model influencesevery other element of the model. Of course,elements can also exert no influence on each other,and a ‘0’ is entered in the matrix where these twoelements intersect. The intersection of two clustersof elements represents a single pairwise comparisonmatrix, and the supermatrix represents thecompilation of all the priorities derived frompairwise matrices (Büyükyazici and Sucu, 2003).An important point is that the supermatrix must becolumn stochastic (Saaty, 1999). This means thatthe sum of each column within the supermatrix mustsum to one so that the supermatrix converges whenraised to an acceptably large power (Büyükyaziciand Sucu, 2003). This can be done by multiplyingthe values within the sub-matrix by the relativeweight of their interaction. Saaty (1999) representsthe supermatrix using the notation: W is thesupermatrix, N represents the number of clusters,CN represents each cluster, eNn represents eachelement of the model, and WNn represents theappropriate sub-matrix weight.

The desirability of an alternative can then becomputed in a similar fashion to AHP whileincorporating the effects of dependence between theelements of the model. While Saaty uses matrixrepresentation for deriving the priority ofalternatives, this process can also be representedmathematically. The following summation wasadapted from the work of Meade and Sarkis(1999) to represent the needed ANP relateddecision variables and represents the desirability ofan alternative for a given control hierarchy.

(1)

Where:Di is the desirability of alternative i.Pj is the relative importance weight of the

criterion j for the control hierarchy,D

kjW is the relative importance weight forelement k of criterion j fordependency between componentlevels of the model.

IkjW is the stabilized relative importance weight as

determined by the supermatrix for element k ofcriterion j for interdependency relationshipsbetween elements of the model.Sikj is the relative preference of alternative i withrespect to element k under criterion j.Kj is the index set of elements for criterion j, and Jis the index set for all criterion.The synthesis step is accomplished after entering allthe pairwise comparisons. The relative priority ofthe alternatives will be displayed graphically and interms of the raw derived priorities.

MODEL BUILDING RESULTS

The following table summarizes the model elementsof the developed modal choice model.

Criteria 1.0 represents the alternatives available tothe decision maker, and criteria 2.0 to 7.0 representelements of the decision. This procedure resulted inthe identification of six main criteria elements and 28sub-criteria elements for consideration in modalchoice decisions (not including the airlift and sealiftalternatives in this total). Throughout the interviewprocess, it became apparent that criteria wereindeed highly influential to each other. For example,the availability of aircraft was influenced to a highdegree by worldwide global demand for airlift.These various relationships are captured byincluding inner dependence and outer dependenceloops within the decision network. Theinterdependencies (both inner and outerdependence) can be seen in the following table bythe many sub-criterion that influence other sub-criterion.

Table 2 shows that the subjects indicated a highdegree of interaction among the elements of themodel. The left most column indicates the elementthat has influence over the criteria to the right.Some elements such as Leader’s Preferences andPlatform Availability have a widespread influenceover many other elements in the model. Thesecriteria, sub-criteria and their dependencies were allentered into Super Decisions 1.6.0 (Super

Journal of Transportation Management16

Decisions, 2006) software to arrive at an overallnetwork model.

Now that the decision model has been developed, itis necessary to evaluate the model. We will examinea specific case to validate the developed model.Complete data for a situation such as is modeledhere is not easy for everyone to obtain, as manyfactors we consider are often either not collected orthe data is not freely distributable. We were able toobtain needed data for validation and used the veryrelevant Department of Defense case of transporting

up-armor kits and armored High MobilityMultipurpose Wheeled Vehicles (HMMWVs).

HMMWV CASE(MODEL VALIDATION)

On December 25, 2003 The New York Timespublished the story “Army Stepping Up ItsHumvee Orders For Troops in Iraq” where theyexpound on how the U.S. Army sent out anurgent call for armored HMMWVs, realizing thatit had not ordered enough to protect its troops.In April 2005, the Government Accountability

Winter 2017 17

Office published a report to CongressionalCommittees identifying nine commodities thatwere subject to systemic deficiencies and fivereasons the shortfalls were realized. Ineffectivedistribution of armored HMMWVs and up-armored kits was cited as one of the prevalentsystemic deficiencies (GAO-05-275, 2005).

Sealift is generally about 1/10th the cost of airlift,and it is the obvious choice for bulk and heavyassets to be moved, especially over long distances.However, monetary considerations were consideredless important because armor kits and vehicles wereconsidered critical to troop survival. News storiesof American casualties due to roadside bombings ofHMMWVs without armor led to heavy politicalinfluence on mode choice. Security and timeconsiderations favor the use of airlift. Geographical

Journal of Transportation Management18

influence on mode choice is mixed. While the itemsare heavy and the distance to move them is great,which would favor sealift, the required location forthese items is inland, which would favor airlift.Another important factor in this decision is thatduring the initial stages of the transportation problemthere were very few armor kits to aggregate to anentire shipload for transportation.

The following table summarizes the data for sub-criteria preference with respect to the alternatives:

The table represents preference for sealift or airliftbased on the HMWVV data and can be interpretedas the relative importance of one criteria overanother with respect to a given criteria. Forexample, in the 2.0 Costs table in Table 3 thebolded entry indicates that operational requirementsare 2 times as important as system limitations withrespect to the costs criteria. When a criterion doesnot have an impact on another criterion, it is notincluded in the pairwise comparison.

Once the steps of ANP are completed, aspreviously discussed, the resulting vector is obtainedfrom the HMMWV data and presented in Table 4.

The bolded priorities within the table for airlift andsealift match the priorities derived using the SuperDecisions 1.6.0 software, and represent apreference for using airlift for mode choice using theHMMWV case data (Super Decisions, 2006). Thederived priorities for the individual sub-criteriapresented in the table indicate the amount ofinfluence each of these elements had in identifyingairlift as the mode of choice. Each of the elementsgrouped in their main criteria display the amount ofinfluence each main criteria had in the overall modechoice.

HMMWV CASE CONCLUSIONS

Overall, the developed ANP decision model showsa relative preference for airlift to deliver theHMMWV armor requirements. In practice, this isthe actual outcome chosen by decision makers. In a

hearing of the House Armed Services Committee,Congressman Hunter relayed that even much later inthe conflict airlift delivered the majority of level 2armor due to the “extreme importance to ourwarfighters” (Hunter, 2006).

Although the choice of airlift or sealift is a seeminglysimple binary choice, it is anything but simple. Oneof the most difficult choices a transportation plannerfaces is deciding when qualitative factors outweighthe quantitative ones. Having a reliable tool tovalidate that choice by including the importantqualitative factors with the quantitative is quitevaluable.

Winter 2017 19

FUTURE RESEARCH

The presented decision model was built using theinputs of 10 experts in the mobility system. Onearea of future research that could add validity to themodel is to evaluate the accuracy and completenessof the model through additional case studies or byusing a Delphi Methodology. Additional casestudies would further validate the model, as would afollow-up survey of all subjects or examining theinputs of new experts. A survey could easily be builtusing WebSurveyor Desktop 4.1, and containLikert scale ratings of each criteria and sub-criteriawith respect to how frequently it impacts the modalchoice, and how important it is to the modal choice.

CONCLUSSIONS

This research presents a unique decision model aswell as a unique method of developing mode choiceusing ANP. Tools such as this decision model andother initiatives serve to aid decision makers byallowing them to make more thoroughly informeddecisions in a systematic way that includes bothquantitative and qualitative inputs not included inprevious modal choice models. This multi-criteriaintegrated methodology and modeling techniquecould be applied to other transportation problemsthat require an important degree of qualitative factordecision making integration. This could includeother military operations, humanitarian assistance/logistics, or disaster relief where many qualitativefactors need to be considered and priorities are notsolely cost based.

REFERENCES

Bergantino, A., and S. Bolis (2004), “An AdaptiveConjoint Analysis of Freight Service Alternatives:Evaluating the Maritime Option”, Chapter 10 inAura Reggiani and Larie Schintler (eds.), Methodsand Models in Transport andTelecommunications: Cross-AtlanticPerspectives., Berlin: Springer-Verlag.

Bhushan, N. and K. Rai (2004), StrategicDecision Making: Applying the AnalyticHierarchy Process, CREAX InformationTechnologies Pvt. Ltd., ISBN: 1-85233-756-7.

Bloomberg BusinessWeek, “TransCanada Wins asObama Keystone Permit Seen”, 2012-10-08.

Bolis, S., and R. Maggi (2003), “Logistics Strategyand Transport Service Choices: An Adaptive StatedPreference Experiment,” Growth and Change,34(4): 490-504.

Brassard, M. (1989), “The Memory Jogger Plus+:Featuring The Seven Management Planning Tools”,Chapter 1, Affinity Diagram, Methuen, MA: Goal/QPC.

Journal of Transportation Management20

Brooks, Mary R. (1990), “Ocean Carrier Selectionin a New Environment,” Logistics andTransportation Review, 26(4): 339-355.

Büyükyazici, M. and M. Sucu. (2003), “TheAnalytic Hierarchy and Analytic NetworkProcesses,” Hacettepe Journal of Mathematicsand Statistics, (32): 65-73.

Cheng, E., and H. Li. (2005), “Analytic NetworkProcess Applied to Project Selection,” Journal ofConstruction Engineering and Management,131(4): 459-466.

Cheng, E, H. Li, and L. Yu. (2005), “The AnalyticNetwork Process (ANP) Approach to LocationSelection: a Shopping Mall Illustration,” Journal ofConstruction Innovation, (5): 83-97.

Coates, Peter A., The Trans-Alaska PipelineControversy: Technology, Conservation, and theFrontier., Lehigh University Press ; London ;Cranbury, NJ : Associated University Presses,c1991.

Coulter, K., and J. Sarkis (2005), “Development ofa Media Selection Model Using the AnalyticNetwork Process,” International Journal ofAdvertising, 24(2): 193-215.

Dyer, J. (1990),”Remarks On The AnalyticHierarchy Process,” Management Science, 36(3):249-258.

Forman, E. and M. Selly (2001), “Decision ByObjectives: How To Convince Others That You AreRight,” World Scientific.

Görener, Ali (2012), “Comparing AHP and ANP:An Application of Strategic Decisions Making in aManufacturing Company,” International Journalof Business and Social Science, 3(11): 194-208.

Harker, P. and L. Vargas (1987), “The Theory ofRatio Scale Estimation: Saaty’s Analytic HierarchyProcess,” Management Science, 33(11): 1383-1403.

Hunter, D. (2006), R-CA, Chairman of the HouseArmed Services Committee. ̄ House ArmedServices Committee Holds Hearing on FY2007Budget: Transportation Command and ComponentCommands” Congressional Hearing Transcripts,Washington DC.

Jeffs, V. and P. Hills (1990), “Determinants ofModal Choice in Freight Transport: A Case Study.”Transportation, 17(1): 29-47.

Shang, J., Y. Tjader, and Y. Ding (2004), “A UnifiedFramework for Multicriteria Evaluation ofTransportation Projects,” IEEE Transactions onEngineering Management, 51(3): 300.

Lee, J. and K. Soung Hie (2000), “Using AnalyticNetwork Process And Goal Programming ForInterdependent Information System ProjectSelection,” Computers & Operations Research,27(4): 367.

Liberatore, M., and T. Miller (1995), “A DecisionSupport Approach for Transport Carrier and ModeSelection,” Journal of Business Logistics, 16(2):85-115.

Lu, Chin-Shan. 2003, “Market Segment Evaluationand International Distribution Centers,”Transportation Research Part E: Logistics andTransportation Review, 39(1): 49-60.

Meixell, M. and M. Norbis (2008), “A Review ofthe Transportation Mode Choice and CarrierSelection Literature,” The International Journal ofLogistics Management, 19(2): 183-211.

McGinnis, M. (1989), “A Comparative Evaluationof Freight Transportation Choice Models.”Transportation Journal, 29(2): 36-46.

Meade, L. & J. Sarkis (1999), “AnalyzingOrganizational Project Alternatives for AgileManufacturing Processes: An Analytical NetworkApproach,” International Journal of ProductionResearch, 37(2): 241-261.

Saaty, R. (2013), “A Validation of the Effectivenessof Inner Dependence in an ANP Model”,Proceedings of the International Symposium onthe Analytic Hierarchy Process.

Winter 2017 21

Saaty, T., and L. Vargas (2000), Models, Methods,Concepts and Applications of the AnalyticHierarchy Process, Boston: Kluwer AcademicPublishers.

Saaty, T. (2001), “Decision Making with the AHP:Why Is the Principal Eigenvector Necessary?,”ISAHP 2001 Proceedings, Bern, Switzerland.

Saaty, T. (2003), “Theory of the Analytic HierarchyProcess, Part 2.1,” The International Journal ofSystems Research and Information Technologies,(1).

Saaty, T. (1990/1997), The Analytic HierarchyProcess, New York: McGraw Hill, 1980 andPittsburgh: RWS Publications.

Saaty, T. (1999), “Fundamentals of the AnalyticNetwork Process,” ISAHP, Kobe, Japan 12-14.

Super Decisions 1.6.0, Freeware available at: http://www.superdecisions.com/~saaty/. 20 Feb 2006.

United States Government Accountability Office(GAO) (2005), “DOD Has Begun to ImproveSupply Distribution Operations, but Further ActionsAre Needed to Sustain These Efforts.” GAO-05-775. Washington.

Young, W., A. Richardson, K. Ogden, and A.Rattray (1982), “Road and Rail Freight ModeChoice: Application of an Elimination-By-AspectsModel,” Transportation Research Board, 838.

BIOGRAPHIES

Professor Anderson is an Assistant Professor in the Department of Information Systems and OperationsManagement, at Ball State University, Muncie, IN. He earned his PhD in Operations Management fromIndiana University – Bloomington. He has published in such journals as the Journal of TransportationManagement, Computers and Industrial Engineering, Mathematical and Computer Modeling, and theJournal of Defense Modeling and Simulation. He was a Logistics Readiness Officer, during a 23-yearUnited States Air Force career. He is now the ‘Point Person’ for the Undergraduate Logistics and SupplyChain Management Major at the Miller College of Business, Ball State University. Email:beanderson@bsu.edu

Tyler T. Prevett is the Deployment and Distribution Operations Center Chief at USTRANSCOM, ScottAFB, IL. He earned his Master in Air Mobility Degree from The Air Force Institute of Technology in2006. He has worked in number of different roles during his career in the United States Air Force (USAF),including several roles at USTRANSCOM. In addition to his current position, he has also been theDeputy Chief of Current Operations and the Deputy Chief of the Joint Operational Support Airlift Centerat USTRANSCOM. He has also held the USAF position of Commander of an Operations SupportSquadron. Email: ttprevett@gmail.com