at SciVerse ScienceDirect

Social Science & Medicine 84 (2013) 53e60

Contents lists available

Social Science & Medicine

journal homepage: www.elsevier .com/locate/socscimed

Local availability of physicians’ services as a tool for implicit risk selection

Amir Shmueli a,*, Esti Nissan-Engelcin b

a The Hebrew University-Hadassah School of Public Health, POB 12272, Jerusalem 91120, IsraelbBen Gurion University, Israel

a r t i c l e i n f o

Article history:Available online 11 February 2013

Keywords:IsraelRisk selectionRisk adjustmentPhysicians’ servicesGeographical disparities

* Tel.: þ972 2 6758514; fax: þ972 2 6435083.E-mail address: amirsh@ekmd.huji.ac.il (A. Shmue

0277-9536/$ e see front matter � 2013 Elsevier Ltd.http://dx.doi.org/10.1016/j.socscimed.2013.02.004

a b s t r a c t

Risk adjustment of the allocated health budget to health plans plays a major role in the functioning ofcompetitive social health insurance systems. Whenever the risk adjusted allocation is below theexpected marginal cost of care for a given person, incentives for risk selection arise. Since coverage isuniversal, risk selection must take on implicit forms such as stinting and distorting quality of healthservices. One of the tools for such selection is to strategically determine the local availability of physiciansbased on the local population. The Israeli competitive national health insurance scheme includes an age(only)-risk adjustment. We argue that the localities’ known characteristics are used by the Israelimanaged care organizations (sickness funds) to adjust the availability of and accessibility to communityhealth services. Consequently, we expect strong competition and high availability of services in healthier-than-average (and richer) towns, and weak competition and low availability of services in sicker-than-average (and poorer) towns. The empirical analysis combines data on the reception hours of physi-cians in five specialties and socio-economic and demographic characteristics (age, mean income, mor-tality rates etc.) of 60 towns in 2004, and strongly confirms that hypothesis, controlling for several otherpossible explanations for such findings. Such a situation clearly represents a regulation failure and aninefficient and inequitable geographic allocation of health services.

� 2013 Elsevier Ltd. All rights reserved.

Introduction

Cream-skimming (also known as explicit risk selection ordumping) is an expected reaction of insurance firms and providersin private unregulated markets to information asymmetry andincomplete risk rating of the premiums (Newhouse, 2002). Againstany social efficiency or equity values, such a selection will result inmarket failures where sick persons might be unable to purchaseinsurance or to be treated by their preferredmedical care providers.To correct for such failures, social insurance might be enacted,where coverage (insurance and treatment options) is universal.Since one of the fundamental principles of social insurance is theseparation between centralized payments according to income anddecentralized provision according to health state e a risk-adjustedmechanism is required, by which the total health budget is allo-cated to the health plans according to the expected cost of care(financial risk) of their insured population (see e.g., Van de Ven &Ellis, 2000). However, whenever the risk adjusted allocation isbelow the expected marginal cost of care for a given person, in-centives for risk selection arise. Since coverage is universal, riskselection must take on implicit forms such as stinting and distorting

li).

All rights reserved.

the quality and quantity of health services (Chalkley &Malcolmson,2000; Ellis & McGuire, 1990; Eggleston, 2000; Glazer & McGuire,2000; Newhouse, 2002; Van de Ven, 2003, Van de Ven et al.,2007). This is a regulation failure, leading to social inefficiencyand inequity of the same nature as the market failure mentionedabove, which the same regulation meant to alleviate.

Implicit risk selection is more feasible in Managed Care Organi-zations (MCOs) e organizations which vertically integrate the in-surance and the provision of services functions e than in indemnityinsurancefirms, since the quantityand qualityof their health servicescan be monitored and managed. In this paper we focus on one par-ticular dimension of managing care e the Israeli sickness funds’ de-cision on the availability of local community physicians’ services. Thisis naturally a keydecisionvariable of theMCOs, andweargue that it isused to regulate the local care provided according to whether thetown is a center of predictable profit or loss, as determined by therisk-adjustment mechanism and the town’s health profile.

Several research lines have produced results which are relatedto the ones examined in this paper. Physicians’ density and avail-ability have been traditionally examined in relation to SupplierInduced Demand (SID) and health care consumption. The findingsindicate that there is a significant positive association betweenphysicians’ density and health care consumption (Leonard,Stordeur, & Roberfroid, 2009). Limiting availability might be a tool

A. Shmueli, E. Nissan-Engelcin / Social Science & Medicine 84 (2013) 53e6054

by the insurers to contain health care cost. Our emphasis is differ-ent; it is on the differential use of limiting availability according tothe locality’s characteristics as a tool of implicit risk selection.

Vast research has documented the income-related inequality indoctors’ use. For example, in an experiment on 12 EU memberstates, Van Doorslaer, Koolman, and Jones (2004) found no incomerelated inequity in GP utilization, while there existed a strong pro-rich inequity in all countries with respect to contacting specialists,controlling for health needs. The common approach has been torelate these gaps in use to considerations on the demand side; suchas medical care being a normal good, lower prices faced by the richdue to wider insurance coverage or simply heterogeneous prefer-ences. The present analysis focuses on considerations of the supplyside which might lead to similar results.

Some 40 years ago, Hart (1971) coined the term “The inversecare law”, according to which the availability of good medical caretends to vary inversely with the need for it in the population servedby the non-competitive English NHS. He explained this phenome-non by the positive correlations between income and access, andincome and health. The present analysis tries to explain a similarphenomenon by the incompleteness of the risk adjustmentmechanism e the “Achilles heel” of the competitive Israeli system.

The paper is organized as follows: in the next sectionwe providea brief description of the Israeli national health insurance system. InSection 3 we describe the theoretical considerations on which theempirical model, discussed and estimated in Sections 4 and 5, isbased. Sections 6 and 7 are a conclusion to the paper.

A brief description of the Israeli national health insurancesystem

The National Health Insurance scheme which was introduced inJanuary 1995 consists of a managed competition model (Enthoven,1978), where four private non-profit sickness funds compete on thequality of care and service of medical care covered by a uniformpackage of benefits defined by the law.

The package of benefits is comprehensive and includes primary,secondary and inpatient care, as well as diagnostic and pharma-ceutical care. The budget of the package of benefits is determinedannually by the government, and is partially indexed to changes ininput prices, demography and technological advances. It is financedby an earmarked health tax, transfers from the general revenuesand co-payments.

The scheme is compulsory (all are insured) and universal (norejection). Citizens are free to switch sickness funds yearly, how-ever, the switching rate is low (1e2% annually). There is no directpremium paid by the members to the sickness funds.

The main source of income for the sickness funds are the risk-equalized payments from the central health fund (the budget ofthe package of services). Risk equalization consists of two separatecomponents: a prospective age-adjustment specifying fixed ratesfor each of 11 age groups (governing the allocation of 94% of thebudget), and a retrospective risk sharing arrangement e governing6% of the budget e by which the sickness funds receive an annualfixed payment per person who is sick with one of five “severeconditions” e Renal failure on Dialysis, Thalasemia major, Gaucher,AIDS, and Hemophilia.

While explicit rejection is against the law, the risk adjustmentsystem is clearly incomplete and leaves incentives for implicit riskselection against high risk (expected cost) individuals, and raisesfinancial difficulties to sickness funds with higher-than-averagerates of high-cost populations.

The sickness funds are Managed Care Organizations, integratingthe insurance and the provision functions. They differ in the waythe care is managed, with which providers they contract and the

terms of the contracts (for further details see Rosen, 2003).Approximately 80% of Clalit Health Services (CHS) members, thebiggest sickness fund in Israel, receive primary care at CHS-ownedand -operated clinics. The clinic-based doctors receive a basemonthly salary and a monthly age-based capitation payment foreach member on their list above a prescribed basic number. Thebase salary, the capitation rate and the prescribed basic list size forclinic-based physicians are all determined in a collective bargainingagreement between the Israeli physicians’ association and CHS.Approximately 20% of CHS members and the members of the otherthree sickness funds receive their primary care from independentphysicians at their clinics. These doctors are paid a passive (perregistered member) or active (per first visit) capitation rates setunilaterally by each of the sickness funds.

Salaried specialists provide the majority of CHS community-based specialist care. Almost all salaried specialists work in clinicsowned and operated by CHS. They can get additional payments for“first-time” patients and for procedures on a contractually agreedlist. All of these parameters are negotiated between the Israeliphysicians’ association and the CHS.

In the other sickness funds, and with a minority of specialists inthe CHS, community-based specialists are independent and typi-cally paid on the basis of a “points” systemwhich takes into accountthe number of sessions they work (3e4-h periods), the number ofvisits and the number and nature of special procedures performed.Senior consultants are paid somewhat differently; most of themwork sessions of 3e4 h in duration and are paid per session, ratherthan according to a points system.

To sum up, two distinct factors combine in Israel to create a se-vere threat to the efficiency and equity of the system. On the onehand, the risk adjustment scheme, by which the central healthbudget is allocated among the four competing sickness funds, isbased on age only and thus leaves considerable incentives for im-plicit risk selection. On the other hand, the sickness funds areManaged Care Organizations that control the supply of medicalservices and the way they are managed. The result of this combi-nation is that the sickness funds are able to strategically determinethe local medical care provided in order to maximize predictableprofits e and minimize predictable losses e implied by the currentrisk adjustment scheme and the locality health profile.

Theoretical considerations and specification

The argument

Since the Israeli risk-adjustment scheme is based on age only,identified sicker-than-average individuals constitute predictableloss in all age groups to the sickness funds. The Law prohibitsexplicit selection by open enrollment, so the four sickness fundsmight bemotivated to “exile” existing and deter potential memberswho seem to be predictable losses by means of implicit selectiontools.

While the Israeli sickness funds operate nationally, since pop-ulation health is produced locally, the sickness funds’ decisions onthe supply and availability of community health services are madeon a local basis (e.g., the number of local physicians contracted).These decisions might be used to implicitly select against collec-tively expensive enrollees by limiting the availability of andaccessibility to community health services in sicker-than-averagelocalities.

Since all sickness funds will find it profitable to engage in suchimplicit selection, in equilibrium, we expect, ceteris paribus, strongcompetition among the sickness funds to attract inhabitants inhealthier-than-average localities, and weak competition to deterinhabitants from joining the sickness fund in sicker-than-average

A. Shmueli, E. Nissan-Engelcin / Social Science & Medicine 84 (2013) 53e60 55

localities. A strong competition will present itself with high avail-ability of medical services while a weak competition will presentitself with low availability of medical services. Such equilibriumclearly represents a regulation (risk adjustment) failure and aninefficient and inequitable allocation of health services acrosslocalities.

We do not, however, expect (nor find) that there will be nophysicians’ services at all in the sickest localities. First, at least thebiggest sickness fund e Clalit Health Services (CHS) e has beenmotivated also by social responsibility to provide some medicalcare everywhere. Second, if there are economies of scale in theprovision of care, a single sickness fund or a sickness fundwith a bigmarket sharemight enjoy a local profit (since the risk adjusted ratesdo not take into account economies of scale) which might cover thehigh expenditures in sick localities.

Additional covariates

Several additional local factors, however, might determinephysicians’ availability and they should be controlled for in order toestimate the relationship between population’s health needs andphysicians’ availability. First, we controlled for age-standardizedsize. Size is an important factor, since if economies of scale existin the provision of care; local provision cost might be lower in biglocalities (or in localities with high market share) even if the healthcondition is poor.

Second, the peripheral status (measured by the new CentralBureau of Statistics (CBS) peripheral index, see below) of the townmight affect services availability. Central locations are moreattractive to health professionals both on grounds of professionalinterests (closer to academic centers), social interests (closer tocultural centers) and jobs and education opportunities for thefamilies. Consequently, sickness funds might have to offer highercompensation for health professionals in peripheral towns, makingmedical care there more expensive for the sickness funds. The factthat health is produced locally enables the sickness funds to limit(expensive) supply in peripheral areas, not because of (possibly)sicker populations (risk selection) but because of higher unit-costand expenditures.

Third, we controlled for the distance of the town from thenearest general hospital. In towns which are close to hospitals,outpatient care in the hospitals might substitute for ambulatoryspecialists’ care, and consequently, local ambulatory physicians’services might provide an underestimate of the availability ofphysicians to the town’s inhabitants. Furthermore, stinting oncommunity services might end up in expensive emergency andinpatient care, the risk for which is inversely related to the distancefrom the nearest hospital. Since the CHS buys inpatient care fromgovernment and other hospitals as well, we controlled for both thedistance to the nearest CHS’s and non-CHS’s hospital even whenthe CHS has a quasi-monopolistic market share.

Fourth, we controlled for religion (Arab vs. Jewish towns). TheIsraeli Arab population is a minority with distinguished epidemi-ology and care patterns, which might affect its community healthservices.

Finally, we needed to control for the possible effect of the localsupply of doctors on the availability of the sickness funds’ com-munity physicians. There are a number of factors determining thelocal supply of doctors. One is the income effect on the demand forprivate community care. Rich localities with high demand for suchcare might attract doctors and increase their supply independentlyof the sickness funds’ actions. Since private community medicalcare e mostly used for second opinion from senior specialists e isalmost entirely financed in Israel by supplementary insurance, weused the local proportion of inhabitants who own supplementary

health insurance to control for the demand effect. We note that thesupplementary insurance is offered by the sickness funds to theirmembers only, so that the competition in the supplementary in-surance market is indistinguishable from that in the universal in-surance. From social point of view, supplementary insurance isknown to be a popular way to select the good risks and its effectreinforces our argument of availability as a selection tool (Van deVen & Ellis, 2000).

More importantly in Israel, as wementioned above with respectto the peripheral gradient, the supply of doctors is mainly deter-mined by the proximity to professional, social and educationalcenters. This is particularly important for doctors who work parttime for the health plans (and are the focus of this paper) and part(or full) time for a hospital, university or a research/care center orfacility. Given the local supply of doctors, the sickness fund contractswith some of them, and this defines the local availability of thesickness fund’s community services. Our argument is that the sick-ness funds contract different levels of availability depending on thelocal health (and wealth) status.

Unfortunately, there is no way to distinguish practically be-tween the availability of the sickness funds’ services which is thefocus of this study and the local supply of doctors. Doctors mightreside in one town but practice in another, and, since completelyprivate practice is rare, this can be observed only if they are con-tracted by one of the health plans, namely are counted in ourmeasure of availability (see below).

Information issues

The way the sickness funds estimate the health conditions ofa specific town is an important issue. Because of the strong (positive)correlation between health and wealth (Fuchs, 2004 and see belowfor the present data set), a commonly used indicator of health state isthe level of income. The patient’s income is not known by the sick-ness fund she belongs to. In fact, the only way the sickness funds cantake into consideration their patients’ socio-economic status is byfocusing on their place of residence. The general socio-economicstatus of the Israeli localities is a common knowledge, and exactinformation on the towns’ socio-economic status can be found in theCBS official publications. Therefore the argument set forth aboveleads to the prediction that in high income towns the competitionamong the sickness funds will be strong and the supply of healthservices will be high, while in poor towns the competition will beweak and supply of health services will be limited.

Standardized Mortality Rate (SMR) is a commonly used directmeasure of population health needs. High SMR indicates excessmortality andmorbidity associated with higher use of medical care.For that reason, the English NHS allocates the central health budgetamong the various districts based, among other factors, on thedistricts’ SMR (Mays & Bevan, 1987). Following the theoretical ar-guments set above, towns with lower SMR are expected to enjoystronger competition and a higher level of physicians’ supply.

Since the sickness funds are interested in the medical conditionof the patients and not in their economic status per-se (beyond thepotential purchase of supplementary insurance offered by thesickness funds), we predict that, controlling for SMR, the effect ofthe mean income on the availability of medical services will drop.

Methods

Data

The sample of localitiesSixty localities with 20,000þ inhabitants were identified from

the Israeli CBS 2004 registry of localities. The reason for limiting the

A. Shmueli, E. Nissan-Engelcin / Social Science & Medicine 84 (2013) 53e6056

sample to towns with a size of at least 20,000 is that age-distribution e a central variable in our analysis e is not availablefor smaller localities. In total there were 1188 Israeli localities(excluding kibbutzim, institutions etc.) in 2004. 1026 localities hadless than 5000 inhabitants, where, in order to prevent inefficientduplications, the National Health Insurance Law (amendment of27.10.1997) specifies that only one sickness fund will operate. Inlocalities with 5000e10,000 inhabitants (55 localities), 1e2 sick-ness funds operate for the same reasons. The concession sicknessfunds are chosen using a tender. A sampling bias might rise fromthe exclusion of the 38 localities with 10,000e20,000 inhabitants.

The availability of physicians’ servicesPhysicians’ hours per week were collected for 2004 from the

service ledgers of the four sickness funds in each town. The serviceledgers of the four sickness funds include the number and range ofhours during which ambulatory services in the community wereprovided.Whenever a doctor had the same reception hours inmorethan one sickness fund’s service ledger, these hours were countedonly once. The community ambulatory service is divided into twomedical spheres: primary and secondary medicine. Primary med-icine consists of three main subcategories: family medicine, pedi-atrics and gynecology. Secondary medicine consists of two mainsubcategories: sub-internal medicine and surgery. Sub-internalmedicine comprises the following medical categories: allergology,oncology, endocrinology, gastroenterology, hematology, neurology,nephrology, skin and venereal diseases, cardiology and rheuma-tology. Surgical medicine comprises otolaryngology, orthopedics,urology, surgery, and ophthalmology. For the analysis, we used theweekly number of physicians’ hours per 1000 age-standardizedinhabitants (see below) in five specialties: family medicine, pedi-atrics, gynecology, sub-internal and surgery medicine and in total(in pediatrics, we used also the weekly hours of pediatricians per1000 children aged 0e15).

The market shares of the sickness fundsThe local market shares of the four sickness funds in Israel were

taken from the databases of the National Insurance Institute.

Towns’ characteristicsAverage income. Since no data on the income of self-employedpersons by localities is available, National Insurance Institute datawas used for the average earnings of salaried people in each town.The salaries are in gross terms per month, representing all the grosspayments made to a salaried employee for a whole month, i.e. basepay, cost of living allowance, seniority allowance, overtime, trans-portation allowance, etc. According to the CBS 2007 data, 90% ofhouseholds’ heads are employees and 10% are self employed. Meanincome of households headed by self-employed is about 15% higherthan that among households of employees, with no significantdifference in education level. Unless there is a strong correlationacross towns between mean salary and percent employed, we donot expect any significant bias from focusing on average earnings ofsalaried persons.

We note that we tried, in the empirical model, to use the CBS’sSocioeconomic Index e which integrates financial resources,housing, home appliances, motorization rate, education and liter-acy, employment and unemployment, socioeconomic distress anddemographic features instead of mean income. The results weregenerally similar, and we chose to use the clear and well defined e

although somewhat narrower e employees’ mean income as thetown’s socio-economic indicator.

The peripheral index. This newCBS index characterizes towns basedon their geographic location and their proximity to the primary

economic activity in the country. The index is valid and reliable,expressing the extent to which a local authority is more central(high value), or more peripheral (low value).

Age-standardized population. For the purposes of age standardiza-tion, the ambulatory component of the Israeli risk adjustmentformula, which describes the relative consumption of ambulatorymedical services by age, was used.

Standardized Mortality Rate (SMR). This CBS index reflects localmortality adjusted by the age structure of the Israeli population.

Jewish and Arab localities. Identified from the National InsuranceInstitute data.

The HerfindahleHirschman Index (HHI). The HHI is a very well-known measure of the level of competition in the market. It iscalculated as the sum of the sickness funds’ local market sharessquared. Lower values indicate higher competition, and a value of 1indicates a monopoly.

Distance to the nearest CHS general hospital. The road distance be-tween the nearest CHS general hospital and the town’s main PostalOffice (traditionally taken as the town location in measuringdistances).

Distance to the nearest non-CHS general hospital. As above, for thenearest general hospital not owned by CHS (a government or otherprivate non-profit hospital).

Ownership rate of supplemental insurance. The local rates werecalculated from the Brookdale Institute’s 2005 national survey(Gross, Brammli-Greenberg, & Matzliach, 2007).

The econometric strategy

The hours per age-standardized 1000 persons of each of the fiveclinical specialties (and in total) and the HHI are considered as(seven) jointly dependent variables. Using the data on the 60 lo-calities, the main analysis explores the variation in the sevendependent variables by the localities’ characteristics. Unfortu-nately, we could not explore all the interdependencies among thedependent variables in a simultaneous equations model. Instead,we used Seemingly Unrelated Regressions Estimation (SURE)(Wooldridge, 2003, chap. 7.7). The SURE procedure estimatesa system of equations which are related through non-zero corre-lations of their error terms. We expect that the error terms will becorrelated across the equations since the decisions on the avail-ability of doctors from different specialties in a given town areexpected to be interdependent.

Since the use of hours per capita might introduce hetero-skedasticity (the error term variance being inversely proportionalto the age-adjusted population squared), the systemwas estimatedwith weights (equal the square root of the age-adjustedpopulation).

Three models were estimated. Model A focuses on the effect ofthe towns’ (natural log of) mean income on the towns’ HHI andavailability of physicians’ services, controlling for age-adjusted size,religion (Arab vs. Jewish), peripheral status, distances to the nearestCHS and non-CHS hospital, and rate of supplementary insuranceownership.

Model B replaces town’s mean income by the SMR. The effect ofSMR is subject to a possible endogeneity bias, since local healthstatus is believed to be affected by the medical care consumed.Model C accounts for this possible bias by using IV estimation, with

A. Shmueli, E. Nissan-Engelcin / Social Science & Medicine 84 (2013) 53e60 57

mean income serving as the instrument for SMR. The Hausman testprovides the statistical test for the endogeneity of SMR.

Results

Table 1 provides the means of the variables included in themodels. Family physicians are by far the most available service witha mean of more than 16 weekly hours per 1000 age-standardizedpersons. They constitute about half the total physicians’ weeklyhours (34.8). Gynecologists and sub-internal medicine specialistsare the least available services, with 2.9 weekly hours per 1000persons on average. Pediatricians are available for 7.7 weekly hoursper 1000 and surgery medicine specialistse for 5.1, on average. Thevariation in availability across towns is quite high. The ratio of themaximal to theminimal availability found is 2 for family physicians,5 for gynecologists, 3 for pediatricians, 30 (!) for sub-internists, 10for surgical physicians, and 2 for total physicians.

The mean HHI across localities is 0.443. It ranges from 0.261indicating intensive competition (the minimal value of HHI is theinverse of the maximal number of active sickness funds operatingin a locality, namely, 0.25) to 0.886 which reflects high concentra-tion (1 being the case of local monopoly).

CHS has the largest market share nationally. However, its localmarket share ranges from 23% to 94%. In some towns some of thethree smaller sickness funds do not operate at all.

Arab localities constitute 12% of the total. The mean SMR is 6.2deaths per 1,000, ranging from 4.2 to 8.6. The mean age-standardized size of the localities analyzed is 87,000, ranging from21,000 to nearly 700,000. The peripheral index ranges from �3(peripheral) to 2.7 (central). The mean is 0.5. The mean of theaverage incomes across localities is 6000 NIS (2004 prices), rangingfrom 3000 to 11,000 NIS. Distances from hospitals are relatively lowin Israel. On average, a CHS hospital can be foundwithin 20 km fromany town, and similarly for a non-CHS hospital. The variation is large,however. The large cities have hospitals within their municipality,while the maximal distance from a CHS hospital is 85 km and froma non CHS hospital is 291 km. The mean supplemental insurancelocal ownership rate is 79%, ranging from 22% to 100% (the nationalownership rate in 2005 was 75%).

The correlations among the variables show that richer townsenjoy lower mortality (the correlation of ln(mean income) with

Table 1Variables’ definitions and means (n ¼ 60).

Variable Description

Dependent: availability measured as weekly hours per 1000 age-standardized personsPSNFAM Family physiciansPSNOB OB/GYN physiciansPSNPED PediatricsPSNINTER InternistsPSNSURG Surgical physiciansPSNTOT Total physiciansHHI The HerfindahleHirschman IndexMarket sharesCHS Clalit Health ServicesLSF Leumit Sickness FundMHS Maccabi Health ServicesMSF Meuhedet Sickness FundIndependentARAB ¼1 if the locality consists mainly of Arab citizensSMR Standardized Mortality Rate (SMR)NSTAN (0000) Number of age-standardized personsPERIFINDX The Peripheral IndexLNINCOM Natural log of average incomeD_CHSHOS Distance (road KM) to the nearest CHS’ hospitalD_OTHHOS Distance (road KM) to the nearest non-CHS’ hospitalSUPPINS Local ownership rate of supplemental insurance

SMR is �0.757), higher competition among the sickness funds(correlation with HHI is �0.595), and higher availability of physi-cians, in particular in internal and surgical medicine (correlationsare 0.628 and 0.526 respectively). Another result was that highermortality (SMR) is associated with less competition (correlationwith HHI is 0.626) and with lower availability of physicians, mainlysub-internal and surgery medicine (correlations are �0.694and �0.650 respectively). These two observations confirm ourpredictions. A third result was that central (vs. peripheral) townsare bigger (correlation of the peripheral index with age-adjustedsize is 0.422), have higher availability per capita of physicians,and in particular, internal and surgery medicine specialists (corre-lations 0.524 and 0.472 respectively), enjoy less concentratedmarkets for medical care (correlation with HHI is �0.377) and arericher (correlationwith ln(mean income) is 0.517). This observationindicates that, as argued above, peripheral status might be a sepa-rate factor leading to risk selection via limiting availability of care.Finally, we note the high correlation between HHI and CHS’s mar-ket share (0.924). High concentration or near-monopoly situationsare associated with a large market share of the CHS.

Table 2 presents the SURE estimation results. Model A focuses onthe identification of the effect of the town’s economic affluence e asa proxy for local health state e on the availability of physicians’services to its inhabitants, controlling for size, religion, peripheralstatus, distance from the nearest hospitals and rate of supplementalinsurance ownership. The results indicate that there is a positiveeffect of the town’s mean income on physicians’ hours per capita.The effects are significant for the availability of gynecologists, sub-internal and surgery medicine specialists, and for the total avail-ability of physicians in the town, but not for family physicians andpediatricians (the results are similar when the availability of pedia-tricians is calculated as hours per 1000 children ages up to 15 insteadof hours per 1000 persons). The effect of mean income on HHI issignificantly negative.

The income elasticity of Obstetricians/Gynecologists’ availabilityis 0.379, namely, a 10% increase in mean income is associated with3.79% increase in gynecologists’ weekly hours. The income elas-ticity of internal medicine physicians’ availability is higher, 0.729.The income elasticity of surgerymedicine specialists’ hours is 0.442and that of total physicians’ availability is 0.219. The income elas-ticity of the HHI is marginally significant at �0.214.

Mean Std. dev. Minimum Maximum

16.148 2.033 11.377 22.1992.928 0.814 0.958 4.9877.663 1.707 4.035 13.2452.932 1.441 0.220 5.8985.076 1.979 1.107 10.528

34.747 5.569 24.384 49.6080.443 0.123 0.261 0.886

0.560 0.143 0.230 0.9400.089 0.053 0.000 0.2400.258 0.128 0.000 0.5200.093 0.087 0.000 0.340

0.117 0.324 0.000 1.0006.163 0.922 4.200 8.6008.727 11.105 2.132 69.4840.525 1.144 �3.011 2.7288.653 0.276 8.119 9.306

20.516 17.015 1.392 85.21721.107 40.269 0.648 290.9530.785 0.209 0.224 1.000

Table 2Estimation of the availability of physicians services and HHI equations.

PSNFAM PSNOB PSNPED PSNINTER

b t-Value b t-Value b t-Value b t-Value

Model A: SURE (LNINCOM)Constant �1.25223 �0.418 �7.16651 �1.026 �1.65183 �0.545 �17.7208 �2.773PERIFINDX 0.11603 0.472 0.12579 1.281 0.57485** 2.024 0.28409 1.509ARAB 1.97716** 2.074 0.01671 0.044 1.75465*** 2.595 �0.66501 �1.251NSTAN 0.01952 1.263 0.00266 0.424 �0.02551** �2.196 0.02287** 2.089LNINCOM 1.31677 1.122 1.10938** 2.484 0.30713 0.312 2.13570*** 3.264D_CHSHOS 0.02547 1.438 0.01658** 2.423 0.03511** 2.421 0.00717 0.706D_OTHHOS �0.00653 �0.822 �0.00184 �0.599 �0.00128 �0.197 �0.00119 �0.261SUPPINS �0.40261 �0.237 �0.31245 �0.476 �0.92995 �0.670 1.19211 1.237R-squared 0.735 0.755 0.749 0.835Model B: SURE (SMR)Constant 19.8403*** 8.832 4.83493*** 5.541 8.67009*** 4.669 8.54298*** 7.151PERIFINDX 0.04078 0.114 0.16208 1.169 0.54498* 1.846 0.1619 0.853ARAB 2.47266*** 2.585 0.22533 0.607 2.11070*** 2.671 �0.09668 �0.19NSTAN 0.01314 0.776 �0.00273 �0.417 �0.02825** �2.025 0.02287** 2.543SMR �0.69400** �2.041 �0.34965*** �2.648 �0.31179 �1.109 �0.91067*** �5.037D_CHSHOS 0.01669 0.961 0.01175* 1.742 0.03107** 2.161 �0.00386 �0.417D_OTHHOS �0.00963 �1.189 �0.00329 �1.048 �0.00264 �0.395 �0.00537 �1.248SUPPINS �0.25797 �0.159 �0.14892 �0.236 �0.79685 �0.595 1.16646 1.372R-squared 0.741 0.758 0.751 0.864Model C: IV (SMR)SMR �1.11168* �1.773 �0.60031** �2.549 �0.54807 �1.072 �1.26797*** �4.066Hausman test p 0.6363 0.2651 0.6122 0.0876*

PSNSURG PSNTOTa HHI

b t-Value b t-Value b t-Value

Model A: SURE (LNINCOM)Constant �14.7802 �1.422 �40.6194 �1.612 1.45984*** 2.979PERIFINDX 0.49974 1.432 1.36884** 2.226 �0.02135 �1.643ARAB �0.96302 �1.125 1.95326 0.819 0.16548*** 3.735NSTAN �0.00327 �0.215 0.02451 0.635 �0.00072 �0.847LNINCOM 2.24478** 2.123 7.62839** 2.938 �0.10971** �2.001D_CHSHOS 0.00621 0.378 0.09053** 2.057 �0.00141* �1.686D_OTHHOS �0.00308 �0.419 �0.01392 �0.705 �0.00027 �0.717SUPPINS 0.34623 0.221 0.4819 0.012 �0.22126*** �2.958R-squared 0.764 0.792 0.846Model B: SURE (SMR)Constant 14.1036*** 7.755 55.4656*** 10.894 0.07972 0.818PERIFINDX 0.13466 0.466 1.05942 1.309 �0.01388 �0.896ARAB 0.39414 0.509 4.88309** 2.253 0.10455** 2.519NSTAN 0.00083 0.061 0.00293 0.075 �0.00089 �1.222SMR �1.41002*** �5.123 �3.58733*** �4.655 0.06100*** 4.135D_CHSHOS �0.00773 �0.549 0.04979 1.263 �0.00077 �1.022D_OTHHOS �0.01037 �1.582 �0.02984 �1.626 .39274D-04 0.112SUPPINS �0.25055 �0.191 �0.13478 �0.037 �0.18847*** �2.847R-squared 0.825 0.825 0.872Model C: IV (SMR)SMR �1.22505** �2.571 �4.64860*** �3.329 0.07363*** 3.661Hausman test p 0.7648 0.3925 0.688

*Significant at 10%.**Significant at 5%.***Significant at 1%.

a Calculated from PSNTOT ¼ PSNFAM þ PSNOB þ PSNPED þ PSNINTER þ PSNSURG.

A. Shmueli, E. Nissan-Engelcin / Social Science & Medicine 84 (2013) 53e6058

One of the novelties of the present analysis is its ability tocontrol for the peripheral status of the town using the CBS’srecently published peripheral index. The results show that, con-trolling for the towns’ mean income, higher values of the index e

higher centrality e are related to higher availability of physicians.However, this effect is significant only for the availability of pedi-atricians and for the availability of all physicians in the town. TheHHI is not affected by the peripheral status of the town.

The age-adjusted size of the town affects the availability of pe-diatrics and internal medicine specialists only. Bigger towns enjoyhigher supply of sub-internists’ hours per capita but lower supplyof pediatricians’ hours.

The results suggest a difference in the availability of physiciansby the town’s religion. Arab towns enjoy higher availability of

family physicians and pediatricians, controlling for size, peripheralstatus and mean income. The HHI in these towns is markedlyhigher by 0.15 than that in Jewish towns, indicating higher con-centration (usually highermarket share of CHS). It seems that whilethe level of concentration is higher in Arab towns, as expected fromtheir low socio-economic ranking, the availability of family physi-cians and pediatricians is higher in Arab towns, controlling forincome.

While the distance to the nearest non-CHS has no effect on theavailability of physicians nor on the level of competition (HHI),towns which are nearer a CHS hospital enjoy lower availability ofgynecologists, pediatricians and of total physicians. It is thusprobable that (at least) the CHS uses the CHS hospitals’ outpatientservices to substitute for community physicians in nearby towns.

A. Shmueli, E. Nissan-Engelcin / Social Science & Medicine 84 (2013) 53e60 59

Ownership of supplemental insurance local rate does not showany effect on the local availability of physicians. It does show,however, a significant effect on the HHI: higher ownership rates areassociated with more intensive competition.

In the next stage of the analysis (not reported), the variableSMR was added to all the equations of the model. The main resultis the lack of significance whatsoever of the effect of the meanincome on physicians’ availability or on the HHI. Instead, a signif-icant negative effect of SMR on the availability of services wasfound. All the covariates’ estimated effects were very similar tothose estimated in Model B below and are not reported. It must benoted that the insignificance of the income effect in this model isnot due to multicollinearity with SMR. The correlation squaredbetween ln(income) and SMR is 0.573, which is much lower thanthe R-squared of the regressions (all above 0.72). Furthermore,multicollinearity usually inflates the standard errors of the co-efficients of the two correlated variables. In all the regressions,wherever the effect of ln(income) on availability was significant inModel A, it lost its significance while the effect of SMR becamesignificant.

Model B includes SMR but not ln(income) as explanatory vari-ables. The effects of SMR on availability are significant and negativefor all types of physicians except pediatricians. The hours per capitaof sub-internal and surgical medicine specialists have the highestelasticity with respect to SMR, about �0.3. The availability of gy-necologists and total physicians has an elasticity of about �0.1. Theavailability of family physicians and of pediatricians has very lowelasticity. The effect of SMR on the level of competition, the HHI, ishighly significant and positive, with elasticity of 0.158.

The effects of the other towns’ characteristics on availability inmodel B are similar to those found in model A above, with someindication that, controlling for SMR instead of mean income, theeffect of the peripheral index lost some of its power, and the gapbetween Arab and Jewish localities increased.

We note that in both models, the R-squared is always higherthan 0.7, indicating that the main part of the variation in availabilityis explained by the models.

The negative relationship between SMR and availability ofphysicians e higher SMR, indicating higher morbidity, being asso-ciated with lower availability of physicians’ services e might sug-gest that SMR is endogenous, namely, it is determined by theavailability of physicians’ services, where lower availability of ser-vices lead to higher morbidity and SMR. In order to examine thispotential bias, the seven equations were estimated by IV method.The instrument for SMR was the mean income of the town. Thefirst-stage equation has an R-squared of 0.654, and the instrumentis proved to be strong (t-value ¼ �5.081). Based on the resultsabove, mean income does not have an effect on availability con-trolling for SMR, and thus it is a valid instrument. The effects of theother covariates on availability are essentially the same as in ModelB, and are not reported.

The IV estimates of the effect of SMR on availability do not differfrom the OLS ones at p¼ 0.05, as is indicated by the Hausman Test preported in Table 2 Model C. We conclude that the negative effect ofSMR on availability of physicians discussed above in Model B is notbiased and confirms our predictions.

Discussion

We hypothesized that facing an incomplete age-based riskadjustment, the sickness funds are incentivized to attract healthiere and to reject sicker e individuals. Since they cannot exerciseexplicit selection (rejection or underwriting) by law, their effortsare expected to turn to implicit selection. Popular tools of implicitselection include skimping and strategic determination of quality

of care in order to attract healthy members. In this paper wefocused on another possible tool for implicit risk selection e

availability of community medical services. Since the availabilitydecisions are made locally and might differ from town to town,and since the health state of the town’s inhabitants can be known,we expect that healthier (and richer) towns would attract thesickness funds, enjoy lively competition, which would result inhigher availability of physicians. The results strongly confirm theseexpectations. Controlling for the peripheral status of the towns,religion, size, the local reliance on hospitals’ outpatient services,and supplemental insurance ownership, we found that richertowns e as indication of better health e enjoy more intensivecompetition and higher availability of community physicians.When the town’s health itself is considered (SMR), mean incomelost its significance, and a strongly significant negative effect onavailability is identified. This loss of the significance of incomewhen SMR is considered directly, confirms the argument that theprevious significant positive income effect on availability is notlikely to originate from demand side considerations, as is sup-ported by the insignificant effect of supplemental insuranceownership on availability as well.

The results indicate that the health effects are particularlystrong on the availability of sub-internal medicine and surgicalphysicians, nonexistent with relation to the availability of pedia-tricians, and weak in relation to the availability of family physiciansand gynecologists. This pattern is consistent with the argument weadvance, since the health care costs associated with the practice ofsub-internal medicine and surgical physicians are higher than forthe other physicians, both because their reimbursement is higher,the care they provide is more expensive, and their SID is moreintensive (The Amoraii Report, 2002).

No effect of mean income or SMR on the availability of pediatricservices was found. Pediatric services might be used to attractyoung and healthy families with children. As is seen in Table 2, theavailability of pediatricians in Arab town, where families are muchlarger than (non-orthodox) Jewish families, is significantly muchhigher, controlling for SMR (no direct measure of the towns’ meanfamily size is available).

The pattern explored above is consistent with e and mightprovide an additional explanation for e the income-related ineq-uity in physicians’ use found in Europe and elsewhere (VanDoorslaer et al., 2004). Repeatedly weak or no income-relatedinequity is found with respect to general practice (family physi-cians), while a strong pro-rich inequity is found with respect tospecialists. Our explanation rests, however, not on demand sidemotives but on pro-rich local availability of the sickness funds’specialists caused by the incomplete risk adjustment scheme.

Recently, a governmental team for the examination of the riskadjustment formula recommended the inclusion of the “peripheralstatus” (distance from the center of the country) in the formula, bywhich the sickness funds will receive higher payments for memberswho reside in peripheral localities. The results of the present paperclearly reiterate that this is not the main issue. The results show thatcontrolling for the localmeanhealth status (or income), theperipheralstatus per-se does not have any effect on the sickness funds’ supplydecisions (andconsequently, theunadjustedeffectof peripherymasksthe effects of the omitted health and income, and is biased(Schokkaert & van de Voodre, 2004)). Availability of health services isnot constrained inperipheral towns because of the higher unit-cost ofservices, but because of higher total cost e and hence higher losseserelated to lower local health status (and to lower mean income).

A few limitations of the study should be mentioned: (a) wementioned earlier that the omission of towns with 10,000e20,000might introduce a sampling bias. We cannot think, however, of anysystematic and predictable way such a bias might affect the findings.

A. Shmueli, E. Nissan-Engelcin / Social Science & Medicine 84 (2013) 53e6060

(b) Large cities are clearly heterogeneous, and comprise of a numberof local neighborhoods where local and different availability de-cisions might be taken. There might be an aggregation bias using bigcities as a single local provision unit. (c) The dependent variable weused is the official availability of physicians, measured as the totaloffice hours registered in the sickness funds’ ledgers. In reality, phy-sicians might stay over-time to treat more patients or not to use theentire reception hours. We believe however that in general thedeclared and the actual hours of work do not differ much, sincedoctors can adjust their patients’ load according to the amount oftime theywante and contractede to allocate toworkwith thehealthplan. (d) We attributed market shares results to the sickness funds’decisions and ignored consumers’ demand for sickness funds alto-gether. While the choice of a particular insurer is a complex process(Laske-Aldershof et al., 2004), since the Israeli sickness funds aremanaged care organizations not competing on price, availability ofcommunity services is likely to be themaindeterminant of the choiceof a particular sickness fund (Shmueli, Achdut, & Bendelac, 2007). (e)We did not account for possible intended substitution between(relatively cheap) family physicians and (relatively expensive) sub-internal (and sometimes surgical) physicians. This potential substi-tutionmight be, however, by itself a sign of stinting quality of care. (f)Finally, we were unable e conceptually and empirically e to controlfor the local supply of doctors. It might be argued that the lowavailability in sick localities is a result of low supply rather than ofintended risk selection. First, with respect to the sickness funds’physicians’ services, the supply of doctors is clearly endogenous, anddepends on the terms offered by the sickness funds. The fact thatnone of the sickness fund tries to attract doctors to sick (and poor)townsmight be an evidence of the argument advanced here. Second,in healthy (and wealthy) towns, facing a high supply of doctors, thesickness funds could contract relatively less doctors and save on cost.The fact that the availability there is so high relative to that in sicktowns serves, we argue, as an evidence of implicit risk selection, bywhich the competing sickness funds try to attract the profitableinsures.

Conclusion

The incomplete Israeli risk adjustment mechanism has beensubjected to strong criticism (Shmueli, Chernichovski, & Zmora,2003). The present study shows yet another facet of this regu-lation failure. The allocation of community physicians’ services isclearly inefficient in not responding to the population needs, andinequitable in discriminating against poor and sick towns. It is themissing health status from the risk adjustment scheme which cre-ates incentives to the sickness fund to select e individually andcollectively e the healthier-than-average and to deter the sicker-than-average from joining in each age group. Israel is about 10years behind systems such as the Netherlands and Germany whichintroduced health state and diagnoses into their risk adjustment

mechanisms in order to reduce the selection threat and its impli-cations to the efficiency and equity of the health system.

Funding source

No funding was received for this study.

Conflict of interest

No conflict of interest.

References

Chalkley, M., & Malcolmson, J. M. (2000). Government purchasing of health ser-vices. In A. J. Culyer, & J. P. Newhouse (Eds.), Handbook of health economics,Amsterdam: North Holland.

Eggleston, K. (2000). Risk selection and optimal health insurance-provider paymentsystems. Journal of Risk and Uncertainty, 67, 175e198.

Ellis, R. P., & McGuire, T. G. (1990). Optimal payment systems for health services.Journal of Health Economics, 9, 375e396.

Enthoven, A. C. (1978). Consumer-choice health plan: a national health insuranceproposal based on regulated competition in the private sector. New EnglandJournal of Medicine, 298, 709e720.

Fuchs, V. R. (2004). Reflections on the socio-economic correlates of health. Journal ofHealth Economics, 23, 653e661.

Glazer, J., & McGuire, T. G. (2000). Optimal risk adjustment in markets with adverseselection. American Economic Review, 90, 1055e1071.

Gross, R., Brammli-Greenberg, S., & Matzliach, R. (2007). Public opinion on the levelof service and performance of the health system in 2005, in comparison withprevious years. Myers-JDC-Brookdale Institute.

Hart, J. T. (1971). The inverse care law. Lancet, 405e412.Laske-Aldershof, T., Schut, F., Beck, K., Greß, S., Shmueli, A., & Van de Voorde, C.

(2004). Consumer mobility in social health insurance markets: a five-countrycomparison. Applied Health Economics and Health Policy, 3, 229e241.

Leonard, C., Stordeur, S., & Roberfroid, D. (2009). Association between physiciandensity and health care consumption: a systematic review of the evidence.Health Policy, 91, 121e134.

Mays, N., & Bevan, G. (1987). Resource allocation in the NHS. London: Bedford SquarePress.

Newhouse, J. P. (2002). Pricing the priceless. MIT Press.Rosen, B. (2003). In S. Thomson, & E. Mossialos (Eds.). Health care systems in tran-

sition: Israel, Vol. 5(10). Copenhagen: European Observatory on Health CareSystems.

Schokkaert, E., & van de Voodre, C. (2004). Risk selection and the specification ofconventional risk adjustment formula. Journal of Health Economics, 23, 1237e1259.

Shmueli, A., Achdut, L., & Bendelac, J. (2007). Who switches sickness funds in Israel?Health Economics, Policy and Law, 2, 251e265.

Shmueli, A., Chernichovski, D., & Zmora, I. (2003). Risk adjustment and risk sharingin Israel. Health Policy, 65, 37e48.

The Amoraii Report. (2002). The report of the Public Committee for the Examination ofthe Public Health System and the Physicians’ status. Tel Aviv.

Van de Ven, W. (Guest Ed.). (2003). Risk adjustment in Europe. Health Policy, 65(1).Van de Ven, W., & Ellis, R. (2000). Risk adjustment in competitive health plan

markets. In A. J. Culyer, & J. P. Newhouse (Eds.), Handbook of health economics,Amsterdam: North Holland.

Van de Ven, W., et al. (2007). Risk adjustment and risk selection in Europe: 6 yearslater. Health Policy, 83, 162e179.

Van Doorslaer, E., Koolman, X., & Jones, A. M. (2004). Explaining income-relatedinequalities in doctor utilization in Europe. Health Economics, 13, 629e647.

Wooldridge, J. M. (2003). Introductory econometrics. Thomson South-Western.