Rwanda Journal Series B: Social Sciences, Volume 3,

Transcription

1 Effect of Health Insurance on Demand for Outpatent Medcal Care n Rwanda: An Applcaton of the Control Functon Approach DOI: Ruhara Mulndabgw Charles, Unversty of Rwanda Urbanus Mutuku Koko, Unversty of Narob Abstract In the 2000 s the Government of Rwanda ntated health sector reforms amng at ncreasng health care access. Despte these reforms there has not been a correspondng ncrease n demand for health servces, as only about 30 percent of the sck persons use modern care (NISR, 2011). The objectve of ths paper s to examne factors nfluencng outpatent care demand n Rwanda and suggest approprate measures to mprove utlzaton of health servces. The source of data s the Integrated Household Lvng Condtons Survey (EICV2) conducted n 2005 by the Natonal Insttute of Statstcs Rwanda (NISR). A structural model of demand for health care s estmated to measure demand effects of covarates. The fndngs ndcate that health nsurance s a sgnfcant determnant of outpatent medcal care. In addton, prce of health care and household ncome are among the man drvers of utlzaton of health care. Beng female s found to ncrease the probablty of seekng outpatent health care. Two man polcy recommendatons emerge from these fndngs. Frst, the government should reduce out-of-pocket health care expendtures (OOPE) through subsdes to publc health facltes. Second, the government should reduce the premums for Communty Based Health Insurance Schemes (CBHIs) to ncrease the coverage rate. Keywords: Outpatent, Health nsurance, User Fees and logt model. 1. Introducton The theoretcal model for analyzng human captal, health, and ts effect on productvty, earnngs and labour supply was frst developed n Grossman (1972). The premse of ths theory s that an ncrease n a person s stock of health rases hs or her productvty n both market and non-market actvtes. There exst large productvty and wages benefts of better health. There s evdence to show that sckness can have adverse effects on learnng, and that these mpacts can later nfluence economc outcomes n lfe (Bhargava et al., 2001). Better health can make workers more productve, ether through fewer days off or through ncreased productvty whle workng. Improved nutrton and reduced dsease, partcularly n early chldhood, leads to mproved cogntve development, enhancng the ablty to learn. Healthy chldren also gan more from school because they have fewer days absent due to ll health. Whle health s determned by many factors ncludng medcal care, food, housng condtons and exercsng, t s accepted that medcal care s one of the key determnants n health producton functon (McKeown, 1976). Rwanda Journal Seres B: Socal Scences, Volume 3,

2 Santerre and Neun, (2010) argued that much as a frm uses varous nputs, such as captal and labour to manufacture a product, an ndvdual uses health care nputs to produce health. When other factors are held constant, an ndvdual health status ndcates the maxmum amount of health that can be generated from the quantty of medcal care consumed. Consderng the mportance of medcal care, both polcymakers and researchers have drected much attenton to the queston of how broad access to health servces can be ensured (Lndelow, 2005). Early polcy and research ntatves focused on the need to mprove physcal access through an expanson of the network of health facltes. Ths conssted of mprovng health care delvery ncludng health care professonals, equpments, and buldngs. A growng lterature on health care has, however, ponted out that supply s not suffcent and ths means that provdng maxmum access to health care remans a challenge for governments n many low ncome countres. In Rwanda, access to health care was dentfed as an mportant objectve n formulatng publc polces snce good health s recognzed as a necessary condton to enjoy economc and socal opportuntes. The country has developed a health care settng open to all Rwandans and that s accessble to everyone regardless of socoeconomc status. For nstance, n the Rwanda Economc Development and Poverty Reducton Strategy (EDPRS, 2008), access to health care s one of the strateges of eradcatng poverty. The strategy s objectve s to promote health care to the entre populaton, ncreasng geographcal accessblty, ncrease the avalablty and affordablty of drugs, and mprove the qualty of servces. Increased accessblty to health care has several benefts partcularly among the poor segments of the populaton (World Bank, 2001a). The Mllennum Development Goals (MDGs) also recognze health as an essental ngredent n socal and economc progress for any country. However, despte the mprovement n access to health care through Communty Based Health Insurance Schemes (CBHIs) and other nsurance provders, t s not known why health care utlzaton has remaned low n Rwanda. To ncrease access to health servces, the government of Rwanda ntated a number of health polces and other economc stmulus efforts some of them targetng supply-sde of the market whle other polces were amed at ncreasng servces utlzaton. The polces ncluded Vson 2020, Economc Development and Poverty Reducton Strategy (EDPRS) , One-Cow-One-Famly, Socal Securty Polcy 2009 and Health Polcy 2004 (Mnstry of Health, 2009). These polces were meant to ncrease access to health servces and hence mprove ultmately the health status of the populaton. The reforms were also meant to strengthen the health care system and make t more accessble (MOH, 2005). Despte these reforms, less than two out of fve sck people seek formal health care n Rwanda (NISR, 2011). The neffectveness of prevous polces amng at ncreasng health care 78 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

3 utlzaton s due to ther mplementaton wthout adequate evdence as to factors nfluencng health servce utlzaton n Rwanda. The am of ths study s to examne the factors that nfluence demand for outpatent health care servces n Rwanda. Although economc theory offers potental factors that nfluence demand for health care, there s lack of quanttatve assessment of ther effects n Rwanda. Evdence on these factors s needed n mplementng polces desgned to mprove health servce utlzaton n the country. To my knowledge, there are no studes n Rwanda that have been done n recent years to determne factors nfluencng health care demand. The only avalable evdence on ths s from studes by Jayaraman et al., (2008) and Shmeles (2010) whch focused on maternal health care and on effects of CBHIs at the dstrct level. In countres n whch estmates of demand for health care exst, research results provde conflctng evdence to demand effects of prce, ncome and nsurance suggestng that more studes are needed. Most studes on demand for health care have not addressed the problems of endogenety (reverse causalty) and heterogenety (varaton n the estmated effect sze due to unobservables). Falure to address these problems leads to based estmates (Rosenzweg and Schultz, 1982; Kabubo-Marara et al., 2009; and Lawson, 2004). Hunt-McCool et al., (1994) ponted out that dfferences n data, model specfcaton, and/or emprcal methods can contrbute to dversty the demand estmates and hnder clarty n health care fnancng polces. The paper addresses these estmaton problems, provdng rgorous evdence on outpatent health care demand determnants n Rwanda that polcy-makers can use to mprove health servce utlzaton across all the regons of the country. 2. Lterature Health care servce s demanded as an nput nto the producton of health that s part of the ndvdual s utlty functon together wth other goods. Emprcally, the analyss of health servces examnes ther determnants based on the mcroeconomc theory of consumer behavour. These determnants nclude factors related to ndvduals, household and communty. Numerous studes have attempted to quantfy how much health care people consume, the types of health care they use, and the factors underlyng utlzaton of health care. Several studes have documented the mpact of nsurance on demand for health care and found that the effect of nsurance on utlzaton vares across the populaton, the level and type of coverage (see Buchmueller et al., 2005; Barros and Galdeano, 2008). The study by Hahn (1994) found that unnsured households had lower average rates of utlzaton compared to persons wth prvate or Medcad coverage. Those wth Medcad for the full year were found to have the hghest rate of health care utlzaton whle the unnsured persons were found to have the lowest mean utlzaton for all types of servces. In a smlar study, Rwanda Journal Seres B: Socal Scences, Volume 3,

4 Barros and Galdeano (2008) estmated the effect of prvate health nsurance coverage beyond a Natonal Health System on the demand for several health servces n Portugal. The study estmated the mpact of havng addtonal coverage on the demand for 3 dfferent health servces; the number of vsts, number of blood and urne tests, and the probablty of vstng a dentst. The results showed large postve effects of coverage for the number of vsts and tests. Smlar fndngs are reported by Jones et al., (2006) who found prvate nsurance to be postvely assocated wth the probablty of health vsts n Ireland, Italy, Portugal, Span and the Unted Kngdom. Morera and Barros (2009) reported comparable results of the mpact of double health nsurance coverage on demand for health servces. Results show that double nsurance ncreases utlzaton of health care. Another study by Shmeles (2010) examned the effects of a CBHI on health care utlzaton at dstrct level n Rwanda. The study used the matchng estmator to address the endogenty problem. As n Hahn (1994), hgher utlzaton of health care servces was reported among the nsured than n unnsured households. The results ndcate that CBHI has a strong postve mpact on access to health care. The results were consstent wth the fndngs by Newhouse, 1981; Saksena et al., 2010; Rashad and Markowtz, 2009; Juttng, 2005 whch found that nsurance was an mportant factor n explanng health seekng behavor. Other studes however found that nsurance may have lttle effect on demand for health care dependng on geographcal locatons (Buchmueller et al., 2005). Cunnngham and Kemper (1998) documented that n areas where there exst a well-functonng health care system, the lack or reducton of nsurance coverage may not mply a sgnfcant lack of access to care. The expanson of coverage would then result n smaller changes n utlzaton than n locatons where the unnsured have fewer optons. For nstance, Mwabu et al., (2003) reported a negatve effect of nsurance suggestng that nsured people make fewer vsts to health facltes relatve to unnsured people. The reason for ths unlkely result was that people wth nsurance may have better health endowments and, thus, demand fewer health care relatve to unnsured people. However, none of the studes controlled for heterogenety of nsurance. Snce the effect of nsurance on utlzaton may vary across populaton, geographcal locaton, the level and type of nsurance coverage, health care demand research needs to handle the problem of heterogenetes to produce relable estmates. There s an extensve lterature n health economcs that sought to estmate the elastcty of ncome on demand for health servces. Most of the lterature show that demand for medcal care was ncome nelastc ndcatng that medcal care was a necessty good (Mocan et al., 2004). The postve sgn of the elastcty ndcates that as ncome ncreases, demand for health servces also ncreases. However, the lterature was nconclusve but noted that ncome effects vary wdely across studes, 80 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

5 countres and regons. Rngel et al., (2002) reports that ncome elastcty of demand usng cross-secton data ranges between 0 to 0.2. Ths knd of magntude suggests however that the effect of ncome on demand s relatvely small. The dfference n estmates across tme frames reles on the ncluson of the effects of changes n medcal technology that use long tme seres data (Rngel et al., 2002). Income elastctes based on cross-sectonal data or on tme seres data coverng a relatvely short perod assumes that the level of avalable medcal technology s constant. As real ncome n the populaton ncreases, the aggregate demand for new medcal technologes and new treatment approaches rses as well. Thus, from the prevous studes on the effect of ncome, no consensus has emerged, and the debate on whether health care s a luxury or necessty good contnues (Blomqvst and Carter, 1997). To account for the prce effect at dfferent levels of vsts rather than the average effect obtaned usng Ordnary Least of Squares (OLS), Mwabu et al., (2003) used quantle regresson method to analyze the effects of prce on demand for health servces n Kenya. The fees were found to have a negatve effect on demand for health care but dfferng across the quantles. The fndngs establshed that an ncrease of 10 shllngs reduced vsts by 0.2 percent. Clearly, the prce elastcty of demand for medcal care was found to be small n magntude and consstent wth Akn et al., (1986) and Sauerborn et al., (1994). The study dd not however address the endogenety and heterogenety problems to produce unbased estmates. Gven that demand for treatment s not determned by the ndvdual alone, several studes have nvestgated the household and communty factors. Lépne and Nestour (2008) controllng for the unobserved effects at the household and communty level that affect health seekng behavor show that household economc status and qualty of health care are mportant determnants of the probablty of seekng treatment from a qualfed provder. In addton, transportaton cost was found to be an mportant determnant of the lkelhood of seekng care as an ncrease of the average transport cost decreased the lkelhood to seek curatve care by 25 percent. Evdence from emprcal studes on the relatonshp between demand for health care and ts man determnants dffered n several ways. In addton, most of prevous studes assumed an exogenous nsurance and dd not consder the reverse causalty that s more lkely to exst between medcal care demand and health nsurance. Ths study provdes new evdence on the factors whch affect demand for health care usng data from Rwanda and handles the endogenety and heterogenety problems to ensure that estmates are unbased and consstent. Rwanda Journal Seres B: Socal Scences, Volume 3,

6 3. Methodology Followng Grossman s (1972a and b), ndvduals maxmze ther utlty over health and other goods subject to market and non-market factors. Health s one of the several commodtes over whch ndvduals have well-defned preferences. The market factors nclude avalablty of health nputs and ther prces, nsurance and household ncome. The nonmarket factors nclude household characterstcs, locaton or dstance and ndvdual characterstcs such as age, educaton, health status, and the percepton they have about the qualty of health servces (Appleton and Song, 1999; Ajakaye and Mwabu, 2007; Bategeka, 2009). Assumng that health care s a consumpton good, the consumer s problem can be expressed as: Max U U ( H, Z, X, Y) (1) where U s the utlty derved from consumpton of dfferent goods; Y s the health related goods that yeld utlty to the sck person and mprove health status; H s the health producton functon; Z stands for health nputs such as health care whle X represents all other goods and servces. The utlty functon s maxmzed subject to the followng constrants: B XP YP ZP (2) x y H H Z, I, S, C, A, h, P, N ) (3) ( s h O z Where Z s defned as n equaton (1) and I s household characterstcs ncludng nsurance; S s the soco-demographc varables ncludng age, sex and educaton; C stands for communty characterstcs ncludng dstance to health faclty; A s the household asset; h s s the sze of the household; P h s the prce of health whle N o s the household nonobservable characterstcs. In the frst constrant, B s the exogenous ncome and P x, P y and P z are, respectvely, the prces of health neutral goods (such as clothng), health related consumer good Z (such as health care) and health nvestment good Y such as exercsng. The maxmzaton problem s then expressed as: Max U U ( H, Z, X, Y) Gven H H( Z, I, S, C, A, hs, Ph, NO ) (4) s.t. B XP YP x y ZP z ) Solvng the maxmzaton problem yelds a demand functon for health care specfed as: D f I, B, A, S, C, h, P, N ) (5) h ( s h o 82 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

7 Where D h refers to the demand for outpatent; I s health nsurance; B s the budget or ncome; A stands for household asset and S stands for soco-demographc varables. C represents the communty characterstcs ncludng dstance to health faclty; h s s the household composton; P h s the prce of health care and N o s the household non-observable characterstcs. Equaton (5) s a structural outpatent health care demand equaton that ncludes an endogenous varable among the ndependent varables. The endogenous varable s health nsurance because of reverse causalty between demand for health care and nsurance whle exogenous varables nclude monetary prces for health care, ncome, age, gender, educatonal attanment of the ndvdual, household sze, locaton as well as regons. In ths study, the demand for outpatent care s dscrete rather than contnuous because patents seek or do not seek health care. In equatons (1) and (2), health nvestment good s purchased only for the purpose of mprovng health so that t enters an ndvdual s utlty functon only through H. In the demand for outpatent model, nsurance s assumed to mprove access to health servces. In addton, the heterogenety of health nsurance due to non-lnear nteracton of demand for health servces wth unobservable and omtted varables could bas the estmates. The study assumes that demand for health servces has only one endogenous varable. In ths study, demand for outpatent refers to any curatve outpatent servce provded by a physcan or any other medcal staff. Gven the dchotomous nature of the outpatent, the estmaton adopts a bnary dscrete model, where health care s ether sought or not. Assumng that the errors are dstrbuted logstcally, we adopt a Logt regresson method to estmate both outpatent and npatent health care demands. The dependent varable takes any two values; l f ndvdual uses outpatent health care and 0 representng the ndvduals who dd not use any health servces. The logt regresson s also preferred because most of the studes n demand for health servces use logt regresson (See Lépne and Nestour, 2008 and Hahn, 1994). Ths relatonshp can be expressed as: servce) 1 f the event takes place (the ndvdual seeks outpatent Y = 0 f the event has not taken place (the ndvdual has not sought treatment) Equaton (5) expressng the demand for health care can be rewrtten as: * y ' (6) x y * where s a latent varable showng the probablty that medcal care s or not sought, x ' s a vector of characterstcs related to the ndvdual, household and communty, and s the error term. Rwanda Journal Seres B: Socal Scences, Volume 3,

8 Y = 1 f y * 0 ( ( x '.e. x ' ) >0 and Y = 0 f y * 0.e. ) <0 The values 0 and 1 are used because they allow the defnton of probablty of occurrence of an event as the mathematcal expectaton of the varable Y. Ths can be expressed as: E[ Y ] Pr( Y 1)*1 Pr( Y )*0 Pr( Y 1) (7) Ths equaton shows that we need to compute the probablty of occurrence (Y=1) over the probablty of no-occurrence (Y=0). Assumng that the error term has an extreme value dstrbuton, ths can be done usng the logt relaton as shown by equaton 8. exp( 0 1X1 2 X 2... k X k ) Pr (Y =1) = 1 exp( X X... X ) In terms of the log-odds, the above expresson can be reformulated as Pr( Y 1 ) Pr( Y 1) ln = ln = ln 1 Pr( Y 1) Pr( Y 0) 1 k 0 j j j 1 = X log t( ) Whch s can be expressed as: log t( ) X X X X (10) 0 3 j 1 j j 0 Where Y s an ndcator for the choce of modern health care (outpatent) by the th household member, X 1 = Vector of characterstcs related to ndvduals lke age, educaton and sex, X 2 = Vector of characterstcs related to household such as ncome, nsurance, X 3 = Vector of characterstcs related to communty level characterstcs such as medcal specalst, and the dstance from household to health faclty. If n equaton (10) 0, then an ncrease n X j (for nstance the j household ncome), whle all other exogenous varables reman unchanged wll ncrease the log-odds rato of ndvdual seekng health servces. If 0, then an ncrease n X j (for example the user fee), wll j reduce the log-odds rato. If 0, then the varable has no effect. j However, n the case of expresson (10), the s ndcate the changes n logstc ndex wth the sgn of ndcatng the drecton of the eventual change n the probablty of seekng care from a gven health faclty. Equaton (10) s the structural form of the probablstc health care k 2 k 3 (8) 3 (9) 84 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

9 demand functon. In ths equaton as n the recent lterature, one of the ndependent varables, health nsurance s endogenous and the estmaton has to address ths problem. Endogenety s due to the reverse causalty between health nsurance and demand for health care. So, n order to obtan unbased and consstent estmates, nstrumentaton of the endogenous varable s requred. The nstrumental varable should be correlated wth the endogenous regressor but unrelated drectly to the dependent varable (Ajakaye and Mwabu, 2007). Health nsurance n equatons (10) s endogenous to the dependent varable. Thus, estmatng the equatons wthout takng nto account ths problem mght encounter the problem of smultanety whch s due to the possblty of reverse causalty between demand functons and health nsurance. Endogenety of health nsurance arses because the decson to purchase health nsurance and the utlzaton of health servces are ntertwned. Frst, snce nsurance reduces the effectve prce of medcal care, nsured people tend to consume more health servces (Rashad and Markowtz, 2009). Second, even f ndvduals cannot perfectly predct ther future health needs, they are lkely to have nformaton about ther health status that could lead them to antcpate hgher use of health servces, and then decde to buy health nsurance. Thus, health care utlzaton not only depends on the ndvdual s health nsurance coverage, but also the level of coverage may be nfluenced by antcpated utlzaton of health servces (Juttng, 2004). Mannng et al., (1987) argue that treatng nsurance as exogenous n demand for health care models produces based results. Ths s because people who antcpate consumng more health servces have an obvous ncentve to obtan nsurance cover ether by selectng a more generous opton at the place of employment by workng for an employer wth a generous nsurance plan, or by purchasng prvately a generous coverage. Exstng lterature suggests useful methods for dealng wth the endogenety problem. Among the common approaches to ths problem s the use of Two Stages Resduals Incluson (2SRI) regresson method whch s approprate for non-lnear models. The procedure s used to address the problems relatng to measurement error, smultanety and omtted varables. Ths method requres dentfcaton an observable varable or nstrument that s correlated wth the endogenous varable but uncorrelated wth the error-term (Koko, 2008; Ajakaye and Mwabu, 2007; Rosenzweg and Schultz, 1982; Strauss and Thomas, 1995; and Wooldrdge, 2002). The problem however, s to dentfy an observable varable, z, that satsfes two condtons. Frst, the selected varable s uncorrelated wth the error-term. Ths means that cov(z, ) = 0, that s, z s exogenous n the estmaton of the endogenous equaton (see Wooldrdge, 2002; Behrman and Deolalkar, 1988; Grlches and Maress, 1998; and Ackerberg and Caves, 2003). The second requrement nvolves the relatonshp between the dentfed nstrument, z, and demand for health servces. Ths means that the Rwanda Journal Seres B: Socal Scences, Volume 3,

10 dentfed varable should not have an mpact on health nsurance;.e., z must be relevant. Ths requres regressng health nsurance aganst all the exogenous varables, ncludng the nstrument (Wooldrdge, 2002; Greene, 2007; Jowett et al., 2004). In the frst regresson, the varables should have sgnfcant coeffcents when the choce varable s regressed on the dentfyng varable together wth all other exogenous varables (Ackerberg and Caves, 2003; Baum and Schaffer, 2003). In the frst stage, we estmated the reduced-form of health nsurance on all exogenous varables ncludng the nstrumental varables. The second stage regressed demand for health care on all ndependent varables plus nsurance and nsurance resduals obtaned from the frst stage regresson (Terza et al., 2008 and Palmer et al., 2008). Followng Ajakaye and Mwabu (2007); Mwabu (2008); Kabubo- Marara et al., 2009 and Bhasn and Bentum, (2010), we can re-formulate the demand for health servces n the form of smultaneous equaton as: D Z1 I,j=1...2 (11) d I j (12) Z 2 j j j where D and I are demand for health care and health nsurance respectvely. Z s a vector of ndependent varables, consstng of Z 1 covarates that belong to the demand for health servces functon and a vector of nstrumental varables that affect nsurance but have no drect mpact on demand for health servces. and are parameters to be estmated and s a dsturbance term. Equaton (11) s the structural equaton to be estmated whle equaton (12) s the lnear projecton of the potentally endogenous varable I, on all the exogenous varables. The system of equatons assumes that there s only one endogenous regressor n the demand equaton. The major challenge of the nstrumental varable approach s the challenge of obtanng vald nstrument for dentfyng the effect of endogenous varables n a structural model. Once potental nstrument s dentfed, t s mportant to test for ts sutablty by assessng whether t has three propertes: relevance, strength and exogenety of nstruments (Stock, 2010; Kabubo-Marara et al., 2009). An nstrument satsfyng all three propertes s sad to be strong and vald nstrument. As used n Meer and Harvey (2004), after testng for valdty and strength, the varables employment status and communty health assocaton membershp were used as nstrument for nsurance. We tested for the endogenety of nsurance and the valdty of nstruments. Frst, we carred out the test for endogenety of health nsurance. If nsurance s exogenous, there would be no justfcaton to estmate the structural model of demand for health care, because the logt models would yeld unbased estmates. We used the Durbn-Wu- Hausman test. The results showed that the Durbn-Wu-Hausman statstc values were sgnfcant at the 10 percent level. 86 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

11 We also conducted the Wald test of exogenety of the nsurance varable whch showed that the values were sgnfcant at the 1 percent level. We then rejected the null hypothess of exogenous nsurance. Second, the coeffcents of nsurance resduals varable were also sgnfcant at the 1 percent level to the demand for medcal care servces. Thrd, we tested the mpact of the nstruments on the dependent varable. These were found to be nsgnfcant. Fourth, the strength of the nstruments was tested by consderng the mpact of the nstruments on endogenous varable. As the coeffcents on nstruments were large and sgnfcant at the 1 percent level, the nstruments were strong. In addton, we conducted the F-test to check the role of the nstruments on the endogenous varable. Whle an F-statstc of at least 10 s recommended (Koko, 2008; Stager and Stock, 1997), the mnmum Egen value statstc for F-test was suggestng that the null hypothess of weak nstrument had to be rejected. A second estmaton ssue s the heterogenety bas whch arses from unobserved factors nteractng wth the varable of nterest and thus basng the results. These are some unobservable preferences and health endowments of ndvduals that nfluence ther demand for health care (Schultz, 2008; Kabubo-Marara et al., 2009). Even wth vald nstruments, t s not easy n practce to separate the mpact of endogenous varable from the effect of unobservables n a structural model. Falure to take nto account heterogenety could lead to unrelable estmates. In ths study, heterogenety may arse from at least three sources. Frst, a rsk reducton effect; where the preferred level of utlzaton s greater because of the fnancal certanty created by nsurance than under uncertanty (Meza, 1983). Second, an access effect; where the nsurance may extend an ndvdual s opportunty set by gvng access to health care that would not otherwse be avalable to them. Nyman (1999) argued that the poolng effect of nsurance provdes access to expensve medcal technologes that would not be affordable. Thrd, an ncome transfer effect where nsurance creates an ex-post transfer of ncome from the healthy to the ll and ths may ncrease utlzaton through an ncome effect on the demand for medcal care (Nyman, 1999). The three sources relate to reasons known by the ndvdual but not by the researcher from whch health nsurance may affect demand for health servces. To handle the problem of heterogenety, we used the Control Functon approach (CFA) (Florens et al., 2008). Ths nvolved estmatng a reduced form nsurance resdual (I*) where the ncluson of the resduals s dentcal to the one obtaned by 2SRI usng an nstrument for nsurance. Assumng the unobserved component s lnear n the nsurance resdual (I*), we ntroduced an nteracton term (of the nsurance and ts resdual (II*)) as a second control varable to elmnate endogenety bas Rwanda Journal Seres B: Socal Scences, Volume 3,

12 even f n the case where the reduced form nsurance s heteroscedastc (Card, 2001). Introducng the control functon varables (nsurance resdual and nteracton) yelds equaton (13). D 0 d Z1 I * II * 1 (13) Where I* s the ftted resduals from the reduced form of the nsurance varable, whch s explaned by Z 1; all other varables are as defned earler. τi* captures the non-lnear ndrect effects of nsurance (I) on demand for health servces (D), because the ftted resduals serve as a control for unobservable varables whch are correlated wth nsurance. Incluson of both I* and the nteracton term II* control for the effects of unobservable factors and therefore purge the coeffcents of the structural equaton of the effects of the unobservables (Card 2001, Ajakaye and Mwabu 2007). If any unobservable varable s lnear n I*, t s only the ntercept n equaton (27) that s affected by the unobservables and therefore the 2SRI estmates are effcent wthout the nteracton term (II*). The 2SRI estmates wll be unbased and consstent f at least one of two condtons holds: Frst, the expected value of the nteracton between nsurance and ts ftted resduals s zero. Second, the expectaton of the nteracton between nsurance and the ftted resduals s lnear (Wooldrdge, 1997). The data used n ths paper s drawn from the Integrated Household Lvng Condtons survey (EICV2) conducted n 2005 by the Natonal Insttute of Statstcs of Rwanda (NISR). Ths natonally representatve survey collected data from 7,620 households and 34,819 ndvduals. Data was collected at the household and the ndvdual level. The EICV2 amed at enablng the government to assess the mpact of the dfferent mplemented polces and programs n mprovng the lvng condtons of the populaton n general. The survey covered all the 30 dstrcts n Rwanda and collected data on a wde spectrum of socoeconomc ndcators, labour, housng, health, agrculture, debt, lvestock, expendture and consumpton n dfferent areas, regons and locatons of the country. Household level nformaton ncluded consumpton expendtures on health, OOPE (consultaton; laboratory tests; hosptalzaton; and medcaton costs). Indvdual level nformaton ncluded soco-economc ndcators and nsurance status. There were also a number of communty varables such as dstance to the nearest health faclty. To mprove relablty of data, the recall perod for the use of health servces was 2 weeks pror to the survey. In ths paper, demand for health care servces was estmated for a sngle vst because the survey dd not capture mult-vsts to health facltes. Hence, the demand for outpatent s lmted to the last consultaton or admsson. 4. Results and dscusson In Table 1, the Wald ch2 tests measurng the goodness of ft ndcate that the estmated models gve an adequate descrpton of the data 88 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

13 because t s hghly sgnfcant mplyng that all models parameters are jontly dfferent from zero. The 2SRI results are reported n columns (4-5) of Table 1 whle the frst stage regresson estmates are gven n Table A2 n the Appendx. Table 1 n columns 6-7 presents results of demand for outpatent care after correctng for heterogenety of nsurance. Due to the ncluson of nsurance resduals and nteracton of nsurance resduals and nsurance, the results reman close to the 2SRI results n terms of sgns of coeffcents although dfferent n magntudes. The sgnfcance of the coeffcent on nsurance resduals suggests that nsurance s endogenous to outpatent medcal demand care. The coeffcent on the nteracton of the nsurance resduals and nsurance s sgnfcant at the 1 percent level ndcatng the presence of heterogenety arsng from nteracton of nsurance wth unobserved determnants of demand for outpatent. For comparson purpose, the baselne model (logt) estmates are also presented n columns (2-3). They appear to be weaker than 2SRI results snce the coeffcent on health nsurance ncreases from 0.49 to 0.9 across model specfcatons (movng from logt to 2SRI) whle the z-value remans statstcally sgnfcant. Ths shows that treatng nsurance as exogenous hghly understates ts mpact on demand for outpatent medcal care. On average hgher user fees reduce the probablty of usng outpatent health servces. Ths fndng s smlar to the results reported by Ltvack and Bodart (1993); Rdde (2003); Dop et al., (1995) and Manj et al (1992) who report negatve effects of user fees on health servce uptake. In partcular, Manj et al., (1992) showed that uptake of treatment n Kenyan schools fell from 75 percent to 19 percent after fees were ntroduced. Ths suggested that the ntroducton of cost-sharng was responsble for the major part of the reducton n uptake. Smlarly, De Bethune et al., (1989) and Yoder, (1989) found the prce of health care to be a sgnfcant hndrance to demand for medcal servces n Swazland. However, ths study has confrmed the results by other cross-secton studes that demand for health care s nelastc to prce. Oxaal and Cook (1998) showed that that the relatonshp between prce and health s nelastc because of falure to dsaggregate ts effect from the one of ncome. The coeffcents on educaton ndcate postve assocaton wth demand for outpatent health servces n Rwanda. The result s consstent wth the work of Katz et al., (2001), whch showed that, the more ndvduals get educated, the more they come nto contact wth other educated ndvduals who have a hgh demand for health care. The socal nteracton whch begns durng schoolng years contnues nto the workplace leads to adopton of health-mprovng behavours, ncludng health servce utlzaton. The evdence from Rwanda s also n lne wth Elo (1992) and Blunch (2004) who observed a strong postve assocaton between educaton and the use of health servces. Rwanda Journal Seres B: Socal Scences, Volume 3,

14 Insurance s found to be an mportant determnant of demand for outpatent medcal servces n Rwanda. Insurance reduces the prce of health care whch makes the servce more affordable than wthout nsurance. The result on nsurance fnds support n fndngs from prevous studes whch addressed the endogenety problem when estmatng demand effect of nsurance (see e.g. Rashad and Markowtz, 2009; Shmeles, 2010; Meer and Harvey, 2004). Smlar results were reported by Phelps and Newhouse (1974) who used data on co-nsurance plans n the Unted States, Canada and the Unted Kngdom. The results were such that the level of senstvty of demand depended on the consurance rate. The evdence presented n the paper reveals that gender s an mportant factor affectng the use of outpatent health servces n Rwanda where females were more lkely to use outpatent servces as compared to men. The results are n lne wth those reported by Mller (1994) who argued that females demand more health care than males because of ther role n chldbearng. Mller (1994) added that some llnesses, such as cardovascular dsease, osteopoross, mmunologc dseases, and Alzhemer s dsease are more prevalent n women than men. In lne wth ths, Ahmad (2001) added that the gender dfferences n health care utlzaton for women were related to specfc dseases such as cardovascular and chronc llnesses. Some research has shown that women use less outpatent health care than men because of the tme they spend takng care of the elderly and other people wth dsabltes. Caregvers, especally women elderly caregvers were found to neglect ther own health n order to fulfll ths responsblty (Fredman et al., 2008). These responsbltes made t dffcult for severely dsadvantaged women to take steps to mprove ther lvng stuatons and health behavors by consumng less health servces than men. Smlarly, Oxaal and Cook (1998) showed that the constrants to access for poor women and grls made them less lkely to have access to approprate care and to seek adequate treatment. Ther paper noted that the range of factors lmtng access for women ncluded soco-economc status of households; tme constrants; composton of households; ntrahousehold resource allocaton and decson-makng, less of educaton and employment; and legal or socal constrants on access to care, heavy work burdens and the opportunty costs of tme n seekng care. Gven the above results, a number of recommendatons emerge. Snce user fees are an mpedment to usng health care n Rwanda, the government should reduce user fees at health facltes through ncreased budget allocatons to all health facltes, partcularly n the publc sector, where the poor go for medcal care. From 2003, OOPE gradually ncreased to reach 32.2 percent of the total health expendtures n Hgh OOPE have a varety of negatve consequences, ncludng household mpovershment. The subsdes on user fees should target the vulnerable groups, such as chldren and women or low ncome 90 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

15 households. The government should also consder subsdzng prvate health facltes to ncrease access to ther hgh qualty servces by low ncome households. The subsdes would help to reduce the effect of ncome nequaltes n health care utlzaton. Health nsurance s an mportant determnant of health care seekng behavour n Rwanda. Thus, polces that ncrease health nsurance coverage would substantally ncrease health servce utlzaton. The 2013 health nsurance coverage rate n Rwanda s 73 percent, the hghest n East Afrcan Communty, but the hgh premums assocated wth ths coverage are not sustanable. The government should subsdze health nsurance to make t accessble to the most dsadvantaged people. The current level of premum of $4.5 for CBHIs per year and per person should be reduced. The premum rate more than doubled n 2011, from $ 1.7 to $ 4.5, and ths reduced the coverage rate from 91 percent to 73 percent. In addton, whle wth the earler premum level, health care expendture represented 10 percent of the total household expendture, holdng other factors constant, wth the new premum, the health care expendture for household would represent 26 percent of the household health expendture. Ths would cause households to ncur catastrophc expendtures and push them nto poverty. Further, wth an average household sze of 6.6 persons per household, ths level of premum per ndvdual does not seem to be sustanable gven that 44.9 percent of the populaton lves wth less than $ 1 per day. Table 2: Logstc Demand Estmates for Outpatent Care: Dependent varable s probablty of an outpatent vst Explanat ory varables Baselne Estmates z- statstcs 2SRI Estmates z- statstcs Control Functon Estmates z- statstcs Househol d ncome *** *** *** User fees *** *** *** Qualty of health care Health nsurance Dstance to the health *** * *** *** *** *** Rwanda Journal Seres B: Socal Scences, Volume 3,

16 faclty Househol d sze ** * ** Age ** * Square age Prmary Secondar y Tertary Male Urban Kgal regon Southern regon Western regon Northern regon Insurance resduals Interacto n of nsurance and nsurance resduals ** ** * * ** ** * * * *** *** ** *** *** *** *** *** ** ** ** *** *** ** *** *** *** 92 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

17 Constant Number of observatons = Durbn-Wu-Hausman ch-sq 0.054* F(1, 5040) = LR *** ch2(19) **** *** Log Lkelhood Note: ***, ** and * = sgnfcant at 1%, 5% and 10% level respectvely Source: Researcher s own constructon REFERENCES 1. Ackerberg D. A. and K. Caves (2003). Structural Identfcaton of Producton Functons: An Applcaton to the Tmng of Input Choce. Department of Economcs, UCLA, Los Angeles, CA Amercan Hosptal Assocaton (2001). Gudelnes for the Evaluaton and Management of Chronc Heart Falure n the Adult. Internatonal Socety for Heart and Lung Transplantaton, Volume 3, Number Ahmad F. (2001). Rural physcans perspectves on cervcal and breast cancer screenng: a Gender-based analyss. J Women s Health Gender-Based Med 5, 2: pp Ajakaye O. and G. Mwabu (2007). The Demand for reproductve health servces: An Applcaton of Control functon Approach. AERC, Narob. 5. Akn J. S., D. K. Grffn and B.M. Popkn (1986). The Demand for Prmary Health servces n The Bcol Regon of the Phlppnes. Economc development and Cultural Change, 34(4): Appleton S. and L. Song (1999). Income and Human Development at the Household level, Evdence from sx countres. Oxford Unversty, Mmeo. 7. Barros P. and P. Machado (2007). Moral hazard and the demand for health servces: a Matchng estmator approach. JEL, C31, l Bategeka L. O., L. Asekeny and J. A. Musme (2009). The Determnants of Brth weght n Uganda. AERC, Narob. 9. Bhargava A., D. Jamson, L. Lau and C. Murray (2001). Modelng the effects of Health on Economc Growth. Journal of Health Economcs 20: Rwanda Journal Seres B: Socal Scences, Volume 3,

18 10. Behrman J.R. and A.B. Deolalkar (1988). Health and Nutrton. Handbook of Development Economcs. Chenery H. and Srnvasan T.N. North Holland, Elsever Scence Publshers BV. 11. Bhasn V. K., C. Obeng and Bentum-Ennn (2010). Fertlty, Income and Household Poverty n Ghana. Journal of Busness and Polcy Research, Volume 5. Number Blomqvst A. G. and R. A. L. Carter (1997). Is health care really a luxury? Journal of Health Economcs, 16(2), Bloom D and D. Cannng (2004). The Effect of Health on Economc Growth: A Producton Functon Approach, World Development V 32, No Blunch N. (2004). Maternal Lteracy and Numeracy Sklls and Chld Health n Ghana. A paper presented at the Northeast Unverstes Development Consortum Conference, HEC Montreal, October Bound J., D. Jaeger and R. Baker (1995). Problems wth Instrumental Varables Estmaton when the Correlaton between Instruments and the Endogenous Explanatory Varables s Weak. Journal of the Amercan Statstcal Assocaton, 90(430): Buchmueller T. C., K. Grumbach, R. Kronck and J. G. Kahn (2005). The Effect of Health Insurance on Medcal Care Utlzaton and Implcatons for Insurance Expanson: A Revew of the Lterature, Medcal Care Research and Revew, Vol. 62 No Card D. (2001). Estmatng the Return to Schoolng: Progress on Some Persstent Econometrc Problems. Econometrca, 69(5): Cunnngham P. and P. Kemper (1998). Ablty to obtan medcal care for the unnsured: How much does t vary across communtes? JAMA 280 (10): De Bethune, X., S. Alfan, and J. P. Lahaye (1989). The nfluence of an abrupt prce ncrease on Health servce utlzaton: evdence from Zare. Health Polcy and Plannng, 4(1), Dop F., A. Yazbeck and R. Btrán (1995).The mpact of alternatve cost recovery schemes on access and equty n Nger. Health Polcy Plan, 10: Elo I. (1992). Utlzaton of Maternal Health-care Servces n Peru; The Role of Women s Educaton. Populaton Studes Center, Unversty of Pennsylvana, PA Florens J.P., M. C. Heckman and E. Vytlacl (2008). Identfcaton of Treatment Effects Usng Control Functons n Models wth Contnuous, Endogenous Treatment and Heterogeneous Effects. Econometrca, 76(5) Fredman L, J.A. Cauley and S. Satterfeld (2008). Caregvng, mortalty, and moblty declne: The Health, Agng, and Body 94 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

19 Composton (Health ABC) Study. Arch Intern Med; 168: Grlches Z. and J. Maresse (1995). Producton Functons: The Search for Identfcaton. Natonal Bureau of Economc Research, Workng Paper No Greene W.L. (2007). Econometrc Analyss. New York. Macmllan Publshng Company. 26. Grossman M. (1972a). On the Concept of Health Captal and the Demand for Health. Journal of Poltcal Economy, 80 (2): Grossman M. (1972b). The Demand for Health: A Theoretcal and Emprcal Investgaton. NBER and Columba Unversty Press, New York. 28. Hahn B. (1994). Health Care Utlzaton: The Effect of Extendng Insurance to Adults on Medcad or Unnsured. MEDICAL CARE, Volume 32, Number 3, pp Hong R. M. Ayad, S. Rutsten, and R. Ren (2009). Chldhood mortalty n Rwanda: levels, trends, and dfferentals. Further analyss of the Rwanda Demographc and Health Surveys. 1992_2007/08. USAID. Calveton, Maryland, USA. 30. Hunt-McCool J., B. F. Kker and Y. C. Ng, (1994). Estmates of the Demand for Medcal Care Under Dfferent Functonal Forms. Journal of Appled Econometrcs, Vol. 9, No Janssen R. (1992). Tme prces and the demand for GP servces. Socal Scences and Medcne 34(7): Jayaraman A. S. Chadndrasekhar and T. Gebreselasse (2008). Factors affectng maternal healthcare seekng n Rwanda. USAID. Workng Paper.Jones A. M., X. Koolman and E. V. Doorslaer (2006). The mpact of Havng Supplementary. Prvate Health Insurance on the Use of Specalsts n European Countres. Anales d Econome et de Statstque. No.83/84, Jowett M., A. Deolalkar and P. Martnsson (2004). Health nsurance and treatment seekng behavour: evdence from a lowncome country. Health Economcs, 13: Jüttng J. P. (2004). Do Communty-Based Health Insurance Schemes Improve Poor People s Access to Health Care? Evdence from Rural Senegal. World Development, 32: Kabubo-Marara J. D. Mwabu and G.K. Nd enge (2009). The Consequences of Fertlty for Chld Health n Kenya: Endogenety, Heterogenety and the Control Functon Approach. AERC, Narob. 36. Katz L., J. Klng and J.Lebman (2001). Movng to opportunty n Boston: Early Results of a Randomzed Moblty Experment. The Quartely Jornal of Economcs 116 (2), Koko M. U. (2009). The Economc Burden of Malara n Kenya: A Household Level Investgaton. PhD Thess, Unversty of Narob. Rwanda Journal Seres B: Socal Scences, Volume 3,

20 38. Krshna A Pathways Out of and Into Poverty n 36 Vllages of Andhra Pradesh, Inda, World Development Vol. 34, No. 2, pp Lawson D. (2004). A Mcroeconomc Analyss of Health, Health Care and Chronc Poverty. Unpublshed, The unversty of Nottngham. 40. Lépne A. and A. Nestour (2008). Health Care Utlzaton n Rural Senegal: the Factors before the Extenson of Health Insurance to farmers. Internatonal labor offce, Research Paper,no Lndelow M. (2002). Health Care Demand n Rural Mozambque: Evdence from 1996/97. Household Survey. Internatonal Food Polcy Research Insttute (IFPRI), FCND, Dscusson Paper, No Ltvack J. I. and C. Bodart (1993). User fees plus qualty equals mproved access to health care: Results of a feld experment n Cameroon. Soc Sc Med ; 37: Mannng W. G., J. P. Newhouse, N. Duan, E. B. Keeler and A. Lebowtz (1987). Health Insurance and the Demand for Medcal Care: Evdence from a Randomzed Experment. Amercan Economc Assocaton Revew, Vol 77, No Manj J.E., S. F. Moses, N.J. Bradley, M.A. Nagelkerke, and F.A. Plummer (1992). Impact of user fees on attendance at a referral centre for sexually transmtted dseases n Kenya. Lancet ; 340: Mathers C.D. (1999). Gans n health expectancy from the elmnaton of dseases among older people. Dsabl Rehabltaton, Vol 21(5 6): Meer J. and S. R. Harvey, (2004). Insurance and the utlzaton of medcal servces. Socal Scence & Medcne 58, McKeown T. (1976). The role of medcne: dream, mrage or nemess? Oxford: Basl Blackwell. 48. Meza D. (1983). Health nsurance and the demand for health care. Journal of Health Economcs 2: Mller L. (1994). Medcal Schools Put Women n Currcula. The Wall Street Journal, p. B Mnstry of Health (MoH), Republc of Rwanda, (2009). Health sector strategc plan.unpublshed. 51. Mnstry of Health Rwanda, (2009). Rwanda Health Fnancng Polcy Revew of Rwanda optons for Unversal Coverage 2009, World Health Organzaton. 52. Mnstry of Health (MoH), Rwanda Health Statstcs Booklet Annual health statstcs booklet Kgal, Unpublshed. 53. Mnstry of Health (MoH), Republc of Rwanda (2012). Annual report Kgal, Unpublshed. 96 Rwanda Journal Seres B: Socal Scences, Volume 3, 2016

21 54. Mnstry of Health (MoH), Rwanda Health Statstcs Booklet Annual health statstcs booklet Kgal, Unpublshed. 55. Mocan N. H., E. Tekn and S. Z. Jeffrey (2004). The Demand for Medcal Care n Urban Chna. World Development, Vol. 32, No. 2, pp , Mwabu G. J., B.Wang ombe and B. Nganda (2003). The Demand for Medcal Care n Kenya. Afrcan Development Bank, Oxford, OX4 2DQ, UK and 350 Man Street, Malden, MA 02148, USA. 57. Natonal Insttute of Statstcs of Rwanda (NISR), (2011). Prelmnary results of nterm 58. demographc and health survey 2010, Kgal: NISR. 59. Nyman J. A. (2005). Health Insurance Theory: The Case of the Mssng Welfare Gan, Unversty of Mnnesot, Unpublshed. 60. Nusselder W.J., K. Van der Velveb, J.L.A., Sonsbeek, M.E. Lenor, G.A.M. Van den Bos. (1996). The Elmnaton of selected Chronc Deseases n a Populaton: The Compresson and Expanson of Morbdty. Amercan Journal of Publc Health: 86: Oxaal Z. and S. Cook (1998). Health and Poverty Gender Analyss. Swedsh Internatonal Development Co-operaton, Unpublshed. 62. Phelps, C.E., and J. P. Newhouse. (1974). Consurance, the Prce of Tme, and the Demand for Medcal Servce. Revew of Economcs and Statstcs 56, pp Rashad I.K. and Markowtz S. (2009). Incentves n Obesty and health Insurance. Inqury, Vol 46: Republc of Rwanda (2010). Unted Natons General Assembly Specal Sesson on HIV and AIDS, Country Progress Report January 2008-December 2009, Unpublshed. 65. Rdde V. (2003). Fees-for-servces, cost recovery, and equty n a dstrct of Burkna Faso operatng the Bamako Intatve. Bull World Health Organ; 81: Rngel, J. S., S. D.Hosek., A. V. Ben, and S. Mahnovsk (2002.) The Elastcty of Demand for Health Care. A Revew of the Lterature and Its Applcaton to the Mltary Health System. Natonal Defense Research Insttute, Unpublshed. 67. Rosenzweg M. R., T. P. Schultz (1982). The Behavor of Mothers as Inputs to Chld Health: The Determnants of Brth Weght, Gestaton, and the Rate of Fetal Growth , n: Fuchs, Vctor R., ed., Economc Aspects of Health, Chcago: The Unversty of Chcago Press. 68. Saksena P., K. Xu, R. Elovano and J. Perrot (2010). Health Servces Utlzaton and Out-of-Pocket Expendture at Publc and Prvate Facltes n Low-ncome Countres. Geneva: World Health Organsaton, Background paper no. 20. Rwanda Journal Seres B: Socal Scences, Volume 3,