WPS I9R8 POLICY RESEARCH WORKING PAPER 1 8 8 8 What Do Doctors Want? Like many large countries, Indonesia has difficulty attracting doctors to service in Developing Incentives for Doctors rural and remote areas. The to Serve in Indonesia's Rural promise of specialist training is and Remote Areas an effective but expensive inducement for service in remote areas. It might help to Kenneth M. Chomitz increase the representation of Gunawan Setiadi Outer Island students in Azrul Azwar medical schools, as they are Nusye Isinail more willing to serve in Widiyarti remote areas. The World Bank Development Research Group March 1998 | POLICY RESEARCH WORKING PAPER 1888 Summary findings Like many large countries, Indonesia has difficulty civil service appointment.) But providing specialist attracting doctors to service in rural and remote areas. training as an incentive to work in remote areas is not To guide the creation of incentives for service in these only expensive, but potentially inefficient, since specialist areas, Chomitz and colleagues analyze two sets of data practice and rural public health management require about physicians: (1) the locations chosen by graduating different skills and attitudes. medical students before and after a major change in the - Moderately (but not extremely) remote areas can be incentive system, and (2) survey data on choices among staffed using modest cash incentives. hypothetical assignments differing in compensation, * Doctors from the Outer Islands are far more willing career prospects, and amenities at various locations. to serve in remote areas than their counterparts from Their findings suggest that: Java. So, it may be worthwhile increasing the The current policy of offering specialist training is representation of Outer Island students in medical incentive enough to make doctors from Java willing to schools (perhaps through scholarships and assistance in serve in remote areas. (It is not necessary to also offer a pre-university preparation). This paper - a product of the Development Research Group - is part of a larger effort in the group to develop methods to analyze public policy in the health sector. The study was funded by the Bank's Research Support Budget under the research project "Incentives for Doctor Placement in Rural and Remote Areas" (Rl-'O 681 -56). Copies of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Teri Charvet, room MC2-522, telephone 202-458-7431, fax 202-522-3230, Internet address mcharvet@worldbank.org. March 1998. (43 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, inteipretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the vimw of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center WHAT DO DOCTORS WANT? Developing Incentives for Doctors to Serve in Indonesia's Rural and Remote Areas Kenneth M. Chomitz' with Gunawan Setiadi2 Azrul Azwar3 Nusye Ismail3 Widiyarti2 1 .Development Research Group, World Bank 2. Ministry of Health, Republic of Indonesia 3. Indonesian Doctors' Association Survey implementation: Sjoijan Dalier and Achmad Noorsubchan Research assistance: Yvonne Ying The findings, interpretations, and conclusions are the authors' own, and are not to be attributed to the World Bank, its Board ofDirectors, or any of its member countries; or to the Government of Indonesia or its Ministry of Health. This study was supported by grant number RPO 681-56from the World Bank's Research Support Budget, and by the Indonesian Ministry of Health. We are particularly grateful to the survey respondents and to the deans of the participating medical schools for their help and cooperation. IYe also thank Howard Barnum for advice and support, and Me ead Over for helpful comments. Address correspondence to: Kenneth M. Chomitz World Bank Washington DC 20433 Email: kchomitz@worldbank.org I 1. INTRODUCTION Understanidably, most physicianis prefer to settle In urban areas offering opportunities for professional developimient, educationl and other amenities for their families, and attractive employmilenit opportun-ities. But it is in rural and remote areas, especially in the developing countries, thiat the most severe public healthi problems are found. As a result, there is a mismatch between the geographic distributioll of physicianis and the perceived need for them. (Anderson and Rosenberg 1990; World Banki 1994, p 142). The geographic distribution of physicians is of particular concerni for Indoniesia. Indonesia's vast size and difficult geography present a tremendous challenige to hiealth services delivery. It is difficult to place doctors in remote island, mountain, or forest locations with few amenities, no opportunlities for private practice, and poor communicationis with the rest of the country. The problem of placing staff in rural areas is further complicated by the rapid growth of lucrative private sector employment prospects in the largest urban areas. In addition, Indolnesia's development goals strongly emphasize equity across regions, with particular stress on improving healthi status in the most remote and poorly served areas. The country's success in placing health centers in all of its subdistricts only increases the challenige of ensurinig that those centers are staffed. To improve the geographical distribution of physicians, governments often hiave used combinationis of compulsory service and incenitives. Incenitives for rural service have been used in the US (Connor et al. 1995), Canada (Bolduc et al forthcomilig; Anderson and Rosernberg 1 990), and Norway (Kristiansen 1992). The evaluationi, and optimal design, of incentive systems requires an understanding of the responsiveness of physicians to these incentives. Little, hiowever, is systematically kinowil about physiciani preferences. For the developed world, there are a handful of studies which apply econometric methods to choice data, most notably Bolduc et al. (forthcominig) for Quebec, and Hurley (1989, 1990) for the United States; see also Kristiansen (1992) for Norway. .For the developing world there are only anecdotal reports. This paper analyzes two complementary sets of data about physician preferences in Indonesia. First, it examines the actual locational choices made by graduating medical students before and 2 after a major change in the incentive system (the revealed preference analysis). Second, it uses survey data on choices among hypothetical assignments to determine physicians' preferences over a set of characteristics describing compensation, career prospects, and locational amenities (the statedpreference analysis.) It builds on, and provides a quaLntitative follow-up, to an earlier, focus-group-based study of contract doctor issues in Indonesia (Soemantri et al 1996). 2. PHYSICIAN DISTRIBUTION AND HEALTH POLICY IN INDONESIA In lndoniesia, most doctors are trained in, and wish to remain in, the cosmopolitan areas of Java. These areas offer good hospital facilities and lucrative opportuniities for private practice. In order to post doctors to the farflung rural health centers of the Indonesian archipelago, the government has long utilized a system of compulsory service for medical school graduates. In the past, those assigned to more remote regions were compensated with shorter periods of service, which ranged from one year at the most remote postings to five years at desirable locations in Java. Following coompletion of compulsory service, doctors were assured positions in the civil service hierarchy, includinig opportunities for specialist training. In 1992, this system was modified. Because of a freeze on new civil service hiring, doctors were hired under contract rather than as civil service employees. Service was (and remains) however a prerequisite for obtaining a license to practice. All service is for a period of three years. Pay is tied to remoteness, however. Currently, doctors serving in 'ordinary' regions are paid Rp 500,000/mo; those in remote regions, Rp 825,000, and those in very remote regions, Rp 1 ,050,000/mo. Because doctors in ordinary regions have much greater opportunities for supplementary earnings from private practice, total income differentials are much smaller. A doctor working in a 24-hour clinic in Jakarta could probably earn about Rp 1,000,000/monti (thoughi this would be illegal if the doctor had not yet completed compulsory service.) Considerinig the difference in amenities associated with remote postings, the salary incentives are relatively modest. Recognizing this, in 1996, the system was further modified. It vvas announced that the doctors serving in very remote regions would be given a 90% chance of subsequent civil service appointment; those serving in remote regions would have a 50% chance, but the probability drops to 10% for those serving in ordinary areas. Because civil servants are eligible for subsidized 3 specialist training, and because specialists enjoy both high prestige and income, this was expected to serve as a powerful inducement. Section 5 of this paper analyzes the impact of this change in incentive structure. A number of important policy questions surround the medium and long-term viability of this policy. From the public viewpoint, the policy of offering civil service appointments and specialist traininlg as an incenitive is effective - as we shall see below -- but expensive. It costs approximately Rp 60 million to provide specialist training, not includinig the value of the student's time. Since the skills appropriate to running a remote health center differ from those appropriate to a clinical specialist, the incentive policy may result in mismatches of personnel to these two quite different positions. Deferring specialist training for the three years of compulsory service reduces tbe return on that training. Finally, usilng civil service positions as an incentive is problematic given the government's policy of zero net growth in the civil service. For these reasons, there is interest in alternative means for attracting doctors to remote locations. One possibility is to assemble alternative incentive packages. Suggested alternatives include higher cash salaries (Soemantri and others 1996), and reduced periods of compulsory service. Another possibility is to intensify recruiting of medical students from rural and remote areas, via scholarships, training programs, or placement of new medical schools. Many doctors are dissatisified with the system. The compulsory nature of the system imposes a particular burden on some graduates, many of them married women, who live in urban Java and are unable for personal reasons to move to other locations. Since there are few openings for compulsory service in these areas, these graduates are indefinitely barred from legally practicing medicine. More generally, graduates of private medical schools question the rationale for compulsory service. So an important question is how much it would cost to attract doctors to health centers if compulsory service were dropped. In many ways the issue parallels the US debate on the all-volunteer army, which also involved considerations of equity, social solidarity, and economic efficiency. Finally, the labor market for doctors is changing rapidly. While there are current concerns about unemployment of young physicians, the rapid growth of incomes in Indonesia, combined with the expansion of health insurance schemes in the formal sector, suggests that there will be very rapid 4 increases in urban demand for physicians. This in turn raises the question of the cost of maintaining physician services in "ordinary" rural regions -- those which are not remote, but which lack urban amenities. Regardless of whether services are financed by public or private sources, it is of interest to forecast the levels of compensating differentials in salaries that might be necessary to retain physicians in these areas. 3. MODELING PHYSICIAN PREFERENCES Revealed and stated preference data Economists' natural tendency is to prefer data on actual choices (revealed preferences) to data on stated preferences. Hurley (1989), Bolduc and others (forthcoming), and section 4 below use this approach. While this approach has the considerable advantage of an incontravertible grounding in reality, it has several disadvantages. Most importantly, the range of variation of the alternatives may be narrow. This can make it impossible to predict doctors' reactions to new policy initiatives which, for instance, offer new types or magnitudes of incentives. Second, it may be difficult to obtain data on the attributes of the choices, as seen by the doctors. Third, there may be considerable multicollinearity among the location-specific attributes of interest, making it difficult to disentangle their impact on utility. Finally, the need to consider a very large number of alternatives requires complex and possibly nonrobust estimation methods. The use of stated-preference data can overcome these problems (Louviere 1994). Stated- preference data has increasingly been used in market research to assess consumer reaction to proposed products, and in environmental economics for valuation of nonmarket goods. In one variant, choice-based conjoint analysis, respondents are asked to choose between hypothetical alternatives, where each alternative is described as a bundle of attributes. Because the researcher has complete freedom to specify the levels of the attributes, these can encompass ranges of policy interest, and can be constructed so as to be free of multicollinearity. Applications of stated-preference discrete-choice models are quite varied. There is a growing literature on dichotomous-choice contingent valuation models (see, e.g. Alberini and others, 1997). These models typically estimate the demand for a hypothetical environmental good or bad. Typically the choices vary only in the presence or absence of the good, and the level of a compensating payment or charge. More akin to the work presented in this paper are studies 5 which look at preferences over bundles of attributes. For instance, Bunch and others (1993) are interested in the potential demand for electric vehicles, and assess consumers' preferences over alternative combinations of purchase price, fuel cost, and range between refueling, in hypothetical electric and gasoline-powered autos. Adamowicz, Louviere, and Williams (1994) analyze demand for water-based recreational site characteristics such as water quality, type and quantity of fish present, and entry fee. The chief risk of the stated preference approach is that stated responses may not accurately reflect behavior. For instance, Cummings and others (1995) asked respondents if they would be willing to purchase specified small items (chocolate, a calculator) at certain prices; the researchers then actually offered the items for sale at those prices. They found that stated willingness to purchase was greater thani actual willingness. However, they note that this may be an unfair test; stated preference respondents may be expressing willingness to purchase at some time in the future, while the actual purchase decision was to be made on the spot. In contrast, Louviere (1994) cites a number of studies showing a good correspondence between predictions derived from stated preference models and actual market behavior. Adamowicz, Louviere, and Williams (1994) retrieved very similar preference functions from a group of respondents using both revealed preference and stated preference analyses. This suggests that respondents can competently manage hypothetical questions, and that responses are at least qualitatively reliable. A more serious problem, especially in the current context, is the potential for strategic misrepresentation, if respondents believe that incorrect information can yield personal gain. In this paper, we utilize both types of surveys, checking for consistency. Utility functions for the nontechnical reader Note: the technical reader may prefer to skip this section. In order to analyze physician preferences, we use an economic approach called utility theory. This approach supposes that doctors choose between alternative assignments as if they have a system for assigning preference scores ('utilities') to each alternative, after which they choose the alternative with the highest score. For instance, suppose that doctors are choosing among alternatives which differ in salary, contract length and remoteness. One doctor might evaluate the positions with a rule like this: add 2 points for every Rp. 1 00,000/month in salary; subtract I I points for every year of contract length; subtract 20 points if the location is remote, and 30 points 6 if it is very remote. Another doctor, with a greater relative desire for cash and tolerance for disamenities, might assign 4 points for every Rp. I OO,OOO/mont:h, but only subtract 5 points for remote location. The methodology used here employs statistical methods to deduice particular scoring systems for particular types of doctors, so that these scoring systems describe doctors' observed choices as well as possible. In the revealedpreference analysis, we look at how doctors' characteristics affect their choices among four broad locational classes of assignments. In the stated preference analysis, we ask doctors to choose among hypothetical assignments described in more detail, focusing on how the characteristics of the assignments affect the doctors' choices among them. Utility function specification Note: nontechnical readers may refer to the previous subsection and skip this one. Utility-theoretic models of physician choice of location have been estimated by Hurley (1989) for the US, Bolduc and others (forthcoming) for Canada. These models are special cases of standard random utility model (see Louviere 1994, Berry 1994): tJ; = f(xi, zj) + uij where Uij is the utility of doctor i for assignmentj, xi are characteristics of doctor i zj are characteristics of assignmentj and uij is a random disturbance term. Doctors are assumed to choose the alternative with the highest utility. The nature of the data, and assumptions about the nature of the disturbance term determine the method of estimation. Strong assumptions about the error terms yield an easy-to-estimate multinomial logit model. The key assumption is the independence of irrelevant alternatives, which is equivalent to assuming that the uij are independent between j; that is, that unobserved factors affecting a doctor's liking for 7 alternative I are uncorrelated with unobserved factors affecting her taste for alternative 2. Wlhere a larger number of closely-related alternatives are considered, this assumption is not tenable. Hurley therefore uses a nested logit model, and Bolduc and others use a sophisticated, computationally demanding multinomial probit model. In the revealed preference analysis below, we address this problem by aggregating the alternatives into arguably distinctive categories. The specification of f depends on the nature of the available data. If there is information about the doctors, but little or no information about attributes of the choices, it is natural to choose the form: f(xi, zj)= xif3 + uii with separate parameter vectors for each of a limited number ofalternativesj. If the disturbances uij are assumed to have independent identical Weibull distributions, then this yields a multinomial logit model, witlh: probability (i chooses j) = exp(xir3j )/V1k xi0k] (where Po are normalized to 0 for some comparison group) This approach was used in the revealed preference analysis. The locational choices were grouped into four categories: j= Java/Bali, the comparison group; outer islands nonremote; outer islanids remote; outer islands very remote. Separate parameter vectors were estimated, via multinomial logit, for each of the categories except the first. Where there is extensive information about characteristics of the choices, it is more convenient to estimate: (]) Uij=Xij+uij where the vector X describes attributes of the choice, the doctor, and their interactioni, and there is now a single parameter vector P. (In practice, separate equations can be estimated for distinct groups, e.g., males and females). This was the approach used in the stated preference analysis. In each choice task, the respondent is presented with two alternatives. Alternative I is chosen if: 8 This condition holds if: (Xi - Xi2) P+ (uijI- Ui2)> 0 If we assume that the disturbance terms u are independently and identically distributed, and define (2) y*= (XiI- Xi2) 3+ (uijI- ui2) y=1 if choice is assignment 1 then we have a simple probit model. Note that there is no constant term, because utilities are defined only up to an additive constant. In addition, the differencing prodedure makes it necessary to exclude one member of each set of mutually exclusive dummy variables to avoid the dummy variable trap. The assumption of independent, homoscedastic error terms facilitates analysis, and has been standard in the literature. It can be thought of as representing a random error as the respondent tries to come up with a utility estimate. It is however a strong assumption, and its failure would mean that estimates are inconsistent. Further, it does not well represent taste heterogeneity, and can be thought of A direction for future work is to allow for taste heterogeneity by through a random coefficients specification (Berry 1993). The earlier framework would be modified to become: Uij = Xij where P3=P* + ei ei -N(O,Q), with ei uncorrelated between individuals Now the vector P* describe mean tastes: the mean coefficient for each attribute. Actual tastes vary between individuals. For instance, individuals with a high marginal utility for income would have a high eincome; those with a high marginal disutility for remote areas would have a negative 9 eremoteness. Tlhe genieral form for the variance-covariance inatrix of e reflects the possibility that tastes for particular attributes are intercorrelated across the population; for instance, the marginal utility of income and remoteness may be negatively correlated. This model can be rewritten in the form of (2) as: y*= (Xi I- Xi2) 3*+ (XiI ei -Xi2ei) where nlow the variance of the error term is a functioni of X I1,Xi2, and Q. This can in principle be estimated by an appropriately-adjusted probit. (Chesher 1995 suggests an alternate but similar approach, heterogeneity-adjusted logit.) However, this becomes computational ly chal lenging where, as in the present case, n would have a large number of parameters to be estimated. Note also that if a respondent replies to multiple choice tasks, the errors will be correlated across those tasks, further complicating estimation. These econometric refinements are left for future work. 4. REVEALED PREFERENCE ANALYSIS1 Background and data Each year, the Ministry of Health assigns two or three batches of graduating medical students to the compulsory contract positions. To facilitate the process, each candidate is asked to specify first, second, and third choices of posting. Each choice consists of both a province and a remoteness category: ordinary or nonremote (biasa), remote (terpencil) and very remote (sangat terpencil). Not all combinations are possible; there are virtually no 'very remote' postinigs in Java, and no 'ordinary' postings in some outlying provinces. To reduce the large number of alternatives to a more manageable set -- and one to which the independence of irrelevant alternatives assumption might plausibly apply -- we aggregated the available choices into four broad alternatives, based on the combination of province and remoteness: Java & Bali Outer Islands Nonremote Alternative 0 Alternative I Remote (comparison Alternative 2 Very remote group) Alternative 3 Note: there are few remote postings and almost no very remote postings in Java-Bali. Primary contributors to this section are Kenneth Chomitz and Gunawan Setiadi. 10 Alternative 0, Java-Bali, coi1sists overwhelmingly of nonremote lpostings. This was used as the comparison group which multinomial logit requires. That is, sets of coefficients were estimated for each of the other three groups. The equation for one of these groups give the predicted logarithm of the odds of choosing that group relative to the compiarison group. We obtained the choice data for the 12th and 13th assignment batches. The latter was the first under the new policy linking service in remote areas to subsequent civil service appointment. Basic demographic and background data were also obtained for each candidate, including age, gender, marital status, number of children, religion, and medical faculty. Because graduates who do not accept an assignment are held over and permitted to bid in subsequent batches, we restricted our attention to those who had registered after the cutolf for the previous batch in order to focus on the choices of new entrants to the process. Appendix table 1 shows definitions and mean values of independent variables for the two batches. Results and discussion Simple cross tabulations (see Table 1) show the powerful impact of the change in incentives. In batch 12, among students graduating from medical faculties in Java/Bali, only 5.6% of males and 1.7% of females volunteered to go to very remote areas in the Outer Islands. In batch 13, those proportions increased to 20.7% and 6.7%. The cross tabulations also show that, regardless of the incentive regime, students from outer island faculties are far more likely to volunteer for remote and very remote outer island assignments than are students from Java/Bali faculties. Multivariate analysis allows us to determine whether these observed relations are due to omitted factors such as age or private schooling. This is potentially important since the mix of doctors (e.g., proportion from private school) varies substantially between batches. The multinomial logit results are shown in appendix table 2 . For both batches 12 and 13, the results are extremely statistically significant. Tables of predicted versus actual choices; are shown below. Predicted versus actual choices: Batch 12 11 Actual I predicted choice choice I Outer Island I Java/Bali nonremote remote very reml Total …__________+…___________________________________________+__________ Java Bali 360 7 12 1 1 380 OI:nonremote 37 23 20 1 1 81 OI: remote 24 11 35 2 1 72 OI: very remote 16 1 20 2 39 …__________+________---___________________________+ …_________ Total I 437 42 87 6 1 572 Predicted versus actual choices: Batch 13 Actual I predicted choice Outer Island I Java/Bali nonremote remote very remote Total …__ _ _ _ _ _ _ _ _ + _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ +…__ _ _ _ _ _ _ _ Java/Bali I 89 31 13 1 1 134 OI: nonremote 43 109 47 23 1 222 OI: remote 19 82 76 21 1 198 OI: very remote 21 23 39 41 1 124 …__ _ _ _ _ _ _ _ _+…__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ +…__ _ _ _ _ _ _ _ Total I 172 245 175 86 1 678 Main results The results are most easily understood by examining the predicted effect of a change in a particular variable on the choice probabilities. Figure I shows the predicted choice probabilities of four classes of students: male/female x Java-Bali medical school /Outer Island medical school, for the two batches2. These predictions closely mirror the cross-tabulation results presented above. There are several important results: 1) Incentives had a large impact on the willingness of Java/Bali graduates to volunteer for remote and very remote posts. Compare the columns marked "Java male 12", "Java male 13". Controlling for other characteristics, the proportion willing to go to ordinary or remote posts in the outer islands 2 Other variables set as follows: graduation year = 1995, birthyear= 1967, unmarried, public school graduate. 12 increases almost threefold, from about 17% to fbouf 50%. The proportion willing to go to very remote posts increases more than fivefold, from 3. 1% to 17.8%. It is particularly striking that incentives also have a strong impact on females, since women are sometimes said to have less flexibility. It is true that women from Java/Bali schools are on average less willing to volunteer for remoter posts (compare the columns "Java male 12" and "Java female 12"). Nevertheless, there was a substantial reponse to the introduction of improved incentives: the proportion willing to go to Outer Island nonremote posts increased from 5.8% to 28.1%; the proportion willing to go to very remote posts increased from 3.5% to 9.5%. 2) Graduates of outer island medical schools are substantially more likely to volunteer for service in remote areas than graduates of Java/Bali schools. This effect is large, statistically significant at the .001 level, ancl especially true in the absence of incentives. In batch 12, for instance, the predicted probability of volunteering for a remote or very remote post is 75% for an outer island male graduate, against 17% for a Java/Bali male graduate. Among women in batch 12, 46% of outer island graduates were predicted to choose outer island remote posts, as opposed to just 5% of Java/Bali graduates. The introduction of improved incentives reduces, but does not eliminate, the differential between outer island and Java/Bali graduates. Differentials by public vs. private schooling In the estimates for batch 13 males, a dummy for private school attendance enters positively and significantly in the equations for nonremote and remote outer island assignments. For batch 12 males, the dummy is positive and significant in the nonremote outer island equation, negative and significant in the very remote outer island equation. The variable is nowhere significant in the estimates for females. The predicted impacts for males, batch 13, are shown in figure 2. Other things equal, males from JavalBali private schools are slightly rnore willing to volunteer for outer island posts, but less willing to volunteer for very remote postings (18% vs. 25%). Marital status and children 13 The effects of marital status and children are weaker thaii might be expected. It is clear from the coefficient estimates that married women with children do not volunteer for posts in very remote areas. Otherwise, however, there are no statistically significant effects of current marital status on women's willingness to volunteer for remote postings. However, it may well be the case that it is expectations about marriage and children which matter, not current status. At this stage in their life cycle, many doctors may anticipate marrying at about the same time that they undertake their compulsory service. Other things equal, currently married men are less willing to volunteer for nonremote outer island and remote outer island posts. For males from Java/Bali schools, being married reduces willingess to go to any outer island post from 68% to 48%. Effect of age and time since graduation In the batch 12 results, age and graduation year had small, statistically insignificant impacts. In batch 13, however, the impacts are profound. Figure 3 shows the predicted probabilities by gender for graduates of Java/Bali schools for three age/graduation date combinations. (The columns are labeled by gender, m/f; age in 1996: and graduation year of '90 or '96.) The results suggest that recent (1996) graduates are far more responsive to incentives than earlier graduates (1990). This is highly plausible, since the latter are more likely to have developed family and employment ties which reduce their flexibility. Age independently reduces the willingness to volunteer for outer island remote postings. Summary Tying service in remote areas to subsequent civil service appointment was, according to this analysis, sufficient to drastically boost the willingess of graduates of Java/Bali schools to serve in outer island and remote areas. However, the post-incentive willingess of these graduates to serve in remote areas was lower than the pre-incentive willingness of graduates from schools in the outer islands. 14 Compare Batch 12 & Batch 13 on predicted probability of choosing outer island assignment, by remoteness (base model) 0.9 0.8 0.7 B,0.6 1 ci 0.5 | 0.4 0.3 0.2 Overy remote 0.1 U remote 0 E riot remote -2 N > 0 0 co I~ ~ ~ C CU CD CD co > > 5 5 Figure 1 Compare base & private school graduates on predicted probability of choosing outer island assignment, by remoteness (batch 13) 0.9 0.8 0.7 0.6 ci0.5 2 0.4 a- 0.3 0.2 0.1 0 very remote 0 E remote Java Java 01 01 Java Java 01 01 Minot remote male male male male female female female female base private base private base private base private Figure 2 15 Predicted probability of choosing outer island assignments, by remoteness: batch 13 for students from Java 0.8 0.7 m 0.5 _ - very remote 2 0.4 E remote a.0.3 0not remote 0.2 0.1 0- LAE O E e E L6 .r e 6 ' C") C, C 10 CN C) D 04 O C'(O 0 a) ) CD 0) a) o2 CD (2 a) co m cao Cam co co c no Figure 3 16 TABLE 1: CHOICES BY STUDENTS OF LOCATIONS FOR PTT SERVICE BATCH 12: STUDENTS FROM JAWA/BALI SCHOOLS Outer Islands I Jawa/Bali biasa terpencil sangat terp. Total …__________+____________________________________________+_-_-…_____ male 1 169 15 18 12 1 214 1 78.97 7.01 8.41 5.61 1 100.00 ___________+____________________________________________+…_________ femalel 202 23 9 4 1 238 1 84.87 9.66 3.78 1.68 1 100.00 ___________+_____________________________________--------_+-______ Totall 371 38 27 16 1 452 1 82.08 8.41 5.97 3.54 1 100.00 BATCH 13: STUDENTS FROM JAWA/BALI SCHOOLS Outer Islands I Jawa/Bali biasa Terpencil sangat terp. Total -----------+_____________________________-- - - ------ + …_________ male 1 63 47 55 43 1 208 I 30.29 22.60 26.44 20.67 1 100.00 ___________+____________________________________________+_________- female 1 68 53 18 10 1 149 1 45.64 35.57 12.08 6.71 1 100.00 …__ _ _ _ _ _ _ _ _+…__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ +…__ _ _ _ _ _ _ _ Totall 131 100 73 53 1 357 1 36.69 28.01 20.45 14.85 1 100.00 BATCH 12: STUDENTS FROM OUTER ISLAND SCHOOLS I Outer Islands I Jawa/Bali biasa terpencil sangat terp. Total _________ -_+--_____________________________+ …_________ male 1 8 18 29 22 1 77 1 10.39 23.38 37.66 28.57 1 100.00 …----------+_____________________________---------------____+…_____ female 1 12 26 20 1 1 59 1 20.34 44.07 33.90 1.69 1 100.00 …-- - - - - - - - -+ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _._ _ _ _ _ _ _ _ _ +…__ _ _ _ _ _ _ _ Totall 20 44 49 23 1 136 1 14.71 32.35 36.03 16.91 1 100.00 BATCH 13: STUDENTS FROM OUTER ISLAND SCHOOLS Outer Islands I Jawa/Bali biasa terpencil sangat terp. Total -----------+___________________________________.…________+ …_________ male 1 3 49 67 60 1 179 1 1.68 27.37 37.43 33.52 1 100.00 -----------+-___________________________________.________+_________- female 1 3 74 59 11 1 147 1 2.04 50.34 40.14 7.48 1 100.00 …__________+_____ ___--___________________.--_____+ …_________ Totall 6 123 126 71 1 326 1. 1.84 37.73 38.65 21.78 1 100.00 17 5. STATED PREFERENCE ANALYSIS Background and motivation The stated-preference analysis was designed to answer questions which could not be answered with the revealed preference analysis, in particular: * How responsive are doctors to increased salary incentives, as opposed to the incentive of civil service appointment? * Is the civil service appointment attractive mostly because of the educational benefits it provides, or also because of the security and status of civil service positions? * Holding constanit the province of assignment, what are the preferences of doctors for different degrees of remoteness? The large sample of the stated-preference survey also allows a more detailed look at the influence of doctor characteristics on doctor preferences. Sample The stated-preference survey was targeted on final-year medical students, who would be shortly facing the task of choosinig locations for their compulsory medical service. The survey covered 14 of lndoniesia's 32 medical schiools, a group comprising 70% of all graduates3. It encompassed public and private schiools, and included three off-Java schools . The survey was administered to 585 final-year students. An effort was made to recruit as many respondents as possible; randomness of selection was therefore sacrificed in order to maximize sample size. Surveys were administered examination-style to groups of students; discussion was prohibited during survey administration. Surveys were conducted over the period May-June 1997. Instrument The survey instrument consisted of two parts: a set of choice tasks, and a series of background questions on respondent characteristics (see appendix B). Each respondent faced a set of eighteen choice tasks. Each task consisted of a choice between two hypothetical job assignments. Based on the latest available data, for 1994. Seven newly-founded schools had no graduates during that year. 18 An assignment was defined by a combination of seven attributes, each with a range of values as follows: Attribute Range Province Jakarta, Central Java, South Sulawesi, Northern Sumatra, Central Kalimantan, East Nusa Tenggara, Irian Jaya Remoteness non-remote, remote, or very remolte. Total monthly income from Rp. 500,000 (approx. $200) to Rp. 3,000,000 (approx. $1200) Length of contract 1, 2, or 3 years Probability of subsequent from 10% to 90% appointment to the civil service Probability of subsequent from 10% to 90% specialist training I An important feature of this framework is the delinking of specialist training from civil service. In practice, these two benefits have always been linked. We know from the revealed preference analysis that the combination of the two serves as a powerful incentive. This analysis allows us to assess the separate impact of each benefit. EXAMPLE; OF A CHOICE TASK Pilihan yang mana lebih disukai? Lingkari I atau 2. Which choice doyou prefer? Circle I or 2. Propinsi (Province) Kalimantan Tengah Nusa Tgara T r Keterpencilan (remoteness) Terpe0#ncilrmote) Sna pe e remote% Tempat ker]a (workplace)_ Puskemas( ealtcter,) Pusesms00 ath0 eter) Lama ikatan kerja (contract 2 tahun I tahun length) (2 years) (1 year,i Pendapatan/hulan, semua Rpl2.500S000 3.0;00M00u0 sumber (monthly income, all sources) : X__________:____:_:_________ Kemungkinan diangkat Pgawa 70% 50% Negeri (probability of appointment as a civilservant) _______:: ___________:______ Kemungkinan spesialisasi 50% 90% (probability of specialist training) _ :_ __D; Box 1 Some combinations of attributes were prohibited as impossible, for instance remote or very remote health centers in Jakarta. (See Annex B4.) The combinations of attributes were randomly generated using the Choice-Based-Conjoint program (Sawtooth Software, n.d.). Fifty different 19 sets of eighteen tasks were generated. An example of a choice task is shown in Box 1. Instructions to the respondents are reproduced in appendix B3. Results Four models were run, for each combination of gender and outer island vs. Java/Bali birthplace. The basic independent variables were differences between choice I and choice 2 in the seven attributes listed above. In addition, private school was interacted with the income, civil service, and specialist training difference variables; the characteristic, "grew up in rural areas or a small town" was interacted with the difference in the very remote and remote dummies; a dummy for "ever failed a course" was interacted with probability of specialist training; and difference variables were created for the variables homeprov and schoolprov. Homeprov (schoolprov) took the value I if the assignment province (school province) was the same as the respondent's birthprovince, zero otherwise. Statistically insignificant interaction variables were dropped from the equations, which were then rerun4. Final estimates are shown in appendix table 4. The robust standard errors allow for intercorrelation among the errors of the 18 responses of a particular respondent. The equations were extremely statistically significant. Approximately 77% of all responses are correctly predicted by the equations.5 The coefficients are almost all plausible and statistically significant. A useful way interpreting them is to express the utility or disutility of each attribute in terms of monetary equivalents. For instance, if the estimated utility function is: U= 10*(salary in thousand rupiahs/mo) - 2000 (contract length in years) + 50000(probability of specialist training) then: * a one year decrease in the contract length is equivalent to a Rp 200 (==2000/1 0) thousand increase in monthly salary; i.e. the cash equivalent of an additional year of contract length is Rp-200,000 (the minus sign indicates that contract length is a disamenity, something that is disliked) 4 Some alternative specifications, including logarithmic terms in income, and interactions of contract length with remoteness of location, did not provide better explanatory power. 5 Keep in mind that about 50% of the responses could be correctly predicted either randomly, or by always choosing '1'. 20 * certainty of receiving specialist training is equivalent to a Rp 5000 (=50000/10) thousand increase in monthly salary; i.e., the cash equivalent of specialty training is Rp 5 million/month. TABLE OF COMPENSATING DIFFERENTIALS This table shows compensating differentials based on the coefficients of appendix table 4. Locational preferences are relative to a private health clinic in Jakarta. Figures shown in italics are noit significantly different from zero6. Note that some private school coefficients are constrained to be equal to the corresponding public school coefficients. Gender Male Male Female Female Male Male Female Female Birthplace Java Java Java Java Outer Is Outer Is Outer Is Outer Is Pub/ Public Private Public Private Public Private Public Private private Length -187 -1,279 -429 -842 +18 -443 -339 (years) I PNS 2189 17 2691 2691 2604 1406 4234 1708 Specialist 9247- 20809- 11349- 11349- 12489.. 13859- 11324- 11324- training 200*age 450*ag 255*age 255*age 322*age 322*age 289*age 289*age e Biasa 36 82 20 20 +272 +272 +539 +539 Terpencil -91 -205 -1303 -1303 -63 -63 -339 -339 Sangat -974 -2194 -3066 -3066 -615 -615 -1475 -1475 Terpencil post in 442 995 835 835 954 954 1821 1821 province of schooling JaTeng -246 -555 +873 +873 0 0 -207 -207 KalTeng -1147 -2514 -1155 -1155 +125 +125 -762 -762 SulSel -829 -1867 -1089 -1089 -263 -263 -1262 -1262 NTT -935 -2105 -1061 -1061 -237 -237 -1619 -1619 SumUt -484 -1088 -727 -727 +39 +39 -748 -748 IrJa -1982 -4462 -3059 -3059 -500 -500 -2472 -2472 6 These significance statements are made on the basis of the coefficient of the variable in question; they should properly be based on the ratio of that coefficient to the coefficient of salary. 21 The results are organized in the table of compensating differentials. Consider the first column, referring to male public school graduates from Java/Bali. On average within this group, and holding location, career prospects and all else equal, graduates would be willing to forgo Rp 187,000 /month in income in order to reduce contract length by one year. (It is important to remember that this represents median preferences -- some would accept a smaller reduction, others would demand a larger one). Private school graduates (second column) have a much higher disutility for contract length. For them, a year of reduced service is worth almost Rp 1.3 million/month. A large value is attached to civil service appointment by public school graduates, even in the absence of specialist training. An increase in the probability of appointment from 0% to 100% is valued on average at Rp 2,189,000/month by male public school graduates from Java/Bali. Continuing down the first column, males from Java/Bali place an extremely high value on specialist training. On average, a 25 year old graduate values this at Rp 4,247,000/month. This is quite reasonable given the increase in earnings associated with specialist education. Value declines with age, reflecting the shorter working lifetime of the training, and perhaps the perceived chance of being deemed to old to enter specialist training. As expected, a huge disutility is attached to very remote (sangat terpencil) location -- this requires a compensating differential of nearly a million rupiah/month, compared to service in a nonremote private clinic. Surprisingly, however, there is no significant disutility attached to service in remote or nonremote health centers relative to nonremote private clinics, controlling for province. This is the most unexpected finding of the analysis. The relative magnitudes of the disutilities attached to the provinces are in accord with popular perceptions of the desirability of these postings for a cosmopolitan Javanese. Central Java is rated as 'worth' Rp 246,000/month less than Jakarta (the comparison location) to a public school graduate, though this difference is not statistically significant. Northern Sumatra, which boasts a large metropolis with good career prospects, has a relatively low disutility attached to it. The remote provinces of Nusa Tenggara Timur (NTT) and Central Kalimantan (KalTeng) require compensating differentials of about Rp 1 million/month. Irian Jaya, the province with the most challenging conditions, requires a differential of nearly Rp 2 million /month, in addition to the 22 differentials for very remote locations. A post in the province of schooling is considered worth Rp. 442,000/month. (A separate variable for posting in home province was not significant.) The second column repeats these calculations for male private school graduates from Java/Bali. There are some striking differences. Essentially no value is attached to a civil service appointment by itself (as opposed to specialist training). The cornpensating differential attached to length of service is very high: Rp -1,279,000/month, versus Rp - 187,000 for the public school counterparts. The differentials attached to service in more remote provinces are about twice as high as those for the public school graduates. Posting in the province of schooling (likely to also be the student's home province) is valued at almost Rp I million/month. The third column describes the preferences of female public school graduates from Java/Bali. Compared to their male counterparts, they have a higher disutility for contract length, with an additional year equivalent to a loss of Rp 429,000/month. As with the males, they place a very high value both on civil service appointment and on specialist training. The locational values show interesting, and expected, contrasts with the males. Unlike the males, the females place a very high disutility on remote (terpencil) locations: these carry a disutility equivalent to a loss of Rp 1,303,000/month. Very remote locations carry an immense penalty of over Rp 3 million/month, three times the disutility expressed by males. The province-specific values are similar to those of the males, with two differences. First, the females actually have a strong preference for Central Java relative to Jakarta. Second, they have on average a much stronger distaste for service in Irian Jaya. Service at a very remote location in Irian Jaya would require a compensating differential of more than Rp 6 million/month in cash, in order to be equivalent to a nonremote post in Jakarta. Turning now to outer island male public school graduates (column 5), there are some interesting contrasts with their Java-born counterparts. The outer island males do not attach any significant disutility to contract length. They prefer nonremote health centers to nonremote private clinics. The disutility they attach to very remote locations (Rp -615,000/rnonth) is one-third less than their Javan counterparts. The province-specific disutilities are quite low compared to the Javans: there is essentially no disutility attached to service in Central Kalimantan, and only Rp - 500,000/month for service in Irian Jaya. Finally, outer island females are intermediate between Javan males and Javan females in terms of their disutility for remote and very remote posts. 23 Outer island females place an extremely higih premium oni service in their province of schooling (Rp 1.8 million/month), but they also place a higih premium on service in their birth province (Rp 1,082,000/monthi, not shown in table). Discussion There are several potential sources of bias in these results. First, the respondents may not have understood, or complied with, the instructionis regarding the choice scenario. For instance, they may believe that choice of a remote location would in fact increase their chance of receiving specialist training, regardless of the probability assigned in the choice task. Second, the respondents may have strategic reasons for misstating their preferences. They may believe, for instance, that they can influence public policy in their favor by overstating the salary or training benefits necessary to induce service in remote areas. Or they may believe that 'socially acceptable' responses indicating willingness to serve in remote areas will in some way yield individual or group benefits. One way to check the results is to compare them to the revealed preference analysis. There is strong qualitative consistency between the two analyses. Both shiow similar differentials in preferences between males and females, between outer island and Java/Bali graduates, and among age or graduation groups. Both show very strong responsiveness to civil-service related incentives. In particular, the increase between batches 12 and 13 in preferences for outer island nonremote and remote locations -- despite the relatively small probabilities of civil service appointment (10% and 50%) -- is consistent with the relatively low disutilities estimated for these positions (except in the least preferred provinces). If there is bias in the results, it is difficult to know in which direction it goes. Consequently, a conservative approach to interpreting the compensating differentials is to accept relative magnitudes, but be cautious about absolute magnitudes. A drawback of the compensating differentials estimates is that, by definition, they describe the value at which respondents with specified characteristics would divide themselves evenly between the two alternatives given. Since doctors may vary in their tastes, it would be of interest to determine how the proportion of doctors choosing, say, service in Irian Jaya as salary increases. This is a complex computational task; our results allow us easily to estimate the 24 proportion who would choose Irian Jaya over Jakarta if there were only those two choices. Of, course, in practice the doctor must choose among a large number of alternatives; our task is to calculate the probability of choosinig each alternative from the complete set, given the predicted preferences between each pair of alternatives. To do this properl y requires a more detailed treatment of heterogeneity of tastes, and will be a topic for follow-on work. Policy implications 1. Problems with the current situation. There are two problems. First, compulsory service is inequitable for medical students, mostly females, who are unable to accept remote or distant postings. The burden imposed by these postings is evident in the huge disutilities attached to these postings-- compulsory service in a very remote area is viewed as equivalent to a 'tax' of Rp 4 million or more a month. These graduates are faced with an unattractive set of alternatives: indefinite unemployment, illegal work as a doctor, or abandonment of their training in favor of a nonmedical career. Alternatively, if public authorities waive the rules and allow these graduates to satisfy their compulsory service in Jakarta or West Java, other graduates may perceive this as inequitable. Second, providing specialist training as an incentive is not only expensive, but inefficient. Doctors who are particularly interested in specialist training may not be much interested in, or suited for, public health work in remote areas. Furthermore, delaying entry into specialist training by three years (of compulsory service) means that doctors do not complete that training until their late thirties or early forties. This significantly reduces the private and social returns to that training. 2. Possible modifications of the current policy a) The estimates suggest that incentive payments for very remote (sangat terpencil) service of the order of Rp I to 1.5 million/month would be attractive to the imedian outer island male graduate. Based on a three year contract, this is less expensive, from the government's point of view, than the cost of paying for specialist training for 90% of these doctors7. The current total W we are assuming that government or social costs of specialist training at public universities are comparable to those at private universities, estimated at about Rp 60 million. 25 supply of outer island male graduates is about 380/year, and the annual number of sangat terpencil posts open is approximately 200. b) However, if the supply of outer island graduates is insufficient to fill the vacancies, then it would be necessary to continue to offer specialist training in order to induce many Javans to work in very remote areas. Cash incentives for graduates from Java would be more expensive thall this training incentive. The results suggest that it is not necessary to offer a civil service appointment in addition to the specialist training, however. c) Remote (te,pencil, as opposed to very remote) posts in the outer islands can be filled by male public school graduates from Java with a salary bonus of about Rp I million/month (above Jakarta private wages -- making a total of about Rp 2 million/month), and from outside Java for a much smaller bonus, of perhaps Rp 300,000 for most provinces. The latter is much smaller, and the former is comparable to, the cost of the current policy of providing specialist training to 50% of the doctors in these posts. d) The above policy could be integrated with one offering indefinite contract renewals to PTT doctors now serving in remote and very remote areas. A companion survey will provide data on1 the potential cost and impact of this policy. e) Perhaps most importantly, both the revealed preference and stated preference results emphasize the much greater willingness of people from the outer islands to serve in remote and very remote locations. This suggests very substantial gains to increasing the representation of outer island students in medical schools. This might be accomplished by scholarships and assistance in pre-university preparation. f) It has frequently been suggested that public healtlh graduates might be trained as healtlh center managers, as an alternative to using physicians as managers. The results presented here would encourage the establishment of such training programs in the outer islands, or for students from the outer islands. These students, lacking the urban employment prospects of doctors, would be expected to be even more willing to volunteer for remote and very remote service than outer island medical students. g) Concurrently with the above policy changes, compulsory service for doctors might be abandoned, at least for students at private schools and for students at public schlools willing to pay unsubsidized tuition fees. 26 REFERENCES Adamowicz, W., J.Louviere and M. Williams. 1994 "Combining Revealed and Stated Preference Methods for Valuing Environmental Amenities." Journal of Environmental Economics and Management, 26(3), 271-292. Alberini, Anna, Barbara Kanninen and Richard T. Carson. 1997. "Dichotomous Choice Contingent Valuation Data". Land Economics 73(3): 309-324. Anderson, M. and M.W. Rosenberg. 1990. "Ontario's underserviced area program revisited: an indirect analysis". Social Science and Medicine 30(1):35-44. Bolduc, D., B. Fortin and M.A. Fournier, forthcoming. "The Impact of lncentive Policies to Influence Practice Location of General Practitioners: a Multinomial Probit Analysis", Journal of Labor Economics. Bunch, David S., Mark Bradley, Thomas F. Golob, Ryuichi Kitamura and Gareth P. Occhiuzzo. 1993. "Demand for Clean-Fuel Vehicles in California: A Discrete-Choice Stated Preference Pilot Project". Transportation Research-A, 27A(3):237-253. Chesher, Andrew and Joao Santos Silva. 1995. Taste Variation in Discrete Choice Models. University of Bristol, Department of Economics. Discussion Paper No. 95/397. Connor, R.A., S.D. Hillson, and J.E. Kralewski. 1995. "An Analysis of Physician Recruitment Strategies in Rural Hospitals". Health Care Management Review 20(1):7-18. Cummings, Ronald G., Glenn W. Harrison, and E. Elisabet Rutstrom. 19,95. "Homegrown Values and Hypothetical Surveys: Is the Dichotomous Choice Approach Incentive-Compatible"? American Economic Review, 85(1): 260-266. Hurley, J. 1990. "Simulated effects of incomes-based policies on the distribution of physicians". Medical Care, 28(3): 221-38. Kristiansen, I.S. 1992. "Medical Specialists' choice of location: the role of geographical attachment in Norway" Social Science and Medicine 34(1): 57-62. Louviere, Jordan J. 1994. "Relating Stated Preference Measures and Models to Choices in Real Markets: Calibration of CV Responses". Paper prepared for the DOE/EPA Workshop on using Contingent Valuation to Measure Nonmarket Values, Herndon VA May 19-20, 1994. Sawtooth Software, n.d. CBC System. Sequim, Washington. Soemantri, S., Cholis Bachroen and Kemal N. Siregar, 1996. Studi Awal Pendayagunaan Dokter PTT. Research and Development Unit, Ministry of Health, Jakarta. Taylor, Donald H. and Thomas C. Ricketts 111. 1992, "Increasing Obstetrical Care Access to the Rural Poor". Journal of Health Care for the Poor and Underserved, 4(1), 9-20. World Bank. 1993. World Development Report 1993: Investing in Health. New York: Oxford University Press. World Bank. 1994. Indonesia's Health Work Force: Issues and Options. Report no. 12835-IND. 27 a) APPENDIX TABLE 1: VARIABLE DEFINITIONS AND STATISTICS, REVEALED PREFERENCE ANALYSIS A. DEFINITIONS age: age in 1996 marital: dummy, married children: number of children oi sch: dummy, medical school outside Java/Bali priv sch: dummy, private medical school spdoctor: dummy, spouse is a doctor graduyr3: year of graduation B. BATCH 12, MALES Variable I Obs Mean Std. Dev. Min Max …________+…____________________________________________________ age 283 29.49823 3.008704 26 49 MARITAL 283 .3038869 .4607492 0 1 CHILDREN I 283 .2155477 .6009791 0 5 oi sch 1 283 .2720848 .4458217 0 1 priv_sch 1 283 .2897527 .4544511 0 1 SPDOCTOR 283 .04947 .2172313 0 1 graduyr3 283 94.72438 .6316189 90 95 C. BATCH 12, FEMALES Variable I Obs Mean Std. Dev. Min Max …________+…____________________________________________________ age 1 290 29.21034 3.096023 25 44 MARITAL 290 .5068966 .5008167 0 1 CHILDREN 290 .3068966 .6326913 0 3 oi_sch 290 .2 .4006914 0 1 priv_sch 1 290 .3827586 .4869003 0 1 SPDOCTOR I 290 .0931034 .2910796 0 1 graduyr3 1 290 94.57931 .8738331 89 95 D. BATCH 13, MALES Variable I Obs Mean Std. Dev. Min Max …________+…____________________________________________________ age 383 28.83812 2.865193 24 43 MARITAL I 383 .2872063 .4530505 0 1 CHILDREN I 383 .227154 .6536999 0 5 oi_sch 1 383 .464752 .4994084 0 1 priv_sch 1 383 .2114883 .4088979 0 1 SPDOCTOR I 383 .0809399 .2730996 0 1 graduyr3 1 383 95.12272 .9588829 89 96 E. BATCH 13, FEMALES Variable I Obs Mean Std. Dev. Min Max _ _ _ _ _ _ _ _-+…-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - age 1 295 28.19322 2.728027 24 44 MARITAL I 295 .4338983 .4964535 0 1 CHILDREN I 295 .2677966 .6542485 0 4 oi_sch 1 295 .4983051 .5008467 0 1 priv sch 1 295 .2542373 .4361715 0 1 SPDOCTOR 295 .1491525 .3568441 0 1 graduyr3 1 295 95 1.020204 88 96 28 APPENDIX TABLE 2A BATCH 12 MALES Multinomial regression Number of obs = 283 chi2(21) = 152.67 Prob > chi2 = 0.0000 Log Likelihood = -233.8589 Pseudo R2 = 0.2461 --------------------------------------------------------------------__-------_ provrmtl I Coef. Std. Err. z P>lzl [95% Conf. Interval] ---------+-______--__________________________________________--_______________ Outer Island nonremote age 1 -.1241292 .1044222 -1.189 0.235 -.328793 .0805346 MARITAL 1 .557725 .5721089 0.975 0.330 -.5635879 1.679038 CHILDREN I -.462538 .5393669 -0.858 0.391 -1.519678 .5946017 oi_sch 1 3.850212 .6010213 6.406 0.000 2.672232 5.028192 priv sch 1 1.336518 .5540064 2.412 0.016 .2506859 2.422351 SPDOCTOR 1 -.3850349 1.183606 -0.325 0.745 -2.704859 1.934789 graduyr3 1 -.3030369 .3156624 -0.960 0.337 -.9217239 .31565 -cons 1 29.15078 30.731 0.949 0.343 -31.08088 89.38245 ________-+------------------------------------------_.---__----------_________ Outer Island remote age 1 -.1405831 .1114062 -1.262 0.207 -.3589352 .077769 MARITAL 1 -.2557866 .6241109 -0.410 0.682 -1.479021 .9674483 CHILDREN 1 -.9073868 .7688577 -1.180 0.238 -2.41432 .5995466 oi_sch 1 3.77406 .5391176 7.000 0.000 2.717409 4.830711 priv sch 1 .2704817 .5528874 0.489 0.625 -.8131577 1.354121 SPDOCTOR 1 .6502684 1.041951 0.624 0.533 -1.391919 2.692455 graduyr3 1 .3249293 .4803456 0.676 0.499 -.6165308 1.266389 cons 1 -29.00467 45.86409 -0.632 0.527 -118.8966 60.88729 ________-+------------------------------------------_----__----------_________ Outer Island very remote age 1 .0332398 .109213 0.304 0.761 -.1808136 .2472933 MARITAL 1 -.6785836 .6438654 -1.054 0.292 -1.940537 .5833695 CHILDREN 1 -.0048397 .4625567 -0.010 0.992 -.9114342 .9017548 oi_sch 1 3.542948 .5598649 6.328 0.000 2.445633 4.640263 priv_sch I -1.674387 .8723153 -1.919 0.055 -3.384094 .0353197 SPDOCTOR 1 -44.20309 . . . . graduyr3 1 -.3902734 .3429819 -1.138 0.255 -1.062506 .2819588 cons 1 33.87223 33.1551 1.022 0.307 -31.11058 98.85503 ----------------------------------------------------------__-----------------_ (Outcome provrmtl==Jawa & B is the comparison group) 29 APPENDIX TABLE 2B BATCH 12, FEMALES Multinomial regression Number of obs = 290 chi2(21) = 115.46 Prob > chi2 = 0.0000 Log Likelihood = -182.25407 Pseudo R2 = 0.2406 provrmtl I Coef. Std. Err. z P>Izl [95% Conf. Interval] Outer Island nonremote age I .013778 .0811935 0.170 0.865 -.1453583 .1729143 MARITAL I -.3688953 .4796584 -0.769 0.442 -1.309009 .5712179 CHILDREN I .2448923 .3496392 0.700 0.484 -.440388 .9301725 oi sch 1 3.437769 .4760645 7.221 0.000 2.504699 4.370838 priv_sch I .5789649 .4831991 1.198 0.231 -.3680879 1.52601-8 SPDOCTOR 1.492807 .60046 2.486 0.013 .3159273 2.669687 graduyr3 -.2098006 .2029136 -1.034 0.301 -.607504 .1879028 -cons 16.83452 19.87388 0.847 0.397 -22.11757 55.7866 …--------+-___________________________________________________________________ Outer Island remote age I .1330548 .0994033 1.339 0.181 -.0617721 .3278817 MARITAL I -1.093825 .641113 -1.706 0.088 -2.350383 .1627334 CHILDREN I .1503781 .4951581 0.304 0.761 -.820114 1.12087 oi_sch I 3.802586 .5881998 6.465 0.000 2.649736 4.955436 priv_sch -.6060551 .7069732 -0.857 0.391 -1.991697 .779587 SPDOCTOR 1.559218 .948813 1.643 0.100 -.300421 3.418857 graduyr3 1 -.0578562 .3555778 -0.163 0.871 -.754776 .6390635 -cons 1 -1.149503 34.18644 -0.034 0.973 -68.1537 65.8547 ---------+-------------------------------------…------------__----____________ Outer Island very remote age I -.093665 .3433381 -0.273 0.785 -.7665954 .5792655 MARITAL 1 -21.81218 1.198711 -18.196 0.000 -24.16161 -19.46274 CHILDREN -39.80506 1.25e+09 0.000 1.000 -2.45e+09 2.45e+09 oi_sch 1.401955 1.250433 1.121 0.262 -1.048849 3.852759 priv sch 1 -.6333743 1.426186 -0.444 0.657 -3.428648 2.161899 SPDOCTOR I 22.55989 . . . . graduyr3 I .3836602 1.07136 0.358 0.720 -1.716167 2.483487 -cons I -36.92709 103.6514 -0.356 0.722 -240.0801 166.2259 (Outcome provrmtl==Jawa & B is the comparison group) 30 APPENDIX TABLE 2C BATCH 13, MALES Multinomial regression Number of obs = 383 chi2(21) = 143.56 Prob > chi2 = 0.0000 Log Likelihood = -449.86333 Pseudo R2 = 0.1376 provrmtl I Coef. Std. Err. z P>Iz| [95% Conf. Interval] Outer Island nonremote age I -.1111011 .0883762 -1.257 0.209 -.2843153 .0621132 MARITAL -.9645574 .5889225 -1.638 0.101 -2.118824 .1897094 CHILDREN I .2506882 .4874255 0.514 0.607 -.7046482 1.206025 oi_sch 1 3.685305 .7065425 5.216 0.000 2.300507 5.070103 priv_sch 1.433465 .5008063 2.862 0.004 .4519027 2.415027 SPDOCTOR .7825769 .7323342 1.069 0.285 -.6527717 2.217925 graduyr3 1 .7742842 .2330639 3.322 0.001 .3174874 1.231081 cons I -70.91739 22.88618 -3.099 0.002 -115.7735 -26.06129 …--------+-___________________________________________________________________ Outer Island remote age I -.2055039 .0901082 -2.281 0.023 -.3821126 -.0288951 MARITAL I -1.401922 .6110201 -2.294 0.022 -2.599499 -.2043445 CHILDREN .4401251 .4939446 0.891 0.373 -.5279886 1.408239 oi_sch 1 3.722087 .6970834 5.340 0.000 2.355828 5.088345 priv sch I .8937679 .5079367 1.760 0.078 -.1017698 1.889306 SPDOCTOR I .9549834 .7578044 1.260 0.208 -.5302859 2.440253 graduyr3 1 .5810809 .218274 2.662 0.008 .1532717 1.00889 -cons I -49.44252 21.49635 -2.300 0.021 -91.5746 -7.310448 ---------+…___________________________________________________________________ Outer Island very remote age I .0014752 .0854615 0.017 0.986 -.1660261 .1689766 MARITALI -.3871185 .5524567 -0.701 0.483 -1.469914 .6956767 CHILDREN I .1742179 .4723127 0.369 0.712 -.7514981 1.099934 oi_sch 1 3.526986 .6991653 5.045 0.000 2.156647 4.897325 priv sch I .4040116 .5191938 0.778 0.436 -.6135895 1.421613 SPDOCTOR -1.970863 1.182279 -1.667 0.096 -4.288088 .3463611 graduyr3 1 1.024132 .2423728 4.225 0.000 .5490897 1.499174 -cons -97.61006 23.76774 -4.107 0.000 -144.194 -51.02614 (Outcome provrmtl==Jawa & B is the comparison group) 31 APPENDIX TABLE 2D BATCH 13, FEMALES Multinomial regression Number of obs = 295 chi2(21) = 145.09 Prob > chi2 = 0.0000 Log Likelihood = -294.10069 Pseudo R2 = 0.1979 provrmtl I Coef. Std. Err. z P>IZI [95% Conf. Interval] Outer island nonremote age I -.1189277 .0936575 -1.270 0.204 -.3024931 .0646376 MARITAL I .0359726 .5088964 0.071 0.944 -.9614459 1.033391 CHILDREN .0711081 .3554449 0.200 0.841 -.6255511 .7677673 oi_sch 3.82478 .6972768 5.485 0.000 2.458143 5.191418 priv sch I .6348682 .5045185 1.258 0.208 -.3539698 1.623706 SPDOCTOR I .833417 .5787424 1.440 0.150 -.3008973 1.967731 graduyr3 I .5998963 .2214786 2.709 0.007 .1658063 1.033986 -cons 1 -54.20297 21.7051 -2.497 0.013 -96.74419 -11.66175 ---------+…___________________________________________________________________ Outer Island remote age I -.1335339 .1093644 -1.221 0.222 -.3478842 .0808164 MARITAL I -.7044551 .6165873 -1.143 0.253 -1.912944 .5040337 CHILDREN I .1420596 .4351483 0.326 0.744 -.7108153 .9949345 oi sch 1 4.928999 .7504803 6.568 0.000 3.458085 6.399913 priv sch 1 .7270435 .6213775 1.170 0.242 -.490834 1.944921 SPDOCTOR 2.103193 .7086708 2.968 0.003 .7142236 3.492162 graduyr3 1 .9449086 .3007622 3.142 0.002 .3554256 1.534392 -cons 1 -87.82871 29.26284 -3.001 0.003 -145.1828 -30.4746 ---------+…___________________________________________________________________ Outer Island very remote age I -.0321541 .1300435 -0.247 0.805 -.2870347 .2227266 MARITAL | -.8309908 .8888162 -0.935 0.350 -2.573038 .9110568 CHILDREN I -37.65063 1.21e+08 0.000 1.000 -2.38e+08 2.38e+08 oi_sch 1 3.505597 .8378328 4.184 0.000 1.863475 5.147719 priv sch .2755101 .8430683 0.327 0.744 -1.376874 1.927894 SPDOCTOR 1.03143 1.153166 0.894 0.371 -1.228735 3.291594 graduyr3 .7891714 .4251077 1.856 0.063 -.0440243 1.622367 cons I -75.77971 41.28481 -1.836 0.066 -156.6964 5.137018 ----------------------------------------------------------__-----------------_ (Outcome provrmtl==Jawa & B is the comparison group) 32 APPENDIX TABLE 3: STATED PREFERENCE VARIABLE DESCRIPTION VARIABLE DEFINITIONS Examples: dlama= delta contract length = contract length in alternative 1 - contract length in alterntative 2 djat = JATENG1-JATENG2 where JATENGi=l if alternative i is in Central Java, =0 otherwise 35. dlama delta contract length 36. dgaji delta salary, 00ORp/mo 37. dpns delta civil service probability 38. dspe delta specialist probability 39. dbias delta nonremote 40. dterp delta remote 42. djat delta Central Java 43. dkalt delta Central Kalimantan 44. dsuls delta South Sulawesi 45. dntt delta East Nusa Tenggara 46. dsumut delta North Sumatra 47. dirja delta Irian Jaya 74. s dlama private school*delta contract length 75. s_dgaji private school *delta salary, 00ORp/mo 76. s_dpns private school*delta civil service probability 77. s_dspe private school *delta specialist probability 91. dhomeprv delta (post=birth province) 92. dsekprop delta (post=school province) 105. dumurspe delta age avail for training*specialist probability 101. ddesst delta rural background*very remote 101. ddesst delta rural background*very remote 106. dluluspe delta ever failed a course*specialist probability 33 MALE JAVA-BALI: SUMMARY STATISTICS Variable I Obs Mean Std. Dev. Min Max --------------------------------------------------------------- dlama 1 3690 .0520325 1.40228 -2 2 dgaji 1 3690 65.78591 1311.786 -2500 2500 dpns 1 3690 2.905149 44.91146 -80 80 dspe 1 3690 -.7750678 45.77866 -80 80 dbias 1 3690 .0168022 .7025798 -1 1 dterp 1 3690 .0192412 .7046363 -1 1 djat 1 3690 -.0075881 .5415227 -1 1 dkalt 1 3690 .0062331 .5260512 -1 1 dsuls 1 3690 -.0054201 .5455385 -1 1 dntt 3690 .0170732 .525811 -1 1 dsumut 3690 .0170732 .5380417 -1 1 dirja 3690 -.0303523 .5239197 -1 1 s_diama 3690 .0162602 .6778501 -2 2 s dgaji 1 3690 24.52575 636.872 -2500 2500 s_dpns 1 3690 1.143631 21.65536 -80 80 s dspe 1 3690 -.6287263 22.22718 -80 80 dhomeprv 1 3690 -.003252 .3192398 -1 1 dsekprop 1 3690 .001626 .362213 -1 1 dumurspe 3690 -25.26287 1305.238 -3110 2980 ddesst 1 3690 -.0081301 .3832195 -1 1 ddesst 1 3690 -.0081301 .3832195 -1 1 dluluspe 1 3690 -1.154472 35.79512 -80 80 FEMALE JAVA-BALI: SUMMARY STATISTICS Variable I Obs Mean Std. Dev. Min Max _________+…____________________________________________________ dlama 1 3276 -.0128205 1.408314 -2 2 dgaji 3276 77.45726 1296.741 -2500 2500 dpns 3276 1.782662 44.34146 -80 80 dspe 3276 -1.117216 45.37448 -80 80 dbias I 3276 .0363248 .7004202 -1 1 dterp 3276 .0155678 .7102749 -1 1 djat 3276 -.0094628 .5421773 -1 1 dkalt 1 3276 .0164835 .5251274 -1 1 dsuls 3276 -.0027473 .5411256 -1 1 dntt 3276 .0030525 .5265383 -1 1 dsumut 3276 .0128205 .5412627 -1 1 dirja 1 3276 -.0357143 .5232959 -1 1 s dlama 1 3276 .01221 .903515 -2 2 s_dgaji 1 3276 11.5232 839.4387 -2500 2500 s_dpns 3276 .3113553 28.47408 -80 80 s_dspe 3276 -.3540904 29.47474 -80 80 dhomeprv 1 3276 .0015263 .3710897 -1 1 dsekprop 3276 .0088523 .4145238 -1 1 dumurspe 3276 -32.83272 1284.412 -2970 2870 ddesst 3276 -.009768 .2794146 -1 1 ddesst 1 3276 -.009768 .2794146 -1 1 dluluspe 1 3276 .4517705 32.66546 -80 80 34 MALE OUTER ISLAND: SUMMARY STATISTICS Variable I Obs Mean Std. Dev. Min Max _________+-____________________________________________________ dlama 2016 .0287698 1.408646 -2 2 dgaji 1 2016 73.53671 1303.728 -2500 2500 dpns 2016 1.964286 44.45323 -80 80 dspe 2016 -.6746032 46.0913 -80 80 dbias 1 2016 -.0049603 .7037476 -1 1 dterp 1 2016 .0262897 .7029208 -1 1 djat 1 2016 -.0292659 .5416972 -1 1 dkalt 1 2016 .0218254 .5229425 -1 1 dsuls 1 2016 .0039683 .5401811 -1 1 dntt 2016 .0034722 .5229121 -1 1 dsumut 1 2016 .0257937 .5459795 -1 1 dirja 1 2016 -.015377 .5274233 -1 1 s dlama 1 2016 -.0059524 .6532714 -2 2 s_dgaji 1 2016 23.80952 612.8209 -2500 2500 s_dpns 2016 1.10119 20.63454 -80 80 s_dspe 1 2016 -.0396825 21.27449 -80 80 dhomeprv 1 2016 .0104167 .4448666 -1 1 dsekprop 1 2016 .0138889 .4999311 -1 1 dumurspe 1 2016 -20.59524 1339.665 -2810 2810 ddesst 1 2016 -.000496 .3707678 -1 1 ddesst 1 2016 -.000496 .3707678 -1 1 dluluspe 1 2016 -.7142857 42.65024 -80 80 FEMALE OUTER ISLAND: SUMMARY STATISTICS Variable I Obs Mean Std. Dev. Min Max _ _ _ _ _ _ _ _ _+…__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _- _ _ _ _ _ _ _ _ _ _ _ _ _ _ dlama 1548 -.0180879 1.403084 -2 2 dgaji 1 1548 123.2235 1269.179 -2500 2500 dpns 1 1548 2.131783 44.61836 -80 80 dspe 1 1548 -.5167959 45.58027 -80 80 dbias 1 1548 .0691214 .698879 -1 1 dterp 1 1548 .005168 .7027321 -1 1 djat 1548 .0090439 .5392622 -1 1 dkalt 1 1548 -.0116279 .5283132 -1 1 dsuls 1 1548 -.005814 .5387071 -1 1 dntt 1 1548 .0180879 .5305737 -1 1 dsumut 1548 .0122739 .5361929 -1 1 dirja 1 1548 -.0335917 .5285961 -1 1 s_dlama 1 1548 -.0103359 .7217351 -2 2 s_dgaji 1548 45.21964 641.2277 -2500 2500 s_dpns 1 1548 .6847545 23.26423 -80 80 s_dspe 1 1548 .0904393 23.05087 -80 80 dhomeprv 1548 .0174419 .4181786 -1 1 dsekprop 1 1548 .0103359 .5096576 -1 1 dumurspe 1 1548 -13.02326 1301.994 -2820 2730 ddesst 1 1548 -.0200258 .3012355 -1 1 ddesst 1 1548 -.0200258 .3012355 -1 1 dluluspe 1548 -.6976744 39.01209 -80 80 35 APPENDIX TABLE 4: STATED PREFERENCE ESTIMATES 4A. MALES, JAVA-BALI Probit Regression with Huber standard errors Number of obs = 3690 Log Likelihood =-1785.0489 Pseudo R2 Grouping variable: id ----------------------------------------------------------------__-----------_ pill I Coef. Std. Err. z P>Izl [95% Conf. Interval] ---------+…___________________________________________________________________ dlama 1 -.0704826 .0351163 -2.007 0.045 -.1393093 -.0016559 dgaji 1 .0003754 .0000291 12.912 0.000 .0003184 .0004323 dpns I .0082162 .0009159 8.971 0.000 .0064211 .0100113 dspe I .0347098 .0116248 2.986 0.003 .0119257 .0574939 dbias I .0136313 .0382208 0.357 0.721 -.0612801 .0885428 dterp I -.0341724 .061449 -0.556 0.578 -.1546103 .0862655 dst -.3659125 .0687266 -5.324 0.000 -.5006141 -.2312109 djat -.0925484 .0683281 -1.354 0.176 -.2264691 .0413723 dkalt -.4192969 .0915411 -4.580 0.000 -.5987142 -.2398796 dsuls -.3114099 .0736019 -4.231 0.000 -.455667 -.1671528 dntt - -.3510476 .0833915 -4.210 0.000 -.5144919 -.1876033 dsumut I -.1815521 .0696759 -2.606 0.009 -.3181144 -.0449897 dirja I -.7443214 .096986 -7.675 0.000 -.9344104 -.5542324 s_dlama I -.1429247 .0497177 -2.875 0.004 -.2403696 -.0454797 s_dgaji -.0002086 .0000536 -3.894 0.000 -.0003135 -.0001036 s_dpns I -.005407 .0015072 -3.588 0.000 -.008361 -.002453 dsekprop I .1659393 .0837393 1.982 0.048 .0018132 .3300654 dumurspe I -.0007512 .0004064 -1.848 0.065 -.0015477 .0000454 4B. FEMALES, JAVA-BALI Probit Regression with Huber standard errors Number of obs = 3276 Log Likelihood =-1464.7073 Pseudo R2 = Grouping variable: id ----------------------------------------------------------------__-----------_ pill I Coef. Std. Err. z P>IzI [95% Conf. Interval] …--------+…___________________________________________________________________ dlama I -.1081921 .0383393 -2.822 0.005 -.1833358 -.0330484 dgaji I .0002524 .0000224 11.272 0.000 .0002085 .0002963 dpns I .0067926 .0006654 10.208 0.000 .0054884 .0080967 dspe I .0286451 .014069 2.036 0.042 .0010704 .0562199 dbias I .0051289 .0412365 0.124 0.901 -.0756933 .085951 dterp I -.3289197 .0728737 -4.514 0.000 -.4717496 -.1860897 dst 1 -.7738035 .0889796 -8.696 0.000 -.9482004 -.5994066 djat I .220269 .0764291 2.882 0.004 .0704706 .3700673 dkalt -.2914487 .1100936 -2.647 0.008 -.5072282 -.0756693 dsuls I -.2748354 .0904313 -3.039 0.002 -.4520775 -.0975933 dntt I -.2678134 .1135186 -2.359 0.018 -.4903057 -.0453211 dsumut -.1834204 .0798705 -2.296 0.022 -.3399637 -.026877 dirja I -.7720885 .1123198 -6.874 0.000 -.9922311 -.5519458 s_dlama I -.1044407 .0448962 -2.326 0.020 -.1924358 -.0164457 dhomeprv I .1820854 .1067371 1.706 0.088 -.0271155 .3912862 dsekprop I .2107087 .1057574 1.992 0.046 .003428 .4179894 dumurspe I -.0006431 .000503 -1.278 0.201 -.001629 .0003428 -----------------------------------------------------------__----------------6 36 4C. MALES FROM OUTER ISLANDS Probit Regression with Huber standard errors Number of obs = 2016 Log Likelihood =-981.98104 Pseudo R2 Grouping variable: id pill I Coef. Std. Err. z P>Izl [95% Conf. Interval] …--------+…___________________________________________________________________ dlama I .0070333 .0445805 0.158 0.875 -.080343 .0944095 dgaji I .000391 .0000365 10.714 0.000 .0003195 .0004626 dpns .0101825 .0012098 8.417 0.000 .0078114 .0125537 dspe I .0488409 .0168881 2.892 0.004 .0157409 .081941 dbias .106396 .0536511 1.983 0.047 .0012419 .2115502 dterp -.024822 .0987274 -0.251 0.801 -.218324 .1686801 dst -.2406674 .109521 -2.197 0.028 -.4553246 -.0260103 djat I .0000529 .0791665 0.001 0.999 -.1551107 .1552164 dkalt I .0490552 .1065622 0.460 0.645 -.159803 .2579133 dsuls -.102688 .0924082 -1.111 0.266 -.2838048 .0784288 dntt -.0925702 .1110943 -0.833 0.405 -.3103109 .1251706 dsumut I .0151435 .0750817 0.202 0.840 -.1320139 .162301 dirja I -.1956504 .1147449 -1.705 0.088 -.4205463 .0292455 s_dlama I -.1802936 .0848115 -2.126 0.034 -.3465211 -.0140661 s_dpns I -.0046865 .0019581 -2.393 0.017 -.0085242 -.0008487 s_dspe .0053577 .0033587 1.595 0.111 -.0012252 .0119406 dsekprop I .3731865 .0861189 4.333 0.000 .2043966 .5419764 dumurspe -.001259 .0005891 -2.137 0.033 -.0024136 -.0001044 4D. FEMALES FROM OUTER ISLANDS Probit Regression with Huber standard errors Number of obs 1548 Log Likelihood = -745.2308 Pseudo R2 Grouping variable: id pill I Coef. Std. Err. z P>Izl [95% Conf. Interval] dlama I -.0924159 .0422767 -2.186 0.029 -.1752767 -.0095551 dgaji .0002725 .0000332 8.197 0.000 .0002073 .0003376 dpns I .0115387 .001592 7.248 0.000 .0084184 .014659 dspe I .0308539 .0148516 2.077 0.038 .0017454 .0599625 dbias .147008 .0616457 2.385 0.017 .0261847 .2678313 dterp I -.1059933 .1039206 -1.020 0.308 -.309674 .0976874 dst I -.4019779 .1192324 -3.371 0.001 -.6356692 -.1682867 djat I -.0563346 .0896582 -0.628 0.530 -.2320614 .1193923 dkalt -.2076515 .1508219 -1.377 0.169 -.503257 .087954 dsuls I -.343934 .109064 -3.154 0.002 -.5576956 -.1301724 dntt I -.4410467 .1576276 -2.798 0.005 -.7499912 -.1321022 dsumut I -.2040547 .1067864 -1.911 0.056 -.4133521 .0052428 dirja I -.6736021 .1497839 -4.497 0.000 -.9671732 -.380031 s_dpns I -.0068558 .0025871 -2.650 0.008 -.0119264 -.0017851 dhomeprv .294753 .1304114 2.260 0.024 .0391514 .5503547 dsekprop .4961757 .1066754 4.651 0.000 .2870958 .7052555 ddesterp .3175753 .1393712 2.279 0.023 .0444128 .5907378 dumurspe I -.0007887 .0005136 -1.536 0.125 -.0017954 .000218 37 APPENDIX B 1: MEDICAL STUDENT SURVEY: QUESTIONNAIRE COVER SHEET Ikatan Dokter Indonesia (PB IDI) PT CORINTHIAN Infopharma Corpora JI. Dr. Sam Ratulangi No. 29 (A member of The CIC Consulting Group) Jakarta 10350 Jl. Raden SaJelie No. 46 - 48 Jakana Pusat 10330 Telp. (021) 3150)679. 3900473 Telp. (021) 324335, 3150345, 3155357, 3155358 SURVEY PREFERENSI PENEMPATAN I Pengurus Besar Ikatan Dokter Indonesia (PB IDI) bekerja sama dengan PT CORINTHIAN Infopharma Corpora, a member of the CIC Consulting Group, suatu perusahaan konsultan independen di bidang riset pemasaran, saat ini sedang melakukan penelitian mengenai preferensi penempatan dokter, dengan mahasiswa Fakultas Kedokteran tingkat klinik sebagai target responden. Penelitian ini bertujuan untuk mengetahui motivasi dokter untuk ditempatkan di suatu daerah tertentu dengan berbagai bentuk kompensasi dan kondisi pekerjaan, sehingga hasil penelitian diharapkan dapat bermanfaat bagi pengembangan usul-usul kebijakan dalam memajukan sistem penempatan dan pendayagunaan tenaga dokter di Indonesia. Oleh karena itu, kami sangat mengharapkan kesediaan saudara sebagai sampel terpilih untuk meluangkan waktu menjawab pertanyaan yang akan kami ajukan. Nama saudara sebagai responden tidak perlu dicantukan sehingga kerahasiaan identitas terjamin dan dikemudian hari tidak mempengaruhi tugas/karir saudara. Atas kesediaan dan partisipasinya, kami ucapkan terima kasih. A. BAGIAN PERTAMA Wilayah penelitian.: 1' Medan I- 1 Bandung P Yogyakarta 6 Bali 4.Jakarta . - Semarang a Surabaya . - - Ujung Pandang U-lniversitas:u; -0; - Unvt6ersitas Sumatera Utara (USU) ; Universitas Tarinmanegara (UNTAR) li2ti Universitas Indonesia (UL) U 1 Universitas Trisakti C Universitas Padjajaran (UNPAD) Universitas Yarsi Cg Universitas Diponegoro (UNDIP) -: -1 Universitas Kristen Indonesia (UKI) I3P Universitas Gajah Mada (UGM) e7 Universitas Kristen Krida Wacana UKRIDA C3 Universitas Airlangga (UNAIR) 2 Universitas Katholik Indonesia Atmajaya O Universitas UDAYANA L1 Universitas lainnya O-/a Universitas Hasanuddin (UNHAS) (sebutkan ..............I....... ...... ......... 38 Draft 1.0. Not for citation or quotation APPENDIX B2: Background questionnaire, survey of medical students Note: a rough English translation has been added in italics C. l. Tempat kelahiran responden Kota ............................ Propinsi . Birthplace: city province C.2. Tahun kelahiran responden 1 l9 El birth year C.3. Catat Jenis kelamin (tanpa ditanyakan): Gender Laki-laki male [01] Perempuanfemale [02] C.4. Agama: Religion Islam [03] Budha [06] Kristen Protestan [04] Hindu [07] Katholik [05] ]ainnya . [081 C.5. Status perkawinan: marital status Kawin married [01] If unmarried, continue at C.9 Belum kawin unmarried [02] Bila belum menikah, dilanjutkan ke C.9 C.6. Bila sudah menikah, apakah istri/suami* anda bekerja? (* coret yang tidak perlu) if married, does your spouse work? Ya yes [01] Tidak no [02] C.7. Jumlah anak: number of children If no children, continue at C. 9 * Belum punya anak none [00] I Bila belum punya anak, dilanjutkan ke C.9 * Sudah punya anak: one or more: Satu orang [Ol] Tiga orang [03] Dua orang [02] Empat orang [04] Lebih dari empat more than 4 [05] C .8. Umur anak tertua : age of oldest child Kurang dari 3 tahun < 3 yrs [01] 7 - 8 tahun [04] 3 - 4 tahun [02] 9 - 10 tahun [05] 5 - 6 tahun [03] Lebih dari 10 tahun > 10 [06] C.9. Tahun masuk Universitas ? Tahun 19 OI L Year entered university C. 10. Sekarang ini, saudara sudah semester berapa ? Semester []LI Current semester in school C. 11. Menurut perkiraan saudara, kapan lulus dari universitas ini ? Tahun 19 LIE expected year of graduation C. 12. Apakah saudara pernah tidak lulus dalam mata kuliah ? Have you ever failed a course? a. Pre Klinik Pernah Yes [Oll; berapa kali ? ..number of times .... kali Tidak pernah [02] b. Klinik Pernah [01]; berapa kali ? .number of times . kali Tidak pernah [02] 39 Draft 1.0. Not for citation or quotation C. 13. Pekerjaan orang tua: Parent's occupation Petani/Nelayan [Oil Karyawan swasta setingkat supervisor [09] farming/fishing supervisory-level employee Pedagang [02] Karyawan swasta setingkat manajer [10] petty trade manager Wiraswasta (TK' < 5 orang) [03] ABRI setingkat Bintara [11] TK = Tenaga Kerja (# employees) armed forces small entrepreneur ABRI setingkat Perwira [12] armed forces Wiraswasta (TK 5 - 50 orang) [04] Tenaga terampil (bengkel/elektronik/dsb) [13] medium entrepreneur skilled wvorker Wiraswasta (TK > 50 orang) [05] Profesional (pengacara, artis, dsb) [14] large entrepreneur Pegawai negeri golongan I-If [06] Civil servant, lower rank Pensiunan/Purnawirawan [15] Pegawai negeri golongan III-IV [07] retiree Civil servant, higher rank Lainnya (sebutkan . ') [16] Karyawan swasta biasa [08] other, specify 'ordinary' wvorker C. 14. Di mana saudara paling lama bertempat tinggal/dibesarkan ( sebelum mahasiswa ) ? Where were you brought up (longest residence, before becoming a student) Desa [00] Ibukota propinsi kecuali Jakarta [03] village provincial capital except Jakarta Ibukota kecamatan [01] Jakarta [O4] subdistrict town Ibukota Kabupaten/Kotamadya [02] Luar Negeri [05] district capital abroad Lainnya, sebutkan .[06] other, specify 40 Draft 1.0. Not for citation or quotation APPENDIX B3: Instructions to respondents, medical student questionnaire SURVEY PREFERENSI PENEMPATAN DOKTER PETUNJUK PENGISIAN A. Bagian Pertama ( halaman depan ) * Merupakan identitas responden secara umum yang terdiri dari wilayah penelitian ( nama kota ) dan universitas. Lingkari jawaban yang sesuai Bila nama universitas tidak tercantum dalam daftar, lingkari nomor 15 ( Universitas lainnya ) dan tulis nama universitas tersebut. B. Bagian Kedua, merupakani pertanyaan utama tentang prefererisi dokter/calon dokter. * Setiap responden mendapat salah satu versi dari total 50 versi yang ada. * Bagian ini terdiri dari 18 pertanyaan yang merupakan pilihan kombinasi. Untuk setiap pertanyaan saudara diharapkan memilih salah satu dari dua pilihan jawaban ( lingkari nomor I atau 2 ). - Setiap pertanyaan merupakan dua kemungkinan penempatan tugas dengan berbagai karakteristik, yang akan dihadapi setelah saudara menyelesaikan pendidikan kedokteran. * Bacalah dengan teliti sebelum menentukan pilihan/jawabain anda karena setiap kombinasi pilihan adalah berbeda. * Saudara harus menentukan pilihan untuk setiap pertanyaan walaupun kombinasinya dirasakan tidak realistik ( Jawaban harus satu pilihan dan tidak boleh kosong ). * Dalam pengisian pertanyaan dilarang untuk saling bertanya/berdiskusi diantara sesama teman dan bila ada hal yang kurang jelas harap ditanyakan langsung kepada petugas. - Setiap pertanyaan terdiri dari 7 butir kata yang masing-masing mempunyai pengertian sebagai berikut: 1. Propinsi, adalah: Lokasi yang menunjukkan tempat penempatan tugas tersebut diatas. 2. Keterpencilan, adalahl: Keadaan keterpencilan dari lokasi penempatan tugas ( biasa, terpencil atau sangat terpencil ). 3. Tempat kerja, adalah: Jenis penempatan tugas ( tempat kerja ) berupa puskesmas atau klinik swasta dalam rangka menyelesaikan wajib kerja. Dalam hal ini, saudara bekerja secara purnawaktu dan tinggal di daerah tempat kerja tersebut kecuali dalam masa cuti. 4. Lama ikatan kerja, adalahl: Lamanya suatu ikatan kerja atau masa kontrak dokter di daerah tersebut (dalam satuan tahun). 5. Pendapatan per bulan adalah: Total penghasilan rata-rata per bulan dari semua sumber, yaitu a. Gaji: yang dibayar secara penuh dan tepat waktu selama saudara memenuhi kewajiban / tugas saudara. b. Praktek swasta. 6. Kemungkinan diangkat pegawai negri, adalah: Kesempatan atau peluang untuk diangkat menjadi Pegawai Negri Sipil ( PNS ) setelah menyelesaikan masa ikatan kerja, yang dinyatakan dalam bentuk persentase. 41 Draft I.0. Not for citation or quotation 7. Kesempatan spesialisasi, adalah Kemungkinan mendapat peluang/kesempatan pendidikan spesialis dengan beasiswa setelah menyelesaikan masa ikatan kerja, yang dinyatakan dalam bentuk persentase. C. Bagian Ketiga ( latar belakang) * Bagian ini merupakan data tentang demografi ( latar belakang responden ) -- cukup jelas * Untuk setiap pertanyaan, isilah jawaban atau lingkari nomor yang sesuai. 42 Draft 1.0. Not for citation or quotation ANNEX B4: Allowed combinationis of remoteness and province ini medical student survey Private clinic nonremote health remote health very remote health center center center Jakarta .7. 7 Central Java Central rHr Kalimantan South Sulawesi East Nusa ^ Tenggara North Sumatra Irian Jaya __r_r_i,,____________________ The shaded combinations were prohibited; all unshaded combinations were allowed. 43 Policy Research Working Paper Series Contact Title Author Date for paper WPS1869 Risk Reducation and Public Spending Shantayanan Devarajan January 1998 C. Bernardo Jeffrey S. Hammer 31148 WPS1 870 The Evolution of Poverty and Raji Jayaraman January 1998 P. Lanjouw Inequality in Indian Villages Peter Lanjouw 34529 WPS1871 Just How Big Is Global Production Alexander J. Yeats January 1998 L.Tabada Sharing? 36896 WPS 1872 How Integration into the Central Ferdinand Bakoup January 1998 L. Tabada African Economic and Monetary David Tarr 36896 Community Affects Cameroon's Economy: General Equilibrium Estimates WPS;1873 Wage Misalignment in CFA Countries: Martin Rama January 1998 S. Fallon Are Labor Market Policies to Blame? 38009 WPS1874 Health Policy in Poor Countries: Deon Filmer January 1998 S. Fallon Weak Links in the Chain Jeffrey Hammer 38009 Lant Pritchett WPS1875 How Deposit Insurance Affects Robert Cull January 1998 P. Sintim-Aboagye Financial Depth (A Cross-Country 37644 Analysis) WP'1876 Industrial Pollution in Economic Hemamala Hettige January 1998 D. Wheeler Development (Kuznets Revisited) Muthukumara Mani 33401 David Wheeler WP',1877 What Improves Environmental Susmita Dasgupta January 1998 D. Wheeler Performance? Evidence from Hemamala Hettige 33401 Mexican Industry David Wheeler WPS1878 Searching for Sustainable R. Marisol Ravicz February 1998 M. Ravicz Microfinance: A Review of Five 85582 Indonesian Initiatives WPS1879 Relative prices and Inflation in Przemyslaw Wozniak February 1998 L. Barbone Poland, 1989-97: The Special Role 32556 of Administered Price Increases WPS1880 Foreign Aid and Rent-Seeking Jakob Svensson February 1998 R. Martin 39065 WPS1881 The Asian Miracle and Modern Richard R. Nelson February 1998 C. Bernardo Growth Theory Howard Pack 31148 WPS1882 Interretional Resource Transfer and Toshihiko Kawagoe February 1998 R. Martin Economic Growth in Indonesia 39065 Policy Research Working Paper Series Contact Title Author Date for paper WPS1883 Intersectoral Resource Allocation and FumihideTakeuchi February 1998 K. Labrie Its Impact on Economic Development Takehiko Hagino 31001 in the Philippines WPS1884 Fiscal Aspects of Evolving David E. Wildasin February 1998 C. Bernardo Federations: Issues for Policy and 31148 Research WPS1885 Aid, Taxation, and Development: Christopher S. Adam February 1998 K. Labrie Analytical Perspectives on Aid Stephen A. O'Connell 31001 Effectiveness in Sub-Saharan Africa WPS1886 Country Funds and Asymmetric Jeffrey A. Frankel February 1998 R. Martin Information Sergio L. Schmukler 39065 WPS1887 The Structure of Derivatives George Tsetsekos February 1998 P. Kokila Exchanges: Lessons from Developed Panos Varangis 33716 and Emerging Markets