Physician brain drain: size, determinants and policy issues 1

Physician brain drain: size, determinants and policy issues 1 Frédéric Docquier a and Hillel Rapoport b a FNRS and IRES, Université Catholique de Louvain b CID, Harvard University, Bar-Ilan University and EQUIPPE The brain drain of health professionals has become a source of concern for many developing countries and international organizations. The World Health Organization estimates the current global shortage of health workers at more than 4 million. From a global perspective, therefore, the medical brain drain could be seen as a matching process through which workers are allocated to places and jobs where they are most productive. This process is not recent but has accelerated recently: it is estimated that expatriated doctors now constitute more than 20 percent of the stock of medical doctors in the OECD (30% in countries such as Ireland or the UK, and more than 50% in the Gulf States); there are currently more Ghanaian-born doctors living in London than in Ghana, and more Filipino nurses in the US than in Manila. In the face of such figures, it is legitimate to ask two simple questions: 1. Could it be that, instead of narrowing the global shortage of healthcare workers, the medical brain drain actually contributes to increase such shortages, especially in developing countries? To a large extent, the answer to this first question will depend on whether the emigration of health workers is caused by push or by pull factors; in the spirit of the new brain drain economics (Stark, 2004, Docquier and Rapoport, 2011), it will also depend on whether migration prospects generate more investment in medical education, i.e. drive more people to invest (or people to invest more) in medical studies. 2. Is market-wage a good proxy for the social contribution of medical workers? As we know, health has a dimension of public good, which implies that the allocation of health workers through market wages may not be globally optimal. Prospective doctors/migrants make their education and migration decisions on the basis of private costs and benefits (including the weight they put of their patients health and wellbeing); an efficient allocation, on the other hand, should be based on the social marginal returns to health professionals education and practice. 1 We thank Xushi Liu for excellent research assistance. This article is part of a research project on "Brain drain, return migration and South-South migration: impact on labor markets and human capital" supported by the Austrian, German, Korean, and Norwegian governments through the Multi-donor Trust Fund on Labor Markets, Job Creation, and Economic Growth administered by the World Bank's Social Protection and Labor unit. 1

In this note we address these two questions focusing on the emigration of physicians from developing countries. We first describe the magnitude and intensity of the physician brain drain (PBD) and analyze its determinants. We then survey the existing literature on the possible existence of a physician brain gain and characterize some of the channels through which physicians emigration can damage health outcomes in sending countries. Finally, we conclude by examining a number of policies that have been suggested recently to curb the PBD in the light of our findings. How big is the physician brain drain? The collection of comparative data on the physician brain drain is very recent; to the best of our knowledge, it includes only three datasets: Clemens and Pettersson (2006), OECD (2006), and Bhargava, Moullan and Bhargava (2010), an extension and harmonization of Bhargava and Docquier (2008). The first data set focuses on African countries and provides information on physicians and nurses emigration stocks and rates by country of birth in 2000; to do so they aggregate data on foreign physicians and nurses from nine important destination countries (UK, US, France, Australia, Canada, Portugal, Belgium, Spain and South Africa). The second dataset also measures physicians and nurses emigration stocks but does so for all the regions of the world, by country of training, and for the year 2005. Finally, the third dataset is also about emigration stocks by country of training for all countries but is for physicians only and adds a panel dimension as stocks are measured on a yearly basis for the period 1991-2004. More precisely, the Bhargava, Docquier and Moullan (henceforth BDM) (2010) dataset contains data on foreign physicians collected from 17 host countries (16 OECD countries and South Africa) and defines migrants according to their country of training. The data are obtained from national medical associations and are available on an annual basis. 2 On total, BDM come up with 14 yearly observations per country covering all the countries of the world for the period 1991-2004. As shown on Figure 1, regional comparisons reveal that PBD rates are highest in Sub-Saharan Africa (with average rates above 20% against 13% in South-Asia and less than 10% in all the other regions), and in smaller Pacific and Caribbean islands. The PBD rates are relatively stable over the period, except in Pacific islands. In absolute terms the main exporters of medical doctors are India and the Philippines, followed by two rich countries (Canada and the UK), as shown on Table 1. Figure 2 lists countries exhibiting the highest physicians emigration rates. Clearly, small islands of the Caribbean (Dominica, Grenada, Saint Lucia, 2 Focusing on the year 2000, the comparison with Clemens and Pettersson (2006) reveals important differences, with a correlation between the two of only.23. The bilateral correlations between physician immigrants stocks in the eight common destination countries are much higher (from 55 percent for South Africa to 97 percent for France and the United States). However, the stock based on country of training is usually much lower than the stock based on country of birth (e.g., 10% in France -- licensure requirements for foreign physicians are more stringent in France than in most other host countries --, 45% in South Africa, 77% in the United Kingdom, and 82% in the United States). 2

Saint Kitts and Nevis, Antigua and Barbuda, etc.) and of the Pacific (Fiji) show the highest PBD rates. There is a risk of overestimation for these countries as they usually host a regional training center. On the other hand, some countries have no medical school and exhibit zero PBD rates (e.g., Bostwana). This is clearly a drawback in using country of training for defining migrants. This bias would be eliminated using data by country of birth, as in Clemens and Pettersson. However, the latter include children migration which most likely represents an important fraction of total immigration for certain countries. Finally, we note that many sub-saharan African countries are among the most affected (such as Liberia, Zimbabwe, Ghana or Malawi), as well as some small European countries (such as Ireland and Iceland). Figure 3 shows the geographical distribution of the physician brain drain computed in BDM (2010) in relative terms (Figure 3a) and in terms of change between 1991 and 2004 (Figure 3b). One can see that physician brain drain rates have increased dramatically in many African countries; overall there is a lot of persistence in PBD rates (regressing 2004 rates on 1991 ones gives a R 2 of.75), and no sign of either convergence or divergence in PBD rates (regressing percent changes between 2004 and 1991 on the initial rates gives a R 2 of.06). [Insert Figures 1-2 -3 and Table 1 about here] What drives the physician brain drain? As is the case for general migration, it is obvious that the emigration of physicians is not an exogenous process. Individual-level surveys in six African countries indicate that more than half of all physicians would like to emigrate to developed countries, in search of better working conditions and more comfortable lifestyles (Awases et al., 2003). The risks associated with caring for HIV/AIDS patients and the possibility of children of healthcare staff contracting HIV as they enter adolescence may exacerbate the physician brain drain (Awases et al., 2003; Bhargava, 2005). It is common in migration studies to divide migration determinants into push and pull factors. For example, Mayda (2008) analyzed the role of push and pull factors in international migration, showing that the impact of push factors on aggregate emigration rates (without educational breakdown) is relatively small compared to distance and to pull factors. Docquier, Lohest and Marfouk (2007) and Docquier and Rapoport (2011), on the other hand, analyzed the determinants of international migration at different education levels. Focusing on the highest segment (tertiary education, or brain drain), they decomposed the high-skill emigration rate between two multiplicative components, the ratio of emigrants to natives (or average emigration rate ) and the ratio of the proportion of highly skilled among emigrants to their proportion among the native-born (or selection bias ). Focusing on developing countries, Docquier, Lohest and Marfouk (2007) found the brain drain to increase with the degree of religious fractionalization at origin (via the selection bias) and to decrease with the level of development at origin 3

(the effect on the selection bias dominates the effect on openness). The size of the country also matters: small states appeared to be more open than large countries. Comparing developing and developed countries, Docquier and Rapoport (2011) found that the coefficients in the two samples have similar signs but different magnitudes. Unsurprisingly, the brain drain from high income countries appeared to be less responsive to distance and to other geographic characteristics, which is probably due to the better ability of rich countries residents to pay for migration costs. The selection bias, on the other hand, is less responsive to immigration policies at destination and to the level of development. The latter result, however, may simply reflect a mechanical effect of human capital on the selection bias. Indeed, when the proportion of tertiary educated people increases, high-skill and total emigration rates tend to converge (from the decomposition above, it is clear that the selection bias mechanically tends to one as the proportion of highly skilled increases). Finally, the degree of openness in rich countries did not appear to depend on the level of development (which is more homogenous in the high-income sample). To explore the determinants of the PBD we will first analyze these determinants in a panel setting for the full sample of countries in the BDM (2010) data set. We will then ask whether the physician brain drain is driven by the same forces as the general brain drain or by different forces. For this we will restrict the analysis to a sample of African countries as we need the data on the general brain drain (defined by country of birth and available only for two years, 1990 and 2000) and from the physician brain drain (taken also from Clemens and Perdersson (2006)) to coincide in both their crosssectional approach, their geographical coverage and their definitions of who is a migrant (defined as foreign-born). Panel estimation for developing countries. We first consider dynamic panel estimation of the determinants of the PBD. For country i at time t, we use the stock of physicians abroad, M t, and at home, H t, taken from Docquier and Bhargava (2006). We define 4 sub-periods of 3 to 4 years: 1991-94, 1995-97, 1998-2000, 2000-2004. Given that when one push-pull factor changes, stocks are likely to adjust only slowly, a beta-convergence model seems appropriate. We therefore estimate the following econometric model: ln M t ln M t-1 = a i + b.ln M t-1 + c.x t, where X t is a set of explanatory variables (population, medical wages, gdp per capita, instability, health variables, etc.), a i is a country fixed effect, and 0<-b<1 captures the speed of convergence toward the steady state. Since M t-1 is used to construct the dependent variable and is also on the right-hand side, there is a risk of endogeneity (Nickel bias). In Table 2, we consider 4 specifications and instrument M t-1 with M t-2 : specification (1) uses the full sample and a limited set of controls (155 countries out of 189 in Docquier and Bhargava (2006) and 309 observations) and a simple IV method (instrumenting physicians emigration by its oneperiod lag). Specification (2) is identical to specification (1) except that it includes two additional controls, the wage for physicians and HIV prevalence; due to missing values for some countries, this reduces the sample to 89 countries (172 observations). Given 4

that the latter specification includes only a relatively small subset of countries (89 out of 189), in specification (3) we run a model with random effects. Finally, in specification (4), we defined all time-variant explanatory variables as predetermined (instrumented by their own lags) and run a 2-stage least square panel analysis with random effects, using a GMM estimator. Table 2b has the same structure but focuses on developing countries (124 countries and 248 observations in the largest sample). [Insert Table 2a and 2b about here] Both for the full sample (Table 2a) and the sub-sample of developing countries (Table 2b), the results show evidence of convergence as the coefficient on past emigration level is negative, very significant and stable over specifications. The number of physicians per thousand inhabitants at home is positive and also very significant, suggesting a supply-push to the PBD, an interpretation supported by the fact that the coefficient on public expenditure on higher education is also positive and significant. In contrast to analyses of the brain drain in general, country size appears to stimulate the PBD (however, this could be an artifact due to our consideration of countries of training only). Finally, as expected, the prevalence of HIV-AIDS is also favoring a higher PBD, but physicians wages do not seem to play any role. This result could be driven by the collinearity between physicians wages and GDP per capita. Indeed, heterogeneity in levels of total factor productivity is the main source variation in skill prices across countries. We will now ask whether these results are weakened or strengthened when we focus of our main region of interest, Africa. African PBD: cross-sectional results. For this second exercise we restrict our sample to African countries. In addition to analyzing the determinants of the African Physician Brain Drain (henceforth APBD), which we perform in the first two columns of Tables 3a and 3b, we also want to analyze whether the APBD is more or less impacted by certain push and pull factors than the general brain drain out of Africa, and why. To do this we apply the method used by Docquier, Lohest and Marfouk (2007) for the general brain drain and decompose the African physician brain drain into two multiplicative components: phys coll phys M phys t M t M t H m t / phys phys + coll coll coll H t M t H t + M t M t H phys t coll t + M + M where the first multiplicative component is the ratio of emigrants to natives among skilled workers (coll stands for college graduates, and phys stands for physician doctors), that is, the average high-skilled emigration rate (or degree of openness to skilled emigration ) and the second multiplicative component is the ratio of the proportion of doctors among highly skilled emigrants to their proportion among the native college graduates (or selection bias ). We consider a cross-section framework because our sources of data for the general brain drain are Docquier and Marfouk (2006) and its extension to correct for age of entry in Beine, Docquier and Rapoport (2007). Docquier and Marfouk (2006) phys t coll t 5

constructed a global bilateral database of South-North and North-North migration (from 195 origin countries to 30 OECD countries) for three levels of education, and for 1990 and 2000. Beine, Docquier and Rapoport (2007) extended this data set for the highest (tertiary) education level; the latter collected data on the age of entry structure of immigration, and used age of entry as a proxy for whether education was acquired in the home or in the host country. Since such data was not available from all OECD receiving countries, their data set combines observations (for 75 percent of the sample size) and estimations of the age of entry structure of the remaining 25 percent using a gravity model. For the African PBD we will use both BDM (2010), as above, and also Clemens and Pettersson (2006), who collected data on foreign physicians from nine important destination countries (UK, US, France, Australia, Canada, Portugal, Belgium, Spain and South Africa) and computed the stock of African-born physicians living abroad by country of birth in 2000. They then evaluate the physician brain drain in relative terms, dividing the number of physicians abroad by the total number of physicians born in each origin country. It is important to consider PBD data by country of birth for two reasons: for robustness and for interpretation of any differences with the results obtained using PBD data by county of training. For the empirical analysis we will match the different data sets as follows. In Table 3a, we match Beine, Docquier and Rapoport (2007) with BDM (2010), that is, the PBD data by country of training coupled with estimates of brain drain rates at age 22 or more (i.e., obtained after excluding those who were younger than 22 at the time of immigration). In Table 3b, we use Docquier and Marfouk (2006) and Clemens and Pedersson (2006), as both define migrants by country of birth. The odd columns in Table 3a and 3b give the results for the full model while the even columns give the results for the parsimonious specifications. 3 [Insert Table 3a and 3b about here] Focusing on the results of the parsimonious specifications for the APBD by country of training (and using the regressions by country of birth for comparison and robustness), a number of interesting results appear: First, GDP per capita has the expected negative sign for both the general and physician brain drain; note that the latter is more impacted by differences in country income levels, as shown by the negative sign for the coefficient on the selection bias (which is not negative enough though to make the coefficient on GDP per capita negative and significant in the PBD regressions by country of birth). Wage differences, on the other hand, are insignificant is all specifications. The latter result can be driven by collinearity between wages and GDP per capita, as stated above. Second, the geographical variables are also generally significant and have the expected signs: negative for landlock status for both the physician and general brain drain (with the former being less negatively impacted in the regressions by country of birth) and 3 Parsimonious specifications are obtained after sequential elimination of insignificant variables. 6

surface area, positive for small island status (but less positive due to a negative coefficient on self-selection) but only in the country of birth regression. Third, the selection bias of physicians vis-à-vis all tertiary educated is significantly affected by the population size: negatively in the regressions by country of birth and positively in the regressions by country of training: this is consistent with a supplypush interpretation, as already suggested for the panel results on developing country. Recall that demographically larger countries are more likely to host medical schools and train physicians from smaller countries. Third, additional push factors seem to play an important role: the prevalence of HIV- AIDS in the regressions by country of training, and the degree of ethnic fractionalization in the regressions by country of birth. The case for a physician brain gain In the spirit of the recent literature on endogenous human capital in a context of migration (e.g., Beine, Docquier and Rapoport (2008)), we may ask whether there is a chance for a net medical brain gain. Regressing the log of domestic physicians per capita on the log of physician emigrants per capita in a sample of African countries, Clemens (2007) found a positive correlation of about 70 percent. Clearly, this correlation can be driven by the simultaneous effects of observed variables (GDP per capita, school enrolment, ethnic conflicts, etc.) or unobserved variables. However, after controlling for observables and instrumenting the number of emigrants, the causal effect of emigration becomes insignificant. This analysis fails to detect any negative effect of health professionals emigration on the supply of healthcare staff in Africa in a cross-section analysis based on 53 observations. The author attributes this result to the positive effect of emigration prospects on enrolment in medical schools. The absence of negative effect of emigration on domestic physicians stocks could also be due to omitted variables such as the size (and quality) of the medical training system. Physician emigration is instrumented with country size and linguistic links. However, data reveal a strong correlation between country size and both the number of medical schools (82 percent) and the annual number of domestically-trained medical graduates (60 percent). In addition, the number of schools and graduates are significantly higher in English-speaking countries and/or UK former colonies. Hence, it is very likely that country size and linguistic linkages exert a direct impact on the domestic supply of health workers. This relationship obviously needs to be explored in more detail in future research. Three other studies examine the interactions between medical education and migration decisions in developing countries; they deliver interesting results for developing countries in general, and for low-income countries in particular. 7

The first study, by Kangasniemi et al. (2007), documents the incentive mechanism in the medical sector using a survey of overseas doctors working in the United Kingdom. The authors show that 28 percent of the Indian doctors surveyed (the largest group in their sample) acknowledge that the prospect of emigration affected their education decisions. This proportion increases to 29 percent for physicians originating from middle-income countries and to 37 for those originating from low-income countries. In addition, the physicians surveyed estimate that among medical students currently studying in the home country, the proportion of students for which emigration prospects affected their decisions as to whether and what to study is very high (estimated proportions are 36 percent for India, 46 percent for low-income countries and 41 percent for middle-income countries). Given these figures, we cannot exclude the possibility that incentive effects are large enough to increase the net supply of physicians in origin countries. A necessary condition for a brain gain is that the additional human capital formed thanks to the prospect of emigration exceeds the human capital lost through actual emigration. The survey responses in Kangansniemi et al. (2007) suggest that such a possibility is not unrealistic for a large number of lowincome countries, including many African ones. This is notwithstanding other potential benefits to home countries through remittances and return migration: the survey also shows that a large fraction of expatriate physicians send substantial remittances, and many intend to return after completing their training or gaining work experience in the UK. In the second study, Defoort (2010) regresses the change in the number of native physicians on past medical emigration rates in logs. She took advantage of the panel structure of the Bhargava-Docquier s data set and worked with 5 observations per country (one observation every 3 years). Using different methods (fixed effects v. random effects, GLS, IV, GMM), she found evidence of a positive incentive effect, especially in Sub-Saharan African countries. Using counterfactual simulations as those used by Beine et al. (2008) for the general brain drain, she found an optimal physician brain drain rate of about 10 percent and concludes that only 20 African countries actually suffer from the physician brain drain while about 30 countries would actually gain (in terms of physicians per capita) from an increase in medical emigration rates. The above result is driven by the log specification of the incentive mechanism. At low levels of PBD, the elasticity of medical training to PBD is very large. Using PBD data by country of training, many developing countries exhibit very low emigration rates. In a third study, Bhargava, Moullan and Docquier (2010) use a slightly different specification with log(1+pbd) at the right-hand side. This avoids unrealistic marginal incentive effects for countries with MBD around or below 1 percent. Again, there appears to be a positive incentive effect of migration prospects on medical training. However, the effect was too small to generate a net brain gain so that MBD mainly reduces the number of physicians in developing countries. Another argument militating against physician brain gain is that migration prospects not only affect the number of physicians but also their fields of study. From this perspective, a clearly negative effect of the brain drain, recognized long ago (e.g., in the pages on the 8

brain drain in Todaro s Economic Development textbook s early editions) is that it can drive prospective doctors toward medical fields which have low social value at home but are highly valued in rich countries. As Lucas (2004) puts it (while observing the case of the Philippines), It is difficult to believe that these high, privately financed enrolment rates are not induced by the possibility of emigration. There are signs that the choice of major field of study... responds to shifts in international demands. Higher education is almost certainly induced to a significant extent by the potential for emigration. For example, medical students contemplating emigration may specialize in geriatrics or in cardio-vascular diseases rather than in pediatrics or tropical diseases, thereby contributing more to brain waste than to brain gain. In summary, the existing literature is not conclusive as to whether the emigration of physicians lowers or increases the net supply of physicians at home: the results from empirical studies depend on data sources or empirical specifications; once incentive effects are taken into account, there is no strong evidence of either a physician brain drain or brain gain. PBD and health outcomes In developing countries, the size and quality of the medical sector is a key determinant of human development and economic performances (see Bhargava et al., 2001, Hagopian et al., 2004, Cooper, 2004, Bhargava and Docquier, 2008). While the number of physicians per 1,000 people is greater than 3 in most industrialized countries, it is lower than 0.25 in a large number of developing countries. Many observers and analysts have pointed to the physician brain drain as one of the major factors leading to the under-provision of healthcare staff in developing countries (see Bundred and Levitt, 2000, Beeckam, 2002, Johnson, 2005, Eyal and Hurst, 2008) and, ultimately, to low health status and shorter life expectancy hence Michael Clemens s (2007) provocative question: do visas kill? Since 1990, the world s countries and leading development institutions have agreed on a set of Millennium Development Goals (MDG). The Millennium Declaration, signed in 2000, established 2015 as the deadline for achieving the MDG. The eight goals include specific health targets: (i) reducing by two thirds the mortality rate among children under five, (ii) reducing by three quarters the maternal mortality ratio and achieving universal access to reproductive health, (iii) combat HIV/AIDS, malaria and other diseases. Much progress has been made in reducing maternal deaths in developing regions, but not in the countries where giving birth is most risky, and many countries are still falling short of meeting the goals. Is the physician brain drain partly responsible for these bad records? Using the methodology described above, Clemens (2007) found no significant causal impact of the numbers of physicians and nurses abroad on child mortality, infant mortality under age one, vaccination rates or prevalence of acute respiratory infections in children under 9

age five. Chauvet et al. (2008) investigated the determinants of child mortality using a sample of 98 developing countries from 1987 to 2004. In their benchmark full-sample regressions, remittances strongly improve health indicators while health aid per capita and the number of physicians per 1,000 people have no significant impact. However, when interacted with the level of development, health aid commitments become significant and help reducing child mortality in poorer countries, while the number of physicians per 1,000 people has no significant impact. Interestingly, the supply of healthcare staff does not significantly reduce infant and child mortality rates. However, the physician brain drain is shown to significantly deteriorate child health indicators. This suggests that emigrants could positively self-select out of the physicians population, with only the most talented obtaining a qualification abroad and leaving. The only study we are aware of directly tackling the issue of physicians self-selection in Africa is a recent paper by De Laat and Jack (2009) who use Ethiopia as a case-study. They take advantage of a feature of the recruitment process of physicians in the public sector of Ethiopia to estimate the extent of adverse selection in that sector. More precisely, physicians first placements occur through a lottery, leading to self-selection into the lottery while non-lottery participants apply mainly to private institutions. The authors argue that such random placement does not allow for efficient signaling of individual ability and therefore leads to adverse selection into the lottery, which is indeed what they find using career and wage records of physicians who remain in the public sector. They also find that within the group of lottery participants, the most able tend to leave and are likely to account for a substantial part (one third) of the physician brain drain out of Ethiopia. Two other studies examine the effect of medical staffing and brain drain on human development. Bhargava and Docquier (2008) assess the effect of PBD on the dynamics of HIV prevalence rates and adult deaths from AIDS. They find no effect of PBD on the long-run level of HIV prevalence, but a significant effect on deaths rates: a doubling of the physician brain drain rate is associated with a 20 percent increase in adult deaths from AIDS. Bhargava, Moullan and Docquier (2010) evaluate the impact of PBD on child mortality and vaccination rates, allowing for quantity and quality effects (i.e., decrease in the numbers and average abilities of the remaining physicians as in Chauvet et al.). They show that infant and child mortality rates decrease with the numbers of physicians per capita when adult literacy rates exceeded 60 percent, which is the case for the majority of countries. The results for DPT and measles vaccinations again underscore the importance of literacy rates and physicians per capita for higher vaccination uptake. However reducing physician brain drain generates only small improvements in human development indicators compared to the stated Millennium Development Goals. In summary and as for the previous section, the existing empirical literature fails to provide strong evidence of the effects of the APBD on health outcomes in Africa. Better health indicators could probably be achieved by exploiting synergies between the numbers of physicians and other factors such as availability of medicines, number of nurses (for which the data are missing for many developing countries and years), and medical equipment (such as hype syringe, latex gloves) and general infrastructure 10

(access to drinkable water, use of mosquito net, etc.). Increasing the supply of physicians might be inefficient for some developing countries if the effective demand is low due to poor infrastructure and/or political instability. Policy implications and conclusion The main insights from the above analysis may be summarized as follows: i) A country s stock of medical human capital is endogenous to the prospect and realization of migration; this induces a theoretical possibility of beneficial physician brain drain. Such a possibility is unlikely to materialize in the case of the African physician brain drain for a least three reasons: first, there is anecdotal evidence of brain waste when prospective doctors invest in fields which may be disconnected from the needs of the local population; second there is suggestive evidence of positive selection into migration among physicians; and third, the empirical evidence on the sign of the net effect (i.e., whether the brain gain effect compensates for the brain drain) is mixed and varies with the choice of econometric specification. ii) iii) Physicians are just one in many arguments of the production function of healthcare and are strategic complements of other factors such as medical infrastructure, nurses and other health care personnel, medical equipment, etc. Increasing the supply of physicians without increasing that of complementary inputs is unlikely to significantly affect health outcomes. Retaining more home-trained physicians implies taking into account the pull and push factors that determine physicians emigration decisions. The evidence here is a mix of good and bad news. On the good side, there does not seem to be evidence of chain migration by which past physician emigration would fuel future waves of expatriation. Also, physicians wages do not seem to be an important drive; this may just reflect the lack of heterogeneity in physicians wages across Africa, however the fact that wage distance to the US is never significant in the panel regression for the sample of developing countries as a whole makes this latter interpretation improbable. On the bad side, the main source of concern is probably the strong push effect of HIV-prevalence that comes out for both the panel regressions for developing countries and the crosssectional results for African countries. Since HIV-prevalence is unlikely to decrease substantially in the short run, this cast doubts on the potential for active policies to effectively retain local physicians or convince expatriated physicians to return. Finally, one should recall that the results summarized in i), ii) and iii) likely affect different countries differently, and so there is no one size fits all optimal policy response to the physician brain drain. Existing policy responses, however, have adopted 11

uniform solutions and either focused on the prevention of physicians emigration or on favoring the return of those who emigrated. We discuss these two policies separately. Should MDs from poor countries be blacklisted? Preventing someone from emigrating is extremely difficult and most likely entails adopting very repressive steps such as sanctions on the remaining family, confiscation of any assets left, deprivation of citizenship or right of return, etc. Still, many have recommended the creation of blacklists of source countries from which it should be forbidden to recruit physicians, nurses, health workers and, more generally, highly-educated individuals. On a general note one should keep in mind that the countries generally cited as candidates for being blacklisted are precisely those where emigration is quite often provoked by political and racial conflicts and/or oppressive and corrupted governments. Constituting such blacklists would deprive many developing countries professionals from their basic right to escape oppression and extortion, and could even lead to higher political and economic repression. 4 Finally, such blacklisting of countries and professions may be based on erroneous appreciations of the role of migration in explaining professional shortages in developing countries. For example, a proposal to ban recruitment of health professionals from Sub-Saharan Africa has gained wide support in many healthcare and media circles; 5 however, as we have seen, it is not clear that the physician brain drain is decreasing the net supply of physicians in Africa, that decreasing physicians emigration is feasible without seriously changing their environment, or that increasing the supply of physician ceteris paribus has a positive health impact in Africa. Is brain circulation a panacea? Another popular, even almost consensual policy proposal is to encourage the brain circulation of MDs from poor countries, that is, their return to the home country after some time, once they have acquired knowledge and professional skills abroad that can eventually serve their home country. This would seem the ideal solution. Most policy reports also emphasize, in line with our findings, that brain circulation should be promoted through incentives to return and be accompanied by an increase in the supply of complementary inputs (such as nurses, infrastructures, medical equipment, etc.) and by policy action to affect the pull and push factors that affected the PBD in the first place. 6 Health organizations now emphasize the need to move from brain drain to brain circulation, to let physicians from poor countries emigrate freely while at the same time designing incentive packages for their return. This is a big improvement over the previous proposal. However, the effectiveness of such policies can be questioned on a number of grounds. 4 Docquier and Rapoport (2003) show that migration as an exit strategy can serve to tame oppressive, rent-seeking governments. 5 See for example the editorial of a recent issue of the British Medical Journal, where the editor, James Johnson, states that the rich countries of the North must stop looting doctors and nurses from developing countries (Johnson, 2005). 6 See for example Physicians for Human Rights (2004). 12

First, wage gaps are so huge that it is unlikely they can be raised in the home country in a way that significantly affects return decisions. 7 Second, physicians who contemplated wage increases when they made their emigration decision will face wage decreases when contemplating return migration; given that preferences evolve over time and people tend to have loss-aversion, any domestic wage increase for local physicians in poor countries would probably affect the decision to migrate more than the decision to return. The same rationale applies for other policy actions aimed at affecting pull and push factors such as improved infrastructures, better governance of the public health system or reduced risks associated with the high-prevalence of HIV-AIDS. Another way of looking at brain gain and brain circulation, however, is possible and may leave more room for optimism; it starts from the realization that ideas can move without people physically moving. From this perspective, physicians do not differ fundamentally from other highly educated professionals in terms of attachment to the home country, willingness to keep return options open, and potential for taking part in medical, scientific, business networks that serve as bridges between their home and host country. A sensible policy recommendation, therefore, is to recognize that preventing physicians emigration is not feasible (nor is it moral) and that hoping for return migration is not a viable strategy unless the conditions at home evolve drastically. As for the brain drain in general, it is preferable instead to favor brain circulation by making emigration and remigration easier (e.g., through dual-citizenship agreements) 8 and, ultimately, by creating domestic networks that allow for quality interactions with the expatriated networks. While this is easier said than done, there seems to be a strong commitment at the bilateral and multilateral level to favor such institutions building aimed at harnessing the diaspora, including the one formed as a result of the physician brain drain. 9 References Awases, M., A. Gbary, J. Nyon and R. Chatora (2003): Migration of health professionals in six countries: A synthesis report, World Health Organization, Regional Office for Africa. Beecham, L. (2002): UK government should stop recruiting doctors from abroad, British Medical Journal, 325. Beine, M., F. Docquier and H. Rapoport (2007): Measuring international skilled migration: new estimates controlling for age of entry, World Bank Economic Review, 21: 249-254. Beine, M., F. Docquier and H. Rapoport (2008): Brain drain and human capital formation in developing countries: winners and losers, Economic Journal, 118: 631-652. 7 Kangasniemi et al. (2007) set the wage gap for Indian physicians in the UK at about eight at current exchange rates and about 4 in PPP. They are about twice those figures for African doctors. 8 See, e.g., Leblang (2009), who shows such dual citizenship agreements are associated with more remittances, capital flows, aid flows, and return intentions. 9 See for example the Global Forum on Migration and Development (http://www.gfmd2009.org/) for a list of policy initiatives in this direction. 13

Bhargava, A. (2005): AIDS epidemic and health care infrastructure inadequacies in Africa: A socioeconomic perspective, Journal of AIDS, 40: 241-242. Bhargava, A. and F. Docquier (2008): HIV Pandemic, Medical Brain Drain, and Economic Development in Sub-Saharan Africa, World Bank Economic Review, 22: 345-66. Bhargava, A., F. Docquier and Y. Moullan (2010): Modeling the effect of physician brain drain on human development, Economics and Human Biology, forthcoming. Bhargava, A., Jamison, D., Lau, L., Murray, C. (2001): Modeling the effects of health on economic growth, Journal of Health Economics, 20: 423-440. Bundred, P. and C. Levitt (2000): Medical migration: who are the real losers?, The Lancet 356, 9225: 245-46. Chauvet, L., F. Gubert, S. Mesplé-Somps (2008): Are remittances more effective than aid to improve child health? An empirical assessment using inter and intra-country data, paper presented at the Annual Bank Conference on Development Economics, Cape Town, South Africa. Clemens, M. (2007): Do Visas Kill? Health Effects of African Health Professional Emigration, Working Paper 114, Center for Global Development. Clemens, M.A. and G. Pettersson (2006): A New Database of Health Professional Emigration from Africa, Working Paper, 95, Center for Global Development. Cooper R.A. (2004): Weighing the evidence for expanding physician supply, Annals of Internal Medicine, 141: 705-14. De Laat, J. and W. Jack (2009) : Adverse selection and the brain drain, Mimeo., Georgetown University. Defoort, C. (2010): To educate or not to educate: the impact of migration perspectives in the medical sector, Working Paper, EQUIPPE, University of Lille. Docquier, F. and A. Marfouk (2006): International migration by educational attainment (1990-2000), in C. Ozden and M. Schiff (eds). International Migration, Remittances and Development, Palgrave Macmillan: New York. Docquier, F. and H. Rapoport (2003): Ethnic discrimination and the migration of skilled labor, Journal of Development Economics, 70, 159-72. Docquier, F. and H. Rapoport (2009): Documenting the brain drain of la crème de la crème : three case studies on international migration at the upper tail of the education distribution, Jahrbucher fur Nationalokonomie und Statistik, 229, 6: 679-705. Docquier, F. and H. Rapoport (2011): Globalization, brain drain and development, Journal of Economic Literature, forthcoming. Docquier, F., O. Lohest and A. Marfouk (2007): Brain drain in developing countries, World Bank Economic Review, 21, 2: 193-218. Eyal, N. and S.A. Hurst (2008): Physician Brain Drain: Can Nothing Be Done?, Public Health Ethics, 1, 2: 180-192. Hagopian A, M.J. Thompson, M. Fordyce, K.E. Johnson, G.L. Hart (2004): The migration of physicians from sub-saharan Africa to the United States of America: measures of the African brain drain, Human Resources for Health, 2-17. 14

Johnson, J. (2005). Editorial: Stopping Africa's medical brain drain, British Medical Journal, 331:2 Kangasniem M., L.A. Winters and S. Commander (2007): Is the medical brain drain beneficial? Evidence from overseas doctors in the UK, Social Science and Medicine, 65, 5: 915-923. Leblang, D. (2009): Harnessing the diaspora: the political economy of dual citizenship, Working Paper, University of Virginia. Lucas, R.E.B. (2004): International migration regimes and economic development, Report for the Expert Group on Development Issues (EGDI), Swedish Ministry of Foreign Affairs. Mayda, A.M. (2010): International migration: A panel data analysis of the determinants of bilateral flows, Journal of Population Economics, 23, 4: 1249-74. OECD (2006): The medical brain drain, Paris: OECD. Physicians for Human Rights (2004): An action plan to prevent brain drain: building equitable health systems in Africa, Boston, MA: June. Stark, O. (2004): Rethinking the brain drain, World Development, 32, 1: 15-22. 15

Table 1. The main exporters of physicians in 2004 Country Emigration1991 Emigration2004 % change India 45375 71290 57.11% Philippines 17158 20000 16.56% Canada 13128 18635 41.95% United Kingdom 15478 17759 14.73% South Africa 10027 16433 63.89% Pakistan 7300 16423 124.97% Germany 8778 13571 54.61% Mexico 10877 13057 20.04% Ireland 11904 11388-4.33% Egypt 5311 8515 60.32% Italy 7037 8402 19.40% Australia 11212 7958-29.02% Spain 6581 7779 18.21% Dominican Republic 4905 7257 47.96% Iran 4985 6620 32.78% Poland 3633 5765 58.67% Nigeria 1519 5499 261.97% China 1655 5417 227.29% Russia 1331 5039 278.51% Grenada 1625 5000 207.69% Top 20 Total 189821 271808 43.19% World Total 288249 411379 42.72% Top20/World 65.85% 66.07% 16

Table 2a. Determinants of the growth rate of the PBD - Full sample (dependent variable = log emphys(t) -minus log emphys(t-1)) Spec 1 Spec 2 Spec 3 Spec 4 Method IV IV GLS GMM Random effect No No Yes Yes Incl. wagerat and hivprev No Yes Yes Yes log(emphys(t-1)) a -0.083 (-8.08)*** -0.087 (-4.48)*** -0.087 (-4.48)*** -0.082 (-3.73)*** log(phys1000(t-1)) 0.081 (3.39)*** 0.165 (3.73)*** 0.165 (3.73)*** 0.169 (3.26)*** log(poptot(t-1)) 0.054 (3.65)*** 0.056 (1.96)** 0.056 (1.96)** 0.048 (1.44) log(scht(t-1)) 0.004 (0.14) 0.017 (0.42) 0.017 (0.42) 0.027 (0.58) log(gdppc(i,t-1)) -0.006 (-0.31) -0.039 (-1.03) -0.039 (-1.03) -0.054 (-1.18) log(wagerat(i,t-1)) - 0.002 (0.09) 0.002 (0.09) 0.021 (0.68) log(hivprev(i,t-1)) - 0.064 (3.09)*** 0.064 (3.08)*** 0.082 (3.44)*** log(pubext(i)) 0.054 (2.23)*** 0.129 (2.92)*** 0.129 (2.92)*** 0.137 (2.79)*** log(distocde(i)) -0.011 (-0.65) -0.028 (-0.93) -0.028 (-0.93) -0.030 (-0.90) log(ethnfr(i)) 0.004 (0.20) -0.017 (-0.38) -0.017 (-0.38) -0.015 (-0.29) linocde(i) 0.044 (1.22) 0.071 (1.09) 0.071 (1.09) 0.066 (0.89) colocde(i) 0.049 (1.08) 0.040 (0.46) 0.040 (0.46) 0.029 (0.30) oilexp(i) 0.120 (2.06)** 0.027 (0.25) 0.027 (0.25) -0.007 (-0.06) smisl(i) 0.251 (4.15)*** 0.341 (3.44)*** 0.341 (3.44)*** 0.354 (3.25)*** landlock(i) -0.002 (-0.03) -0.069 (-0.97) -0.069 (-0.97) -0.104 (-1.29) Constant -0.405 (-1.22) -0.334 (-0.51) -0.335 (-0.51) -0.111 (-0.14) Chi2(15) - - 87.78 77.89 Prob > chi2 - - 0.0000 0.0000 Rsquare (Adj.) 0.3037 0.2989 0.3604 0.3449 # of countries 155 89 89 83 # of observations 309 172 172 156 Note. a log(emphys(t-1)) instrumented by log(emphys(t-2)). 17

Table 2b. Determinants of the growth rate of the PBD - Developing countries (Dependent variable = log emphys(t) -minus log emphys(t-1)) Spec 1 Spec 2 Spec 3 Spec 4 Spec 5 Method IV IV GLS GMM GMM Random effect No No Yes Yes Yes Incl. wagerat and hivprev No Yes Yes Yes Yes log(emphys(t-1)) a -0.095 (-7.25)*** -0.086 (-3.86)*** -0.086 (-3.86)*** -0.081 (-3.21)*** -0.078 (-4.31)*** log(phys1000(t-1)) 0.082 (2.94)*** 0.154 (3.11)*** 0.154 (3.11)*** 0.145 (2.42)** 0.152 (4.9)*** log(poptot(t-1)) 0.078 (3.73)*** 0.064 (1.86)* 0.064 (1.86)* 0.055 (1.35) 0.052 (1.85)* log(scht(t-1)) 0.033 (1.01) 0.022 (0.49) 0.022 (0.49) 0.027 (0.53) log(gdppc(i,t-1)) 0.020 (0.72) 0.005 (0.10) 0.005 (0.10) 0.002 (0.03) log(hivprev(i,t-1)) 0.062 (2.61)*** 0.062 (2.61)*** 0.086 (3.14)*** 0.056 (2.93)*** log(wagerat(i,t-1)) 0.007 (0.24) 0.007 (0.24) 0.023 (0.68) log(pubext(i)) 0.088 (2.80)*** 0.143 (2.94)*** 0.143 (2.94)*** 0.151 (2.79)*** 0.117 (3.27)*** log(distocde(i)) -0.005 (-0.22) -0.036 (-0.93) -0.036 (-0.93) -0.063 (-1.36) log(ethnfr(i)) -0.003 (-0.09) -0.007 (-0.12) -0.007 (-0.12) -0.033 (-0.44) smisl(i) 0.305 (4.22)*** 0.355 (3.30)*** 0.355 (3.30)*** 0.349 (2.95)*** 0.256 (3.2)*** oilexp(i) 0.075 (1.09) -0.017 (-0.14) -0.017 (-0.14) -0.034 (-0.25) linocde(i) 0.052 (1.15) 0.053 (0.69) 0.053 (0.69) 0.042 (0.47) colocde(i) 0.080 (1.5) 0.066 (0.69) 0.066 (0.69) 0.034 (0.32) landlock(i) 0.011 (0.2) -0.046 (-0.60) -0.046 (-0.60) -0.092 (-1.07) Constant -1.200 (-2.58)** -0.775 (-1.01) -0.775 (-1.01) -0.423 (-0.46) -0.639 (-1.52) Chi2(15) 72.09 64.46 62.39 Prob > chi2 0.0000 0.0000 0.0000 Rsquare (Adj.) 0.2809 0.2811 0.3550 0.3434 0.2716 # of countries 124 76 76 71 90 # of observations 248 147 147 133 168 Note. a log(emphys(t-1)) instrumented by log(emphys(t-2)). 18

Table 3a. Determinant of aggregate PBD rates in 2000 for African Countries Analysis by country of training Log(PBD_Train) Log(BD22+) Log(PBD_Train/BD22+) Full Parsim Full Parsim Full Parsim Log(gdppc) -1.176-0.700-0.671-0.338-0.257 (-1.57) (-2.56)** (-3.08)*** (-2.95)*** (-0.43) Log(scht) 0.754 0.077 0.505 (1.39) (0.40) (1.18) Log(poptot) 0.629 0.283 0.439 0.546 (1.47) (1.96)* (1.29) (3.51)*** Log(georeg) -2.023-5.527 1.342-4.444 (-0.28) (-2.16)** (0.70) (-0.78) landlock -1.469-1.581-0.968-0.853-0.004 (-1.95)* (-2.75)*** (-2.89)*** (-3.15)*** (-0.01) smisl 2.241 2.822 1.073-2.977 (0.83) (3.43)*** (2.87)*** (-1.39) linocde 0.356 1.487 0.801-0.426 (0.45) (3.86)*** (2.60)** (-0.67) opec 0.816 0.247 0.236 (2.04)* (1.68)* (0.74) oilexp -0.835-0.145-0.338 (-0.97) (-0.54) (-0.49) Log(hivprev) -0.096 0.670-0.578 0.764 0.274506 (-0.11) (3.66)*** (-2.48)** (1.13) (2.23)** Log(ethnfr) -2.424-2.545-0.192 (-1.07) (-2.63)** (-0.11) Log(relfr) 1.255 1.189-0.378 (1.00) (2.10)** (-0.38) Log(phys1000) 0.050 0.237 (0.07) (0.42) Log(wagerat) -0.359-0.158 (-0.73) (-0.41) Cons -5.181 11.023-7.008-0.674 1.990 (-0.27) (2.04)** (-1.37) (-0.89) (0.13) observations 33 36 43 51 33 35 R-square 0.53 0.36 0.59 0.41 0.62 0.40 Adj. R-square 0.16 0.27 0.43 0.36 0.33 0.36 Notes: OLS regressions, robust t-statistics in parentheses. *,**,*** significant at 10, 5 and 1%. Data sources: BD22+ from Beine et al. (2007), PBD by country of training from BDM (2010). Variables definitions: Gdppc=gdp per capita. Linocde=Linguistic proximity with OECD. Landlock=land locked country. Smisl=small islands developing country. Georeg= geographic area. Wagerat= Average wage rate of physicians (US=1). Relfr=religious fractionalization. Ethnfr=ethnicity fractionalization. 19

Table 3b. Determinant of aggregate PBD in 2000 for African Countries Analysis by country of birth Log(PBD_Birth) Log(BD0+) Log(PBD_Birth/BD0+) Full Parsim Full Parsim Full Parsim Log(gdppc) -0.237 - -0.574-0.492-0.028 - (-1.11) (-2.79)*** (-4.09)*** (-0.10) Log(scht) 0.107-0.091 - -0.102 - (0.66) (0.50) (-0.48) Log(poptot) 0.117-0.257 - -0.152-0.179 1.03 (1.89)* (-1.03) (-1.95)* Log(georeg) -3.229-2.806 0.638 - -3.276 - (-2.06)** (-3.04)*** (0.35) (-1.59) landlock -0.431-0.313-0.814-0.737 0.562 0.445 (-1.70)* (-1.72)* (-2.57)** (-2.9)*** (1.69)* (1.8)* smisl 1.543 0.804 2.588 1.570-1.658-0.961 (2.22)** (2.47)** (3.33)*** (3.41)*** (-1.82)* (-2.01)* linocde 0.426-1.340 0.984-0.770-0.680 (1.49) (3.68)*** (3.3)*** (-2.05)** (-2.49)** opec 0.129 - -2.523-1.462 0.964 1.007 (1.15) (-2.76)*** (-2.04)** (0.99) (2.1)** oilexp -0.307-1.155 0.888-0.586 - (-1.50) (2.16)** (1.85)* (-1.07) Log(hivprev) 0.106-0.179 - -0.032 - (0.61) (1.29) (-0.22) Log(ethnfr) -1.028 - -0.017 - -0.249-0.461 (-1.39) (-0.07) (-0.93) (-2.49)** Log(relfr) 0.597 1.289-0.500-0.188 0.598 0.346 (1.43) (3.58)*** (-2.27)** (-1.71)* (2.63)** (2.97)*** Log(phys1000) -0.247-0.301 - - 0.317 - (-1.07) (-3.27)*** (1.04) Log(wagerat) -0.131 - - - -0.063 - (-0.95) (-0.35) Cons 2.635 2.473-5.426 0.095 10.980 4.180 (0.66) (1.72)* (-1.12) (0.13) (2.09)** (2.64)** observations 42 45 43 47 42 48 R-square 0.52 0.36 0.59 0.53 0.52 0.45 Adj. R-square 0.27 0.28 0.43 0.45 0.27 0.35 Notes: OLS regressions, robust t-statistics in parentheses. *,**,*** significant at 10, 5 and 1%. Data sources: BD-Docquier and Marfouk (2006), PBD by country of birth-clemens and Pedersson (2006). Variables definitions: Gdppc=gdp per capita. Linocde=Linguistic proximity with OECD. Landlock=land locked country. Smisl=small islands developing country. Georeg= geographic area. Wagerat= Average wage rate of physicians (US=1). Relfr=religious fractionalization. Ethnfr=ethnicity fractionalization. 20

Figure 1. Evolution of PBD by region (1991-2004) Note. PBD = physician emigration rate in percent. Pacific islands exclude Australia and New Zealand. Latin America = Central America + South America (excluding Caribbean islands). Asia =East, South-East and South Asia. MENA =Middle East and Northen Africa. 21

Figure 2. Largest PBD rates in 2004 22