Interaction Between Economic and Political Factors In the Migration Decision*

Transcription

1 Interaction Between Economic and Political Factors In the Migration Decision* Kit-Chun LAM Department of Economics Hong Kong Baptist University Renfew Road, Kowloon Tong Hong Kong kclam@hkbu.edu.hk September, 2001 *Funding support under Earmarked Research Grant GSC/90-91/06 of the University and Polytechnic Grants Committee is gratefully acknowledged. I am indebted to Mark Rosenzweig, An-Chi Tung and other participants of the Taipei International Conference on Labor Market Transition and Labor Migration in East Asia for valuable comments. I am also indebted to the comments of two anonymous referees. Remaining errors are my own. i

2 Interaction Between Economic and Political Factors In the Migration Decision Abstract While the traditional human capital model regards differential lifetime net earnings as the key motivation for emigration, in reality migrants often earn substantially less in the receiving country when they leave a country facing possible changes in political and economic systems. In this paper we incorporate the interaction of economic and political factors in the emigration decision. Empirically we allow economic and political confidence to interact in a discrete choice model of migration decision and test it with the data of Hong Kong when she was facing a transition to a new regime of one country and two systems. Our results show that the lack of political confidence increases emigration propensity significantly. Lack of economic confidence also increases emigration propensity though by a lesser extent, whereas expected decrease in income abroad deters emigration by only a small magnitude. Our analysis is pertinent to migration from countries facing political uncertainties such as the LDCs, Asian and former Soviet bloc countries. JEL Classification numbers: J31, J61, O15 1

3 Interaction Between Economic and Political Factors In the Migration Decision I. Introduction In this paper we incorporate the interaction of economic and political factors in the emigration decision of an individual. Our study is motivated by the observation that while the traditional human capital model regards differential lifetime net earnings as the key motivation for emigration, it is not uncommon to find migrants earning substantially less in the receiving country than in their country of origin. Even though this observation could be rationalized on the basis that immigrants initial expectation on future earnings in the receiving country could be unrealized, leading some of them to return migrate or onward migrate to a third country (see DaVanzo (1983) and Lam (1989)), yet many stay on in the receiving country, despite a large reduction in earnings after migration. It is apparent that economic gains (actual or anticipated) could not be the only driving force for migration. In many instances, political consideration is an important factor in migration decision. This is obvious in the extreme case of refugees leaving a country to escape political persecution. More importantly, it also applies to the more common cases of migrants leaving countries which are facing changes in political and economic systems. Economic analysis of migration decision has a long tradition in the migration literature. Hicks (1932) argued that differences in net economic advantages, chiefly in wages, are the main causes of migration. Sjaastad (1962) put the migration decision in a human capital framework, such that a person will migrate if the present value of expected increased earnings exceeds the present value of investment costs. Almost all modern analyses of migration decisions follow this framework 1. In most migration decision models, differential net present value of migrants earnings in the receiving country and the country of origin is the key determinant. However, as a rule, individuals do not have perfect information on future earnings in different countries. Their decisions to emigrate depend very much on their expectations about future earnings but information on subjective expectations of future earnings is 2

4 difficult to obtain. Therefore, in most analyses, expected earnings in the receiving country are imputed according to the individual characteristics of the potential migrant. Recognizing the role of political factors in the emigration decision, there have been several attempts to incorporate political factors in the migration decision, using macro-data of the sending and receiving countries. In his study on professional indirect immigration to the United States, Huang(1987) constructed an index of the lack of political and civil rights in his study, and found that this variable displays striking significance in explaining non-returns. The effect of income differentials was found to be quite small. Gani and Ward(1995) included the variable political instability in their study of migration of professionals from Fiji to New Zealand. It was found that political instability is significant and positive related to migration flow. The economic factor income in New Zealand was also found to be significant. Gani(1998) again included the variable political instability together with other economic variables in his study of migration from Fiji to New Zealand from However, living standard differential and political instability do not exert a statistically significant impact upon migration while wage and unemployment differentials have the expected signs. The use of micro-data in the study of the role of political factors has been scanty because of the unavailability of data. Borjas(1984) tried to make a differentiation between economic and political migrants and studied the economic assimilation of Cuban political migrants as compared to other migrants. However, the distinction between political and economic migrants may not be clear-cut as such. Besides being important on its own, political stability affects the emigration decision because it impacts on economic stability. To the extreme, political upheavals may result in loss of wealth through confiscation or violent seizures. Thus we expect that people s expectations of political and economic conditions are interrelated, and the two factors interact to affect an individual s emigration decision. Under such circumstances, to classify an individual as purely a political migrant or an economic migrant, therefore, may not be appropriate. To contribute to the understanding of the interaction between political and economic factors in the micro-decision of potential migrants, in this paper we allow economic and political confidence to 3

5 interact in a discrete choice model of migration decision and test it with specially designed micro-data of Hong Kong. Hong Kong went through an extended period of political uncertainty after Britain signed an agreement with China in late 1984 to hand over sovereignty of Hong Kong on 1 July Despite China s promise to allow a high degree of autonomy under the One Country, Two Systems arrangement and to maintain stability and prosperity of Hong Kong, many Hong Kong people did not find that promise credible. They were worried that the basic structure of the political system would change after 1997 and that Hong Kong s economic prosperity under capitalism would not continue when sovereignty is handed over to communist China. After the future of Hong Kong was settled with the signing of the Sino-British Agreement in late 1984, the number of emigrants leaving Hong Kong increased sharply from the pre-agreement annual flow of about 20,000 (0.4% of the population) to over 60,000 (1% of the population) after the crackdown of student demonstration in Beijing on June 4, 1989 (see Figure 1). The most popular destination countries are Canada, U.S. and Australia, followed by U.K. and New Zealand. Owing to the selection criteria of these receiving countries, emigrants are typically young, well-educated, English speaking and employed as professionals or managers. Many emigrated and accepted much lower paid jobs in the receiving countries. [Insert Figure 1 here] The sizeable outflow of emigrants from Hong Kong over an extended period of more than a decade is rather unusual. It offers an opportunity for us to test our model to ascertain the relative importance of political and economic confidence and how they interact in the migration decision. Our results show that the lack of political confidence increases emigration propensity significantly. Lack of economic confidence also increases emigration propensity though by a smaller extent, whereas expected decrease in income abroad deters emigration by only a small magnitude. The rest of the paper is organized as follows. In Section II, we give a theoretical framework for an individual s emigration decision. The data used in this study are described in section III. The 4

6 empirical framework is presented in section IV. The empirical results are discussed in Section V. We conclude in section VI with some implications of our findings. II. Theoretical Framework The utility function of an individual depends on his lifetime net earnings and the lifetime value of political good he can enjoy. These vary across locations and are different for individuals with different socio-economic characteristics. We define the utility derived from the choice of location 0 (the country of origin), and location 1 (the receiving country) as U io = z i0 δ + w i γ 0 + e io (1) U i1 = z i1 δ + w i γ 1 + e i1 where z i0 and z i1 are vectors of the alternative locations' characteristics as perceived by individual i, and w i is a vector of socioeconomic characteristics of the i th individual. The returns to w, such as returns to schooling, may be different in the two different locations. e io and e i1 are random disturbances. The individual will choose location 1, that is to migrate, only if U i1 > U io, or if the latent variable y i * = U i1 - U io > 0. Thus if we define y i as an observable dichotomous variable which takes the value of 1 if the individual chooses to migrate, and the value of 0 otherwise, the values of the observable random variable y i are determined as y i = 1 if y i * > 0 (2) y i = 0 if y i * 0 Rewrite y i * = (z i1 - z i0 ) δ + w i (γ 1 - γ 0 ) + (e i1 - e i0 ) (3) = x i β + e i * then Pr[y i = 1] = Pr[y i * > 0] = Pr[e i * > -x i β] (4) We will use the logit distribution for our empirical estimation in Section IV 2. A special interest of this paper is on the location-related variables z i1 and z i0. We take them to be the economic stability and political conditions in the receiving country and the country of origin. 5

7 These variables are particularly important for the migration decisions of those living in a country facing potential future economic and political instabilities, as in some LDCs, Asian and former Soviet bloc countries. The incorporation of these variables modifies the human capital theory of migration. Since z i1 - z i0 enters into the decision criterion, if it is strongly positive, an individual will migrate even if the differential lifetime earnings as represented by w i (γ 1 - γ 0 ) is negative. In fact, this was observed in Hong Kong when she was facing the change-over to communist China in A lot of people chose to migrate to another country even though they had to suffer a reduction in lifetime earnings for the pursuit of the assurance of political freedom. The greater is the anticipated loss in political freedom and economic stability in the country of origin, the larger will be the propensity to outmigrate. Besides, the subjective expectations of an individual regarding political and economic conditions are likely to be related, and thus they enter into the migration decision in an interactive way. For example, the contribution of political good to y* depends on the economic stability an individual can enjoy at that particular location. In other words, z i1 and z i0 contain interactive variables of location-specific political and economic conditions. For simplicity, we can regard w i (γ 1 - γ 0 ) as the differential lifetime net earnings in the receiving country and the country of origin. When the cost of migration is held constant, expected increase in income would increase the propensity to emigrate. We have in our study a direct measure of the expected change in income abroad DY. The variable DY may not be able to fully capture the effect of change in life-time net earnings, and other socio-economic variables are included in w i to explain the differential net earnings that varies across individuals, through their effects on the earnings stream and the cost of migration. The differential lifetime net earnings also depends on the location parameters γ 1 and γ 0. One important factor in determining the differential lifetime net earnings and hence migration propensity is education. Education affects the likelihood of migration in two ways, first, through the earnings gap between the two countries and second, through the migration cost. If the return to schooling is higher in the receiving country than in the country of origin, individuals with higher levels of schooling will experience a higher 6

8 earnings gap, and thus will be more likely to migrate than individuals with lower levels of schooling. On the other hand, if the return to schooling is higher in the country of origin, people with lower levels of education will be more likely to migrate. On the cost side, it is generally accepted that individuals with a higher level of education are more efficient in collecting information and thus have a lower migration cost. This factor results in a higher probability of migration for the more educated. If subjective expectation about change in income is held constant, the cost effect of education will dominate, and we expect education to have a positive effect on the propensity to emigrate. Furthermore, it makes a difference where the schooling was acquired. If the potential migrant has previously acquired some education in the receiving country, the cost of migration will be much lower. At the same time, the embodiment of country-specific human capital in the receiving country will increase his lifetime earnings there and thus increase his propensity to migrate. On the other hand, schooling acquired in the country of origin may not be as productive in the receiving country and thus contribute less to emigration. Another factor that affects the expected earnings streams of the potential migrant is working experience. The effect of working experience again depends on the differential returns to experience in the receiving countries and the country of origin. A longer working experience may also imply a greater loss in specific human capital invested in on-the-job training if the individual migrate to another country. Base on this consideration we may expect those with more working experience to have a lower propensity to emigrate, other things being the same. However, taking into account the earnings effect and the specificity effect, the overall effect of working experience on emigration propensity is ambiguous. The occupation of the individual is another human capital variable that may affect the differential net earnings in the receiving country and the country of origin. The demand for workers of different occupations and skill types may be different in different countries. If the individual is in an occupation that is in greater demand in the receiving country, the potential migrant may expect a greater increase in expected net earnings upon migration, and thus have a higher propensity to outmigrate. 7

9 In the traditional human capital model of migration, age is expected to have a negative effect on the propensity to migrate since an increase in age will reduce the amount of time left for recouping the return to migration, and the cost of migration is expected to be higher for people of older age. However, in our model, it is possible for an individual with negative differential lifetime earnings between the receiving country and the country of origin to migrate, as long as this is more than offset by the gain in political good. If this in fact is true, other things being the same, older people who migrate will suffer from a smaller negative differential in lifetime earnings because the period remaining for the negative differential is smaller. On the other hand, lifetime improvement in political good is also smaller. When we combine these two effects together with the effect of age on migration costs, the overall effect of age on the propensity to migrate will be ambiguous. One other factor that is often considered in the migration literature is the network effect related with the presence of family members abroad. If an individual has family members already abroad, it may increase the propensity to migrate by reducing the cost of migration. III. The Data The data were collected in a survey conducted in the summer of 1991 to study emigration patterns in Hong Kong addresses were drawn by a random sampling method with the help of the Census and Statistics Department. The details of the sampling procedures are described in the Appendix. After eliminating observations with inconsistencies and missing observations, we further restricted our sample to those working individuals under the age of 65. We ended up with 572 observations for our analysis, of which 355 are male and 217 are female. Our data set contains some socio-economic variables which are important to emigration decisions but not available in other data sets. The information on emigration tendency was obtained by asking the respondents whether they intend to emigrate. We have several variables which may measure political factors in emigration decisions. Information on the expectation of the respondents about future economic and political situations was obtained by asking whether they have confidence towards the 8

10 future economic and political conditions of Hong Kong 3, and whether they have confidence towards the Hong Kong Government. We have direct information on the expected change in income of the respondents should they emigrate. Our data set also contains information on the kind of passport held by the respondent and the number of years of schooling abroad. The definition of all the variables used in our study are presented in Table 1, while their mean and standard deviations are given in TableA1 [Insert Table 1 here] When we compare the intended 'movers' with the 'non-movers', we can see from Table A1 that the movers are more educated with more schooling acquired in foreign countries. The average age of the two groups are very close, with the movers being slightly younger. However, there is a notable difference between the two groups on their confidence about economic and political future. A high proportion of the movers do not have economic or political confidence % of them do not have economic confidence as compared to 33.2% for the non-movers, and 65.1% do not have political confidence, as compared to only 43.2% for the non-movers. Similarly, 55.9% of the movers do not have confidence towards the Hong Kong government, as compared to a lesser 42.4% among non-movers. Another important observation is that in Hong Kong, many individuals would like to emigrate even though by so doing, their expected change in income is negative. From the table, movers on average expect their income abroad to be 3.8% lower than their current income It supports our postulation that if there is political uncertainty in the country of origin, individuals may want to emigrate even if they expect a decrease in income upon emigration. The expected decrease in income abroad is even higher among non-movers, which amounts to 5.6%.These figures suggest that an expected decrease in income has played a role in holding back the desire of some individuals to emigrate, but not strong enough for other individuals. The marginal effect of a change in income on emigration propensity is analyzed in greater detail in the probability model described below. 9

11 IV. Empirical Framework To model empirically how the emigration decision of an individual is affected by political and economic confidence alongside with other socio-economic variables, we have to deal with the problem of high correlation between political and economic confidence if we are going to estimate these two variables together. We expect that an individual who is more optimistic about the political future of Hong Kong also has greater confidence regarding her economic future. This is confirmed by the statistics in Table 2, which shows that the correlation coefficient between the variables NPCONF and NECONF is indeed quite high, being 0.57, and is statistically significant. [Insert Table 2 here] Another issue is whether the variables NECONF and NPCONF are highly dependent on other explanatory variables in the emigration decision like age and schooling. The results of the likelihood tests as presented in Table A2-A3 show that, contrary to expectation, NECONF can only be explained significantly by NPCONF, and vice versa, whereas each of these two variables cannot be explained adequately by all other variables even together, like age, schooling and other socioeconomic variables. The evidences presented do not support the estimation of a reduced form equation, which would find a substitution for NECONF or NPCONF by other socioeconomic variables. We resolved the problem of a high correlation between NPCONF and NECONF by creating four interactive dummies NENP, NEP, ENP and EP. These interactive dummies capture the interactive effect of political good and economic stability as discussed in our theoretical framework. The definitions of these variables are given in Table 1. The variable EP is left out in subsequent regressions so that the corresponding group of respondents who have both economic and political confidence serves as the reference group. From Table 3, we can see that the correlations between these interactive dummies are much smaller than that between NECONF and NPCONF, ranging from only from 0.1 to [Insert Table 3 here] 10

12 Besides the variables on confidence towards the political and economic futures, we also have another measure of political factor no confidence towards the Hong Kong government NCHKG. We expect those who do not have confidence towards the Hong Kong government to have a higher propensity to emigrate, other things being the same. Another variable on the place of birth may also reflect the political orientation of an individual, especially those who were born in the Mainland of China. From our data set we create a dummy variable CHINA which takes the value of 1 for individuals born in the Mainland of China. Some of these individuals may have been political migrants or refugees who once chose to leave the Mainland China because their perceived value of political goods are lesser there, and they may now have a higher propensity to emigrate from Hong Kong for similar reason. Our data set contains valuable information on the expected change in income should the individual emigrate. This enables us to study how the expected differential income in the original and receiving countries affects the emigration decision, and also the partial effect of various socio-economic variables on emigration, as well as the effect of political and economic stability, holding constant the expected income differential. A dummy variable on occupation called PROF is included to study whether being in the professional managerial position will affect the emigration propensity when other factors are being held constant. The variable PASS which is a dummy on possession of passport of the five favorite destination countries (i.e. Australia, Canada, New Zealnad, U.K. and U.S.) is included partly to capture the insurance effect. In Hong Kong, it is often suggested that the holding of a foreign passport serves as an insurance so that the passport holder can stay in Hong Kong if the political situation is stable and leave in case the political situation turns bad and becomes unacceptable. The insurance effect thus implies that individuals who have already acquired foreign passports have a lower propensity to migrate now. On the other hand, the acquisition of a foreign passport may indicate a self-selection effect that foreign passport holders have a higher preference to emigrate. The overall effect is thus ambiguous. The variable 11

13 FAMABR which measures the number of family members living abroad is included to capture the network effect described in the previous section. We estimated the effect of various variables on the emigration decision of an individual by the logit model. The dependent variable move is dichotomous, taking the value 1 if the individuals choose to move, and 0 otherwise. The maximum likelihood estimates of the logit models are reported in Table 4. V. Empirical Results We estimated three logit models in Table 4. Model 1 includes various conventional human capital variables. Model 2 includes three additional interactive dummies on political and economic confidence. Maximum likelihood test confirms that this group of interactive dummies are statistically significant. It can be seen that the sign of the logit estimates remain unchanged upon the inclusion of these interactive dummies, though the magnitude of the variables change slightly. This is consistent with our results in Tables A2 and A3 that political and economic confidence are not statistically dependent on other socioeconomic variables. In Model 3, we include an alternative measure of confidence NCHKG as described in the previous section. [Insert Table 4 here] If the motivation for emigration is mainly an expected increase in net income, we should find the estimate of DY to be positive and significant. In Table 4, it is shown that though the estimated coefficient of DY is indeed positive and statistically significant, its magnitude is very small. An increase in expected income abroad by 10% will only raise emigration propensity by about 1.6%. On the other hand, the effects of the interactive variables on political and economic confidence are of much greater importance. The estimated coefficient of the variables NENP, NEP and ENP are all statistically significant and of relatively large magnitude. Specifically, if an individual has neither economic nor political confidence in the future of Hong Kong, he or she will have 24.9% higher 12

14 propensity to migrate than those with political and economic confidence. It is interesting to observe that the incremental effect of having no political confidence on emigration is higher than that of having no economic confidence. If an individual has no political confidence, even though he has economic confidence, he has a 21.4% higher propensity to emigrate than those with both political and economic confidence. On the other hand, if an individual has no economic confidence but has political confidence, his propensity to emigrate is only 17.6% higher than those with both political and economic confidence in the future of Hong Kong. These findings seem to indicate that the lack of political confidence is a stronger push factor than the lack of economic confidence in the emigration decision. The alternative measure of political factor NCHKG also indicates that if an individual has no confidence towards the Hong Kong government, his propensity to emigrate will be 10% higher than the reference group. The consideration of economic factors alone as found in most previous migration studies would not be adequate for the study of the emigration phenomenon in Hong Kong and other places with political uncertainty. Of course, political stability may be related to the economic well being of individuals in the long run, and might have a strong impact on wealth if serious political instability occurs. It has been observed in previous studies that socio-political instability creates uncertainty in property rights protection 4. In this perspective, even the political factor that shows a dominant effect over the economic factors in our study may have some economic basis. Despite this consideration, the argument for the importance of including political factors in emigration models remains intact 5. The effect of other socio-economic variables on the emigration decision by and large conforms with the analysis in our theoretical framework. The effect of schooling is positive, whether it is acquired locally or abroad. The marginal effect of schooling acquired abroad is about 2.8%, and is as expected higher than the effect of schooling acquired locally, which is about 2.2%. Note that the schooling effect here cannot be explained by differences in expected change in income or differences in expectation on political and economic stability across schooling groups, since these other variables are already held constant in our model. Instead, it supports the hypothesis that the more educated are more efficient in gathering and processing information overseas, thus reducing the cost of emigration. The positive 13

15 correlation between education and some unobserved characteristics favorable to emigration, such as language ability and higher motivation, can also contribute to the positive effect of schooling on emigration propensity. For instance, in Hong Kong where English is taught in all schools, it is most likely that education is positively correlated with proficiency in English. The latter is important in reducing the cost of migration and increasing the potential lifetime earnings in the receiving country. In our model where English proficiency is not entered as an independent variable, the coefficient of schooling partly captures the positive effect of English proficiency on the propensity to migrate. Working experience has an U-shaped effect on the propensity to emigrate. Its effect is negative at sample mean, indicating that at sample mean, an increase in working experience will decrease the propensity of an individual to outmigrate. This occurs when the effect of the loss in specific human capital upon migration dominates. The effect of the occupation variable PROF on emigration propensity is not statistically significant. It does not mean that professionals do not have a higher propensity to emigrate compared to other occupations. As a matter of fact, there are more professional and managerial workers among the movers than non-movers, as can be seen from Table A1. Most of these professionals are highly educated senior staff, and we have found that education does have a significant positive effect on emigration. What our findings suggest is that when we net out the effect of schooling and other human capital variables, occupation in itself does not have a statistically significant effect on emigration propensity 6. The effect of age on the propensity to emigrate is non-linear. The emigration propensity of individuals aged between 25 and 35 is higher than those aged below 25, while the estimated coefficient of other age groups are not statistically significant. This finding cannot be explained adequately by the traditional human capital model of migration, which hypothesize that emigration propensity will decrease with age since migration cost increases with age, and there is a shortening of the period for the migrant to recoup the return to investment in migration as age increases. On the other hand, this imposes no problem of interpretation for our model. In the first place, we have included explicit variables on expected change in income so that the age effect measures something other than the income effect. It is 14

16 true that migration cost still increases with age and thus age has a deterring effect on emigration because of that. However, in our model it is possible for an individual to emigrate even if the differential lifetime net earnings are negative, since he may be prompted by expected improvement in political good in the country of destination. Under such circumstances, an increase in age will reduce both the number of periods left for an individual to face a decrease in lifetime earnings, and at the same time the number of periods left for the individual to enjoy an improvement in political good. It is thus uncertain which effect will dominate. Nevertheless, it is conceivable that migration cost increases more rapidly as age increases, and this effect may dominate to deter the older individuals to emigrate. The coefficient of the variable FAMABR is not statistically significant, showing that the network effect found in some previous studies is absent in our case. The effect of holding a passport of the five most popular destination countries do not affect the emigration propensity as well. It indicates that any positive selection effect of holding a foreign passport is roughly offset by the negative insurance effect as discussed in Section III. The estimated coefficient of the sex dummy MALE is not statistically significant, indicating that there is no gender differences in emigration propensity in Hong Kong. VI. Conclusion and Discussions In this paper, we study how the economic and political factors affect the probability of an individual to emigrate in Hong Kong. An individual maximizes his utility in his migration decision, which depends on political and economic goods in the sending and receiving countries. Thus lifetime net earnings are not the only consideration in his migration decision as implied by the traditional human capital approach. An individual may choose to migrate even if his lifetime net earnings is lower in the receiving country. This is especially probable in countries facing political uncertainties and possible loss of political freedom, as in Hong Kong and some LDCs. In our sample, the intended movers on the average expect their income abroad to be 3.8% lower than their current income. One interesting finding of this paper is the interactive effect of the factors of political and economic conditions as perceived by individuals. As expected, individuals with neither economic nor 15

17 political confidence have the highest probability to migrate, being 24.9% greater than those with both economic and political confidence. If an individual has no political confidence, even though he has economic confidence in the future of Hong Kong, his propensity to migrate is about 21% higher than those with both political and economic confidence. On the other hand, if an individual has political confidence but no economic confidence, his probability to migrate is 17.6% higher those with both political and economic confidence, other things being equal. This finding suggests that the incremental effect of political confidence is higher than that of economic confidence in an individual s emigration decision. Using another measure of political confidence, if an individual have no confidence towards the Hong Kong Government, the probability to emigrate is 10% higher than others, other things being the same. Those who were born in Mainland China have again an 11% higher probability of emigration, other things being the same. Note that the push effect of a lack of political and economic confidence is of a relatively large magnitude It is much greater than the marginal effect of an expected change in income abroad. A 10% change in expected income will only change the probability of emigration by 1.7%. Our study indicates that for countries which are facing uncertain political changes, as in many LDCs and former Soviet bloc countries, differential lifetime earnings may not be the key factor in the migration decision. It is important to incorporate the factor of political good into the utility function in the analysis of migration decision. The probability of emigration does not decrease with age as expected by the traditional human capital model. Individuals aged between 25 and 35 are found to have a higher emigration propensity than those who are aged below 25. There is no difference in emigration propensity among other age groups. The positive age effect on emigration propensity which exists among tyounger age groups cannot be explained adequately by the traditional human capital model which predicts that individuals emigrate in response to an expected increase in income abroad. But this is possible when individuals emigrate to take advantage of a gain in political good despite a decrease in expected income. 16

18 The effect of schooling on emigration is consistent with the human capital model. The marginal effect of schooling on emigration probability is 2.4% for schooling acquired locally and 2.9% for schooling acquired abroad. (Education in general reduces the cost of gathering information and the cost of migration. This is more so if the individual has acquired more education abroad. From the cost considerations, individuals with more schooling abroad have a lower cost of migration because they tend to have more information about the receiving countries, have built up more personal effort and have a lower psychic cost of moving. Besides, we expect the return to schooling abroad in the receiving country to be higher than schooling acquired in Hong Kong because part of the schooling investment is countryspecific.) Compared to previous studies, the minor effect of income differential on migration decision and the pushing effect of political instability in the sending country should not come as a total surprise. For example, Devoretz and Maki (1983) found that the effect of income in Canada for a particular occupation on immigration flow of a particular occupation is not statistically significant. The limitation of his income variable is that it only measures the income in the receiving country without taking into account the income in the country of origin. Besides, political factors are not controlled for in his estimation. Consistent to our study, Huang(1987) found a positive but very small effect of income differentials. It is the political variable which displays striking significance. In his study of migration of professional from Fiji to New Zealand during the period , Gani and Ward (1995) found that the variable income in New Zealand is significant and positively related to the migration flow. However, just as Devoretz and Maki, the income variable measures the income in the receiving country alone, and does not reflect the income differential that is more relevant in migration decisions. The variable political instability is again found to be significant and positively related to migration flow. However, in another study that uses a different empirical model and for a different period, Gani (1998) found that the living standard differential and Fiji s political instability do not exert a statistically significant impact 17

19 upon migration. The author s explanation was that the measures of the variables may not capture the intended factors. All of the above studies employ time-series macro data for their analysis. Our present study is the only study which employs micro-data which can capture individual differences in individuals emigration decisions more accurately. Besides, the design of the model enables us to study the interaction between political and economic confidence, which are highly correlated, and the incremental effect of political and economic confidence on emigration decision. However, as the data set is collected under the scenario of potential drastic changes of political conditions, our conclusions are meant to apply only to sending countries that are facing similar potential drastic changes, rather than to all countries in general. Our finding that education has a positive effect on emigration implies that countries facing political instability may suffer from important welfare loss to the receiving countries through the emigration of the highly educated and the professionals. Education is in general heavily subsidized in most countries, and is an important form of social capital. When an individual chooses to emigrate to maximize his own personal utility and carries with him the embodied human capital, the sending country suffers from a loss in social capital and income generating capacity in the future. On the other hand, the receiving countries gain in welfare because of the increase in productivity due to the inflow of human capital which did not cost to them to invest. Besides, the brain drain problem can result in the shortage of high-skilled workers in LDCs in their attempt to transform into a knowledge-based economy, and thus the future development of the economy 7. The small effect of income differential on emigration effect implies that economic incentives will not effective to retain the potential migrants from outmigrating. Instead, it is of paramount importance that the government foster a stable political and economic environment to reduce emigration. 18

20 Figure 1. Number of Emigrants Departing Hong Kong Emigrants Year Source: Hong Kong Monthly Digest of Statistics, 1980(January) 1998(April) 19

21 Table 1 Definition of Variables EXP EXPSQ SCHABR SCHLO FAMABR DY number of years of working experience EXP * EXP number of years of schooling abroad (other than in China and Hong Kong) number of years of local schooling = total number of years of schooling SCHABR number of family members abroad expected percentage change in income =3 if expected income abroad is 30 percent or more higher than current income =2 if expected income abroad is 20 percent higher than current income =1 if expected income abroad is 10 percent higher than current income =0 if expected income abroad is same as current income =-1 if expected income abroad is 10 percent lower than current income =-2 if expected income abroad is 20 percent lower than current income =-3 if expected income abroad is 30 percent or more lower than current income Dummy Variables MOVE =1 if the respondent chose to definitely move or probably move before or after 1997 = 0 otherwise PROF = 1 PASS=1 SINGLE=1 CHINA = 1 MALE=1 if the occupation of the respondent is managers and professionals if the respondent hold the passport of Australia, Canada, New Zealand, U.K. or U.S. if the respondent is never married if the respondent was born in Mainland China if the respondent is a male AGE2=1 if 25 =< age < 35 AGE3=1 if 35 =< age < 45 AGE4=1 if 45 =< age < 55 AGE5=1 if 55 =< age Reference group: age < 25 NECONF=1 NPCONF=1 NENP=1 NEP=1 ENP=1 NCHKG=1 if the respondent is not confident about economic future if the respondent is not confident about political future if NECONF=1 and NPCONF=1 if NECONF=1 and NPCONF=0 if NECONF=0 and NPCONF=1 Reference group: NECONF=0 and NPCONF=0 if the respondent is not confident towards the Hong Kong Government 20

22 Table 2 Frequency Table of Economic and Political Confidence NPCONF 0 1 Total NECONF Total Correlation Coefficient = Significance Probability < Note: Significance Probability = Prob > R under H 0 : RHO = 0 21

23 N=572 Table 3 Correlation Coefficient Among Variables NENP, NEP and ENP (Significance probability in parenthesis) NENP NEP ENP NENP (0.0000) (0.0003) (<0.0001) NEP (0.0000) (0.0124) ENP (0.0000) Note: Significance Probability = Prob > R under H 0 : RHO = 0 22

24 Table 4. Logit Estimates --- Marginal Effects (t-statistics in parenthesis) Model 1 Model 2 Model 3 Marginal Marginal t-statistics Marginal t-statistics Variable Effect t-statistics Effect Effect CONST (-4.100) (-5.084) (-4.455) EXP (-2.361) (-2.113) (-2.167) EXPSQ (2.284) (2.038) (2.062) SCHLO (3.138) (2.981) (3.105) SCHABR (3.091) (3.041) (3.182) PROF (0.215) (-0.023) (0.176) DY (1.913) (1.851) (1.966) FAMABR (0.886) (0.420) (0.566) PASS (-1.329) (-1.327) (-1.539) SINGLE (-1.233) (-1.425) (-1.288) CHINA (2.246) (2.374) (2.413) MALE (-0.616) (-0.056) (-0.449) AGE (2.282) (2.116) (2.091) AGE (1.581) (1.429) (1.387) AGE (1.324) (1.214) (1.328) AGE (-0.631) (-0.691) (-0.549) NENP (5.990) NEP (2.210) ENP (4.262) NCHKG (2.677) LOGL N

25 Footnotes 1. Stark (1991) proposed an alternative theory of migration theory which models migration decisions as family decisions to spread risk by spreading their labour assets over different markets. 2. The marginal effect of an explanatory variable on the probability to migrate would be non-linear in both cases. In particular, for logit distribution, Pr/ X j = β j e -Xβ /(1+ e -Xβ ). 3. If the respondent answered Í am very confident or I am quite confident, the respondent is regarded as having confidence. It may be desirable to study the two groups separately, but the sample size for those who are very confident is too small for separate analysis. As a sensitivity analysis, we include only those who are quite confident and take away the very confident group. The results remain basically the same. 4. See Alesina, Alberto and Roberto Perotti(1996) 5. We have tried other models which include interactive variables of political and economic confidence with other socio-economic characteristics. The estimates of all these interactive variables are not statistically significant. 6. When we estimated the logit model with PROF but not other human capital variables nor the confidence variables, we found the gross effect of PROF to be positive and statistically significant. 7. Lam(2000) gave an account of the impact of emigration of the highly educated professionals on the shortage of high-skilled workers in Hong Kong and related policy issues. 24

26 Table A1. Summary Statistics - Mean (Standard Deviation) MOVERS NON-MOVERS ALL VARIABLE MEAN STD DEV MEAN STD DEV MEAN STD DEV AGE (8.797) (10.191) (9.833) AGE (0.501) (0.486) (0.492) AGE (0.445) (0.453) (0.451) AGE (0.299) (0.287) (0.290) AGE (0.161) (0.223) (0.208) EXP (9.454) (9.730) (9.662) EXPSQ ( ) ( ) ( ) SCHLO (3.883) (3.514) (3.624) SCHABR (3.337) (1.966) (2.420) PROF (0.496) (0.471) (0.479) DY (2.144) (2.261) (2.230) FAMABR (1.196) (1.106) (1.134) PASS (0.081) (0.109) (0.102) SINGLE (0.471) (0.484) (0.481) CHINA (0.452) (0.438) (0.441) MALE (0.489) (0.485) (0.486) NECONF (0.502) (0.446) (0.471) NPCONF (0.478) (0.478) (0.496) NENP (0.497) (0.414) (0.448) NEP (0.237) (0.228) (0.230) ENP (0.414) (0.340) (0.363) NCHKG (0.498) (0.495) (0.499) N

27 (a) Unrestricted Model (UR): Table A2. Likelihood Ratio test for different specifications of logit models Dependent Variable: NECONF NECONF = b 0 + b 1 NPCONF + b 2 X 2 where X 2 = { EXP, EXPSQ, SCHLO, SCHABR, PROF, DY, FAMABR, PASS, SINGLE, CHINA, AGE2,AGE3,AGE4,AGE5} (b) Restricted Model 1 (R1): Hypothesis 1: b 1 = 0 s.t. NECONF = b 0 + b 2 X 2 (c) Restricted Model 1 (R2): Hypothesis 2: b 2 = 0 s.t. NECONF = b 0 + b 1 NPCONF (d) Likelihood Ratio Tests N 572 LogL UR LogL R LogL R (LogL R1 - Log UR ) (LogL R2 - Log UR ) χ 2 ( 1, 0.05) = 3.84 χ 2 (15, 0.05) = χ 2 (15, 0.01) = Hypothesis 1 Hypothesis 2 rejected accepted 26

28 (a) Unrestricted Model (UR): Table A3. Likelihood Ratio test for different specifications of logit models Dependent Variable: NPCONF NPCONF = b 0 + b 1 NECONF + b 2 X 2 where X 2 = { EXP, EXPSQ, SCHLO, SCHABR, PROF, DY, FAMABR, PASS, SINGLE, CHINA, AGE2,AGE3,AGE4,AGE5} (b) Restricted Model 1 (R1): Hypothesis 1: b 1 = 0 s.t. NPCONF = b 0 + b 2 X 2 (c) Restricted Model 1 (R2): Hypothesis 2: b 2 = 0 s.t. NPCONF = b 0 + b 1 NECONF (d) Likelihood Ratio Tests N 572 LogL UR LogL R LogL R (LogL R1 - Log UR ) (LogL R2 - Log UR ) χ 2 ( 1, 0.05) = 3.84 χ 2 (15, 0.05) = χ 2 (15, 0.01) = Hypothesis 1 Hypothesis 2 rejected accepted 27

29 APPENDIX Sample Design and Sampling Procedure Sample Design A total of 3098 addresses was drawn by systematic sampling method with the aid of Census and Statistics Department in April The listing we used was the frame of living quarters (LQF) which consisted of both permanent and temporary living quarters in Hong Kong. Only records of permanent ones were selected. The sampling of such survey covered the whole territory excluding outlying islands. Selection of Household and Respondent Since each record contained no household and personal information but the address of a living quarter, we had to make use of a multi-stage design, with three different stages of selection: Stage 1. Selection of Addresses In April, a total of 3098 addresses was drawn from 7 systematic replicated samples of the subframe of the LQF. Each replicate contained about 470 records of address. Stage 2. Selection of Household The selection was carried out by interviewers. Interviewers had to call at each selected address and list all households living at that living quarter. A random table pre-attached to each address was used to select one household. Stage 3. Selection of Respondent For each selected household the interviewer was required to list all those eligible in the sample, that is, all persons currently aged 18 or above and resident at the selected household at least half of the time at the selected household in the past 3 months or in the coming 3 months. Only a respondent would then be selected according to the random selection grid(a modified Kish grid). Response Statistics A total of 1595 successful interviews were made. However, 43 of them contained illogical and/or inconsistent data which could not be corrected. As a result, the number of valid cases that we could make use of was