Journal of Asian Economics 10 (1999) 91–109

Labor market participation of urban women in Southeast Asia by migration status Evidence from Microdata J. Malcolm Dowling, Christopher Worswick* Department of Economics, University of Melbourne, Parkville, Victoria, Australia 3052

1. Introduction The operation of urban labor markets and the migration into these markets has been studied from a number of perspectives. The dual labor market theory of Todaro (1969) and its extensions (Harris and Todaro, 1979; Corden and Findlay, 1975; Williamson, 1989; Rosenzweig, 1989) suggests that migrants who come to the city are prepared to wait for openings in high paying jobs in the formal sector. The restrictive assumptions of this class of models have been criticized, and its implications do not hold up to empirical tests in a number of developing countries in Asia and Latin America (see Ghatak et al., 1996, for example). Nevertheless, these models continue to be the standard by which other models of individual migration behavior are judged. Alternatively, researchers have stressed the importance of family rather than individual behavior in decisions relating to migration behavior. This family perspective has been used to explore the returns to migration and remittances (Lucas and Stark, 1985; Stark, 1991; Knowles and Anker, 1981; Hoddinott, 1994; Stark and Bloom, 1985; Vijverberg and Zeager, 1994) and intergenerational transfers and altruistic motives (Ghatak et al., 1996). The relative importance of macroeconomic variables vis-a`-vis human capital has also been explored by Tunali (1996) and the problem of selectivity bias has been examined by Robinson and Tomes (1983). Many of the more recent articles, which have adopted the household theoretic approach, have tested their models using household survey data and longitudinal data sets; and have used multinomial choice models to explore the probabilities of migration.

* Corresponding author. Tel.: ⫹61-3-9-3447098; fax: ⫹61-3-9-3446899. E-mail address: [email protected] 1049-0078/99/$ – see front matter © 1999 Elsevier Science Inc. All rights reserved. PII: S 1 0 4 9 - 0 0 7 8 ( 9 9 ) 0 0 0 0 8 - 1

92

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

The diversity of the models and tests reported in the literature suggests that the decision to migrate may be very complex and not easily explained by simple models of individual choice. Family considerations and individual human capital endowments as well as local conditions in rural areas, including population pressure, employment opportunities, and wage levels, are important push factors. At the same time, labor market segmentation, economic growth, and employment are important pull factors, which stimulate migration from rural to urban localities.

2. The economics of women’s migration Interestingly enough, almost all models in the development literature reviewed are silent on what differences there may be in migration behavior between men and women. However, nearly all of the sample data sets relate to men and the models of circular migration and intergenerational transfer motives are couched in terms of young male migrants. There are a few studies involving women but they tend to be concerned with discrimination issues. Cohen and House (1993), for example, find that occupational discrimination rather than pay discrimination was responsible for earnings differentials in Khartoum, Sudan. Eaton (1992) finds that the Todaro model is not supported by data from the Brazilian northeast. Eaton argues that, contrary to Todaro, women who are not in the labor force while residing in rural areas begin to work in the informal sector as soon as they reach the city. This shortage of analysis of women’s migration and labor market behavior is puzzling, given the considerable attention that women’s education and labor force participation has been given in the development literature. Generally, women are viewed as playing an important role in the process of development. A virtuous cycle or paradigm has been suggested wherein resources devoted to lifting the educational attainment of women bear fruit in a number of ways. Women’s education increases labor market participation and provides better employment opportunities for women, and hence raises their incomes. This raises the status of women both in society and within the family. There are also positive externalities to such a process including a reduction in fertility and population growth, improved health and life expectancy of children, reduced infant mortality, and a reduction in environmental degradation (Asian Development Bank, 1989, pp. 170 –171; and Todaro, 1994, Chapter 11). The process by which women enter the labor market is critical to an understanding of how improvements in women’s income levels and social status in developing countries can be facilitated and supported by government policy. Evaluation of the benefits of educating women has been carried out in a number of studies (Durand, 1975; Psacharopoulous, 1985; Subbarao and Raney, 1995; among others). These results have led many development economists to argue that educating females yields substantial economic benefits— higher economic returns than comparable expenditures on men. At the same time, there has been a limited number of studies on the labor market experience of migrant women (Schultz, 1983; Fields and Schultz, 1982; for Colombia being an exception). As far as we know, no studies for Asian countries have been conducted that examine the relationship between migration, education, and labor force participation of

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

93

women. To do this, an analysis of the determinants of labor force participation, using household level data, is necessary. It may be possible to model and empirically test women’s models of decisions to migrate as well as subsequent entry into the labor force within the context of a household decisionmaking model. We have developed such a model with respect to the impact of education on labor market outcomes (see Cameron et al., 1997). In the present paper we have a more modest objective, primarily because data limitations preclude estimation of a fully specified model of women’s migration, which could then be tested using data from several countries. Alternatively, a series of hypotheses are proposed and then tested using several cross-section data sets from three Southeast Asian countries. These hypotheses flow either from a simple model of household maximizing behavior or from similar maximizing behavior for an individual. Generally, the presence of additional women migrants in the urban labor force both contributes to labor supply and, perhaps, underemployment and unemployment. However it also enlarges the pool of labor willing to work at low wages, and in this way, assists in the process of development through concentration on labor intensive industrialization and supports export-led growth by keeping wages low. In this paper, the labor force participation decisions of urban women are analyzed with the focus on differences in participation decisions between migrant and non-migrant women. The word migrant refers to a woman who, in the survey, indicated that she is currently residing in an urban area and her childhood residence was in a rural area. The word non-migrant refers to a woman who, in the survey, indicated that she is currently residing in an urban area and that her childhood residence was in an urban area.

3. Description of data sources The data come from two sources: 1) the World Fertility Surveys (WFS) conducted by the International Statistical Institute; and 2) the Demographic and Health Surveys conducted by the United States Agency for International Development. The use of the WFS data enables a comparison of countries with different social norms and development experience in the early stage of industrialization. The goals and focus of the DHS were similar to those of the WFS and information on many of the same variables were collected, including age, family size, labor market outcomes, and education. The fact that the WFS surveys were collected in the mid-1970s, while the DHS surveys were collected in the late 1980s and early 1990s, means that we are able to investigate the effects of economic development over the intervening period on the participation decisions of migrant and non-migrant women living in urban areas. While data from these two surveys are available for a number of countries, this study focuses on Indonesia,1 the Philippines, and Thailand, three countries in Southeast Asia that experienced rapid growth and structural changes during the period encompassed by the surveys. These countries are similar in several respects, as they are tropical primary product producers that made the transition to industrialization at roughly the same time. Per capita incomes are also comparable, although Thailand has drawn away from the Philippines in

94

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

recent years. Indonesia, which started out with a much lower per capita income than either Thailand or the Philippines, was drawing closer by the end of the sample period.

4. Hypotheses to be tested Four main hypotheses are analyzed. The first two relate to general relationships between the woman’s personal and family characteristics and her decision to work. The next two hypotheses relate to differences in participation behavior between migrant and non-migrant women. 4.1. Higher education levels increase women’s participation probabilities Human capital theory predicts that increases in the level of formal education will lead to an increase in wage offers in urban labor markets. The higher wage offers resulting from higher education levels will make women more likely to choose to work (see Williamson, 1989). 4.2. The presence of children reduces women’s participation probabilities Women with children present in the household are less likely to work in the labor market because they may need to care for their children. This is a common finding in developed countries and in Asian developing countries (see Cameron et al., 1997). 4.3. Migrant women are more likely to work than are non-migrant women After controlling for socioeconomic characteristics, women migrants to urban areas are more likely to work than are non-migrants. The motivation for this hypothesis is that migration into urban labor markets by women is largely a function of economics. Urban labor markets are more fluid and flexible than their rural counterparts and are attractive to women looking for work. 4.4. Participation decisions of migrant women are more sensitive to macroeconomic conditions than those of non-migrant women Labor force participation rates of both migrants and non-migrants are likely to be higher in periods of buoyant aggregate economic activity and in countries which have been industrializing and experiencing rapid structural changes in output and employment. However, the marginal effect of improved macroeconomic conditions on participation decisions is likely to be larger for migrant women compared with non-migrant women, since the migrant women may have moved to the urban area so as to take advantage of the improvement in labor market opportunities there. Non-migrants will be more prone to enter into home production, ceteris paribus.

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

95

5. Reduced form participation model Probit estimation of a binary choice model over women’s participation in the urban labor market is carried out over three pooled samples for each of the three countries. In each case, data from the WFS survey and the DHS survey are pooled. The specification of the index of the binary choice model is: I i共t兲 ⫽ X i共t兲 ␤ ⫹ ␤ 1Migrant i ⫹ ␤ 2DHS i ⫹ ␤ 3共Migrant i兲 ⫻ 共DHS i兲 ⫹ v i共t兲

(1)

where woman i is employed in period t if Ii(t) ⱖ 0, and out of the labor force otherwise;2 ␤ is a parameter vector, ␤1, ␤2, and ␤3 are parameters and vi(t) is a mean zero, normally distributed error term. The vector of personal characteristics, Xi(t), contains controls for the individual’s age, education, literacy, marital status, presence of children, and fertility history. The variable Migranti identifies women who had a childhood residence that was in a rural area.3 The variable DHSi is a dummy variable that identifies women from the DHS data (which is the more recent survey).4 This allows for an intercept shift in the index to take into account the changing situation over time in the country. Finally, the DHSi variable and the Migranti variable are included as an interaction. This allows for the participation probability of immigrants to be different in the two survey years, ceteris paribus. 6. Sample selection and analysis of sample means The sample is restricted to women who reported that they were living in an urban area (defined as town or city in the questionnaire) at the time of the survey. Since growth in employment has been primarily in urban areas in Southeast Asia, an analysis of urban women seems appropriate. To facilitate the analysis, the sample was also restricted to ever-married women between 15 and 49 years of age. In each of the WFS and the DHS surveys, women over the age of 49 were not surveyed. In the DHS for Indonesia and Thailand, female household members who were under the age of 15 were interviewed but this was not the case in any of the other surveys. In all of the surveys except for the DHS of the Philippines, only ever-married women were sampled. In order to allow for comparability across the surveys, it was therefore necessary to impose the restriction that only ever-married women between the ages of 15 and 49 years be included in the analysis. A separate analysis of never-married women in the DHS of the Philippines is carried out in section 9 of the paper. Finally, each survey is a weighted sample of the populations of women in each country at the time of each survey. In the analysis of this paper, we employ the weights variables provided in the data sets to allow for generalizations of the results to the entire population of urban women in each country. The sample means recorded in Table 1 for the DHS samples reveals a number of interesting comparisons between women who have migrated from rural areas (migrants) and women who have always been urban dwellers (non-migrants). Labor market participation rates are lower for migrants than non-migrants in the Philippines and higher in Indonesia. In the Thai data, the sample participation rates of migrant and non-migrant women are very

96

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

Table 1 Sample means from DHS data Indonesia 1987 Variable Labor market participation Age (years) Education None Primary Secondary Post-secondary Illiterate Family Children last five years Children Births Married Sample

Philippines 1993

Thailand 1987

Non-Migrants

Migrants

Non-Migrants

Migrants

Non-Migrants

Migrants

0.3344 32.70

0.4226 32.75

0.6120 33.77

0.5451 33.72

0.6084 32.66

0.5894 32.17

0.0673 0.4652 0.4081 0.0594 0.0795

0.1869 0.5924 0.2010 0.0198 0.2431

0.0084 0.1999 0.3809 0.4105 0.0125

0.0233 0.3980 0.3736 0.2051 0.0376

0.0615 0.4715 0.2981 0.1689 0.0447

0.0563 0.6993 0.1641 0.0803 0.0598

0.9463 2.995 3.332 0.9174 2551

0.9160 2.930 3.386 .9012 1990

1.077 2.907 3.078 0.9268 2402

1.137 3.315 3.573 0.9430 2275

0.6878 2.054 2.136 0.9066 1188

0.5924 2.040 2.143 0.9263 1208

close at approximately 60 percent. Therefore, the comparison of the sample means support Hypothesis 4.3 in the Indonesian case but not in the case of the other two countries. With respect to education, the Philippines has the best educational record, having a much higher proportion of secondary and post-secondary graduates than either Thailand or Indonesia for both sample groups. In Indonesia and Thailand, the means of the three measures of birth history (children born in the last five years, living children, and total number of births) are similar for migrants and non-migrants. In the Philippines data, migrant women have significantly larger mean values for each variable. Table 2 contains the equivalent sample means taken from the earlier WFS samples. Comparing the WFS sample means with the DHS sample means reveals several interesting features. Educational attainment of both migrants and non-migrants improved quite dramatically in Indonesia between 1976 and 1987. The proportion of illiterate women declined by 12 percentage points among non-migrants and 13 percentage points among migrants. The proportion of women with no education also dropped substantially. There was also a dramatic rise in the proportion of women with primary and secondary level education. In Thailand, the improvement was less dramatic. The illiteracy and zero-completion rates fell but they were already low in the 1970s. There was also an increase in the percentage of women with post-secondary education but surprisingly, a decline in secondary levels among the non-migrants probably as a result of a large increase in the proportion of women who had post-secondary level education. In the Philippines, in both the migrant and non-migrant samples, there was a rather dramatic drop in the proportion women who ended their education at the secondary level and this was not offset by increases in post-secondary education as in Thailand. Rather, the percentage of women who ended their studies at the primary level increased. The reasons for the rather dramatic differences in educational achievement among the three countries is partially the result of differences in public policy. Indonesia made a

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

97

Table 2 Sample means from WFS data Indonesia 1976 Variable Labor market participation Age (years) Education None Primary Secondary Post-secondary Illiterate Family Children last five years Children Births Married Sample

Philippines 1978

Thailand 1975

Non-Migrants

Migrants

Non-Migrants

Migrants

Non-Migrants

Migrants

0.3490 31.90

0.4174 32.04

0.4419 33.80

0.4187 33.76

0.6583 32.91

0.5859 32.77

0.3793 0.3446 0.2447 0.0314 0.2024

0.5608 0.3096 0.1179 0.0117 0.3721

0.0152 0.0744 0.5879 0.3225 0.0297

0.0209 0.1764 0.6597 0.1430 0.0471

0.1170 0.4571 0.3416 0.0807 0.0656

0.1435 0.7104 0.1162 0.0233 0.1133

0.9211 3.145 3.688 0.8831 1927

0.9003 2.881 3.497 0.8508 996

0.9724 3.582 3.875 0.9478 2697

1.052 3.788 4.155 0.9576 1897

0.7738 2.889 3.058 0.9024 270

0.7842 3.044 3.389 0.9135 305

concerted effort after the second oil shock to increase both educational standards and coverage and devoted resources to accomplishing this objective. Given that it started from a low base, its achievements were quite outstanding. Thailand, on the other hand, also experienced rapid growth over this period. However, it achieved considerably less improvement. There was a modest movement toward higher achievement but primary education remained the standard among rural migrants (70 percent) despite 16 years of rapid economic growth. In the Philippines, educational standards were already much higher than in either of the other countries. However there was a deterioration in educational attainment among women in both parts of the sample. Secondary completion rates went down and primary completion rates went up. This was partially compensated by a rise in post-secondary enrollment. Nevertheless, the proportion of women in either of the two lowest education categories (none and primary) increased. While public policy in the Philippines may be partly to blame, the overall deterioration in the macroeconomic environment over the intervening years between the two samples is more directly responsible. For much of the period, per capita income in the Philippines stagnated or even declined. Only in the past few years have incomes started to rise again. In Fig. 1, the growth paths of real GDP per capita are plotted for the three countries, with 1975 as the base. Thailand experiences the highest growth over the sample period, with much of it occurring after 1985. Indonesia experienced a similar rate of growth until the mid-1980s and then had slower growth than Thailand. The Philippines experienced slow, and at times negative, growth through until 1985 when growth increased somewhat. Labor force participation rates for women rose dramatically in both migrant and nonmigrant groups in the Philippines (increase of 17 percentage points for non-migrants and 13 percentage points for migrants). This is consistent with women entering the urban labor market in response to the growth experienced in the Philippines after the mid-1980s. However, participation rates remained steady for migrants and fell for non-migrants in Thailand and to a lesser extent in Indonesia. These results are altogether rather striking and

98

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

Fig. 1. Growth paths of GSP per capita by country (1975 ⫽ 100, in 1987 U.S.$). Source: Maddison (1995), Table B-10e.

unexpected. A priori we would have expected the Philippines to have the most modest increase in participation for macroeconomic reasons. Nevertheless, these results do suggest that economic opportunity in the late 1980s and early 1990s in the Philippines may have been a push factor toward labor market participation, whereas there was an income effect operating on urban residents to reduce the rate of labor force participation in the other two countries. 7. Results from probit estimation Since a woman’s decision to work may depend on a number of factors such as age, education, presence of children, and marital status, the differences in participation rates between migrants and non-migrants in each country may partially reflect mean differences in these demographic characteristics. In order to carry out tests of the hypotheses (4.1– 4.4), it is necessary to control for these characteristics, which can be done through Probit estimation of the binary choice model defined in Eq. (1). In Table 3, marginal probabilities are presented, which are derived from the Probit estimates.5 The marginal probabilities are evaluated at the mean values of the exogenous variables defined over the pooled sample from all three countries. The variables included in the estimation can be organized into three groups. The first group contains variables likely to affect labor market earnings. The second group of variables contains proxies for the value of the woman’s time in home production. The third set of variables contains controls for whether the woman is from the first or second survey, her migrant status, and an interaction of these two variables. The coefficients on these variables allow us to see 1) the change in participation over time (which can be interpreted in terms of the economic growth over the period); 2) the difference in participation behavior between migrant and non-migrant women; and 3) how the later relationship has changed between the surveys. Direct estimates of market earnings were not collected in either the WFS or the DHS. Alternatively, earnings are assumed to be a concave function of age and the results are

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

99

Table 3 Marginal probabilities from probit estimation of urban women’s participation decisionsa Indonesia Intercept Age Age2/100 No education Secondary education Post-secondary education Illiterate Number of children born in last 5 years Number of children present Number of births Widowed, separated, or divorced DHS survey Migrant (Migrant)b (DHS Survey) Percent correct predictions Sample size

⫺1.190 (⫺13.7) 0.0636b (11.7) ⫺0.0848b (⫺10.5) 0.0011 (0.058) 0.0156 (1.10) 0.1621b (5.15) 0.0729b (3.68) ⫺0.0465b (⫺6.50) ⫺0.0130 (⫺1.78) ⫺0.0077 (⫺1.21) 0.2267b (12.1) ⫺0.0232 (⫺1.45) 0.0501b (2.63) 0.0361 (1.52) 66.2 7393 b

Philippines

Thailand

⫺1.070 (⫺11.2) 0.0582b (10.1) ⫺0.0751b (⫺9.03) ⫺0.1167b (⫺2.21) ⫺0.0197 (⫺1.29) 0.2100b (12.1) ⫺0.0139 (⫺0.344) ⫺0.0562b (⫺8.74) ⫺0.0269b (⫺2.92) 0.0096 (1.14) 0.1970b (8.01) 0.1500b (10.0) 0.0254 (1.62) ⫺0.0401 (1.85) 64.4 9271

⫺1.037b (⫺7.60) 0.0685b (8.24) ⫺0.0931b (⫺7.52) ⫺0.0774 (⫺1.78) 0.0659b (3.15) 0.3095b (9.68) 0.0395 (0.835) ⫺0.0496b (⫺4.55) ⫺0.0111 (⫺0.532) 0.0084 (0.435) 0.2120b (6.40) ⫺0.0910b (⫺2.94) ⫺0.0375 (⫺0.975) 0.0577 (1.36) 65.0 2971

b

Notes: a Asymptotic t statistics are in parentheses. b Denotes significance at 5% level. Marginal probabilities are evaluated at the sample means of the variables over the pooled sample from all three countries in order to facilitate comparisons across countries. The standard errors are derived by the Delta method.

consistent with this assumption for all three countries. Both age and the square of age are significant with the correct signs. Education is used as a proxy for human capital. Primary education is the default category and other categories are no schooling, secondary and post-secondary. There is also a dummy variable that indicates illiteracy. While this variable is correlated with non-education, it is included explicitly to reflect the extent to which literacy is a requirement in the labor market. The most striking result is the importance of postsecondary education on labor force participation. It is highly significant and increases the probability of entering the work force by 16, 21, and 31 percentage points for Indonesia, the Philippines, and Thailand, respectively, relative to women with primary education in each country. Women with no education have participation probabilities which are 11 and 8 percentage points lower than those of women with primary education in the Philippines and Thailand

100

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

respectively.6 Also, in the case of Thailand, secondary education increases participation probabilities by 7 percentage points relative to women with only primary education. For all three countries, the results give strong support for Hypothesis 4.1, that higher education levels increase women’s participation probabilities. The second set of variables relate to home production as an alternative to market participation. The primary determinant of home production is the structure of the family, particularly the numbers and ages of children. Care is required for all children and this is primarily the responsibility of the mother in Asian societies. Therefore, home production involving childcare is likely to make labor force participation more costly in terms of the woman’s and the family’s welfare as suggested in Hypothesis 4.2. To model this, the total number of children and the number of children born in the previous five years are included in the model. Each additional child born in the previous five years leads to a drop in the woman’s participation probability by 5 to 6 percentage points and these effects are significant in all three countries. However, the total number of children variable was less of a determinant for labor force participation. This variable was only significant in the Philippines (at the 5 percent level) and in Indonesia at the 10 percent level, implying a drop in the participation probability of 3 and 1 percentage points, respectively, for each additional child. The marginal impact of the total children variable on labor force participation may represent an averaging of the inhibiting force of young children on labor market participation and the stimulating impact of older children who could be used to take care of younger siblings. In addition, the number of live births was included. It was expected that this variable would proxy for time spent out of the labor market and so have a negative effect on wages and hence, labor market participation of women, due to loss of human capital formation or atrophy. This variable was not statistically significant for any of the countries. In Asian labor markets, these human capital effects appear to be small because the opportunities for skilled work are limited. A final variable identifies women who are widowed, separated, or divorced. The coefficient on this variable is positive and significant for all three countries. This reflects the fact that women, having been married and now living without a spouse, have significantly higher labor market participation rates. The third set of variables includes a dummy variable to account for differences in participation decisions, ceteris paribus, across the two survey dates in each country.7 The coefficient is not significant in the Indonesian case. In the Philippines case, it is positive and significant implying a 15 percentage points increase in participation probability in the 1993 DHS data compared with the 1978 WFS data. This is consistent with the large increase in participation rates of women between the two surveys found in the sample means. In the case of Thailand, the coefficient is negative and significant, implying a participation probability that is 9 percentage points lower, ceteris paribus, in the 1987 DHS data compared with the 1975 WFS data. Based on the evidence from Fig. 1 of rapid income growth in Thailand, we expected a positive and significant coefficient on this variable. A possible explanation for this result is that there was a general income effect that tended to reduce labor force participation. Nevertheless, these results show that labor force behavior among women varies significantly despite apparent similarities in economic performance and living standards. Finally, two variables are included in the specification to allow for differences in participation decisions between migrants and non-migrants. A variable to reflect whether the

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

101

respondent is a migrant or a non-migrant is included along with an interaction of the variable with the DHS dummy variable. In Indonesia, migrants have a significantly higher labor force participation probability than non-migrants, which is consistent with Hypothesis 4.3. The coefficient on migrant indicates that migrants in the first survey had participation probabilities that were 5 percentage points higher than non-migrants, ceteris paribus. The coefficient on the interaction variable is positive but not significant in the Indonesian case, indicating that the difference in participation probabilities between migrants and non-migrants did not fall between the two surveys by 5 percentage points. The results for the Philippines and Thailand indicate that migrants and non-migrants do not differ in their participation probabilities, ceteris paribus, which is evidence against Hypothesis 4.3.8 Also, the empirical evidence does not support Hypothesis 4.4 since the difference between the participation probability of a migrant woman and a non-migrant woman with similar characteristics are not found to vary between the two survey periods. The importance of using microdata in analyses of this type is immediately apparent. The significant variation in labor market participation decisions of women with different demographic and economic characteristics indicates the importance of using large household-level data sets in analyzing the determinants of women’s labor market behavior in Southeast Asia. The complex relationships found in this analysis would not be possible using more aggregate data sources.

8. The impact of years since migration In this section of the paper, we explore the relationship between years since migration (YSM) from the rural area to the urban area and the woman’s labor market participation decision. Women who have recently migrated to an urban area from a rural area may need to find work quickly, so as to support their families while their husband or other family members search for work. With greater years since migration, migrant women may be less likely to participate in the labor market as their family members may find suitable work, freeing up the wife’s time for home production. Acting in the opposite direction is the tendency for women to be more likely to participate in the labor market as their wages grow in response to their increased human capital due to the acquisition of skills valued in the local labor market, and through their improved knowledge of the local labor market. The net effect of these two relationships between the probability of participating and the woman’s years since migration is an empirical matter. Duleep and Sanders (1993) and Worswick (1996) find evidence consistent with this pattern of behavior for immigrant women in the U.S. and Canada, respectively. The WFS data do not contain information on duration of residence in the urban areas; however, the DHS data do contain information on duration. Unfortunately, the duration of residence information is not ideal in that the question relates to the number of years of residence in the woman’s current locality. A second question asks whether the previous place of residence was in a rural or an urban area. In the former case, we interpret the duration of residence information as giving the number of years since migration from the woman’s (rural) childhood place of residence to the urban area. In the latter case, the years of residence

102

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

information relates to the date of an inter-urban migration, which must have occurred after the woman migrated from her (rural) childhood place of residence. These two groups of migrant women are identified separately in the estimation. The model defined in Eq. (1) is estimated over the DHS data using the following specification: I i共t兲 ⫽ X i共t兲 ␣ ⫹ ␣ 1Migrant i ⫹ ␣ 2YSM i ⫹ ␣ 3共YSM i兲 2 ⫹ ␣ 4NOYSM i ⫹ u i共t兲

(2)

where Ii(t) and Xi(t) are defined as before, ␣ is a parameter vector, ␣1, ␣2, ␣3, and ␣4 are parameters and ui(t) is a mean zero, normally distributed error term. YSMi is the number of years since migration for migrant women who reported that their previous place of residence was in a rural area. The variable NOYSMi identifies migrant women who reported that their previous place of residence was another urban area. For these women, we do not know how many years it has been since they moved from their childhood rural area; therefore, this dummy variable controls for these women in the analysis. The years since migration variables are set to zero for non-migrant women and for migrant women whose previous place of residence was an urban area (NOYSMi ⫽ 1). The results of estimation over the ever-married women in each DHS sample are presented in Table 4. The coefficients on the controls for the woman’s demographic characteristics are in general very similar to those found in the estimation over the pooled sample presented in Table 3. For the Indonesian case, the migrant coefficient implies a 7 percentage points higher probability of working for migrant women than non-migrant women, and this effect is significant at the 10 percent level. For Thailand and the Philippines, the coefficient on the migrant dummy variable is not significant as was found in Table 3. The years-sincemigration variables are not significant for any of the three countries, indicating that the participation probability of migrant women does not differ between women who have recently arrived and those who have resided for a number of years in the urban area.9 9. Migration decisions of never-married women in the Philippines As a final investigation of differences in participation behavior between migrant and non-migrant women, we exploit the fact that never-married women were surveyed as part of the DHS of the Philippines. Estimation of a modified version of Eq. (2) is carried out over this sample and the marginal probabilities are presented in Table 5. The marginal probabilities for both age and the square of age have increased dramatically in magnitude compared with the previous results. This likely reflects the fact that these women are on average 21 years old, compared with the ever-married samples where the mean values for age are in the 30 to 35 years range. Illiterate women have 27 percentage points lower probabilities of working than otherwise similar literate women. This is in contrast with the previous results. It may be that illiterate single women are contributing to home production since the market value of their time is low. The education controls appear on their own and as interactions with the migrant variable.10 Women with secondary or post-secondary education have 20 to 22 percentage points lower probabilities of labor market participation than do women with primary education.

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

103

Table 4 Marginal probabilities from probit estimation with controls for years of residencea

Intercept Age Age2/100 No education Secondary education Post-secondary education Illiterate Number of children born in last 5 years Number of children present Number of births Widowed, separated, or divorced Migrant Year-since-migration (Year-since-migration)2/100 Migrants without YSM data Percent correct predictions Sample size

Indonesia

Philippines

Thailand

⫺1.160 (⫺9.31) 0.0613b (8.07) ⫺0.0830b (⫺7.51) ⫺0.0923b (⫺2.56) 0.0198 (1.15) 0.1928b (5.06) 0.1166b (3.54) ⫺0.0467b (⫺5.17) 0.0048 (0.455) ⫺0.0224b (⫺2.42) 0.1617b (6.30) 0.0659 (1.75) ⫺0.0017 (⫺0.280) 0.0148 (0.724) 0.0433 (1.13) 65.0 4470

⫺0.4351 (⫺3.35) 0.0329b (4.19) ⫺0.0415b (⫺3.65) ⫺0.0665 (⫺0.847) ⫺0.0023 (⫺0.123) 0.1559b (7.35) ⫺0.0650 (⫺1.02) ⫺0.0631b (⫺7.57) ⫺0.0245b (⫺1.73) 0.0018 (0.142) 0.2079b (6.31) ⫺0.00001 (0.0003) ⫺0.0007 (⫺0.111) 0.0027 (0.138) ⫺0.0274 (⫺0.739) 63.4 4668

⫺1.096b (⫺6.52) 0.0671b (6.43) ⫺0.0910b (⫺5.86) ⫺0.1022 (⫺1.84) 0.0657b (2.62) 0.3373b (9.11) 0.0722 (1.22) ⫺0.0584b (⫺4.47) 0.0282 (0.939) ⫺0.0286 (⫺1.01) 0.2476b (6.10) 0.0010 (0.028) 0.0056 (0.985) ⫺0.0158 (⫺0.833) ⫺0.0211 (⫺0.449) 65.1 2375

b

b

Notes: a The analysis is carried out using the DHS data; asymptotic t statistics are in parentheses. The marginal probabilities are evaluated at the sample means of the variables over the pooled sample from all three countries to facilitate comparisons across countries. The standard errors are derived by the Delta method. b Denotes significance at 5% level.

This is evidence against Hypothesis 4.1. This result may follow from the fact that, for never-married women, higher education levels act as a proxy for high family income or wealth. If this is the case, then never-married women in these households may not need to work to support their families while the women with lower levels of education may have a greater need to participate in the labor market. The results for never-married women in the Philippines sharply contrast with those found for ever-married women in each of the three countries studied. This high level of heterogeneity in the determinants of women’s labor market participation decisions is further evidence in support of the use of microdata sets, with detailed controls for demographic and economic characteristics, in analyzing labor market outcomes of women in Southeast Asia. In the second column of Table 5, the marginal probabilities for the migrant variables are

104

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

Table 5 Marginal probabilities from probit estimation of participation model: Never-married women in the Philippinesa Variable name Intercept Age Age2/100 Illiterate No education Secondary education Post-secondary education Percent correct predictions Sample size

Variable name ⫺2.392 (⫺19.4) 0.1770b (19.2) ⫺0.2620b (⫺16.2) ⫺0.2338b (⫺1.99) ⫺0.0555 (⫺0.357) ⫺0.1030b (⫺2.17) ⫺0.2159b (⫺4.55) 72.9 3172 b

Migrant Year-since-migration (Year-since-migration)2/100 Migrants without YSM data (No education) migrantb (Secondary) migrantb (Post-secondary) migrantb

0.5100b (7.92) ⫺0.0800b (⫺5.11) 0.3085b (3.18) ⫺0.4026b (⫺9.60) ⫺0.1599 (⫺0.709) ⫺0.1252b (⫺1.99) ⫺0.1800b (⫺2.74)

Notes: a The analysis is carried out using the DHS data. Asymptotic t statistics are in parentheses. The marginal probabilities are evaluated at the sample means of the variables over the sample means of the variables for the never-married sample from the Philippines. The standard errors are derived by the Delta method. b Denotes significance at 5% level.

presented. The interactions of the education variables with the migrant dummy variable indicate that the negative effect of higher education on participation is even larger for migrant single women than for non-migrant single women. This may well reflect the fact that migrant women with low levels of education are migrating so as to find work, while migrant women with higher levels of education may have migrated with their families, who in turn support their acquisition of higher levels of education.11 The controls for years-since-migration indicate that recent, never-married, migrant women are much more likely to participate in the labor market than are non-migrant women, with the difference being approximately 51 percentage points for migrant women who have just arrived in the urban area. The effect of years-since-migration is to reduce this difference but the effect of YSM squared is to diminish this effect. In Fig. 2, the Probit estimates are used to derive the predicted difference in the probability of participating in the labor market between a migrant woman and a non-migrant woman, and these are plotted against YSM. Consistent with the marginal probabilities,12 the figure indicates that recent arrivals are much more likely to work than are non-migrants and this difference is generally smaller for migrants with longer duration of residence in the urban area. This pattern is consistent with a number of hypotheses, which unfortunately cannot be disentangled using the available data. One possibility is that never-married migrant women in the Philippines move to an urban area and must find work immediately upon arrival so as to support themselves or their families still residing in the rural area. With more years of

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

105

Fig. 2. Predicted difference in probabilities of participation between migrants and non-migrants by years since migration.

residence, they may be joined by other family members whose added incomes allow a larger proportion of them to withdraw from the labor market. Another possibility is that the estimated YSM pattern reflects different types of migrant women. The first group may be women who leave their families in the rural area so as to find jobs in the city and send money home to the rural areas. If these women stay only for short periods of time in the urban area before returning to the countryside, then this would be consistent with the very high participation probabilities of the recent arrivals. The second group of migrant never-married women may be those who move with other family members (possibly as children) and then reside with other family members who support them financially. These women would be more likely to remain in the urban area for long periods of time than the first group of womens since their families would now live in the urban area. We would therefore expect the first group of women to be observed in the data with low levels of YSM and the second group of women to appear at both low and higher levels of YSM. In this case, Fig. 2 should not be interpreted as the participation experience of a typical never-married migrant woman in the Philippines. Instead, it is a cross-sectional picture of the participation probabilities of different migrant women with different levels of YSM.13

10. Suggestions for design of future surveys In order to evaluate these different explanations, one would ideally use panel data so that one could follow the migrant women through time and take into account return migration. However, panel data sets are very expensive to conduct and suffer from attrition bias. In the absence of suitable panel data, multiple cross-sections from household surveys (like the ones used in this study) may be of use in order to track cohorts of individuals through time. This study has highlighted a number of deficiencies in the WFS and DHS data sets and it may be useful to list these here. Information on women’s income and family income would

106

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

be very useful in that it would allow us to control for income effects on labor supply decisions. Also, the women’s earnings information would allow us to construct an offered wage rate that could be incorporated into the labor supply model. Other useful information would be: 1) whether the woman currently attends school; 2) whether the woman is an unpaid worker in a family business; 3) information on the husband’s characteristics, wages, education, and occupation; and 4) a complete history of the woman’s migration history giving the relevant dates and locations.

11. Conclusions Estimation of a binary choice model of the participation decisions of urban women has been carried out using microdata from three Southeast Asian countries spanning the period 1975 through 1993. The evidence indicates that ever-married migrant women are generally at least as likely to work in the urban labor market as are non-migrant women, ceteris paribus. In Indonesia and the Philippines, these migrant women were found to have 19 to 30 percentage points higher probabilities of working, while no significant differences in participation probabilities were found in Thailand between migrants and non-migrants. The evidence is consistent with ever-married migrant women in lower income countries migrating primarily to find work so as to support themselves and their families. On the other hand, ever-married migrant women in higher income countries appear to spend more time in home production, so as to free up time for other family members to work in the labor market. A detailed investigation has been carried out of the dynamic relationship between the labor market participation decisions of migrant women and their duration of residence in the urban area. Duration of residence was not found to significantly affect labor market participation of ever-married women. However, for the case of the Philippines, never-married women who had recently arrived in the urban area are much more likely to work than are either non-migrant women or never-married women with longer duration of residence. An investigation of the determinants of this relationship is left for future work. The evidence indicates that migrant women are at least as active in the urban labor market as are non-migrant women, ceteris paribus. Therefore, migrant women provide an important component of the expansion of the supply of labor services in the growing Southeast Asian labor market. The heterogeneity of the participation probabilities according to age, education, presence of children, and duration of residence indicate the need for large household surveys in undertaking analyses of the labor market activities of women in Southeast Asia.

Acknowledgment We thank Lisa Cameron, Patrick Kenny, Medhi Krongkaew, Thammarak Moenjak, Chalongphob Sussangkarn, and an anonymous referee for helpful suggestions as well as seminar participants at the Thailand Development Research Institute.

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

107

Notes 1. The WFS of Indonesia only surveyed women residing in Bali and Java. The region surveyed in the 1987 DHS for Indonesia is larger; however, women residing in regions of Indonesia other than Bali and Java were excluded to allow for comparability with the WFS data. 2. It should be stressed that since the participation dummy variable is generated from the question: “Are you currently working?” in both the WFS and the DHS data sets, we are actually modeling employment probabilities rather than participation probabilities. Women who were unemployed at the time of the survey will fall into the non-participation group in the estimation; therefore, Ii(t) ⬍ 0. 3. Recall that the sample is restricted to women who currently reside in urban areas. Therefore, this variable identifies women who migrated from rural areas at some point in their lives. 4. The WFS data were collected in 1975, 1976, and 1978, while the DHS data were collected in 1987, 1987, and 1993 in Thailand, Indonesia and the Philippines, respectively. 5. These are the partial derivatives of the cumulative distribution function (derived using the estimated parameters) with respect to each of the explanatory variables. 6. In the case of Thailand, the effect is only significant at the 10 percent level. 7. The survey dummy variable equals one of the woman was surveyed as part of the DHS, and equals zero if the woman was surveyed as part of the (earlier) WFS. 8. A number of other specifications were investigated. The effect of education level on participation probabilities was allowed to differ by migrant status by introducing interactions of the Migrant variable with the education controls. For each country, these variables were insignificant. Interactions of the Migrant variable with the presence of children variables were also included but found to be insignificant. Finally, controls for birth year were included to investigate whether different age cohorts have difference participation probabilities. There was some evidence that women born in the earliest birth period (the 1930s) were significantly less likely to work than women from later birth periods, ceteris paribus; however, the remaining results were not sensitive to the inclusion of this control. Therefore, we have opted for the more parsimonious specification, (1). 9. Since the estimation is carried out over a single cross-section, it is not possible to distinguish between changes in the participation probability with duration of residence from unobserved differences in participation tendencies across different migration cohorts (defined in terms of arrival periods). For a discussion of this issue see Borjas (1985). 10. We investigated a number of different interactions of the migrant dummy variable with the other controls. The interactions with age and illiteracy were insignificant. 11. The data sets do not contain information on whether the respondent currently attends school. In order to investigate whether the results reflect current decisions over whether to attend school, we re-estimated after excluding women from our working samples who were age 15–20. The results were qualitatively unchanged.

108

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

12. The estimates are not exactly comparable; however, since these are discrete differences in probabilities while the marginal probabilities are the partial derivatives. 13. Information on re-migration to rural areas is not available in the data sets. We have also been unable to find research on the magnitude of re-migration of migrant women in Southeast Asia. Therefore, an exploration of this issue is left for future work.

References Asian Development Bank (1987). Asian Development Outlook. Oxford University Press: Hong Kong. Borjas, G. J. (1985). Assimilation, change in cohort quality, and the earnings of immigrants. Journal of Labor Economics 3(4), 463– 489. Cameron, L., Dowling, J. M., & Worswick, C. (1997). Education and Labor Market Participation Decisions of Women in Asia: Theory and Evidence, research paper number 561. Melbourne, Australia: Economics Department, The University of Melbourne. Cohen, B. & House, W. J. (1993). Women’s labour market status in developing countries: how well do they fare in Khartoum, Sudan? Journal of Development Studies 29(3), 461– 483. Corden, W. & Findlay, R. (1975). Urban unemployment, intercessor capital mobility and development policy. Economica 42(165), 59 –78. Duleep, H. & Sanders, S. (1993). “The Decision to Work by Married Immigrant Women.” Industrial and Labor Relations Review 46(4), 64 – 80. Durand, J. D. (1975). The Labor Force in Economic Development. Princeton, NJ: Princeton University Press. Eaton, J. P. (1992). Rural-urban migration and underemployment among females in the Brazilian Northeast. J Econ Issues 26(2), 385–395. Fields, G. & Schultz, T. P. (1982). Income generating functions in a low income country: Colombia. Rev Income Wealth 28(1), 71– 87. Ghatak, S., Levine, P., & Price, S. W. (1996). Migration theories and evidence: an assessment. Journal of Economic Surveys 10(2), 159 –198. Harris, R. & Todara, M. P. (1970). Migration, employment and development: a two sector analysis. American Economic Review 60(1), 126 –142. Hoddinott, J. (1994). A model of migration and remittances applied to western Kenya. Oxford Economic Papers 46(3), 459 – 476. King, E. M. & Hill, M. A. (Eds.). (1993). Women’s Education in Developing Countries: Barriers, Benefits and Policies. Baltimore: Johns Hopkins Press. Knowles, J. C. & Anker, R. (1981) An analysis of transfers in a developing country: the case of Kenya. J Develop Econ 8(2), 205–226. Lucas, R. E. B. (1991). Migration from Botswana. Economic Journal 95(378), 358 –382. Lucas, R. E. B. & Stark, O. (1985). Motivations to remit: evidence from Botswana. J Political Econ 93(5), 901–918. Maddison, A. (1995). Monitoring the World Economy 1820 –1992. OECD Development Centre Studies. Organization for Economic Cooperation and Development. Nelson, J. M. (1976). Sojourners versus new urbanites: causes and consequences of temporary versus permanent migration in developing countries. Econ Develop Cultural Change 24(4), 721–757. Psacharopoulos, G. (1985). Returns to education: a further international update and implications. J Hum Resources 20(4), 583– 604. Robinson, C. & Tomes, N. (1983). Self-selection and interprovincial migration in Canada. Can J Econ 15(3), 474 –502. Rosenzweig, M. R. (1989). Labor markets in low-income countries. In H. Chenery & T. N. Srinivasan (Eds.), Handbook of Development Economics (Vol. 1). Amsterdam: North-Holland.

J.M. Dowling, C. Worswick / Journal of Asian Economics 10 (1999) 91–109

109

Schultz, T. P. (1989). Education investments and returns. In H. Chenery & T. N. Srinivasan (Eds.), Handbook of Development Economics (Vol. 1). Amsterdam: North-Holland. Schultz, T. P. (1983). Migrant Behavior and the Effect of Regional Prices: Aspects of Migration Selection in Colombia, discussion paper no. 443. Yale Economic Growth Center. Yale University. Stark, O. (1995). Altruism and Beyond: An Economic Analysis of Transfers and Exchanges Within Families and Groups. Cambridge: Cambridge University Press. Stark, O. & Bloom, D. (1985). The new economics of labor migration. Am Econ Rev 75(2), 173–178. Standing, G. (1978). Labor Force Participation and Development. Geneva: International Labour Organization. Subbarao, K. & Raney, L. (1995). Social gains from female education: A cross-national study. Econ Develop Cultural Change 44(1), 105–128. Todaro, M. P. (1994). Economic Development. New York: Longman. Todaro, M. P. (1969). A model of labor migration and urban employment in less developed countries. Am Econ Rev 59(1), 138 –148. Tunali, I. (1996). Migration and remigration of male household heads in Turkey 1963–73. Econ Develop Cultural Change 45(1), 31–57. Vijverberg, W. P. M. & Zeager, L. A. (1994). Comparing earning profiles in urban areas of an LDC—rural to urban migrants versus native workers. J Develop Econ 45(2), 177–199. Williamson, J. G. (1989). Migration and unemployment. In H. Chenery & T. N. Srinivasan (Eds.), Handbook of Development Economics (Vol. 1). Amsterdam: North-Holland. Worswick, C. (1996). Immigrant families in the Canadian labour market. Can Public Policy 22(4), 378 –396.