Economic Development and the Role of Women in Rural China

Transcription

1 Economic Development and the Role of Women in Rural China Dwayne Benjamin* Loren Brandt* Daniel Lee** Social Science Division Hong Kong University of Science & Technology Clear Water Bay Kowloon Hong Kong Tel. (852) Fax (852) (Attn. Daniel Lee) 26 September 2002 * Department of Economics, University of Toronto ** Hong Kong University of Science and Technology

2 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 1 Abstract Economic reform has brought many changes to China s rural economy in the past two decades, as reflected in a combination of rapid economic growth, rural industrialization, structural change, and sharp reductions in fertility. This paper explores the impact of economic growth and market development on the changing economic role of women in rural China. In China, recent studies examining female relative wages in industry and urban areas (Meng 1998; Liu et al and Meng et al. 1995), and female labor force participation in the rural non-farm sector (Entwisle et al. 1995; Mathews and Nee 2000) find that while market development reduces the male -female wage differential, its effect on female labor participation is less clear. As a result, the link between market development and the contribution of women to household income is unclear. The major challenge in the empirical work is the measurement of individual contributions to farm production, given that only household farm income is observed. This problem I resolve by using individual labor supply in agricultural production to estimate the relative contributions of women and men. A panel household-level data set collected by CHNS in 1991, 1993, and 1997 that contains more than 1400 households is used to conduct the empirical analysis. Both fixed effect and random effect models that eliminate the effect of unobserved household characteristics are estimated. The most remarkable observation is that both women and men are moving out of agriculture, confirming that women are not left on the farm in the process of economic development. However, the effect of economic development, as proxied by village per- capita income, is more significant for men. Consequently, the gap between male and female contributions to household income is widening in high economic growth areas.

3 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 2 1. Introduction Economic reform, accompanied by rural industrialization, has generated millions of lucrative non-agricultural, off-farm working opportunities for people in rural China. An important question surrounding the impact of these changes is the extent to which the role of women has evolved. Prior to the reforms, women worked mainly in farming, raised livestock, or did housework. Now, there are many stories about single women working in factories, remitting their income home to improve the living standard of their families. Possibly, the status of women will rise along with their economic contributions to household income. On the other hand, there is some concern about whether women are being left on the farm while their husbands and sons work off the farm, particularly in low-income provinces where many have out-migrated to search for better off-farm jobs. Thus, while economic development may be beneficial to the rural households as a whole, the women may not be fully share in the gains. The purpose of this paper is to document changes, and evaluate the impact of economic development on women s labor supply and earnings. In this paper, we use a household-level panel data set collected by the China Health and Nutrition Survey in the 1990s to explore the impact of economic development on the pattern of labor allocation and the relative income contribution of women to the households. The major challenge in the empirical work is the measurement of individual earnings given that most earnings are reported at the household level. We address this problem by focusing on hours of work, which are observed at the individual level. We use household income by various sources to impute earnings to men and women, assuming that wages are the same for men and women. As we explain more fully later, most of the variation in economic contribution comes from hours worked (as opposed to wages). There are several hypotheses that we would like to address. What is the effect of economic development on time allocation of women and men? Are women more likely to be relegated to low return activities while men are more likely to engage in more

4 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 3 lucrative jobs? Is there any interaction between the family structure and the time allocation of women? What is the role of human capital in the allocation of labor to different economic activities? In which activity do women make their highest contribution to household income? By examining the results of an exhaustive set of regressions, we find that women have indeed benefited from the process of economic development in rural China. However, men have benefited more. This paper proceeds as follows. Section two reviews the previous literature on the effect of economic development on women. Section three discusses the problem of measuring individual income in farming. We will discuss the strategy of estimating individual income and the potential problems associated with this strategy. Section four provides a description of the China Health and Nutrition Survey. Section five presents the descriptive statistics of labor supply and earnings for the women and men in our sample. This provides a broad picture of the changes in labor allocation in different economic activities across the sample period. Section six describes some of the key village-level characteristics by province, by which we hope to capture economic development. This section also provides a better sense of the variations in economic growth and structural transformation across the CHNS provinces. Section seven presents the household-level regression analysis. Here, we explain the determinants of the total labor supply, total income, and their components: farming, raising livestock, family businesses, and off-farm wage employment. For each employment and income variable, we examine determinants of the levels for women, as well as the differences between women and men: did the rising tide of economic development raise all boats equally, at least in the male-female dimension? Section eight offers some conclusions.

5 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 4 2 Previous Work The easiest way to link economic development to men s and women s economic contributions is through the associated income and substitution effects that may accompany increasing male and female wages. 1 If the female wage rate rises faster than the male wage, this will induce higher labor participation, and a greater economic contribution of women. Goldin and Sokoloff (1982) explore the linkages between early industrialization of the American Northeast and the economic role of women, by highlighting regional differences in the productivity of women in agriculture. Goldin (1990) studies the wage gap between women and men, and provides supportive evidence that the annual earnings of women increased from 46 percent of men s income in 1890 to about 60 percent of men s income in Craig and Field-Hendrey (1993) also confirm this Goldin- Sokoloff relative productivity hypothesis that the relative wage of women increased in agriculture as well as manufacturing during that period of time. Studies that focus on the interaction effect of education and economic development on the working opportunities of women predict a U-shape relationship between economic development and women s labor participation. Women s labor supply will decline at the beginning of economic development (because of an income effect) and will increase later. Goldin (1995) argues that at the beginning of economic development, which in most cases is characterized by rural-urban migration, most of the jobs available for women are manual in nature, and customs and norms may prevent women from taking these jobs (their husbands will prefer them to stay at home). This explains the decline in women s labor participation in the initial stage of economic development. Later, when women have received more education and more white-collar jobs are available, their labor participation will increase. This hypothesis has been confirmed by Goldin (1995) and Mammen and Paxson (2000). However, Horton (1996) examines the female participation rate of seven Asian countries and finds that women s 1 Increase in wages will have a positive substitution effect, but a negative income effect on labor participation. However, only those who worked in the labor market will have the negative income effect, and consequently, the substitution effect is likely to dominate the income effect.

6 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 5 participation rises through the two decades that these countries have experienced rapid growth. Horton (1996) explains that the increase in women s labor participation has offset the decrease due to rural-urban migrations in those countries. Hence we may not see a decline in women s labor participation in export-oriented Asian countries, including China. The first part of the Ushape may not be relevant. In rural economies in which markets may not be highly developed, and women are largely confined to work on the family farm, the farm household models of Singh, Squire and Strauss (1986) predict that there are patterns of intra-household substitution of labor, that depend crucially on the structure of families, skills possessed by, and opportunities available, for other family members. If markets are poorly developed, we cannot rely on prices alone to identify the substitution and income effects. In rural China there is a concern that men may be in a better position to gain off-farm working opportunities, and that women are left on the farm while their husbands and sons work off the farm (Boserup 1970; Judd 1995; Parish et al. 1995; Croll 1997; Jacka 1997 and Rozelle et al. 2000). Hare (1999) confirms that in rural China, the probability of obtaining wage employment depends on household size and composition. She finds that only single women are observed in wage-employment. The probability of working in wage employment is negatively correlated with per worker endowments of land, suggesting that women are in high demand in farming. Barrett et al. (1991) also find the women s labor participation increases with the share of agricultural employment. Within the non-farm sector, women do not only have a lower chance to work in wage employment, Entwisle et al. (1995) find that households with more men are more likely to have family business. And conditional on having family business, men are more likely to work in it. However, the hypothesis that women s contributions to household income rise with the level of development receives support from both historical and contemporary studies (Johnson, Parish and Lin (1987) and Benjamin and Brandt (1995) for the 1930s; and Mathews and Nee (2000) for the 1980s). In China, recent studies examining female relative wages in industry and

7 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 6 urban areas (Meng 1998; Liu et al and Meng et al. 1995), and female labor force participation in the rural non-farm sector (Entwisle et al. 1995; Mathews and Nee 2000) find that while market development reduces the male-female wage differential, its effect on female labor participation is less clear. In summary, in the context of China, there is no clear evidence for the relationship between market development and the contribution of women to household income. 3. Measuring the Contribution of Women The main challenge for determining the relative female -male contribution to household incomes arises from the difficulty of identifying individual contributions in certain economic activities. Except for wage earnings, income is only recorded at the household level, as farming, home gardening, fishing, raising livestock, and family businesses are properly regarded as householdlevel activities. However, detailed information on individual labor supply in each economic activity is collected, and this allows the construction of male and female labor supply by activity. How can we use this information (on household income and individual hours) to estimate individual income? Let Y H be total household income, which is the sum of males (indexed by i) contributions to k activities, i k w h m ik ik, plus females (indexed by j) contributions, j k w h f jk jk. Household income is then given by: (1) Y = w h + w h m f H ik ik jk jk i k j k In Equation 1, m wik and h ik are respectively the wage rate and the number of hours of labor supply of male i in the k th economic activity. Similarly f w jk and h jk are the wage rate and hours of labor supply of female member j in the k th economic activity. If we have measures of the

8 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 7 wage rate of individuals in each economic activity, the determination of the relative female -male contributions is a simple accounting exercise. However, individual-specific wages (earnings) are only observed for wage employment. Thus, the problem of determining the relative female - male contributions turns out to be a problem of assigning wages for males and females in various activities within the household. One way to solve this problem would be to estimate the average returns in each economic activity separately for male and female by OLS: Y = α h + β h + υ (2) Hk k ik k jk Hk i j Analogous to a production function, male labor in the k th economic activity, and α k is the estimate of the mean wage rate (returns) of β k is the estimate of the mean wage rate (returns) of female labor in the k th economic activity. However, there are a number of serious problems with this approach. First, this method constrains all households to have the same rate of return in each economic activity. Second, there is almost certainly serious measurement error in reported hours worked, so estimates of α k and β k will be biased downwards, in unpredictable relative directions for men and women. Instead of estimating a single return to all households, we assume that the rate of return in each activity is the same for all members of the households. Let the total hours of family labor supply in the k th economic activity. H k be (3) H = h + h k ik jk i j In a rural economy, the amount of labor supplied by family members in farming and raising livestock may be a reasonable proxy for their contribution to that activity. Moreover, rational households will allocate their members to family businesses in order to maximize profits. We can thus expect that labor supply in family businesses will also reflect the relative contributions of the individuals. We thus create household-specific average wages by activity,

9 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 8 w ˆ k, by dividing the net return in each activity by the total hours of household labor supply (see Equation 4). Then we multiply the wage rate by hours of labor supply to estimate an individual s contribution to household income (Equations 5 and 6): m f wˆ = wˆ = Y H (4) k k Hk k y = w h (5) m m k k ik i y = w h (6) f f k k jk j While this solves some of the problems associated with the approach of Equation (1), it comes with problems of its own. First, when using the number of hours of labor supply to obtain the wage rate, we assume equal productivities for all members of the household. This imposes the obviously strong assumption that men and women are equally productive. If men are more productive than women, we will overestimate the wage rate of women, and consequently, their contributions to household income. Conversely, if women are more productive, we will underestimate their contribution to household income. In the context of rural China, men are possibly more productive on the farm, and we will overestimate the contributions of women. 2 We can come up with similar stories for any activity. Clearly, we will not be able to identify the changes in this fundamentally unobservable quantity. However, our main objective is to weight a woman s hour spent in agriculture versus an hour spent off the farm. Average returns in these activities will still serve as reasonable weights. Furthermore, in the regression analysis relating economic development to economic contribution, some of the effect of this assumption will be absorbed by other variables, especially when we look at differences between men and women. 3 Similarly, any trends in the 2 For example, in the collective era, women were rewarded 80 percent of the work points of men for the same hours of work. 3 Another potential problem is that if the difference in female-male productivity is correlated with the level of income, we will have a biased result.

10 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 9 male-female difference in productivity will be captured by the time trend in the regressions. Another difficulty apparent with the approach above is that we assume equal productivity of household members even of the same sex (for example, of different ages). In the regression analysis, we include a set of household demographic and other control variables that should reduce this problem. The second problem is that when labor supply is used to measure the relative femalemale contributions, we assume that other farm inputs, such as farmland and farm capital, have the same cross partial effect on men and women s return. w m / k = w f / k If women use capital more efficiently, we will underestimate their relative contribution. Moreover, if economic development is positively correlated with the amount of farm capital used, then by assuming equal male-female productivity, we will underestimate the effect of development on women s contribution to household income. While we cannot exclude this possibility, there seems to be no a priori reason to believe that either women or men are more efficient with their use of capital. Third, in using the hours of labor supply to measure the relative contributions of men and women, we also neglect the productivity difference due to variations in human capital. While men always have the advantage in performing physical labor, highly educated women could also be productive. In the regression analysis, we will control for differences in male and female human capital. In the end, our most direct and non-contentious estimates are based on labor supply, which is observable. Nevertheless, estimates of income are very helpful in allowing us to weight hours in various activities. Given that income and hours will move closely together, it is easy to qualify the conclusions regarding income, in case economic development affects the relative productivity of men and women (the biases are straightforward to understand, even if

11 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 10 unobservable). Note, however, that pricing hours of work is important in evaluating questions like whether women are consigned to low-paying or lousy jobs. 4 The China Health and Nutrition Surveys We use data from the China Health and Nutrition Survey (CHNS) conducted by the University of North Carolina and the Institute of Nutrition and Food Hygiene of the Chinese Academy of Preventive Medicine during three successive periods 1989, 1991, and Eight provinces that representing the geographical diversity and income variations of China were carefully chosen, including Liaoning, a prosperous, heavyindustry-based northern province, Henan, an agriculture-based middle-income eastern province, and Guizhou, a poor southwestern province (see Figure 1). We will discuss some problems with the data that eliminate some observations for empirical analysis. First, the 1989 survey fails to ask a question about the number of weeks worked in wage employment. There is thus no way to estimate the individual labor supply in wage employment, and we drop the entire year of data in Second, the 1997 CHNS replaced the survey site of Liaoning with Heilongjiang. Since we restrict our analysis to panel households (observed all three years) we have to exclude the sample from Lia oning and Heilongjiang in our analysis, which means a loss of one-eighth of the sample. Third, we find that some survey communities were classified as rural in 1993, but were reclassified as urban in 1997.This may be the result of increases in the number of residents, changes in the sources of income, or other reasons. 5 Instead of following the CHNS definition, we define rural areas as places in which no 4 A fourth wave of surveys was conducted in 1997, but the data have not yet been made available. 5 Urban population is defined as those who live in an urban area or tow nship. In 1984, township was defined according to the population size (over 20000), and according to the share of income from nonfarm

12 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 11 less than 25 percent of the households have members working in agriculture. Three villages are reclassified as urban based on this criterion. Lastly, as with most panel surveys, there is a high attrition rate. We document the loss of samples for different reasons in Appendix I. After these changes, we are left with 1420 households for which we have data for all three years. 5 Labor Supply and Contribution of Income We begin by describing the most readily observable indicators of economic activity. Table 1 provides simple descriptive statistics for the labor market information and imputed income for different economic activities for women and men in 1991, 1993, and In the calculation, we include only those people aged 15 years or older who are not in school. All the mean hours of labor supply and income are calculated at the household level. Note that as the average size of households fell considerably (from 5.0 in 1991 to 4.2 in 1997), one must be cautious in interpreting household levels across years (the variables are not per capita adjusted). In the regressions that follow, we control for the effect of demographic adjustment across time by including a set of variables that measures the number of household members in different gender and age groups. This provides a more flexible adjustment than simply looking at per capita income. Also note that all income measures are expressed in real 1990 terms. The top panel of Table 1 summarizes labor market participation and hours. Note that all figures refer to household (not individual) levels and rates. However, for ease of exposition, we will occasionally drop the household -level qualifier. The household labor participation rate is high throughout the three years. In 1991, almost all households had at least one woman and one sources (over 10 percent) (Statistical Yearbook of China 1998). Many areas defined as rural in the past changed to townships when their share of nonfarm income surged during the economic reform.

13 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 12 man engaged in some economic activity (96 percent for women and 97 percent for men). We observe that even after two decades of economic reform, agriculture is still the main place in which rural households work, as reflected by the high percentage of women (90 percent) and men (85 percent) working on the farm. Also, significantly more women raise livestock than do men (80 percent compared to 26 percent). This suggests a clear pattern of household division of labor, namely, that women are responsible for this sideline production. Men appear to be in a better position than women regarding non-farm jobs. There are about three times more households in which men have non-farm work than households in which women have non-farm work (23 percent versus 8 percent). This supports the findings in Entwisle et al. (1995), that in households with a family business, men are more likely to work in it. Men appear to have similar advantages in wage employment. Whereas 28 percent of households have a man working in wage employment, only 12 percent of households have a woman working in wage employment. Over the 6-year period, we observe a clear trend of structural transformation. Many households moved out of the agricultural sector, as represented by a decline in both women and men s participation in farming and raising livestock. We observe a particularly substantial decline in the number of women in raising livestock (80 percent in 1991, 65 percent in 1997). At the same time, women increased their participation in non-farm work in both family businesses (8 percent in 1991, 12 percent in 1997) and wage employment (12 percent in 1991, 16 percent in 1997). As men s participation in non-farm work is steady over this period of time, women have narrowed the gap with men in terms of participation in non-farm work. The middle panel of Table 1 reveals a striking decline in total hours of labor supply. In 1991 women worked more than 3000 hours, but in 1997, they worked less than 2300 hours, a 25 percent drop in labor supply. Although women increased their labor supply in both family business and wage employment, this increase is far from enough to compensate for a sharp drop in labor supply from farming (2011 hours in 1991, 1194 hours in 1997), and a modest reduction

14 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 13 of labor supply in livestock raising (546 hours in 1991, 424 hours in 1997). A similar, though less dramatic, reduction in men s total hours of labor supply is also observed, the main source of which is also farming. The male labor supply fell less than female because men devoted less time to agriculture. What about the relative female-male household labor supply? The main driving force of the pattern of change in labor supply is from the agricultural sector, particularly farming. 6 In 1991, women worked more than men because they worked more intensively on the farm. In 1997, women reduced their labor supply on the farm, and as both women and men increased their labor supply in non-agriculture to a similar extent, men worked more than women. The combination of the sharp reduction in hours and reduced participation in farming suggests that agriculture is declining in importance, which is what we expect in the process of economic development. The reduction of female -male relative labor supply may suggest, consistent with Goldin (1995), that at the beginning of economic development, jobs suitable for women may be short in supply, in which case women simply reduce their labor supply. Whether or not women have more working opportunities in the process of economic development is the key question that we will address in detail in the regression analysis below. We also find a remarkable difference in the ways women and men allocate their time. Compared to men--who spent more than 30 percent of their time in wage employment-- women spent less than 20 percent of their time in that activity. Even in 1997 women still spent 72 percent of their time farming and raising livestock, whereas men, spent just slightly more than half of their time in the agricultural sector. The lower panel of Table 1 presents the imputed income for women and men from different economic activities (using the approach described in the previous section). The first observation is the striking increase in total income for women (1424 yuan in 1991, 1597 yuan in 6 Notice that family size also dropped from 5 in 1991 to 4.2 in 1997, which partially accounted for some of the reduction in labor supply. However, this may not be significant because most labor supply is from people in their prime age, not from old or young people.

15 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA ) and, especially, men (1786 yuan in 1991, 2493 yuan in 1997) from 1991 to This increase is mostly attributable to gains in the income from wage employment and family businesses. That household income increases as household labor supply declined substantially implies an increase in average hourly returns ( wages ) for both women and men. This increase in wages even happened within the agricultural sector. For instance, during this period women reduced their labor supply to farming by 40 percent while their estimated farm income declined only by 25 percent. Similarly, men reduced their farm labor supply by 34 percent while their income only dropped by 11 percent. This increase in the implicit hourly wage in farming can be partially explained by the increase in the prices of main crop products, and more efficient allocation of farm labor. 7 Except for wage earnings, our estimates of income are largely based on hours worked in a given activity, and changes in income will largely reflect changes in labor supply. Nevertheless, it is apparent that hourly returns in wage employment and family businesses have increased significantly. If women work mostly in farming, while men take advantage of opportunities in more lucrative wage employment or family enterprises, the relative contribution of women to household income will decline over time, which, in fact, is what we observe in the difference between women s and men s income. The income gap between women and men increases from 362 yuan in 1991 to nearly 900 yuan in In relative terms, women s contributions to household income also decreased from 44 percent in 1991 to 39 percent in In summary, this descriptive table paints a mixed picture for the role of women. We observe that women s absolute contribution to household income has increased, and like men, they are moving out of farming. However, men appear to gain more from the process of rural industrialization, as their relative contribution rose over the period. 7 We found that the price of main crops increased slightly in this period.

16 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 15 6 Village-Level Characteristics Time trends are not enough evidence upon which to draw inferences about development and the role of women. We wish to exploit spatial variation in the pace and nature of economic development in order to see what types of development correlate with the absolute and relative contributions of women. To begin, we explore some of the dimensions along which village economic experiences varied over this time period. Table 2 presents the summary statistics of the key village-level variables by sample provinces. These include the real per capita income, and the share of income from non-agriculture, that proxy the effect of economic development, and a set of prices--real male and female wage rates, and the real price of main crops--that may capture other important economic conditions. There is significant variation in the level of per capita income among the sample provinces. For instance, in 1991, the per capita income in a high-income province (Jiangsu, 900 yuan) is about twice the level of the income in a low-income province (Guizhou, 463 yuan). Nevertheless, all provinces experienced fairly rapid growth between 1991 and 1997, the average rate of which was about 45 percent for all sampled villages. Some provinces, such as Jiangsu, Henan, and Guangxi, shared similar growth rates, but otherwise, there were differential growth rates across provinces. 8 The per capita income in Jiangsu grew by nearly 59 percent over this period, but for Guizhou, per capita income increased by only 26 percent. Figure 2 arrays per capita income for the seven provinces. For Hubei, a middle-income province in the sample, there is virtually no growth of per capita income over the sample period. Thus, there is no clear trend that provinces with high income in 1991 grew faster, nor any sign of convergence in the level of incomes across provinces. 8 A similar estimate of the growth rate is also found in another survey over a 10-province sample of about 7000 households over the same period. For the panel households in the 10-province survey, the mean is 47.9 percent. In general, our estimates in levels and growth line up with these data.

17 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 16 The share of income from non-agricultural work (wage employment plus family enterprises) rose over the period. These sources of income accounted for about one-third of total income in 1991, playing a more important role in Jiangsu, Shandong, and Guangxi. In 1997, nonagricultural income accounted for more than half of the total income in the sample villages. In some provinces, like Hunan and Guangxi, non-agricultural income accounted for nearly twothirds of village income. This suggests a significant structural change in these economies over this period. In addition, we also observe considerable variation in the rate of nonagricultural structural transformation across provinces. Figure 3 shows that with the exception of Hubei, there is a clear corresponding relationship between structural change and economic growth in each province; that is, in each province, the level of income increases with the share of income from non-agriculture, though the share of non-agriculture income cannot explain differences in the level of income across provinces. Turning to the price variables, we find considerable variation of wage rates. The pattern of male wage rates is similar to the pattern for village per capita income, perhaps suggesting that improvement in male wage employment is the driving force behind the rising village incomes(figure 4). There is a similar change in the female wage rate. Figure 5 shows that provinces that experienced more rapid growth, like Jiangsu, Henan, and Hunan, have higher female wage rates. Over this period, female wage rates grew faster than male wage rates. In 1991, the ratio of female to male wages for the whole sample was 0.68, increasing to 0.79 in Figure 6 plots the female-male wage ratio across provinces. One notable observation is that this ratio reaches a high of 0.96 in Hunan, a province that experienced rapid economic growth and considerable structural change. This provides supportive evidence for the findings of Mathews and Nee (2000), that in areas experiencing rapid rural industrialization, shortages of male workers may disproportionately drive up women s wages.

18 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 17 Another important price variable, especially in determining the returns to agriculture, is the price of major crops. We computed an average crop price per jin of output. 9 Since the major crop grown is grain, changes in this price will affect the income from agriculture. The spatial variation in crop prices is much smaller than for changes in wage rates over the same period, possibly reflecting the fact the market for crop output is more highly developed and integrated compared to labor markets, which may be more localized (Figure 7). On the other hand, there is substantial variation of crop prices across time. The average real unit price fell by 15 percent between 1991 and 1993, and then rose by about 20 percent between 1993 and This intertemporal price change reflects the effect of inflation and the cyclical nature of farm product prices. 7 Household-Level Regressions 7.1 The Econometric Model In this section, we combine the household-level information outlined in Table 1 with the villagelevel information from Table 2 in order to explore the possible linkages between economic development and the role of women. The base model for household i in year t is given by: y = α' D + γ T + γ T + X ' φ + DEV ' β + P ' η+ ε (7) it it it it it it The dependent variable, y it, refers to the various labor supply and income measures by activity. We report results for women, and for the difference between men and women (women minus men, at the household level). Clearly, we expect some factors to affect the level of income or hours for women, but possibly differentially for men and women, and the difference specification allows us to look at the position of women relative to men. 9 Since there is no way to identify the main crop in each village, changes in price might be due to changes in the crop grown. Thus, this price change might also reflect shifts to higher-valued crops. However, the main crop in a particular village is mainly determined by weather and spatial considerations. A change in the crop price due to change of main crop will be insignificant.

19 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 18 Regarding the covariates, D it is a vector of household demographic characteristics, namely the number of family members in various age-sex groups. This is similar to the method employed by Craig and Field-Hendrey (1993) and Benjamin and Brandt (1995), to estimate women s contribution to total household farm production. The demographic variables capture differences in income or labor supply that can be attributed to changes and differences in household structure. For example, we expect households with more women to have more income from women. Less mechanically, we can explore the possible cross-effects of the number of men or children in the household on female activities. The time trend is captured by T93, T 97, respectively, for the cumulative time effect between 1993 and 1991, and the marginal time effect between 1993 and X it is a set of variables to control for differences in household endowments, including the household average years of male and female education, and household (per capita) arable land. The level of economic development is represented by DEV, which includes two proxies described earlier, namely, the logarithm of the village per capita income (lnvy) and the share of income from non-agriculture (ratnf). β captures the effect of economic development on women s labor supply and income. Both lnvy and ratnf are constructed by calculating the average value of the measure of all households in the village excluding household i itself. This construction breaks any simple mechanical relationship between the household and the village measure of the same variable, for example, due to common measurement error. P it is a vector of market prices, including the logarithm of the real village male wage rate (lnwm) and female wage rate (lnwf), as well as the real price of major crops (mvmp). We estimate two specifications for each dependent variable. The first specification includes only demographic variables and a time trend. This provides a clear picture of the time 10 These are not simply year dummies. T 93 is a dummy variable equal to one if the year is greater than or equal to 1993, and zero otherwise. T 97 is a dummy variable equal to one if the year equals 1997, and zero otherwise. it

20 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 19 trend, and essentially corresponds to a more rigorous version of Table 1, with adjustments for household demographics. In the second specification, we include the broader set of covariates. For each specification, we also present random and fixed-effects estimates. The random effects specification will take into account the common error components of repeated observations for each household, while the fixed effects specification will control for unobserved household (fixed) heterogeneity. One way to think about the differences between fixed and random effects is to note that the random effects estimator will exploit permanent, cross-village differences in income levels in estimating the impact of village characteristics on female outcomes, whereas the fixed effects specification will net out these differences, using only within-village changes over time (i.e., growth) to identify these effects. Finally, as households do not partic ipate in all activities, there are plenty of zeroes. To deal with this, we estimate the random effect Tobit model to account for this censoring The Regression Results for Labor Supply Total Hours In the following discussion, we will focus on the second specification (with covariates) in order to highlight the most important results,. Furthermore, in most cases, there is no significant difference between the random effects and fixed effects models. In the majority of cases, we will thus discuss the results from the random effects model, and only mention the fixed effects results when they are different. 11 The fixed effect Tobit model is inconsistent, mainly because the individual effects cannot be factored out of the likelihood function (see Maddala (1987). We therefore only show the random effect Tobit model. Heckman and MaCurdy (1980) suggest estimating the log-likelihood function using iterative methods. However, their estimation requires at least eight panel observations per household for the estimator to behave well.

21 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 20 Table 3 presents the regression results for total hours of labor supply for women, and the relative labor supply of women relative to men. There are three notable observations from the coefficients on the demographic variables. First, the number of young children (0 to 15), both boys and girls, are positively correlated with the total hours of female labor supply. Should this represent a negative income effect from having more children, we expect that families with more children will spend more time on activities with higher returns, such as family businesses and wage employment. To examine the sources of increase in female labor supply associa ted with children, we assemble the coefficients of the demographic variables obtained from different economic activities of the random effect model in the second specification in Table 4. Notice that the coefficients represent the marginal effect of having one more family member in different age groups. The share of time spent on different economic activities in Table 1 represents the average effect. This explains why in Table 1 the women spent on average 10 percent of their time on wage employment, while in Table 4, the contribution of wage employment is more than 20 percent in all age groups. Except for rounding discrepancy, the coefficients in individual activities should sum up to the coefficient of total labor supply. Thus, this table decomposes the sources of contribution of female total hours of labor supply. 12 Table 4 shows that young children increase women s hours of labor supply in farming, and reduce it in wage employment. This suggests that families who primarily engage in farming have more children. It may also be the case that childcare and farming are complementary, so that women substitute farm work for wage employment when they have young children. Note however, that only the coefficients for young females are significant in the difference between women s and men s total labor supply, 12 There are still some problems in interpret ing the contributions of older women. For instance, it seems more appropriate to interpret the presence of an old woman as releasing young females for wage employment.

22 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 21 suggesting that most of the effect of children is direct ; that is, young women contribute to women s household labor supply, and young men to men s total hours. 13 Second, we observe a life cycle pattern of labor supply. For instance, an additional woman aged will increase female labor supply by 1037 hours. Labor supply increases with age until 60, and thereafter declines (Table 4, column one). Third, while young women are more likely to engage in wage employment, older women are more likely to raise livestock. The negative coefficients on T 93 and T 97 show a clear negative time trend between 1991 and 1993 and, to a certain extent, between 1993 and This confirms the pattern of rapid decline in female labor supply observed in Table 1. However, there is no significant time trend in the difference between women s and men s labor supply, suggesting that men and women reduced their hours to the same extent. Male education has a positive effect on the level of women s labor supply, but apparently no effect on the relative labor supply of women. This suggests that women work more in families with more highly educated men. By contrast, we find that female education has no effect on the total hours of women s work, nor on the relative labor supply of women. However, female education may affect the kind of activity in which women work, a question that we address below. Consistent with the finding of Hare (1999), we find that women in families with more arable land work more, presumably on the farm. That per capita land cannot explain the difference between women s and men labor supply suggests that farming is a household (not just a women s) activity. Economic development seems to have a positive impact on women s labor supply, as reflected by the positive coefficient on the per capita income. However, this coefficient is not significant in the fixed effect model, suggesting that this positive effect does not hold within villages. Thus, it appears that women in richer villages supply more labor, and in poorer villages, 13 As we observe later, young children only increase the difference between women s and men s labor supply in farming, which again suggests that families predominantly in farming have more young children.

23 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 22 less labor. In addition, we find in the fixed effect model that per capita income has a negative effect on the relative labor supply of women. While the share of income from non-agriculture has no effect on the level of women s labor supply, it has a negative impact on the labor supply of women relative to men. Taken together, it appears that economic development provides women with more working opportunities, but men, particularly those in areas with significant structural transformation, are in a better position to take advantage of these opportunities (at least as far as hours of work are concerned). Wages and crop prices have no perceivable effect on the total hours women work or on the labor supply of women relative to men. We will examine if they have any effect on individual economic activities below. Farm Hours We present the regression results for women s labor supply to the farm in Table 5. Since over 95 percent of the families have at least one woman and one man farming, there is virtually no censoring problem and the results of the random effect regression and Tobit models are basically the same. Most of the demographic effects on women s la bor supply to farming, such as the effect on the level of women s labor supply and the life cycle features, are similar to those for total hours of women s labor supply discussed previously. The only exception is that young men aged increase the level of women s labor supply to farming. At the same time, they reduce the relative farm labor supply of women, suggesting that women and young men may be complements in farming. We also observe a significant time trend between 1991 and 1993, and a fairly modest trend between 1993 and 1997, confirming the observation of a sharp decline in women s hours in farming over this period of time. However, with the exception of the time trend between 1993 and 1997 in the random effects model, there is no time trend in the relative labor supply of women, suggesting common trends for both women and men. Men s education has a positive effect on

24 ECONOMIC DEVLEOPMENT AND THE ROLE OF WOMEN IN RURAL CHINA 23 women s farm labor supply, both in absolute and relative terms. It therefore seems that women substitute for more highly educated men as farm labor. 14 Paralleling this, women s education has a negative impact on the level of women s hours on the farm, though it has no effect on the relative labor supply of women in farming. Farm size increases women s farm labor supply, but has no effect on their relative farm labor supply. However, the negative coefficient of farm size, though it is statistically insignificant, suggests that in some cases when there is substitution, the extra land goes to men. Turning to the development variables, in the random effect model, the negative coefficient of the share of income from non-agriculture suggests that women do leave the farm when more non-farm work is available. However, both the per capita income and the share of income from non-agriculture increase the relative labor supply of women to farming, suggesting that men move out of farming even more in villages that are growing and developing quickly. To this extent, women are indeed being left on the farm, though it is in a relative, not absolute sense. There is little relationship between farm prices and women s farm labor supply. This should come as no surprise, since we saw a large drop in farm hours, accompanied by only modest price changes. The positive coefficient on the male wage, however, suggests that in villages with higher male wage rates, women supply more farm labor. But as the male wage rate has no effect on the relative labor supply of women to the farm, this suggests that the male wage rate captures a common (to men and women) effect of off-farm returns on farm labor supply. Livestock Table 6 presents the results for hours in livestock raising, possibly a traditional female activity. Demographic structure has two notable effects on the time women spent caring for livestock. First, additional young men (aged 16-20) increase the number of hours women spend in 14 This variable is not significant in explaining the level of women labor supply on farm in the fixed effect model. This is not surprising as family education level is more or less fixed within the households.