Chris Spohr is a Ph.D. candidate in the Department of Economics at MIT.This research was assisted by an International Predissertation Fellowship from the Social Science Research Council and the American Council of Learned Societies with funds provided by the Ford Foundation. Assistance by Dr. C. C. Li of the Ministry of Education in Taipei and the Office of the Directorate-General of Budget, Accounting, and Statistics have been particularly instrumental, as have discussions with Abhijit Banerjee (MIT) and Yih-chyi Chuang (National Cheng-chi University). Thanks also to the Lin's of Kaohsiung for giving me a more intricate perspective of family business in Taiwan. All errors are the responsibility of the author. Comments on this doctoral work-in-progress are welcome: email This e-mail address is being protected from spam bots, you need JavaScript enabled to view it

Note: please click here to view supporting figures and tables for this article in another window

Development economists have long recognized the potential role of formal education and human capital accumulation in determining individuals' labor market outcomes. Much of the literature studying this relationship is devoted to estimating "wage returns" to schooling: effectively, the correlation between wages and formal education among those reporting such income.¹ Not surprisingly, perhaps, this literature generally finds education-in particular primary and secondary schooling-to have far higher returns in less developed countries (LDCs) than in industrialized economies (see for example Psacharopoulos, 1994). By contrast, relatively few studies have addressed the somewhat more fundamental question of whether and how formal schooling might determine individuals' labor market participation in the first place, particularly for LDCs.² The degree of enfranchisement, defined here in terms of individuals' access to various forms of productive employment, may itself be a potent indicator of welfare and of the pace of modernization within LDC economies. Moreover, it is likely to bear upon a much broader collection of issues in fields outside of economics.³ With such issues in mind, we utilize data from Taiwan to explore the relationship between formal education and individuals' access to the diversity of possible employment outcomes coexisting at any point in that modernization process. Particularly, we look at the effects of the 1968 education reform on labor participation among different cross sections of Taiwanese society.

Earlier research in economics has explored education and various definitions of the participation/non-participation threshold in a variety of economic contexts, including Eckstein & Wolpin (1989) in the US, Funkhouser (1996) in Central America, Banerji (1992) across Indian states, Yang (1995) in rural mainland China, and Jiang (1992) in Taiwan. These generally suggest positive and significant effects of formal schooling on access to jobs in "preferred" sectors of the economy.⁴ However, as widely recognized in the labor economics literature, it may be dangerous to infer causality from such observed correlations in light of the econometric problem of "joint endogeneity" (briefly described in Section 2).⁵ Evidence from a wide variety of social science disciplines, although strictly speaking anecdotal, suggests that both education and formal labor market participation are strongly linked with status across income classes/castes and also within the household. Thus, disentangling the causal linkage between formal education and labor market outcomes should also be of interest in fields beyond the scope of economics.

Taiwan's experience may also shed light on another question which has generally received only ancillary attention in the mainstream economics literature: namely, the importance of the distribution of education above and beyond population-wide measures of human capital such as average years of schooling or the aggregate stock of university graduates.⁶ During the early 1990s, equity and education were frequently noted as keys to East Asia's widely acclaimed "miracle".⁷ While the onset of the East Asian financial crisis caused an abrupt and largely pessimistic reevaluation in the popular wisdom on Asia, Taiwan's ability to largely evade these economic woes is noteworthy. In particular, this crisis affected Taiwan's "top heavy" competitors among the region's most advanced economies (i.e. the Japanese and South Korean systems built around keiretsu and chaebol conglomerates, respectively) and also many nations of Southeast Asia who had largely relied on cheap labor.⁸ In short, the distribution of schooling and of workforce opportunities in Taiwan may help explain both historic growth and also dynamic flexibility to shifts in technology and market conditions in an economy still driven by its small and medium enterprises (SMEs). Secondly, returning to welfare per se (i.e., how the gains to aggregate-level growth are distributed across sub-segments of the population), Taiwan presents an ideal laboratory for studying the role of schooling in access to the "formal" labor market within the broader context of "growth with equity". (Fei et al., 1979)

Within the field of economics, policies are often primarily of interest as the analog of true experiments in the physical sciences. Thus, research described here utilizes Taiwan's universal extension of free, "compulsory" education from 6 to 9 years in 1968 principally as a shock to individual's schooling (operating through household decision-making processes). Using 18 consecutive cross-sectional surveys from Taiwan's extensive Directorate-General of Budget, Accounting, and Statistics (DGBAS) household surveys, the upward shift in average schooling (the "first-stage" outcome) is positively correlated to labor force participation outcomes.⁹ Comparing the first six cohorts of children born late enough to be affected by the reform relative to respondents born earlier, we find a sizeable and statistically significant upward shift in years of schooling and in labor force participation for these respondents relative to the preexisting trends across earlier cohorts. The policy increased average schooling for all males in Taiwan by almost 0.5 years, versus closer to 0.3 years for females. On the other hand, the increase in labor participation rates was nearly twice as large for females (roughly 2 percentage points). For both sexes, increases in enfranchisement are traced to waged employment (versus entrepreneurial activity) and a shift toward manufacturing and service occupations. Patterns are found to be qualitatively similar in urban and non-urban sub-samples, though effects are larger in the latter. Finally, these results may help to clarify Jiang's (1992) observations into labor force shifts in participation and wage patterns which accompanied the influx of more educated workers into the Taiwan's workforce: average years of formal schooling leapt from 6.54 to 8.37 for males and 5.41 to 7.98 for female workers between 1971 and 1981.¹⁰

Education: Taiwan-style

At the same time, the case of Taiwan's 1968 expansion of compulsory schooling may be of much broader interest in its own right. For example, establishing the linkage between shifts in schooling and workforce participation is also of consequence for policy analysis, since it may contribute to our more general understanding of the efficacy of law in LDC economies: i.e., the direct, indirect, and in some cases unintended outcomes of intervention from the center. In addition, it is noteworthy that this Taiwanese reform occurred at a critical juncture in that economy's transformation. In particular, the targeted expansion of minimal education standards (primarily affecting opportunities for low income/status groups) may explain how Taiwan achieved balanced development as it shifted towards a strategy of export orientation.¹¹

We consider two aspects of this fact. First, the extent to which Taiwan "harnessed" its rural population has been widely documented. Ho (1978) notes the prevalence of part-time agricultural households in rural Taiwan (i.e. at least one member contributes to most households' income via off-farm labor); Otsuka (1996) provides evidence for the argument that rural industrialization can create side benefits for local employment creation and self-sufficiency; finally, Chen (1994, in mandarin) presents a sociological evaluation of industrial diffusion via the sub-contracting system (waibao). While the island's small size and adequate infrastructure may have been advantageous in providing rural residents access to nearby urban labor markets, Taiwan's broad schooling base is likely to have been critical.¹² Second, Taiwan provides a laboratory for studying precisely how the division of labor by gender (in addition to urbanization level) co-evolves with movements in human capital and broader sectoral shifts. Such analysis may convey valuable lessons for other LDCs in Asia and beyond, since the Taiwanese experience of grass roots development represents a more widely viable model than examples of top-down industrialization. Given the cultural similarities and more recent transfer of entire modes of economic production across the Taiwan Strait, Taiwan's experience with compulsory schooling may be of particular relevance to continued development in mainland China.¹³

Overview of the Reform, Data, and Mechanical Issues in Estimation

The reform of compulsory education in 1968 extended the duration of constitutionally guaranteed, tuition-free education for all citizens by three years to include junior high schooling.¹⁴ In terms of public resource commitment, the first three years of the policy reform were the most ambitious, as indicated in Ministry of Education (MOE, 1997) statistics. Educational expenditures by the central government alone, which had increased at a rate of roughly 15 percent annually during the previous decade, leapt by almost 70 percent in school year 1968/69 to almost NT$ 977 million. They reached NT$ 1.23 billion the next year and NT$ 1.82 billion in school year 1970/71 before falling back to just above NT$ 1 billion during the next three school years. Increases in educational expenditures at the level of hsien (county) and city governments were slightly larger in nominal terms, though smaller in relative terms since expenditures at this level constituted roughly one-half of all public educational expenditures during this era. Across all administrative levels, the total government expenditures on education climbed from NT$ 4.47 billion in school year 1967/68 to NT$ 9.07 billion in school year 1970/71 (comprising 15.3% and 16.5% of all government expenditures in those years).

Ideally, we would like to verify the claim that the policy caused any observed shifts in schooling and labor patterns by precisely pinpointing educational inputs as well as those outcomes. Unfortunately, because junior and senior high schools were initially jointly administered, we cannot distinguish between upper and lower secondary levels prior to 1968 in Ministry of Education (1997) figures for physical infrastructure, numbers of full-time teachers, or expenditures. The commitment of resources to junior high education per se is clearly evident, however, using total numbers of classes (not classrooms) at all levels. For public junior highs, this figure exploded during the first three school years under the reform, reaching 14584 classes or an increase of 85% from school year 1967-68. Available data show fairly steady ratios of students to either full-time teachers or numbers of classes within secondary levels, suggesting the absence of any large "overcrowding" effects. These statistics provide independent support for key methodological assumptions regarding both the linearity of schooling profiles across pre-1955 cohorts and the focus of the reform's apparent impact on junior high school (JHS) enrollments in particular. Moreover, these figures are important as indicators of roughly unchanging quality of education, since we implicitly assume comparability between one year of JHS schooling before and after the reform.

In tracking effects of the 1968 reform, analysis described herein utilizes data from Taiwan’s Household Income and Expenditure Surveys (jiating shouzhi diaocha), compiled and provided by the Executive Yuan’s Directorate-General of Budget, Accounting, and Statistics (DGBAS) in Taipei. The modern, computer-tabulated DGBAS household surveys have been collected annually since 1976, evolving and expanding from an earlier survey project described by Fei et al. (1979). After omission of the first three surveys, the data set consists of responses from 18 consecutive surveys (years 1979 through 1996), and thus remains slightly larger than the 1976-1990 data sets used by Deaton (1997) and Fwu (1993).

In practice, we are restricted to using cohorts born in and after 1947, in order to avoid demographic and socioeconomic anomalies associated with both the end of Japanese occupation in 1947 and the mass exodus of “mainlander” Nationalist supporters to Taiwan with the communist victory in China in 1949. Available evidence (see also, for example, Liu, 1972) suggests that the gross share of respondents included in our final samples (i.e. those born 1947 to 1960) were born in Taiwan and raised under unbroken guomindang (Nationalist) administration. On the other hand, we exclude all respondents younger than age 26, in order to avoid under-sampling highly educated males (and their spouses) due to deferred participation in Taiwan’s compulsory military service. The final sample pooling the 1979-1996 DGBAS surveys consists of 258,454 adults (128,460 males and 129,994 females) in the 26-49 age range and born between 1947 and 1960.

Before tracing the reform’s impact upon schooling and adult workforce outcomes using this data, we briefly address some rather mechanical issues. First, we distinguish “opportunity costs”, a concept which may be unfamiliar to non-economists. In the context of schooling, this can essentially be treated as the value to the household of children’s time (potentially) spent in school activities. Opportunity costs are of particular importance for two reasons. First, while policies such as Taiwan’s 1968 reform can eliminate pecuniary costs such as tuition and even provide free textbooks to poor children, they may nonetheless do little to address “costs” in terms of alternative activities foregone due to school attendance. Second, opportunity costs may provide the strongest disincentives for schooling for precisely those groups of Taiwanese for whom the direct (i.e. monetary) costs would appear to be the greatest constraint. Namely, time expenditure related to schooling more directly competes with children’s obligations / contributions to farm households. Likewise, with respect to girls’ opportunities for schooling, multiple economic studies including Thomas (1990) have noted that girls’ time is especially costly in poor families or where the mother works full-time, since older daughters typically must bear a greater burden of raising younger siblings and helping with various household tasks. In fact, as discussed in Section 4, this may be particularly important in Taiwan, because the Chinese tradition of daughters’ marital exit shortens the time horizon for their contribution to their birth family (see also Wolf, 1972). The consequences for schooling are actually noted by Thornton et al. (1994), “This historical cultural system might tend to limit the amount of education that parents would be willing to provide for their daughters.”

Finally, before proceeding to the results, we discuss the issue of “endogeneity” (versus exogeneity) introduced in motivation of this study in Section 1. With a minimum of econometric detail, the following essentially addresses the question of why we don’t just compare trends in employment to trends in average educational attainment across surveys, or alternately run a straightforward regression of labor force participation upon years of schooling and other covariates for all individuals in a given survey(s). Since the interrelationship of brideprice/dowry with a variety of women’s outcomes such as education and marital matching has received a fair amount of attention in economics and other social sciences, we utilize the specific example of zifangqian in the Taiwanese context for demonstration.

Like many other sociological “variables” related to family or individual status, zifangqian - literally “private money” deriving from a bride’s net dowry, as described at length by Chen (1985) and Rubie Watson (1984) - may be strongly correlated to both educational opportunities and employment outcomes (for the same female at different stages of life), albeit in a rather complex manner. For example, a female born into a very poor family is likely to receive little formal schooling or marital dowry, as compared to a female born into a moderate income family. In light of evidence from a variety of contexts, one might expect that the former would also be less likely to report income following marriage: perhaps becoming an informal homeworker or having no autonomous control over income from outside work due to her low status. First, note that in this hypothetical comparison no reference has yet been made to educational attainment. Even in the absence of any gap in average schooling attainments for girls in poor versus non-poor households, this disparity in adult responses emerges as a result of factors related to status and marital matching (proxied here by the unobservable variable zifangqian). Of course, the effects of actual schooling gaps may compound this effect on labor force opportunity.

In practice, zifangqian is never disclosed, an anthropological fact that has considerable econometric consequences. As a result, the "naïve" regression model that includes schooling but omits such family and individual characteristics will misinterpret any differential in workforce outcomes as arising entirely from schooling. The straightforward approach will overstate the perceived role of education in the preceding scenario if we assume that schooling, participation, and the omitted status variable(s) are all mutually positively correlated.²³ A second point, however, is that the effects of such unobserved individual or household level characteristics are likely to be complex in reality: i.e. even if some proxy for zifangqian were available, it would be difficult to hypothesize about its relationship with education and employment. On one hand, evidence suggests that wives with the least financial autonomy are more likely to work but less likely to report individual income.²⁴ On the other hand, the relationship might be reversed at the other extreme: women from the wealthiest/highest status families are likely to receive more schooling than those from median income households, but less likely to be compelled to augment husbands' earnings through regular contribution of their own labor (due to "private money").

In the example above, the principal difficulty in estimation arises from individual-level characteristics: investments in schooling are themselves endogenously determined to the extent that they reflect such unobserved but wide-reaching factors. However, endogeneity of explanatory variables may also be encountered at the aggregate-level. A variety of factors such as Taiwan's shift towards an export-oriented strategy, a decline in gender biases over time, or various demographic effects associated with the (largely male) mass migration from the mainland might affect the opportunities and/or incentives affecting individuals' education and later employment for some sub-group or for the entire population.²⁵ In either case, in the absence of acceptable proxies for such tertiary factors, one econometric solution is to replace actually reported values of the hypothesized causal variable with an exogenous "instrument". In the present context, this variable should (obviously) be strongly correlated with education, but statistically independent of any unobservable individual-level characteristics or economy-wide trends such as those noted above. Effectively, this "instrumental variable" (IV) regression captures the true effect of schooling (at least that according to the model), while screening away the effects of the omitted factors which would bias the coefficient on the original schooling variable. Though the computational mechanics and the assumptions underlying this procedure are well outside the scope of this discussion, we return briefly to the interpretation of IV estimates in Section 4.²⁶

The Impact of the Policy on Education

Applying the technique proposed by Deaton (1997) to any desired sub-sample of respondents allows us to observe movements in average schooling across birth cohorts of that sub-sample. Sets of "cohort effects" for rural males and females born between 1947 and 1960 are plotted for demonstration in Figure 1. Note first that years of formal education can be fit fairly closely to an increasing linear trend for children who reached age 13 (the typical JHS entry age in Taiwan) too early to be covered by the reform (i.e. for cohorts born in 1947-54). The crucial cohort is the 1955 one because they were the first group affected by the 1968 reform when they reached junior high age. The second key observation is the marked increase in average schooling attainments for the first six cohorts to be potentially affected by compulsory JHS, lying well above the simple linear forecast based on earlier cohorts. These profiles are qualitatively similar for other sub-samples of males and females for each of the 5 urbanization units used in this study. Quantitatively, both the steep-ness of the "preexisting" trend and the size of the upward shift associated with birth in or after 1995 (i.e. precisely those cohorts principally "treated" by the policy reform) increase as we shift focus down the urbanization spectrum from Metropolitan Taipei to rural Taiwan-sheng.²⁷

We assume that the linear trend observed for unaffected 1947-54 cohorts would have continued to roughly characterize increases in average schooling for cohorts born immediately after the critical 1955 cutoff in the counterfactual case where no reform had been enacted.²⁸ The impact of the policy is thus simply the vertical deviation between actually observed cohort averages and the trend line (for either sex) continuing to the right in Figure 1. Subtracting out this trend for clarity, which the actual estimation procedure essentially does, the policy-era shift in schooling attainments resembles a rounded plateau, as shown in the top frames of Figure 2 for males and females in entire Taiwan sample.²⁹ The timing of this upward deviation provides some evidence for the additional assumption that the abrupt shift in policy led to these increases, which can thus be treated as independent from any unobserved factors which might endogenously affect relative opportunities and/or incentives for either schooling or employment compared across birth cohorts.³⁰ Figure 2 demonstrates that the deviation in average years of total schooling is almost entirely explained by expansion of JHS education alone (the level targeted by the reform), offering further support for this claim.³¹

Uniform national education policies such as Taiwan's 1968 reform might be expected to be the hardest to implement/enforce in non-urban areas, where they have the greatest potential impact but also face stronger norms (especially sexual biases) and often higher transportation and/or opportunity costs associated with schooling (see Section 2). This may be particularly true of rapidly developing economies where urban living standards may outpace those in more rural areas.³² Below, we explore more carefully both pre- and post-reform comparisons in schooling across urbanization levels, with the important caveat that these estimates do not account for relationships between individuals' education and their propensity to migrate.³³ Specifically, more educated rural children may be more likely than otherwise similar rural residents to permanently migrate to urban areas at some point in their adult lives; sons often relocate with the fenjia division in the natal household, while daughters might first relocate at marriage.³⁴

This is demonstrated most simply in an (unrealistic) hypothetical scenario, which is nonetheless inspired by Parish & Willis' (1992) research into birth order norms and schooling, in addition to a variety of anthropological evidence. Consider a hypothetical small farming community wherein each family has two sons: eldest sons (the zhangzi) are less likely to receive education, regardless of schooling policy, and in most cases can expect to inherit much of the household's non-divisible fixed assets. By contrast, parents are more likely to educate second sons (the cizi), particularly if tuition is suddenly eliminated. Cizi are also certainly much less tied to the land if not actually under pressure to find employment elsewhere.³⁷ Oversimplifying this hypothetical example still further, suppose that all cizi born after 1954 migrate to urban areas as young adults thanks to free JHS education, such that (years later) the pool of rural respondents from each of these affected cohorts consists of only zhangzi; the policy would then be estimated to have had zero (or even negative) effect on either schooling or employment rural areas. The more general point is that the extent to which guaranteeing free JHS schooling opens up opportunities for migration to urban areas, this important outcome is not verifiable using the DGBAS data. Our "non-urban" estimates consider only current residentsthose who have not migrated by the time surveyed as adults. Thus, these represent a lower bound on the impact of the policy on schooling and employment outcomes among the population actually raised in non-urban areas.³⁸

With this important caveat, we group together intercept and cohort trend estimates from regressions on both urban and non-urban sub-samples in Table 1. These provide a rough "snapshot" of the state of education in Taiwan prior to the 1968 reform.³⁹ Intercepts essentially represent the average schooling attainments among respondents born only one year too early to be affected (generally reflecting the minimum urban-rural disparity for the pre-policy cohorts); the trends capture the average rate of increase observed in each of these measures for all the unaffected cohorts included in the sample (1947-54). In addition to total years of schooling, these are also tabulated (without standard errors to conserve space) for the share of each relevant population completing each level or type of schooling ranging from primary to tertiary. Not surprisingly, schooling attainments in non-urban areas of Taiwan-sheng lagged far behind those in urban areas, with the non-urban/urban disparities typically appreciably larger than either those across non-urban localities or those across cities in Taiwan-sheng and the two designated metropolitan areas of Taipei and Kaohsiung. Given historical completion rates at the junior high level and beyond, a much larger share children in rural villages and townships were potentially affected by the extension of compulsory schooling to 9 years than in urban areas. The same holds for women vis-à-vis men; moreover, gender bias in schooling appears stronger in less-educated areas, as observed in many developing societies (see King and Hill, 1993).⁴⁰

Among cohorts born prior to 1955 as a whole, schooling opportunities had generally expanded more rapidly for girls relative to boysthis is also evident in MOE (1997) enrollment figuresand also for less urbanized locales. Comparing sex-specific trends for all 5 urbanization levels in Table 1a, average rates of increase for girls' schooling outpaces that among successive cohorts of boys by roughly 0.05 in rural Taiwan to 0.13 years attained per cohort in Kaohsiung-shi. Columns II through VII report corresponding results for the likelihood of completing distinct levels of schooling. A principal finding relates to the persistence of gender bias under the earlier 6 year compulsory guomin jiaoyu system in non-urban areas in particular. Above the primary level, increases in girls' schooling were at best only on par with that for boys in townships or rural villages of Taiwan-sheng. In the latter, girls born in cohort c were more likely than those born in cohort c-1 to complete primary school by more than two percentage points, versus a corresponding increase of only 0.53 % for successive cohorts of boys; by contrast, the cohort-by-cohort rate of increase in girls' likelihood of completing 9 years of schooling among unaffected cohorts had been roughly 2.1%, compared to a marginally faster rate of 2.5% among rural boys.⁴¹

In light of the magnitude of public resources committed to implementation of reform in the first several years, outlined in Section II, the topmost frames in Figure 2 provide a quick indication of the program's efficacy in terms of actual schooling increases. For males, "phase-in" of the program (i.e. the uptake of compulsory junior high schooling) was almost immediate. On average, males born in 1956 (i.e. the second affected cohort) received at least half an additional year of schooling relative to those born two years earlier, or slightly almost than 0.4 years more than they would have been predicted to complete based upon 1947-54 trends.

For females, phase-in was more gradual, and the overall shift relative to (faster) underlying trends was somewhat smaller: we estimate the policy's impact to have reached only 0.3 years for 1957-60 girls versus an increase of almost 0.5 years for boys born in the same years (corresponding increases among the non-urban sub-sample were roughly 0.4 and 0.6 years respectively). Deviation from prior trends only becomes highly significant in the 1957 cohort. Two explanations seem most plausible. First, secondary schooling was traditionally viewed as more of a luxury for girls than in the case of boys. Daughters' education was likely considered as primarily an indicator of social status, and largely of socialization value. Various anthropological accounts support the view that girls' schooling was considered a luxury good (in particular, see Wolf, 1972 and Thornton et al., 1994), as do MOE (1997) figures for level-specific enrollments. Moreover the relative sluggishness of female advancement in JHS education, particularly in rural areas, adds to the discussion in Section 2 of the opportunity costs of daughters' time prior to marital exit from the natal household. Within the Chinese family system, and perhaps under similar cultural norms elsewhere, provision of free primary school may result in large female uptake because it might, if anything, actually ease parents' child-rearing burden.⁴² By contrast, efforts to school adolescent girls may meet more resistance since schooling compromises potentially substantial contributions to the household.⁴³ Looking at MOE (1997) statistics on school continuation rates for the era bracketing the reform, while many more girls than boys terminated schooling immediately upon completing primary schooling, among the (smaller) share who actually completed JHS, girls throughout this era were slightly more likely than boys to continue into upper secondary schooling: among the last cohort of students "in the pipeline" at the time of the 1968 reform, these shares were 84% for girls versus just below 82% for boys graduating junior high schools in school year 1967/68.⁴⁴ A second and rather complementary explanation for delayed uptake involves a double standard related more closely to Parish & Willis' (1992) evidence on sibling birth order and intra-household dynamics. Simply put, where a daughter born after 1955 has an older brother who had been unable to complete at least junior high school, parents would be less likely to comply with the policy than where the birth orders were reversed.

In sum, the preceding results suggest that the 1968 reform of the guomin jiaoyu system did indeed have disproportionate impact on schooling in rural areas, though the efficacy of raising minimal educational standards in terms of promoting equity is tempered by the disappointing results for females' schooling. Though formal cost-benefit is not pursued here, aggregate-level estimates of the policy's impact on junior high school completion itself in Figure 3 may be suggestive. Light-colored bars show the share of all males or females born in each year receiving less than 9 years of schooling. For cohorts born after 1954, the darkened portion represents the reduction in this share arguably caused by the reform (relative to forecasted shares): the remaining light-colored bars represent non-compliance as a share of the total cohort population. The chart for females is particularly pessimistic as to the policy's influence over parents' decisions to school daughters: even for the sixth affected cohort (women born in 1960), non-compliance rates were roughly 23.4% or ten percentage points larger than the fraction of that cohort estimated to have received 3 additional years of schooling as a direct result of the policy shift.

Labor Force Outcomes of Policy-treated Cohorts

Evidently, while the 1968 compulsory school reform had substantial effects on the distribution of education in Taiwan, many children were "left behind" by the reform: especially girls. Alternatively, it might be argued that roughly 4 out of 5 children who received publicly-funded junior high schooling in the first years of the reform would have attained at least 9 years of schooling even in the absence of the policy shift. Nonetheless, it is important to consider the consequences for these cohorts of children as adults in terms of labor force outcomes. As noted in the introduction, the most commonly studied of such outcomes is wages. Interestingly, wage regressions on Taiwan tend to find fairly low returns to schooling relative to other developing economies.⁴⁵ We find that results strongly depend on whether entrepreneurial income is included; since "selection" into waged employment (and possibly systematic disparities in reporting) appears to be very problematic, such regressions are much less informative than analysis of participation outcomes itself.

Interestingly, straightforward regression techniques (using either OLS or probit models) on the entire sample of males or females suggest that formal schooling has only a nominal impact on the likelihood of individuals reporting positive work income in the DGBAS survey.⁴⁶ For females this is more surprising since, even at peak ages for employment, at most 50% have historically reported work income (versus fairly stable and near universal labor participation rates among males surveyed in their thirties).⁴⁷ Moreover, level-specific estimates for the impact of schooling from similar regressions (not tabulated) cast still further doubt on the ability of compulsory junior high schooling to considerably affect employment prospects. For males, these regressions suggest that any positive effect of schooling vanishes beyond the primary level. Females actually show a reversed pattern. Women completing junior high appear no more likely to report income than are unschooled/primary dropouts (controlling for time trends, age, etc.). Schooling has a large effect beginning at the upper secondary schooling level, increasing participation by roughly 12 percentage points, while junior college or tertiary schooling are associated with an increase 2-3 times as large. In short, endogeneity of schooling and other econometric flaws notwithstanding, these results would suggest that the policy could have had little meaningful effect on labor force outcomes for either males or females.

Without going into any more details of the IV technique or the actual two-stage least squares (2SLS) estimation procedure used to re-estimate the impact of schooling on labor force outcomes, it is clear that any such impact should ap pear if we simply were to repeat the so-called "first-stage" analysis from the preceding section, but replace schooling variables with the labor force outcome(s) of interest. When examining the corresponding patterns for pre-policy cohorts with respect to participation, there appears to have been little underlying trend in various participation outcomes: positive income response rates actually marginally declined for successive unaffected cohorts of males.⁴⁸ For example, the share of women born in 1954 reporting income (when surveyed near the prime working age) ranges from a low of 41% in Kaohsiung-shi to 50% in Taiwan-sheng townships and 52% in villages. Examination of subsequent cohorts of women demonstrates that the apparently higher rates of enfranchisement in less urban areas actually coincide with higher shares of women in entrepreneurial activity but lower rates of waged employment.⁴⁹ For males, we likewise find that work (mostly low-skilled labor) in primary and infrastructure-related sectors (i.e. construction, utilities, etc.) is most important in rural localities. Finally, the share of males reporting positive income roughly doubles that for females of similar agenearing universal response in the most urban areaswith even larger disparities in entrepreneurship: among the latter, virtually all bosses are male.

For the non-urban sub-sample, Figure 4 captures deviations for three labor force participation outcomes associated with the onset of coverage under the schooling reform, corresponding to Figure 2 for schooling. The first key observation is that these correspond fairly closely to movements induced in schooling for the same cohorts. Women affected by the reform demonstrate a particularly large increase in response of positive income, on the order of a 2 percentage point rise among females born between 1957 and 1960. Moreover, for males or females, net rises in enfranchisement conceal still larger increases in access to modern sector employment and waged versus entrepreneurial occupations. Similar, though somewhat more muted, effects appear in the urban sub-sample and in aggregated estimates. It should also be noted that the relationship between movements in schooling and participation across affected cohorts contrasts the absence of any clear association between deviations across the 1947-54, and is surprising in light of the simple OLS regressions noted above.

In sum, cohort-by-cohort analysis indicates a strong association between our instrument for years of schooling (a set of dummy variables for birth in each year between 1955 and 1960) and the following mutually-consistent shifts in adults' workforce participation:

1) Increased participation, defined simply as reporting any positive work-related income (POSINC)

2) A large increase in shares reporting positive income in modern sectors (MANFSERV pools manufacturing and services), with a smaller shift out of traditional sectors (PRIMINFRA combines respondents in primary and infrastructure-related industries).

3) Expansion of opportunities for waged employment (captured by the variable EMPLOYEE), associated with a shift away from entrepreneurial activities (ENTREP).

Before concluding, we provide some explanation for the important result that the instrumental variables estimates are generally much higher than those found by OLS. This disparity resembles a broad range of IV estimates for wage returns to schooling in the labor economics literature.⁵⁰ In the context of participation outcomes here, the large albeit only marginally significant disparities between 2SLS and OLS estimates also likely reflect the fact that IV essentially presents an estimate of "local average treatment effects" (see Imbens and Angrist, 1992). Accordingly, IV estimates identify the treatment effect of marginal schooling (specifically, at the JHS level) upon groups of the population most sensitive to movements in the instrument: socially marginal families with children born in or after 1955.⁵¹ In simplest terms, under our assumptions, these estimates capture the effect of schooling on precisely those children who would not have received JHS schooling in the absence of the law change (i.e. those comprising the dark shaded regions of bar graphs in Figure 3).

While these results do not necessarily imply that, for example, the expansion of schooling led to an expansion of modern sector industries in rural areas of Taiwan, they at the minimum offer strong evidence that relative attainment of formal schooling acted as a strong and persistent determinant of individuals' occupational prospects in an economy undergoing rapid growth and modernization. Moreover, despite the high rates of noncompliance among potential beneficiaries of free JHS schooling (especially serious for females), our estimates suggest considerable welfare benefits for those whom the policy lifted above economic or cultural barriers to schooling. These data do not allow us to directly test macro-level or "general equilibrium" effects on Taiwan's comparative advantage, though our results support the conclusion that the reform was important in broadening Taiwan's industrial capability. Likewise, we cannot strictly challenge Ibrahim's (1989) pessimistic conclusion that, "replacement [of men in lower-tier jobs] by women does not reflect women's advances in education, training or skills, but is instead related directly to a degradation process affecting the jobs they move into." However, our estimates on balance suggest that Taiwan's uphill struggle to enroll women in compulsory schooling nevertheless contributed to more equitable labor market outcomes for women.⁵²

Finally, the magnitudes of estimates for the impact of access to education on type of employmentpoint estimates are largest for rural females, but sizeable even in regressions for urban malesare rather striking in light of the Taiwanese ideal of the "heishou" entrepreneur and the noted vivacity of Taiwan's SMEs.⁵³ 2SLS estimates for the non-urban sector suggest that each additional year of schooling for females lowered the likelihood of reporting income as an entrepreneur by roughly 3.5%. This clearly represents a lower bound on the degree to which women exited from inferior employment activities as own-account workers, given that many low status women in such occupations do not report income. The coefficient of 0.707 (std. error 0.264) in column VII underscores the huge impact of compulsory schooling on women who actually received it: three years of additional schooling are associated with roughly a 21% increase in the likelihood of reporting waged employment. For males, estimates generally are smaller in magnitude but show a similar pattern.⁵⁴ A notable partial exception is the regression for non-urban males: the 1.76% drop in entrepreneurship per year of schooling appears to arise from shifts out of the "boss" rather than "own-account worker" category. Along with the almost identical decline in labor in primary or infrastructure-related jobs (i.e. the variable PRIMINFRA in column VI), this provides at least weak evidence that a fair number of males forwent opportunities to be "employers" in small-scale rural sideline industries once given access to employment in modern sector jobs.

Taiwanese Junior High School Students, Taibei, 1997

Though there is clearly a need to pursue cleaner estimates using alternative data sources, our preferred 2SLS estimates are generally larger than those obtained by simple OLS regressions on respondents born in 1947-60. Non-urban estimates in particular remain quite large despite a downward bias associated with selective migration, noted above; in other words, formal schooling appears to have sizeable effects on employment opportunities even for those remaining outside urban areas.⁵⁵

Conclusion

In sum, like contributions by Jiang (1992) and others, the research described here addresses the question of "how growth of education in Taiwan has helped boost economic development." However, our focus on Taiwan's 1968 compulsory schooling reform and changes in labor participation outcomes is rather distinct. Not surprisingly, our conclusions are also somewhat different. In particular, we note that the policy had less impressive effects on girls' education, raising average years of schooling by approximately 0.3 years, versus a roughly half year increase for boys in all of Taiwan. This is taken as evidence of the persistence of gender bias, which nonetheless represents in some respects economically "rational" parental decisions within the socioeconomic context. At the same time, for the minority of women fortunate enough to have received free junior high schooling, the policy substantially lowered barriers to participation in preferred occupations: 2SLS estimates for all females in Taiwan suggest that one year of schooling increased the likelihood of participation in manufacturing or service jobs by 7.52%, and that of reporting regular waged employment by 5.32%. Moreover, the reform's effects on access to both schooling and workforce opportunities were largest outside urban areas. While economic, institutional, and cultural factors have clearly mediated the relationship between formal schooling and workforce participation in Taiwan, the successes and shortfalls of compulsory JHS reform in the ROC may nonetheless convey important lessons for growth and distribution in other modernizing economies. 

References

• Amsden, Alice. 1989. Asia's Next Giant: South Korea and Late Industrialization. New York: Oxford University Press.
• Bates, Robert H. July, 1993. "'Urban Bias': A Fresh Look," The Journal of Development Studies, 29(4) 219-28.
• Becker, Gary S. January, 1985. "Human Capital, Effort, and the Sexual Division of Labor," Journal of Labor Economics Pt. 2, 3(1), S33-58.
• 1995, in Febrero and Schwartz (eds.). The Essence of Becker. Stanford: Hoover Institution Press.
• Birdsall, Nancy, and Sabot, Richard H. (eds.). 1996. Opportunity Foregone: Education in Brazil. Washington DC: Inter-American Development Bank.
• Behrman, Jere. 1996. "Measuring the Cost-Effectiveness of Schooling Policies in Developing Countries: Revisiting Issues of Methodology," Economics of Education Review Vol. 15, No. 4.
• Cameron, Stephen and Taber, Chris. 1999. "Borrowing Constraints and the Returns to Schooling," (working paper)
• Card, David. 1998. "The Causal Effect of Education on Earnings, " (working paper) University of California, Berkeley.
• Chen, Chung-min. 1985. "Dowry and Inheritance," in Hsieh & Chuang, eds., The Chinese Family and Its Ritual Behavior. Taiwan: Institute of Ethnology, Academia Sinica.
• Collins, Susan M. and Bosworth, Barry P. 1996. "Economic Growth in East Asia: Accumulation versus Assimilation," Brookings Papers on Economic Activity, 2, 135-203.
• Court, David and Kinyanjui, Kabiru. 1986. "African Education: Problems in a High-Growth Sector," in Berg and Whitaker (eds.), Strategies for African Development. Berkeley: University of California Press, pp. 361-92.
• Deininger, Klaus, and Squire, Lyn. 1996. "A New Data Set Measuring Income Inequality," World Bank Economic Review 10: 565-91.
• Directorate-General of Budget, Accounting, and Statistics, Executive Yuan, Taiwan, R.O.C. 1979, 1980, 1981. jiating shouzhi diaocha: Survey of Family Income and Expenditure
• Fei, John C. H., Ranis, Gustav, and Kuo, Shirley W. Y. 1979. Growth with Equity: the Taiwan Case. New York: Oxford University Press.
• Funkhouser, Edward. November, 1996. "The Urban Informal Sector in Central America: Household Survey Evidence," World Development, 24(11), 1737-51.
• Fwu, Bih-jen. 1993. "Long-Term Changes in Rates of Return to Education, Occupational Attainment, and Married Female Labor Force Participation: A Case Study of Taiwan, 1976-1990," (doctoral dissertation), UCLA.
• Griliches, Zvi. 1977. "Estimating the Returns to Schooling: Some Econometric Problems," Econometrica, 45(1), pp. 1-22.
• Ho, Samuel P. S. 1978. Economic Development of Taiwan, 1860-1970. New Haven: Yale University Press
• Ibrahim, Barbara. July, 1989. "Policies Affecting Women's Employment in the Formal Sector: Strategies for Change," World Development, 17(7), 1097-1107.
• Imbens, Guido and Angrist, Joshua. 1994. "Identification and Estimation of Local Average Treatment Effects," Econometrica, 62(2), 467-75.
• James, Estelle and Benjamin, Gail. 1986. "Educational Distribution and Income Redistribution through Education in Japan," Journal of Human Resources 22(4), 469-489.
• Jiang, Feng-fuh. April, 1992. "The Role of Educational Expansion in Taiwan's Economic Development," Industry of Free China. 77, 37-68.
• King, Elizabeth and Hill, Ann, eds. Women's Education in Developing Countries: Barriers, Benefits, and Policies. Baltimore, MD: Johns Hopkins University Press.
• Liu, Paul K. C. 1972. "Population Redistribution and Economic Development in Taiwan, 1951-1965. Taipei: Academia Sinica.
• Mazumdar, Dipak. 1989. Microeconomic Issues of Labor Markets in Developing Countries: Analysis and Policy Implications. Washington, DC: World Bank.
• Ministry of Education, Taiwan, R.O.C. 1997. Education Statistics of the Republic of China. Taipei.1998. Education Statistical Indicators: Republic of China. Taipei.
• Otsuka, Keijiro. 1996. "Rural Industrialization in East Asia, " mimeo, Tokyo Metropolitan University.
• Parish, William L. and Willis, Robert J. 1992. " " Economic Research Center / NORC Discussion Paper 92-8.
• Psacharopoulos, George. 1994. "Returns to Investment in Education: A Global Update," World Development 22(9), pp. 1325-43.
• Rosenzweig, Mark, and Wolpin, Kenneth. 1988. "Migration Selectivity and the Effects of Public Programs," Journal of Public Economics 37, 265-289.
• Shieh, Gwo-shyong. 1992. "Boss" Island: The Subcontracting Network and Micro-Entrepreneurship in Taiwan's Development. Washington, D.C.: American University Studies.
• Speare, Alden Jr., Liu, Paul K. C., and Tsay, Ching-lung. 1988. Urbanization and Development: the Rural-urban Transition in Taiwan. Boulder, CO: Westview Press (in cooperation with the Population Studies and Training Center, Brown University).
• Thomas, Duncan. 1990. "Intra-Household Allocation: an Inferential Approach," Journal of Human Resources, 25(4)
• Thornton, Arland and Lin, Hui-Sheng (eds.). 1994. Social Change and the Family in Taiwan. Chicago: U. of Chicago Press.
• United Nations. 1991. The World's Women: Trends and Statistics, 1970-90. New York.
• Vogel, Ezra. 1991. The Four Little Dragons: The Spread of Industrialization in East Asia. Cambridge: Harvard University Press.
• Watson, James L. 1989. "Self Defense Corps, Violence, and the Bachelor Sub-Culture in South China," Proceedings of the Second International Conference on Sinology. Taipei: Academia Sinica.
• Watson, Rubie S. 1984. "Women's Property in Republican China: Rights and Practice," Republican China 10(1) 1-12.
• Willis, Robert J. May, 1987. "What Have We Learned from the Economics of the Family?," American Economic Review (Papers and Proceedings), 77(2) 68-81.
• Wolf, Margery. 1972. Women and Family in Rural Taiwan. Stanford, CA: Stanford University Press.
• World Bank. 1994. Enhancing Women's Participation in Economic Development. Washington, DC.
• 1993. The East Asian Miracle: Economic Growth and Public Policy. Oxford: Oxford University Press.
• Yan, Yunxiang. 1996. The Flow of Gifts: Reciprocity and Social Networks in a Chinese Village. Stanford, CA: Stanford University Press.
• Zhao, Yaohui. November, 1997. "Labor Migration and Returns to Rural Education in China," American Journal of Agricultural Economics 79(4), 1278-87.
• Zveglich, Joseph E. Jr., Rodgers, van der Meulen, and Rodgers, William M. III. July 1997. "The Persistence of Gender Earnings Inequality in Taiwan," Industrial & Labor Relations Review, 50(4), 594-609.

In Chinese (note: excepting author's names, romanization uses the "pinyin" system, as in the text)

• Chen, Hsuan-shu. 1994. xieli wanglu shenghuo jiegou: taiwan zhongxiaoqiye de shehui jingji fenxi. Taichung: Tunghai University.
• Shieh, Gwo-shyong. 1990. "heishou bian toujia: taiwan zhizaoye zhong de jieduan liudong" Taiwan shehui yanjiu jikan, 2(2), 11-54.
• Zhongguo qingnian bao. September 10, 1998. "zoulai de yi ge guilang de jiushijiu ge." (12th article in a China Youth Daily series available via internet: http://www.cyd.com.cn/zqb/qnsw/llsj/2^1201.htm).

Notes

1. In addition to inconsistent reporting of wage and entrepreneurial income (where the latter is included at all), this approach often ignores much of the population who report no labor income despite (semi-) regular work: omitting disproportionate shares of women and other socially-marginal groups. This is particularly problematic in analyzing less developed countries (LDCs), due to prominence of the "informal sector" (see also Deiniger & Squire, 1996).
2. Svejnar (1989) notes that "wage and employment determination in the modern sector is one of the least understood areas of development economics."
3. In a chapter on mechanisms of family change in Taiwan, Thornton et al.(1994) note the broader implications of these issues and the need for further research, stating that "the lifetime influence of school attendance on children's contributions to the family economy is less clear. One obvious question is whether there is a net positive economic return on the investment in schooling." This became particularly true with Taiwan's shift "from a labor-pooling to a wage-pooling family economy," though the authors note the importance of returns other than immediate wages.
4. This result is less robust for females. While Kravdal (1992) concludes that more educated wives are less likely to drop out of the labor market due to childbearing, Unni (1994) finds participation rates for rural Indian women to be negatively correlated with both parents' schooling. In one highly relevant analysis of education policy in Peru, King (1996) finds that educational catch-up has been associated with increased participation but falling returns to schooling in the labor market for females.
5. Endogeneity of female's workforce participation is discussed in Willis (1987). Moreover, enfranchisement in LDC "formal" labor markets is likely to be more strongly influenced by unobserved individual, family, and environmental factors than are wage or entrepreneurial income comparisons within positive income respondents.
6. See Behrman (1996).
7. Bosworth and Collins (1994), for example, stress East Asia's early advantages in schooling relative to Latin America and other LDCs. Cassel (1996) suggests education to be a critical mechanism in income distribution, while Birdsall and Sabot (1994) describe the distribution of education as part of a "virtuous cycle" of economic growth. See also World Bank (1993) for a general discussion of education and other determinants of the East Asian Miracle.
8. The 1968 reform also departed from Japan's model of a "high-quality, low-quantity public system." (James & Benjamin, 1986). Guomindang emphasis on equity was at least in part dictated by social and political conditions.
9. Here the outcomes of interest are all treated as binary variables. For example, the variable POSINC takes on a value of 1 for individuals reporting positive income from work (regardless of industrial sector or whether such income is reported to be employee wages or entrepreneurial returns), and 0 otherwise. Alternative measures of workforce participation are treated similarly, and are outlined in Table 0.
10. Using Taiwan's Manpower Utilization Surveys from 1978 and 1986, Jiang finds that education is positively correlated to higher wages, but also higher labor mobility. Moreover, his conclusion that post-elementary education has on balance exerted a "disequalizing effect" is noteworthy.
11. Statistics presented by Fei et al. (1979) demonstrate two key trends: particularly after the mid-1960s, growth of export share was accompanied by a shift away from "traditional" products towards manufacturing and eventually the service sector as well.
12. More generally, Mazumdar (1989) notes that the dynamics of intra-rural linkages "is strongest in Asian economies with a highly developed off-farm sector." This suggests that the benefits of Asia's egalitarian educational structure performance may have operated far beyond the scope of its widely-cited impact on urban industrialization (see also Collins & Bosworth, 1996, or World Bank, 1994).
13. On paper, mainland China has a similar system of 9 years of compulsory, free schooling. In practice, however, there is an alarming and widening rift in educational opportunities between urban and rural areas. The situation facing rural children and schools (reports of teachers going unpaid for at least 6 months are not uncommon) has attracted the attention of private donors domestically and abroad, as epitomized by a sponsorship program called xiwang guocheng ("Project Hope"). Nonetheless, state-level educational planners may continue to prioritize higher education, as demonstrated by recent efforts to restructure the college examination system and to allow application over the internet. For perspectives on this, see for example Zhongguo qingnian bao (1998, in mandarin).
14. This system of guomin jiaoyu also entitles children of poor families to free textbooks.
15. Most original documentation available for the surveys is in Mandarin Chinese, perhaps explaining why the DGBAS data has not been widely exploited by western economists. See R.O.C. (1989) for a description of the Surveys of Personal Income Distribution in Taiwan (China), which incorporates income information from the full DGBAS surveys of income and expenditure, and is the basis of analysis in Deaton (1997).
16. 1976-78 surveys are excluded primarily due to some data concerns raised by DGBAS personnel, and because of the major rezoning between 1978 and 1979 surveys. Most notably, Kaohsiung-shi becomes a distinct administrative unit (Taiwan's second municipality) beginning in the 1979 survey, and it is desirable to distinguish this from other urban areas outside of Taipei.
17. While migrants also included Nationalist bureaucrats and wealthy industrialists and merchants, many were young male soldiers, forced to abandon their families and any property. The male/female ratio of respondents for each birth cohort in the unrestricted sample provides additional evidence on the demographics of these migrants. Consistent with published estimates, we find a sharp peak around the birth year 1927. The gender ratio reaches roughly 1.4 in these years but then falls very rapidly, effectively reaching one by the 1935 cohorts.
18. The remainder of this section outlines some econometric issues and introduces the concept of "opportunity costs". These are both discussed within the specific context of Taiwan, and both are important in interpreting the basic results presented later. Nonetheless, the explanation may be too rudimentary for those knowledgeable on either econometrics or on Taiwanese society, and can be skipped.
19. There is some evidence that this is changing with modernization in Taiwan, and also under the one-child policy in mainland China. The preference for sons seems to be fading due to parents growing expectation that daughters will better care for them in old age.
20. Becker's (1985) more general model of schooling, intra-household status, and the sexual division of labor has been fairly influential within the field of economics.
21. Skepticism for models of economic rationality in human behavior may be well-deserved in some cases. Nonetheless, marital norms offer a striking, if somewhat humorous, example of rationality in Taiwanese culture. Wolf (1972) describes the barter-like exchange of bride price (itself consisting of small and large payments, or siou tia and tua tia in Taiwanese) and dowry in terms similar to textbook descriptions of adverse selection theory, noting that "a bride who comes as a bargain will be presumed flawed by the neighbors".
22. For example, even in urban parts of Turkey, virtually no female homeworkers surveyed in Cinar (1994) reported themselves to be "working", though over 20% had been active in production for at least 5 years.
23. This is directly analogous to the problem of biases due to unobserved individual "aptitude" in regressions for wage returns to schooling, more commonly found in the economics literature (see Grilliches, 1977).
24. Wolf (1972) discusses women's contributions within Taiwanese households, though this is a fairly general result.
25. See Liu (1972) on migrational patterns. Here we focus on individuals born between 1947 and 1960, hence the primary concern is any impact that the aging population of displaced "mainlanders" might have in terms of labor market competition.
26. We adapt a technique proposed by Deaton (1997) for analysis of data available in consecutive household surveys. Deaton thoroughly describes the mechanics of decomposing individual-level variation into discrete, average effects associated with age, birth cohort, and survey year, and demonstrates this using DGBAS income data.
27. In addition to recording hsien or county location, the DGBAS survey discriminates 3 basic levels of urbanization: urban, township, and rural. We subdivide the former to differentiate between both metropolitan areas (the capital city of Taipei-shi, and the large industrial port of Kaohsiung-shi) and all other urban areas of Taiwan-sheng.
28. Forecasting exercises inevitably entail fairly strong assumptions. Nonetheless, here we specifically look at movements across birth cohorts, controlling for survey year fluctuations with an independent set of regressors.
29. The IV technique outlined in Section 2 essentially generates an alternative variable for education with values given by the height of this deviation: for all respondents born in the 1947-54 cohorts, it is identically zero.
30. Essentially, while it is conceivable that some other institutional or cultural discontinuity might affect a portion of the population on the basis of year(s) of birth, it is hard to imagine how such shifts might affect incentives for schooling at the junior high level alone.
31. These estimates based upon cohort-by-cohort analysis are also consistent with national-level MOE (1997) statistics for enrollments by school year.
32. For example, see Bates' (1993) discussion of "urban bias".
33. The difficulty arises because DGBAS data does not include place of birth and schooling. See Speare et al. (1988) on general migration and demographic patterns in Taiwan. With the partial exception of Liu (1972), there is very little documentation on differential educational characteristics of individual migrants relative to those remaining in rural areas.
34. In one suburban county of Beijing, Zhao (1997) finds that demand for high school education is driven by expectations of migration to urban areas. By contrast, returns to schooling in rural sector employment is very low.
35. Of course, given the stigma against males marrying women with more formal schooling, wives' education also generally conveys information about husbands' schooling attainments.
36. This would also be consistent with more general evidence that girls' education decreases their adult fertility because it raises potential wages and thus the opportunity costs of time spent raising children (see Thomas, 1990).
37. Evidence on schooling is presented in Section 3. Inheritance norms are lie outside the scope here. For examples of variations in contemporary culture, see Watson's (1989) case study of the Man, or Yan (1996) on the innovation of the danguo system as an alternative to fenjia household division in contemporary China.
38. For urban areas this is more complicated: urban-born children are on average more educated, but the influx of the most educated adults from non-urban areas appears to raise average schooling. In addition, the direction of bias caused by selective migration is more ambiguous for our "second-stage" estimates relating schooling directly to employment outcomes at the end of Section 4.
39. For example, the regression for the non-urban males allows both preexisting schooling trends and intercept terms to vary for the two non-urban units ("rural" and "township" areas of Taiwan-sheng respectively), but assumes that the 1968 policy had identical effects on males in both of these units.
40. Our estimates may additionally be of value because Ministry of Education (1997) statistics do not separately tabulate urban and non-urban enrollments, and in some cases do not differentiate by students' gender. Where comparable figures exist, our estimates appear consistent with those published.
41. This is generally consistent with national-level MOE (1997) statistics. Among children graduating from elementary schools in 1956, boys were more likely to enter junior high school by roughly 14.8 percentage points (53.67% versus 38.91%). Among 1967 graduates, the gap had increased to almost 21 percentage points (73.96% of boys entered junior highs versus only 52.01% for girls).
42. Even if child labor laws are weakly enforced or nonexistent, very young children's efficiency as laborers is relatively low in most contexts, hence a low opportunity cost to their time relative to slightly older children. 
43. This argument is generally consistent with general models of intra-household roles in developing economies (see for example Thomas, 1990). Where either daughters formally exit their birth family at marriage or face limited prospects for formal employment, parents are likely to be more myopic with respect to girls' schooling.
44. The share of junior high graduates immediately enrolling in higher levels of schooling (senior highs, vocational schools, and in some cases, directly into junior colleges) dropped by more than 10 percentage points as the first cohorts of students directly affected by the policy graduated junior high school in school year 1970-71. For most years thereafter, the gap remained at least 2 %, and has generally continued to increase over time: of 1996 graduates, 92.9 % of girls and 88.6% of boys completing compulsory education entered a higher level.
45. Our preliminary estimates (not reported) are fairly close to those reported, for example, by Psacharopoulos (1994).
46. These OLS results are compared to preferred 2SLS estimates in Table 3. We introduce distinctions below for type of labor participation: employer, waged employee, or own-account worker.
47. On the other hand, Zveglich et al. (1997) have argued that gender bias may have been accentuated by structural change since 1978.
48. Jiang (1992) presents one explanation for this decline, also found in the 1986 Manpower Utilization Survey.
49. Low rates of employment in Kaohsiung are an interesting exception in the overall pattern.
50. See for example Card (1998), Cameron & Taber (1999), and Duflo (1998).
51. By contrast schooling decisions for more affluent families would not be affected by elimination of junior high school tuition, though social stigma could have influenced the minority of these families who might have chosen not to enroll their children in the absence of the policy.
52. Even for Asia, there is little consensus in the debate on schooling and women's status. Our results seem closer in accord with evidence from Horton (1999), who cites increases in women's participation rates, including recent advances in clerical and other white collar jobs. Mitter (1996) presents a similar perspective, while Ibrahim argues that "women are taking jobs no longer acceptable to men."
53. A pragmatic spirit of rags-to-riches entrepreneurship, epitomized in the phrase "heishou bian toujia" (the "black hand" rising from manual laborer to boss), coexists alongside oft-cited "Confucian ethics". See Shieh (1990, 1992).
54. For the urban and non-urban regressions, 2SLS coefficients are statistically only marginally significant, due to higher standard errors than in the aggregate regression.
55. As noted previously, the latter includes a large number of non-urban residents employed in the urban sector. See also Ho (1978).