Occupational segregation and the gender wage gap in a dynamic East Asian economy.

• Author: Zveglich, Joseph E., Jr.

1. Introduction

   Labor markets in the East Asian "miracle" economies of Hong Kong, Singapore, South Korea, and Taiwan have undergone profound changes in recent decades. Their comparative advantage in low-wage labor, which helped fuel their export-led development, has diminished over time. In the 1980s and 1990s, labor markets in these economies shifted toward higher-skill manufacturing and services, while lower-skill, labor-intensive industries moved abroad. The benefits of these changes in labor market structures have been slow to reach women. Female workers, who held a disproportionate share of the low-skilled jobs during the early stages of development, continued to earn substantially less than men despite the rapid pace of macroeconomic growth and relative gains in education and experience. (1) This observation raises an interesting question: To what extent are trends in East Asia's gender wage gaps explained by shifts in their occupational structures?

   The key elements of this question--gender differences in wages and occupations--are the fodder of numerous existing wage-decomposition studies based on a Oaxaca (1973) type of approach. These studies typically include summary measures of occupational segregation among a host of detailed worker characteristics to explain differences in wages. Common summary measures are dummy variables for occupation and industry categories or proxies for occupational characteristics, such as physical demands, work conditions, and demographic composition. (2) When applied to large data sets containing information for detailed occupations and establishments, the standard decomposition procedure can achieve a high level of disaggregation and results close to the level of detail needed to apply corrective labor market policies. For example, Bayard et al. (1999) use matched employer-employee census data on the sex composition of detailed occupations, industries, establishments, and job cells (occupation establishment) and find that a large portion of the U.S. gender wage gap is explained by pay discrepancies within narrowly defined occupations within establishments. However, as the authors point out, such rich data for narrowly defined job cells are difficult to find, particularly outside the United States.

   A small group of studies have utilized an alternative decomposition method with less stringent data requirements for clarifying the role of occupational segregation in explaining the gender wage gap. The method, first devised in Brown, Moon, and Zoloth (1980), henceforth BMZ, decomposes a wage gap into a component due to differences in employment shares across occupations and a component due to within-occupations pay gaps. Each component is further divided into a "justifiable" and an "unjustifiable" portion, based on regressions performed with individual years of cross-section data and a few occupation categories. Since its initial application to U.S. data, this useful technique has been used to analyze labor markets in a wide range of countries, including Britain (Miller 1987), Sudan (Cohen and House 1993), Australia (Kidd 1993), and China (Meng and Miller 1995). A similar procedure, with further corrections for index number problems that occur in the choice of the base wage structure, is used to examine gender wage gaps in Ethiopia, Cote d'Ivoire, and Uganda (Appleton, Hoddinott, and Krishnan 1999).

   Our study develops a decomposition technique that extends the BMZ approach by providing new information on occupational structures at lower computational costs. In particular, our technique adds a trend analysis that allows the across- and within-occupations gaps to each have two dimensions--a pay dimension and an employment dimension--while the BMZ across-occupations gap has just an employment dimension and the BMZ within-occupations gap has just a pay dimension. Unlike BMZ, the trend analysis does not formally model the factors affecting women's occupational attainment, so it can say little about the justifiable and unjustifiable distinction for the employment-dimension terms. In avoiding the high computational costs of making such a distinction, the new procedure can be performed at a much finer level of occupational disaggregation than encountered in the BMZ study and its subsequent applications.

   For any two groups identified by some demographic characteristic, such as gender or race, the framework developed in this study yields an across-occupation effect that encompasses group differences in employment distributions across occupations and the associated market returns to these occupations. For example, the across-occupation effect can lead to an overall gender wage gap if pay structures within occupations are equitable but women are relatively concentrated in lower-paying occupations. The framework yields a within-occupations effect that encompasses pay differences across demographic groups within occupations and the degree to which workers hold more equitably compensated positions. For example, the within-occupations effect can contribute to an overall gender pay gap if women work in occupations that have inequitable pay structures, even if these occupations (such as senior officials and managers) are generally higher-paying occupations. Through its application to regression-adjusted wage gaps, the procedure controls for a wide range of worker productivity characteristics.

   To examine the link between changes in wages and occupational structures in East Asia, this study applies the occupational-decomposition technique to Taiwan labor data covering the period 1978-2000. The Taiwan data set is extremely comprehensive and reliable by developing country standards, with detailed demographic and employment information on individual workers. The analysis is performed using up to 59 detailed occupation categories. As in other East Asian economies, women in Taiwan have been catching up to men in terms of education levels, work experience, and labor force participation. In fact, East Asian economies stand out among most other countries for the speed with which they have closed the gender gap in school enrollment rates and average years of schooling. Furthermore, women's growing attachment to the labor force has placed East Asia ahead of most other regions in terms of women's share of the total labor force (World Bank 2000). However, a sizable wage gap has persisted in Taiwan despite declining occupational segregation since the late 1970s. The decomposition analysis sheds light on this puzzle by demonstrating that considerable pay inequity between men and women within detailed occupations explains most of the overall wage gap. This conclusion holds both for absolute levels and for trends over time. More broadly, investigating women's relative wages and occupational status in Taiwan enriches our knowledge of labor market dynamics in an environment where women's roles can be substantially different from those of women in the United States and Europe.

2. Description of the Data

   The study uses repeated cross sections of household survey data from the Manpower Utilization Survey produced by the Directorate-General of Budget, Accounting, and Statistics, Executive Yuan (DGBAS), Taiwan. This comprehensive data set has detailed information on individual worker earnings, hours worked, educational attainment, tenure, occupation, industry, location, and personal characteristics. The government has collected this data every year since 1978 by surveying an average of 57,000 people each year. In order to obtain a representative sample of wage earners, we reduced the data set to a sample of civilian, nonfarm, paid employees aged 15 to 65, from 1978 to 2000. Individuals who are unemployed or not in the labor force constitute the largest excluded category. (3) The final sample of paid employees contains an average of 19,500 observations in each year, and women constitute roughly 39% of the employees. While the analysis is performed with every year of data, results are reported as averages across time periods to limit the profusion of numbers and to average out business cycle effects.

   Industry and occupation data are categorized according to two-digit codes from the Standard Industrial Classification of the Republic of China (SIC) and the Standard Occupational Classification of the Republic of China (SOC), both of which are based on international standards. Some two-digit occupations have been combined in the analysis to avoid zero values for these occupation cells in some years, while other occupations have been further disaggregated by industry. As revisions in the SOC and SIC codes beginning in 1993 cannot be reconciled with earlier codes, the data aggregated by occupation have a structural break at that point. As a result, the analysis is performed using 59 detailed occupation categories for the years 1978-1992 and 38 categories thereafter. The Appendix provides more detail on the steps taken to maximize consistency over time in sorting the data by occupation and industry.

   In addition to analyzing overall employment and wages, the study also examines results for four broad occupational groupings, henceforth referred to as "clusters": (i) senior officials, managers, professionals, and technicians; (ii) sales, service, and clerical workers; (iii) light production workers and laborers; and (iv) heavy production workers and laborers. (4) This classification of light and heavy production work is consistent with a ranking scheme based on capital intensities by Dollar and Sokoloff (1994). Light production occupations correspond with Dollar and Sokoloff's "traditional light manufactures" or "food products" categories, and heavy production occupations correspond with Dollar and Sokoloff's "heavy industries" or "high-skill light manufactures" categories. This breakup of production occupations can be justified from a development perspective--countries generally move up the product ladder from...