High Technology Agglomeration and Gender Inequalities By Elsie Echeverri-Carroll and Sofia G Ayala * The high-tech boom of the last two decades overlapped with increasing wage inequalities between men and women in the United States. The extraordinary growth of high-tech industries increased the demand for college-educated workers, thus pushing up their wages. However, these industries tended to employ mainly skilled men, so similarly qualified women may have been left out of the employment and wage benefits associated with that extraordinary growth. This paper illuminates the degree to which gender inequalities in the United States emanate from the spatial concentration and labor demand characteristics of high-tech firms. I. Literature Review on Gender Inequalities The predominant model in the literature analyzing gender wage inequalities is the human capital model which argues that differences in wages (in the context of perfect competition) are explained by differences in human capital variables (e.g., education and experience) and the relative treatment of women by employers (e.g., discrimination usually measured by the error term). Not only are gender-specific factors a main culprit for the wage gap, but, as noted by Chinhui Juhn, Kevin Murphy, and Brooks Pierce (1993), so is the wage structure the array of prices set for various labor market skills (both observed and unobserved) determined by the interaction of supply and demand and the rents received for employment in particular sectors of the economy. Increases in wage inequalities among men in the 1980s were mainly explained by the increase in the returns to skills resulting from an increase in the demand for but abrupt slowdown in the supply of skilled men (Juhn et al. 1993). To the extent that this trend 1
reflects rising returns to skills and employment in high-wage sectors, it could be expected to disadvantage women as a group relative to men. Francine Blau and Lawrence Kahn (1997) found that improvements in women s human capital compensated for factors that affected women in a negative way during this period. A slowing convergence in the gender wage gap characterized the 1990s compared to the 1980s (Blau and Kahn 2004). Why? Using Juhn et al. s (1993) decomposition technique, Blau and Khan (2004) find that the unexplained portion of the regression analysis fully accounts for the slowdown in wage convergence in the 1990s. The unexplained portion could be a proxy for several factors, including a large reduction in demand for female labor, which may help explain why the difference between women s and men s wages was greater in the 1990s than the 1980s. David Autor, Lawrence Katz, and Melissa Kearney (2005) compare wage inequalities within the upper-tail wage group (90 th to 50 th percentiles) and within the lower-tail wage group (50 th to 10 th percentiles) for men and women. They find that wage inequalities within the upper-tail group between those in the middle and those at the top increased dramatically for men and women from 1979 to the present. They also find that wage inequalities within the lower-tail wage group increased rapidly in the first half of the 1980s for men and women, but after 1987, that inequality leveled off for women and decreased for men. Autor et al. (2005) also document the expansion of educational wage differentials, which increased sharply in the 1980s but grew more slowly in the 1990s, with particularly large increases in the earnings of college graduates. Both trends point to demand forces that have favored high-income college-educated male and female workers from the 1980s to the present. 2
Two factors are usually identified in the literature as the raison d être for the increase in the demand for skilled workers relative to the demand for unskilled workers: (1) the rise of high-tech industries and (2) technological changes in all industries, especially the increasing use of computers. But, as Autor et al. (2005) note, that second factor would not predict a deceleration in relative demand for skills in the 1990s, given the continuous rapid spread of information technology. Could high-tech industries explain it? II. High-Tech Industries Some Stylized Facts Figure 1 shows that the number of workers in high-tech industries increased at a rate of 20 percent in the 1980s and 1990s. The explosive growth in the second half of the 1990s was fueled by the emergence of numerous start-ups, many financed by venture capitalists. U.S. venture capital disbursement rose gradually from the early 1980s until 1994 when it reached over $4 billion, then rose rapidly to $22 billion by 1998 and soared beyond $100 billion in 2000 at the height of the dot-com boom (NSF 2004). In their study of the impact of industrial restructuring on wage inequalities, William Dickens and Lawrence Katz (1986) suggest that industries differ along one major dimension, wages. High-wage industries have higher labor productivity, fewer women, and more highly-educated workers, and low-wage industries show opposite characteristics. We find a similar division among high-tech and low-tech industries in our study of gender inequalities. Our weighted sample of over 35 million male and 27 million female full-time workers from the 2000 Census of Population i shows that 42 percent of full-time collegeeducated male workers worked in a high-tech industry in 2000 (the height of the hightech boom), while only 14 percent of women with similar qualifications worked in this 3
sector. High-tech industries paid on average $35.30/hour to their full-time collegeeducated male workers, but only $25.62/hour to similarly qualified women. These wage differences are maintained even after controlling for gender-specific effects and other industry and city variables that affect wages (see Section V). Finally, high-tech industries employed only 11 percent of non-college educated workers. What are high-tech industries? The Office of Technology Assessment (1982) describes high-technology firms as those engaged in the design, development, and introduction of new products and/or innovative manufacturing processes through the systematic application of scientific and technical knowledge. Most studies believe that high-techness is best captured by industries that have a proportion of workers in technology-oriented occupations ii above the national average (Daniel Hecker 1999; Pingkang Yu 2004). To calculate the number of technology-oriented workers in each manufacturing and service industry, we use the 2002 Occupational Employment Statistics (OES) from the Bureau of Labor Statistics. Our methodology leads to an underestimation of the number of technology-oriented workers for a few industries. iii We classify a manufacturing or service industry as high tech if its percentage of technology-oriented workers is at least 6 percent or twice the national average. iv Why do high-tech industries pay higher wages? Studies suggest that one important explanation might be the location characteristics of high-tech industries, which tend to concentrate in few cities with a relative abundance of R&D universities and venture capital firms and a large supply of skilled workers (AnnaLee Saxenian 1993). There is convincing evidence that workers are more productive in cities with a large concentration 4
of skilled workers because they benefit from space-bound knowledge spillovers (James Rauch 1991). Ross DeVol (1999) suggests that the most appropriate approach to measure high-tech spatial concentration is to develop an index that combines the location quotient (the degree of high-tech employment concentration in a metropolitan area s economy) with the metro area s share of national high-tech employment in a multiplicative fashion. We use this index in our models. III. Multi-Level Models of Wage Inequalities When observations are highly correlated within sub-groups, using OLS can yield biased standard errors because it assumes that random errors are independent and have constant variance. We analyze the relationship between high-technology (industry and spatial agglomeration) and wage inequality using a multivariate and multilevel model. The model accounts for unobservable similarities among individuals within the same city. The first level of our model calculates adjusted individual wage gaps (i.e., after controlling for an individual s observable characteristics). The second level is a citylevel equation. Both levels have error terms that are uncorrelated across levels. A. Level 1 The first level of the model is represented by: lnw ij = β 0j + β 1j E ij + X ij θ + ε ij lnw ij is the log hourly wage of individual i working in city j and is defined as the natural logarithm of annual wage and salary earnings, divided by the product of weeks worked and usual weekly hours. X ij is a vector of human capital variables whose effects are assumed to be fixed across cities. E ij is a binary education variable that is 1 if the 5
individual i in city j has a college degree and zero otherwise. β 1j is a randomly varying coefficient for each city j associated with the variable E. Thus, the excluded category is non-college graduate, and its effect is captured by β 0j for each city j when all X ij variables are centered around their labor market (metro-level) means to eliminate their individual effects. β 0j represents returns to non-college-educated workers, and β 1j represents wage gaps measured by the log difference between the hourly wages for non-college graduates (β 0j ) and the hourly wages of college graduates respectively for each city j. The individual error term is ε ij Ν(0,σ 2 ε ). B. Level 2 In the second level, we estimate the effect of high-tech agglomerations on variation across cities in the hourly wages of the non-college-educated worker (β 0j ) and in the wage gap between the non-college-educated and college-educated worker (β 1j ): β 0j = γ 00 + γ 0 H j + K j ρ 0 + S j η 0 + R j τ 0 + ν 0j β 1j = γ 10 + γ 1 H j + K j ρ 1 + S j η 1 + R j τ 1 + ν 1j H j represents a city s level of high-technology agglomeration; K j is the vector of human capital externalities comprised of the average years of experience of workers in a city; S j is the vector of variables that control for the effect of city factors that previous studies have suggested are the main sources of urban wage inequalities. In particular, S j is comprised of a city s percentage of unemployed; the proportion of college-educated migrants in the college-educated workforce of the city; the proportion of employment in import-sensitive industries; the unionization rates; and the natural logarithm of the median house price in a city. R j is the vector of three binary variables defining the region 6
of each city: Northeast, Midwest, West. The South is excluded to control for the effect of unobserved region-specific factors that affect wage gaps. The fixed coefficients (γ 0, γ 1 ) represent the effect of a city s high-technology agglomeration on the portion of wage gaps remaining after controlling for the distribution of observable and unobservable characteristics within cities. The importance of the fixed coefficients is that their statistical significance and their positive sign test the hypothesis that returns to the skilled (college-educated) would increase in cities with a greater level of high-tech agglomeration, but returns to the unskilled would decrease or remain the same. IV. Addressing Endogeneity Problems Our model assumes that high-technology agglomeration (H j ) is an exogenous variable in the second level model. However, studies (e.g., Saxenian 1994) show that other variables are important determinants of high-technology agglomerations. Hence, changes in these variables have a direct effect on the level of high-tech agglomeration that, in turn, affects a city s productivity and returns to education. We allow for endogeneity by specifying our model as a two-equation model in H j and logw ij. This model allows us to estimate covariances σ V ν 0j and σ V ν 1j to test for endogeneity. Our first equation is the multilevel model introduced in the previous section for logw ij while our second linear equation for the high-tech agglomeration index (H j ) is the following: H + V j =! 0 +! 1T1 j +! 2T2 j +! 3T3 j +! 4T4 j Here, T 1j is city s R&D expenditure, T 2j is venture capital investment, T 3j is average education, and T 4j is defense expenditure. These are deviations from their corresponding j 7
national averages to capture the idea that the higher the level of a city s pro-high-tech variables relative to the national average, the more attractive that city is for high-tech agglomeration. V. Results Table 1 shows that, after controlling for individual and city variables that affect wages, there is a 12-percent wage gap between male and female workers who worked for a high-tech industry. As expected, wages are highest for workers living in high-tech cities (HTA). Also, productivity-enhancing effects from living in a high-tech city are larger for male college-educated workers than for females (HTA*college). Are our results biased because HTA is endogenous? Results from our analysis (using aml software) for the male sample (column 3) show that the covariances across equations (σ V ν 0j and σ V ν 1j ) were not significant, indicating a lack of endogeneity for the HTA variable. We have no convincing reason to assume that the situation is different for the female sample, so we propose that our estimates from the models in columns 1 and 2 are not biased. In sum, the growth of the high-technology industry contributed to a polarization of work by increasing the demand for skill among highly-educated male workers and leaving women and the lower echelons of the wage distribution comparatively unchanged. 8
FIGURE1. HIGH-TECH EMPLOYMENT AND THE GENDER WAGE GAP, 1979-2002 Sources: Current Population Survey and Discontinued CES data on SICs, Bureau of Labor Statistics. Notes: Wages are defined in real weekly earnings. High-tech employment is calculated by adding annual employment for all high-tech SICs. Two high-tech employment series are calculated. One series accounts for employment in 18 of 30 high-tech SICs for which data could be obtained for the entire 1979-2002 period. Beginning in 1988, employment data could be obtained for 7 new high-tech SICs; the second series accounts for 25 of the 30 high-tech SICs. Data for the remaining 5 HT-SICs could not be obtained for any year. 9
TABLE 1. ESTIMATED COEFFICIENTS FROM THE REGRESSIONS OF WAGES ON DEMOGRAPHIC AND LABOR MARKET CHARACTERISTICS Variables Men: College grads compared to noncollege grads Women: College grads compared to noncollege grads Men: College grads compared to noncollege grads (correcting for endogeneity) Intercept Coefficient 2.514 *** 2.383 *** 2.496 *** College Educated Dummy 0.315 *** 0.378 *** 0.362 *** High-Tech Industry Dummy 0.195 *** 0.198 *** 0.179 *** HTA 0.012 ** 0.014 *** 0.007 HTA * College 0.019 *** 0.009 ** 0.036 ** 0j 0 0 0.285 1j 0 0-0.118 Number of Observations 1,582,304 1,243,066 10,048 Significance Levels: ***p 0.01, **p 0.05, and *p 0.10. HTA = High-Tech Agglomeration Index Model 3 did not control for some city variables: house price, experience, and R j 10
References Autor, David H.; Katz, Lawrence F. and Kearney, Melissa S. Trends in U.S. Wage Inequality: Re-Assessing the Revisionist. Working paper 11627, National Bureau of Economic Research, 2005. Blau, Francine D. and Kahn, Lawrence M. Swimming Upstream: Trends in the Gender Wage Differential in the 1980s. Journal of Labor Economics, 1997, 15(1), pp. 1-42.. The U.S. Gender Pay Gap in the 1990s: Slowing Convergence. Working paper 10853, National Bureau of Economic Research, 2004. DeVol, Ross C. America s high-tech economy: growth, development, and risks for metropolitan areas. Santa Monica, CA: Milken Institute, 1999. Dickens, William T. and Katz, Lawrence F. Interindustry Wage Differences and Industry Characteristics. Working paper 2014, National Bureau of Economic Research, 1986. Hecker, Daniel. High-Tech Employment: A Broader View. Monthly Labor Review, June 1999, 122(6), pp. 18-28 Juhn, Chinhui; Murphy, Kevin M. and Pierce, Brooks. Wage Inequality and the Rise in Returns to Skills. Journal of Political Economy, 1993, 101(31), pp. 410-442. National Science Foundation (NSF). Science and Engineering Indicators. Washington, DC: U.S. Government Printing Office, 2004. Rauch, James E. Productivity Gains from Geographic Concentration of Human Capital: Evidence from Cities. Working paper 3905, National Bureau of Economic Research, 1991. 11
Saxenian, AnnaLee. Regional advantage: culture and competition in Silicon Valley and Route 128. Cambridge, MA: Harvard University Press, 1994. Yu, Pingkang D. Focus on high-tech-what's in a name? Gauging high-tech activity. Regional Review, 2004, 14, pp. 6-9. Federal Reserve Bank of Boston. 12
End Notes * McCombs School of Business, University of Texas at Austin, Austin, Texas 78713. i Our sample comes from the 5-percent Public Use Microdata Sample (PUMS) of the 2000 Census of Population and includes workers, ages 18-65, working full time (at least 35 hours per week), neither self-employed nor working without pay, who had worked for at least fourteen weeks during the year prior to the census. ii Technology-oriented occupations (TOO) are: engineers, life and physical scientists, computer professionals and mathematicians (except actuaries), and engineering, computer, and scientific managers. As Hecker (1999) notes, workers in these occupations need in-depth knowledge of theories and principles of science, engineering, and mathematics. Such knowledge is generally acquired through specialized post-high school education ranging from an associate degree to a doctorate in some field of technology. iii In the 184 manufacturing and services sectors (where high-tech industries are concentrated) in the OES database, only 14 industries have more than 30 percent unreported employment in the TOO. In contrast, 90 industries have less than 10 percent unreported employment in these occupations. iv We find 38 high-tech NAICS in the manufacturing and service sectors by 4-digit 2002 NAICS. We compressed our list of 38 high-tech 4-digit NAICS to 33 3- and 4-digit NAICS used by the Census. 13