Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please contact us to request a format other than those available.
The male-female wage gap
The issue of male-female wage inequality is complex and requires analysis from a number of different perspectives. The goal of this article is not to provide a single, definitive estimate of the wage gap, but to demonstrate the importance of measurement, decomposition techniques, and differences in the gap within the distribution of wages.
Measures of earnings
For three decades, the Survey of Consumer Finances (SCF) was the primary source of data on the earnings of Canadians. As of 1997, the Survey of Labour and Income Dynamics (SLID) replaced the SCF. According to the SCF, among full-year, full-time (FYFT) workers, women working full-year, full-time earned 72.5% of what men earned (Table 1). Using the 1997 SLID, the ratios varied from 61.6% (using annual wages and salaries for all employees in their main job) to 68.3% (using annual wages and salaries for full-year, full-time employees in their main job) to 80.4% (using the hourly wage rate for all employees in all jobs).
Why the large differences in earnings ratios?
The SLID ratio, based on the hourly wages of all employees, was eight percentage points greater than the SCF ratio for full-year, full-time workers. This sizeable disparity arises from several conceptual differences between SLID and the SCF:
First, the ratios are for different populations of workers. The SCF ratio considers both employees and the self-employed, while the SLID ratio reflects only employees. Analyses based solely on FYFT workers tend to exclude a significant portion of the labour force—especially women. According to 1997 SLID data, roughly 76% of men and 60% of women worked full year, full time. Ratios based on hourly wage rates avoid these definition problems and are representative of female-male pay differences.
Second, the definition of earnings is different for each survey. SCF annual earnings include wages and salaries from all jobs, and net income from self-employment. SLID annual earnings include only wages and salaries. Hourly wage rates are job-specific, thus facilitating comparisons between the wages of men and women in comparable jobs.
Third, pay ratios based on annual earnings do not accurately account for differences in work volume. Even among men and women working full year, full time, the number of hours worked per week varies considerably. According to the Labour Force Survey, men employed full time in 1997 worked 43.1 hours per week, while women worked 39.0 hours per week. Ratios based on hourly wage rates overcome this problem.
The remainder of this article focuses on 1997, when women earned on average $15.12 per hour and men earned $18.81. In other words, women earned about 80% of the average male hourly wage. 1
Measures of experience
A major limitation of previous studies is the lack of sufficient information on the determinants of the wage gap. One such determinant is the quantity of lifetime work experience. Often age or the Mincer measure of experience (age — schooling — 6) is used as a proxy for the acquisition of general human capital skills or for potential work experience. However, proxy measures tend to overstate the actual work experience of women by not accounting for interruptions related to parenting (that is, complete withdrawals from the labour market) or for any restrictions on the number of hours worked per week or the number of weeks worked per year.
SLID collects information on work experience starting with the first full-time paid job. It asks respondents the years in which they worked at least six months (full year), as well as the years in which they did not work at all, since starting to work full time. The remaining years are coded as working part year. Respondents are then asked the years in which they worked six months or more, and if they worked full time (30 or more hours per week), part time, or both. Part-time or summer jobs while in school are not included. From the information provided, a measure of full-year, full-time equivalent (FYFTE) work experience is calculated:
Men and women bring different work experience to the labour market (Table 2). The Mincer proxy for potential work experience shows little difference in the work experience of men and women (19.5 and 19.1 years respectively). A different story emerges when actual labour market experience is applied. The average FYFTE work experience of men is 18.3 years compared with 14.4 for women. Men also spend a greater proportion of their potential years of work experience working full year, full time (94% versus 75% for women).
In absolute terms, the actual experience of young women is similar to that of their male counterparts but, with age, the gap widens (Table 3). This may be partially because older women (aged 55 to 64 in 1997) were less inclined to combine work and family than later cohorts (age 25 to 34 in 1997). As well, young men and women are new to the labour market and have not yet experienced the effects of career interruptions.
The SLID measure of experience, although a welcome addition to the study of pay differentials, is far from perfect. Information is missing on the continuity of work experience, the duration of labour force withdrawals, and the frequency and timing of withdrawal. These factors may influence the pay women receive in several ways (Corcoran and Duncan, 1979).
First, men and women differ considerably in the amount of time they work and in the continuity of their work experience. Women are more likely to combine periods of paid work with periods of labour force withdrawal for family-related reasons. This affects job tenure—a factor that influences wages.
Second, human capital skills may depreciate during long periods of labour force withdrawal. Women returning to the same employer after an interruption in employment may be less likely to be promoted. Or, women not returning to the same employer may have to accept lower wages than they received prior to their withdrawal.
Third, women expecting several withdrawals from the labour force may postpone training, or may decide to accept low-wage jobs in industries or occupations that are easy to enter and exit.
Fourth, the timing of labour force withdrawals may affect wages. Job-related skills are usually acquired at the start of careers—which generally coincides with decisions regarding children. A significant portion of real lifetime earnings growth has been found to occur during the first years after graduation (Murphy and Welch, 1990). If so, the timing of labour force withdrawals may have important long-term implications for future earnings patterns.
Using definitions of full year, full time to calculate FYFTE work experience may be problematic. Full year refers to working at least six months in a calendar year and full time refers to working at least 30 hours per week. As noted earlier, among full-time workers, the number of hours usually worked per week varies considerably. For example, Person A works 45 hours per week for 12 months, while Person B works 32 hours per week for 7 months. Under the SLID measure of FYFTE, each person would have one year of FYFTE experience despite the significant variation in work hours. As well, the measure of FYFTE work experience does not account for short-term (less than six months) labour force withdrawals in a given year. Almost 40% of working women take less than six months off work after giving birth (Marshall, 1999). If these unrecorded work interruptions and hours of work become more important as experience increases, then the measure of FYFTE work experience may become less accurate in reflecting the relative amount of work done by men and women.
The unadjusted wage gap is small for workers with less than two years of experience (4%) but grows larger as years of work experience increase. 2 A study of recent postsecondary graduates from the National Graduates Survey found that the gender pay gap was relatively small two years after graduation (7%) but widened two to five years after graduation (16%) (Finnie and Wannell, 1999): "These findings have interesting implications for the longer-term earnings profiles of graduates as they suggest that longer run (permanent) reductions in the earnings gap amongst cohorts of graduates might not be nearly as great as immediate post-graduation records suggest."
Education and major field of study
Canadian women have increased their educational attainment, matching and at some levels surpassing men. The importance of human capital characteristics—notably education—in the wage determination process has been firmly established in the economics literature. SLID contains information on educational attainment as well as major field of study. Not surprisingly, many fields of study are dominated by one of the sexes. For instance, graduates of engineering, applied sciences, technologies, and trades fields are mostly men. On the other hand, women are overrepresented among graduates with a commerce or business administration degree, and in the health and education fields. Since wages differ by major field of study, the choices made by men and women may account for some of the pay gap.
Explaining the gap
In describing the causes of labour market differentials, economists look first at the attributes different workers bring to the workplace. There is no universally accepted set of conditioning variables that should be included. However, the consensus is that controls for productivity-related factors such as education (level and major field of study), experience, experience squared, job tenure, marital status, presence of children, part-time status, union status, firm size, region, and urban size should be included. It is debatable whether occupation and industry should be included. If employers differentiate between men and women through their tendency to hire into certain occupations, then occupational assignment is an outcome of employer practices rather than an outcome of individual choice (Altonji and Blank, 1999). Analyses that omit occupation and industry may overlook the importance of background and choice-based characteristics on wage outcomes, while analyses that fully control for these variables may undervalue the significance of labour market constraints on wage outcomes. (see Accounting for male-female pay differences).
Using actual rather than potential experience allows more of the wage gap to be explained (Table 4). The explained component is about 9.2% to 37.4% in the conventional model (using potential work experience) and 29.2% to 49.3% in the augmented model (using actual work experience). This finding is novel for Canada but similar to findings in other countries (for example, Wright and Ermisch, 1991, for the United Kingdom; O'Neill and Polachek, 1993, for the United States).
The portion of the wage gap attributable to differences in work experience between men and women is severely underestimated when potential instead of actual labour market experience is used. Differences in actual work experience account for at most 12% of the gap, while potential work experience explains less than 1%. This may be explained as follows: First, as stated earlier, men and women differ little in the mean characteristics of potential experience but they differ significantly in actual FYFTE work experience. 3 Second, although potential and actual work experience are highly correlated, an additional year of FYFTE experience gives greater returns than a year of potential work experience. So, when FYFTE experience is used, both the difference in means and the difference in returns produce a greater explained component than when potential experience is used.
Differences in men's and women's educational attainment reduce the explained component by at most 5%, while differences in the major field of study by educational attainment account for no more than 5% of the wage gap. The contribution of each major field of study to the wage gap varies considerably. About 15% 4 can be explained because men are more likely to graduate from engineering and applied sciences programs, and degrees in these programs yield high returns. However, the prevalence of women graduating from health science and education programs—occupations associated with high earnings—reduces the explained component by 5% to 9%.
Despite the long list of productivity-related factors used in this study, a substantial portion of the wage gap cannot be explained. The same results were found in other notable Canadian studies (Baker et al., 1995; Gunderson, 1998; Christofides and Swidinsky, 1994). Such large, unexplained differences may be related to productivity-related factors, labour market decisions, or skill measures that are not captured by SLID.
Questions related to pay differentials are often framed in a manner that examines the extent to which women are paid the same as comparable men. For this reason, the male wage structure is often considered the 'competitive' or the 'non-discriminatory' wage structure. Alternative competitive wage structures may be employed. If the female pay structure is used, the question would be: What is the hypothetical wage men would receive if they were paid according to the female wage structure? Other approaches hint that the competitive norm falls somewhere between the male and female wage structure. One suggestion is that the competitive wage structure should be more similar to the structure for the larger group (Cotton, 1988). That is, the returns are estimated as the weighted average of the male and female coefficients, where the weights are simply the proportion of the total population that are men and women. Another suggestion is that the least squares estimates from a combined or pooled model be used as the estimate of the competitive wage structure (Neumark, 1988).
When the male wage structure is adopted as the competitive standard, 51% of pay differentials result from differences in the returns to wage-determining characteristics, while 49% result from differences in the endowments of wage-determining characteristics. When the female wage structure is adopted, male-female differences in characteristics explain 6%, while differences in the returns to wage-determining characteristics account for 94% of the pay differential.
By construction, the weighted average method yields decomposition techniques that are bounded by those obtained from the separate wage structures of men and women. This method yields an explained component of 30% and an unexplained component of 70%. Finally, the pooled method attributes a smaller proportion (39%) of the wage differential to differences in the returns to wage-determining characteristics, and a larger proportion to gender differences in characteristics, than the other three competitive wage structures.
These results suggest that adopting alternative comparative pay structures can lead to quite different interpretations of the components of male-female pay differentials.
The pay gap differs within the wage distribution
A major drawback in the preceding methodology is that it considers only the information of the average pay differential, and it assumes that the size of the wage gap and its components are constant along the whole wage distribution.
In relation to the absolute predicted wage gap based on the attributes of the typical worker, about 47% of the difference in pay at the 90th percentile compared with 57% at the 25th percentile was explained by male-female differences in observable characteristics. This suggests that the 50% of the wage gap due to men and women possessing different wage-determining characteristics at the mean fails to accurately represent the differences encountered along the wage distribution. As well, the contribution of relevant factors in explaining pay differentials varies throughout the wage distribution. Work experience, job responsibilities and industry contribute more to understanding pay differentials at the upper end of the distribution relative to the lower end, while occupation, union status and part-time status explain more at the lower end of the distribution relative to the upper end. 5
Other factors may contribute
The regression models used in the decomposition analysis account for no more than half of the variation in the hourly wages of men and women. The fit of the model to the SLID data can be improved by including other variables deemed to influence wages. The data typically used come from household surveys. Researchers have been unable to document the potential effect of firm characteristics—other than industry and firm size—on the wages of men and women. The new Workplace and Employee Survey will allow researchers to move beyond the individual worker to consider the importance of the workplace. But as noted earlier, pay differentials can be explained only if the factor in question influences wages and differences exist in the distribution of the factor among men and women.
People differ in their preference for particular types of work (that is, paid work or self-employment, hours, location, responsibilities). Differences between men and women in the labour market may reflect genuine differences in preferences, pre-labour market experiences, expectations, or opportunities. It is therefore difficult to distinguish between choice-based decisions and differential treatment based on sex. For example, pre-market experiences—which are related to expectations based on sex, both at home and in the education system—may influence the level of educational attainment and the choice of major field of study, labour force participation, job selection, and work habits.
Much of the literature has emphasized the importance of imperfect information about worker attributes. Employers are constantly making decisions regarding hiring and promotion and may use sex to predict future work commitment. Some firms may hesitate to hire women because women have, on average, more career interruptions and more absences for family reasons than men. Consequently, statistical discrimination may result (Arrow, 1973). This would be especially true in firms with substantial hiring or training costs, or where wages are higher. Insofar as employer behaviour changes as a result of women's increased participation in the labour force, women's access to high-paying jobs will improve due to decreases in statistical discrimination.
While earnings differences between men and women have narrowed since the 1970s, they continue to be remarkably persistent. Measurement and methodological issues play important roles in studying these differences.
Estimates of the pay ratio are sensitive to the measurements used. Conceptual differences in data sources and earnings measures that do not fully account for differences in work volume lead to different estimates of the ratio. The portion of the wage gap attributable to differences in work experience between men and women is severely underestimated when proxies for experience are used instead of actual labour market experience. Education level plays little role in explaining wage differentials, but choice of major field of study furthers our understanding of the issue.
Conclusions are often sensitive to the methodology chosen by the researcher. Questions are often framed in a manner that examines the extent to which women are paid the same as comparable men. Most studies use the male pay structure to make comparisons. However, alternative pay structures can be employed, with estimates of the unexplained portion of the differential ranging from 39% to 94%.
One drawback is that only the average difference is considered. When the size of the wage gap and its components along the whole wage distribution are examined, a different story emerges.
Marie Drolet is with the Business and Labour Market Analysis Division. She can be reached at (613) 951-5691 or firstname.lastname@example.org.
Appendix: Accounting for male-female pay differences
The wage structures of men and women were examined by estimating the relationship between hourly wage rates and observed characteristics in semi-logarithmic form:
where the natural logarithm of wages is used as the dependent variable, is a vector of worker and employer characteristics, is a vector of coefficients measuring the returns to those characteristics, and is the error term. Each coefficient is the percentage change in hourly wage rates associated with a one-unit change in the explanatory variable.
From the estimated regressions, the difference in the mean log wages between men and women is decomposed into three terms (Table A1):
where is the choice of comparative wage structure. The choice of comparative wage structure is examined in the section The choice of non-discriminatory wage structure matters. The first term represents the explained portion, which includes male-female differences in worker characteristics evaluated at the competitive wage structure . The residual or unexplained component is the proportion of the gap due to differences in the returns to wage-determining characteristics and consists of a male advantage (second term) and a female disadvantage (third term). This decomposition is made possible by the ordinary least squares (OLS) property that the sample average wage, is equal to the product of the average vector of characteristics and the estimated regression coefficients .
Previous work has shown that men and women differ in ways that may affect their productivity, but these differences do not necessarily explain the pay gap. It can only be explained if the differences in the factors in question are themselves important determinants of the pay received. The wage gap between men and women may be the result of either differences in their productivity-related characteristics or differences in the compensations they receive for the same productivity-related characteristics.
Quantile regression methods reveal dispersions in the wage differential that cannot be captured by OLS models. The properties of the OLS model used above (Equation 1) ensure that the average wage, is equal to the sample average characteristics evaluated at the estimated regression coefficients . However, the quantile regression model does not have a comparable property and as a result, no exact decomposition is possible. The difference in the log wage between men and women can be formulated (following the notation of Mueller, 1998) as:
where is the natural logarithm of hourly wages for sex evaluated at quantile , is a vector of estimated coefficients for sex evaluated at quantile , is a vector of average characteristics of worker ( = m, denotes male and female respectively and = 0.10, 0.25, 0.50, 0.75 and 0.90).
Two sets of predicted values can be calculated from the coefficients of the quantile regression model of the log hourly wages. The first is conditioned on a vector of characteristics of men and women around the percentile, while the second is conditioned on the vector of average characteristics of all men and women (Table A2). To calculate the predicted wages of men and women at the 25th percentile, the first set of conditioning variables includes the characteristics of individuals falling within the 20th - 30th percentile. The second set of conditioning variables used to calculate the predicted wages is simply the average characteristics of all men and women respectively.
For each percentile, both the actual and predicted wages are always greater for men than for women. The predicted wage gap can deviate from the actual wage gap. For instance, the actual wage gap at the 10th percentile is about 23%. But the predicted wage gap is 18% when the characteristics of the typical worker are used, and 19% when the average characteristics are used. Using the characteristics of the average worker, the predicted wage gap increases along the wage distribution. However, a similar conclusion cannot be reached when the predicted gap is based on the characteristics of a typical worker in the percentile or when the actual wage gap is computed.