Notes

See Katz and Autor (1999), Card and DiNardo (2006), and Lemieux (2008) for syntheses of the various explanations put forward.

For example, falling demand for manufactured goods (resulting from increases in the offshoring of manufacturing activities, a weak economy among Canada’s trading partners, or the rising value of the Canadian dollar) may reduce employment and wages in manufacturing relative to other industries, thereby leading to slower wage growth (or greater wage declines) for manufacturing workers than for other workers. If the relative supply of highly-educated workers rises (as a result of rapidly rising numbers of Canadians receiving university degrees, or as a result of growing immigration by highly-educated workers), wages of university-educated workers may fall compared with those of other workers. Relatively low interest rates can boost demand for housing and labour demand for construction workers, thereby raising wages of construction workers relative to those of individuals employed in other industries. Increases in world oil prices may increase wages of workers employed in the oil extraction industry to a greater extent than is observed in other sectors of the economy.

Researchers who attempt to quantify the role of the aforementioned factors face several challenges: 1) technological change is unobserved in most conventional data sets; 2) technological change and international trade affect each other; 3) identifying the causal impact of changes in groups-specific labour supplies (e.g., young workers, university graduates, women, immigrants) requires strong assumptions about the degree of substitution between various groups of workers; 4) data on international trade by cells jointly defined in terms of commodity, industry, and country of origin is relatively scarce; and 5) measuring the impact of changes in social norms is a complex task. Furthermore, the relative contribution of these various factors may have differed across decades.

Workers’ main job is the job with the greatest number of work hours per week.

In these surveys, hourly wages are obtained by dividing the job-specific earnings reported by respondents for a given time interval (e.g., one week, one month, one year) by the number of hours worked during this time interval. The question asked in order to obtain information about respondents’ earnings refers to the "usual wage or salary before taxes and other deductions." As subsection 9.1 shows, some surveys use different earnings concepts or different hours concepts. For instance, the Labour Force Survey (LFS) explicitly includes tips and commissions in the calculation of earnings and explicitly excludes overtime in the calculation of work hours. In contrast, all surveys prior to 1987 make no explicit reference to tips or commissions when calculating earnings and make no explicit reference to overtime in the calculation of work hours. These differences should be kept in mind throughout the paper.

The Standard Industrial Classification of 1980 (SIC 1980) and the Standard Occupational Classification of 1980 (SOC 1980) were used in the Survey of Work History (SWH) of 1981, the Survey of Union Membership (SUM) of 1984, the Labour Market Activity Survey (LMAS) for the years 1986 to 1990, and some Tabs files of the LFS of 1997 and 1998. The North American Industry Classification System of 2007 (NAICS 2007) and the National Occupational Classification for Statistics of 2006 (NOC-S 2006) are currently used in all Tabs files of the LFS. The disaggregation of the last three decades into these two sub-periods is made possible because some Tabs files of the LFS of 1997 and 1998 have information on SIC 1980 and SOC 1980 while others have information on NAICS 2007 and NOC-S 2006.

Readers should also note that, because education categories changed in the early 1990s, a consistent definition of education can be obtained from 1981 to 1998 only by comparing individuals who have a university degree with other individuals. More detailed education controls can be used from 1998 to 2011.

For 1984, the sample consists of full-time paid workers employed in their main job in December. To maximize sample sizes while having independent observations from the LFS, Section 5 uses samples of workers who are employed in their main job in March or September.

RPP coverage is available on a consistent basis from 1984 to 1998, using SWH 1981, SUM 1984, LMAS for the years 1986 to 1990, and the Survey of Labour and Income Dynamics (SLID) for the years 1993 to 1998. However, the wording of SLID questions regarding RPP coverage changed frequently after 1998, thereby raising issues of comparability over time. See Morissette and Ostrovsky (2006) for details. Indicators of temporary work are available from the General Social Surveys for the years 1989 and 1994 and from the LFS starting for the year 1997. See Cranford et al. (2003) and Morissette and Johnson (2005) for analyses of temporary work.

Contrary to Baker and Drolet (2010), who analyzed the evolution of the gender wage gap from 1981 to 2008 using a single set of industry and occupation controls, regression analyses are performed for two distinct sub-periods: 1981-to-1998 and 1998-to-2011. Separating the 1981-to-2011 period into these two sub-periods allows the use of two sets of industry and occupation controls that are more detailed than those used by Baker and Drolet (2010). This in turn allows one to explain a higher fraction of the gender wage gap observed during the late 2000s than Baker and Drolet (2010) did. See subsection 9.2 for details.

Readers should keep in mind that gender differences in industry and occupation may result either from gender differences in preferences or from sectoral and occupational segregation.

Regardless of the metric used, most of the decline in the gender wage gap is found to have occurred starting in the early 1990s (Chart 4).

The results are based on the dynamic Blinder-Oaxaca decomposition outlined by Baker and Drolet (2010, p. 450).

Among workers aged 17 to 64, the gender wage gap narrowed by 0.073 log points. Roughly speaking, this corresponds to a narrowing of the gap by about 7 percentage points.

Because observed factors—e.g., changes in the provincial distribution of employment—and unobserved factors tended to widen the wage gap between men and women between 1981 and 1998, the total contribution of changes in job tenure, industry, occupation, educational attainment, and unionization in narrowing the gender wage gap exceeds 100%.

The increased participation of female university graduates in high-paying fields of study—a dimension unobserved in the data sets used in this study—is one of the unmeasured factors that may have tended to narrow gender wage differences.

More recently, Boudarbat et al. (2010) documented some recovery in the wages of young men between 2000 and 2005.

A similar pattern is observed among men aged 55 to 64.

The data used in this study suggest that age-wage differences increased mainly during the early 1980s. Data from the Survey of Consumer Finances show a more continuous increase in age-earnings differences between 1981 and the mid-1990s (Morissette 1998; Picot 1998). However, both sets of data show a significant increase in age-earnings differences from 1981 to the mid-1990s.

A similar conclusion holds when comparing wage growth of men aged 55 to 64 to that of men aged 17 to 24.

About one-third of the decline in the unionization rate of younger men and women can be accounted for by changes in the type of occupation or industry in which they were employed.

Compositional effects account for even less (about one-quarter) of the increase in the age-wage gap between men aged 17 to 24 and men aged 55 to 64 (See Table 25).

The introduction of the personal computer and the advent of computer-based technologies were one example of technological change.

Of the five dimensions of the wage structure discussed in this study, wage differences between highly-educated workers and their less-educated counterparts likely receive the most attention in both the popular and academic press, perhaps along with the gender wage gap. This is because knowledge of the evolution of this particular wage gap imparts very useful information to a variety of users. The gap in wages between the less-educated and the highly-educated is often described as the economic returns to higher levels of education (e.g., a university degree), since it represents the additional wages that university- (or other) educated individuals earn beyond those of high school graduates. This wage gap is also often referred to as the "university wage premium." Education policy analysts use information on the university or college wage premium to assess the benefits of investments in postsecondary education. Prospective students and their families use such information to determine the economic advantage of attending a postsecondary institution. Academics have used this statistic to better understand the causes of the rising earnings inequality between the less-educated and the highly-educated observed in many Western nations, including Canada. Immigration policy analysts use the information on the change in the wage gap as an indicator of the rising (or falling) demand for highly-educated workers, and may adjust immigration policy accordingly.

Because the data sets used in this study contain educational categories that changed during the early 1990s, they do not allow for an analysis of the evolution of wage differences across education levels from 1981 to 2000.

All numbers in this section are based on the March and September files of the LFS.

Average weekly wages grew by about 8% among men with trades certificates and by about 2% among males with bachelor’s degrees.

In line with Boudarbat et al. (2010), a quartic in potential work experience is used in year-specific regression analyses. Potential work experience is defined as a person’s age minus the number of years of completed schooling, minus 6. In these regression analyses, the dependent variable is either the natural logarithm of hourly wages or the natural logarithm of weekly wages. The omitted educational category consists of high school graduates. As Charts 8 and 9 show, a good portion of the narrowing of wage differences for men took place from the late 1990s/early 2000s to the mid-2000s. Since LFS data allow a distinction between immigrant and Canadian-born workers starting only in 2006, whether the narrowing that occurred from the late 1990s/early 2000s to the mid-2000s is observed for both groups of workers cannot be investigated.

These numbers result from year-specific regression analyses performed on full-time male workers aged 17 to 64. They are obtained by taking the antilog of the bachelor’s degree coefficient, minus 1.

When weekly wages are used as a measure of pay rates, average weekly wages of male bachelor’s degree holders were, all else equal, 37% higher than those of male high school graduates in 2011, down from 43% in 2000. The corresponding result for women in 2011 was 55%, down from 62% in 2000.

From 2001 to 2004, employment in the computer and telecommunications sector fell by about 15%.

It is assumed that, for these industries, the reduction in training costs and/or the increase in productivity induced by higher-than-average wages more than offset the increase in labour costs associated with the payment of these wages, thereby leading to higher profits.

For Canada, Gera and Grenier (1994) concluded that inter-industry wage differentials are relatively stable over time, and are consistent with the notion that firms in some industries have higher profits per worker and share these gains with workers. For the United States, Alexopoulos (2001) also concluded that some form of efficiency wages play a major role in the differences in wages among industries.

Other factors may be important: firms’ exposure to import competition or degree of export orientation; the magnitude of foreign investment; the degree of value-added achieved in the commodities or services produced by the industry; and capital intensity. Grey (1993) examined the importance of these factors for Canada, focusing on the manufacturing sector. He found that the degree of value-added explained more of the inter-industry variation than other variables. As expected, import orientation was negatively correlated with industry-level wages while export orientation was positively, but weakly, associated with industry-level wages.

For instance, a reduction in interest rates may help support demand for construction while an increase in the value of the Canadian dollar on the foreign-exchange market may reduce the global demand for exports of manufactured goods.

There are a few exceptions. For instance, industry codes for some services-producing industries, such as finance and retail trade, are fairly consistent.

Business services include finance industries, insurance carriers, insurance agencies, real estate industries, and services to business management.

In this regard, the following should be noted: a) in the absence of controls, average log wages grew 0.199 points faster in finance than in retail trade; and b) after controlling for changes in occupation and worker characteristics, average log wages grew 0.108 points faster in finance than in retail trade.

Average log wages in finance grew 0.228 points faster than those in construction (Column 1). After controlling for changes in worker and job characteristics, this difference is reduced by about one-half, to 0.119 points (Column 3). Likewise, while average log wages in finance grew 0.140 points faster than those in manufacturing (Column 1), this difference drops to 0.064 points after accounting for changes in worker attributes and occupations (Column 3). Hence, movements towards highly paid occupations and the relatively fast upgrading of the skills of workers employed in the finance industry explain, to a large extent, the relatively strong wage growth observed in this sector between 1981 and 1998.

Wage growth in this sector may also reflect the challenge of attracting workers to jobs located in small and remote communities.

The relatively strong wage growth in natural and applied sciences is consistent with the high and rising returns in the fields of engineering and mathematics and computer science at the college and university undergraduate levels. See Drewes (2010) and Walters and Frank (2010).