Studies on Gender and Intersecting Identities
Measuring and Analyzing the Gender Pay Gap: A Conceptual and Methodological Overview

Warning View the most recent version.

Archived Content

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.

by Melissa Moyser, PhD

Release date: August 30, 2019

Skip to text

Text begins

Acknowledgements

This study is funded by the Department for Women and Gender Equality Canada.

1. Introduction

Pay inequality between women and men is a persistent phenomenon in Canada and around the world.Note Note Focusing on Canada, female labour force participation has risen dramatically since the 1960s, driven simultaneously by an increase in the proportion of employed women at any given time, and a decrease in the propensity of employed women to withdraw from the workforce upon marriage or motherhood.Note Note Note Women are now better educated, on average, than their male counterparts, having made substantial gains in educational attainment over the past three decades by increasingly acquiring university degrees at the bachelor level or above, and doing so at a faster pace than men.Note Note Note At post-secondary institutions, women have diversified their fields of study, making inroads in lucrative science, technology, engineering, and mathematics (STEM) specialties.Note Note Recent cohorts of women also seem to have stronger attachments to the labour market than their predecessors, as they are more likely to return to employment after a work interruption (e.g., maternity/parental leave), and to be employed when they have young children.Note Note Although women remain the minority, their representation in leadership positions (i.e., senior management) and in the top 1% of income earners has grown since the 1980s.Note Note Finally, social-policy support for women’s employment, in the form of job-protected maternity/parental leave and partial income replacement for that leave, has expanded, along with legislation addressing unfair treatment in the labour market.Note Note Note Note

While these trends are associated with a reduction in gender differences in pay in Canada, women aged 15/16 years and older earned from $0.69 to $0.89 for every dollar earned by men in 2017, depending on how the gender pay gap is measured. Pay inequality between women and men is an important social, economic, and political issue for at least four reasons. First, earnings are the main determinant of economic well-being for employed individuals, and of potential gains to employment for those who are not currently in the workforce.Note In this way, earnings serve as a significant input into a variety of decisions, from labour supply to family formation, as well as a factor affecting bargaining power and relative status within the household.Note

Second, there have been dramatic changes in the structure of families in Canada: dual-earner families now outnumber single-male-earner families; single-female-earner families and lone-parent families (usually headed by mothers) have become increasingly common; and relationship dissolution is prevalent.Note Note Note Note While women’s contribution to household income has grown, gender differences in earnings make it harder for them to financially support themselves and their families.Note Further, a lifetime of pay inequality between women and men means that women are disproportionately retiring into financial insufficiency and even poverty.Note

Third, the gender pay gap is symbolic of gender-based discrimination and injustice. Women spend more time than men on unpaid housework, childrearing, and caregiving (i.e., domestic labour), and they often reduce their labour force participation accordingly, putting them at a disadvantage in the labour market and making them vulnerable to low income, particularly in the event of union dissolution and during old age.Note Note Women’s disproportionate responsibility for domestic labour enables men to prioritize paid work and spend more time doing it than women. Although both paid work and domestic labour are integral to “social reproduction”—that is, the interdependent processes by which societies organize themselves to produce and reproduce material life on a daily and intergenerational basis—only production for the market is economically rewarded.Note It follows that the unpaid work performed disproportionately by women creates broad social benefits, yet they pay the price for doing that work in the form of lower earnings, relative to men.

There is also a connection between the work that women do at home and their paid work, in that women’s employment is concentrated in industries and occupations that parallel their traditional gender roles.Note Note Female-dominated jobs tend to be underpaid, even when they involve the same level of skill as male-dominated jobs.Note Note

Fourth, pay inequality between women and men perpetuates the gender division of labour. When women earn less than their male counterparts, even when they work the same amount of time, it is economically disadvantageous for individual families to move toward greater sharing of paid and unpaid work.

For these reasons, reducing pay inequality between women and men is a key priority, both nationally and internationally, for achieving gender equality. In Budget 2018, the Government of Canada introduced the Gender Results Framework, which includes the gender pay gap as a measure of progress toward the strategic goal of women’s equal and full participation the economy. Internationally, the principle of equal pay for work of equal value is included in the 2030 Sustainable Development Agenda under Goal 8, “decent work and economic growth.”

Documenting gender inequality in pay and tracking progress in this regard for policy purposes requires at least one indicator. The gender pay gap—the difference between the employment earnings of women and men, expressed either as a proportion of men’s earnings (i.e., the “gender pay ratio”) or one minus the gender pay ratio—typically serves this purpose. At present, there are no internationally-recognized standards for measuring the gender pay gap, leaving considerable scope for political choice.Note The purpose of this paper, then, is to inform the development of international standards for measuring the gender pay gap by explaining the assumptions underlying, and the implications following from, various methods.

This paper has two additional purposes: One is to increase literacy about the meaning and interpretation of different estimates of the gender pay gap.  Understanding the distinction between the unadjusted and adjusted gender pay gap is particularly important, as confusion in this regard contributes to debate as to the very existence of the gender pay gap. The other purpose is to bring together various explanations for the gender pay gap, and highlight the importance of taking context into account.

This paper is organized as follows: in Section 2, measurement of the gender pay gap in terms of who estimates are based upon, what type of earnings are counted, and how “typical” earnings are calculated are discussed. Section 3 addresses analysis of the gender pay gap, focusing on prevailing explanations for earnings inequality between women and men; the statistical technique used to determine the relative importance of different explanations (i.e., Oaxaca-Blinder decomposition); and considerations when studying trends in the gender pay gap. Section 4 concludes, providing an overview of key points.

Début de l’encadré

Text box 1
Sex, gender, gender identity, and gender expression

The complexity of sex and gender has become increasingly evident in the past decade, as growing numbers of people challenge binary and mutually-exclusive conceptualizations of female and male, feminine and masculine. For this reason, it is important to define these terms from the outset.

Sex is assigned to an individual at birth, typically based on his/her biology: anatomy, hormones, and physiology.Note In contrast, gender is a social category, entailing roles, behaviours, activities, and attributes that a society deems to be appropriate for a given sex.Note Gender refers to the social construction of female and male as oppositional categories with distinct characteristics and unequal social value and power.Note Individuals develop gendered selves by internalizing, through socialization beginning in childhood, social norms and cultural expectations regarding appropriate roles, behaviours, activities, and appearances for their sex.Note In their social interactions, individuals perform sex-appropriate gender roles (i.e., they “do gender”), thereby creating and sustaining the prevailing gender order.

There are also gender identity and gender expression. Gender identity refers to personal identification of one’s gender according to an internal awareness that can fall anywhere on the gender spectrum between female and male.Note Gender expression pertains to how an individual outwardly expresses or presents his/her gender.Note

Although sex, gender, gender identity, and gender expression are conceptually distinct, the terms tend to be conflated based on the understanding that most biological women are female/feminine, while most biological men are male/masculine (i.e., cisgender).Note In other words, gender, gender identity, and gender expression are equated with sex because they are associated, and generally align, with that biological base.Note This is necessary when it comes to measuring the gender pay gap, as information relevant to making distinctions between sex, gender, gender identity, and gender expression has not previously been collected. For the purposes of this paper, then, women and men are distinguished on the basis of sex, with the assumption that their gender, gender identity, and gender expression are consistent with expectations for people of their sex (i.e., they are cisgender).Note Note Importantly, Statistics Canada recently established new sex and gender standards (2018), which distinguish between sex assigned at birth and gender identity or expression. In alignment with these new standards, a new gender-inclusive question has been recently implemented on national surveys, including the 2019 Census of Population content test, allowing transgender persons and those with non-binary gender identities or expressions to report their gender.

Fin de l’encadré

2. Measuring the gender pay gap

Relative earnings often signify how different groups are valued socially and economically.Note For this reason, the unadjusted gender pay gap—the raw difference between the earnings acquired by women and men through their paid work, which favours the latter—has often been used as a call to action for gender equality and women’s economic empowerment. Not all women earn less than men; in fact, women’s representation among high-income earners has increased: 20.3% of high-income earners were women in 2016, compared with 8.5% in 1981.Note Note However, among those with earnings, most women earn less than men (Table 1). For this reason, women’s average annual earnings from wages, salaries, and commissions were lower than men’s in 2017: $40,600 vs. $59,200 (a difference of $18,600).


Table 1
Annual wages, salaries and comissions of employed women and men aged 16 and older, 2017
Table summary
This table displays the results of Annual wages Women, Men and Difference (% women - % men), calculated using percent and dollars units of measure (appearing as column headers).
Women Men Difference (% women - % men)
percent
Under $5,000 (excluding zero earnings and losses) 8.1 6.4 1.7
$5,000 to $9,999 8.4 5.8 2.6
$10,000 to $19,999 15.6 10.4 5.2
$20,000 to $29,999 13.6 9.3 4.3
$30,000 to $39,999 12.7 10.2 2.5
$40,000 to $49,999 11.5 10.4 1.1
$50,000 to $59,999 7.8 9.2 -1.4
$60,000 to $79,999 11.0 14.3 -3.3
$80,000 to $99,999 6.6 9.6 -3.0
$100,000 and over 4.8 14.3 -9.5
Total 100.0 100.0 Note ...: not applicable
dollars
AverageTable 1 Note 1 40,600.00 59,200.00 -18,600.00
Median 33,400.00 47,500.00 -14,100.00

There are two ways of presenting the gender pay gap, whether unadjusted or adjusted, both of which are based on the gender pay ratio—that is, women’s earnings expressed as a proportion of men’s earnings. The gender pay ratio is calculated by dividing women’s earnings by men’s earnings (see Calculating the Gender Pay Gap). It can be interpreted as the number of cents that women earn for every dollar earned by men. Alternatively, the gender pay ratio can be subtracted from one and then multiplied by 100, such that it represents how much less women earn than men in percentage terms.

Début de l’encadré

Text box 2
Calculating the Gender Pay Gap

G e n d e r   p a y   r a t i o = W o m e n s   a v e r a g e   o r   m e d i a n   e a r n i n g s M e n s   a v e r a g e   o r   m e d i a n   e a r n i n g s MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaacbmaeaaaaaaaaa8qacqWFhbWrcqWFLbqzcqWFUbGB cqWFKbazcqWFLbqzcqWFYbGCcqWFGcaOcqWFWbaCcqWFHbqycqWF5b qEcqWFGcaOcqWFYbGCcqWFHbqycqWF0baDcqWFPbqAcqWFVbWBcqGH 9aqpdaWcaaWdaeaapeGaem4vaCLaem4Ba8MaemyBa0MaemyzauMafm OBa42dayaafaWdbiabdohaZjabcckaGkabdggaHjabdAha2jabdwga LjabdkhaYjabdggaHjabdEgaNjabdwgaLjabcckaGkabd+gaVjabdk haYjabcckaGkabd2gaTjabdwgaLjabdsgaKjabdMgaPjabdggaHjab d6gaUjabcckaGkabdwgaLjabdggaHjabdkhaYjabd6gaUjabdMgaPj abd6gaUjabdEgaNjabdohaZbWdaeaapeGaemyta0KaemyzauMafmOB a42dayaafaWdbiabdohaZjabcckaGkabdggaHjabdAha2jabdwgaLj abdkhaYjabdggaHjabdEgaNjabdwgaLjabcckaGkabd+gaVjabdkha YjabcckaGkabd2gaTjabdwgaLjabdsgaKjabdMgaPjabdggaHjabd6 gaUjabcckaGkabdwgaLjabdggaHjabdkhaYjabd6gaUjabdMgaPjab d6gaUjabdEgaNjabdohaZbaaaaa@ACCF@ G e n d e r   p a y   g a p = ( 1 G e n d e r   p a y   r a t i o ) * 100 MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaacbmaeaaaaaaaaa8qacqWFhbWrcqWFLbqzcqWFUbGB cqWFKbazcqWFLbqzcqWFYbGCcqWFGcaOcqWFWbaCcqWFHbqycqWF5b qEcqWFGcaOcqWFNbWzcqWFHbqycqWFWbaCcqGH9aqpdaqadaWdaeaa peGaeGymaeJaeyOeI0Iaem4raCKaemyzauMaemOBa4MaemizaqMaem yzauMaemOCaiNaeiiOaaQaemiCaaNaemyyaeMaemyEaKNaeiiOaaQa emOCaiNaemyyaeMaemiDaqNaemyAaKMaem4Ba8gacaGLOaGaayzkaa GaeiOkaOIaeGymaeJaeGimaaJaeGimaadaaa@6CB6@ = ( ( M e n s   a v e r a g e   o r   m e d i a n   e a r n i n g s w o m e n s   a v e r a g e   o r   m e d i a n   e a r n i n g s ) M e n s   a v e r a g e   o r   m e d i a n   e a r n i n g s ) * 100 MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqGH9aqpdaqadaWdaeaapeWaaSaa a8aabaWdbmaabmaapaqaa8qacqWGnbqtcqWGLbqzcuWGUbGBpaGbau aapeGaem4CamNaeiiOaaQaemyyaeMaemODayNaemyzauMaemOCaiNa emyyaeMaem4zaCMaemyzauMaeiiOaaQaem4Ba8MaemOCaiNaeiiOaa QaemyBa0MaemyzauMaemizaqMaemyAaKMaemyyaeMaemOBa4MaeiiO aaQaemyzauMaemyyaeMaemOCaiNaemOBa4MaemyAaKMaemOBa4Maem 4zaCMaem4CamNaeyOeI0Iaem4DaCNaem4Ba8MaemyBa0MaemyzauMa fmOBa42dayaafaWdbiabdohaZjabcckaGkabdggaHjabdAha2jabdw gaLjabdkhaYjabdggaHjabdEgaNjabdwgaLjabcckaGkabd+gaVjab dkhaYjabcckaGkabd2gaTjabdwgaLjabdsgaKjabdMgaPjabdggaHj abd6gaUjabcckaGkabdwgaLjabdggaHjabdkhaYjabd6gaUjabdMga Pjabd6gaUjabdEgaNjabdohaZbGaayjkaiaawMcaaaWdaeaapeGaem yta0KaemyzauMafmOBa42dayaafaWdbiabdohaZjabcckaGkabdgga HjabdAha2jabdwgaLjabdkhaYjabdggaHjabdEgaNjabdwgaLjabcc kaGkabd+gaVjabdkhaYjabcckaGkabd2gaTjabdwgaLjabdsgaKjab dMgaPjabdggaHjabd6gaUjabcckaGkabdwgaLjabdggaHjabdkhaYj abd6gaUjabdMgaPjabd6gaUjabdEgaNjabdohaZbaaaiaawIcacaGL PaaacqGGQaGkcqaIXaqmcqaIWaamcqaIWaamaaa@CA82@

In 2017, the average hourly wages of all female and male workers aged 15 years and over in Canada were $24.28 and $28.00, respectively.

2017   g e n d e r   p a y   ( w a g e )   r a t i o = $ 24.28 $ 28.00 = 0.87 MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqaIYaGmcqaIWaamcqaIXaqmcqaI 3aWnieWacqWFGcaOcqWFNbWzcqWFLbqzcqWFUbGBcqWFKbazcqWFLb qzcqWFYbGCcqWFGcaOcqWFWbaCcqWFHbqycqWF5bqEcqWFGcaOdaqa daWdaeaapeGae83DaCNae8xyaeMae83zaCMae8xzaugacaGLOaGaay zkaaGae8hOaaQae8NCaiNae8xyaeMae8hDaqNae8xAaKMae83Ba8Ma eyypa0ZaaSaaa8aabaWdbiabcscaKiabikdaYiabisda0iabc6caUi abikdaYiabiIda4aWdaeaapeGaeiijaqIaeGOmaiJaeGioaGJaeiOl a4IaeGimaaJaeGimaadaaiabg2da9iabicdaWiabc6caUiabiIda4i abiEda3aaa@709D@ 2017   g e n d e r   p a y   ( w a g e )   g a p = ( 1 0.87 ) * 100 = 13 % MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqaIYaGmcqaIWaamcqaIXaqmcqaI 3aWnieWacqWFGcaOcqWFNbWzcqWFLbqzcqWFUbGBcqWFKbazcqWFLb qzcqWFYbGCcqWFGcaOcqWFWbaCcqWFHbqycqWF5bqEcqWFGcaOdaqa daWdaeaapeGae83DaCNae8xyaeMae83zaCMae8xzaugacaGLOaGaay zkaaGae8hOaaQae83zaCMae8xyaeMae8hCaaNaeyypa0ZaaeWaa8aa baWdbiabigdaXiabgkHiTiabicdaWiabc6caUiabiIda4iabiEda3a GaayjkaiaawMcaaiabcQcaQiabigdaXiabicdaWiabicdaWiabg2da 9iabigdaXiabiodaZiabcwcaLaaa@6C4E@ 2017   g e n d e r   p a y   ( w a g e )   g a p = ( ( $ 28.00 $ 24.28 ) $ 28.00 ) * 100 = ( $ 3.72 $ 28.00 ) * 100 = 13 %   MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqaIYaGmcqaIWaamcqaIXaqmcqaI 3aWnieWacqWFGcaOcqWFNbWzcqWFLbqzcqWFUbGBcqWFKbazcqWFLb qzcqWFYbGCcqWFGcaOcqWFWbaCcqWFHbqycqWF5bqEcqWFGcaOdaqa daWdaeaapeGae83DaCNae8xyaeMae83zaCMae8xzaugacaGLOaGaay zkaaGae8hOaaQae83zaCMae8xyaeMae8hCaaNaeyypa0ZaaeWaa8aa baWdbmaalaaapaqaa8qadaqadaWdaeaapeGaeiijaqIaeGOmaiJaeG ioaGJaeiOla4IaeGimaaJaeGimaaJaeyOeI0IaeiijaqIaeGOmaiJa eGinaqJaeiOla4IaeGOmaiJaeGioaGdacaGLOaGaayzkaaaapaqaa8 qacqGGKaajcqaIYaGmcqaI4aaocqGGUaGlcqaIWaamcqaIWaamaaaa caGLOaGaayzkaaGaeiOkaOIaeGymaeJaeGimaaJaeGimaaJaeyypa0 ZaaeWaa8aabaWdbmaalaaapaqaa8qacqGGKaajcqaIZaWmcqGGUaGl cqaI3aWncqaIYaGma8aabaWdbiabcscaKiabikdaYiabiIda4iabc6 caUiabicdaWiabicdaWaaaaiaawIcacaGLPaaacqGGQaGkcqaIXaqm cqaIWaamcqaIWaamcqGH9aqpcqaIXaqmcqaIZaWmcqGGLaqjcqGGGc aOaaa@8C96@

Fin de l’encadré

This section discusses how implicit or explicit decisions that are made when measuring the unadjusted gender pay gap affect its size, and therefore its perceived seriousness:

  1. Who estimates are based upon? In other words, which workers are included in the sample from which the gender pay gap is estimated? Key considerations in this regard are:
    1. Age
    2. Labour supply (i.e., hours and weeks worked)
  2. What type of earnings are counted (i.e., annual earnings or hourly wages)?
  3. How “typical” earnings are calculated (i.e., average or median)? Note Note

In addition to affecting the size of the gender pay gap, these decisions are important because they can either justify or challenge the status quo of gender differences in employment and earnings.

2.1. Who are estimates based upon?

For the purposes of measuring the gender pay gap, workers are generally defined as paid employees in their main job. Self-employed individuals are excluded because they typically work for themselves as a business owner, freelancer, or independent contractor for another company. Earnings of self-employed individuals usually come directly from the business, instead of from wages, salary or commission-based reimbursement, and they are therefore not comparable to earnings from paid employment.

Ideally, individuals who were enrolled as full-time students during the reference period or the past year would also be excluded, as work was not their primary activity.

Other considerations when defining the sample upon which estimates of the gender pay gap are based—namely, age group and labour supply (i.e., hours and weeks worked)—are discussed below.

2.1.1. Age group

Empirical studies of earnings often focus on a segment of the employed adult population in what are considered to be the core working ages, generally defined as 25 to 54. The logic of focusing on the core-working-age population is that younger and older adults are often full-time students or retired, respectively. To the extent that younger and older workers are systematically different from workers in the core-working ages (e.g., they have higher or lower socioeconomic status), excluding them from the analytical sample may bias estimates of the gender pay gap. “Selection bias” is discussed in more detail in Section 3.4.2.

In this paper, data pertain to the employed population aged 15 years and older, unless otherwise indicated, for comparability with international indicators of the gender pay gap. However, it is recognized that the gender pay gap differs by age group. As discussed in Section 3.4.3, earnings inequality between women and men tends to increase with age, both because women experience more employment interruptions related to family responsibilities than men, and because older women generally have lower levels of educational attainment and work experience than younger women (i.e., “cohort effects”). For these reasons, the gender pay gap is narrower among those aged 15/16 to 24 than it is among those aged 25 to 54 and 55 years and older.

Comparison of the average earnings of women and men aged 15/16 years and older and 25 to 54 over the past 30 years demonstrates that the gender pay gap is virtually the same for both age groups, whether measured in terms of annual earnings from wages, salaries and commissions or hourly wages (Chart 1). Given that gender-based earnings inequality tends to increase with age, as the workforce ages (due to both the aging of Canada’s population and the increasing labour force participation of older age groups, estimates of the gender pay gap based on those aged 15/16 years and older and those aged 25 to 54 may eventually diverge.Note

Chart 1

Data table for Chart 1 
Data table for Chart 1
Table summary
This table displays the results of Data table for Chart 1 Average annual earnings, workers aged 16 years and older, Average annual earnings, workers aged 25 to 54, Average hourly wages, workers aged 15 years and older and Average hourly wages, workers aged 25 to 54, calculated using gender pay ratio (women's earnings : men's earnings) units of measure (appearing as column headers).
Average annual earnings, workers aged 16 years and older Average annual earnings, workers aged 25 to 54 Average hourly wages, workers aged 15 years and older Average hourly wages, workers aged 25 to 54
gender pay ratio (women's earnings : men's earnings)
1997 0.63 0.63 0.82 0.81
1998 0.63 0.63 0.81 0.81
1999 0.60 0.60 0.81 0.81
2000 0.60 0.60 0.81 0.80
2001 0.62 0.62 0.81 0.81
2002 0.62 0.64 0.82 0.82
2003 0.61 0.62 0.82 0.82
2004 0.63 0.64 0.83 0.83
2005 0.64 0.64 0.84 0.84
2006 0.66 0.67 0.84 0.84
2007 0.65 0.66 0.84 0.84
2008 0.66 0.66 0.84 0.84
2009 0.69 0.70 0.85 0.85
2010 0.69 0.69 0.85 0.85
2011 0.68 0.68 0.86 0.86
2012 0.68 0.67 0.86 0.86
2013 0.69 0.70 0.86 0.86
2014 0.68 0.69 0.86 0.86
2015 0.68 0.69 0.86 0.86
2016 0.68 0.68 0.86 0.86
2017 0.69 0.69 0.87 0.87

2.1.2. Labour supply

It is well established that women and men differ with respect to characteristics that determine earnings from employment, particularly work hours. According to data from the Labour Force Survey, employed women aged 15 years and older usually worked an average of 32.9 hours per week at their main job in 2017—5.5 fewer hours than employed men.Note Also, employed men were three times as likely as employed women to work long work hours—defined as 50 hours or more per week—at their main job (11.0% vs. 3.6%) .

Based on data from the 2016 General Social Survey on Canadians at Work and Home, employed women and men aged 15 years and older both worked an average of 44 weeks at all jobs in the past year.Note However, women were less likely than men to be employed on a full-time, full-year basis, defined as mostly working 30 or more hours per week for 49 to 52 weeks in a given year (43.6% vs. 56.4%).

Although women spend less time, on average, than men on paid work, they spend more time on housework, childrearing, and caregiving.Note Based on data from the 2015 General Social Survey on Time Use, women aged 15 years and older spent an average of 25.2 hours per week on domestic labour—8.4 hours more than men.Note Even when women were employed on a full-time basis, they spent more time on domestic labour than their male counterparts: 21.0 vs. 16.1 hours per week (a difference of nearly 5 hours).

Estimates of the gender pay gap can be based on either a sample consisting of all workers, regardless of hours and weeks worked, or a sample consisting of workers employed on a full-time, full-year basis. These samples imply different conceptualizations of women’s decision-making about their labour supply.

One perspective, coming from economics, is that individuals are rational, self-interested, and preference or “utility” maximizing, meaning that they will always select the course of action from which they derive the most satisfaction, given their relatively stable preferences.Note Note  According to this perspective, women are biologically predisposed to place greater value on family than men and they are more efficient at performing housework, childrearing, and caregiving. For these reasons, women “choose” to perform less, if any, paid work, focusing instead on unpaid work.Note Viewed from this perspective, earnings inequality between women and men that is related to gender differences in labour supply is justified as the outcome of personal autonomy.Note It therefore makes sense to limit the sample from which the gender pay gap is estimated to full-time, full-year workers.

An alternative perspective, coming from sociology, is that individuals’ choices are constrained by social norms and cultural expectations related to gender, and the gendered structural conditions of everyday life. Different attitudes, appearances, and behaviours are communally deemed to be appropriate for biologically-based sex categories.Note Individuals are held morally responsible for “doing gender”—that is, conducting themselves in ways that are consistent with and demonstrate their sex category—lest their competence as members of society be questioned.Note When women specialize in housework, childrearing, and caregiving and men specialize in breadwinning, according to this perspective, they are doing gender.Note Individuals are also doing gender when they select the “right” field of study and occupation for their sex.Note Note Note Note Note Note Note

According to previous research, the gender division of labour is both reflected and reinforced in the labour market through employers’ discriminatory recruitment and promotion practices, which block women from stereotypically “male” occupations and limit their career advancement; gender-based pay inequality that makes it economically disadvantageous for couples to divide domestic labour and paid work more equitably; and work organizations’ preferences for employees who can prioritize paid work due to their limited family responsibilities (usually men).Note Note Note Note To the extent that women expect to experience gender-based discrimination and inequality in the labour market, it may affect their decisions regarding how much and which type of education and training to invest in and which occupations to pursue. Studies show that, through these feedback effects, the gender division of labour becomes self-reinforcing: women have lower-value human capital (e.g., education) and lower-status occupations, and they therefore earn less, which reduces their bargaining position relative to men in the household, making them more likely to be the one who performs a greater share of housework, childrearing, and caregiving.Note

Previous research supports this interpretation of women’s decisions about their labour supply, as it demonstrates that women’s employment status and work hours are negatively affected by their ideological support for gender-differentiated roles.Note Note Note Note Note Further, Pedualla and Thébaud find that the preferences of young, unmarried, childless women and men in the United States regarding their future gender division of labour are responsive to institutional constraintsNote . Specifically, women and men generally prefer egalitarian relationships, but, in the absence of supportive work-family policies, women and men with higher levels of education and men with lower levels of education have a “fallback plan” of neo-traditional relationships (i.e., the female spouse/partner is primarily responsible for managing the household, while the male spouse/partner is primarily responsible for breadwinning).Note Less educated women have a fall-back plan of self-reliance (i.e., themselves as the primary breadwinner).Note

Under these circumstances, women (and men) make constrained choices about how much paid work they do.Note Estimates of the gender pay gap that are based on samples that are limited to full-time, full-year workers omit the gendered processes involved in women's decisions about their labour supply.

2.2. What type of earnings are counted?

There are three main ways of counting earnings: (1) the annual earnings of all workers; (2) the annual earnings of workers employed on a full-time, full-year basis; and (3) the hourly wages of all workers.Note Annual earnings are defined here as the sum of wages, salaries and commissions from all jobs in a given year before taxes among individuals whose main job was as a paid employee. Employees with zero annual earnings are excluded. Hourly wages are defined as the rate that an employer pays an employee per hour worked in his/her main job.

Three limitations characterize all three ways of counting earnings. First, earnings are only measured for individuals in paid employment. If employed women and men differ systematically from their counterparts who are not employed with respect to characteristics that affect earnings, the gender pay gap will be biased, irrespective of what type of earnings are counted (see Section 3.4.1).

Second, earnings derived from (self-)employment in the underground economy are excluded from all three measures. By definition, economic activities within the underground economy escape measurement because they are unreported, hidden or illegal. In 2016, the underground economy accounted for 2.5% (or $51.6 billion) of Canada’s Gross Domestic Product.Note It is unclear whether women and men in Canada participate equally in the underground economy, where workers are often poorly paid and lack access to promotional opportunities, legal protections, and employment benefits, because important illegal industries—namely, drugs and sex work—are excluded from national estimates.Note Note Also, national estimates of the underground economy rely on assumptions, weak indicative information, and various indirect methods because it is difficult to obtain information on hidden, illegal, and informal activities.Note With these caveats in mind, the three industries that accounted for more than half of measured underground economic activity in Canada in 2016 were residential construction (26.6%), retail trade (13.5%), and accommodation and food services (12.1%).Note These industries have been the main contributors since 1992, the first year of study.Note

Third, other sources of compensation related to employment, such as (supplementary) medical and dental benefits, pension plans, stock options, and use of company cars, are not counted as earnings. Previous research suggests that women may have less access to some of these employer-provided perks, relative to men, because they are more likely to work part-time, and they are less likely to hold higher-status occupations and work for large corporations.Note Note Note However, employed women in Canada were somewhat more likely than men to be covered by employer-sponsored pension plans in 2015 (40% vs. 36%), due to the fact that they are overrepresented in the three industries with the highest rates of pension coverage: education, health, and public administration.Note Consequently, women may have more discretionary income than men after saving for retirement—even with the same earnings.Note Gender inequality in non-monetary employee compensation is not reflected in any estimate of the gender pay gap that is currently available.

2.2.1 Comparing the annual earnings of all women and men, regardless of how much they work

The most inclusive, and therefore the largest, measure of the gender pay gap involves comparison of the annual earnings of all employed women and men, including those who worked on a part-time and/or part-year basis (Chart 2). By implication, this measure captures gender differences in both pay (i.e., the price of labour) and hours and weeks worked (i.e., labour supply).Note It can therefore be thought of as an unrestricted measure of the gender pay gap.

Chart 2

Data table for Chart 2 
Data table for Chart 2
Table summary
This table displays the results of Data table for Chart 2 Average annual earnings, all workers, Average annual earnings, full-time, full-year workers, Average hourly wages, all workers and Average hourly wages, full-time workers only, calculated using gender pay ratio (women's earnings : men's earnings) units of measure (appearing as column headers).
Average annual earnings, all workers Average annual earnings, full-time, full-year workers Average hourly wages, all workers Average hourly wages, full-time workers only
gender pay ratio (women's earnings : men's earnings)
1997 0.63 0.69 0.82 0.83
1998 0.63 0.70 0.81 0.83
1999 0.60 0.67 0.81 0.83
2000 0.60 0.69 0.81 0.82
2001 0.62 0.69 0.81 0.82
2002 0.62 0.68 0.82 0.83
2003 0.61 0.69 0.82 0.84
2004 0.63 0.69 0.83 0.85
2005 0.64 0.70 0.84 0.86
2006 0.66 0.72 0.84 0.85
2007 0.65 0.71 0.84 0.85
2008 0.66 0.72 0.84 0.85
2009 0.69 0.74 0.85 0.87
2010 0.69 0.73 0.85 0.87
2011 0.68 0.73 0.86 0.88
2012 0.68 0.73 0.86 0.88
2013 0.69 0.73 0.86 0.88
2014 0.68 0.73 0.86 0.88
2015 0.68 0.72 0.86 0.88
2016 0.68 0.73 0.86 0.88
2017 0.69 0.73 0.87 0.89

The typical criticism of the unrestricted measure of the gender pay gap is that it is “confounded” by gender differences in hours and weeks worked. The counterargument is that the unrestricted measure of the gender pay gap captures the full scope of the financial implications of gender, which partly result from women’s reduced labour supply, relative to men, given their greater family responsibilities.

Annual earnings have an advantage over hourly wages when it comes to conveying the economic significance of the gender pay gap, as it speaks to the women’s and men’s different command over goods and services. As Lips explains: “When a woman applies for a mortgage or a car loan, she is not asked about her hourly income. The income statistic that affects whether or not she gets the loan, and indeed what kind of life she is able to afford, is her annual income.”Note Measuring the gender pay gap in terms of annual earnings provides an index of gender differences in purchasing power and material well-being that the gender wage gap does not. In other words, the gender pay gap, as measured from annual earnings, speaks to women’s overall economic well-being.

2.2.2. Comparing the annual earnings of women and men employed on a full-time, full-year basis

Traditionally, the gender pay gap has been measured from the annual earnings of women and men who worked full-time, defined as a minimum of 35 hours per week, for 50 to 54 weeks in the reference year. Measured as such, the gender pay gap will be smaller than it would be if workers employed a part-time and/or part-year basis—who are predominantly women—are included in the estimate (Chart 2). However, gender differences in work hours are only partially removed when the gender pay gap is measured from the annual earnings of full-time, full-year workers. The reason is that women work fewer hours each week than men, on average, even when they are employed on full-time, full-year. Based on data from the Labour Force Survey, women aged 15 years and older who were employed on a full-time basis usually worked an average of 38.5 hours per week at their main job in 2017, while their male counterparts usually worked an average of 41.5 hours (a difference of three hours per week).

The exclusion of workers employed on a part-time and/or part-year basis from measurement of the gender pay gap implies the idealization of full-time, full-year employment—a pattern typified by men, both historically and currently.Note Studies show that the male pattern of employment itself depends on housework, childrearing, and caregiving disproportionately performed by women.Note Note Note Requiring full-time, full-year employment can therefore be construed as marginalizing mothers, who often experience work interruptions related to childrearing, and other women for whom it is impractical to fit that pattern, due to their greater family responsibilities, relative to men.Note Note Further, it renders invisible domestic labour, its financial cost to women, and its benefits to the economy.Note

2.2.3 Comparing the hourly wages of women and men

Although there is no agreement internationally on a standard way of counting earnings for the purposes of measuring the gender pay gap, there is a strong preference for hourly wages. The ratio of women’s hourly wages to men’s hourly wages—known as the gender wage gap—is the most restricted measure of the gender pay gap, as it captures only the per-unit price of labour, and it is therefore unaffected by gender differences in labour supply.Note For this reason, hourly wages yield the smallest estimate of the gender pay gap (Chart 2). For example, between 1997 and 2017, the gender wage ratio in Canada ranged from 0.81 to 0.89, while the earnings-based gender pay ratio (all workers) ranged from 0.60 to 0.74. In many years the gender wage ratio was higher than the earnings-based gender pay ratio by nearly 20 percentage points.

In theory, it is not necessary to exclude part-time workers when estimating the gender wage gap in order to control for gender differences in work hours, as labour supply is irrelevant when earnings are measured on a per-unit basis. Indeed, the move from all workers to full-time workers makes little difference when the gender pay gap is measured in terms of hourly wages, as opposed to annual earnings (Chart 2).

The fact that the gender wage gap is not confounded by gender differences in labour supply can be viewed as a drawback. Treating gender differences in work hours as a separate issue from the gender pay gap rests on the assumption that women freely choose to work fewer hours than men, with the implication being that their lower earnings are deserved. In this way, the gendered processes that give rise to women’s reduced labour supply, relative to men, and the contribution of those processes to the gender wage gap are hidden from view.

Many workers are not paid on an hourly basis; instead, they are paid on the basis of an annual salary.Note According to the Labour Force Survey, 41.4% of women aged 15 years and older were salaried employees in 2017, as were 47.1% of men. For salaried employees, estimates of hourly wages are generally made by either assuming a standard 40 hour work week, or dividing their annual salary by their usual weekly work hours multiplied by the number of weeks in a year. However, 16.0% of salaried employees reported that their work hours varied from one week to the next in 2017, and 13.6% of salaried employees reported usual work hours in excess of 40 per week. It follows that, for many workers, estimates of hourly wages on the basis of an annual salary and usual work hours will be less accurate.

Notably, hourly wages do not reflect contingent or “pay-for-performance” that is included in annual earnings.Note Women are less likely than men to receive contingent pay for a number of reasons.Note First, they are less likely to be employed in leadership positions and higher-status occupations, particularly in the private sector, that offer contingent pay.Note  Second, negotiation plays a greater role in determining bonuses and stock options than it does in fixed salaries. Note Note  Women are less likely than men to negotiate more favorable employment terms with respect to base salary, stock options, and bonuses.Note Note Third, women tend to have lower risk tolerance in financial decision-making than men, with the implication that they are less likely to take jobs in which a significant component of compensation is dependent on performance.Note

In addition to women being less likely than men to receive contingent pay, previous research demonstrates that the gender pay gap is more pronounced for contingent pay than fixed salaries, possibly because the latitude for management discretion is greater in the case of contingent pay.Note Note Note Note For these reasons, the gender wage gap may underestimate to some extent earnings inequality between women and men.

2.3. How typical earnings are counted: Mean or median?

In addition to which workers are included in the samples from which the gender pay gap is estimated and what type of earnings are counted, the size of the gender pay gap depends on the statistic chosen to represent “typical” earnings: the mean (more commonly known as the average) or the median. To calculate the mean, the earnings of all workers of a given sex are summed and the resulting value is divided by the number of workers of that sex. The mean is interpreted as the pay that each worker of a given sex would receive, if the earnings of all workers of that sex were divided evenly among them.

To calculate the median, the earnings of all workers of a given sex are put in ascending/descending order, and the middle of that earnings distribution is identified. The median separates the half of workers who earn less than that amount from the other half of workers who earn more.

Averages have the advantage of taking into account the earnings of every worker, which medians do not. The downside of this characteristic is that averages can be skewed by workers with very low or very high earnings to a greater extent than medians.Note At the same time, it can be argued that the overrepresentation of women among workers with very low earnings, and the underrepresentation of women among workers with very high earnings, are central to the issue of gender-based pay inequality, and should therefore be reflected in the statistic used to measure central tendency.Note Further, previous research demonstrates that the gender pay gap increases over the earnings distribution, being greater at the top than the bottom.Note Note Note Note Note Note Note Note Measuring the gender pay gap at the median minimizes the impact of the segment of the workforce with the highest earnings.Note

Another advantage of the mean, over the median, is that the multivariate methods of data analysis that are typically used to examine the gender pay gap—ordinary least squares regression and standard decomposition techniques—are based on the mean.Note  

Chart 3 shows the gender pay ratio calculated from average and median annual earnings (all workers) and hourly wages. Notably, since the late 2000s, the average and median gender pay ratios have generally been more similar when annual earnings are used, as opposed to their hourly wages.

Chart 3

Data table for Chart 3 
Data table for Chart 3
Table summary
This table displays the results of Data table for Chart 3 Average annual earnings, all workers, Median annual earnings, all workers, Average hourly wages, all workers and Median hourly wages, all workers, calculated using gender pay ratio (women's earnings : men's earnings) units of measure (appearing as column headers).
Average annual earnings, all workers Median annual earnings, all workers Average hourly wages, all workers Median hourly wages, all workers
gender pay ratio (women's earnings : men's earnings)
1997 0.63 0.60 0.82 0.79
1998 0.63 0.60 0.81 0.78
1999 0.60 0.58 0.81 0.80
2000 0.60 0.58 0.81 0.77
2001 0.62 0.60 0.81 0.80
2002 0.62 0.60 0.82 0.80
2003 0.61 0.61 0.82 0.80
2004 0.63 0.61 0.83 0.83
2005 0.64 0.63 0.84 0.81
2006 0.66 0.64 0.84 0.82
2007 0.65 0.63 0.84 0.82
2008 0.66 0.64 0.84 0.84
2009 0.69 0.69 0.85 0.84
2010 0.69 0.69 0.85 0.83
2011 0.68 0.68 0.86 0.84
2012 0.68 0.65 0.86 0.85
2013 0.69 0.71 0.86 0.84
2014 0.68 0.69 0.86 0.87
2015 0.68 0.68 0.86 0.83
2016 0.68 0.69 0.86 0.85
2017 0.69 0.70 0.87 0.84

2.4. Comparison of different measures over time

Different measures of the gender pay gap yield different stylized facts regarding the evolution of earnings inequality between women and men in Canada over time, and different implications for future progress. Based on the average annual earnings of full-time, full-year workers, the prevailing narrative highlights the fact that the gender pay ratio hovered around 0.70 between 1992 and 2008 (Chart 4).Note This contrasts with the preceding 15 years, during which time there was a steady narrowing of the gender pay gap (from 0.39 in 1976 to 0.32 in 1991).Note Since 2009, the gender pay ratio has been marginally higher than that observed between 1992 and 2008, averaging 0.73. Given similar findings in the United States, some scholars wonder whether progress toward equal pay between women and men has stalled.Note

Chart 4

Data table for Chart 4 
Data table for Chart 4
Table summary
This table displays the results of Data table for Chart 4 Gender pay ratio (average annual earnings, full-time, full-year workers) and Gender wage ratio (average hourly wages, all workers), calculated using gender pay ratio (women's earnings : men's earnings) units of measure (appearing as column headers).
Gender pay ratio (average annual earnings, full-time, full-year workers) Gender wage ratio (average hourly wages, all workers)
gender pay ratio (women's earnings : men's earnings)
1976 0.61 0.68
1977 0.63 0.73
1978 0.63 0.71
1979 0.64 0.74
1980 0.64 0.73
1981 0.63 0.76
1982 0.64 0.75
1983 0.64 0.75
1984 0.65 0.78
1985 0.65 0.77
1986 0.66 0.78
1987 0.66 0.79
1988 0.65 0.77
1989 0.66 0.80
1990 0.66 0.77
1991 0.68 0.77
1992 0.70 0.81
1993 0.72 0.84
1994 0.69 0.80
1995 0.73 0.84
1996 0.72 0.81
1997 0.69 0.82
1998 0.70 0.81
1999 0.67 0.81
2000 0.69 0.81
2001 0.69 0.81
2002 0.68 0.82
2003 0.69 0.82
2004 0.69 0.83
2005 0.70 0.84
2006 0.72 0.84
2007 0.71 0.84
2008 0.72 0.84
2009 0.74 0.85
2010 0.73 0.85
2011 0.73 0.86
2012 0.73 0.86
2013 0.73 0.86
2014 0.73 0.86
2015 0.72 0.86
2016 0.73 0.86
2017 0.73 0.87

Using the hourly wages of all female and male workers, instead of annual earnings, reveals that the pace of growth in the gender pay ratio slowed considerably between 1992 and 2000, but did not stall.Note The gender pay ratio picked up again in 2001, signalling the resumption of a strong, long-term trend towards convergence in women’s and men’s pay.

Début de l’encadré

Text box 3
Selecting a measure of the gender pay gap

Although all of the previously-discussed measures of the gender pay gap are meaningful, each one approaches the issue from a different angle, and is therefore more or less informative, depending on the question/s being addressed.  For example, pay equity legislation mandates equal pay for work of equal or comparable value to the employer, regardless of employees’ work hours and terms of employment (i.e., permanent, temporary, seasonal, or casual). It follows that the gender pay gap in hourly wages is more relevant to pay equity legislation than the gender pay gap in annual earnings, as the latter is confounded by gender differences in work hours. On the other hand, when used as a reflection of the realities of women’s and men’s lives, the gender pay gap is best measured in terms of the annual earnings of all workers, as doing so encompasses all of the different ways in which gender results in women having lower pay, and ultimately less purchasing power, than men.

For the purposes of monitoring progress toward gender equality, as per the Government of Canada’s Gender Results Framework, it is recommended that two measures of the gender pay gap be presented: one based on the annual earnings of all workers, and the other based on the hourly wages of all workers. Doing so highlights the contribution of gender differences in work hours to the gender pay gap, without implying that women’s route to pay equity with men lies in them adopting stereotypically-male patterns of work and labour-force attachment.

Similar contextual “fit” should guide the use of mean or median. In many cases, it makes sense that the gender pay gap be measured in terms of the mean, as opposed to the median. While the mean can be skewed by workers with either very high or very low earnings, the overrepresentation of women at the bottom of the earnings distribution and their underrepresentation at the top are central to the gender pay gap, as is the widening of the gender pay gap as one moves up the earnings distribution. Also, standard decomposition techniques used to identify the factors affecting the gender pay gap and the magnitude of their effects apply to the mean (see Appendix A). However, when undertaking a more thorough analysis of the gender pay gap, it is recommended that the earnings of women and men be examined at different points along the earnings distribution, including the median (see Appendix A).

Fin de l’encadré

3. Analyzing the gender pay gap

No matter which measure is used, the unadjusted gender pay gap reflects, among other influences or “predictors,” gender differences in education, work experience, job tenure, occupation, and industry, as well as gender-based discrimination. Untangling the mechanisms that give rise to the gender pay gap requires the use of statistical control in the context of ordinary least squares regression or some other multivariate method of data analysis. Statistical control enables us to estimate earnings, holding constant the effects of various predictors. If the gender pay gap is calculated from these earnings, it is “adjusted.” Importantly, the adjusted gender pay gap reflects a hypothetical situation in which female and male workers are equally skilled and evenly distributed across occupations and industries. It should therefore not be interpreted as evidence that the gender pay gap is inconsequential or does not exist.

In this section, explanations for the gender pay gap, and changes in the relative importance of related predictors, are discussed. Considerations when studying trends in the gender pay gap are also presented. Although a brief overview of the statistical technique typically used to analyze the gender pay gap is provided here, readers are referred to Appendix A for more information.

3.1. Traditional economic explanations for the gender pay gap

Traditional economic explanations for the gender pay gap have largely focused on (1) gender differences in human capital, such as education, work experience and job tenture; (2) occupational/industrial gender segregation; and (3) gender differences in the treatment of similarly-qualified and similarly-positioned workers (i.e., discrimination).Note Note

3.1.1. Gender differences in human capital

Human capital refers to individuals’ stock of skills, knowledge, competencies, and other intangible assets, whether innate or acquired, that increase their productivity and therefore their economic value (i.e., earnings). Although many factors contribute to one’s human capital, education, work experience, and job tenure are among the most influential and amenable to measurement.

In the context of neoclassical economic theories of human capital and rational choice, women’s lower wages, relative to men’s, are understood as a consequence of their lower productivity.Note Women may be less productive in the workplace because of the traditional division of labour within the family, whereby women specialize in housework, childrearing, and caregiving, and men specialize in earning.Note Given these gender roles, women can rationally expect shorter and more discontinuous careers.Note Women therefore have less incentive to invest in market-oriented formal education and on-the-job training; they choose occupations and industries where such investments are less important and where pecuniary penalties for work interruptions are smaller; they avoid jobs that require significant investments in firm-specific skills, as returns to those investments accrue only as long as one remains with the firm; and they put less effort into their work.Note Note Note Also, these theories suggest that women’s human capital depreciates during their relatively frequent employment interruptions.Note Note

Contrary to the hypotheses of the human capital model, women in Canada have sustained a long-term trend toward higher education.Note Based on data from the Census of Population, the proportion of women aged 25 to 34 with at least a bachelor’s degree increased by 27.0 percentage points between 1986 and 2016, from 13.7% to 40.7%.Note  The proportion of men in the same age group with at least a bachelor’s degree also increased, but to a lesser extent: 13.6 percentage points, from 15.5% in 1986 to 29.1% in 2016.

Although recent cohorts of women have surpassed the educational attainment of their male counterparts, gender differences persist in field of study. According to data from the 2016 Census of Population, women constituted the majority of Canadians aged 25 to 34 with at least a bachelor’s degree in all broad fields of study, with the exception of the lucrative engineering, computer science, and mathematics fields (Chart 5).Note

Chart 5

Data table for Chart 5 
Data table for Chart 5
Table summary
This table displays the results of Data table for Chart 5. The information is grouped by Field of study (appearing as row headers), Women and Men, calculated using percent units of measure (appearing as column headers).
Field of study Women Men
percent
Engineering 23.0 77.0
Computer and information science 29.2 70.8
Mathematics and related studies 45.2 54.8
Business and administration 51.4 48.6
Science and science technology 57.5 42.5
Law 59.0 41.0
Medicine, dentistry, optometry and veterinary medicine 60.5 39.5
Pharmacy and related programs 64.5 35.5
Arts and humanities 64.5 35.5
Social and behavioural sciences 67.0 33.0
Health care 78.0 22.0
Education and teaching 80.7 19.3
Nursing 89.0 11.0

Data show, however, that women are more likely than men to experience work absences and interruptions—both long-term, scheduled absences related to childbearing and childrearing, and short-term, sporadic absences related to a child’s illness or a major household appliance in need of repair. Note Note Previous research documents the stigmatization of work absences and interruptions, as well as fewer subsequent promotion opportunities and earnings increases.Note Note

The 2011 General Social Survey on Families tracks respondents’ work history retrospectively, given its interrelationship with family transitions. In the context of this survey, a work interruption is defined as being away from work for more than three months since the respondent first started working for a period of six months or longer. A greater proportion of men aged 15 years and older than women had either no work interruptions (67.3% vs. 31.6%) or one work interruption (32.1% vs. 24.3%). Conversely, more women than men had two or more work interruptions (36.3% vs. 8.3%). Given women’s greater number of work interruptions, relative to men, they had an average of three fewer years of work experience (19.6 vs. 22.7 years).

In addition to having less work experience, women tend to have slightly less job tenure—that is, fewer consecutive months or years working for their current (or most recent) employer. Based on data from the Labour Force Survey, employed women aged 15 years and older had an average of 100.2 months with their current employer in 2017—6.2 months fewer than men.

3.1.2. Occupational/industrial gender segregation

Occupational gender segregation, also known simply as “gender segregation,” refers to the uneven distribution of women and men within and across occupations. It takes two forms—vertical segregation and horizontal segregation—both of which contribute to the gender pay gap. Vertical segregation describes the concentration of women at the bottom of the occupational hierarchy and men at the top, and it is suggestive of the extent to which women face obstacles to career advancement that ultimately prevent them from reaching highly-paid positions (i.e., the proverbial "glass ceiling").Note Note Using data from the Labour Force Survey, occupations (four-digit National Occupational Classification) were sorted in ascending order by average hourly wages. In this context, 14.4% of women aged 15 and older had an occupation with average hourly wages in the bottom 20% of the occupational distribution in 2017, compared with 9.6% of men. On the other hand, 1.6% of women had an occupation with average hourly wages in the top 20% of the occupational distribution in 2017, compared with 2.7% of men.

In addition to women’s underrepresentation in higher-status occupations, the gender pay gap may be greater at the top of the occupational hierarchy than the bottom. Previous research demonstrates that wage inequality between women and men is greater and declined more slowly over time at the top of the wage distribution than the bottom.Note Note Note Note Note Note Note Based on data from the Labour Force Survey, the unadjusted gender pay ratio at the highest end of the wage distribution (i.e., the 95th percentile) was 0.86 in 2017, compared with 0.97 at the lowest end (i.e., the 5th percentile). These findings suggest that, even within higher-status occupations, women may encounter glass ceilings that prevent them from reaching the earnings of the highest-paid men.

Horizontal segregation refers to the concentration of women and men in different occupations and industries. It affects the gender pay gap to the extent that female-dominated occupations and industries are paid less than male-dominated occupations and industries, even when they involve the same skill levels.

Although breadwinning has become a central and enduring role for most women in Canada, their employment often parallels traditional female gender roles.Note In other words, what is typically designated as "women’s work" in the private sphere tends to be designated as such in the public sphere as well.Note

Many women in Canada are employed in traditionally-female occupations—teaching, nursing and related health occupations, social work, clerical or other administrative positions, or sales and services—in which women have been concentrated historically. This is reflected in Table 2, showing the proportion of women and men aged 15 years older in the 20 occupations (of 140, based on the three-digit National Occupational Classification) with the greatest concentration of women in 2017. About 27% of women were employed in these occupations, many of which involve the “5 Cs” of caring, clerical, catering, cashiering, and cleaning. In contrast, 5.2% of men were employed in the 20 occupations with the greatest concentration of women. Conversely, 11.5% of men were in the 20 occupations men in 2017, as were less than one percent of women (Table 3). In addition to illustrating how women’s paid work tends to parallel their unpaid work, these findings demonstrate that women are concentrated in sex-typed occupations to a greater extent than men (27.1% vs. 11.5%).


Table 2
Proportion of women and men aged 15 and older employed in the 20 occupations with the greatest concentration of women in 2017, Canada
Table summary
This table displays the results of Proportion of women and men aged 15 and older employed in the 20 occupations with the greatest concentration of women in 2017. The information is grouped by Occupations (appearing as row headers), Women, Men and Difference (% women - % men), calculated using percent units of measure (appearing as column headers).
Occupations Women Men Difference (% women - % men)
percent
Office administrative assistants - general, legal and medical 2.36 0.15 2.22
Professional occupations in nursing 2.16 0.21 1.95
Home care providers and educational support occupations 1.54 0.15 1.38
Other occupations in personal service 0.55 0.07 0.48
General office workers 2.30 0.30 2.00
Assisting occupations in support of health services 2.07 0.30 1.77
Therapy and assessment professionals 0.44 0.07 0.37
Paraprofessional occupations in legal, social, community and education services 2.90 0.49 2.41
Medical technologists and technicians (except dental health) 0.96 0.17 0.78
Financial, insurance and related administrative support workers 1.39 0.27 1.12
Technical occupations in dental health care 0.24 0.05 0.19
Cashiers 2.30 0.54 1.76
Court reporters, transcriptionists, records management technicians and statistical officers 0.13 0.03 0.10
Finance, insurance and related business administrative occupations 1.36 0.34 1.02
Specialized occupations in personal and customer services 0.72 0.20 0.52
Machine operators and related workers in textile, fabric, fur and leather products processing and manufacturing 0.15 0.04 0.11
Library, correspondence and other clerks 0.27 0.08 0.19
Secondary and elementary school teachers and educational counsellors 2.60 0.81 1.78
Librarians, archivists, conservators and curators 0.09 0.03 0.06
Administrative and regulatory occupations 2.52 0.85 1.67
Total 27.05 5.16 Note ...: not applicable

Table 3
Proportion of women and men aged 15 and older employed in the 20 occupations with the greatest concentration of men in 2017, Canada
Table summary
This table displays the results of Proportion of women and men aged 15 and older employed in the 20 occupations with the greatest concentration of men in 2017. The information is grouped by Occupations (appearing as row headers), Women, Men and Difference (% women - % men), calculated using percent units of measure (appearing as column headers).
Occupations Women Men Difference (% women - % men)
percent
Mine service workers and operators in oil and gas drilling 0.01 0.08 0.08
Central control and process operators in processing and manufacturing 0.01 0.14 0.13
Train crew operating occupations 0.00 0.06 0.05
Contractors and supervisors, industrial, electrical and construction trades and related workers 0.06 1.08 1.01
Other installers, repairers and servicers 0.03 0.49 0.46
Other transport equipment operators and related maintenance workers 0.02 0.29 0.28
Underground miners, oil and gas drillers and related occupations 0.02 0.35 0.33
Logging machinery operators 0.00 0.08 0.08
Contractors and supervisors, mining, oil and gas 0.01 0.19 0.18
Supervisors, logging and forestry 0.00 0.03 0.03
Machining, metal forming, shaping and erecting trades 0.05 1.31 1.26
Masonry and plastering trades 0.02 0.57 0.55
Automotive service technicians 0.04 1.06 1.03
Heavy equipment operators 0.03 0.90 0.87
Electrical trades and electrical power line and telecommunications workers 0.04 1.31 1.27
Machinery and transportation equipment mechanics (except motor vehicle) 0.04 1.49 1.46
Plumbers, pipefitters and gas fitters 0.01 0.62 0.60
Carpenters and cabinetmakers 0.02 1.11 1.09
Other mechanics and related repairers 0.00 0.21 0.21
Crane operators, drillers and blasters 0.00 0.14 0.14
Total 0.40 11.52 Note ...: not applicable

In a similar vein, in 2017, the three industries (of 20, based on the two-digit North American Industry Classification System) with the greatest share of women were (1) health care and social assistance, (2) educational services, and (3) accommodation and food services. The proportion of women aged 15 years and older who worked in these three industries combined was 40.1%, compared with 14.1% of men. By comparison, the three industries with the greatest share of men were (1) construction, (2) forestry, fishing, mining, oil and gas, and (3) manufacturing. The proportion of men who worked in these three industries combined was 23.2%, compared with 5.1% of women. As with occupation, women were concentrated in sex-typed industries to a greater extent than men (40.1% versus 23.2%).

Female-dominated occupations and industries tend to be compensated less than male-dominated ones—even when they involve the same skill level.Note For example, using data from the 2017 Labour Force Survey, the effect of being employed in a female-dominated occupation (at the 4-digit level of the National Occupational Classification) on the (log) hourly wages of workers aged 15 and older was estimated. Workers of both sexes employed in occupations in which 90% of incumbents were female earned an average of $6.60 less per hour than workers employed in occupations in which 10% of incumbents were female ($29.78 vs. $36.38). The negative effect of being employed in a female-dominated occupation on hourly wages remained the same when educational attainment was controlled.

Why do women end up in different occupations than men? There are both “supply (employee)-side” and “demand (employer)-side” processes at play. The former pertains to employees’ tastes and preferences for different kinds of work, while the latter pertains to employers’ greater demand for workers of a certain sex and differential valuation of occupations depending on their gender composition.Note

According to some theories, individuals act in rational ways, and therefore select courses of action that they believe will lead to the most desirable outcome.Note Individuals’ preferences influence the means that they choose to pursue their goals.Note Given social norms and cultural expectations that assign primary responsibility for housework, childrearing, and caregiving to women, and primary responsibility for breadwinning to men, women may place more importance on the non-pecuniary attributes of occupations than men.Note Note Note They may therefore trade off higher earnings for occupations that require less time, energy, and emotional input, in order to conserve these resources for their family responsibilities; occupations that engender greater intrinsic, altruistic, and social rewards; and/or occupations that offer amenities like temporal flexibility that facilitate work-family balance.Note Note Note Note

Gender differences in psychological attributes may also make some occupations more or less attractive to women, relative to men.Note For example, previous research demonstrates that women tend to be more risk averse and less competitive than men.Note Note These psychological attributes may lead women and men to choose occupations that differ with respect to job security, earnings stability, injury and fatality risk, performance evaluation, and competitiveness.Note To the extent that these occupational characteristics are associated with higher earnings, gender differences in risk aversion and competitiveness may contribute to the gender pay gap through choice of occupation.Note

Employers’ conscious or unconscious preferences for employees of a particular sex for certain occupations can also generate and maintain occupational segregation through their effects on hiring and promotion.Note Note However, formal human resource practices and policies may limit the extent to which employers’ discretion, and therefore gender bias come into play in occupational placement and advancement.Note

Why do occupations dominated by women pay less than those dominated by men? The two main alternative accounts of the relationship between gender segregation and gender-based wage inequality are (1) “crowding” and (2) the cultural devaluation of women’s work. Both imply discrimination by employers. The crowding hypothesis posits that the exclusion of women from stereotypically “male” occupations through employers’ discriminatory hiring and promotion practices results in an oversupply of labour for “female” occupations, thereby depressing the wages of equally-productive workers.Note Note

The theory of the cultural devaluation of women’s work posits that female-dominated occupations are under-rewarded because of their association with women.Note Note This theory suggests that women, as a group, tend to be valued less than men in the cultures of western industrialized countries, so anything associated with them, including occupations, are valued less by extension.Note Note Note Note Female-dominated occupations also tend to involve temporal flexibility (e.g., part-time work and telework) that does not conform to the organizational norms (i.e., full-time work, overtime, and a constant presence, except when travelling for work), which are predicated on male patterns.Note Note As England explains, this is a form of discrimination whereby “employers see the worth of predominantly female jobs through biased lenses and, as a result, set pay levels for both men and women in predominantly female jobs lower than they would be if the jobs had a more heavily male sex composition.”Note This account of the relationship between gender segregation and gender-based wage inequality is supported by previous research, which shows that the average earnings within occupations decrease as the proportion of women within occupations increases.Note Note

3.1.3. Gender differences in the treatment of similarly-qualified and similarly-positioned workers

Even with the same amount of human capital and employment in the same occupations and industries, women and men may be paid differently due to gender-based discrimination.Note Standard economic models suggest that discrimination can arise in a variety of ways. As per Becker, employers, coworkers, and customers or clients may have personal prejudice against members of particular groups, manifested in a desire to maintain social distance from them.Note Such prejudice will cause a wage differential between women and men when

Discriminatory employers will only hire women at a sufficient wage discount that compensates them for the disutility of employing women. Discriminatory male workers will demand a wage premium to work with women, thus raising men’s relative wages, and the reluctance of discriminatory customers or clients to buy goods or services provided by women will make women less productive in terms of revenue brought in, thus depressing their relative wages.Note

In theory, competitive forces should reduce or eliminate employer discrimination in the long run because the least discriminatory firms, which would hire more female workers because the price of their labour is less than that of male workers, would have lower costs of production and should therefore drive the more discriminatory firms out of business.Note However, discrimination has persisted, and “statistical discrimination” has been used to explain its persistence. Given employer uncertainty about either the productivity of prospective employees or the reliability with which their productivity can be predicted, employers observe prospective employees’ group identity (e.g., sex/gender, Aboriginal status, immigrant status, visible minority status, and/or disability status), and then make hiring, placement, or remuneration decisions on the basis of the expected or “average” level of productivity of that group.Note Employers may discriminate on the basis of that average.Note

3.2. Determining the relative importance of various explanations for the gender pay gap: Decomposition

Faced with the unadjusted gender pay gap, we usually want to know how much of that gap is attributable to gender differences in human capital, occupation, and industry. To make this determination, researchers often begin by using ordinary least square regression to model the logarithmic (log) earnings of workers as a function of sex—coded as one for women and zero for men—and other predictors.Note  The baseline model usually includes only the indicator for sex, such that its coefficient represents the unadjusted difference in the log earnings of female and male workers, with a negative number indicating that the former earn less than the latter on average. Other predictors of log earnings are added in successive models. In these models, the regression coefficient for sex measures the extent to which the log earnings of female and male workers are different—all other predictors in the model being equal (i.e., statistically controlled). In this way, what the gender pay gap would be were it not for gender differences in various characteristics can be determined. For example, based on Labour Force Survey data, the unadjusted gender log wage gap among workers aged 15 years and older was -0.13 in 2017, while the adjusted gender log wage gap was -0.11 (net of province, human capital, union status, public-sector employment, occupation, and industry).

The contribution of human capital, occupation, and industry to explanation of the unadjusted gender pay gap can be then be more formally estimated using the Oaxaca-Blinder decomposition method (see Appendix A for a more technical discussion of this method). After estimating the log earnings of women and men separately as a function of human capital and other productivity-enhancing characteristics, a counterfactual exercise is undertaken, whereby one implicitly asks: what would women’s average earnings be if they received the same remuneration to their human capital, occupation, and industry as men? The portion of the gender pay gap that is “explained” by compositional differences between women and men in terms of their human capital, occupation, and industry is then calculated by subtracting women’s average earnings under the counterfactual scenario from men’s average earnings. The “unexplained” portion of the gender pay gap is the difference between the counterfactual earnings of women and their actual earnings, capturing different returns to women’s and men’s human capital, occupation, and industry. Although this portion of the gender pay gap is typically used as an estimate of gender-based discrimination in the labour market, it also includes the effects of omitted predictors of the gender pay gap.Note

An important caveat regarding these empirical methods for analyzing the gender pay gap is that the predictors do not “explain” earnings inequality between women and men in a social or cultural sense.Note Each predictor is embedded in several layers of context that shape its meaning and effect on earnings.Note In other words, each predictor is itself complicated and requires explanation.Note Also, seemingly neutral predictors may incorporate elements of the discriminatory gender system, and therefore the distinction between the explained and unexplained portion of the gender pay gap is imprecise.Note Gender differences in education and occupation, for example, themselves need to be understood as the outcomes of gendered processes that channel women and men toward different types of degrees, fields of study, and jobs.

What proportion of the gender pay gap in Canada is explained vs. unexplained? As seen in Section 2.4, the gender wage gap in Canada has narrowed over time, although the pace slowed considerably during the 1990s. There is no single, ongoing source of Canadian data on hourly wages for an extended period of time prior to 1997, when the Labour Force Survey began collecting these data.Note For this reason, the gender wage gap among all workers aged 15 years and older from the late 1990s onward is analyzed here (Table 4). Nearly half of the decrease in the gender wage gap that occurred between 1997 and 2017 was explained by changes in the relative characteristics of women and men. Particularly important in this regard were occupation and education. Specifically, the increased representation of women in high-paying managerial and professional occupations was the driving force behind the narrowing of the gender wage gap, followed by the increased proportion of women with a university degree at the bachelor level or above. These two factors are related, as a university degree is typically required for managerial and professional occupations.Note


Table 4
Decomposition of change in the gender wage gap among workers aged 15 and older, Canada, 1997 to 2017
Table summary
This table displays the results of Decomposition of change in the gender wage gap among workers aged 15 and older.
Change in real log gender wage ratio -0.062
Gender gap in log wages in 2017 0.134
Gender gap in log wages in 1997 0.196
Percent due to change in gender differences in characteristics ("explained") 46.8
Province -3.3
Education 13.1
Age -5.3
Job tenure 2.4
Union status 11.2
Public-sector employment 8.9
Occupation 27.7
Industry -8
Percent due to changes in gender differences in returns to characteristics ("unexplained") 53.2

While less important to explaining the decrease in the gender wage gap that occurred between 1997 and 2017, women’s increased job tenure, the reduced proportion of men with union coverage, and the increased proportion of women employed in the well-paid public sector and the decreased proportion of men employed there also played a role.Note

Notably, as the gender wage gap has narrowed over time, what remains is increasingly unexplained. Nearly 70% of the gender wage gap in 2017 could not be explained by gender differences in province, human capital, union status, public-sector employment, occupation, and industry. By comparison, focusing of full-time workers aged 25 to 54, Baker and Drolet find that 61% of the gender wage gap in 1981 was unexplained by a similar model specification.Note In effect, as women have surpassed men in terms of educational attainment and increased their representation in managerial and professional occupations, the remaining gender wage gap has largely become an issue of the unequal returns to such characteristics that women and men receive in the labour market.

3.3. Explanations for the remaining gender pay gap

As compositional differences between women and men in terms of human capital and occupation have been vastly reduced, if not eliminated (i.e., the gender education gap), it has become necessary to consider other explanations for the remaining gender pay gap. In addition to discrimination (previously discussed), motherhood earnings penalties, the high “price” of temporal flexibility in certain occupations, and gender differences in personality traits, psychological attributes, and non-cognitive skills are plausible candidates.

3.3.1. Motherhood earnings penalties

Previous research demonstrates that the gender pay gap is virtually non-existent during early adulthood, but increases with age thereafter (see Section 3.4.3).Note Note It also demonstrates that childless women have higher earnings than women with children; the earnings of childless women are almost equal to those of their male counterparts; and men with children earn as much, if not more, than childless men. Note Note Note Note Note Note Taken together, these findings are suggestive of gender differences in the effects of parenthood on earnings. Specifically, women with children experience “motherhood earnings penalties,” while men with children do not, and may even benefit from “fatherhood earnings premiums.”Note

There are a number of ways that motherhood can negatively affect women’s earnings, both before and after having children. Women may pre-emptively invest less in human capital or select family-friendly, but lower-paying, career paths in anticipation of becoming a mother.Note Note In response to becoming a mother, women may reduce their work hours or switch their occupation or firm to a more child-friendly one.Note Note Note Now that women invest in human capital to a similar extent as men and they have increased their representation in higher-status occupations, post-motherhood effects may be more important for understanding motherhood earnings penalties than pre-motherhood effects. In other words, women may largely be incurring earnings penalties after they have children, rather than before.

How does women’s employment change after they have children? The transition to parenthood tends to move previously-egalitarian couples toward a more traditional gendered division of labour.Note Note Note For this reason, mothers often adjust their labour supply in a downward fashion—leaving the workforce, temporarily or permanently, or reducing their work hours if they remain employed—to accommodate their greater responsibility for childrearing, relative to fathers.Note Note These strategies negatively affect earnings in both the short-term, by reducing time spent at work, and the long-term, by reducing work experience and job tenure.Note However, previous research demonstrates that motherhood earnings penalties persist after controlling for differences in overall work experience and full-time and part-time work experience.Note Note

Mothers may also be less productive at, or committed to, work—or perceived to be so by employers (i.e., statistical discrimination)—insofar as their family responsibilities leave them exhausted and/or distracted, or they put less effort forth at work in order to conserve their energy for later use at home.Note Note Exploring this issue, Kmec finds that mothers and fathers in the United States in the mid-2000s were similar on five out of seven “pro-work” behaviours and conditions: work effort; frequency with which the respondent’s responsibilities at home reduced his/her work effort; the extent to which the respondent’s home life helps with him/her relax and feel ready for the next day’s work; the frequency with which the respondent’s activities and chores at home prevented him/her from getting the needed amount of sleep to do his/her job well; and the frequency with which providing what was needed at home made the respondent work harder at his/her job.Note When they differed, mothers reported greater job engagement and work intensity than fathers.Note Further, mothers were no different from childless women and men on all outcomes.Note

Given mother’s disproportionate responsibility for childrearing, it may be the case that they trade off higher earnings for greater “non-pecuniary amenities” that enable to them balance work and family life, such as less demanding tasks, flexible work hours, few demands for travel or weekend or evening work, and/or on-site childcare.Note Note This mechanism, operating through occupation, industry and/or firm, is elaborated below.

3.3.2. Trading off earnings for temporal flexibility

The majority of the current gender pay gap in Canada derives from earnings inequality between women and men within occupations, as opposed to the uneven distribution of women and men across occupations.Note Based on similar findings for the United States, Goldin argues the gender pay gap is wider in some occupations than others, largely due to differences in the importance attributed to long work hours, set schedules and “face time,” and job tenure.Note Some occupations, particularly in the corporate, financial, and legal worlds, have high penalties for even short work interruptions, and earnings that increase disproportionately or “non-linearly” with hours worked.Note In these occupations, temporal flexibility—fewer and more discretionary work hours—comes at a high price in terms of earnings.Note Due to women’s greater family responsibilities, they may desire more temporal flexibility than men, and they are therefore willing to trade off earnings for that amenity, contributing to gender-based earnings inequality within occupations.Note

In a similar vein, Cha and Weeden suggest that the increasing prevalence of long work hours, defined as 50 or more hours per week, and rising earnings returns to overwork contribute to the gender pay gap, as men are more likely than women to perform long work hours.Note Notably, they demonstrate that the effects of overwork on trends in the gender pay gap in the United States between 1970 and 2009 were most pronounced in managerial and professional occupations, where long work hours are particularly common and the norm of overwork is deeply embedded in organizational practices and occupational cultures.

3.3.3. Gender differences in personality traits, psychological attributes, and non-cognitive skills

While it is unclear whether gender differences in personality traits, psychological attributes, and non-cognitive skills are the product of genetics, culture, or a combination of the two, studies have found that, compared with men, women tend to place a lower value on money; to be less willing to negotiate and compete; to be more risk averse; to have lower self-esteem; to be less self-confident; and to be more agreeable.Note Personality traits, psychological attributes, and non-cognitive skills may affect earnings directly, by increasing productivity like other forms of human capital, and/or indirectly, by influencing human capital, occupation, and industry.Note

Notably, the same personality trait, psychological attribute, or non-cognitive skill in women and men may be rewarded differently in the labour market. For example, Mueller and Plug find that women’s greater agreeableness (i.e., being more trusting, straightforward, altruistic, compliant, modest, and sympathetic), relative to men, is not related to their wages.Note However, men earned a premium for being disagreeable.Note There is also anecdotal and empirical evidence that women are often in a no-win situation at work, in that they face potential penalties for behaving in gender-stereotypical ways (e.g., not negotiating over salaries, raises, or promotions), but, if they try to act more like men, they may elicit negative or less positive responses than men, as they are behaving in “unfeminine” ways.Note Note

3.4. Considerations when studying trends in the gender pay gap

When analyzing and interpreting trends in the gender pay gap, whether using annual earnings or hourly wages, it is necessary to consider changes over time in both the characteristics of workers and the prevailing macroeconomic conditions. Specifically:

  • Female labour force participation has increased dramatically in the past 60 years, and therefore the composition of the female workforce—that is, the characteristics of employed women—may have changed in non-random ways (i.e., selection into employment based on shared characteristics).Note
  • Meaningful comparisons of earnings at different time points require that observed earnings be adjusted for inflation, such that they are expressed in constant or “real” dollars.Note
  • Earnings are affected by the business cycle—the pattern of expansion, contraction, and recovery in the economy—in gendered ways. There is evidence from the United States that women’s earnings decrease, relative to men’s, during business-cycle expansions (i.e., the gender pay gap increases) and increase during recessions (i.e., the gender pay gap decreases).Note
  • Technological changes, particularly advances in information and computer technologies, and globalization have affected the wage structure—that is, returns to human capital, occupation, and industry—and contributed to increasing earnings inequality in Canada since the 1980s.Note Note Note Given gender differences in human capital and occupational/industrial gender segregation, technological changes may have affect women’s and men’s earnings differently.
  • The gender pay gap at a given time simultaneously reflects age, period, and cohort effects. Understanding trends in the gender pay gap therefore requires that these effects be distinguished.  

This section explores the implications for trends in the gender pay gap of (1) selection into employment, (2) overall earnings inequality and the wage structure, and (3) age, period and cohort effects, as these issues have received considerable attention in the academic literature.

3.4.1. Selection into employment

Labour force participation is central to understanding the gender pay gap, both because receipt of earnings is conditional on employment, and because work experience is a predictor of earnings.Note Given that earnings data are only available for workers, and additional restrictions may be placed on the analytic sample of workers (i.e., workers employed on a full-time, full-year basis), it is possible that observed earnings reflect a non-random or self-selected group of individuals that chose paid work over staying home.Note Selection bias has generally been considered to be more influential for women’s earnings than men’s, as women are less likely than men to participate in the labour market.Note Note As Blau and Kahn explain, “…the closer the wage sample is to 100 percent of the underlying population, the smaller the selection bias.”Note

Women with the potential to earn more in the labour market due to their human capital, who therefore face higher opportunity costs to staying home in the form of foregone wages, may be more likely to be employed (i.e., positive selection).Note Note Furthermore, if women receive lower returns to their human capital, occupation, and industry than men, they may rationally and disproportionately choose not to enter the labour market. In this case, the observed gender pay gap will be understated or biased downward (and the gender pay ratio will be biased upward), relative to the corrected estimate.

Alternatively, women with the potential to earn more in the labour market may be less likely to be employed (i.e., negative selection). It is well-established that individuals tend to marry/cohabit with those who are similar to themselves in terms of education (i.e., “educational homogamy”), due to both preferences and greater opportunities to meet such spouses/partners by virtue of attending the same type of educational institutions. Note Note Note Note Note Given that earnings are positively associated with education, educational homogamy implies that women with more education are better positioned financially to remain home, insofar as their spouse/partner has relatively high earnings. In this case, the observed gender pay gap will be overstated or biased upward (and the gender pay ratio biased downward), relative to the corrected estimate.

As the labour force participation of women has increased and women have become more similar to men with respect to labour supply and human capital, scholars have questioned the continuing relevance of selection bias to the gender pay gap.Note Note Using a statistical technique to control for selectivity (i.e., Heckman’s method), as outlined in Appendix B, the gender wage gap among workers aged 15 and older in 2017 was estimated from the Labour Force Survey.Note There was no evidence that the gender wage gap was affected by selection bias.

When studying longer time series (or different geographies), correcting for selection bias may be necessary to make valid inferences.Note Note  Women’s employment has increased dramatically since the 1960s, and therefore the characteristics of employed women may be different in 2017, as compared to say 1977, in ways that affect estimates of women’s earnings and the gender pay gap (Chart 6).Note The characteristics of employed women may also change over time, even if female labour force participation remains constant, in response to changes in earnings returns to human capital, occupation and industry (see Section 3.4.2). For example, Mulligan and Rubinstein provide evidence that the selection of women into full-time, full-year employment in the United States shifted from negative in the 1970s to positive in the 1990s.Note Further, they demonstrate that the majority of the apparent narrowing of the gender pay gap in the United States during that period reflected these changes in the characteristics of employed women.Note Baker, Benjamin, Cegep, and Grant reach the same conclusion regarding the gender pay gap in Canada between 1970 and 1990.Note

Chart 6

Data table for Chart 6 
Data table for Chart 6
Table summary
This table displays the results of Data table for Chart 6 Women and Men, calculated using employment rate (percent) units of measure (appearing as column headers).
Women Men
employment rate (percent)
1946 24.1 82.0
1947 23.7 83.0
1948 23.1 83.0
1949 23.1 82.5
1950 22.6 80.7
1951 23.0 81.8
1952 23.1 80.8
1953 23.0 81.1
1954 23.0 77.9
1955 23.3 78.0
1956 24.4 79.0
1957 25.2 77.9
1958 25.3 75.0
1959 25.9 75.4
1960 26.9 74.2
1961 27.6 73.1
1962 28.1 73.6
1963 28.6 73.5
1964 29.6 74.0
1965 30.5 74.4
1966 32.7 76.7
1967 33.6 75.9
1968 34.0 74.7
1969 34.7 74.5
1970 34.6 73.1
1971 35.2 72.3
1972 35.8 72.6
1973 37.4 73.9
1974 38.5 74.5
1975 39.0 73.0
1976 41.9 72.7
1977 42.3 71.9
1978 43.4 72.0
1979 45.1 73.1
1980 46.4 72.8
1981 47.7 72.8
1982 46.5 68.4
1983 46.8 67.4
1984 47.7 68.0
1985 49.1 68.7
1986 50.3 69.6
1987 51.3 70.3
1988 52.7 70.9
1989 53.6 71.1
1990 53.8 69.9
1991 52.8 66.9
1992 51.9 65.0
1993 51.5 64.6
1994 51.9 65.1
1995 52.2 65.4
1996 52.1 65.0
1997 52.6 65.5
1998 53.7 66.0
1999 54.6 66.7
2000 55.4 67.3
2001 55.7 66.8
2002 56.6 67.1
2003 57.5 67.6
2004 57.7 67.7
2005 57.7 67.7
2006 58.1 67.6
2007 59.0 67.9
2008 59.0 68.0
2009 58.0 65.1
2010 57.9 65.3
2011 57.8 65.7
2012 57.9 65.5
2013 58.1 65.6
2014 57.6 65.4
2015 57.4 65.3
2016 57.5 64.9
2017 57.9 65.4

3.4.2. Overall earnings inequality and wage structure

Since 1980, earnings inequality (i.e., dispersion) in Canada has increased significantly, as the real earnings of individuals at the top of the earnings distribution have risen dramatically. Note Note At the same time, the earnings of middle-income earners have stagnated, and the earnings of low-income earners have fallen.Note Increasing earnings inequality tends to offset to some extent trends toward reducing the gender pay gap, as women are underrepresented at the top of the earnings distribution and overrepresented at the bottom.Note Note Note In effect, women have been “swimming upstream” by narrowing the gender pay gap despite economy-wide forces working against them.Note

Changes in the wage structure—that is, changes in returns to human capital, occupation, and industry—have contributed to increasing earnings inequality.Note New technologies, particularly computers in the workplace, tend to enhance productivity and therefore earnings in high-skill occupations involving cognitive tasks (e.g., managerial and professional occupations), while replacing workers who perform routine tasks (often found in middle-skill secretarial and clerical occupations), resulting in lower demand and earnings for those workers.Note Note Although low-skill sales and services occupations that entail non-routine tasks and personal interaction are less likely to be affected by greater use of computers, earnings in these occupations may decrease as job loss in middle-skilled occupations and manufacturing increases the supply of available workers.Note Note To the extent that these changes favor men over women or vice versa, given gender differences in human capital and occupational/industrial gender segregation, they may affect trends in the gender pay gap.Note

While weekly-wage returns to work experience have not changed in Canada since the 1980s, Foley and Green demonstrate that wage returns to education increased throughout the 1980s and 1990s, even as the supply of more educated workers grew. Note This pattern can be expected to have benefited women to a greater extent than men, as the share of university graduates rose faster among women than men after 1990.Note

In the 2000s, wage returns to university degrees, relative to high school diplomas, declined among men, and the trend toward increasing educational attainment stalled.Note Foley and Green attribute these findings to the resource (oil and gas) boom in western and Atlantic Canada, in that it increased demand for middle- and low-skilled workers, particularly young men, thereby increasing their wages.Note Notably, this shift in demand can be expected to have benefitted men, relative to women. However, in the wake of lower oil prices, Foley and Green predict that demand for workers in the resource sector will decline.Note

According to Lemieux and Riddell, certain occupations and industries drove the growth in top incomes in Canada (around 80% of which came from earnings) between 1981 and 2011.Note Specifically, senior managers in the finance and insurance industry, individuals in business and finance occupations, and individuals in natural and applied sciences occupations came to represent a much larger proportion of the top 1 percent, and their incomes grew much more rapidly than those of other top-income earners.Note The share of top earners also increased significantly in oil and gas extraction, while it decreased in manufacturing.Note Given women’s underrepresentation in leadership positions, STEM occupations, and resource extraction, increasing returns to these occupations and industries is increasingly unfavorable to women, and puts them at a growing disadvantage when it comes to narrowing the gender pay gap.Note

3.4.3. Age, period and cohort effects

It is well-established that the wages of young women and men are more equal than the wages of older women and men, and that the gradual convergence of the gender pay gap over time is driven by the replacement of older cohorts by younger ones.Note Note Note Note In order to understand these patterns, the effects of age, period, and cohort must be distinguished. In the simplest terms, age effects refer to variations produced by the physiological or social processes of aging.Note Studies consistently show that the gender pay gap is minimal at the time of entry into the labour market, but grows as women and men age and, in the process, form families (Table 5, reading across the rows).Note Since women’s employment is shaped to a greater extent by their caregiving roles and/or employers’ presumptions of these roles than men’s, women’s wages are more affected by growing older than men’s.Note As they age, women work fewer hours outside of the home than men, and they experience more work interruptions and absences related to childcare.Note Women are also less likely than men to have high-wage jobs with promotional opportunities, and they may face barriers to advancement.Note


Table 5
Gender wage ratio as measured from the average hourly wages of employed women and men by period and age group, Canada, 1997, 2007 and 2017
Table summary
This table displays the results of Gender wage ratio as measured from the average hourly wages of employed women and men by period and age group 15 to 24, 25 to 34, 35 to 44, 45 to 54, 55 to 64 and 65 years and over, calculated using ratio units of measure (appearing as column headers).
15 to 24 25 to 34 35 to 44 45 to 54 55 to 64 65 years and over
ratio
1997 91.2 87.6 80.7 75.3 73.8 77.9
2007 89.4 89.0 82.2 81.0 79.1 81.1
2017 93.0 89.5 85.7 84.8 83.4 89.6

There is evidence that the gender wage gap is not increasing with age among recent cohorts, as it did among their predecessors.Note In other words, the negative association between aging and the gender pay ratio is weakening over time.Note Based on data from the Labour Force Survey, in 1997, the gender wage ratio was 16.4 percentage points lower among seniors aged 65 years and older (0.78) than it was among 15-to-19-year-olds (0.94). By 2017, the difference between the gender wage ratio of younger and older workers was reduced to 5.6 percentage points. These findings suggest that cohort replacement drives the convergence of the gender pay gap over time: as younger cohorts replace older ones, the overall gender pay gap decreases simply because the gender pay gap is smaller among younger cohorts than older ones.Note

Period effects refer to variations produced by events that simultaneously affect all ages and cohorts.Note A number of dramatic social and economic changes have occurred in Canada since the 1970s, which may have contributed to the decrease in the gender pay gap over time. For example, the federal government has offered more legal protections against gender-based discrimination (e.g., Employment and Pay Equity legislation), and gender attitudes have become more egalitarian.Note At the same time, broad structural changes in the Canadian economy that have disproportionally affected men—such as the shift away from manufacturing toward the provision of services, and declining union coverage rates—have worsened their labour market position, thereby reducing the gender pay gap.Note

Period effects are difficult to separate from cohort effects: variations produced by being born at a particular time, within a particular social context.Note Recent cohorts of women have more education, work experience, and representation in higher-status occupations than their predecessors. They also have greater control over their fertility and more egalitarian personal relationships.Note For these reasons, the gender wage gap has decreased across successive cohorts (Table 5, reading down the columns), indicating that each new cohort of women has fared better than the previous one.Note

Although the narrowing of the gender pay gap is primarily related to the entry of new cohorts, within-cohort earnings growth has also played a role for older cohorts.Note The relatively high gender wage ratios for younger women (under the age of 35) tend to decline as they age (Table 5, reading diagonally), largely due to women’s tendency to interrupt or discontinue their labour force participation for family reasons. However, the gender wage ratios of older women (aged 35 years and older) tend to increase as they age, likely because the demands of motherhood ease as children mature.

4. Conclusion

The unadjusted gender pay gap is an important indicator of gender equality and women’s economic empowerment. Decisions made by analysts regarding who estimates are based upon (i.e., age and labour supply), what earnings are counted (i.e., annual earnings vs. hourly wages), and how typical earnings are calculated (i.e., average or median) affect the size of the gender pay gap and its perceived seriousness.

Whichever measure of the unadjusted gender pay gap is used, it reflects, among other predictors, gender differences in human capital, occupation, and industry, as well as gender-based discrimination. Statistical control in the context of ordinary least squares regression enables estimation of the gender pay gap adjusted for the effects of various predictors. The adjusted gender pay gap reflects a hypothetical situation in which female and male workers are equally skilled and evenly distributed across occupations and industries. It should therefore not be interpreted as evidence that the gender pay gap is inconsequential or does not exist.

Traditional economic explanations for the gender pay gap focus on gender differences in human capital, occupation and industry, and returns to these characteristics (i.e., discrimination). Women’s increased educational attainment, both overall and relative to men, stronger attachments to the labour market over the life course, and greater representation in managerial and professional occupations have contributed to the narrowing of the gender pay gap over time. However, the remaining gender pay gap is largely unexplained by gender differences in human capital, occupation, and industry, making it necessary to consider other explanations. In addition to gender-based discrimination, motherhood penalties, the high “price” of temporal flexibility in certain occupations, and gender differences in personality traits, psychological attributes, and non-cognitive skills are plausible candidates.

It is important to recognize that, in the context of ordinary least squares regression and Oaxaca-Blinder decomposition, predictors of the gender pay gap do not “explain” it in a social or cultural sense. Gender differences in education and occupation, for example, themselves need to be understood as the outcomes of gendered processes that channel women and men toward different types of degrees, fields of study, and jobs.

Also, when analyzing trends in the gender pay gap, it is necessary to consider changes over time in both the characteristics of workers and the prevailing macroeconomic conditions, including the degree and direction of selection bias, overall earnings inequality, and age, period and cohort effects.  

Appendix A: Decomposing the gender pay gap

Oaxaca-Blinder decomposition

Faced with an unadjusted pay gap between women and men (on average), researchers typically endeavor to determine how much of that gap can be explained by gender differences in human capital and other productivity-related characteristics. To do so, they use the Oaxaca-Blinder decomposition method in the context of an ordinary least squares regression, estimating the logarithmic (log) mean wages of women and men separately, as a function of productivity characteristics (equations a and b). After estimating an earnings equation for women and one for men, a counterfactual exercise is undertaken, whereby one implicitly asks: what would women’s average earnings be if they received the same remuneration to their human-capital characteristics as men (equation c)? The portion of the gender pay gap that is explained by compositional differences between women and men with respect to their human-capital characteristics is then calculated by subtracting women’s average earnings under the counterfactual scenario from men’s average earnings (equation d). The unexplained portion of the gender pay gap is the difference between the counterfactual earnings of women and their actual earnings, capturing different returns to women’s human capital (equation e).

  1. Earnings equation for women: E ¯ w = α w + β w X ¯ w + μ w MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaWgaaWcbaWd biabdEha3bWdaeqaaOWdbiabg2da9iabeg7aH9aadaWgaaWcbaWdbi abdEha3bWdaeqaaOWdbiabgUcaRiabek7aI9aadaWgaaWcbaWdbiab dEha3bWdaeqaaOWdbiqbdIfay9aagaqeamaaBaaaleaapeGaem4DaC hapaqabaGcpeGaey4kaSIaeqiVd02damaaBaaaleaapeGaem4DaCha paqabaaaaa@4E32@
  2. Earnings equation for men: E ¯ m = α m + β m X ¯ m + μ m MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaWgaaWcbaWd biabd2gaTbWdaeqaaOWdbiabg2da9iabeg7aH9aadaWgaaWcbaWdbi abd2gaTbWdaeqaaOWdbiabgUcaRiabek7aI9aadaWgaaWcbaWdbiab d2gaTbWdaeqaaOWdbiqbdIfay9aagaqeamaaBaaaleaapeGaemyBa0 gapaqabaGcpeGaey4kaSIaeqiVd02damaaBaaaleaapeGaemyBa0ga paqabaaaaa@4DCE@
  3. Counterfactual equation for women: E ¯ w * = α m + β m X ¯ w + μ w MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaqhaaWcbaWd biabdEha3bWdaeaapeGaeiOkaOcaaOGaeyypa0JaeqySde2damaaBa aaleaapeGaemyBa0gapaqabaGcpeGaey4kaSIaeqOSdi2damaaBaaa leaapeGaemyBa0gapaqabaGcpeGafmiwaG1dayaaraWaaSbaaSqaa8 qacqWG3bWDa8aabeaak8qacqGHRaWkcqaH8oqBpaWaaSbaaSqaa8qa cqWG3bWDa8aabeaaaaa@4EE7@
  4. “Explained” variation: E ¯ m E ¯ w * MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaWgaaWcbaWd biabd2gaTbWdaeqaaOWdbiabgkHiTiqbdweaf9aagaqeamaaDaaale aapeGaem4DaChapaqaa8qacqGGQaGkaaaaaa@4251@
  5. “Unexplained” variation: ¯ E w * ¯ E w MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaGaeg4Ra8keaaaaaaaaa8qacqWGfbqrpaWaa0baaSqa a8qacqWG3bWDa8aabaWdbiabcQcaQaaakiabgkHiT8aacqyGVcWRpe Gaemyrau0damaaBaaaleaapeGaem4DaChapaqabaaaaa@470E@

Change in the unadjusted gender pay gap over time (equation f) can also be decomposed into an explained portion attributable to changes in the average human capital of women and men (equation g), and an unexplained portion attributable to changes in women’s and men’s returns to human capital (equation h).Note

  1. Change in the gender pay gap between two time points (t and t-1): ( E ¯ m t E ¯ m t 1 ) ( E ¯ f t E ¯ f t 1 ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qadaqadaWdaeaapeGafmyrau0dayaa raWaaSbaaSqaa8qacqWGTbqBpaWaaSbaaWqaa8qacqWG0baDa8aabe aaaSqabaGcpeGaeyOeI0Iafmyrau0dayaaraWaaSbaaSqaa8qacqWG TbqBpaWaaSbaaWqaa8qacqWG0baDcqGHsislcqaIXaqma8aabeaaaS qabaaak8qacaGLOaGaayzkaaGaeyOeI0YaaeWaa8aabaWdbiqbdwea f9aagaqeamaaBaaaleaapeGaemOzay2damaaBaaameaapeGaemiDaq hapaqabaaaleqaaOWdbiabgkHiTiqbdweaf9aagaqeamaaBaaaleaa peGaemOzay2damaaBaaameaapeGaemiDaqNaeyOeI0IaeGymaedapa qabaaaleqaaaGcpeGaayjkaiaawMcaaaaa@5756@
  2. Explained variation between two time points: [ β m t ( X ¯ m t X ¯ m t 1 ) β w t ( X ¯ w t X ¯ w t 1 ) ] MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qadaWadaWdaeaapeGaeqOSdi2damaa BaaaleaapeGaemyBa02damaaBaaameaapeGaemiDaqhapaqabaaale qaaOWdbmaabmaapaqaa8qacuWGybawpaGbaebadaWgaaWcbaWdbiab d2gaT9aadaWgaaadbaWdbiabdsha0bWdaeqaaaWcbeaak8qacqGHsi slcuWGybawpaGbaebadaWgaaWcbaWdbiabd2gaT9aadaWgaaadbaWd biabdsha0jabgkHiTiabigdaXaWdaeqaaaWcbeaaaOWdbiaawIcaca GLPaaacqGHsislcqaHYoGypaWaaSbaaSqaa8qacqWG3bWDpaWaaSba aWqaa8qacqWG0baDa8aabeaaaSqabaGcpeWaaeWaa8aabaWdbiqbdI fay9aagaqeamaaBaaaleaapeGaem4DaC3damaaBaaameaapeGaemiD aqhapaqabaaaleqaaOWdbiabgkHiTiqbdIfay9aagaqeamaaBaaale aapeGaem4DaC3damaaBaaameaapeGaemiDaqNaeyOeI0IaeGymaeda paqabaaaleqaaaGcpeGaayjkaiaawMcaaaGaay5waiaaw2faaaaa@64D7@
  3. Unexplained variation between two time points: [ X ¯ m t 1 ( β m t β m t 1 ) X ¯ f t 1 ( β f t β f t 1 ) ] MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qadaWadaWdaeaapeGafmiwaG1dayaa raWaaSbaaSqaa8qacqWGTbqBpaWaaSbaaWqaa8qacqWG0baDcqGHsi slcqaIXaqma8aabeaaaSqabaGcpeWaaeWaa8aabaWdbiabek7aI9aa daWgaaWcbaWdbiabd2gaT9aadaWgaaadbaWdbiabdsha0bWdaeqaaa Wcbeaak8qacqGHsislcqaHYoGypaWaaSbaaSqaa8qacqWGTbqBpaWa aSbaaWqaa8qacqWG0baDcqGHsislcqaIXaqma8aabeaaaSqabaaak8 qacaGLOaGaayzkaaGaeyOeI0IafmiwaG1dayaaraWaaSbaaSqaa8qa cqWGMbGzpaWaaSbaaWqaa8qacqWG0baDcqGHsislcqaIXaqma8aabe aaaSqabaGcpeWaaeWaa8aabaWdbiabek7aI9aadaWgaaWcbaWdbiab dAgaM9aadaWgaaadbaWdbiabdsha0bWdaeqaaaWcbeaak8qacqGHsi slcqaHYoGypaWaaSbaaSqaa8qacqWGMbGzpaWaaSbaaWqaa8qacqWG 0baDcqGHsislcqaIXaqma8aabeaaaSqabaaak8qacaGLOaGaayzkaa aacaGLBbGaayzxaaaaaa@68CB@

Although there is no universally accepted set of predictors that should be included in a wage analysis, there are a number of widely-used predictors: personal characteristics, including presence of children in the household, age of the youngest child in the household, and province of residence; productivity-enhancing characteristics, including standardized test scores, highest level of educational attainment, work experience in quadratic terms, and job tenure; and work characteristics, including job permanence, class of worker (private or public sector), and union status.Note Note Presence of children and age of the youngest child in the household are intended to capture family responsibilities that can affect labour supply, but they may also be a harbinger of discrimination (i.e., motherhood earnings penalties).. These variables may also be considered to be endogenous to women’s decisions about their labour supply, and therefore exclude from analysis of the gender pay gap.Note

Province of residence controls for differences between local labour markets in wage structure and tightness (i.e., the number of job vacancies relative to the number of unemployed people).Note  Standardized test scores, education, and work experience (typically approximated by age) are included in the gender wage gap decomposition as productivity-enhancing characteristics. Work experience squared captures diminishing returns to work experience as its duration lengthens. Job tenure with one’s current employer, as distinct from work experience, measures both job-specific knowledge and loyalty. Job permanence, class of worker, and union status account for differences in wage structures between different types of jobs (i.e., seasonal, temporary, and permanent jobs; private- and public-sector jobs; and unionized and non-unionized jobs).

Assuming that productivity-related differences between women and men are precisely measured, and that they are not themselves the product of discriminatory behavior by employers, the unexplained portion of the gender pay gap can be interpreted as evidence of discrimination.Note In effect, women and men with the same human capital are paid unequally on the grounds of a characteristic that has no direct effect on their productivity (i.e., gender). However, any analytic approach that relies on the statistical residual is open to question as to whether all of the relevant predictors were included in the regression.Note When human capital characteristics explain only a portion of the gender pay gap, the possibility exists that unmeasured productivity-related characteristics, such as effort, may be reflected in the residual.Note Note If men are more highly endowed with those characteristics than women, the extent of discrimination will be overestimated.Note Until recently, work experience has been particularly problematic in this regard. In the absence of retrospective data on work histories, as collected through the General Social Survey on Families until 2011, for example, “potential” experience is approximated by age or estimated by subtracting from respondents’ current age the number of years spent in school plus four (corresponding to the pre-school years). This solution tends to overestimate women’s actual experience because they have more frequent and longer lasting career interruptions than do men.Note However, as continuous labour force participation over the life course has become more typical among recent cohorts of women, relative to their predecessors, the accuracy of estimates of women’s potential experience has improved.Note Note

Additional and somewhat controversial predictors of the gender pay gap include field of study, occupation, and industry. Inclusion of any one of these characteristics (or all of them) in the analysis takes into account occupational/industrial gender segregation.Note However, if occupational/industrial gender segregation reflects gendered processes and gender-based discrimination, the unexplained portion of the gender pay gap will be underestimated when field of study, occupation, and/or industry are included. Given that these variables are controversial predictors of the gender pay gap, a worthwhile strategy is to present the results of a series of model specifications that alternately include and exclude them.

Gender pay gap at different points across the earnings distribution

Although most research focuses on the gender pay gap at the mean, it is worthwhile to examine the gender pay gap across the entire earnings distribution. Studies that have done so reveal that earnings inequality between women and men has decreased more at the bottom of the wage distribution than it has at the top of the wage distribution. Note Note Note Note Note Note Note

Decompositions can be performed at various points (i.e., percentiles) along the wage distribution. Doing so requires use of the unconditional quantile regression method, recently developed by Firpo, Fortin, and LemieuxNote Note and Fortin, Lemieux, and Firpo.Note This method involves finding women’s and men’s (log) wages at a given percentile (separately)—typically the 10th and/or 90th percentile, representing the bottom and top of the wage distribution, respectively. Each gender gap can then be decomposed as per the standard Oaxaca-Blinder approach, using a transformed dependent variable, known as the “re-centered influence function,” in the OLS regressions. For a given percentile (j), the recentered influence function (RIF) is computed for each individual (i) in the pooled male-female sample as such:

R I F j i = w j + j 1 { w i w j } f Y ( w j ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqWGsbGucqWGjbqscqWGgbGrpaWa aSbaaSqaa8qacqWGQbGAcqWGPbqAa8aabeaak8qacqGH9aqpcqWG3b WDpaWaaSbaaSqaa8qacqWGQbGAa8aabeaak8qacqGHRaWkdaWcaaWd aeaapeGaemOAaOMaeyOeI0IaeGymaeZaaiqaa8aabaWdbiabdEha39 aadaWgaaWcbaWdbiabdMgaPbWdaeqaaOWdbiabgsMiJoaaciaapaqa a8qacqWG3bWDpaWaaSbaaSqaa8qacqWGQbGAa8aabeaaaOWdbiaaw2 haaaGaay5Eaaaapaqaa8qacqWGMbGzpaWaaSbaaSqaa8qacqWGzbqw a8aabeaak8qadaqadaWdaeaapeGaem4DaC3damaaBaaaleaapeGaem OAaOgapaqabaaak8qacaGLOaGaayzkaaaaaaaa@5C66@

Where, w j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqWG3bWDpaWaaSbaaSqaa8qacqWG QbGAa8aabeaaaaa@3DA6@ is the wage at a given percentile (j); { w i w j } MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaWaaiWaaeaaqaaaaaaaaaWdbiabdEha39aadaWgaaWc baWdbiabdMgaPbWdaeqaaOWdbiabgsMiJkabdEha39aadaWgaaWcba WdbiabdQgaQbWdaeqaaaGccaGL7bGaayzFaaaaaa@44DC@ is an indicator function taking the value of one, if an individual’s wage is less than the wage at the j t h MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqWGQbGApaWaaWbaaSqabeaapeGa emiDaqNaemiAaGgaaaaa@3EEB@ percentile, and a value of zero otherwise; and f Y ( w j ) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqWGMbGzpaWaaSbaaSqaa8qacqWG zbqwa8aabeaakmaabmaabaWdbiabdEha39aadaWgaaWcbaWdbiabdQ gaQbWdaeqaaaGccaGLOaGaayzkaaaaaa@423D@ is the density of the wage at the j t h MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacqWGQbGApaWaaWbaaSqabeaapeGa emiDaqNaemiAaGgaaaaa@3EEB@ percentile, estimated by kernel density.

Appendix B: Correcting for selection bias

Simple method

The possibility that the gender pay gap reflects selection bias can be assessed using a technique delineated by Baker, Benjamin, Cegep, and Grant and Drolet.Note Note After estimating the earnings equations for employed women and men at a given time point (t), as per equations a and b in appendix A, the mean log wages of those who were not employed (and who were therefore not part of the analytic sample) at that time are estimated by substituting their mean values ( X ¯ w   o r   X ¯ m   ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qadaqadaWdaeaapeGafmiwaG1dayaa raWaaSbaaSqaa8qacqWG3bWDcqGGGcaOa8aabeaak8qacqWGVbWBcq WGYbGCcqGGGcaOcuWGybawpaGbaebadaWgaaWcbaWdbiabd2gaTbWd aeqaaOWdbiabcckaGcGaayjkaiaawMcaaaaa@4AB1@ for the corresponding regression coefficients ( β w   o r   β m   ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qadaqadaWdaeaapeGaeqOSdi2damaa BaaaleaapeGaem4DaChapaqabaGcpeGaeiiOaaQaem4Ba8MaemOCai NaeiiOaaQaeqOSdi2damaaBaaaleaapeGaemyBa0MaeiiOaakapaqa baaak8qacaGLOaGaayzkaaaaaa@4B51@ .

To look at changes over time, one year is designated as the base year (t = 0). For all of the subsequent years, a new weighted estimate of women’s log wages are calculated as:

E ¯ t w = ω t w E p t w ¯ + [ ( 1 ω ¯ t w ) * E ¯ n t w ] MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaqhaaWcbaWd biabdsha0bWdaeaapeGaem4DaChaaOGaeyypa0ZdamaanaaabaWdbi abeM8a39aadaqhaaWcbaWdbiabdsha0bWdaeaapeGaem4DaChaaOGa emyrau0damaaDaaaleaapeGaemiCaaNaemiDaqhapaqaa8qacqWG3b WDaaaaaOGaey4kaSYaamWaa8aabaWdbmaabmaapaqaa8qacqaIXaqm cqGHsislcuaHjpWDpaGbaebadaqhaaWcbaWdbiabdsha0bWdaeaape Gaem4DaChaaaGccaGLOaGaayzkaaGaeiOkaOIafmyrau0dayaaraWa a0baaSqaa8qacqWGUbGBcqWG0baDa8aabaWdbiabdEha3baaaOGaay 5waiaaw2faaaaa@5DBD@

Where,

E ¯ p t w = M e a n   l o g   w a g e s   o f   e m p l o y e d   w o m e n   a t   t i m e   t MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaqhaaWcbaWd biabdchaWjabdsha0bWdaeaapeGaem4DaChaaOGaeyypa0Jaemyta0 KaemyzauMaemyyaeMaemOBa4MaeiiOaaQaemiBaWMaem4Ba8Maem4z aCMaeiiOaaQaem4DaCNaemyyaeMaem4zaCMaemyzauMaem4CamNaei iOaaQaem4Ba8MaemOzayMaeiiOaaQaemyzauMaemyBa0MaemiCaaNa emiBaWMaem4Ba8MaemyEaKNaemyzauMaemizaqMaeiiOaaQaem4DaC Naem4Ba8MaemyBa0MaemyzauMaemOBa4MaeiiOaaQaemyyaeMaemiD aqNaeiiOaaQaemiDaqNaemyAaKMaemyBa0MaemyzauMaeiiOaaQaem iDaqhaaa@7E3A@ ω ¯ t w = E m p l o y m e n t   r a t e   o f   w o m e n   a t   t i m e   t E m p l o y m e n t   r a t e   o f   w o m e n   a t   t i m e   0 MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuaHjpWDpaGbaebadaqhaaWcbaWd biabdsha0bWdaeaapeGaem4DaChaaOGaeyypa0ZaaSaaa8aabaWdbi abdweafjabd2gaTjabdchaWjabdYgaSjabd+gaVjabdMha5jabd2ga TjabdwgaLjabd6gaUjabdsha0jabcckaGkabdkhaYjabdggaHjabds ha0jabdwgaLjabcckaGkabd+gaVjabdAgaMjabcckaGkabdEha3jab d+gaVjabd2gaTjabdwgaLjabd6gaUjabcckaGkabdggaHjabdsha0j abcckaGkabdsha0jabdMgaPjabd2gaTjabdwgaLjabcckaGkabdsha 0bWdaeaapeGaemyrauKaemyBa0MaemiCaaNaemiBaWMaem4Ba8Maem yEaKNaemyBa0MaemyzauMaemOBa4MaemiDaqNaeiiOaaQaemOCaiNa emyyaeMaemiDaqNaemyzauMaeiiOaaQaem4Ba8MaemOzayMaeiiOaa Qaem4DaCNaem4Ba8MaemyBa0MaemyzauMaemOBa4MaeiiOaaQaemyy aeMaemiDaqNaeiiOaaQaemiDaqNaemyAaKMaemyBa0MaemyzauMaei iOaaQaeGimaadaaaaa@A308@ E ¯ n t w = M e a n   log w a g e s   o f   w o m e n   w h o   w e r e   n o t   e m p l o y e d   a t   t i m e   t MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaqhaaWcbaWd biabd6gaUjabdsha0bWdaeaapeGaem4DaChaaOGaeyypa0Jaemyta0 KaemyzauMaemyyaeMaemOBa4MaeiiOaaQaeeiBaWMaee4Ba8Maee4z aCMaem4DaCNaemyyaeMaem4zaCMaemyzauMaem4CamNaeiiOaaQaem 4Ba8MaemOzayMaeiiOaaQaem4DaCNaem4Ba8MaemyBa0MaemyzauMa emOBa4MaeiiOaaQaem4DaCNaemiAaGMaem4Ba8MaeiiOaaQaem4DaC NaemyzauMaemOCaiNaemyzauMaeiiOaaQaemOBa4Maem4Ba8MaemiD aqNaeiiOaaQaemyzauMaemyBa0MaemiCaaNaemiBaWMaem4Ba8Maem yEaKNaemyzauMaemizaqMaeiiOaaQaemyyaeMaemiDaqNaeiiOaaQa emiDaqNaemyAaKMaemyBa0MaemyzauMaeiiOaaQaemiDaqhaaa@8FBE@

New weighted estimates of men’s log wages would be similarly calculated.

Although the term, E ¯ n t w MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfKttLearuatH9givLearmWu51MyVXgatC vAUfeBSjuyZL2yd9gzLbvyNv2CaeHbd9wDYLwzYbItLDharyavP1wz ZbItLDhis9wBH5garqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbb L8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpe pae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaam aaeaqbaaGcbaaeaaaaaaaaa8qacuWGfbqrpaGbaebadaqhaaWcbaWd biabd6gaUjabdsha0bWdaeaapeGaem4DaChaaaaa@405C@ , controls for observable differences between employed women and those who are not employed, it is also possible to control unobservable differences by multiplying that term by a scale factor (k).Note Assuming that those who are not employed would receive lower wage offers than those who are employed, a k value less than 1.0 would be appropriate. Baker and his colleagues present adjusted results for both k = 1.0 and k = 0.9.

Heckman two-step correction

Heckman’s two-step correction makes it possible to assess whether selection bias is present, identify its determinants, and control for it in estimating an outcome of interest.Note To begin, the substantive models of the outcomes of interest—in this case, women’s wages—are estimated by means of OLS regression. In the first step of the “Heckman,” a selection model is developed for women, which estimates the likelihood of participation vs. non-participation—in this case, (1) employed versus (0) not employed—by means of binary probit regression.Note Note Multiple predictors of women’s employment status will be included in the selection model, most of which will also be included in the substantive models of women’s wages. It is important that the selection model contain at least one variable that can be legitimately excluded from the substantive models, so as to safeguard against collinearity between these models and properly identify the effect of selection. In effect, such a variable (or “instrument”) must strongly predict women’s employment status, but not their wages: marital status, family type, and the availability or level of non-earned income within the household can play this role.Note

In the second step, the residuals (error terms) from the selection model are used to derive a new variable called the Inverse Mills Ratio (IMR).Note The IMR is then added to the substantive model to capture both observed and unobserved variables that affect selection. Here, two factors are of interest in terms of determining whether there is significant sample-selection bias that results from the exclusion of non-working respondents from the substantive models. First, if the IMR is significant, it is suggestive of sample-selection bias in the substantive models. A second factor to consider is whether there have been significant changes in the parameter estimates of the other predictors in the substantive models before and after correction.

Appendix C: Data Sources

Statistics Canada has a vast repository of data sources that can be used to examine the gender pay gap. Since the variables included in these data sources differ, the appropriateness and usefulness of a given dataset depends on the empirical questions being addressed. Below, an overview of relevant datasets is provided.

Census of Population

On one specific day every five years, Statistics Canada conducts the Census of Population in order to develop a statistical portrait the people in Canada by their demographic, social and economic characteristics. While all households receive a short-form questionnaire, a sample of approximately 25% of Canadian households receive a long-form questionnaire (with the exception of 2011). The short-form census enumerates the entire Canadian population, which consists of Canadian citizens (by birth and by naturalization), landed immigrants and non-permanent residents and their families living with them in Canada.Note The long-form census includes the same target population as the short-form census, with the exception of Canadian citizens living temporarily in other countries; full-time members of the Canadian Forces stationed outside Canada; persons living in institutional collective dwellings such as hospitals, nursing homes and penitentiaries; and persons living in non-institutional collective dwellings such as work camps, hotels and motels, and student residences.

For the first time, the 2016 Census included administrative data from the Canada Revenue Agency (CRA) on annual income from various sources in 2015 for those who completed the short-form census.Note From these data, the gender pay gap can be calculated in terms of annual earnings. Using information on the number of weeks worked in the year preceding the census day and the number hours worked in the week preceding the census day, annual income can be converted to hourly (or weekly) wages—if one assumes that both the number of weeks worked in the year preceding the census day, and the number of hours worked in the week preceding the census day, are typical. Even with that assumption, information on weeks worked and work hours is only available for a sub-sample of the Canadian population (i.e., those who completed the long-form census).

The long-form census includes a number of variables that correspond to key predictors of the gender pay gap, including: age, family composition, education, labour market activity, full-time and part-time weeks worked, class of worker, occupation, and industry. It also includes indicators of diversity characteristics beyond sex: Aboriginal status; immigrant status, birthplace, year of immigration, and immigrant generation and birthplace of parents; visible minority status; and religion (asked every five years, with the next time being in 2021). Importantly, the large sample size for long-form census enables analysts to produce reliable estimates of outcomes among small population groups or for low levels of geography.

Labour Force Survey (LFS)

The Labour Force Survey (LFS) is a household survey conducted monthly by Statistics Canada. It provides timely information on major labour market trends by dividing the working-age population (aged 15 and older) into three mutually exclusive categories—employed, unemployed and not in the labour force—and collecting data on a variety of demographic, socioeconomic, and employment characteristics.

The LFS is based on a sample of approximately 56,000 households (corresponding to around 100,000 individuals) each month. It uses a rotating panel sample design, meaning that households remain in the sample for six consecutive months, with one-sixth of the sample being replaced each month. Persons living on reserves and other Aboriginal settlements in the provinces, full-time members of the Canadian Armed Forces, the institutionalized population, and households in extremely remote areas with low population density are not covered by the LFS.

Since 1997, the LFS has collected data on weekly and hourly wages. It is the only data source to do so. The LFS contains a number of variables that correspond to key predictors of the gender pay gap, including: age, age of the youngest child in the household, education, labour force status, work hours, class of worker (i.e., self-employed, unpaid family worker, public-sector employee, and private-sector employee), job tenure, union status, occupation, and industry. However, the LFS is more limited when it comes to identity factors beyond sex. Whilethe LFS includes Aboriginal status, immigrant status, birthplace, and year of immigration, there is no variable pertaining to racialization.

It is often useful to work with annual estimates, instead of monthly ones, particularly for descriptive purposes. Calculated on the basis of 12 months of data, annual estimates avoid the problem of seasonality that affects unadjusted monthly estimates. Annualizing the monthly data can be done in three steps:

  1. For a given calendar year, stack or append its 12 months of data (i.e., merge by adding cases).
    • There will now be one large dataset, containing a year’s worth of responses.
    • Some respondents will have been observed up to six times, depending on the rotation schedule.
      • The survey date (SYEAR and SMTH) in combination with household ID (HHLDID) and person ID (LINE) can be used to uniquely identify respondents, if need be.
  2. Divide the weight variable (FINALWT) by 12.
  1. Apply the weight variable, and calculate estimates as you would using a single month of data.Note

Canadian Income Survey (CIS)

The Canadian Income Survey (CIS) provides information on the annual income and income sources of Canadians, along with their individual and household characteristics. The CIS is administered to a sub-sample of LFS respondents, consisting of the outgoing rotation groups in January, February, March and April. Each of these rotation groups consists of approximately 8,400 respondents.

The CIS gathers information about labour market activity, school attendance, disability, support payments, child care expenses, inter-household transfers, personal income, and characteristics and costs of housing. This content is supplemented with information on individual and household characteristics from the LFS. CRA data on annual income and income sources are also combined with the CIS data.

Although the first CIS was conducted in 2012, the data series can be extended using its predecessors, including the Survey of Labour and Income Dynamics (SLID) and the Survey of Consumer Finances (SCF).

Longitudinal and International Study of Adults (LISA)

The Longitudinal and International Study of Adults (LISA) is a household survey that has collected data every two years since 2011 from the non-institutionalized population aged 15 or older living in Canada’s ten provinces. LISA is made up of two parts: (1) the interview portion, which collects information on respondents’ education, work, health, life events, pensions and finances, and (2) its linkage to administrative data from a variety of sources, as appropriate.

While the LISA is a relatively new data source, its longitudinal nature lends itself to being one of the richest data sources available in the study of the gender pay gap. In addition to personal and productivity-enhancing characteristics, the LISA includes other variables that may help us to better understand the gender pay gap: actual work experience, field of study, and variables pertaining skills used at work and job flexibility. It also includes indicators for identity factors beyond sex: Aboriginal status; immigrant status, year of immigration, birthplace, and birthplace of parents; and visible-minority status.

General Social Survey (GSS)

Since 1986, the General Social Survey (GSS) has gathered information on social trends in order to monitor changes in the living conditions and well-being of Canadians. The target population consists of all non-institutionalized persons 15 years of age or older, living in the 10 provinces of Canada. Survey content rotates, with themes generally recurring every five years. Notably, GSS cycles on “families” collect retrospective data on union formation and dissolution, childbearing, and work history. Although income data in the GSS pertain only to the reference year, Cycle 25 of the GSS (2011) has recently been linked to administrative tax data, enabling analysts to study the gender pay gap longitudinally in relation to time-varying variables. Standard predictors of the gender pay gap (i.e., demographic and productivity-enhancing characteristics, occupation, and industry), as well as a variety of identity factors, including sexual orientation, are available in the GSS.

Canadians at Work and Home, cycle 30 of the GSS (2016), provides information that is not typically found in labour/income surveys, yet which can yield greater knowledge of the challenges women face with respect to paid work: work load, work ethic, work environment, perceived unfair treatment at work, and division of chores within the household.

Longitudinal Worker File (LWF)

The Longitudinal Worker File (LWF) is an administrative data set consisting of a 10% random sample of all Canadian workers. It is constructed from four separate data sources: the T4 and T1 files from Canada Revenue Agency (CRA), the Record of Employment (ROE) files from Human Resources and Skills Development Canada (HRSDC), and the Longitudinal Employment Analysis Program (LEAP) file constructed by Statistics Canada. The current version of the LWF provides longitudinal information on individuals over the period from 1983 to 2015. In addition, an older, but comparable, version covers the period from 1978 to 1989.

The LWF provides the sample size necessary to examine the gender pay gap at the upper end of the earnings distribution. However, the lack of predictors other than sex and age limits its usefulness.


Date modified: