AEA 2018 Meetings, Philadelphia, January 5th 2018

Top Earnings Inequality and the : , , and the

Nicole Fortin School of and Canadian Institute for Advanced Research

1 AEA 2018 Meetings, Philadelphia, January 5th 2018

Top Earnings Inequality and the Gender Pay Gap: Canada, Sweden, and the United Kingdom and with the collaboration of Aneta Bonikowska and Marie Drolet with Brian Bell, Kings’ College London and Michael Boehm, University of Bonn

2 Data

Appeal to administrative/income tax data to capture the highest incomes Use all earnings data from income tax data available in the Canadian Longitudinal Worker Files (LWF, 1983-2010) Utilize similar annual earnings from administrative data from Sweden (LISA, 1990-2013) and for the United Kingdom (ASHE, 1999-2015) To include additional covariates, the analysis is supplementedby hourly wage data from public use Canadian (CAN-LFS, 1997-2015) and UK Labour Force Survey (UK-LFS, 1993-2015) Focus on workers 25 to 64 years old, exclude self-employment income and too low earners.

Go to data details 3 Slower Convergence in Share of Women among Top Earners in Canada

Share of Women in Selected Percentiles of Annual Earnings 0.60

0.50

0.40

0.30

0.20

0.10

0.00 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 All Bottom 90% Next 9% Next 0.9% Top 0.1%

Source: Fortin, Drolet and Bonikowska (2017), LWF 1983-2010, 25-64 years old, Annual earnings from all jobs 4 Similar Trends in Female Shares in Sweden and the UK

Share of Women in Selected Percentiles of Share of Women in Selected Percentiles of Annual Earnings - Sweden Annual Earnings – United Kingdom 0.60 0.60

0.50 0.50

0.40 0.40

0.30 0.30

0.20 0.20

0.10 0.10

0.00 0.00 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014

All Bottom 90% Next 9% Next 0.9% Top 0.1% All Bottom 90% Next 9% Next 0.9% Top 0.1%

5 Increasing Earnings Inequality in Top Incomes and the Gender Pay Gap

Questions of interest: 1) What are the consequences of the under-representation of women in top jobs (top decile) for the overall average gender pay gap? 2) Given recent starkling increases in top incomes, is this under- representation contributing to the slowdown in the convergence of female/male pay? 3) Are the popular “Women on Board” policies and disclosures are effective to improve this under-representation?

6 Increasing Earnings Inequality in Top Incomes and the Gender Pay Gap

Several papers (Blau and Kahn, 1992, 1994; Fortin and Lemieux, 2000) have explored the consequences of rising wage dispersion (increased variance) for the gender pay gap When residual inequality experienced stupendous increases in the 1980s, Blau and Kahn (1997) coined the term “swimming upstream” to characterize women’s pursuit of pay equality in the face of countervailing currents. Have recent increases in top incomes (increased skewness) lead to similar effects, therefore accounting for the slower progress in the gender pay and growing unexplained (by traditional factors) share?

7 Larger Increases for Top Earners in Canada

Canadian Annual Average Earnings ($CAN 2010) 500,000 2,300,000 450,000 6%/yr 2,100,000 400,000 1,900,000

350,000 1,700,000

300,000 1,500,000

250,000 3%/yr 1,300,000 200,000 1,100,000 150,000 1.25%/yr 900,000 100,000 700,000 50,000 0.5%/yr 500,000 0 300,000 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 Bottom 90% Next 9% Next 0.9% Top 0.1% (right axis) Source: Fortin, Drolet and Bonikowska (2017), LWF 1983-2010, 25-64 years old, Annual earnings from all jobs

Source: LWF 1983-2010, 25-64 years old, Annual earnings from all jobs 8 Some Increases in Lower Groupings in Sweden

Swedish Annual Average Earnings (SEK 2010)

1600 5000

1400 4500 4000 1200 5.8%/yr 3500 1000 3.1%/yr 3000 800 2.3%/yr 2500 SEK 100,000 SEK 10,000 600 2000 1500 400 1000 200 1.8%/yr 500 0 0 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 Bottom 90% Next 9% Next 0.9% Top 0.1% (right axis)

Source: Fortin, Bell, and Boehm (2017), LISA data. 9 Top Income Groups Hit by the Financial Crisis in the UK

United Kingdom ─ Average Annual Earnings (₤2010)

900,000 200,000

175,000 800,000 0.5%/yr 150,000 700,000

125,000 600,000

100,000 0.11%/yr 500,000 75,000 400,000 50,000 1.1%/yr

25,000 300,000 0.65%/yr 0 200,000 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 Bottom 90% Next 9% Next 0.9% Top 0.1% (right axis)

Source: Fortin, Bell, and Boehm (2017), ASHE data. 10 Increasing Earnings Inequality in Top Incomes and the Gender Pay Gap

Apply the approach used in the analysis of earnings inequality in top incomes (developed by Thomas Piketty, Emmanuel Saez, and co-authors) to characterize top earners Use positional ranks to construct a proxy of vertical segregation at the top of the earnings distribution (Fortin and Lemieux, 1998; Bayer and Charles, 2016) Apply reweighing techniques à la DiNardo, Fortin and Lemieux (1996) [DFL] to construct counterfactual gender pay gaps

Caveat: Analysis remains an accounting “what if” exercise to quantify the relative importance of the potential swimming upstream effects

11 Measure of the gender pay gap

“Hourly Wage” ratio is the preferred measure to consider whether employers treat women fairly and should be used in statements “women earn 85 cents out (86 öre/82p) of every $1 (1kr/£1) men earn” “Annual (Weekly) Earnings of Full-Time Workers” ratio ≈ 70% in Canada and ≈ 64%* in the UK Because many women working full-time full-year work less hours a week than men mixes the number of hours worked with hourly pay But for the very top income groups, the “All Annual Earnings” measure is the only one available (from tax data) “Annual Earnings” ratio ≈ 65% in Canada, ≈ 74% in Sweden, and 62%* in the UK It gives a better idea of costs of women’s lower labour supply or impact of bonuses

Source: Sweden: Eurostat (2015); *UK: Dias, Elming and Joyce (2016) 12 Generational Effects in the Gender Pay Gap

Canada - Gender Ratio Hourly Wages Canada - Gender Ratio in Annual Earnings

0.95 0.95 31 31 0.85 0.85 41 31 31 41 0.75 0.75

0.65 0.65

0.55 0.55

0.45 0.45 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 1983 1986 1989 1992 1995 1998 2001 2004 2007 2010 1935-39 1940-45 1946-53 1935-39 1940-45 1946-53 1954-58 1954-58 1959-65 1966-75 1959-65 1966-75 1976-85 All

Source: Fortin (2017), LFS public use data, ages 25 to 64 year, 3-year moving average annual earnings from all jobs 13 Gender Gap in Top Incomes

Follow Guvenen, Kaplan, and Song (2014) in using the thresholds of the wage and earnings distribution for men and women combined Depart from the traditional literature on the glass ceiling which compares the pay gap at percentiles of the gender-specific distributions Depart from most of the literature which uses the logarithm of wages or earnings in order to emphasize the top end Allow for the construction of counterfactuals to study the under- representation of women in top income groups: what if women were distributed in the top decile in the same way men are?

14 Under-representation of women in top jobs makes for a less favorable overall gender pay ratio

Female-Male Average Hourly Wages Ratios - Female-Male Average Annual Earnings Ratios - Canada Canada 1.10 1.10

1.00 1.00

0.90 0.90

0.80 0.80

0.70 0.70

0.60 0.60

0.50 0.50 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 1983 1986 1989 1992 1995 1998 2001 2004 2007 All Bottom 90% Next 9% All Bottom 90% Next 9% Next 0.9% Top 0.1% Median Next 0.9% Top 0.1%

Source: LFS 1997-2015, 25-64 years old, Hourly wages from the main job Source: LWF 1983-2010, 25-64 years old, Annual earnings from all jobs 15 Similar Differences in Ratios in Sweden and the UK No Upward Trend in Gender Earnings Ratio in Top 0.1%

Female-Male Earnings Ratios - Sweden Female-Male Earnings Ratios - United Kingdom 1.20 1.20

1.10 1.10

1.00 1.00

0.90 0.90

0.80 0.80

0.70 0.70

0.60 0.60

0.50 0.50 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014

All Bottom 90% Next 9% Next 0.9% Top 0.1% All Bottom 90% Next 9% Next 0.9% Top 0.1%

Source: Fortin, Bell, and Boehm (2017), workers ages 25 to 64, LISA data for Sweden, ASHE data for the UK. 16 Table 1. Average Professorial Salaries at UBC in 2010 Gender Rank Numbers % of All % of Average Female/ Gender women Salary Male Ratio Gap Men All 968 100 134955 0.89 14332 Women All 419 100 30.2 120623 Men Full 501 51.8 152494 0.96 6446 Women Full 130 31 20.6 146048 Men Associate 297 30.7 121483 0.94 6888 Women Associate 184 43.9 38.3 114595 Men Assistant 170 17.6 106806 0.93 7097 Women Assistant 105 25.1 38.2 99709

If the proportion of women across professorial ranks was identical to men, the overall counterfactual average female salary would be: 51.8/100×146048 + 30.7/100×114595 + 17.6/100×99709 =128259.3, and the overall ratio would be 128382/134955(*100)=95%  The salary gap explained by rank is 128259.3 120623.1 =7636.2  More that 53% of the gap is accounted for by the gender differences in the proportion of faculty members across rank. − Table 2. Oaxaca-Blinder Decomposition of Average Professorial Salaries at UBC in 2010

Variables: Model 1 % of gap Model 2 % of gap

Raw Gender Salary Differentials 14332.24 *** 14332.24 *** Accounted for by differences in characteristics Professorial Rank 7636.226 *** 53.28% 6647.376 *** 46.38% CRC, DUP 546.2663 * 3.81% Years in Rank 1180.126 ** 8.23% Departmental Dummies 3093.223 ** 21.58% Total Explained 7636.226 *** 53.28% 11466.99 *** 80.01% Total Unexplained 6696.018 *** 46.72% 2865.253 *** 19.99% Note: Using female coefficients. *** p<0.01, ** p<0.05, * p<0.1 See UBC (2011) for alternative specifications.

 The more complete specification accounts for 80% of the gap, 46% of which from vertical segregation and 22% from horizontal segregation.  This leaves an unexplained gender gap of 2.2% of average professorial salary 18 If the shares of women in percentiles grouping* were the same as men’s, the gap in annual earnings would be 20 point lower

Female-Male Annual Earnings Ratios - Canada 1.0 0.42 0.33 0.9

0.8 0.19 = 58%

0.7 0.19 = 45% 0.6

0.5

0.4 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 Simulated Ratio Actual Ratio *percentiles grouping: Source: Fortin, Drolet and Bonikowska (2017), LWF 1983-2010, 25-64 years old, Annual earnings from all jobs bottom 90%, next 9%, next 0.9%, top 0.1%

19 If the shares of women in percentiles grouping* were the same as men’s, the gap would be 50% lower

Female-Male Annual Earnings Ratios - Female-Male Annual Earnings Ratios – Sweden United Kingdom 1 1 0.43 0.34 0.32 0.26 0.9 0.9

0.8 0.14 = 53% 0.8 0.15 = 45% 0.18 = 53% 0.7 0.7 0.22 = 48% 0.6 0.6

0.5 0.5

0.4 0.4 1990 1993 1996 1999 2002 2005 2008 2011 2014 1990 1993 1996 1999 2002 2005 2008 2011 2014 Actual ratio Simulated ratio Actual ratio Simulated ratio

Source: Fortin, Bell, and Boehm (2017), workers ages 25 to 64, LISA data for Sweden, ASHE data for the UK. Go to alternative partitions 20 If the shares of women in percentiles grouping* were the same as men’s, the gap would be 6-9 points lower

Counterfactual Hourly Wage Gender Ratio - Canada 1.00 0.19 0.15 0.95

0.90 0.07 = 45% 0.85 0.08 = 41% 0.80

0.75

0.70 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 Actual Ratio Simulated Ratio *percentiles grouping: Source: Fortin (2017), LFS 1997-2015, 25-64 years old, Hourly wages on the main job bottom 90%, next 9%, next 0.9%, top 0.1%

21 Against traditional factors in O-B decomposition, centile groupings remain dominant and growing over time 120%

100% Total Unexplained 80% Selected Centiles Occupation 60% 48% Industry 36% 40% 44% Part-time, Tenure 36% 17% 37% 20% Demographics

0% Region 1990 2010 1997 2015 1997 2015 -20% Sweden - LISA UK- LFS Canada - LFS Note: Entries are male/female differences in the explanatory variables multiplied by the corresponding female coefficients as a share of the gender pay gap. 22 In provinces with Pay Equity, the explanatory power of occupations, like education, has gone negative 130%

110% Total Unexplained 90% Selected Centiles 70% Occupation Industry 50% 34% Part-time, Tenure 30% 34% 28% Education 30% 29% 37% 11% Demographics 10% 4% Province -6% -2% -5% -10% 1997-98-99 2013-14-15 1997-98-99 2013-14-15 1997-98-99 2013-14-15 -27% -30% Quebec Ontario ROC Note: Entries are male/female differences in the explanatory variables multiplied by the corresponding female coefficients as a share of the gender pay gap. 23 Findings

In Canada (except for Quebec), Sweden, and the United Kingdom, the under-representation of women in partitions of the top decile accounts for a predominant and growing share of the gender pay gap. By itself, this under-representation accounts for 45% (circa 1990) to 58% (circa 2010) of the gender earnings gap in the three countries Pitted against traditional explanatory factors, it stills account for 36% (circa 2000) to 48% (circa 2015) of gap (Sweden and UK), from 17% (circa 1985) to 37% (circa 2010) in Canada Increasing Women’s Representation in Top Jobs

With increasing earnings inequality in top incomes, further improvements in vertical segregation, “relatively more women in top jobs” will be likely be even more important for further decline in the gender pay gap in the 21st century How do we get there? Country fixed effects DD models of the impact of quotas for “Women on Boards” show direct effects of 50%, but no significant trickle down Leaving open how to improve women’s representation in top jobs?

25 Women’s Quotas for Corporate Boards?

Women on Boards and Employment Share 40

NONONO

30 NO SESE FIFI SESE FI

20 FI PH DKFR FRNL DK NZ NLDK PLAUAUNZUSDE DK PL USDENL GBCACACA PH NZATUSGB THESTH ATDE GB

10 CH THCH ATCHESBE BEHU HKFR IE GRPH ITCZPLAUIE ESGBHK IEIE Womenon Boards (%) MY GR GR DE FR ID MY SG NLSGCHBE AT HU MY BECZ RURU ID GR ESIT IT HU PT RU IT PT JPJPJP PT 0

40 45 50 Female Employment Share (%)

No Regulation or Quota Pre-Quota Post-Quota Pre-Regulation Post-Regulation Fitted values

Years: 2006, 2009, 2011, 2013, 2014

Source: Dizik, 2015 Source: Fortin, Bell, and Boehm (2017) 26 Higher Representation of Women in Tops Jobs Do Quotas Help?

Yes, for Dependent Variable: Women on Boards Women in Senior Management Women on Mean 11.14 12.12 29.73 29.86 Explanatory Variables Boards Quotas 5.22 5.48 0.15 0.83 (1.17) (1.25) (1.05) (0.94) but no Disclosure Rules 2.15 2.31 (1.12) (1.05) evidence of (0.95) (1.09) (0.69) (0.69) Relative Female 50.66 53.77 (6.87) (12.41) trickle down Employment Rate (20.38) (21.2) (37.81) (37.79) in country- Log GDP per capita 2.57 5.64 (0.62) (0.25) (PPP) (4.26) (6.69) (3.61) (3.20) fixed effects R-square 0.27 0.36 0.13 0.29 models No. of observations 224 173 213 195 OECD only No Yes No Yes No.of countries 40 29 27 23 Note: Dependent variables are the share of women on corporate boards from BoardEx data (European PWN, 2008) from 2006 to 2009, from GMI data (Gladman and Lamb, 2013) from 2009 to 2014, and the share of women in senior management from ILO (2014). The data on the relative female employment rate, computed as the ratio of female employment rate to the total employment rate, is from the World Bank. Estimates from country Source: Fortin, Bell, and fixed-effects models with robust standard errors clustered at the country level. ∗*∗p < 0.01, ** p < 0.05, * p < 0.1. Boehm (2017)

27 Thank you!

28 Steep Growth in Women’s Labour Force Participation* Followed by a Leveling-Off

Canadian Labour Force Participation Rate - Ages 25 to 64 1.0

0.9

0.8 8.7 0.7

0.6

0.5 Participation Rate Participation 0.4

0.3

0.2 1975 1979 1983 1987 1991 1995 1999 2003 2007 2011 2015 Men Women Source: Fortin, Drolet and Bonikowska (2016), LFS Public use files, 1976-2015 *Labour force participants include employed (at work or on-leave) and unemployed individuals 29 Less Convergence in Gender Gap in Hours

Canada - Gender Ratio in Average Total Weekly Hours 1.00

0.95

0.90

0.85

0.80

0.75

0.70

0.65

0.60 1975 1979 1983 1987 1991 1995 1999 2003 2007 2011 2015 <1920 1921-39 1940-45 1946-53 1954-58 1959-65 1966-75 1976-85 1985-90 All

Source: Fortin (2017), LFS data, ages 25 to 64 year, employed with positive hours of work, usual hours from all jobs 30 Canadian Data

Longitudinal Worker File (LWF) Labour Force Survey (LFS) Public Use LWF is a 10% random sample of all Monthly survey on approximately Canadian workers 100,000 individuals rotating 6- Years: 1983-2010 months panel sample design integrates data from the T1 and T4 files of Canada (CRA) and the LEAP Years: 1997-2015 (Statistics Canada) Hourly wage of employees from Annual earnings from all jobs, main job include bonuses, honorariums, etc. Selected if > half the minimum Selected if > half of minimum wage earnings equivalent wage Select workers age 25 to 64 Select workers age 25 to 64

31 Canadian Data

Longitudinal Worker File (LWF) Labour Force Survey (LFS) Public Use No self-employment income No self-employment income No labour supply information Number of weeks worked unavailable Top coded at P99.99 ≈ $2,000,000 in 1983 to ≈$10,000,000 in 2000 Imputed > P99.9 from ≈$95/hour in 1997 ≈ $125/hour in 2015 Available covariates: union At 2080 (=52wk*40hrs) hrs/year, coverage, age, industry from $200,000 to $260,000 CPI adjusted to 2010$CAN Available covariates: age, union, education, occupation, industry, firm size, etc.

32 Swedish Data British Data

Longitudinell Integrationsdatabas för Annual Survey of Hours and Sjukförs äkrings-och Arbetsmarknadsstudier Earnings (ASHE) (LISA) 1% panel of workers based on Integrates data from Statistics Sweden, social security number the Social Insurance Agency and the Statutory filing required by Swedish Agency for Innovative Systems employers annual earnings data are from the Years: 1999-2015 largest source of income, include Annual earnings includes all cash compensation, including bonuses performance pay and bonuses, exclude etc. self-employment income Selected if > half of minimum Years: 1990-2013 wage annual earnings equivalent Select workers age 25 to 64 Select workers aged 25 to 64 2.5-3 million observations per year Go back Average Female/Male Earnings Ratio Alternative Partitions 50 60 70 80 90 100 1980 b90 p9095 p9599t01 n09 b90 p9099 n09 t01 Ratio Actual 1985 1990 1995 A.Canada Year 2000 b80 p8090 p9099t01 n09 b90 p9099 n05 t05 Ratios: Simulated 2005 2010 2015

Average Female/Male Earnings Ratio 50 60 70 80 90 100 1980 Actual Ratio Actual b90 p9099 n09 t01 b90 p9095 p9599t01 n09 1985 1990 1995 B.Sweden Year 2000 Go back Simulated Ratios: Simulated b90 p9099 n05 t05 b80 p8090 p9099t01 n09 2005 2010 2015 Alternative Partitions

Average Female/Male Earnings Ratio 50 60 70 80 90 100 1980 b90 p9095 p9599t01 n09 b90 p9099 n09 t01 Ratio Actual 1985 1990 C.United Kingdom 1995 Year 2000 b80 p8090 p9099t01 n09 b90 p9099 n05 t05 Ratios: Simulated 2005 2010 2015