gender identity and relative income within households ... · economic theory says that education...
TRANSCRIPT
2019-06
Gender Identity and Relative Income within Households: Evidence from Canada
Maéva Doumbia Marion Goussé
Juin / June 2019
Centre de recherche sur les risques les enjeux économiques et les politiques publiques
www.crrep.ca
ABSTRACT Bertrand, Kamenica, and Pan (2015) show that among married couples in the US, the distribution of the share of the household income earned by the wife exhibits a sharp drop just to the right of 50%. They argue that this drop is consistent with a social norm related to gender identity prescribing that a man should earn more than his wife. We investigate this phenomenon over the years 1996 to 2011 in Canada at the national level but also across provinces where cultural values and gender norms appear to be different. We find a similar discontinuity at the 50% threshold and show that it comes from both matching patterns and time use behavior. We find that couples form less when the probability that a woman earns more than her husband increases. We also find that wives with greater probability of earning more than their husbands are less likely to participate in the labor force and that if they participate, they have a higher probability of earning less than their potential income. Finally, we find that when women exceeded the spouse’s income equality threshold, they also increased their time spent in housework, but this effect has disappeared in recent years. We find a stronger effect of gender roles in Quebec and in the Prairies but a less important effect in British-Columbia and in Ontario. JEL Classification: D1, J1. Keywords: Gender Wage Gap, Gender Identity, Intra-Household Allocation. Maéva Doumbia : Department of Economics, Université Laval. Marion Goussé : Department of Economics, Université Laval.
Marion Goussé gratefully acknowledges financial support from FRQ-SC. Part of the analysis presented in this paper was conducted at the Quebec Interuniversity Centre for Social Statistics which is part of the Canadian Research Data Centre Network (CRDCN). The services and activities provided by the QICSS are made possible by the financial or in-kind support of the Social Sciences and Humanities Research Council (SSHRC), the Canadian Institutes of Health Research (CIHR), the Canada Foundation for Innovation (CFI), Statistics Canada, the Fonds de recherche du Québec - Société et culture (FRQSC), the Fonds de recherche du Québec - Santée (FRQS) and the Quebec universities. The views expressed in this paper are those of the authors, and not necessarily those of the CRDCN or its partners.
1. Introduction
Whereas there has been a substantial reduction in the gender pay gap in most economically ad-
vanced nations, wage convergence, increases in female labor-force participation rates and reductions
in occupational segregation by sex have plateaued or slowed since the 1990s (Blau and Kahn, 2017).
Canada is no exception. There is a persistent pay gap in every province and in every major occu-
pational group. The gap in annual earnings between men and women has barely budged over the
past two decades, even as education levels among women have surpassed those of men (Baker and
Drolet, 2010; Moyser, 2017; Schirle, 2015; Bonikowska, Drolet, and Fortin, 2019; Fortin, 2019).
It is now well known that women’s career interruptions and prevalence of part time work are the
most significant factors in explaining the differences in earnings between men and women. A more
challenging question is the existence of a persistent gender gap in the annual number of working
hours. In spite of the rise in female participation in the labor market, men spend more time in
paid work, and women more time in housework: the gender specialization within couples remains
a quasi-universal norm. New research look into how the compliance to “gender roles” may explain
the gender division of labor (Bertrand, Kamenica, and Pan, 2015; Fortin, 2015; Bursztyn, Fujiwara,
and Pallais, 2017; Lippmann, Georgieff, and Senik, 2019). Bertrand, Kamenica, and Pan (2015),
hereafter BKP, show that the distribution of the share of income earned by women in American
households falls strikingly at the right of 0.5, the point where the wife and the husband contribute
equally to the total income. This drop highlights a phenomenon of missing couples, a lack of couples
where the wife earns more than her husband. They argue that it is consistent with a social norm
related to gender identity prescribing that a man should earn more than his wife. Compliance to
gender identity may prevent some couples to form or may impact time use behavior of women who
reduce their participation on the market and increase their housework when their income exceed the
income of their husband.1
In this paper, we investigate the effect of gender identity on marriage and time use behavior of
married women in Canada at the national level but also across provinces. We define gender identity
1Akerlof and Kranton (2000) import the concept of identity into economics. In particular, theyimport the concept of gender identity and the potentially competing identities of “homemaker” andof “career woman”.
2
as the traditional belief that the man should have a higher income than his wife in order to preserve
the traditional structure of the household in which the man is the main financial support of the
family. Our contribution is twofold. First, we show that the discontinuity at the spouse’s income
equality threshold also exists in Canada and comes from the impact of gender identity on both
matching pattern and time use behavior. Second, we document variations of gender norms across
Canadian provinces and show how they are related to different behaviors on the marriage market
and labor household division.
The existence of an aversion toward a situation where the wife has an income higher than that
of her husband is clearly identified in the wave 3 of the World Value Survey collected in 1995. Re-
spondents of the survey are asked whether they agree or not with the statement “If a woman has a
higher income than her husband, it will necessarily cause problems”. In the US, where BKP high-
light the existence of the missing couples phenomenon, 37.3% of American respondents agreed to
this statement. In Germany the agreement rate reached 43.4% and there also exists a discontinuity
in the distribution of share of the income earned by women at the 50% threshold (Wieber and Holst,
2015; Lippmann, Georgieff, and Senik, 2019). Interestingly, the discontinuity is much higher in West
Germany than in East Germany where gender norms are much less traditional.2 On the contrary,
Hederos Eriksson and Stenberg (2015) do not find such a discontinuity in Sweden although the agree-
ment rate was 30.9% among Swedish respondents. In Brazil, the agreement rate was 33.5% and a
large discontinuity is also found at the 50% threshold (Codazzi, Pero, and Albuquerque Sant’Anna,
2018).
In this paper, we first document gender norms in Canada using the European Values Study (EVS)
and the World Value Survey (WVS). Unfortunately, Canada has not been surveyed in 1995 so we do
not know the agreement rate to the statement “If a woman has a higher income than her husband,
it will necessarily cause problems” but we consider other statements related to gender norms in our
analysis. We find that the provinces of the Prairies3 have relatively more traditional gender norms
2Using the 41-year division of Germany as a natural experiment, Lippmann, Georgieff, and Senik(2019) show that East Germany gender-equal institutions created a culture that has undone themale breadwinner norm and its consequences on women labor force participation. By contrast, thenorm of higher male income is still prevalent in West Germany.3The Prairies include the provinces of Manitoba, Saskatchewan and Alberta.
3
than the other provinces whereas British-Columbia and Ontario relatively less. Second, we show
that the discontinuity at the 50% threshold also exists in Canada and is present in all provinces
except in the Atlantic provinces4 and in British-Columbia. We find a higher discontinuity in Quebec
and in the Prairies.
Two main channels may explain this missing couples phenomenon. The first channel is sorting
patterns and marriage rates as couples where the woman has a higher earning capacity than the man
may not form, or may divorce more. The second channel is time use behavior as women who would
be able to earn more than their husband may reduce their labor supply not to exceed the earnings
of their partner and comply to social norms. We investigate in turns these two mechanisms.
First, we look at how gender norms may affect marriage rate and matching patterns. Standard
economic theory says that education and earning capacity are important in determining partner’s
choice to maximize the household utility function and domestic production (Becker, 1973), however
it says nothing on which partner has to be the principal earner when both partners have the same
productivity at home and on the market. Who has to do the most of domestic production only comes
from efficiency and comparative advantages. Yet, it seems that who does what is not innocuous in
family behavior. Economic and sociological studies have shown that couples in which women work
more hours than their husband are more likely to separate (Kalmijn, Loeve, and Manting, 2007;
Heckert, Nowak, and Snyder, 1998). Experimental economics has also shown that women may hide
their career ambition in order to find a partner more easily. Bursztyn, Fujiwara, and Pallais (2017)
find that single female students report lower desired salaries and lower willingness to travel and
work long hours when they expect their classmates to see their preferences whereas married females
do not. Bertrand, Kamenica, and Pan (2015) find that an increase of one percentage point in the
probability that women earn more than men in a marriage market decreases by 0.22 percentage point
the marriage rate in this market. In this paper, using a similar methodology, we find a negative
effect of 0.26 percentage point in Canada. We find a stronger negative effect in Quebec and in the
Prairies and a non-significant effect in Ontario and in British-Columbia.
4The Atlantic provinces include Newfoundland and Labrador, Nova Scotia, Prince Edward Islandand New Brunswick.
4
Second, we look at how gender norms may affect women attachment to the labor market. In her
study on women’s participation in the labor market in OECD countries, Fortin (2005) finds that the
impact of gender identity is significant. Her analysis of the World Value Survey data suggests that
countries where people mostly agree to the statement “When jobs are scarce, men must have priority
over women in the granting of work” have low employment rates among women and have a higher
wage gap between men and women. Discriminatory attitudes towards women continue to play an
important role in slowing down access to equality between men and women in the labor market. In
a more recent paper, Fortin (2015) documents the evolution of gender role attitudes and women’s
labor market participation in the US. She first shows that traditional gender roles after a steady
decline in the 1970s and 1980s began to stabilize in the mid-1990s, even increasing in the late 1990s
for some cohorts. She shows that the progression of gender role attitudes explains at least a third of
the recent leveling-off in female labor force participation. Bertrand, Kamenica, and Pan (2015) find
that an increase of one percentage point in the probability that a wife earns more than her husband
decreases by 0.14 percentage point her labor force participation. Wieber and Holst (2015) also find
a negative effect in Germany but only in West Germany for women working full time. In this paper,
we find decrease of 0.09 percentage point in Canada using a similar methodology. We find similar
effects across provinces.
Finally, we look at how gender norms may affect time spent in housework. Despite their grow-
ing presence in the labor market and the relative rise in their incomes, women remain primarily
responsible for domestic work (Lachance-Grzela and Bouchard, 2010). This behavior contradicts
standard theory where women should reduce the amount of time they spend doing housework as
their income and the time they spend on the job market increase. Greenstein (2000) finds that
women housework decreases with their labor income until their labor income reaches the level of
their husband’s labor income, then their time spent in housework ceases to decrease and even in-
creases significantly. This behavior has been shown in the US (Bertrand, Kamenica, and Pan, 2015)
and in Germany but only in West Germany and not in East Germany (Lippmann, Georgieff, and
Senik, 2019). Using confidential micro-data files of the General Social Survey of 2010 and 2015 on
time use, we find a similar behavior in Canada in the 2010 survey but not in the 2015 survey anymore.
5
In the next section, we show how gender norms vary across Canadian provinces. In section 3, we
present the distribution of within couples relative income and investigate the role of gender identity
on marriage rates in section 4. In section 5 we look at the impact of gender identity on women labor
supply then we focus on housework in section 6 before concluding.
2. Gender norms in Canada
To document gender norms in Canada, we use the European Values Study (EVS) and the World
Value Survey (WVS) which are two large-scale, cross-national and longitudinal survey research
programs. They include a large number of questions, which have been replicated in 364 surveys.
Both programs have been conducted from 1981 to 20145. Canadian data are only available on
the 1990 wave of the EVS and on the wave 2000 and 2005-2008 of the WVS. The questions on
gender role attitudes are included in a section that asks respondents whether they agree strongly,
agree, disagree or disagree strongly with a number of statements. Fortin (2005) has identified four
statements which are correlated with women labor force participation in the OECD countries. She
shows that agreement with the statement that “When jobs are scarce, priority must be given to
men in the granting of work” appears to be the most powerful explanatory factor of cross-country
differences in female employment and in the gender pay gap. This statement capture perceptions
of the man as the main breadwinner, as well as anti-egalitarian views or discriminatory attitudes
against working women. Perception of the woman as homemaker, measured as agreement with the
statement “Being a housewife is just as fulfilling as working for a salary” also has a significant
explanatory power. For individual choices, she also finds that agreement with the statement that
“A working mother can establish just as warm and secure a relationship with her children as a
mother who does not work” is closely associated with the employment status and is a measure of
some literature calls “mother’s guilt”. We also look at the statement “Husband and wife should both
contribute to income”.
Using those statements, we build four dummies to compute the share of individuals who have
traditional view of gender norms. We build the dummy “Men first” which is equal to 1 if the
5EVS (2015): European Values Study Longitudinal Data File 1981-2008 (EVS 1981-2008). GESISData Archive, Cologne. ZA4804 Data file Version 3.0.0, doi:10.4232/1.12253WVS (2015). World Value Survey 1981-2014 official aggregate v.20150418, 2015. World ValuesSurvey Association (www.worldvaluessurvey.org). Aggregate File Producer: JDSystems, Madrid.
6
individual agrees to the statement “When jobs are scarce, priority must be given to men in the
granting of work” and which is equal to 0 if the individual answers that he disagrees or that he
neither disagrees nor agrees. We build the dummy “Housewife fulfilling” which is equal to 1 if
the individual answers that he agrees or strongly agrees to the statement “Being a housewife is
just as fulfilling as working for a salary” and which is equal to 0 if the individual answers that he
disagrees or strongly disagrees. We build the dummy “Warm if not working” which is equal to 1 if
the individual answers that he disagrees or strongly disagrees to the statement “A working mother
can establish just as warm and secure a relationship with her children as a mother who does not
work” and which is equal to 0 if the individual answers that he agrees or strongly agrees. Finally
we build the dummy “Both should not contribute” which is equal to 1 if the individual answers
that he disagrees or strongly disagrees to the statement “Husband and wife should both contribute
to income” and which is equal to 0 if the individual answers that he agrees or strongly agrees.
Table 1 then shows the share of individuals who have traditional view of gender norms in Canada
(at the national level and at the provincial level) and in other countries overtime. We find that
agreement rates to give priority to men in a context of scarce jobs are decreasing over time for
all countries. However, they are much lower in North America and Sweden than in France and
Germany (except in the Quebec province where the rate is particularly high). Similarly the share
of individuals who agree with the statement that both members should not contribute is decreasing
with time in all countries and provinces. This share is higher in North America than in Europe
(except in the Quebec province where the share is lower). This rate is also particularly high in
the Prairies. The share of individuals agreeing with the statement that women who do not work
may establish a warmer and more secure relationship with their children than working mother is
also decreasing with time in all countries and provinces. In 2000, it is close to 20% in all countries
(except in Sweden where it is slightly lower at around 15%). Surprisingly, in North America, the
agreement rate to the statement that being a housewife is as fulfilling than working is much higher
than in Europe and has increased overtime whereas it has decreased in Europe. In 2006, around
80% of individuals agreed to this statement in North America whereas they were only 40% to 50%
in Europe (in 1990, they were 70% in North America and between 40% and 60% in Europe). This
fact has also been noted by Fortin (2005). All in all, we remark that gender norms appear to be less
7
traditional overtime. They also appear to be less traditional in Sweden where gender identity has
no effect on within household income according to Hederos Eriksson and Stenberg (2015) and more
traditional in the US where BKP find a strong effect of gender identity on within-household relative
income. As gender norms in Canada seem to be close to the gender norms in the US, we expect to
find similar gender behavior.
Table 1. Agreement rates to gender norms statements
Men first Housewife Both should Warm iffulfilling not contribute not working
1990 2000 2006 1990 2000 2006 1990 2000 1990 2000
Canada 16.1 12.2 10.2 69.9 81.6 80.3 32.0 23.7 27.5 19.6Atlantic provinces 12.7 13.1 6.9 66.7 82.5 87.0 28.3 25.4 31.9 19.2Quebec 20.5 15.2 16.8 80.7 86.1 79.6 25.9 13.6 23.7 18.1Ontario 17.1 8.7 9.2 63.3 75.1 76.4 33.0 25.1 30.0 19.7Prairies 13.8 10.7 6.9 70.6 85.7 80.1 37.1 35.2 27.7 22.4British-Columbia 8.7 13.3 9.4 67.3 77.6 74.6 37.3 27.0 24.4 20.6United-States 20.9 8.1 5.6 72.2 78.9 79.1 33.5 30.8 25.2 19.8France 31.4 19.7 15.6 56.3 61.3 50.3 20.6 20.3 25.6 21.0Germany 27.0 22.6 16.5 41.0 34.1 39.0 25.3 17.1 50.6 23.0Sweden 6.6 1.6 1.7 60.6 47.8 47.3 13.5 11.4 27.2 15.2
Data are from the integrate survey of EVS and WVS 1981 to 2014. The sample includes individuals between 18 and
65 years of age. Atlantic provinces include Newfoundland and Labrador, Nova Scotia, Prince Edward Island andNew Brunswick. The Prairies include the provinces of Manitoba, Saskatchewan and Alberta.
Provinces of Canada differ in several characteristics which may be correlated with values. To
obtain a measure of the strength of gender norms in each province, we perform a linear regression of
each of our gender norm dummies on age, education6, sex, year and provinces fixed effects. Ontario
is the province of reference. We also perform a regression on an index built as the sum of the four
dummies previously defined. The index may then takes five values from 0 to 4. The higher the
index, the more traditional the individual. As all statements are not available each year, it is only
computed for individuals present in wave 1990 and 2000. We present the province fixed effects for
the five regressions in figure 1 (all regression results are presented in appendix in table 11). Results
do not show a clear pattern of gender norms across Canada. The province fixed effect for the index
shows that the provinces of the Prairies may be more traditional than the others. We could expect
6Education level is only available in the WVS. We use the information on the leaving school age,which is available in both the EVS and the WVS age to control for education.
8
that gender identity will play an important role in women behavior in the Prairies. The province
of Quebec is particular as it appears to be significantly more traditional than Ontario with respect
to give priority to men for jobs and with respect to the statement that housewifery is as fulfilling
as working. However, it is significantly less traditional than Ontario with respect to the statement
that working women may secure a warm relationship with children and with the statement that
both members should contribute. However, Fortin (2005) show that the former two statements are
those which matter the most in explaining women employment and full-time participation. We could
expect that gender identity will also play an important role in women behavior in the province of
Quebec. The Atlantic provinces are less traditional than Ontario with respect to give priority to
men for jobs but significantly more traditional with respect to the statement that being as housewife
is as fulfilling as working for pay. It is non significantly different from Ontario otherwise. British-
Columbia is not significantly different from Ontario. We expect gender identity to have a less
important role in British-Columbia and Ontario than in other provinces.
Figure 1. Gender norms fixed effects across Canadian provinces
-0.03*
0.09***
-0.02
0.000.01
0.04***
0.10***
-0.05**
-0.09***
0.01
-0.01
0.07***
0.01
0.08***
0.13**
-0.02
0.01
-0.02
0.04
0.01
Men F irst Housewife fu l f i l l ing
Warm i f not working
Both should not contr ibute
Index
Atlantic Quebec Prairies British-Columbia
Note : Data are from the integrate survey of EVS and WVS 1981 to 2014. The sample includes individuals between18 and 65 years of age. Atlantic provinces include Newfoundland and Labrador, Nova Scotia, Prince Edward Islandand New Brunswick. The Prairies include the provinces of Manitoba, Saskatchewan and Alberta. The bars show the
province fixed effects obtained after regressing gender norms dummies on age, sex and education of respondents,year and province fixed effects. All regression coefficients are presented in appendix in table 11. ***significant at 1%
level, **at 5%, *at 10%.
9
3. Relative earnings within household
In this section, we are interested in computing the distribution of the relative income of women
in Canadian households that is the share of the household income which is earned by the wife. We
use the 1996 to 2011 cross-sections of the Survey on Labor and Income Dynamics (SLID) which
provides data on employment and earnings of Canadian individuals and families. We restrict the
data to couples7 aged between 18 and 65 year old where both members earn positive labor market
earnings. We then compute the relative income as
relativeIncomei =wifeIncomei
wifeIncomei + husbIncomei
where wifeIncomei and husbIncomei respectively represent the individual labor market earnings
of the wife and the husband in couple i and includes income from self-employment.
The distribution of these relative incomes are presented in figure 2. The relative income is divided
by twenty bins of 0.05. The blue dots on the densities represent the fraction of couples in each bin.
We use a lowess function, and estimate the distribution of the relative income to the left and right
of the 0.5 point. We observe a large discontinuity at the 0.5 threshold as in BKP.
The estimation of the discontinuity is sensitive to the presence of a mass of couples at the 0.5
threshold in the distribution (Binder and Lam, 2018; Hederos Eriksson and Stenberg, 2015). Ex-
cluding couples in which both members earn the same income may eliminate the discontinuity. In
our data, we observe an important mass point at the 0.5 threshold as 1.3% of couples earn the
same income. We will consequently test for the presence of the discontinuity both in including and
in excluding same-income couples. We perform McCrary tests (McCrary, 2008) to ensure that the
observed fall corresponds to a break in the distribution density. The principle of this test is to esti-
mate non-parametrically the density to the left (f−) and to the right (f+) of the threshold where a
discontinuity is suspected. The size of the discontinuity is then equal to θ = ln f+ − ln f−. We test
for a break at the 0.5 threshold when we exclude same-income couples whereas we test for the break
at the right of the threshold (at 0.501) when we include same-income couples. We present in figure
3 the resulting estimated densities with solid lines curves. Figure 3.A presents the results when we
include same-income couples from the sample and figure 3.B the results when we exclude them. We
7We include married and unmarried couples.10
observe on figure 3.A the presence of a mass point at 0.5 where the wife and the husband earn the
same income. In both graphics, we observe a discontinuity. Binder and Lam (2018) show that the
densities estimated with McCrary algorithm and consequently the test results are sensitive to the
choice of the bandwidth. We present test results for two different bandwidths: the optimal band-
width h∗ chosen by the McCrary algorithm and a smaller bandwidth equal to 0.75h∗. We present in
table 2 the size of the discontinuities estimated both with and without same-income couples and with
different bandwidths. In all cases, we find a significant discontinuity at the 0.5 threshold. However,
the discontinuity is smaller when we exclude same-income couples. Using the optimal bandwidth as
in BKP, we find a discontinuity of -0.102 (pvalue < 0.01) in Canada which is very close to the -0.12
found by BKP. The discontinuity varies across provinces. It is much higher in the Prairies (-0.176,
pvalue < 0.01) and it is non significant in the Atlantic provinces and British-Columbia8
We have identified a missing couples phenomenon at the 50% threshold. We now want to inves-
tigate how it is related to gender identity that is to a social norm prescribing that a man should
earn more than his wife. It may come from the impact of gender identity on matching patterns or
on time use behavior. We investigate these two channels in the following sections.
Figure 2. Distribution of Relative Income
Data are from the 1996 to 2011 SLID cross-sections. The sample include married and cohabiting couples where bothpartners earn positive income and are between 18 and 65 years of age. Each dot is the fraction of couples in a 0.05
relative income bin. The vertical line indicates the relative income share = 0.5. The solid line is the lowess smoother
applied to the distribution allowing for a break at 0.5.
8In appendix, we present on figure 4 and 5 the estimated densities across provinces.11
Figure 3. Discontinuity at 0.5 in Relative Income
05
1015
-.5 0 .5 1 1.5
(a) With same income couples
02
46
8-.5 0 .5 1 1.5
(b) Without same income couples
Data are from the 1996 to 2011 SLID cross-sections. The sample include married and cohabiting couples where bothpartners earn positive income and are between 18 and 65 years of age. Estimated densities to the left and right of
the threshold are estimated using McCrary algorithm with a bin size of 0.0011. On the left panel, same income
couples are included, and the break is set at 0.501. The optimal bandwidth h∗ = 0.159 is used. On the right panel,same income couples are excluded, and the break is set at 0.5. The optimal bandwidth h∗ = 0.155 is used.
Table 2. Discontinuity estimates
Canada Atlantic Quebec Ontario Prairies British-Columbia
Same-income couples included, break at 0.501h∗ -0.353*** -0.237*** -0.345*** -0.398*** -0.408*** -0.328***
(0.015) (0.028) (0.035) (0.027) (0.032) (0.054)0.75× h∗ -0.399*** -0.219*** -0.393*** -0.426*** -0.453*** -0.400***
(0.017) (0.032) (0.040) (0.031) (0.036) (0.061)Bin size 0.0011 0.0025 0.0025 0.0020 0.0022 0.0037Automatic bandwidth h∗ 0.159 0.259 0.162 0.143 0.143 0.136
Same-income couples excluded, break at 0.5h∗ -0.102*** 0.015 -0.116*** -0.122*** -0.176*** -0.058
(0.016) (0.037) (0.038) (0.031) (0.031) (0.055)0.75× h∗ -0.064*** 0.060 -0.096** -0.068* -0.143*** -0.018
(0.019) (0.044) (0.044) (0.036) (0.036) (0.063)Bin size 0.0011 0.0025 0.0025 0.0020 0.0022 0.0038Automatic bandwidth h∗ 0.155 0.166 0.156 0.145 0.175 0.154
Data are from the 1996 to 2011 SLID cross-sections. The sample include married and cohabiting couples where both
partners earn positive income and are between 18 and 65 years of age. Atlantic provinces include Newfoundland and
Labrador, Nova Scotia, Prince Edward Island and New Brunswick. The Prairies include the provinces of Manitoba,Saskatchewan and Alberta. Standard errors are reported in parenthesis. ***significant at 1% level, **at 5%, *at
10%.
12
4. Marriage market
We analyze in this section how the marriage rate varies with the distribution of potential incomes
of men and women. We group individuals into marriage markets under the criterion of homophily.
How the changes in the woman’s income relative to that of a man in a marriage market affects the
marriage rate in this market ? We first define M marriage markets where individuals have the same
age group, level of education and province. We define three levels of education (primary education,
secondary education and some superior education) and three age groups which differ for men and
women as usually men marry younger women. We define the age group by 18-35, 36-45 and 46-65
year old for men and 18-33, 34-43 and 44-65 year old for women. We then obtain 3 × 3 × 10 = 90
marriage markets each year, that is 90×16 = 1440 observations. In table 12 in appendix, we present
how couples sort within different groups. More than 70% of men marry within their age group and
more than 50% of men marry within their education group.
For each marriage market m ∈ [1;M ] and year t, we calculate the probability that when a woman
encounters a man, she has an higher income than his. We proceed as follows. We randomly select
couples among each market. That is, in each market for each year, we replicate 20 times each
individual observation and we match men and women randomly. Then we compute the rate of women
earning a higher income than their assigned husband to obtain the variable PrWomanEarnsMoremt
which is then the probability that women earn more than men in the marriage market m at time
t. For each marriage market and year t, we also compute the observed marriage rate of men. Table
13 in appendix presents descriptive statistics of these variables. The average marriage rate is 66.8%
and the average probability that a woman earns more than a man is 35.5%. Then we perform a
linear regression of the marriage rate on the variable PrWomanEarnsMoremt. The causal effect
is identified if there are no omitted variables that are simultaneously correlated with the marriage
rate and the probability that women earn more than men. We therefore add several sets of variables
of controls and fixed effects. All regressions include the log of the average men income, the log
of the average women income and the deciles of the men and women income distributions in the
market each year. We add marriage market fixed effects, and year fixed effects interacted with age
groups, education groups and provinces. The unit of observation is a marriage market in a year.
Standard errors are clustered by province, and each observation is weighted by the number of women
13
in the marriage market. Results for this baseline specification are presented in column (1) of table
3. Column (2) adds a control for average women’s income divided by the sum of average men’s and
women’s income and includes several additional controls: the sex ratio (the share of men among the
market population), the average number of years of schooling for men and women, and the number
of men and women in the market. The impact of PrWomanEarnsMoremt on the marriage rate is
negative and significant equal to -0.28 (pvalue < 0.01).
One may still worry that the variable PrWomanEarnsMoremt is endogenous as women earn-
ings depend on their marital situations. Following BKP, we use two alternatives measures of women
income to compute the variable PrWomanEarnsMoremt and the variables of controls for income.
First, we use a measure of predicted income based on demographic characteristics. For each demo-
graphic group (gender, age, education, province and year)9, we compute the distribution of income
for working men and women. Then, we assign potential income to each men and women by drawing
from the earning distribution of those in their demographic group. Results are presented in columns
(3) and (4) of table 3. The coefficients are very stable. The impact of PrWomanEarnsMoremt
on the marriage rate is still negative and significant and is equal to -0.25 (pvalue < 0.01). Fi-
nally, we build a distribution of incomes derived from a Bartik instrumental methodology that
isolates the variation in relative income which is plausibly unrelated to the factors that directly
affect the marriage market. We use the variation of wages among different industries as shares of
men and women are different by industry. In particular, we compute predicted Bartik wages at the
p = {5th, 10th, ...95th} percentiles the following way
wB,pgeaprt =
∑j
γjgepr,1996× wj,p
geat,−pr
where g indexes gender, e education group, a age group, j industry10, t the year and pr the province.
Variable wj,pgeat,−pr
is the pth percentile of the national income distribution in year t in industry j
for workers of a given gender, education, and age group, excluding province pr. γjgepr,1996is the
9Demographic groups are defined more precisely than marriage markets. We group age into sixyear intervals and we define 5 groups of education: less than secondary, secondary education, non-university post secondary education, partial university education, university degree.10We consider 11 industry groups: agriculture and forestry; construction; manufacturing; trans-portation; trade; finance, insurance, and real estate; business, personal, and repair services; teachingservices, health care and social assistance, other services; and public administration.
14
fraction of individuals with gender g and education e in province pr who are working in industry
j, as of the base year 1996. Those percentiles are strongly correlated with predicted percentiles by
demographic group but should not be correlated with marriage market rates. Results are presented
in columns (5) and (6) of table 3. The impact of PrWomanEarnsMoremt on the marriage rate is
still negative and significant. One point of percentage increase in the probability that women earn
more than men in a marriage market decrease by 0.16 percentage point the marriage rate in this
marriage market.
We now investigate the effect of gender identity on marriage rates across provinces and we add
in the regressions the interaction variable area × PrWomanEarnsMore. We present the effect
of gender identity on marriage rate in each province in table 411. The first row gives the es-
timate for this effect that is the sum of the coefficients of variables PrWomanEarnsMore and
area×PrWomanEarnsMore when actual earnings are used and all control variables are included.
The second row gives the F-stat of a significativity test on that sum. Rows (3) and (4) show the
results when predicted income by demographic group is used in the estimation whereas rows (5)
and (6) show the results when the Bartik specification is used. We find a strong and significant
negative effect in each area when actual income is used. When using predicted income and Bartik
instrument, we find a strong and significant negative effect in the Atlantic provinces, in Quebec and
in the Prairies but a non-significant effect in British-Columbia and even a positive effect in Ontario.
Our results then highlight a negative effect of gender identity on marriage rates as in BKP. As
women have strongly increased their labor income over the last fifty years, a part of the decline in
marriage rates overtime may be explained by compliance to gender identity.
11All parameters are presented in table 14 in Appendix.15
Table 3. Potential relative income and marriage rate in Canada
Dependent variable : marriage rateActual earnings Predicted earnings Bartik instrument(1) (2) (3) (4) (5) (6)
PrWomanEarnsMore -0.324*** -0.285*** -0.276*** -0.250*** -0.126** -0.161***(0.0435) (0.0411) (0.0421) (0.0354) (0.0403) (0.0416)
Log average in men’s income 0.0218 0.0627** 0.0129 0.0555* -0.00652 -0.0564(0.0182) (0.0260) (0.0174) (0.0266) (0.0553) (0.0587)
Log average in women’s income 0.0388* 0.00347 0.0454* 0.00947 -0.0279 -0.0152(0.0200) (0.0319) (0.0222) (0.0375) (0.0795) (0.0726)
Share of Women Income 0.152 0.154 -0.134(0.119) (0.136) (0.0939)
Men average schooling 0.0268 0.0325 0.0400(0.0255) (0.0250) (0.0247)
Women average schooling 0.0384** 0.0453** 0.0392**(0.0164) (0.0149) (0.0152)
Number of men 2.38e-05 9.96e-06 9.05e-05(7.87e-05) (7.01e-05) (8.01e-05)
Number of women 4.83e-06 1.75e-05 -5.27e-05(7.80e-05) (6.97e-05) (7.64e-05)
Sex ratio -0.153 -0.158 -0.245**(0.102) (0.0925) (0.0929)
Constant 0.543*** 0.195 0.501*** 0.0999 0.245*** 0.0805(0.0405) (0.152) (0.0370) (0.144) (0.0680) (0.173)
Observations 1,440 1,440 1,440 1,440 1,440 1,440R-squared 0.984 0.985 0.984 0.985 0.983 0.984
Data are from the 1996 to 2011 SLID cross-sections. The unit of observation is a marriage market by year. All
specifications include marriage market fixed effects, year fixed effects, and the year interacted with the age group,the education group, the province and deciles of men’s and women’s income. Regressions are weighted by the
number of women in the marriage market. Standard errors clustered at the province level are in parenthesis.***significant at 1%, **at 5%, *at 10%.
Table 4. Potential relative income and marriage rate in Canada. Ef-fects by province.
Atlantic Quebec Ontario Prairies British-ColumbiaActual earnings -0.361*** -0.189*** -0.244*** -0.287*** -0.097***
[36.60] [15.39] (0.051) [23.78] [16.11]Predicted earnings -0.322*** -0.217*** -0.064 -0.281** -0.061
[28.88] [19.66] (0.070) [6.18] [3.21]Bartik instrument -0.172** -0.348** 0.122* -0.223** -0.031
[9.62] [6.27] (0.061) [8.17] [0.09]F-stats are reported in brackets. Standard errors in parenthesis. ***significant at 1% level, **at 5%, *at 10%.
16
5. Women labor supply
5.1. Women labor force participation. In this section, we want to estimate the effect of gender
identity on participation of married women in the labor market. We use the SLID cross-sections
from 1996 to 2011 and we restrict the data to couples where the husband earns a positive income.
As in the previous section, we also construct a predictive measure of income according to demo-
graphic characteristics. For each couple i, we estimate the distribution of the potential income of
a married woman according to her demographic group. For this purpose, we define the percentiles
of income that we denote wpi , p = {5th, 10th, ...95th} among the population of working women in
the demographic group of married women (age group, level of education, province of residence). We
define the probability that a married woman will have a higher income than her husband, on the
basis of the demographic group to which she belongs computing the following variable:
PrWifeEarnsMorei =1
19
∑p
1(wpi >husbIncomei)
where husbIncomei is the income of her current spouse. Whatever her current professional status,
this calculation captures the probability of her earning more income than her spouse if her income
was randomly drawn from the labor force of women in her demographic group.
We establish wifeLFPi, wife’s labor market participation for a couple i, as a binary variable
that takes the value 1 when the married woman is in the labor force all year. We present summary
statistics of our computed variable in table 15 in appendix. The average probability of wives earning
more than their husband is equal to 27% and the participation rate of women is of 67%. We estimate
the effect of gender identity on women labor force participation using a linear probability model.
Formally, we simply run the following regression:
wifeLFPi = β0 + β1PrWifeEarnsMorei + β2logHusbIncomei + β3Xi + βdwdi + εi
where logHusbIncomei is the logarithm of the husband’s income, wdi controls for the potential
income of the wife at each decile. And Xi is a matrix of non-income controls: age group, education
level, province of residence. Results are presented in column (1) of table 5. One point of percentage
increase in the probability that a wife earns more than her husband decreases by 0.02 percentage
17
point (pvalue < 0.05) the probability that she participates to the labor market. To control for the
husband income, we add a cubic polynomial of the logarithm of the income of the husband (column
(2)), and in our last specification (columns (3)), we add the presence of children, the interaction
variable logMedianWifeIncome× logHusbIncome and the wife demographic groups fixed effects.
We find that one point of percentage increase in the probability that a wife earns more than her
husband decreases by 0.09 percentage point (pvalue < 0.01) the probability that she participates to
the labor market which is close to the -0.14 point found in BKP.
In columns (4) to (6), we investigate the effect of gender identity on labor force participa-
tion of women across provinces and we add in the regressions the interaction variable area ×
PrWifeEarnsMore. We present the effect in each province in table 6. The first row gives the
estimate of the effect that is the sum of the coefficients of the variables PrWifeEarnsMore and
area×PrWifeEarnsMore when all control variables are included. The second row gives the F-stat
of a significativity test of that sum. We find strong and significant similar negative effects across
provinces.
18
Table 5. Potential relative income and wife’s labor force participation
Dependent variable : wife in the labor force all year(1) (2) (3) (4) (5) (6)
PrWifeEarnsMore -0.0230** -0.0562*** -0.0923*** -0.0222 -0.0665*** -0.0868***(0.0112) (0.0137) (0.0140) (0.0156) (0.0173) (0.0172)
Atlantic× PrWifeEarnsMore 0.00534 0.0298* -0.00575(0.0157) (0.0157) (0.0150)
Quebec× PrWifeEarnsMore -0.0246 -0.00540 -0.0216(0.0194) (0.0192) (0.0176)
Prairie× PrWifeEarnsMore 0.0202 0.0260* 0.0116(0.0149) (0.0149) (0.0136)
British Columbia× PrWifeEarnsMore -0.0197 -0.0150 -0.0265(0.0168) (0.0173) (0.0167)
logHusbIncome -0.00814** -0.495*** -0.553*** -0.00793** -0.500*** -0.551***(0.00380) (0.0663) (0.0669) (0.00382) (0.0666) (0.0670)
SquareLogHusbIncome 0.0690*** 0.0586*** 0.0696*** 0.0582***(0.00778) (0.00761) (0.00782) (0.00763)
CubicLogHusbIncome -0.00300*** -0.00265*** -0.00302*** -0.00263***(0.000290) (0.000282) (0.000291) (0.000283)
logMedianWifeIncome × logHusbIncome 0.0147*** 0.0148***(0.00141) (0.00145)
Has children -0.0465*** -0.0466***(0.00591) (0.00591)
Constant 0.223*** 1.201*** 1.405*** 0.219*** 1.205*** 1.401***(0.0423) (0.179) (0.176) (0.0422) (0.179) (0.176)
Observations 181,631 181,631 181,631 181,631 181,631 181,631R-squared 0.075 0.078 0.090 0.075 0.078 0.090Wife demog. group No No Yes No No Yes
Data are from the 1996 to 2011 SLID cross-sections. The sample consists of couples where both the wife and the
husband are between 18 and 65 years old and the husband is working. All regressions include controls for log
husband’s income, deciles of the wife’s potential income, wife’s and husband’s education (five categories), wife’s andhusband’s six-year age group, year, and province fixed effects. Standard errors are reported in parenthesis.
***significant at 1% level, **at 5%, *at 10%.
Table 6. Potential relative income and wife’s labor force participa-tion. Effects by province
Atlantic Quebec Ontario Prairies British-ColumbiaCoefficient -0.0926*** -0.1084*** -0.0868*** -0.075*** -0.133***
[31.38] [32.81] (0.017) [22.41] [34.64]F-stats are reported in brackets. Standard errors are reported in parenthesis.
***significant at 1% level, **at 5%, *at 10%.
5.2. Difference between income earned by the wife and her potential income. The cost
of a definitive departure from the labor market is very high. To comply with gender identity, a
woman may instead only reduce her participation at the intensive margin which would be less costly
financially. In this section, we intend to estimate the difference between the income a married
woman achieves and her potential income, that is the income she could realize according to her19
demographic characteristics if she did not modify her behavior on the labor market in order to
preserve the traditional structure of her household. We compute for each women i the variable
incomeGapi =wifeIncomei − wifepotentiali
wifepotentiali
where wifeIncomei is the actual income earned by the wife and wifepotentiali is the prediction
of the potential income of the woman according to her demographic group. In this section, we
restrict our estimation to married women who work and earn a positive income. We use the same
linear probability model and the same explanatory variables as in section 5.1 and run the following
regression:
incomeGapi = β0 + β1PrWifeEarnsMorei + β2lnHusbIncomei + β3Xi + βdwdi + εi
Results are presented in table 7. One point of percentage increase in the probability that a wife
earns more than her husband decreases by 0.19 percentage point the income gap between her current
income and her potential income. To control for the husband income, we add a cubic polynomial of
the logarithm of the income of the husband (column (2)), and in our last specification (columns (3))
we add the presence of children, the interaction variable logMedianWifeIncome× logHusbIncome
and the wife demographic groups fixed effects. We find smaller estimates when we add these controls.
Our estimate becomes non-significant in our last specification.
In columns (4) to (6), we investigate the effect of gender identity on the income gap across
provinces and we add in the regressions the interaction variable area × PrWifeEarnsMore. We
present the effect for each province in table 10. The first row gives the estimate of the effect that
is the sum of the coefficients of the variables PrWifeEarnsMore and area× PrWifeEarnsMore
when all control variables are included. The second row gives the F-stat of a significativity test of
that sum. We find a strong and significant negative effect Quebec and in the Prairies only. In those
regions, we find that one point of percentage increase in the probability that a wife earns more than
her husband decreases by 0.07 her income gap which is close to the -0.11 point found in BKP.
Our results then highlight the existence of a negative effect of gender identity on women labor
force participation.
20
Table 7. Potential relative income and income gap
Dependent variable : Income gap(1) (2) (3) (4) (5) (6)
PrWifeEarnsMore -0.185*** -0.0502* -0.0333 -0.178*** -0.0274 -0.0116(0.0220) (0.0263) (0.0272) (0.0274) (0.0334) (0.0340)
Atlantic× PrWifeEarnsMore 0.0312 0.000623 0.0307(0.0266) (0.0277) (0.0271)
Quebec× PrWifeEarnsMore -0.0191 -0.0452 -0.0560**(0.0283) (0.0317) (0.0280)
Prairie× PrWifeEarnsMore -0.0417 -0.0508* -0.0600**(0.0260) (0.0269) (0.0268)
British Columbia× PrWifeEarnsMore -0.00844 -0.0129 -0.0174(0.0449) (0.0458) (0.0455)
logHusbIncome 0.0297*** 0.0717 0.390*** 0.0291*** 0.0661 0.379***(0.00814) (0.136) (0.138) (0.00812) (0.136) (0.138)
SquareLogHusbIncome -0.0288* -0.0342** -0.0282* -0.0326**(0.0161) (0.0161) (0.0161) (0.0160)
CubicLogHusbIncome 0.00194*** 0.00222*** 0.00192*** 0.00216***(0.000625) (0.000622) (0.000624) (0.000620)
logMedianIncWife × logHusbIncome -0.0281*** -0.0284***(0.00272) (0.00269)
Has children -0.143*** -0.142***(0.0135) (0.0135)
Constant 0.310*** 0.823** -0.342 0.303*** 0.834** -0.311(0.0879) (0.383) (0.380) (0.0875) (0.382) (0.379)
Observations 120,072 120,072 120,072 120,072 120,072 120,072R-squared 0.032 0.035 0.050 0.032 0.036 0.051Wife demog. group No No Yes No No Yes
Data are from the 1996 to 2011 SLID cross-sections. The sample consists of couples where both the wife and the
husband are between 18 and 65 years old and the husband is working. All regressions include controls for loghusband’s income, deciles of the wife’s potential income, wife’s and husband’s education (five categories), wife’s and
husband’s five-year age group, year, and province fixed effects. Standard errors are reported in parenthesis.
***significant at 1% level, **at 5%, *at 10%.
Table 8. Potential relative income and income gap. Effects by province
Atlantic Quebec Ontario Prairies British-ColumbiaCoefficient 0.0191 -0.0677** -0.0116 -0.0716** -0.0290
[0.40] [4.45] (0.034) [5.21] [0.56]F-stats are reported in brackets. Standard errors in parenthesis.
***significant at 1% level, **at 5%, *at 10%.
6. Relative income and housework
In this section, we analyze the effect of gender identity on women’s time spent on housework.
For this purpose, we use the confidential micro-data files of the General Social Survey of 2010 and
2015 on time use. These data provide information on time spent on unpaid work, on paid work
and provides for the exact amount of income earned by the respondent and the exact amount of the21
total income earned by the household. However, the income of the spouse is not directly observed,
so we infer it from the difference between the total household income and the respondent’s personal
income. To reduce the risk of errors in calculating spouses’ income, we keep only the households in
which both spouses work, and we remove households in which more than two members contribute
to the total household income. We measure domestic work, as the sum of the amount of time spent
in household chores or maintenance.12 It is measured in hours per week for each respondent i by
our variable HouseworkT imei. We restrict the data to individuals aged between 18 and 65, who
are married or living in a common-law relationship and who have a non-zero income. Finally, we
build the variable wifeEarnsMorei which is equal to 1 if the woman has a higher income than
her spouse. Summary statistics are presented in table 15. In 2010, wives spend an average of 18.8
hours a week on chores, compared with 12.7 hours for husbands and the wife earns more than the
husband in 25 percent of the couples. In 2015, wives spend an average of 19.7 hours a week on
chores, compared with 13.9 hours for husbands and the wife earns more than the husband in 30
percent of the couples. From 2010 to 2015, the ratio of husbands’ housework over wives’ housework
has then slightly increased from 68% to 71%. In the GSS, we observe information on time use for
only one respondent per household. Thus, to analyze how relative income impacts the division of
home production, we focus, as in BKP, on the interaction between the impact of gender and the
impact of relative income on time use. Specifically, we consider the baseline OLS model:
HouseworkT imei =β0 + β1wifeEarnsMorei + β2femalei × wifeEarnsMorei + β3femalei
+ β4lnWifeIncomei + β5femalei × lnWifeIncomei
+ β6lnHusbIncomei + β7femalei × lnHusbIncomei
+ β8lnTotalIncomei + β9femalei × lnTotalIncomei
+ β10Xi + β11femalei ×Xi + εi
where femalei is a dummy indicating if the individual is a woman, and Xi includes levels of
education of both spouses (three categories), a quadratic polynomial of the age of both spouses, a
12It includes meal preparation, house cleaning, dish washing, tidying, laundry, ironing, repair, orga-nizing, doing the groceries, gardening, ...
22
cubic polynomial of the logarithm of income of the husband, the wife and the total income. We
also add the income share of the woman, dummies indicating whether the household has zero child,
one child, two children, three children or more than three children, dummies indicating whether the
household has zero child under four year old, one child, two children, three children or more than
three children under four year old, province fixed effects and day of the week fixed effects. We only
keep weekdays observations.
Our coefficient of interest is given by β2. A positive coefficient indicates that, ceteris paribus, in
couples where the wife earns more than the husband, she also spends more time on housework. Table
9 presents our results. Column (1) presents our estimates for observations the year 2010 whereas
column (3) presents our estimates for 2015. We find a significant and positive effect of 4.10 hours
per week in 2010 (pvalue < 0.05). This effect is higher than the effect of 2.7 hours found in BKP.
The effect is not significant anymore in 2015.
In columns (2) and (4), we investigate the effect of gender identity on housework across provinces
and we add in the regressions the two interaction variables female× area× wifeEarnsMore and
area × wifeEarnsMore. Our coefficient of interest is then given by the sum of the coefficients
of the variables female × area × wifeEarnsMore and female × wifeEarnsMore. We present
this estimate and its F-stat for each province and each year in table 10. We only find a significant
and positive effect in the Atlantic provinces (4.91, pvalue < 0.1) and in British-Columbia (6.63,
pvalue < 0.1) for the year 2010. We do not find any significant effect in 2015.
Our results show that in 2010, when women exceeded the spouse’s income equality threshold,
they also increased their time spent in housework. This behavior is in contradiction with standard
theory. Compliance to gender identity is the most likely explanation. This behavior does not exist
anymore in 2015, which may be related to a decline in traditional gender norms.
23
Table 9. Relative income and wife’s housework
Dependent variable : Housework (Hours per week)2010 2015
(1) (2) (3) (4)
wifeEarnsMore -2.42* -0.51 1.40 2.55(1.48) (2.05) (1.30) (1.89)
female× PrWifeEarnsMore 4.10** 2.21 0.28 0.19(1.93) (2.73) (1.64) (2.47)
Atlantic× wifeEarnsMore -1.70 -3.40(2.70) (2.37)
Atlantic× female× wifeEarnsMore 2.71 2.47(3.55) (3.16)
Quebec× wifeEarnsMore -4.25 0.59(3.27) (2.42)
Quebec× female× wifeEarnsMore 1.79 -3.26(4.31) (3.26)
Prairie× wifeEarnsMore -2.44 -2.60(2.71) (2.39)
Prairie× female× wifeEarnsMore 2.87 2.19(3.70) (3.28)
British Columbia× wifeEarnsMore -3.83 0.10(3.26) (3.10)
British Columbia× female× wifeEarnsMore 4.42 -1.87(4.37) (4.14)
Observations 2585 4075R-squared 0.108 0.110 0.114 0.115
Data are from confidential micro-data files of the GSS 2010 and 2015 on time use. The sample consists of coupleswhere both the wife and the husband are between 18 and 65 years old and earn a positive income. All regressions
include controls for a cubic polynomial of the logarithm of the incomes of the husband, the wife and the totalincome, for wife’s and husband’s education (three categories), quadratic polynomial of wife’s and husband’s age,
controls for children, day of the week and province fixed effects. We only keep weekdays. Standard errors are
reported in parenthesis. ***significant at 1% level, **at 5%, *at 10%.
Table 10. Relative income and wife’s housework. Effects by province
Canada Atlantic Quebec Ontario Prairies British-Columbia2010 4.10** 4.91* 3.99 2.21 5.07 6.63*
(1.93) [2.63] [1.11] (2.73) [2.37] [2.63]2015 0.28 2.65 -3.07 0.19 2.38 -1.68
(1.64) [1.02] [1.30] (2.47) [0.72] [0.19]F-stats are reported in brackets. Standard errors are reported in parenthesis.
***significant at 1% level, **at 5%, *at 10%.
7. Conclusion
In this paper, we show that the distribution of the share of income held by the woman among the
Canadian households drops to the right of 0.5. As in BKP, we believe that this drop may be linked
to gender identity, that is the traditional belief that the man should have a higher income than his
wife in order to preserve the traditional structure of the household. We find that couples form less24
when the probability that a woman earns more than her husband increases and that the wives with
greater probability of earning more than their husbands are less likely to participate in the labor
force. If they participate, they have a higher probability of earning less than their potential income.
We find a stronger effect of gender norms in Quebec and in the Prairies but a less important effect
in British-Columbia and in Ontario.
Other explanations of the missing couple pattern have been raised in the recent literature and
complement the gender norm hypothesis. One is related to the existence of gender difference in
reporting income in surveys. Murray-Close and Heggeness (2018) analyse this phenomenon with US
data and Roth and Slotwinski (2018) with Swiss data. These papers find that the gap between a
husband’s survey and administrative earnings is higher if his wife earns more than he does, and the
gap between a wife’s survey and administrative earnings is lower if she earns more than her husband
does. These findings suggest that gendered social norms can influence survey reports of seemingly
objective outcomes and that their impact may be heterogeneous not just between genders but also
within gender. Another explanation is related to what women think is a fair distribution of relative
working hours within the household. Fleche, Lepinteur, and Powdthavee (2018) present evidence
that in addition to the fact that women may have preferences for not wanting to out-earn their
husband, there is also an aversion to a situation where women work significantly longer hours than
their husbands. They show that women’s propensity to opt-out of the labor force is significantly
higher in couples where the wife’s working hours exceed the husband’s. They also show that life
satisfaction is significantly lower among women whose work hours exceed their partners, holding the
share of wife’s income constant. Men, by contrast, are not affected by working longer or fewer hours
than their wives. They argue that these patterns are best explained by perceived fairness of the
division of labor within the household.
All these results highlight the impact of gender-unequal norms on men and women behavior. This
is important as traditional gender norms such as the male breadwinner norms are prevalent around
the world and are still present in developed countries where they have stabilized (Fortin, 2015). As
norms are cultural and can be done and undone by institutions (Lippmann, Georgieff, and Senik,
2019), policy-makers have a role to play in the construction of these norms and their consequence
on individual behavior.
25
References
Akerlof, G. A., and R. E. Kranton (2000): “Economics and identity,” The Quarterly Journal
of Economics, 115(3), 715–753.
Baker, M., and M. Drolet (2010): “A new view of the male/female pay gap,” Canadian Public
Policy, 36(4), 429–464.
Becker, G. S. (1973): “A theory of marriage: Part I,” Journal of Political Economy, 81(4), 813–
846.
Bertrand, M., E. Kamenica, and J. Pan (2015): “Gender identity and relative income within
households,” The Quarterly Journal of Economics, 130(2), 571–614.
Binder, A. J., and D. Lam (2018): “Is there a male breadwinner norm? The hazards of inferring
preferences from marriage market outcomes,” Discussion paper, National Bureau of Economic
Research.
Blau, F. D., and L. M. Kahn (2017): “The gender wage gap: Extent, trends, and explanations,”
Journal of Economic Literature, 55(3), 789–865.
Bonikowska, A., M. Drolet, and N. M. Fortin (2019): “Earnings Inequality and the Gender
Pay Gap in Canada: The Role of Women’s Under-representation Among Top Earners,” Economic
Insights, Statistics Canada.
Bursztyn, L., T. Fujiwara, and A. Pallais (2017): “’Acting Wife’: Marriage Market Incentives
and Labor Market Investments,” American Economic Review, 107(11), 3288–3319.
Codazzi, K., V. Pero, and A. Albuquerque Sant’Anna (2018): “Social norms and female
labor participation in Brazil,” Review of Development Economics, 22(4), 1513–1535.
Fleche, S., A. Lepinteur, and N. Powdthavee (2018): “Gender Norms and Relative Work-
ing Hours: Why Do Women Suffer More Than Men from Working Longer Hours Than Their
Partners?,” in AEA Papers and Proceedings, vol. 108, pp. 163–68.
Fortin, N. M. (2005): “Gender role attitudes and the labour-market outcomes of women across
OECD countries,” Oxford review of Economic Policy, 21(3), 416–438.
(2015): “Gender role attitudes and women’s labor market participation: Opting-out,
AIDS, and the persistent appeal of housewifery,” Annals of Economics and Statistics/Annales
d’Economie et de Statistique, (117/118), 379–401.
26
(2019): “Increasing earnings inequality and the gender pay gap in Canada: Prospects for
convergence,” Canadian Journal of Economics/Revue canadienne d’economique.
Greenstein, T. N. (2000): “Economic dependence, gender, and the division of labor in the home:
A replication and extension,” Journal of Marriage and Family, 62(2), 322–335.
Heckert, D. A., T. C. Nowak, and K. A. Snyder (1998): “The impact of husbands’ and
wives’ relative earnings on marital disruption,” Journal of Marriage and the Family, pp. 690–703.
Hederos Eriksson, K., and A. Stenberg (2015): “Gender Identity and Relative Income within
Households: Evidence from Sweden,” .
Kalmijn, M., A. Loeve, and D. Manting (2007): “Income dynamics in couples and the disso-
lution of marriage and cohabitation,” Demography, 44(1), 159–179.
Lachance-Grzela, M., and G. Bouchard (2010): “Why do women do the lions share of
housework? A decade of research,” Sex roles, 63(11-12), 767–780.
Lippmann, Q., A. Georgieff, and C. Senik (2019): “Undoing Gender with Institutions. Lessons
from the German Division and Reunification,” .
McCrary, J. (2008): “Manipulation of the running variable in the regression discontinuity design:
A density test,” Journal of Econometrics, 142(2), 698–714.
Moyser, M. (2017): Women in Canada: a Gender-based Statistical Report. Women and Paid
Work. Statistics Canada.
Murray-Close, M., and M. L. Heggeness (2018): “Manning up and womaning down: How
husbands and wives report their earnings when she earns more,” US Census Bureau Social, Eco-
nomic, and Housing Statistics Division Working Paper, (2018-20).
Roth, A., and M. Slotwinski (2018): “Gender Norms and Income Misreporting within House-
holds,” .
Schirle, T. (2015): “The Gender Wage Gap in the Canadian Provinces, 1997-2014,” Canadian
Public Policy / Analyse de Politiques, 41(4), 309–319.
Wieber, A., and E. Holst (2015): “Gender identity and womens’ supply of labor and non-market
work: Panel data evidence for Germany,” .
27
8. Appendix
Table 11. Gender norms
(1) (2) (3) (4) (5)Men First Warm If Not Working Housewife fulfilling Both should not contribute Index
Female -0.00884 -0.114* 0.0481 0.0920 -0.0393(0.0383) (0.0589) (0.0496) (0.0621) (0.130)
25-39 year old -0.0168 0.0839** 0.0370 0.140*** 0.223**(0.0303) (0.0413) (0.0399) (0.0437) (0.0926)
40-54 year old -0.00316 0.137*** 0.0864** 0.208*** 0.379***(0.0317) (0.0432) (0.0417) (0.0458) (0.0971)
55-65 year old 0.0513 0.242*** 0.190*** 0.284*** 0.750***(0.0364) (0.0504) (0.0474) (0.0533) (0.113)
Female × 25-39 year old 0.0561* 0.00246 0.0135 0.0452 0.136(0.0309) (0.0481) (0.0400) (0.0509) (0.106)
Female × 40-54 year old 0.0500 0.0149 0.0375 -0.00204 0.168(0.0310) (0.0493) (0.0400) (0.0522) (0.109)
Female × 55-65 year old 0.0158 -0.0574 0.0105 -0.0905 -0.126(0.0351) (0.0577) (0.0453) (0.0611) (0.128)
Leave school [17-21] -0.0685*** -0.0241 -0.00762 0.0834** -0.0547(0.0219) (0.0342) (0.0286) (0.0360) (0.0760)
Leave school after 21 -0.128*** -0.0375 -0.0140 0.111*** -0.111(0.0218) (0.0346) (0.0286) (0.0364) (0.0772)
Female × Leave school [17-21] -0.0279 0.0189 -0.0304 -0.121** -0.134(0.0291) (0.0454) (0.0377) (0.0480) (0.100)
Female × Leave school after 21 -0.0484* -0.0205 -0.0503 -0.128*** -0.227**(0.0292) (0.0462) (0.0378) (0.0489) (0.102)
2000 -0.0314 -0.0135 0.213*** 0.0163 0.179*(0.0341) (0.0441) (0.0443) (0.0466) (0.0978)
2006 -0.0403 0.142***(0.0344) (0.0449)
Female × 2000 0.00867 0.0781** -0.0854*** 0.0207 0.0315(0.0241) (0.0312) (0.0312) (0.0332) (0.0693)
Female × 2006 -0.0177 -0.0350(0.0241) (0.0312)
25-39 year old × 2000 -0.0145 -0.0710 -0.00916 -0.105** -0.190*(0.0371) (0.0479) (0.0480) (0.0507) (0.106)
25-39 year old × 2006 0.0126 0.0189(0.0379) (0.0492)
40-54 year old × 2000 0.00191 -0.136*** -0.113** -0.141*** -0.396***(0.0381) (0.0491) (0.0493) (0.0521) (0.109)
40-54 year old × 2006 0.0166 -0.0706(0.0384) (0.0498)
55-65 year old × 2000 0.0138 -0.157*** -0.124** -0.187*** -0.443***(0.0434) (0.0561) (0.0558) (0.0595) (0.124)
55-65 year old × 2006 -0.0158 -0.123**(0.0427) (0.0552)
Atlantic -0.0256* -0.0189 0.0896*** 0.00150 0.0118(0.0144) (0.0241) (0.0186) (0.0256) (0.0536)
Quebec 0.0405*** -0.0464** 0.0989*** -0.0880*** 0.0148(0.0130) (0.0203) (0.0167) (0.0215) (0.0446)
Prairies -0.0127 0.00541 0.0708*** 0.0758*** 0.130**(0.0150) (0.0233) (0.0194) (0.0251) (0.0523)
British-Columbia -0.0217 -0.0192 0.0131 0.0413 0.0109(0.0171) (0.0273) (0.0219) (0.0290) (0.0603)
Constant 0.233*** 0.260*** 0.575*** 0.0889* 1.231***(0.0335) (0.0481) (0.0440) (0.0507) (0.107)
Observations 4,706 3,018 4,507 2,967 2,798R-squared 0.049 0.036 0.040 0.042 0.050
Standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
28
Table 12. Percentage of wives marrying husbands of particular ageand education
Husband Age groupWife Age group 18-35 36-45 46-6518-33 0.8092 0.1744 0.016434-43 0.1263 0.6983 0.175444-65 0.0090 0.1106 0.8804
Husband Education groupWife Education group Secondary Non-university University
and less post secondarySecondary and less 0.5246 0.4052 0.0701Non-university, post secondary 0.3070 0.5418 0.1512University 0.1240 0.3750 0.5009
Data are from the 1996 to 2011 SLID cross-sections. Each cell reports the percentage of wives and husbands in eachcategory.
Table 13. Summary statistics - Marriage market sample
Mean Standard DeviationPrWomanEarnsMore - Actual 0.355 0.479PrWomanEarnsMore - Predicted 0.316 0.465PrWomanEarnsMore - Bartik 0.335 0.472Male marriage rate 0.668 0.202
Data are from the 1996 to 2011 SLID cross-sections. Each observation is a marriage market defined by the
interaction of province, year, age-group and education-group. N = 1440.
29
Table 14. Potential relative income and marriage rate in Canada, byprovince
Dependent variable : marriage rateActual earnings Predicted earnings Bartik instrument(1) (2) (3) (4) (5) (6)
PrWomanEarnsMore -0.334*** -0.244*** -0.174** -0.0642 0.199* 0.122*(0.0498) (0.0511) (0.0630) (0.0700) (0.0898) (0.0610)
Atlantic× PrWomanEarnsMore -0.0598 -0.117 -0.156 -0.258** -0.338*** -0.294***(0.108) (0.104) (0.0996) (0.105) (0.0993) (0.0621)
Quebec× PrWomanEarnsMore 0.106 0.0557 -0.0690 -0.153* -0.460*** -0.469***(0.0682) (0.0593) (0.0620) (0.0795) (0.105) (0.114)
Prairie× PrWomanEarnsMore 0.0144 -0.0430 -0.139 -0.216 -0.410*** -0.344***(0.0837) (0.0803) (0.145) (0.133) (0.0714) (0.0467)
British Columbia× PrWomanEarnsMore 0.215** 0.147* 0.117** 0.00343 -0.165* -0.153*(0.0724) (0.0676) (0.0518) (0.0616) (0.0736) (0.0797)
Log average in men’s income 0.0197 0.0581* 0.0137 0.0529 0.000876 -0.0528(0.0188) (0.0267) (0.0165) (0.0315) (0.0536) (0.0580)
Log average in women’s income 0.0400* 0.00737 0.0434* 0.0126 -0.0244 -0.00950(0.0201) (0.0340) (0.0214) (0.0412) (0.0793) (0.0719)
Share of Women Income 0.138 0.132 -0.142(0.124) (0.148) (0.0948)
Men average schooling 0.0267 0.0329 0.0362(0.0258) (0.0247) (0.0261)
Women average schooling 0.0404** 0.0481** 0.0401**(0.0177) (0.0148) (0.0148)
Number of men 2.47e-05 2.28e-05 8.74e-05(7.90e-05) (7.55e-05) (7.39e-05)
Number of women 2.00e-06 4.63e-06 -5.98e-05(7.84e-05) (7.56e-05) (7.41e-05)
Sex ratio -0.144 -0.157 -0.249**(0.102) (0.0914) (0.0900)
Constant 0.572*** 0.221 0.531*** 0.134 0.241*** 0.0993(0.0513) (0.152) (0.0335) (0.140) (0.0725) (0.183)
Observations 1,440 1,440 1,440 1,440 1,440 1,440R-squared 0.985 0.985 0.984 0.985 0.983 0.984
Data are from the 1996 to 2011 SLID cross-sections. The unit of observation is a marriage market by year. All
specifications include marriage market fixed effects, year fixed effects, and the year interacted with the age group,
the education group, the province and deciles of men’s and women’s income. Regressions are weighted by thenumber of women in the marriage market. Standard errors clustered at the province level are in parenthesis.
***significant at 1%, **at 5%, *at 10%.
Table 15. Summary statistics - Time uses
N Mean St. Dev.Data source : SLID. 1996 - 2011PrWifeEarnsMore 181631 0.270 0.305Wife Labor Force Participation 181631 0.672 0.470Income Gap 120072 0.101 0.691Data source : GSS. 2010WifeEarnsMore 3988 0.25 0.43Wife Housework time (hours per week) 3292 18.82 16.69Husband Housework time (hours per week) 2691 12.74 15.45Data source : GSS. 2015WifeEarnsMore 5948 0.30 0.46Wife Housework time (hours per week) 3218 19.66 17.86Husband Housework time (hours per week) 2746 13.93 17.30
Data are from the 1996 to 2011 SLID cross-sections and from the GSS cross-sections available in CDR.
30
Figure 4. Discontinuity at 0.5 in relative income by provinces
02
46
-.5 0 .5 1 1.5
(a) Atlantic. With same inc. couples
01
23
4-.5 0 .5 1 1.5
(b) Atlantic. Without same inc. couples
02
46
8
-.5 0 .5 1 1.5
(c) Quebec. With same inc. couples
01
23
4
-.5 0 .5 1 1.5
(d) Quebec. Without same inc. couples
02
46
810
-.5 0 .5 1 1.5
(e) Ontario. With same inc. couples
01
23
4
-.5 0 .5 1 1.5
(f) Ontario. Without same inc. couples
31
Figure 5. Discontinuity at 0.5 in relative income by provinces (continued)
02
46
8
-.5 0 .5 1 1.5
(a) Prairie. With same inc. couples
01
23
4-.5 0 .5 1 1.5
(b) Prairie. Without same inc. couples
01
23
45
-.5 0 .5 1 1.5
(c) British-Columbia. With sameinc. couples
01
23
-.5 0 .5 1 1.5
(d) British-Columbia. Withoutsame inc. couples
32