Full text of DOI:10.1080/00036840122961

Applied Economics
ISSN: 0003-6846 (Print) 1466-4283 (Online) Journal homepage: http://www.tandfonline.com/loi/raec20
An investigation into sexual orientation
discrimination as an explanation for wage
differences
Suzanne Heller Clain & Karen Leppel
To cite this article: Suzanne Heller Clain & Karen Leppel (2001) An investigation into sexual
orientation discrimination as an explanation for wage differences, Applied Economics, 33:1,
37-47
To link to this article: http://dx.doi.org/10.1080/00036840122961
Published online: 09 Nov 2010.
Submit your article to this journal
Article views: 301
View related articles
Citing articles: 43 View citing articles
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=raec20
Download by: [Emory University]
Date: 15 March 2016, At: 21:52

Applied Economics, 2001, 33, 37±47
An investigation into sexual orientation
discrimination as an explanation for wage
diŒerences
SUZANNE HELLER CLAIN and KAREN LEPPEL{
Department of Economics, Villanova University, 800 Lancaster Avenue, Villanova, PA
1
9085, USA; e-mail: suzanne.clain@villanova, edu;
Department of Economics, Widener University, 1 University Place, Chester,
{
PA 19013, USA; e-mail: karen.leppel@widener.edu
This study explores the eŒects of sexual orientation on earnings. It is found that,
ceteris paribus, men living with male partners tend to earn less than other men, and
women living with female partners tend to earn more than other women. These
earnings diŒerentials tend to vary by region. They also vary by education and
occupation for men, and with the presence of minor children for women. In addition,
the age-earnings pro®les of women living with female partners tend to be higher,
¯
atter, and less concave than those of other women.
I. INTRODUCTION
®ed on the grounds of e ciency and equity. While it may
be argued that current legislation has not been 100% eŒec-
tive in eliminating discrimination against women and
minorities, there is evidence that it has had a measurable
impact (Beller, 1982; Leonard, 1984).
Labour market discrimination exists when two equally
quali®ed individuals are treated diŒerently in the labour
market, on the basis of some personal characteristic unre-
lated to productivity. Discrimination can take the form of
diŒerences in compensation, or diŒerences in hiring and
promotion practices. When discrimination occurs, individ-
uals may not be employed in the job where they can con-
tribute the most. Consequently, discrimination is
ine cient, from a social perspective. Furthermore, because
discrimination results in two equally quali®ed individuals
being treated diŒerently, discrimination is also counter to
the interests of equity.
In order to detect labour market discrimination based on
sexual orientation, one must inspect the data for diŒerences
in labour market outcomes by sexual orientation. If such
diŒerences exist and cannot be fully explained by diŒer-
ences in productivity, discrimination based on sexual orien-
tation may be present. The unexplained portion of the
diŒerential is commonly used as an estimate of the impact
of labour market discrimination.
While there have been numerous economic studies of
race and gender discrimination (e.g. Gunderson, 1989;
Smith and Welch, 1989), the issue of discrimination on
the basis of sexual orientation has been largely neglected
by the economic literature. The exception is Badgett’s
(1995) study. She found that gay and bisexual males earn
from 11±27% less than heterosexual males controlling for
experience, education, occupation, marital status, size of
SMSA, and region of residence. She also found that lesbian
and bisexual women earn less than heterosexual women,
but the ®nding was not consistently statistically signi®cant
In the past few decades, legislators have moved towards
a public policy that the labour market treatment of indi-
viduals should be based on their productivity, rather than
on their personal characteristics. Laws prohibit acts of
labour market discrimination based on race, colour, reli-
gion, gender, and national origin. They also prohibit acts
of labour market discrimination based on age and physical
or mental disability. However, labour market protection at
the federal level in the USA does not extend to gays and
lesbians. If discrimination against gays and lesbians exists,
widening the scope of federal legislation to ®ght it is justi-
1
1
across speci®cations.
Badgett’s sample included only 34 (4.9% ) lesbian or bisexual women, and only 47 (5.2%) gay or bisexual men, so insigni®cant results
are not surprising.
Applied Economics ISSN 0003±6846 print/ISSN 1466±4283 online # 2001 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
37

3
8
S. H. Clain and K. Leppel
In Badgett’s earnings equations for women, the eŒect of
adverse job consequences, may avoid disclosing their sex-
ual orientation. The practice of hiding one’s sexual prefer-
ence is likely to reduce the measurable impact of
discriminatory behaviour.
sexual orientation was re¯ ected in two ways. It was
re¯ ected by a dummy variable and by the interaction of
the dummy variable with the variable measuring experi-
ence, to capture diŒerences in the eŒectiveness of the
proxy for experience for lesbian and heterosexual women.
No interactions of the dummy variable with any other
independent variables were included. In Badgett’s earnings
equations for men, the eŒect of sexual orientation was
re¯ ected simply by a dummy variable. The implication is
that the eŒect of sexual orientation is in the form of a
parallel shift of earnings equations. The eŒects of sexual
orientation on earnings, however, may be more complex.
To explore other possibilities, the present study examines
many interaction terms in the earnings equations of both
the men and the women.
Hiding one’s sexual preference, however, is likely to
cause anxiety and stress. Concealing one’s sexual identity
may also result in reduced opportunities to meet a compa-
tible individual with whom a strong relationship can be
developed. Many gays and lesbians may, therefore, reveal
their sexual inclinations despite the possible negative eŒects
on career advancement and income. Disclosure of sexual
orientation may also occur indirectly or involuntarily.
Inferences may be drawn based on conversations or other
information. In some cases, a voluntary disclosure to one
co-worker may become an involuntary disclosure to an
employer or other co-workers.
In this study, the results for males corroborate those of
Badgett. It is found that men living with male partners earn
less than men not living with partners, ceteris paribus. Men
living with male partners earn less than men living with
female partners, if they are college educated, working in
the Midwest or working in blue collar occupations. For
the most part, the estimated eŒects on earnings are within
the range calculated by Badgett.
If employers, co-workers and/or consumers have tastes
for discrimination and can distinguish gay and lesbian
employees from heterosexual employees, the result may
be lower earnings for gays and lesbians. There is a beha-
vioural response, however, that could produce a positive
impact of homosexuality on earnings level. Gays and les-
bians may respond to the threat of employment discrimi-
nation by working harder. Many believe that if they are
su ciently productive, they could overcome the stigma of
sexual orientation. Under these circumstances, the dummy
variable indicating their sexual orientation would be cap-
Results that are statistically signi®cant and generally
opposite in sign to Badgett’s are found for women, how-
ever. Women living with female partners tend to earn more
than women not living with partners, ceteris paribus. This is
especially true if they are living in the Midwest or living
with their own minor children, in which case they also earn
more than comparable women with male partners.
Section II contains a discussion of the theoretical frame-
work of discrimination on the basis of sexual orientation.
A description of the data follows in Section III. Estimation
results are presented and discussed in Section IV. The
primary conclusions of this research are reviewed in
Section V.
3
turing a productivity advantage.
III. DATA AND DESCRIPTIVE STATISTICS
Reliable and representative data on sexual orientation are
extremely rare. Badgett (1995) used three years of the
General Social Survey, which includes answers to questions
about the gender of one’s sexual partners. The present
study uses data from the 1/1000 Public Use Microdata
Sample (PUMS) of the 1990 Census of Population and
Housing. In this data set, an unrelated individual who is
living in the household may be classi®ed as either (i) a
roomer, boarder or foster child, (ii) a housemate or room-
mate, (iii) an unmarried partner, or (iv) other non-relative.
Persons of the same sex who are classi®ed as unmarried
partners are, for the purposes of this study, presumed to be
II. THEORETICAL FRAMEWORK
If the attitudes of employers, co-workers and/or consumers
are anti-homosexual, these individuals could exhibit a taste
2
for discrimination against gays and lesbians. For this taste
4
to be manifested in discriminatory behaviour, however,
these individuals must be able to distinguish gays and les-
bians from heterosexuals. Gays and lesbians, fearing
involved in a gay or lesbian relationship.
Persons living in group quarters were eliminated from
the study. Moreover, this analysis was restricted to civilian
2
For an economic analysis of tastes for discrimination, see Becker (1971). Becker noted that perfect competition would drive employers
with tastes for discrimination out of the market. However, discrimination stemming from the tastes of co-workers or consumers can
persist, with negative impacts on the wages of those who are the victims of discrimination.
3
For a discussion of this reaction, see Woods (1993: 209).
Note that the sexual orientation of individuals who are not living with a partner is not identi®able. It is also possible that lesbian and
4
gay couples may have, intentionally or unintentionally, misreported their relationships. The eŒects of these classi®cation errors on the
results are considered in a later footnote.

Sexual orientation discrimination in wage diŒerences
39
workers, 65 years old or younger, who were employed at
least 35 hours per week and at least 40 weeks per year,
because earnings levels, measured as annual wages or sal-
ary income, are lower for employees who work part-time or
speci®cation, their coe cients measure the impact of
being in these occupations, compared to working in farm-
9
ing, forestry or ®shing.
Other factors that in¯ uence earnings are unrelated to
productivity. The conditions of the local labour market
are examples. In this paper, the variables used to re¯ ect
these conditions were region and urban residence. Urban
residence was captured by a dummy variable (CITY) with
value one if the area in which the respondent lived was
partially or fully in a central city location, MSA or
PMSA, and zero otherwise. Regional eŒects were measured
using three dummy variables for geographic location.
NEAST was set equal to one if the individual lived in
New England or the Mid-Atlantic states, and zero other-
wise. MDWEST was equal to one if the individual lived in
the North Central states, and zero otherwise. WEST was
equal to one if the individual lived in the Mountain or
Paci®c states, and zero otherwise. When all three of these
variables appear in a speci®cation, their coe cients meas-
ure the impact of living in these states, compared to living
in the south.
5
,6
only part of the year.
There are numerous factors besides discrimination that
can in¯ uence earnings. Some of these factors pertain to
individual productivity. The productivity variables used
in the study were age, education, ¯ uency in the English
language, health status, and occupation. The variable
COLLEGE was set equal to one if the respondent has
completed college and zero otherwise. The variable HS
was set equal to one if the respondent has completed
high school, and zero otherwise. The coe cients on these
variables measure the eŒects of degree completion, com-
7
pared to workers who did not complete high school. The
variable AGE measured the individual’s years of age. To
allow for a non-linear relationship between wage and age,
the square of AGE (AGESQ) was also used in the analy-
8
sis. The variable FLUENT was set equal to one if the
individual speaks English well or very well, and zero other-
wise. To capture possible eŒects of disability and disease
Personal characteristics that trigger discriminatory feel-
ings also in¯ uence earnings. These characteristics may
include gender, race, and sexual orientation. In this
study, the sample was sorted by gender, and men and
women were analysed separately. The race variable was a
dummy variable (WHITE) with value one if the respondent
was white, and zero otherwise. Individuals with unmarried
partners of the same sex were assumed to be gay or lesbian
(SAMESEX= 1). Individuals not reporting such living
arrangements were assumed to either be heterosexual or
to have disclosed nothing about sexual orientation
(SAMESEX= 0).
(
(
including HIV and AIDS),
a
dummy variable
HLTHLIM) for ill health was considered. If the indivi-
dual’s activities were limited by poor health, the variable
was set equal to one; otherwise it was zero. Three dummy
variables for occupational categories (WHCOLL,
BLCOLL, and SERVICE) were included in the analysis.
The variable WHCOLL was set equal to one if the indivi-
dual’s occupation was among white collar occupations,
and zero otherwise. The variable BLCOLL was set equal
to one if the individual’s occupation was among blue collar
manufacturing occupations, and zero otherwise. The vari-
able SERVICE was set equal to one if the individual’s
occupation was among service occupations, and zero
otherwise. When all three of these variables appear in a
Finally, personal characteristics that re¯ ect household
composition may in¯ uence the individual’s relative tastes
for non-pecuniary job characteristics. The presence of an
5
To avoid possible problems of bias from the selection of these full-time workers, a two-stage estimation procedure proposed by
Heckman (1976) is applied. In the ®rst stage of this procedure, a probit analysis is applied to estimate a function determining whether
an individual works full-time year round. This process yields an estimate of lambda, the inverse of Mill’s ratio, known as the hazard rate
in reliability theory. The estimated lambda is then used as a regressor in the earnings function, along with other factors that in¯ uence
earnings. According to Heckman, the results of this procedure can be quite close to those yielded by maximum likelihood estimation.
While the resulting estimator is not e cient (in the statistical sense) compared to a maximum likelihood estimator, it is much simpler
computationally. This correction is thought to be more critical for women than for men, since women display more variability in work
behaviour. However, for the sake of comparison, the correction is used for both genders here.
6
The data set reports accurate measurements of earnings up to $139999. Income levels above $140000 are censored and reported as the
state median of incomes in excess of $140000.
7
X
Let ­ represent the coe cient on variable X in the estimated earnings equation. Then high school graduates earn an extra amount ­ HS,
compared to those who did not complete high school. College graduates earn an extra amount ­ HS‡ ­ COLLEGE, compared to high-school
dropouts. Dummy variables based on degree completion are easily constructed from the data set; years of schooling cannot be accurately
inferred in all cases.
8
Information on actual years of labour market experience is not available in this data set. It is quite common to measure potential
experience as age ± years of schooling ± 5. However, this practise has been questioned in recent research (Murphy and Welch, 1990; Light
and Ureta, 1995). Given these concerns and the imprecise measurement of years of schooling in the PUMS data, the present speci®cation
uses AGE and education variables separately, rather than in linear combination.
9
The use of broad occupational categories in wage studies is common. However, to the extent that there are wage diŒerences among the
occupations within a broad category, these diŒerences may be mistakenly attributed to sexual orientation, if the distribution of indi-
viduals among occupations within the broad category is systematic, and not uniform, by sexual orientation.

4
0
S. H. Clain and K. Leppel
Table 1. De®nitions of variables
Variable name
De®nition
LNINC
PINCOME
SAMESEX
WHITE
COLLEGE
HS
Natural logarithm of 1989 wages or salary
1989 income of spouse or unmarried partner, if any
1 if individual has a same-sex unmarried partner; 0 otherwise
1 if race of individual is white; 0 otherwise
1 if individual completed college; 0 otherwise
1 if individual completed high school; 0 otherwise
Years of age, if 90 or less; 90 otherwise
AGE
AGESQ
CITY
HLTHLIM
Square of age
1 if individual resides in a central city location, MSA or PMSA; 0 otherwise
1 if individual is limited in kind or amount of work, has a mobility limitation, or has a personal care limitation;
otherwise
1 if individual is ¯ uent in the English language; 0 otherwise
0
FLUENT
DKIDS
NEAST
MDWEST
WEST
1 if individual is female and living with minor children; 0 otherwise
1 if individual lives in the northeast; 0 otherwise
1 if individual lives in the midwest; 0 otherwise
1 if individual lives in the west; 0 otherwise
1 if individual’s occupation is among precision production, craft, or repair occupations, or the individual works
as an operator, fabricator or labourer; 0 otherwise
BLCOLL
WHCOLL
1 if individual’s occupation is among managerial or professional specialties, or the individual works in a
technical, sales or administrative support position; 0 otherwise
1 if individual is in a service occupation; 0 otherwise
SERVICE
LAMBDA
Inverse of the Mill’s ratio, estimated from probit equation results reported in the Appendix
adult partner who works may allow the individual to
accept a lower wage in exchange for more pleasant working
conditions, ceteris paribus, compared to individuals with-
out such partners. On the other hand, the presence of a
partner may indicate a greater need for household income,
the other explanatory variables were added to the equa-
1
tions.
3
For convenience, the de®nitions of the variables are sum-
marized in Table 1. Table 2 provides descriptive statistics
for males living with male partners, males living with
female partners, and males not living with partners.
Table 3 provides the same information for females.
1
0
resulting in the opposite eŒect. The 1989 income of one’s
adult partner (PINCOME), if any, is used to help sort out
the eŒects of partnership. For women, the presence of one’s
own minor children (DKIDS= 1) may force the individual
to seek more ¯ exible working schedules, and accept corre-
spondingly lower wages, compared to individuals with no
It can be seen from these tables that households of gay
and lesbian couples are quite rare or under-reported.
Substantial undercounting of gays and lesbians must be
occurring, if one is to believe the ®gure of 10% frequently
1
1,12
14
such children (DKIDS= 0).
cited (Woods, 1993; p. 274).
The eŒect of sexual orientation on earnings may not
take the form of simply a parallel shift of the earnings
equation. Therefore, terms interacting SAMESEX with
It can also be seen that the percentage of non-white
persons is smaller among men living with partners of the
same sex than among other men. The diŒerence in racial
1
1
0
1
Empirical evidence for men shows the latter tendency, while that for women shows the former tendency (Light and Ureta, 1995).
This variable is based on a question that was asked only of women. While it should, theoretically, be possible to scan all records of
persons in each household to impute a measurement for men, it would greatly add to the complexity of the computer programming.
Moreover, it is not clear that biological paternity could be deduced, in the way that biological maternity is addressed directly for women.
Therefore, this option was not pursued.
1
2
Wood et al. (1993) found that mothers earned no less than women without children, once hours and experience were controlled. In the
absence of accurate information on the latter, DKIDS picks up the eŒects of reduced hours and/or experience.
1
3
With no a priori basis for limiting these interactions to a select few, a speci®cation search was conducted to determine which interaction
terms were empirically important, given the speci®c samples.
While it has been assumed that gays and lesbians who are not living in households with unmarried partners have disclosed nothing
1
4
about their sexual orientation in the workplace, it is perhaps na õÈv e to do so. Badgett presented arguments that errors in measuring the
extent of workplace disclosure result in a bias against ®nding discrimination. For example, it could be argued that gay and lesbian
individuals who are not living with partners might have nevertheless disclosed sexual orientation in the workplace and so might face
discrimination at work. If so, it would lower (marginally)the average earnings of the group against which individuals with partners of the
same sex are measured. One may also argue that those in partnership may not have disclosed their living arrangements or sexual
orientation to employers, co-workers and consumers, and so do not face discrimination at work. Moreover, these individuals may
possess personal characteristics, both measured and unmeasured (e.g. greater sense of commitment), which make them more desirable
employees as well as partners. Again, these eŒects would result in a bias against ®nding discrimination.

Sexual orientation discrimination in wage diŒerences
41
a
Table 2. Descriptive statistics : males
With
male
partners
With
female
partners
With
no
partners
Mean earnings (natural log)
Percentage white
Percentage living in cities
Percentage high school graduates
Percentage college graduates
Mean age
10.12
10.26
10.12
91.21%
95.60%
93.41%
46.15%
33.87
87.34%
74.47%
84.42%
26.34%
40.69
84.17%
79.53%
87.59%
29.26%
37.70
Percentage with health limitations
Percentage living in northeast
Percentage living in midwest
Percentage living in west
Percentage blue collar
Percentage white collar
Percentage in service occupations
Percentage ¯ uent
9.89%
18.68%
9.89%
35.16%
15.38%
73.63%
9.89%
5.56%
6.61%
21.11%
24.97%
20.11%
42.15%
48.39%
7.61%
19.79%
23.38%
24.44%
37.31%
50.58%
9.87%
98.90%
98.05%
97.85%
Mean partner’s income
Number of observations
$28723
91
$12554
31 153
±
5585
a
Notes: In the interest of greater comparability among workers, this study includes only persons
who worked at least 35 hours per week and at least 40 weeks per year.
a
Table 3. Descriptive statistics : females
With
female
partners
With
male
partners
With
no
partners
Mean natural logarithm of earnings
Percentage white
Percentage living in cities
Percentage high school graduates
Percentage college graduates
Mean age
10.17
9.80
9.87
93.10%
86.21%
94.83%
53.45%
37.57
85.31%
74.71%
88.53%
22.23%
39.02
77.42%
81.72%
86.89%
25.75%
40.90
Percentage with health limitations
Percentage living in northeast
Percentage living in midwest
Percentage living in west
Percentage blue collar
Percentage white collar
Percentage in service occupations
Percentage ¯ uent
3.45%
4.18%
6.58%
39.66%
15.52%
29.31%
15.52%
75.86%
8.62%
19.93%
23.97%
19.82%
13.45%
76.34%
9.81%
22.06%
20.93%
20.66%
13.27%
73.39%
12.99%
98.49%
30.58%
96.55%
12.07%
98.10%
47.08%
Percentage living with minor children
Mean partner’s income
Number of observations
$26222
58
$29951
18 367
±
7 603
a
Notes: In the interest of greater comparability among workers, this study includes only persons
who worked at least 35 hours per week and at least 40 weeks per year.
composition is similar for women. The percentage of per-
sons living in urban areas is greater for women living with
partners of the same sex than for other women; the gap is
much greater for men. Individuals living with partners of
the same sex are more educated than other individuals.
Individuals living with partners of the same sex are also
Unintentional misreporting by gay and lesbian couples, systematically linked with low education and/or low income, would also
contribute to a bias against ®nding discrimination. Gay and lesbian couples intentionally not reporting themselves to be living with
partners of the same sex are likely to have hidden their sexual orientation from their employers and co-workers, as well. If they have not,
and face discrimination as a result, the end product would be, if anything, a bias against ®nding discrimination. Most likely, the
magnitude of this bias would be mitigated by the characteristics that make these individuals more desirable employees as well as
partners, and the small size of this group, relative to heterosexuals in the sample. As a singular exception, a bias towards ®nding
discrimination would exist if there were self-selection of high-income couples into intentional misreporting.

4
2
S. H. Clain and K. Leppel
younger than other individuals, on average, in this sample;
the gap is greater for men than for women.
for males and females, respectively. Terms with low t-
1
6
statistics were dropped. The equations reported show
the results for two samples: a sample of individuals living
with married or unmarried partners, and a sample of indi-
viduals living without partners or with partners of the same
The average earnings of men living with male partners
are virtually identical to those of women living with female
partners. The average earnings level was greater among the
women living with partners of the same sex than among
other women. The reverse was true for men. Given the
educational advantage, individuals living with partners of
the same sex should experience higher earnings.
Furthermore, the tendency of these individuals to live in
more populous areas, where employment prospects are
generally better and cost of living is higher, could further
increase the level of earnings. Also, because individuals
with partners of the same sex are less likely to be non-
white, racial discrimination would tend to increase the
earnings advantage for these individuals relative to others.
However, ceteris paribus, discrimination on the grounds of
sexual orientation may result in lower earnings for individ-
uals with partners of the same sex. Statistical analysis that
takes these other variables into consideration is necessary
to determine whether individuals living with partners of the
same sex suŒer from lower earnings than otherwise com-
parable individuals.
1
7
sex.
The results show that men living with partners of the
same sex have lower earnings than otherwise similar men.
It is found that these men earn about 22% less than men
1
8
not living with partners, ceteris paribus. Men living with
male partners earn 16% (27% ; 35% ) less than comparable
men living with female partners, if they are college-
educated (working in blue collar occupations; living in
1
9
the Midwest).
However, women living with partners of the same sex
tend to have higher earnings than otherwise similar
women. At most ages, women with female partners have
earnings that are signi®cantly greater than those of women
not living with partners, ceteris paribus. This is particularly
true if they are living in the Midwest or with their own
minor children, in which case they also earn more than
women living with male partners. Between the ages of
37±48, however, women with female partners are at a slight
disadvantage, if childless and not living in the Midwest;
they have earnings which are at most 4% lower than
2
0,21
IV. ESTIMATION RESULTS
women not living with partners, ceteris paribus.
One explanation for the observed earnings diŒerential
between individuals living with partners of the same sex
and other individuals is that employers, co-workers, and/
or consumers discriminate against gays in favour of hetero-
The statistical package SAS was employed to estimate pro-
1
5
bit and earnings equations. Tables 4 and 5 report the
results of the statistical estimation of earnings equations
As Badgett pointed out, however, statistical correction for errors in categorizing those facing possible discrimination on the basis of
sexual orientation is not feasible, without direct information on disclosure and, in the present case, sexual orientation of those reportedly
not living with partners.
1
1
5
6
The results of the probit analysis are discussed in the Appendix.
To facilitate the speci®cation search, a stepwise regression procedure with backward elimination is used. The approach leads to
diŒerent speci®cations for diŒerent samples. This outcome should not be surprising as an empirical possibility, given that there may
be less variability in X or higher multicollinearity of X with the other X’s (and hence no ®nding of a signi®cant eŒect of X ) in one
k
k
k
sample than in the other.
It is not obvious, a priori, whether individuals living with partners of the same sex have more intangible characteristics in common with
1
7
individuals living with heterosexual partners or with individuals living with no partners. In the former case, the individuals are likely to
be more similar with respect to the intangible characteristics that contribute to the decision to form a partnership. (This issue is
mentioned in note 14.) In the latter case, the individuals are likely to be more similar with respect to on-the-job behavioural character-
istics associated with low or no risk of fertility. Given these considerations, the investigation is conducted using both samples.
1
8
This result is calculated as antilog (70.25097267)71, using the coe cient for SAMESEX. Note that this gap is narrower for men
whose male partners earn income. (It decreases by 5.3 percentage points for every 10 000 dollars of the partner’s income.)
These results are calculated using the coe cients of interaction terms individually. For example, for college-educated men, the result is
1
9
calculated as antilog (70.17957414)71, assuming that BLCOLL= 0 and MDWEST= 0, so that the interaction terms involving SAME-
SEX and these other variables contribute nothing additional.
In comparison to women not living with partners, the eŒect of homosexuality on the natural logarithm of females’ earnings is
2
0
2
.2547632570.10908765AGE+ 0.00129331AGESQ, for the base case, assuming MDWEST= 0 and DKIDS= 0. This eŒect is positive
prior to age 37 and after age 48. It is minimized at approximately 42 years of age, when the expression equals 70.04. The reported
percentage gap is calculated as antilog(70.04)71. The variation with age may be caused by the diŒerences in the nature of the
comparison group of women not living with partners, at diŒerent ages. Among the younger cohorts are those less committed to the
labour force, desiring marriage, but not yet married. Their number dwindles with age, as these individuals marry out of the sample. An
opposing eŒect is occurring among the older cohorts, as the number of those surprised by divorce, unprepared for the labour force,
increases.
2
1
A female partner may contribute to home production in a way that enhances the mother’s market productivity. If so, the estimated
coe cient on the interaction of SAMESEX and DKIDS could be a re¯ ection of her reward, in the absence of variables measuring her

Sexual orientation discrimination in wage diŒerences
43
a
Table 4. Results for estimation of earnings equations : males
sexual men and against heterosexual women in favour of
lesbians.
b
c
Variable
INTERCEPT
WHITE
Coe cient
Coe cient
One interpretation of this ®nding is that employers, co-
workers, and/or consumers are discriminating in favour of
the personality characteristics of the stereotypical hetero-
sexual male, rather than discriminating against gays and
heterosexual females. Perhaps the more closely an
employee ®ts that mould, the more favourably the individ-
ual is looked upon. In their review of the literature on
psychological test results and female homosexuality,
Riess et al. (1974) found that non-projective studies showed
lesbians to be `more dominant, autonomous, assertive and
detached ± in fact, more like the stereotypic male’. Thus,
employers, co-workers, and/or consumers may prefer the
personality characteristics of lesbian employees to those of
heterosexual women employees.
7.166***
8.165***
(0.0001)
0.116***
(0.0001)
0.256***
(0.0001)
0.173***
(0.0001)
0.048***
(0.0001)
70.0004***
(0.0001)
0.175***
(0.0001)
±
(
0.0001)
0.178***
0.0001)
0.282***
0.0001)
0.210***
0.0001)
0.082***
0.0001)
(
(
(
(
COLLEGE
HS
AGE
AGESQ
70.0008***
(
0.0001)
0.195***
0.0001)
CITY
(
HLTHLIM
FLUENT
NEAST
70.175***
(
(
(
(
(
(
(
(
0.0001)
0.280***
0.0001)
0.122***
0.0001)
0.056***
0.0001)
0.091***
0.0001)
0.410***
0.0001)
0.294***
0.0001)
0.133***
0.0001)
±
0.188***
(0.0001)
0.135***
(0.0001)
0.051***
(0.0064)
0.076***
(0.0001)
0.307***
(0.0001)
0.198***
(0.0001)
Other explanations for the earnings diŒerentials are
possible, however. To a large extent, these explanations
revolve around the immeasurable or unobservable factors
that may contribute to the earnings gap. Correlation of
sexual orientation with these omitted variables may cause
the SAMESEX variable to re¯ ect their eŒects.
MDWEST
WEST
WHCOLL
BLCOLL
SERVICE
PINCOME
SAMESEX
One factor concerns the amount of on-the-job training
that the individual has. Relative to other females, females
living with female partners may have invested more heavily
in on-the-job training, because of a stronger commitment
±
5.12/10 **
0.0480)
6
22
to life-long labour force participation. This diŒerence
(
may be responsible for the observed wage gap, but is not
being attributed appropriately, in the absence of a variable
that measures investment in on-the-job training.
±
70.251***
(
0.0063)
±
SAMESEX*COLLEGE 70.180**
0.0263)
SAMESEX*MDWEST 70.426**
0.0187)
SAMESEX*BLCOLL 70.315**
(
Another explanation for the observed earnings dispari-
ties is suggested by the recent work of Goldsmith et al.
(1997) that shows that self-esteem, as psychological
human capital, has a positive eŒect on earnings. The eŒect
of this unobservable factor may possibly be re¯ ected in the
positive (negative) impact of living with a partner of the
same sex for women (men). Lesbians (gays) may have
greater (less) self-esteem in market work, because they
are more (less) like the stereotypic heterosexual male
worker, in comparison to observationally-equivalent
women (men) who are not living with partners of the
same sex.
±
±
(
(
0.0216)
LAMBDA
0.066
0.4744)
0.295
70.272***
(0.0001)
0.282
148.3***
(0.0001)
(
2
R
F statistic
726.4***
0.0001)
31 244
(
n
5676
a
bNotes: p-values are in parentheses.
sample of males with partners.
c
sample of males with male partners or no partners.
signi®cant at the 10% level;
*
*
*
*
The existing evidence is consistent with all three of the
23
above explanations. Further study of this question with a
signi®cant at the 5% level;
signi®cant at the 1% level.
**
hours of work and experience, or her partner’s time spent in home production. Given the small number of women on which the estimate
is based, however, one should not place too much emphasis on the magnitude of the eŒect. Moreover, one should realize that the result
could be caused by simultaneity bias: lesbians with higher incomes are more likely to have children. They are more likely to be able to
maintain custody of children born as a result of a previous heterosexual relationship, and more likely to be able to aŒord arti®cial
insemination or be approved for adoption.
2
2
2
3
These women may have also chosen career tracks (with higher starting wages) in ®rms with delayed-compensationincentive systems.
It must be remembered that the bias against ®nding discrimination, discussed in note 14, may be responsible for some of the ®ndings for
women.

4
4
S. H. Clain and K. Leppel
a
Table 5. Results for estimation of earnings equations : females
richer data set would help to determine which explanations
are correct.
b
c
Variable
INTERCEPT
WHITE
COLLEGE
HS
Coe cient
Coe cient
In the current study, one cannot ignore the clues revealed
by the diŒerences between individuals with partners of the
same sex and others, in the eŒects on earnings of other
independent variables. For example, the age-earnings pro-
®les of women with female partners are higher, ¯ atter, and
less concave than those of other women. This result may be
associated with diŒerences in job choices and/or patterns of
investment in on-the-job training, and not necessarily the
outcome of discrimination.
8.169***
0.0001)
±
6.857***
(0.0001)
0.065***
(
(
0.0001)
0.305***
0.338***
0.0001)
0.158***
0.0001)
0.036***
0.0001)
(
(
(
(0.0001)
0.220***
(0.0001)
0.090***
(0.0001)
AGE
AGESQ
CITY
70.0004***
70.0010***
(0.0001)
0.260***
(0.0001)
70.271***
On the other hand, consider the regional variations in
the earnings of individuals living with partners of the same
sex, relative to others. Regional variations may re¯ ect
regional diŒerences in attitudes towards homosexuality.
(
0.0001)
0.185***
0.0001)
±
(
HLTHLIM
FLUENT
NEAST
MDWEST
WEST
(
0.0001)
0.199***
24
Such diŒerences are documented by opinion polls. They
are also re¯ ected in state-by-state diŒerences in the legal
0.189***
0.0001)
0.120***
0.0001)
0.023**
0.0169)
0.116***
0.0001)
0.276***
0.0001)
0.175***
0.0001)
±
(
(
(
(
(
(
(0.0001)
0.114***
(0.0001)
±
2
5
treatment of gays and lesbians. While correlation of these
diŒerences with the observed coe cients on regional inter-
action variables in Tables 4 and 5 is not obvious, it does
provide food for thought. It is quite plausible that regional
diŒerences in discriminatory behaviour are responsible for
these latter results, and so contribute to the earnings gap
between gays and lesbians and heterosexuals.
0.077***
(0.0001)
0.394***
(0.0001)
0.272***
(0.0001)
±
70.155***
(0.0001)
±
WHCOLL
BLCOLL
SERVICE
DKIDS
70.032**
(
0.0310)
6
PINCOME
3.53/10 ***
0.0001)
±
V. CONCLUSIONS
(
SAMESEX
2.255**
Analysis of the data from the 1/1000 Public Use Microdata
Sample of the 1990 Census of Population indicates that,
ceteris paribus, men living with male partners tend to earn
less than other men, while women living with female part-
ners earn more than other women. In addition to ®nding
an earnings gap between individuals with partners of the
same sex and others, the current study also reveals diŒer-
ences in the eŒects on earnings of other personal character-
istics. First, there tend to be regional variations in the
earnings diŒerentials between individuals living with part-
ners of the same sex and other individuals. Second, these
diŒerentials tend to vary by education and occupation for
men, and with age and presence of children for women.
These results may be due in part to omitted variables
correlated with sexual orientation. Further research with
a richer data set would help to clarify the role of these
variables. However, it could be argued that the results
re¯ ect the existence of regional diŒerences in discrimina-
(
7
(
0.0299)
0.109**
0.0455)
0.0013*
0.0616)
0.313*
SAMESEX*AGE
SAMESEX*AGESQ
SAMESEX*MWEST
SAMESEX*DKIDS
LAMBDA
±
±
(
0.330**
0.0406)
0.371**
0.0430)
(
(
(0.0755)
0.513**
(0.0106)
0.253*
(0.0573)
0.305
70.085**
0.0364)
0.282
451.9***
0.0001)
18 425
(
2
R
F statistic
176.6***
(0.0001)
7661
(
n
a
bNotes: p-values are in parentheses.
sample of females with partners.
c
sample of females with female partners or no partners.
signi®cant at the 10% level;
*
*
*
*
signi®cant at the 5% level;
signi®cant at the 1% level.
**
2
4
For example, the 1990 General Social Survey has been investigated and empirical evidence was found that suggests that disapproval of
homosexuality is greater in the Northeast and the Midwest than in the West, but not as great as in the South. Intraregional diŒerences
complicate the interpretation of these patterns. Speci®cally, in the South, there is greater disapproval in the South Central states than in
the South Atlantic states. In the West, there is greater disapproval in the Mountain states than in the Paci®c states. A ®ner classi®cation
of geographical location would be ideal, but would further split samples which are already small in this study. (For example, only ten
lesbians in the PUMS data set live in the South.)
2
5
In the early 1990s, only seven states (California, Connecticut, Hawaii, Massachusetts, New Jersey, Vermont and Wisconsin) had laws
prohibiting discrimination in employment on the basis of sexual orientation. One state (Colorado) had a law prohibiting any protection
on the basis of sexual orientation. For more information, see Norris and Randon (1993). Assuming no sudden shifts in popular opinions

Sexual orientation discrimination in wage diŒerences
45
tory behaviour. It remains to be seen whether the federal
government will amend its legislation so as to prohibit such
discriminatory behaviour.
employed full-time year round, versus not full-time or not
full year. The sample does not include individuals who
never worked or who last worked prior to 1985, inasmuch
as there is no occupational information for these individ-
uals.
REFERENCES
For the most part, the signs of the coe cients make
sense. The probability of being a full-time year round
worker increases with age, but at a rate that diminishes
with age. City dwellers and individuals ¯ uent in the
English language are more likely to be full-time year
round workers, presumably because of greater labour mar-
ket opportunities for individuals with these characteristics.
Individuals who are in farming, forestry, or ®shing are less
likely to work full-time year round; seasonality of demand
in these occupations surely is a factor. Women in house-
holds with their own minor children are less likely to be
full-time year round workers; the presence of minor chil-
dren may make it di cult for mothers to hold a full-time
job.
Badgett, M. V. L. (1985) The wage eŒects of sexual orientation
discrimination, Industrial and Labor Relations Review,
48, 726±39.
Becker, G. S. (1971) The Economics of Discrimination, University
of Chicago Press.
Beller, A. H. (1982) Occupational segregation by sex: determi-
nants and changes, Journal of Human Resources, 17, 371±92.
Goldsmith, A. H., Veum, J. R. and Darity, W., Jr. (1997) The
impact of psychological and human capital on wages,
Economic Inquiry, October, 35, 815±29.
Gunderson, M. (1989) Male±female wage diŒerentials and policy
responses, Journal of Economic Literature, March, 27, 46±72.
Heckman, J. (1976) The common structure of statistical models of
truncation, sample selection and limited dependent variables
and a simple estimator for such models, Annals of Economic
and Social Measurement, 5, 475±92.
Leonard, J. S. (1984) Antidiscrimination or reverse discrimi-
nation: the impact of changing demographics, title VII and
a rmative action on productivity, Journal of Human
Resources, Spring, 19, 145±74.
Light, A. and Ureta, M. (1995) Early-career work experience and
gender wage diŒerentials, Journal of Labor Economics,
January, 13, 121±54.
The negative sign of COLLEGE requires some explana-
tion. College-educated individuals are less likely to be
full-time year round workers than individuals with a
high school diploma. Perhaps these individuals are more
likely to be self-employed and/or in¯ uenced by the income
eŒect of high wages in making their labour supply
decisions, compared to individuals with only a high school
diploma. Nevertheless, in comparison to individuals
who have not completed high school, those who are
college-educated are more likely to be full-time year
round workers, since the summed eŒect of their degrees
is positive.
Murphy, K. M. and Welch, F. (1990) Empirical age-earnings
pro®les, Journal of Labor Economics, April, 8, 202±29.
Norris, M. P. and Randon, M. A. (1993) Sexual orientation and
the workplace: recent developments in discrimination and
harassment law, Employee Relations Law Journal, Autumn,
19, 233±46.
Riess, B. F., Safer, J. and Yotive, W. (1974) Psychological test
data on female homosexuality: a review of the literature,
Journal of Homosexuality, Fall, 1, 71±85.
Smith, J. P. and Welch, F. R. (1989) Black economic progress
after Myrdal, Journal of Economic Literature, June, 27,
The results of the ®rst stage of estimation are employed
in the second stage for both genders, for purposes of com-
parison. The adjustment proves to be signi®cant in three of
the four earnings equations, the exception being the equa-
tion involving men living with female partners. For the
sample of men with male partners or no partners, the coef-
519±64.
Wood, R. G., Corcoran, M. E. and Courant, P. N. (1993) Pay
diŒerences among the highly paid: the male±female earnings
gap in lawyers’ salaries, Journal of Labor Economics, July, 11,
417±41.
Woods, J. D. (1993) The Corporate Closet: The Professional Lives
of Gay Men in America, The Free Press, New York.
®
cient of lambda is signi®cantly negative. Thus, for this
group, the individuals are negatively self-selected; those
who are more likely to work full-time year round are
more likely to have lower earnings, ceteris paribus.
For the sample of women with male or female partners,
there is similar negative self-selection. For the sample
of women with female or no partners, there is positive
self-selection.
APPENDIX
Tables A1 and A2 show the results of probit estimation for
males and females, respectively. The coe cients measure
the in¯ uence of the variables on the probability of being
between 1989 and the early 1990s, these diŒerences in state laws re¯ ect diŒerences in attitudes towards homosexuality that could have
aŒected 1989 earnings.

4
6
S. H. Clain and K. Leppel
a
Table A1. Results for estimation of probit equations : males
Table A1. continued
b
c
b
c
Variable
Coe cient
Coe cient
Variable
Coe cient
Coe cient
INTERCEPT
72.677***
73.767***
(0.0001)
0.095**
log likelihood
n
715 533.7
37 958
No. full-time year round 31 244
74006.0
7728
5676
(
0.0001)
0.191***
0.0001)
WHITE
(
(0.0296)
70.124***
(0.0040)
0.254***
(0.0001)
0.201***
(0.0001)
70.0024***
(0.0001)
0.148***
(0.0002)
70.733***
(0.0001)
0.110
(0.2725)
70.005
(0.9175)
0.015
(0.7335)
70.096**
(0.0264)
0.379***
(0.0001)
0.241**
a
bNotes: p-values are in parentheses.
sample of males with partners.
COLLEGE
70.035
(
(
(
0.1347)
0.222***
0.0001)
0.156***
0.0001)
c
sample of males with male partners or no partners.
signi®cant at the 10% level;
HS
*
*
*
*
signi®cant at the 5% level;
signi®cant at the 1% level.
AGE
**
AGESQ
70.0020***
(
(
0.0001)
0.124***
0.0001)
CITY
a
Table A2. Results for estimation of probit equations : females
b
c
HLTHLIM
70.876***
Variable
Coe cient
Coe cient
(
(
(
(
0.0001)
0.252***
0.0001)
0.060***
0.0094)
0.034
FLUENT
INTERCEPT
72.430***
(0.0001)
70.239***
(0.0001)
0.017
(0.3419)
0.187***
(0.0001)
0.133***
(0.0001)
70.0017***
(0.0001)
0.089***
(0.0001)
70.519***
(0.0001)
0.097**
74.164***
(0.0001)
0.072**
NEAST
WHITE
(0.0166)
70.130***
(0.0001)
0.230***
(0.0001)
0.210***
(0.0001)
70.0025***
(0.0001)
0.166***
(0.0001)
70.700***
(0.0001)
0.075
MDWEST
COLLEGE
HS
0.1195)
WEST
70.097***
(
(
(
(
(
(
(
(
0.0001)
0.334***
0.0001)
0.161***
0.0012)
0.136**
0.0136)
1.73/10 ***
0.0065)
3.762
0.9993)
0.313
0.4985)
0.369
0.2810)
WHCOLL
AGE
BLCOLL
AGESQ
(0.0108)
0.195*
(0.0588)
SERVICE
CITY
6
6
PINCOME
78.52/10
(0.1113)
HLTHLIM
FLUENT
SAMESEX
4.684
(0.9995)
0.513
(0.2725)
0.610*
(0.0450)
70.116***
(0.0001)
70.079***
(0.0001)
70.123***
(0.0001)
0.630***
(0.0001)
0.663***
(0.0001)
0.075
(0.4121)
70.052
SAMESEX*WHITE
SAMESEX*COLLEGE
SAMESEX*HS
SAMESEX*AGE
SAMESEX*AGESQ
SAMESEX*CITY
SAMESEX*HLTHLIM
SAMESEX*FLUENT
SAMESEX*NEAST
SAMESEX*MDWEST
SAMESEX*WEST
SAMESEX*WHCOLL
SAMESEX*BLCOLL
SAMESEX*SERVICE
NEAST
(0.1343)
70.066*
(0.0516)
70.122***
(0.0004)
0.515***
(0.0016)
0.539***
(0.0011)
0.030
MDWEST
WEST
(0.0908)
70.509
(0.4209)
0.034
(0.7273)
70.0005
(0.6865)
710.804
(0.9992)
0.211
70.361
(
0.5601)
0.056
0.5597)
WHCOLL
BLCOLL
(
70.0007
0.5457)
710.273
(
SERVICE
(
(
(
0.9987)
0.429
0.3327)
0.688
0.5949)
(0.3823)
70.551***
(0.0001)
(0.8528)
70.498***
(0.0001)
DKIDS
(0.6400)
1.051
6
6
PINCOME
SAMESEX
SAMESEX*WHITE
SAMESEX*COLLEGE
SAMESEX*HS
SAMESEX*AGE
SAMESEX*AGESQ
75.80/10 ***
8.44/10
(0.4496)
70.723
(0.1514)
71.483***
(0.0028)
70.569
(0.1827)
5.222
(0.9994)
5.571
(0.9994)
5.195
(0.9994)
(0.0001)
77.232
(0.9995)
1.450
(0.1857)
71.674
(0.1772)
2.374
(0.9996)
1.104*
(0.0826)
70.0115***
(0.1605)
(0.6350)
77.110
(0.9997)
1.481
(0.2047)
71.832
(0.1924)
2.107
(0.9998)
1.125
(0.1067)
70.0117
(0.1919)
70.745
(
0.1311)
71.325***
(
0.0058)
70.618
(
(
(
(
0.1377)
4.966
0.9991)
5.384
0.9991)
5.035
0.9991)

Sexual orientation discrimination in wage diŒerences
47
Table A2. continued
b
c
Variable
Coe cient
Coe cient
SAMESEX*CITY
1.530
0.4525)
74.229**
1.360
(0.6059)
74.354**
(0.0391)
76.959
(0.9997)
1.062
(0.4796)
71.269
(0.5470)
73.223
(0.1157)
711.600
(0.9991)
±
(
SAMESEX*HLTHLIM
SAMESEX*FLUENT
SAMESEX*NEAST
SAMESEX*MDWEST
SAMESEX*WEST
(
0.0307)
76.907
(
(
0.9995)
1.297
0.3563)
70.789
(
0.6941)
72.760
(
0.1160)
SAMESEX*WHCOLL
711.511
(
0.9985)
SAMESEX*BLCOLL
SAMESEX*SERVICE
±
710.834
711.112
(0.9992)
7.485
(
(
0.9986)
6.861
0.9989)
SAMESEX*DKIDS
(0.9993)
Log likelihood
n
No. full-time year-round
723 012.9
35 866
18 425
76 872.9
12 044
7661
a
Notes: p-values are in parentheses.
b
sample of females with partners.
sample of females with female partners or no partners.
signi®cant at the 10% level;
c
*
*
*
*
signi®cant at the 5% level;
signi®cant at the 1% level.
**