Content uploaded by Sergey Alexeev
Author content
All content in this area was uploaded by Sergey Alexeev on Sep 29, 2022
Content may be subject to copyright.
Wage equation misrepresents
gay wage discrimination:
overlooked evidence from Russia
This document is a post-print; please cite the published version:
Alexeev, S. “Wage equation misrepresents gay wage discrimination: overlooked
evidence from Russia.” International Journal of Manpower (2022)
https://doi.org/10.1108/IJM-08-2021-0475
Published under the Creative Commons Attribution Non-commercial International Licence 4.0
(CC BY-NC 4.0). Any reuse is allowed in accordance with the terms outlined by the licence.
To reuse the AAM for commercial purposes, permission should be sought by contacting
permissions@emeraldinsight.com.
Sergey Alexeev
sergei.v.alexeev@gmail.com
University of New South Wales, NSW, Australia
February, 2022
Abstract
Purpose – Only data from developed countries were used to estimate
the sexual orientation difference in wages. This paper is the first to
identify the wage discrimination of gay men in Russia – a country where
institutional discrimination and ignorance against gay men are known to
present.
Design – Gays are identified as men who reported having sex with other
men in several waves of the national household survey. A wage equation
is used to estimate the gay wage penalty. Extending the wage equation
to implement a difference-in-difference design, the paper also evaluates
the effect of the gay-propaganda law of 2013 on gay wages.
Finding – No wage discrimination is identified. The law also has no
adverse effect on gay wages.
Originality – These results are implausible and add to existing evi-
dence that gay discrimination measured with wage equation suffers from
endogeneity and should be interpreted with caution. Particular caution
should be exercised in cross-sectional and time-series comparisons, as a
tendency to report the orientation honestly and unobserved confounders
vary by location and time.
Practical implications – Cross-country comparison and theoretical
generalizations are premature, and better identification strategies are
needed to understand sexual orientation differences.
Social implications – Policymakers should be aware that in both dis-
criminatory and equitable environments, there may be hidden inequality
even if researchers do not detect it.
Keywords Sexual Orientation, Discrimination, Earnings
Paper type Research paper
1 INTRODUCTION 2
1 Introduction
Following the work by Badgett (1995), that applied contemporary at
the time race discrimination statistical methods to study sexual orienta-
tion discrimination, estimating wage differences between gay and straight
men became a firmly-established research question with hundreds of con-
tributions. From 1989 to 2014, on average, gay men were paid less by
9% (Drydakis 2019) and from 2012 to 2020 by 6.8% (Drydakis 2021).
This reduction is taken as evidence that the acceptance of gay men is
increasing.
This evidence is used to support various versions of the minority stress
theory (MST) posited by Meyer (2003). The theory states that adverse
outcomes experienced by gay men are a direct or indirect result of social
oppression. The examples of theories that accentuate the direct conse-
quences of prejudice include distastes to gay men (Becker 1957), uncer-
tainties concerning gay men performance, or perceived lack of leadership
qualities (Ashenfelter and Rees 2015; K¨ollen 2016). Examples of indirect
harm of prejudice include situations where gay men engage in risky be-
havior to cope with the stress caused by prejudice. The examples include
an excessive drug (Prestage et al. 2009) and alcohol use (Peralta 2008),
and unprotected sex with a large number of sexual partners (Crossley
2004).
The sexual orientation difference is a challenging empirical question.
The empirical literature, following the original contribution by Badgett
(1995), utilizes a standard wage equation, which assumes that a worker’s
characteristic space is exhausted by age and education, and that race (a
clearly visible attribute) and sexual orientation (a deeply private matter)
can be treated similarly from a data modeling perspective. Kitagawa-
Oaxaca-Blinder decomposition – which is based on the same assumptions
as the wage equation and, thus, does not improve the equation’s valid-
ity (Kr¨oger and Hartmann 2021) – occasionally supplements estimates
drawn from the wage equation.
There are only two papers that, instead of replicating an almost 30
years old paper by Badgett (1995), acknowledge the complexities of this
empirical question. Urwin, Mason, and Whittaker (2021) utilize cor-
related random coefficients (a combination of random and fixed effects
that identifies the parameters of time-invariant variables) and show that
ignoring unobserved confounders and orientation fluidity severely misrep-
resent sexual orientation differences. Another work by Alexeev (2022),
notes that interviewers are quasi-randomly assigned to respondents, and
then constructs an instrumental variable for gay status exploiting this
assignment. He then shows that, once endogeneity is resolved, the dif-
ferences disappear, suggesting that the question of sexual orientation
1 INTRODUCTION 3
differences may be ill-imposed.
Neglect of endogeneity issues is particularly deplorable because, un-
like some research questions that exist exclusively as a matter of academic
curiosity, the degree of discrimination is a vital statistic that directly
informs the public and policymaking community. It is crucial to under-
stand whether the acceptance of gays is increasing in countries that put
effort and resources into creating equitable environments. Lamentably,
the current methods do not deliver estimates which allow for time-series
and cross-sectional comparisons. As a result, policymakers do not know
whether they have done enough or more is needed to create an equal
opportunity for everyone. Consequently, the discrimination becomes a
purely political matter, where words and catchy slogans are the means of
societal advancements; instead of impartial scientific quantities and an
effective evidence-based public policy.
The contribution of my paper is to show that the wage equation and
a disregard to the complexity of estimating wage discrimination may
deliver grotesquely incorrect results. More specifically, I apply the wage
equation to the data from Russia – a country where institutional discrim-
ination and ignorance against gay men are well documented (Horne and
White 2020; Hylton et al. 2017; Kondakov 2019). That is, MST holds
trues in Russia; thus, a penalty must be present. I am unable to identify
a penalty. No penalty can also be found following the gay-propaganda
law of 2013. Applying the same interpretation as in previous studies, my
estimates suggest no gay male discrimination in Russia. This can not be
true. The wage penalty cannot be identified because the wage equation
suffers from at least three sources of endogeneity, which are discussed in
Section 2. These sources of endogeneity are present in all countries, but
in Russia, they take on an obvious form.
The results from Russia are particularly important because existing
empirical evidence comes from a handful of developed countries. A re-
searcher from a developed country may not realize that the mere fact
that gay pay is researched already reflects a form of institutional ac-
knowledgment. For example, a famous Russian European University in
St. Petersburg has been closed down because of the publishing of several
research papers that discuss the problems of the LGBT community in
Russia (Kelly 2017). That is how a member of the Russian parliament,
Vitaly Milonov, who lodged an official complaint against the university,
expresses the motivation for it ‘I personally find [gender studies] disgust-
ing, it’s fake studies, and it may well be illegal.’ Views like these permeate
societies that are hostile to gays. This extends to the inability to receive
governmental research grants to, for example, collect data. This is the
essence of institutional discrimination. As a result, the current litera-
ture on gay pay is populated with estimates that, because of models’
2 METHODOLOGICAL ISSUES 4
endogeneity, identify parameters of unknown nature from countries that
generally do not have an overwhelming objection against gay men.
The paper proceeds as follows. Section 2reviews gender studies liter-
ature that point to the methodological complexities of estimating sexual
orientation differences. Section 3introduces the data. Gay men are iden-
tified as men who affirmed that they had sex with another man. Section 4
reviews the methodology. I apply a standard wage equation controlling
for human capital measures. Section 5presents results. No statistically
significant sexual orientation differences in the Russian labor market can
be estimated. Section 6summarize the finding and their significance.
2 Methodological issues
If MST fully reflects reality, then risky health behavior is caused by dis-
crimination. Therefore, controlling for characteristics such as substance
abuse in the wage equation should be avoided as it produces a collider
bias; thereby underestimating the wage penalty. There are, however,
other theories apart from MST that suggest a different causal pathway
structure. Evolutionary psychologists argue that a large number of sex-
ual partners among gay is typical male behavior. The full argument is
that gay men illustrate what men’s sexual behavior would be like without
women to slow them down. In contrast, lesbians, who engage in sex rel-
atively infrequently, demonstrate what women’s sexual behavior would
be like without men to urge them forward (David 2007; Waldis, Borter,
and Rammsayer 2021).
The frequent use of drugs or alcohol by gay men to lower inhibitions
and facilitate intimate contact with strangers (‘chemsex’) is also well
documented (ACSF investigators 1992; Billy et al. 1993; Bourne et al.
2015). Under this causal pathway structure, omitting measures of risky
health behavior – which is common to all single, uncommitted men re-
gardless of sexual orientation and directly affects workplace productivity
– overestimates discrimination.
In 20 years since the introduction of MST, sexologists have also made
further progress in understanding sexual orientation differences. Newer
models inform econometricians of the risk of omitted variable bias when
estimating sexual orientation differences. A limitation of MST that be-
came apparent over time is that it struggles to explain why an increase
in acceptance does not translate into the improvement in well-being by
nonheterosexuals (Bailey 2019). A rejection sensitivity model formulated
by Feinstein (2020) offers an explanation.
This model is based on the notions of rejection sensitivity – an expec-
tation of rejection leads to hypervigilance for signs of rejection. Rejection
sensitivity is well documented and experimentally confirmed defensive
2 METHODOLOGICAL ISSUES 5
mechanism that protects individuals from rejection by preparing them
to detect and respond to social threats. This mechanism can become
maladaptive if activated indiscriminately in response to minimal threat.
In individuals with poor emotional control, the rejection sensitivity may
create a self-fulfilling prophecy in which individuals, in turn, elicit rejec-
tion from others, including lower wages. Past victimization, rejection in
the family, or the dominance of MST in media or public debate heighten
expectations of rejection. In individuals with particular qualities, victim-
ization, dysphoria, or an impression of discrimination might happen even
without social oppression. To sum up, according to this model, sexual
orientation is not a causal factor for lower wages.
The rejection sensitivity model accentuates the importance of individ-
ual resilience and subjective aspects of stress. It addresses the limitations
of MST, which Meyer (2003) also explicitly discusses in his work. He
points out that the MST inevitably describes gays as victims of oppres-
sive social conditions. This ignores individual agency and resilience and
conflicts with an impressive body of research that demonstrates the ca-
pacity of people to cope with stresses. The author intentionally makes an
explicit distinction between objective and subjective stresses. The sub-
jective view of stress highlights individual differences in appraisal and, at
least implicitly, places more responsibility on the individual to withstand
stress. It highlights, for example, processes that lead resilient individu-
als to see discriminatory circumstances as less (or not at all) repressive,
implying that less adaptive individuals are responsible for their stress
experience.
More reasons to worry about the omitted variable bias is that gay
men tend to be younger siblings, with up to 29% of gay men owing their
orientation to being a later-born son (Ablaza, Kab´atek, and Perales 2022;
Blanchard 2018; Blanchard and Lippa 2020). This is known as the fra-
ternal birth order effect – the most consistent biodemographic predictor
of sexual orientation in men. Since a large body of evidence shows that
later-born children are less educated, a sibship size or the birth order
are potentially omitted variables (Black, Devereux, and Salvanes 2005;
Booth and Kee 2009). Apart from sexologists who have the best under-
standing of the subject matter, the economics literature suggests several
omitted variables. Gay men have a distinct pattern of labor market
choices, accumulation of human and health capital, occupational prefer-
ences, specialization within households, and decisions about geographic
location (e.g., Badgett, Carpenter, and Sansone 2021; Black, Sanders,
and Taylor 2007).
Another econometric problem is that revealing one’s sexual orienta-
tion in general and during a survey or in the workplace, in particular, is
a personal choice. This choice correlates with a host of individual char-
2 METHODOLOGICAL ISSUES 6
acteristics. Ignoring the nature of observed gay status, the regression
of outcomes showing the impact of sexual orientation would be biased,
regardless of whether the regression model controlled for all relevant vari-
ables (Greene 2018, Ch. 19.4). For gay men, we only observe a particular
range of wages, where for the rest of the range, the men prefer to conceal
their orientation. Evidence of incidental truncation can be found in the
fact that less-educated gay men tend to conceal their sexual orientation
during the surveys (Black, Gates, et al. 2000). It is also regularly ob-
served that gay men are better educated (Black, Sanders, and Taylor
2007), but at the same time, as just mentioned, likely to be younger sib-
lings. As later-born children are known to be lesser-educated, these facts
can only coexist in the presence of the survivorship bias: only particular
gay men reveal their sexual orientation.
Reverse causality is another potential source of endogeneity. Up to
50% of gay men might change their sexual orientation (Katz-Wise 2015)
for reasons potentially related to wages. This is known as sexual orienta-
tion fluidity and may include situations where men might start denying
their orientation (to themselves or others) if it hurts their professional
and private life opportunities.
Figure 1: Directed acyclic graph.
Gay (true) Wage penalty
Gay (error)
Unobserved confounders
OLS
To summarize, Figure 1encodes the relationship between homosexual
status and wage discrimination in a directed acyclic graph. Variables in
dashed boxes are unobserved. The dotted arrow refers to the correlation
captured with the OLS estimate of the wage equation. Arrows ema-
nating from ‘unobserved confounders’ refers to omitted variable bias, a
bi-directed arrow between the true gay status and wage refers to simul-
taneity, and an arrow between ‘gay (true)’ and ‘gay (error)’ to incidental
truncation.
The empirical exercise of my paper hinges on the observation that the
aforesaid sources of endogeneity became evident in data from developed
relatively gay-affirmative countries. Institutional discrimination and ig-
norance are not the conditions for them. Thus, the Russian context is
not qualitatively distinct; rather, it is likely a special case where all three
sources of endogeneity are present and take on an obvious, particularly
3 DATA 7
informative form. It is, however, impossible to give a data-driven answer
which particular source of endogeneity is dominant relative to some other
countries with better data on gay pay.
3 Data
The choice of a sexual orientation measure in an empirical study has
historically been dictated by its availability within a particular dataset.
For example, the gender of a spouse has often been used as a way to infer
an individual’s sexual orientation (Laurent and Mihoubi 2012; Sabia,
Wooden, and Nguyen 2017; Sansone 2019; Saxby, de New, and Petrie
2020).
Today, a common practice in the national household surveys is to use
a directly self-reported orientation within a multiple-choice question. The
usual choices are ‘straight,’ ‘gay,’ and ‘bisexual.’ These choices are not
considered optimal by the sexologists. From the clinical assessment point
of view, asking about past sexual partners provides a better assessment
of sexual orientation than asking about sexual orientation or identity.
Friedman and Downey (2008) and Salomaa and Matsick (2019) and many
others discuss that a better assessment of sexual orientation includes
items about numbers of past male and female sexual partners and items
about relative attraction to males and females genitalia. A good datum
in clinical assessment asks what the respondent fantasizes about during
masturbation, especially as he approaches orgasm.
The self-identified bisexuals are particularly problematic from both
clinic and statistical angles. In practice, self-reported bisexual men are a
highly heterogeneous category. It could include heterosexually married
men who have sex with guys outside their marriage (or who consistently
desire this) and men who formerly dated women but are now solely inter-
ested in men. They could also include heterosexual men who have had
only a few incidental encounters with other men, but who think they
should use the label ‘bisexual’ to be completely accurate, and men who
interact sexually with other men only in specific contexts, for example,
having oral or anal sex with other men while their wives watch or direct
the action. Because of these ambiguities in definitions and lack of clinical
context, survey assessment of sexual orientation should go well beyond
self-identification.
A categorization that assumes as little as possible about human sex-
uality is heterosexual and nonheterosexual. It is at least ad hoc and
has some advantages. In particular, heterosexual men never report that
they are anything but heterosexuals; this takes care of the orientation
fluidity problem, which occurs within the nonheterosexual category. An-
other advantage is that the target of discrimination is everyone who is
3 DATA 8
not straight, which simplifies the interpretation of the estimates.
This paper uses waves 1994-2015 of the individual questionnaires of
the Russia Longitudinal Monitoring Survey (RLMS). RLMS is a high-
quality national survey used in dozens of publications each year across
all social sciences. The survey is conducted by the University of North
Carolina at Chapel Hill, National Research University Higher School of
Economics, and the Federal Center of Theoretical and Applied Sociology
of the Russian Academy of Sciences. For a history of the survey, an
outline of the sample design and the replenishment of sample designs,
the loss to follow-up, and other key factors, see the data resource profile
in Kozyreva, Kosolapov, and Popkin (2016).
Table 1: Orientation by year
Year Straight Gay Total
1985 945 10 955
1990 1,261 13 1,274
1994 1,515 12 1,527
1995 1,586 10 1,596
1996 1,697 12 1,709
1998 1,951 17 1,968
2000 2,288 16 2,304
2001 2,734 40 2,774
2002 2,759 26 2,785
2003 2,911 47 2,958
2004 2,557 22 2,579
2005 2,374 20 2,394
2006 2,217 17 2,234
2007 2,054 18 2,072
2008 1,871 14 1,885
2009 1,798 16 1,814
2010 1,745 13 1,758
2011 1,628 12 1,640
2012 1,515 10 1,525
2013 1,417 11 1,428
2014 1,312 10 1,322
2015 1,238 10 1,248
Total 41,373 376 41,749
Notes: Gays are identified from the
survey questions asked in 2001 and
2003.
Source: RLMS 1994-2015 (1985
and 1990 collected retrospectively in
2000 and 2001).
In 2001 and 2003, males were asked if they ever had sex with another
man. These questions were not added into RLMS as a move towards gay-
affirmative data collection, but as a part of a medical questionnaire on
the Russian HIV epidemic (Stuikyte, Barbosa, and Kazatchkine 2019).
For this study, those who replied ‘yes’ are interpreted as gay and ‘no’ as
straight. The information for the missing years was obtained through a
panel component of the data (the same respondents respond each year).
The sample was then restricted to men who replied ‘yes’ or ‘no’ in 2001
or 2003. Table 1shows the number of gay and straight men by year.
It is impossible to determine to what extent the fraction of gay men
in the sample is under-reported. The true share of gays is uncertain
3 DATA 9
even for countries with better data (Black, Sanders, and Taylor 2007).
In addition, my work is the only one that uses nationally representative
Russia data; therefore, I can not directly compare my share with share
from convenient follow-up studies (cf. Kon and Riordan 1993). Notwith-
standing this, we do know that the sample size of gay men in the survey is
not small in absolute terms. Smaller sample sizes have been used before
to judge gay wage penalties (e.g., Drydakis 2011). The sample used in
the current study is is larger than some other studies that use national
household surveys and identify gay men using the same-sex partner (e.g.,
Laurent and Mihoubi 2012).
The data for 1985 and 1990 was collected retrospectively in 2000 and
2001. Sabirianova Peter (2003) provides an assessment of the recall bias
(a potential threat to internal validity) in RLMS by comparing answers
on occupation with the official enterprise reports. She concludes that the
recall is not significant due to stable salaries and the strong attachment
of workers to one job in the Soviet period. Age and educational levels
are not available for 1985 and 1990 but can be reconstructed using the
panel nature of the data. RLMS reports the year when each educational
level was conferred, which allows reconstructing the educational levels.
The age is reconstructed using birth year.
Table 2: Descriptive statistics
Straight Gay Mean difference
Mean SD Min Max Mean SD Min Max Estimate t-stat
Log of wage 8.83 1.76 2.08 16.59 8.57 1.70 4.29 15.42 -0.268 (-1.32)
Age 33.42 12.16 4.00 67.00 33.56 10.15 12.00 55.00 0.134 (0.10)
Schooling 15.71 3.82 0.00 23.00 15.17 3.89 7.00 23.00 -0.545 (-0.95)
N 23,792 213
Notes: Differences are estimated by regressing variables on a constant and a dummy for gay status,
clustering error on the respondent level.
Source: RLMS 1994-2015 (1985 and 1990 collected retrospectively in 2000 and 2001).
The preferred dependent variable is a log of monthly contractual after-
tax wages at the primary workplace. This choice of the dependent vari-
able is standard for the dataset and is considered the best choice to
proxy wages in Russia (e.g., Kyui 2016). In Russia, the wage is calcu-
lated monthly rather than annually, and the firms pay all labor taxes.
The main idea is that monthly contractual after-tax wages are what Rus-
sians report when asked about wages. There is no habit of keeping track
of hours worked; that is why the number of hours worked contains many
missing and out-of-normal-range values and is not used in empirical work.
In particular, hours worked are entirely unavailable for 1985–1990, as ev-
eryone was on a state-provided job. Table 2provides the descriptive
statistics for the variables considered in this study.
4 METHODS 10
4 Methods
Following Badgett (1995) and many others, I estimate the sexual orien-
tation differences in wages using a human capital equation:
Yit =βGayi+X
t
X
h
γh
t(Xh
it ×λt) + λt+εit.(1)
This specification is applied to annual data covering 1985-2015. The
variable Gayiis an indicator for being gay, Yit is the log of wage, i
indexes individuals, and tindexes years. The equation also has an array
of controls, Xh
it. It includes standard human capital measures, age, age
square, and the number of years of education. It is common not to
extend the wage equation by including, for example, the industry of
occupation or occupational level because the wage discrimination could
be underestimated if some of the controls already reflect the impact of
discrimination (Blau and Kahn 2017). The parameter λtis year fixed
effects. These effects partial out the effect of inflation and tax reforms on
wages. The variance of the control variables is partialled out with time-
variant parameters, allowing for a change in the variables’ composition
from year to year. The parameter βis the focal point of interest. Because
Gayiis a dummy variable, and not gay is the omitted category, βis the
controlled percent difference in mean outcomes between gay and straight
men.
To estimate the impact of the gay-propaganda bill signed into law in
2013 on the gay wage premium, I implement the difference-in-difference
(DD) estimator with the following specification:
Yit =β(Gayi×Aftert) + X
t
X
h
γh
t(Xh
it ×λt) + λt+Gayi+εit.(2)
The new variable Aftertis an indicator function for observation after
2013. Under the common trend assumption, the parameter βhere deliv-
ers a causal effect of the bill on the gay wage premium.
Equation (1) implies that the premium is time-invariant. To relax
this assumption, I then estimate the following equation:
Yit =X
t
βt(Gayi×λt) + X
t
X
h
γh
t(Xh
it ×λt) + λt+εit.(3)
Here the vector of time-dependent parameters βtshows the annual trends
in gay wage premium. Because so few gay men are observed each year,
this specification delivers noisy parameters. The primary purpose of this
model is to show that no penalty can be found even for the years where
the number of observed gay men is maximal (i.e., 2001 – 2003).
For all three models, to account for the error term’s autocorrelation
5 RESULTS 11
over different years for the same individual, I cluster standard errors by
individuals and use a robust covariance matrix.
5 Results
Figure 2: Wage trends: 1985-2015
Notes: The figure plots the log of monthly contractual wage
after tax by year and sexual orientation. The vertical dashed
line demarcates the introduction of the gay-propaganda bill.
Source: RLMS 1994-2015 (1985 and 1990 collected retro-
spectively in 2000 and 2001).
Figure 2presents the graphic evidence on the gay wage difference
in Russia. The spike in wages in the 1990s is the hyperinflation and a
subsequent re-denomination of currency in 1997 (all prices were divided
by 1,000). The wages of both gay and straight men evolve in a remarkably
similar manner. The gay male wages are more fragile, reflecting a smaller
sample size. Notably, the common trend assumption holds near the gay-
propaganda law of 2013. This assumption is required for the model (2).
Table 3reports estimates using the three models presented in the
section above. Column (1) reports the sexual orientation difference es-
timated with a single parameter leveraging information from all data.
Column (2) reports the effect of the gay-propaganda bill of 2013 on gay
wages. Both columns report statistically insignificant results. The table
also reports the 95% confidence intervals (referred to in the table as 95%
CI).
The interval for the wage difference in Column (1) is shifted towards
positive values. One-sided test Ho:β≤0 produces a p-value of 0.21,
whereas Ho:β≥0 produces a p-value of 0.787. It is noticeably harder to
reject a positive value. This indirectly suggests a gay wage premium, not
a penalty. A similar conclusion can be made regarding the bill’s effect
on gay wages reported in Column (2).
5 RESULTS 12
Table 3: The estimated sexual orientation difference
Dependent variable (DV): log of monthly wages
(1) (2) (3)
Estimate SE Estimate SE Estimate SE
95% CI 95% CI 95% CI
Gay 0.0928 (0.116) 0.0809 (0.116)
[-0.136; 0.321] [-0.147; 0.308]
Gay×After 0.252 (0.265)
[-0.267; 0.771]
Gay×1985 0.0686 (0.193)
[-0.310; 0.448]
Gay×1990 -0.381 (0.262)
[-0.896; 0.133]
Gay×1994 -0.384** (0.144)
[-0.667; -0.101]
Gay×1995 0.122* (0.0542)
[0.0156; 0.228]
Gay×1996 1.788*** (0.127)
[1.538; 2.038]
Gay×1998 -0.777 (0.655)
[-2.061; 0.507]
Gay×2000 0.0615 (0.561)
[-1.039; 1.162]
Gay×2001 0.185 (0.280)
[-0.365; 0.734]
Gay×2002 0.307+ (0.164)
[-0.0143; 0.629]
Gay×2003 -0.0927 (0.190)
[-0.466; 0.280]
Gay×2004 0.00636 (0.228)
[-0.440; 0.453]
Gay×2005 0.418+ (0.252)
[-0.0769; 0.912]
Gay×2006 0.275 (0.185)
[-0.0864; 0.637]
Gay×2007 0.240 (0.262)
[-0.274; 0.754]
Gay×2008 0.126 (0.199)
[-0.263; 0.516]
Gay×2009 0.187 (0.177)
[-0.159; 0.534]
Gay×2010 -0.210 (0.224)
[-0.649; 0.229]
Gay×2011 0.413 (0.254)
[-0.0861; 0.911]
Gay×2012 0.333 (0.248)
[-0.153; 0.820]
Gay×2013 0.297* (0.149)
[0.00543; 0.588]
Gay×2014 0.235 (0.241)
[-0.237; 0.706]
Gay×2015 0.497 (0.360)
[-0.209; 1.203]
Controls Yes Yes Yes
Time effects Yes Yes Yes
Observations 24,005 24,005 24,005
Clusters 3,122 3,122 3,122
Adjusted R20.782 0.782 0.782
Mean of DV 8.831 8.831 8.831
Min of DV 2.079 2.079 2.079
Max of DV 16.59 16.59 16.59
Notes: The table shows various estimated wage differences between gay and straight
men.
+p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001.
Source: RLMS 1994-2015 (1985 and 1990 collected retrospectively in 2000 and 2001).
6 CONCLUSION 13
Results reported in Column (3) breakdown the wage differences by
the survey year. A wage penalty is detected for the year 1994. For
the remaining years, the difference is either imprecise or in favor of gay
men. The results for 2001 – 2003 are particularly important as this is
when the sample size of gay men is at its maximum. The model detects
gay wage premium. Remarkably, if the year 1994 is dropped from the
sample, the model (1) detects a premium with 76% confidence. The wage
equation fails to detect discrimination in a setting where discrimination
is obviously present.
To align these results with the literature reviewed in Section 2, I also
confirm endogeneity issues applying the Durbin-Wu-Hausman test. For
all models test shows endogeneity with p-value of 0.000.
6 Conclusion
A large body of empirical literature seeks to answer whether gay men’s
discrimination is present in the labor market. Instead of applying statisti-
cal methods to answer whether discrimination existed, this paper inverts
the research question and evaluates the methods using the data where
discrimination is known to present. This is also the first paper that esti-
mates the gay male wage penalty using a nationally representative sample
from a country with a particularly hostile environment towards gay men
(in Soviet and present-day Russia). This unique opportunity allows to
‘stress test’ the statistical methods that are commonly used to measure
discrimination. The estimates from the human capital equation – widely
utilized in this context statistical method – do not detect discrimination.
Instead, a gay wage premium is present with 79% confidence.
My results provide a reminder that the OLS estimates of wage equa-
tion is not an internally valid modeling choice for estimating sexual orien-
tation difference in pay. Nor are the DD estimates of the policy changes
that affect gays. The DD design is not robust to the survivorship bias.
One evident consequence of only better-educated gays revealing their
sexual orientation is that the effects on the most vulnerable gays are
ignored. For example, consider a fundamental question whether gay-
affirmative policies lead to backlash (more hatred after change) (e.g.,
Valencia, Williams, and Pettis 2022). The DD design applied to data
where gays most likely to experience hate are misclassified as straight
underestimates the backlash effects and misinforms the policymakers.
Ultimately, even though the literature on sexual orientation differences
is extraordinarily extensive, it is likely to be of little use, as the endo-
geneity issues are ignored. Reliable sources of exogenous variation are
needed to produce time and location comparable estimates.
The methods based on the quasi-random assignment of interviewers
REFERENCES 14
seem most promising (Alexeev 2022). The fraternal birth order effect
is another potential source of a valid instrument. Whereas the number
of older brothers is unlikely to be random, a fraction of older brothers
in sibship is plausibly random and can be used as an instrument. The
comparable sibling indexes have a long tradition in applied econometrics
(e.g., Booth and Kee 2009). For example, Ablaza, Kab´atek, and Perales
(2022) use the Dutch census to confirm that the number of older brothers
is a reliable predictor of sexual orientation in males. Theirs, or any other
population size dataset, can provide sufficient variation in the sibling
index to accurately estimate sexual orientation differences.
The empirical finding of the current paper is not directly informative
in relation to the ongoing theoretical debate between rejection sensitivity
and MSTs (e.g., Maiolatesi, Clark, and Pachankis 2022). My reasoning
hinges on the assumption that MST is the relevant theory for an op-
pressive environment. Thus, according to the conventions of the gay pay
literature, the penalty must be detected, but in the Russian data, the
penalty is absent. I then claim that the lack of penalty results from the
models’ endogeneity. To support this claim and its validity across the
settings, I review the relevant literature.
An alternative way to interpret the lack of penalty, entirely consistent
with existing literature, is to reject MST in favor of the alternatives. The
only well-established option is the rejection sensitivity model. That is,
not the gay status, but some other factor such as poor emotional control
is the causal agent for the difference. This line of reasoning is implausible.
The mechanics of the rejection sensitivity model do not have first-order
significance in explaining orientation differences in Russia.
My work is centered around the methods and quality of empirical
evidence on sexual orientation differences; thus, the estimates presented
do not have direct policy implications, apart from a reminder that the
wage equation will fail in a truly discriminatory environment. This pa-
per should not be taken as a minimization of the hardships that sexual
minorities are facing. Discrimination is damaging not only to its victims
but also to society, and policy must counter it. There is no need to look
for theoretical or empirical support for a non-discriminatory public pol-
icy because it is the only just and lawful option. Russian anti-gay laws
are appalling and require immediate revision. Sadly, this is not the only
deplorable and inhuman public choice we witness today in Russia.
References
Ablaza, C., J. Kab´atek, and F. Perales (2022). “Are Sibship Character-
istics Predictive of Same Sex Marriage? An Examination of Fraternal
Birth Order and Female Fecundity Effects in Population-level Admin-
REFERENCES 15
istrative Data from the Netherlands”. In: The Journal of Sex Research
0.0. PMID: 35040387, pp. 1–13. doi:10.1080/00224499.2021.1974330.
eprint: https://doi.org/10.1080/00224499.2021.1974330.url:https:
//doi.org/10.1080/00224499.2021.1974330 (cit. on pp. 5,14).
ACSF investigators (Dec. 1992). “AIDS and sexual behaviour in France”.
In: Nature 360.6403, pp. 407–409. issn: 1476-4687. doi:10 . 1038 /
360407a0.url:https://doi.org/10.1038/360407a0 (cit. on p. 4).
Alexeev, S. (2022). “Estimating sexual orientation differences using ran-
domly assigned interviewers”. In: ResearchGate working paper (cit.
on pp. 2,14).
Ashenfelter, O. and A. Rees (2015). Discrimination in Labor Markets.
Princeton Legacy Library. Princeton University Press. isbn: 9781400867066.
url:https://books.google.com.au/books?id=vEt9BgAAQBAJ (cit.
on p. 2).
Badgett, M. V. L. (1995). “The Wage Effects of Sexual Orientation Dis-
crimination”. In: Industrial and Labor Relations Review 48.4, pp. 726–
739. issn: 00197939-2162271X. url:http://www.jstor.org/stable/
2524353 (cit. on pp. 2,10).
Badgett, M. V. L., C. S. Carpenter, and D. Sansone (May 2021). “LGBTQ
Economics”. In: Journal of Economic Perspectives 35.2, pp. 141–70.
doi:10.1257/jep.35.2.141.url:https://www.aeaweb.org/articles?
id=10.1257/jep.35.2.141 (cit. on p. 5).
Bailey, M. (Dec. 2019). “The Minority Stress Model Deserves Reconsid-
eration, Not Just Extension”. In: Archives of Sexual Behavior.issn:
1573-2800. doi:10.1007/s10508-019-01606-9.url:https://doi.org/
10.1007/s10508-019-01606-9 (cit. on p. 4).
Becker, G. (1957). The Economics of Discrimination. Economic Research
Studies. University of Chicago Press. isbn: 9780226041049. url:https:
//books.google.com.au/books?id=50qHcSNVVEMC (cit. on p. 2).
Billy, J. O., K. Tanfer, W. R. Grady, and D. H. Klepinger (1993). “The
sexual behavior of men in the United States”. In: Family planning
perspectives, pp. 52–60 (cit. on p. 4).
Black, D., G. Gates, S. Sanders, and L. Taylor (May 2000). “Demo-
graphics of the gay and lesbian population in the United States: Evi-
dence from available systematic data sources”. In: Demography 37.2,
pp. 139–154. issn: 1533-7790. doi:10 . 2307 / 2648117.url:https :
//doi.org/10.2307/2648117 (cit. on p. 6).
Black, D., S. G. Sanders, and L. J. Taylor (June 2007). “The Economics
of Lesbian and Gay Families”. In: Journal of Economic Perspectives
21.2, pp. 53–70. doi:10.1257/jep.21.2.53.url:https://www.aeaweb.
org/articles?id=10.1257/jep.21.2.53 (cit. on pp. 5,6,9).
Black, S. E., P. J. Devereux, and K. G. Salvanes (May 2005). “The More
the Merrier? The Effect of Family Size and Birth Order on Children’s
REFERENCES 16
Education”. In: The Quarterly Journal of Economics 120.2, pp. 669–
700. issn: 0033-5533. doi:10 . 1093 / qje / 120 . 2 . 669. eprint: https :
//academic.oup. com/qje/article-pdf/120 /2/669/5449009/ 120-2-
669.pdf.url:https://doi.org/10.1093/qje/120.2.669 (cit. on p. 5).
Blanchard, R. (Jan. 2018). “Fraternal Birth Order, Family Size, and
Male Homosexuality: Meta-Analysis of Studies Spanning 25 Years”.
In: Archives of Sexual Behavior 47.1, pp. 1–15. issn: 1573-2800. doi:
10.1007/s10508-017-1007-4.url:https://doi.org/10.1007/s10508-
017-1007-4 (cit. on p. 5).
Blanchard, R. and R. A. Lippa (Oct. 2020). “Reassessing the Effect of
Older Sisters on Sexual Orientation in Men”. In: Archives of Sexual
Behavior.issn: 1573-2800. doi:10.1007/s10508-020-01840-6.url:
https://doi.org/10.1007/s10508-020-01840-6 (cit. on p. 5).
Blau, F. D. and L. M. Kahn (2017). “Women’s Work and Wages”. In: The
New Palgrave Dictionary of Economics. London: Palgrave Macmillan
UK, pp. 1–14. isbn: 978-1-349-95121-5. doi:10 .1057/ 978- 1 - 349 -
95121-5 2207-1.url:https://doi .org/10 .1057/978 - 1- 349-95121-
5 2207-1 (cit. on p. 10).
Booth, A. L. and H. J. Kee (Apr. 2009). “Birth order matters: the effect of
family size and birth order on educational attainment”. In: Journal
of Population Economics 22.2, pp. 367–397. issn: 1432-1475. doi:
10.1007/s00148-007-0181-4.url:https://doi.org/10.1007/s00148-
007-0181-4 (cit. on pp. 5,14).
Bourne, A., D. Reid, F. Hickson, S. Torres-Rueda, and P. Weatherburn
(2015). “Illicit drug use in sexual settings (‘chemsex’) and HIV/STI
transmission risk behaviour among gay men in South London: findings
from a qualitative study”. In: Sexually Transmitted Infections 91.8,
pp. 564–568. issn: 1368-4973. doi:10 .1136 / sextrans- 2015 - 052052.
eprint: https://sti.bmj.com/content/91/8/564.full.pdf.url:https:
//sti.bmj.com/content/91/8/564 (cit. on p. 4).
Crossley, M. L. (2004). “Making sense of ‘barebacking’: Gay men’s narra-
tives, unsafe sex and the ‘resistance habitus’”. In: British Journal of
Social Psychology 43.2, pp. 225–244. doi:https://doi.org/10.1348/
0144666041501679. eprint: https: / / bpspsychub.onlinelibrary.wiley.
com/doi/pdf/10.1348/0144666041501679.url:https://bpspsychub.
onlinelibrary.wiley.com/doi/abs/10.1348/0144666041501679 (cit. on
p. 2).
David, S. (2007). “Sexual Strategies Across Sexual Orientations”. In:
Journal of Psychology & Human Sexuality 18.2-3, pp. 183–214. doi:
10.1300/J056v18n02\06. eprint: https://doi.org/10.1300/J056v18n02
06.url:https://doi.org/10.1300/J056v18n02 06 (cit. on p. 4).
Drydakis, N. (2011). “Women’s Sexual Orientation and Labor Market
Outcomes in Greece”. In: Feminist Economics 17.1, pp. 89–117. doi:
REFERENCES 17
10.1080/13545701 . 2010.541858. eprint: https: //doi.org/ 10 . 1080/
13545701.2010.541858.url:https://doi.org/10.1080/13545701.2010.
541858 (cit. on p. 9).
Drydakis, N. (2019). “Sexual orientation and labor market outcomes”.
In: IZA World of Labor.url:doi:%2010.15185/izawol.111.v2 (cit. on
p. 2).
— (2021). “Sexual Orientation and Earnings. A Meta-Analysis 2012-
2020”. In: Journal of Population Economics.url:https : / / www .
econstor.eu/bitstream/10419/235027/1/GLO-DP-0862.pdf (cit. on
p. 2).
Feinstein, B. A. (Oct. 2020). “The Rejection Sensitivity Model as a
Framework for Understanding Sexual Minority Mental Health”. In:
Archives of Sexual Behavior 49.7, pp. 2247–2258. issn: 1573-2800.
doi:10 .1007 / s10508- 019 - 1428 - 3.url:https://doi. org/10.1007/
s10508-019-1428-3 (cit. on p. 4).
Friedman, R. and J. Downey (2008). Sexual Orientation and Psychody-
namic Psychotherapy: Sexual Science and Clinical Practice. Columbia
University Press. isbn: 9780231504898. url:https://books.google.
com.au/books?id=vRw2AAAAQBAJ (cit. on p. 7).
Greene, W. (2018). Econometric Analysis. Pearson. isbn: 9780134461366.
url:https://books.google.com.au/books?id=Gqh2AQAACAAJ (cit.
on p. 6).
Horne, S. G. and L. White (2020). “The return of repression: Mental
health concerns of lesbian, gay, bisexual, and transgender people in
Russia.” In: Perspectives on sexual orientation and diversity. Wash-
ington, DC, US: American Psychological Association, pp. 75–88. isbn:
9781433830914. doi:10.1037/0000159-006.url:https://doi.org/10.
1037/0000159-006 (cit. on p. 3).
Hylton, E. et al. (June 2017). “Sexual Identity, Stigma, and Depression:
the Role of the “Anti-gay Propaganda Law” in Mental Health among
Men Who Have Sex with Men in Moscow, Russia”. In: Journal of Ur-
ban Health 94.3, pp. 319–329. issn: 1468-2869. doi:10.1007/s11524-
017-0133-6.url:https://doi.org/10.1007/s11524-017-0133-6 (cit. on
p. 3).
Katz-Wise, S. L. (2015). “Sexual fluidity in young adult women and
men: associations with sexual orientation and sexual identity devel-
opment”. In: Psychology & Sexuality 6.2, pp. 189–208. doi:10.1080/
19419899.2013 .876445. eprint: https:/ /doi.org/10.1080/19419899.
2013.876445.url:https: //doi.org/10.1080/19419899.2013.876445
(cit. on p. 6).
Kelly, ´
E. (Sept. 14, 2017). “European University in Russia faces clo-
sure”. In: Science—Business.url:https://sciencebusiness.net/news/
REFERENCES 18
european-university-russia-faces-closure (visited on 10/14/2021) (cit.
on p. 3).
K¨ollen, T. (2016). Sexual Orientation and Transgender Issues in Organi-
zations: Global Perspectives on LGBT Workforce Diversity. Springer
International Publishing. isbn: 9783319296234. url:https://books.
google.com.au/books?id=60kWDAAAQBAJ (cit. on p. 2).
Kon, I. and J. Riordan (1993). Sex and Russian Society. Indiana Univer-
sity Press. isbn: 9780253332011. url:https://books.google.com.au/
books?id=DgGjOjRQAPIC (cit. on p. 9).
Kondakov, A. (2019). “The influence of the ‘gay-propaganda’ law on
violence against LGBTIQ people in Russia: Evidence from criminal
court rulings”. In: European Journal of Criminology.doi:10.1177/
1477370819887511. eprint: https://doi.org/10.1177/1477370819887511.
url:https://doi.org/10.1177/1477370819887511 (cit. on p. 3).
Kozyreva, P., M. Kosolapov, and B. M. Popkin (2016). “Data Resource
Profile. RLMS Phase 2: Monitoring the Economic and Health Situa-
tion in Russia 1994 – 2013”. In: International Journal of Epidemiology
45.2, pp. 395–401. url:https://academic.oup.com/ije/article/45/2/
395/2572590 (cit. on p. 8).
Kr¨oger, H. and J. Hartmann (2021). “Extending the Kitagawa-Oaxaca-
Blinder decomposition approach to panel data”. In: The Stata Journal
21.2, pp. 360–410. doi:10.1177/1536867X211025800. eprint: https:
//doi.org/ 10.1177/1536867X211025800.url:https: //doi.org/10 .
1177/1536867X211025800 (cit. on p. 2).
Kyui, N. (2016). “Expansion of Higher Education, Employment and
Wages: Evidence from the Russian Transition”. In: Labour Economics
39, pp. 68–87. issn: 0927-5371. doi:https://doi.org/10.1016/j.labeco.
2016.01.001.url:http://www.sciencedirect.com/science/article/pii/
S0927537116000026 (cit. on p. 9).
Laurent, T. and F. Mihoubi (Dec. 2012). “Sexual Orientation and Wage
Discrimination in France: The Hidden Side of the Rainbow”. In: Jour-
nal of Labor Research 33.4, pp. 487–527. issn: 1936-4768. doi:10.
1007/s12122-012-9145-x.url:https://doi.org/10.1007/s12122-012-
9145-x (cit. on pp. 7,9).
Maiolatesi, A. J., K. A. Clark, and J. E. Pachankis (2022). “Rejec-
tion sensitivity across sex, sexual orientation, and age: Measurement
invariance and latent mean differences.” In: Psychological Assess-
ment, No Pagination Specified–No Pagination Specified. doi:10 .
1037/pas0001109.url:https: / /doi.org/10. 1037 / pas0001109 (cit.
on p. 14).
Meyer, I. H. (Sept. 2003). “Prejudice, social stress, and mental health in
lesbian, gay, and bisexual populations: conceptual issues and research
evidence”. eng. In: Psychological bulletin 129.5. 2003-99991-002[PII],
REFERENCES 19
pp. 674–697. issn: 0033-2909. doi:10 .1037 / 0033- 2909.129 . 5.674.
url:https://doi.org/10.1037/0033-2909.129.5.674 (cit. on pp. 2,5).
Peralta, R. L. (2008). ““Alcohol Allows You to Not Be Yourself”: Toward
a Structured Understanding of Alcohol Use and Gender Difference
among Gay, Lesbian, and Heterosexual Youth”. In: Journal of Drug
Issues 38.2, pp. 373–399. doi:10.1177/002204260803800201. eprint:
https://doi.org/10.1177/002204260803800201.url:https://doi.org/
10.1177/002204260803800201 (cit. on p. 2).
Prestage, G., J. Grierson, J. Bradley, M. Hurley, and J. Hudson (2009).
“The role of drugs during group sex among gay men in Australia”. In:
Sexual Health 6.4, pp. 310–317. doi:10.1071/SH09014.url:https:
//doi.org/10.1071/SH09014 (cit. on p. 2).
Sabia, J. J., M. Wooden, and T. T. Nguyen (2017). “Sexual Identity,
Same-Sex Relationships, and Labour Market Dynamics: New Evi-
dence from Longitudinal Data in Australia”. In: Southern Economic
Journal 83.4, pp. 903–931. doi:https://doi.org/10.1002/soej.12181.
eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/soej.12181.
url:https://onlinelibrary. wiley.com/ doi/abs /10.1002/soej.12181
(cit. on p. 7).
Sabirianova Peter, K. (2003). “Skill-biased Transition: The Role of Mar-
kets, Institutions, and Technological Change”. In: IZA Discussion
Paper No. 893; William Davidson Institute Working Paper No. 616.
url:https://ssrn.com/abstract=459582 (cit. on p. 9).
Salomaa, A. C. and J. L. Matsick (2019). “Carving sexuality at its joints:
Defining sexual orientation in research and clinical practice.” In: Psy-
chological assessment 31.2, p. 167. url:https://pubmed.ncbi.nlm.
nih.gov/30211575/ (cit. on p. 7).
Sansone, D. (2019). “Pink work: Same-sex marriage, employment and
discrimination”. In: Journal of Public Economics 180, p. 104086.
issn: 0047-2727. doi:https : / / doi . org / 10 . 1016 / j . jpubeco . 2019 .
104086.url:https:/ / www.sciencedirect.com/ science /article/pii/
S0047272719301471 (cit. on p. 7).
Saxby, K., S. C. de New, and D. Petrie (2020). “Structural stigma and
sexual orientation disparities in healthcare use: Evidence from Aus-
tralian Census-linked-administrative data”. In: Social Science and
Medicine 255, p. 113027. issn: 0277-9536. doi:https : / / doi . org /
10.1016/j.socscimed.2020.113027.url:https://www.sciencedirect.
com/science/article/pii/S027795362030246X (cit. on p. 7).
Stuikyte, R., I. Barbosa, and M. Kazatchkine (2019). “Getting to grips
with the HIV epidemic in Russia”. In: Current Opinion in HIV and
AIDS 14.5. issn: 1746-630X. url:https://journals.lww.com/co-
hivandaids / Fulltext / 2019 / 09000 / Getting to grips with the HIV
epidemic in Russia.7.aspx (cit. on p. 8).
REFERENCES 20
Urwin, S., T. Mason, and W. Whittaker (2021). “Do different means of
recording sexual orientation affect its relationship with health and
wellbeing?” In: Health Economics n/a.n/a. doi:https://doi.org/10.
1002/hec.4422. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.
1002/hec .4422.url:https: //onlinelibrary.wiley. com/doi/abs/10.
1002/hec.4422 (cit. on p. 2).
Valencia, Z., B. Williams, and R. Pettis (2022). “Pride and prejudice:
Same-Sex marriage legalization announcements and LGBT Hate-Crimes”.
In: Journal of Law and Economics.url:https://dx.doi.org/10.2139/
ssrn.3362835 (cit. on p. 13).
Waldis, L., N. Borter, and T. H. Rammsayer (2021). “The Interactions
Among Sexual Orientation, Masculine and Feminine Gender Role Ori-
entation, and Facets of Sociosexuality in Young Heterosexual and Ho-
mosexual Men”. In: Journal of Homosexuality 68.12. PMID: 31990236,
pp. 2003–2023. doi:10.1080/00918369.2020.1717837. eprint: https:
//doi.org/10.1080/00918369.2020.1717837.url:https://doi.org/10.
1080/00918369.2020.1717837 (cit. on p. 4).
