When evaluating the regression equations, it is necessary to take into account that women's participation in the labor force is endogenous, since the sample excludes women who did not want to work for the wages offered in the labor market, but chose not to work at all. Such assumptions lead to a bias in the estimates of the coefficients obtained by least squares method. Offset correction is performed using the Heckman procedure (Heckman, 1979). However, our Heckman procedure showed that the probability of being included in the sample, for both women and men, is not related to the level of wages. Therefore, the simulation results below were obtained without correcting for sample bias. The lack of influence of wages on selection for employment in the countries of the former USSR is also noted in the works (Oshchepkov 2007; Khitarishvili 2019). The obtained results indicate a different nature of the selection for work in countries with economies in transition compared with developed economies.
The first additive component which is on the right represents the explained share of the gender wage gap attributable to observed worker characteristics. The second and third terms are the unexplained share of the gender pay gap, which characterizes the difference in returns to characteristics. Newmark's modification makes it possible to solve the problem of establishing the wage structure at the same productivity, by determining it by coefficients from the regression equation common for men and women, and also to obtain more accurate estimates of the unexplained part of the decomposition.
Another problem is sensitivity of decomposition methods to the choice of base for categorical variables. To solve this problem, we normalized categorical variables using Yun's method (Yun, 2005), according to which the “true” contribution of the variables to the wage difference is the mean of the estimates of the coefficients of equations with different base categories.
A limitation of the Oaxaca-Blinder method is its application only for the decomposition of average wages. Decomposition methods based on quantile regressions are used to decompose the differences in wages between men and women in other parts of the distribution. In accordance with the method (Chernozhukov et al. , 2008), a linear probabilistic model is constructed for group A.
Based on the obtained estimates of the parameters β ̂_A, the predicted values of the probabilities Λ(X_i β ̂_A (y)) are calculated, где Λ(·) - is a linear relationship function and averaging is performed over the predicted probabilities to obtain a counterfactual distribution F ̂_(Y_A^C ) (y):
F ̂_(Y_A^C ) (y)=1/n_B ∑_(i∈B)▒〖Λ(X_i β ̂_A (y))〗. (3)
After obtaining the counterfactual distribution, counterfactual quantiles are found by inverting the estimated distribution function:
Q_(A, θ)^C=F ̂_(Y_A^C)^(-1) (θ). (4)
Using the decomposition method described above, it is possible to perform an aggregate decomposition of pay differentials.
The results of the decomposition based on quantile regressions indicate significant differences in the estimates of the gender gap at different levels of the wage distribution. As we move from low-paying jobs to high-paying jobs, we see an increase in both the overall pay gap and the effects of the structure and composition of wages. In contrast to the results of the study (Newell, Reilly 2001), where the authors recorded a difference in the size of the gap between the 1st and 9th percentiles at 0. 33 log points for Kazakhstan, we found a decrease in this difference to 0. 13 log points. This indicates a decrease in the differentiation of the gender wage gap between quantile groups of workers. However, the problem of the largest gap for high-paid men and women remains at 33% versus 18% for low-paid workers. The highest value of the unexplained part of the gender gap at the 90% quantile indicates differences in remuneration of men and women, other things being equal, and may indicate the presence of a "glass ceiling" in the Kazakhstani labor market. The different signs for the explained gender wage gap in different parts of the distribution indicate different contributions of the variables to the size of the gap. Consequently, a more detailed analysis of the causes of the gender gap for groups of workers with different income levels is required, which indicates further prospects for studying gender inequality in labor market of the Republic of Kazakhstan.
However, analysis of the gender pay gap at different distribution quantiles showed significant differences between low-paid and high-paid workers. The wage gap between men and women is widening as the transition from low-paid to high-paid workers continues to grow. The largest return on decile 9 of the distribution of wages indicates the “glass ceiling” problem in the Kazakhstani labor market.
When evaluating the
regression
equations, it is necessary to take into account that women's participation in the
labor
force is endogenous, since the sample excludes
women
who did not want to
work
for the wages offered in the
labor
market
,
but
chose not to
work
at all. Such assumptions lead to a bias in the
estimates
of the coefficients
obtained
by least squares
method
. Offset correction
is performed
using the
Heckman
procedure (
Heckman
, 1979).
However
, our
Heckman
procedure
showed
that the probability of
being included
in the sample, for both
women
and
men
, is not related to the level of wages.
Therefore
, the simulation
results
below were
obtained
without correcting for sample bias. The lack of influence of wages on selection for employment in the countries of the former USSR is
also
noted in the works (
Oshchepkov
2007;
Khitarishvili
2019). The
obtained
results
indicate
a
different
nature of the selection for
work
in countries with economies in transition compared with developed economies.
The
first
additive component which is on the right represents the
explained
share of the gender wage
gap
attributable to observed
worker
characteristics.
The
second and third terms are the unexplained share of the gender
pay
gap
, which characterizes the
difference
in returns to characteristics.
Newmark
's modification
makes
it possible to solve the
problem
of establishing the wage structure at the same productivity, by determining it by coefficients from the
regression
equation common for
men
and
women
, and
also
to obtain more accurate
estimates
of the unexplained
part
of the decomposition.
Another
problem
is sensitivity of
decomposition
methods
to the choice of base for categorical variables. To solve this
problem
, we normalized categorical variables using
Yun
's
method
(
Yun
, 2005), according to which the “true” contribution of the variables to the wage
difference
is the mean of the
estimates
of the coefficients of equations with
different
base categories.
A limitation of the Oaxaca-Blinder
method
is its application
only
for the
decomposition
of average wages.
Decomposition
methods
based on quantile
regressions
are
used
to decompose the
differences
in wages between
men
and
women
in other
parts
of the
distribution
. In accordance with the
method
(
Chernozhukov
et al.
,
2008), a linear probabilistic model
is constructed
for group A.
Based on the
obtained
estimates
of the parameters β ̂_A, the predicted values of the probabilities Λ(
X_i
β ̂_A (y))
are calculated
,
где
Λ(·)
-
is a linear relationship function and averaging
is performed
over the predicted probabilities to obtain a counterfactual
distribution
F ̂_(Y_
A^C
)
(y):
F ̂_(Y_
A^C
)
(y)=1/n_B ∑_(
i∈B
)
▒〖Λ
(
X_i
β ̂_A (y))〗. (3)
After obtaining the counterfactual
distribution
, counterfactual quantiles
are found
by inverting the estimated
distribution
function:
Q_(A, θ)
^C
=F ̂_(Y_
A^C
)^(-1) (θ). (4)
Using the
decomposition
method
described
above, it is possible to perform an aggregate
decomposition
of
pay
differentials.
The
results
of the
decomposition
based on quantile
regressions
indicate
significant
differences
in the
estimates
of the gender
gap
at
different
levels of the wage
distribution
. As we
move
from low-paying jobs to high-paying jobs, we
see
an increase in both the
overall
pay
gap
and the effects of the structure and composition of wages.
In contrast
to the
results
of the study (Newell,
Reilly
2001), where the authors recorded a
difference
in the size of the
gap
between the 1st and 9th percentiles at 0. 33 log points for Kazakhstan, we found a decrease in this
difference
to 0. 13 log points. This
indicates
a decrease in the differentiation of the gender wage
gap
between quantile groups of
workers
.
However
, the
problem
of the largest
gap
for high-paid
men
and
women
remains at 33% versus 18% for low-paid
workers
. The highest value of the unexplained
part
of the gender
gap
at the 90% quantile
indicates
differences
in remuneration of
men
and
women
, other things being equal, and may
indicate
the presence of a
"
glass ceiling
"
in the
Kazakhstani
labor
market
. The
different
signs for the
explained
gender wage
gap
in
different
parts
of the
distribution
indicate
different
contributions of the variables to the size of the
gap
.
Consequently
, a more detailed analysis of the causes of the gender
gap
for groups of
workers
with
different
income levels
is required
, which
indicates
further
prospects for studying gender inequality in
labor
market
of the Republic of Kazakhstan.
However
, analysis of the gender
pay
gap
at
different
distribution
quantiles
showed
significant
differences
between low-paid and high-paid
workers
. The wage
gap
between
men
and
women
is widening as the transition from low-paid to high-paid
workers
continues to grow. The largest return on
decile
9 of the
distribution
of wages
indicates
the “glass ceiling”
problem
in the
Kazakhstani
labor
market
.