Higher School of Economics, Moscow October 14, 2008

Oaxaca-Blinder (1973):

Difference in characteristics + Difference in coefficientsAssumptions: Index number problem: we first use males wages as a non-discriminatory benchmark (in line with other studies on Russia), and then use a pooled model (Neumark, 1988, Oaxaca and Ransom, 1994)

It is based on the OLS property that the mean wage conditional on average characteristics is equal to the unconditional mean wage that does not hold in the context of quantile regression:

Machado-Mata (2005)

Denote by the wage of individual i with characteristics X who leaves behind a fraction of individuals () with the same characteristics (Koenker and Basset, 1978).

Then the gender gap can be decomposed using the Machado-Mata (2005) methodology and dropping superscript i as follows:

[in words: total gap = difference in characteristics + difference in coefficients + residual]

Assumptions: we use males wages as a non-discriminatory benchmark

Juhn, Murphy, Pierce (1991)

To decompose changes in the wage gap over 1997-2002 we use Juhn, Murphy, Pierce (1991)Assume that the residual consists of two parts where is the standard deviation of the residual of the male wage equation, and is the standardized residual with a mean of zero and a standard deviation of 1.

1st term (the Observed Xs effect): the effect of changes in observable characteristics over time2nd term (the Observed Prices effect): the contribution of changes in the prices of the observed skills of men3rd term (the Gap effect): the effect of changes in the relative position of women in the male residual wage distribution, 4th term (the Unobserved Prices effect): the contribution of the widening or narrowing of the male residual distribution

Assumptions: first use males coefficients as a benchmark, and then experiment also with the pooled model

Development and Reform Research Team University of Bologna

Proportion of females in the firm


Occupational distribution (%)


Earnings by gender, 1997 and 2002:All employees Workers


Evolution of the GEG inside the firm


OB decomposition, all employees

Determinants of wages, 1997 and 2002

Determinants of wages, 1997 and 2002 (contd)


GEG at the meansAt best one third of the gap is explained by differences in productive characteristics

GEG decreased between 1997 and 2002 by approx. 20 points

GEG for the entire workforce is driven by the earnings differentials for engineers and production workers

GEG is small and for the most part insignificant for managers (in line with Lazear and Rosen, 1990) and (in some years) for service staff

Workers have by far the highest gaps, little of which is explained by differences in observed characteristics


GEG at the quantiles: raw and adjusted gaps

Determinants of wages, 1997 and 2002

Determinants of wages, 1997 and 2002 (contd)


GEG at the quantiles: MM (2005) total gap and gap due to coefficients1997 2002

Machado-Mata: All employees


GEG at the quantiles: MM (2005)In general, GEG has roughly an inverted U-shape profile across wage distribution, apart from 2002

There is evidence for an increase of a glass ceiling effect by 2002

The highest quantile in 1997 and the lowest in 2002 exhibit particularly low gender differentials

The main portion of the GEG is due to the differences in coefficients


Potential explanations of the change in GEG: 1997-2002


Change in GEG at the mean: JMP (1991)

Juhn, Murphy, Pierce (1991)

To decompose changes in the wage gap over 1997-2002 we use Juhn, Murphy, Pierce (1991)Assume that the residual consists of two parts where is the standard deviation of the residual of the male wage equation, and is the standardized residual with a mean of zero and a standard deviation of 1.

1st term (the Observed Xs effect): the effect of changes in observable characteristics over time2nd term (the Observed Prices effect): the contribution of changes in the prices of the observed skills of men3rd term (the Gap effect): the effect of changes in the relative position of women in the male residual wage distribution, 4th term (the Unobserved Prices effect): the contribution of the widening or narrowing of the male residual distribution

Assumptions: first use males coefficients as a benchmark, and then experiment also with the pooled model


Change in GEG at the mean: JMP (1991)About 29 percent of the decrease can be explained by changes in observed characteristics and prices

Changes in observed characteristics about four times as important as changes in observed prices

About 6 points of the reduction of the gap is because women improve their position in the male residual earnings distribution

About 8 points are due to a narrowing of this distribution

The joint contribution of gender-specific effect has the most weight (contrary to the early years of transition, see Brainerd, 2000)


Change in GEG at the quantiles: MM (2005)

Raw gap fell more at the bottom than at the top. Is that due to changes in Xs or changes in s?___________________________________________________1 The actual gap is the coefficient on the male dummy in the quantile regressions without covariates.




Change in GEG at the quantiles: WomenIf the distribution of womens Xs had not changed from 1997, the gap would have decreased at the bottom, but would have stayed almost the same throughout the rest of the distribution (row 6). Thus, womens characteristics were better in 1997 at the bottom, but not in the rest of the distribution. That does not help to explain the larger fall at the bottom.

If women in 2002 had the returns to their characteristics as in 1997, the gap would have been even negative at the top (benefiting women over men) and would have risen a lot at the bottom. Changes in s contributed to the large reduction in the gap at the bottom and an increase at the top. Thus, a large increase in the prices of womens characteristics at the bottom (i.e. decrease in discrimination) is an explanation of the larger fall of the GEG at the bottom.




Change in GEG at the quantiles: MenIf men in 2002 had characteristics of 1997, the gap would have been slightly larger at the bottom 10th percentile and almost the same in the rest of the distribution.Thus, at the very bottom mens Xs were slightly better in 1997 than in 2002, and worsening in mens Xs contributed to the fall in the gap there (however, to a small extent). The best from the bottom have moved away.

If men in 2002 had 1997 s, the gap would have been larger everywhere. Mens s in 1997 were better than in 2002 and decline in rewards for men contributed to reducing the gap throughout the whole distribution. The reduction in s, however, is higher at the top than at the bottom. It is increased rewards of women at the bottom + a slight worsening in mens characteristics


What have we learned so far?There exist a GEG inside a Russian firm, which is the largest for production workers and is absent for managersThe gap is largerly unexplained by productivity characteristics at the mean and at the quantilesThe gap declines from 1997 to 2002, and the glass ceiling effect emergesPotential explanations of the decline: change in prices and composition effectIt is not the less-skilled women who separate (Hunt, 2002) The quality of new hirees is slightly worse for both genders, average quality of female employees does not improve over time nor is changing composition of males at lower end driven by hirings 1/3 of the fall of GEG at the mean is explained by changes in observed characteristics and prices.

Probability to separate (Pooled): ME from Probit

Probability to separate (contd)


What have we learned so far?The decline of GEG is largely due to a decline in the lowest part of the distribution.The reasons: men with better characteristics leave the bottom of the wage distribution, which also improves relative position of women in residual male wage distribution; decreased rewards for men;mainly: the rewards to characteristics for women improve disproportionately at the bottom of the distribution.


Potential explanations of the existence of the GEG


Potential reasons behindThe GEG declines from 36% to 17% between 1997 and 2002, however is still present

Potential reasons:BonusesArrearsTrade-off between job security and wagesDiscriminationSegregation.


Potential explanations of the GEG: bonusesNO, since the decomposition and regression results for total compensation are very similar to those of the GEG.


Potential explanations of the GEG: wage arrearsExisted only in 1998 in this firm W.a.s in 1998 are very small and lowest for workers (0.05 months of 1997 wages vs. around 2 months in Russia, see e.g. Lehmann and Wadsworth, 2007);

Dohmen, Lehmann and Schaffer (2008): in this firm no gender difference in incidence of w.a.s for all employees;

Decomposition results for employees with wages paid in full are very similar.NO


Potential explanations of the GEG : trade-off between secure jobs and wagesIn the firm, the majority of separations are quits (79% among all separations)

After having controlled for productivity characteristics and occupations, females have on average 3 p.p. higher probability to quit than males

They have also 1 p.p. higher probability to be laid-off NO, but it is not a direct test of self-selection


Potential explanations of the GEG : segregationProduction workers have the highest GEG that contributes most to the overall gap

Production workers have jobs that are linked to levels - 8 for primary workers and 6 for auxiliary workers: so far for 2002 only

Controlling for such hierarchical levels is a descriptive exercise because of the endogeneity of these levels

Ransom and Oaxaca (2005): But this makes the male/female wage difference that we observe all the more startling: among these workers , although wages were set by a collective bargaining that was, ostensibly, gender neutral, a large wage differential arose because women were placed in jobs different from those assigned to similar men


Distribution of workers by wage levelsAuxilliary levels:


Distribution of workers by wage levelsPrimary levels:


Results for workers including levels at the means: Oaxaca-Blinder decomposition


Results for workers at the quantiles with and w/o levels

No levels With levels

Machado-Mata: Workers 2002


SegregationThere exist virtually no GEG within the job levels for production workersControlling for these levels leads to disappearance of the gapThese levels explain the whole wage differentialThis (descriptive) exercise points out that, in spite of a seemingly gender-neutral wage policy of the top management, large earnings differentials arises because overwhelming numbers of women are placed in low-paid job levels

However, gender difference in occupational distribution may reflect promotion discrimination or unequal occupational access. If it does, then it cannot be used to explain the GWGThe results with no levels in the regressions can be viewed as an upper bound for the extent of discrimination, and the results with levels - as a lower bound (Arulampalam et al., 2006).


Segregation or unequal job assignments ?Probit regression results show that females have 84 p.p. lower probability to be in the primary levels (even those with university education)

Fairlie (2003) decomposition shows that only 11% of this job assignment can be explained by the observed characteristics

Gender differences in promotion rates and in entry-level jobs (to be completed)


What have we foundThere exists an intra-firm GEG that declines over 1997-2002, which is driven by the GEG for production workers

Increased rewards for women at the lower end of the distribution (and outflow of men with better characteristics at the bottom) seem to be a reason behind the decline

Bonuses, wage arrears or wages-secure jobs tradeoff do not seem to be reasons behind the existence of the GEG

For production workers the gap is almost completely explained when workers levels are included into the regressions

Job levels explain about 45-59% of all the variation in wages (R2 from the respective regressions)


Conclusions and future researchComposition effect and increase in rewards for women at the bottom of the wage distribution (decrease in discrimination?) are the reasons behind the decrease in GEG

Consistent with the increasing competition that firm faces as well as with the reduction on childcare facilities in the second half of 1990s

The potential explanation of the existence of the GEG seems to be existence of segregation in the internal labor market in Russia

However, the lower job assignment of women could only to a small degree be explained by individual productivity characteristics and deserves further explorations

Current research agenda: lower entry-level jobs vs. lower promotion opportunities

Higher School of Economics, Moscow October 14, 2008

Documents

Transcript of Higher School of Economics, Moscow October 14, 2008