Modelling the Gender Pay Gap

1
Modelling the Gender Pay Gap

• By Wendy Olsen and Sylvia Walby
• (Part of a 3-part project on Modelling Gendered
  Pay and Productivity, EOC 2003-5)

2
Publication

• www.eoc.org.ukWorking Paper No. 17
• For the report that was dated 2002, by the same
  authors, using similar techniques with 2000 data,
  see
• http//www2.umist.ac.uk/management/ewerc/equalpay/
  walbyolsenreport.pdf

3
Introduction

• Re-thinking the dichotomy between human capital
  and discrimination
• Regression was used.
• Then fixed effects modelling,
• And decomposition of the pay gaps causes.
• Critique of Oaxaca
• Using simulation to do decomposition
• What accounts for the gender wage gap?

4
Human capital and discrimination are not mutually
exclusive

• Re-thinking the dichotomy
• Human capital theory is re-estimated
• Part-time work is associated with no rise in wage
• Interruptions are associated with lower wages
• What is the place of institutions?
• Re-interpretation of the coefficients
• One interpretation focuses on the variables
• Other interpretations are suffused with theory,
• E.g. the labour market rigidities
  interpretation
• And the EOCs discrimination and other factors
  interpretation which is misleading

5

• Regression results
• The main factors influencing wage rates for women
  and men
• Female 8.9 lower wages if female
• Education (years) 5.7 higher wages for each year
  of FT education
• Years of full-time employment (curved) 2.6
  higher wages for each year of FT work
• Years of part-time employment (curved) 0.8 lower
  wages for each year of PT work
• Unemployment (years) 2.2 lower wages per year
  of unemployment
• Family care (years) 0.8 lower wages for each
  year of interruptions to employment for childcare
  and other family care
• Recent education not employer funded 5.9 lower
  among those funding their own training

6

• Regression results
• Further (institutional) factors influencing wage
  rates
• Segregation (male percent x10) 1.3 higher wages
  per 10 more males in that occupation
• Firm size 500 workers 11.7 higher wages if firm
  size is over 500 workers
• Firm size 50-499 workers 6.2 higher wages if
  firm size is 50-499 workers
• In public sector 8.0 higher wages if working in
  public sector
• In union or staff association 6.2 higher wages
  if union member
• (These are the same regression continued. That
  regression also has SIC and REGION in it)

7

• Regression results
• The results for female of 9 are re-affirmed
  using ten years of data. (See Appendix of EOC
  Working Paper No. 17)
• Panel data set for 1992/3, 1993/4, 1998/9,
  1999/2000, 2000/2001, and 2001/2 from BHPS
• I merged the annual work-life histories for the
  people who are in this data set continuously or
  who enter the data-set as young people later in
  the panel.
• The work-life history data and annual data are
  used together, to re-calculate a fixed-effects
  regression, which shows a huge female factor (a)
  due to preferences or motivation or
  discrimination (Kim Polachek). We calculated
  the 9 figure from their technique for estimation
  of the gender component of the fixed-effects
  individual heterogeneity.

8
(No Transcript)
9
The Human Capital Results
Variables Education (Scaled in years) The
length of the working-life that was spent in
full-time work The length of the working-life
that was spent in part-time work The length of
time spent in interruptions of the working-life
for caring and family work Other periods
Unemployment Longterm sick/disabled
periods. Training on the job that is
employer-funded or at the place of
employment Training during the past year that is
not employer funded nor on the premises of the
employer
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
Oaxaca

• Operationalises the dichotomy between human
  capital and discrimination
• Poor grasp of institutional causes of gender
  wage gap (Juhn, Pierce Murphy extension)
• Estimates of discrimination unstable and
  arbitrary, depending on choice of comparator
  men, women, all. (ORansom Neumann)
• Inclusion of 3rd term to represent average
  improves but does not eliminate problems
• Separate regressions omit gender despite its
  significance and considerable effect.

15
Equations

• Traditional Oaxaca two-term equation
• Mens wage rate relative to womens wage rate
  human-capital effect a residual discrimination
  effect.
• The full decomposition of the wage gap equation
  is offered by
• ln wm ln wf (Xm - Xf) ?m (?m - ?f)Xf (Eq.
  2)
• where the Xi's refer to the mean for men and
  women of each variable. The ?i are the slope
  coefficients for the men and women respectively.
• Hence wm/wf exp(Xm - Xf) ?m (?m - ?f)Xf
  (Eq. 3)

16
Equations

• Oaxaca three-term equation (OR, 1988, 1994)
• Ln (gap1) (Xm - Xf) ? (?m - ?)Xm (? -
  ?f)Xf (Eq. 4)
• productivity differential male wage
  advantage female wage disadvantage

17
Beyond Oaxaca Originality in the Research So Far

• A single, full (integrated by sex) regression,
  with institutional as well as individual factors
  included
• Gender a variable in that regression
• Heckman to eliminate potential sample selection
  bias also done in panel
• Simulation to estimate size of components of
  gender wage gap

18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
Problems with Oaxaca-Blinder

• 1) The labelling of slope and levels components
• (endowments Oaxaca and Ransom 1999
• discrimination vs. productivity, OR 1994)
• 2) Interpretive contradictions
• a) descriptive contradictions, where the
  operationalisation of discrimination is found
  both in both the discrimination and the
  productivity terms
• b) normative contradictions, where the approval
  of one term has as its dual the disapproval of
  the other term

22

• 3) Arbitrary reference point of the male wage
  equation (Applies only to two-term Oaxaca, not to
  3-term version found in OR 1988 Neilsen 2000)
• 4) Arbitrary reference point of one category,
  e.g. lowest level of educational qualification
• 5) Oaxaca discourages adding up the three terms
  (or two terms) horizontally to see the net effect
  of each associated factor
• 6) Not well adapted to the factors other than
  human capital inherently individualistic.

23

• 7) Does not handle nicely the factors which are
  present for one sex but not for the other
• 8) Considers womens slopes only in relation to
  other womens returns -- but the slope is higher
  whilst the intercept is lower than men
• 9)Considers mens slopes only in relation to
  other men lacks a sex term in equation.

24
Summary What makes a difference to rates of pay?

• Gender
• Motherhood (current and former)
• Employment experience (nuanced)
• Part-time (not pro-rata, not neutral, but
  negative)
• Interruptions for child and other family care
• Training, tenure
• Segregation
• Institutions firm size, public sector, union
  membership
• Region and industry

25
The Next Two Stages of Research

• 1. We have simulated the effects of changing the
  values of X-variables, e.g. education, training,
  occupational segregation, and the work-histories.
• 2. We give results for each type of woman.
• 3. The aggregation of results is costed out (as
  a cost-benefit analysis) for 4 stakeholder groups.