Title: Pay%20Differences%20by%20Gender%20of%20University%20Faculty
1Pay Differences by Gender of University Faculty
- Jenny Hunt, Daniel Parent, and Michael R. Smith
2Institutional context
- Non-union staff association (MAUT) discusses
salary with the University administration. - Everyones a professor (sort of) about 1,400 in
the various analyses. - Normal career progression involves
Assistant-Associate-Full Professor.
3(No Transcript)
4The response
- A little less than 1,000,000 assigned over three
years to correct female salaries, informed by
position of individual salary with respect to
relevant regression line. - It was a agreed that the study would be repeated
in 2003 to determine whether or not the gap had
re-emerged
5Slippage between 2003 and 2009 (1)
- A report on regression analysis commissioned from
a faculty member (Fall, 2003). - Responsibility for the analysis assigned to a
private consulting firm contract signed in
2005 The distance and lack of familiarity with
McGill data and structure has also, in my
opinion, hampered the firms ability to
interpret the data. Academic Salary Gender Study
Status Report, July 25th. - In early 2006 the contract with the private
consulting firm was terminated and responsibility
for the production of an analysis assigned to
Mary Mackinnon and Michael Smith, who approached
Jenny Hunt and Daniel Parent for assistance.
6Slippage between 2003 and 2009 (2)
- Sometime during 2006 the project was unilaterally
appropriated by the Office of the Provost.
Various fragments of analysis were presented to
the Committee on Academic Salary Policy but no
report was produced. - At the request of the SSCOW, in December 2007 a
question was asked in Senate on the whereabouts
of the report. - Shortly after that, responsibility for the
analysis was returned to a group of people
associated with the MAUT.
7(No Transcript)
8The 2000 study was of great value, but
- It didnt tell us much about the sources of
gender differences in earnings. - There may have been biases in the 2000 analyses
leading to either an underestimate or
overestimate of the gender disadvantage. - This sort of analysis usually examines log
earnings. - CRCs and related awards have become more
important since 2000. - A premise a single coefficient is unlikely to
suitably inform us on pay differences by gender. -
9Possible sources of bias
- Bias that may increase the measured gender effect
. - Department groupings may not fully reflect market
differences across fields of study. - Bias that may reduce the measured gender effect
. - Rank was controlled in these models but may be
endogenous with respect to gender. (If youre
going to discriminate against me in pay, why
wouldnt you discriminate against me in
promotion?) - Department groupings may reflect downward
pressure on pay through the devaluation of
womens work.
10Possible sources of gender difference
- systematic disadvantage of some sort or another
- differences in mobility potential (Blackaby,
Booth, and Frank, 2005) - differences in quality.
- We bear these in mind in the analysis and return
to them in the conclusion.
11McGill pay policy is likely to influence outcomes
.
- Entry-level pay varies substantially, matching
market processes common across North American
research universities. - Annual pay increases are substantially tied to
merit judgments. - Pay may be increased to retain a faculty member
who has received an offer from another
university. - Pay has (erratically) increased with promotion to
full professor. - Pay is further increased through the award of
federal and McGill chairs. - Plus, heterogeneity in practices across McGill.
12Choosing how to describe the data
- Avoid complicated ways of incorporating indirect
effects e.g., that gender may have a direct
effect on earnings but also an indirect effect
through time to promotion. - Instead, look at some descriptive information on
parts of the McGill pay determination process
that might account for differences in earnings by
gender. - Then regress log pay on gender and controls.
- Include awards.
- Exclude administrative stipends.
- Exclude part timers.
13(No Transcript)
14(No Transcript)
15(No Transcript)
16(No Transcript)
17(No Transcript)
18(No Transcript)
19(No Transcript)
20(No Transcript)
21(No Transcript)
22(No Transcript)
23(No Transcript)
24(No Transcript)
25(No Transcript)
26(No Transcript)
27(No Transcript)
28(No Transcript)
29(No Transcript)
30(No Transcript)
31What do these slides tell us?
- (Remember, the slides include no controls!)
- Longer time to promotion may contribute to lower
female pay i) the proportion of women in Arts
and Education is higher than it is in
Engineering, Science, and Medicine and the time
to promotion is longer in Arts and Education ii)
women in Medicine take longer to be promoted than
men. - Merit pay appears not to be a cause of lower pay
for female faculty. - Men are more likely to have their pay increased
with an external award. - Much of the average difference in pay between
males and females is produced by the very high
pay received by a small number of male faculty
members. - Anomaly and retention no satisfactory evidence
on this.
32No controls in the charts so move on to
regression analysis
- Log salary evidently a technical advantage
where the distribution is skewed. - Experience - consistently measured as years since
Ph.D. both number of years, and the square of
number of years. - Enter the variables of interest consecutively to
see what happens to the gender effect, as
consecutive controls are added, adding
potentially endogenous factors at the end - Two regression techniques OLS and median
regression. The first is sensitive to extreme
values, the second isnt. - Maximum control for department.
33Statistical significance
- P-values routinely reported in this work and used
in interpretation. - We have a population.
- However, the population in 2007 could be viewed
as a sample from a hypothetical set of possible
McGills. - In practice I shall take 0.1 (two-tailed) as the
significance threshold. (Cf. Oxana Marmer and
Walter Sudmant, Statistical analysis of UBC
faculty salaries Investigation of differences
due to sex or visible minority status, UBC
Panning and Institutional Research.
34(No Transcript)
35What does the university level analysis show?
- Using OLS, controlling for all departments
- women earn 3 less than men before controlling
for rank - Women still earn almost 2 less than men after
rank is controlled - The difference becomes weaker or insignificant
after either appointed as full professor or
holds award is added - Using median regression, controlling for all
departments - the pay disadvantage of women becomes clearly
insignificant once rank is added. - The differences between the OLS and median
regression results show that the extreme values
we saw in the density functions are influencing
the OLS results. - Neither OLS nor median regression is right.
36What about within (large) faculty results?
- In the previous study and in the reanalysis using
the same model, disadvantage was concentrated in
the faculties of Arts and Medicine. Our analysis
modifies the method used in the previous study in
potentially significant ways. In particular - we use the log of salaries rather than salary in
dollars - So far weve reported results that have
controlled for all departments.
37(No Transcript)
38(No Transcript)
39What do these tables show?
- Median regressions eliminate disadvantage in all
three faculties. - OLS suggests female disadvantage in Arts, up to
addition of rank. But remember, years to
promotion does not seem to be longer for women in
Arts and Education. - There is no evidence of a female earnings
disadvantage in Science. - In Medicine, after controlling for departments,
the OLS result becomes insignificant. This is
surprising given that, using the 2000 model,
there were significant or approximately
significant differences in Medicine in 2006 and
2007, in which the controls were experience,
departments, and rank. Possible explanations i)
the department dummies used ii) the switch from
raw to log dollars.
40(No Transcript)
41The previous table replaces log earnings with
dollar earnings
- Using OLS, with the following variables in the
model gender, experience, various department
controls. - Women are disadvantaged they earn about 4,000
less - in all models. - The different department specifications do not
substantially change the conclusion.
42Some other effects
- At University level, maternity leave has no
significant effect. - At University level, (married to a
professor)(gender) has no significant effect.
43Back to possible explanations for disadvantage
- Quality merit awards suggest, if anything,
higher female quality. - Limits on mobility of women i) gendermarried to
McGill professor interaction the right sign,
but insignificant ii) the effect of hired at
rank of full professor on the gender coefficient
might be consistent with this. - Systematic disadvantage i) women have a lower
rate of promotion in Medicine larger
proportions not promoted in faculties with more
women (Arts, Education) ii) fewer women get
awards iii) external recruitment brings in fewer
women.
44Some broader conclusions
- The relative advantages of institutionally
specific versus general surveys. - The issue of the tails of the distribution
- 1. Income earnings data sources top coded or
extreme values deleted. - 2. Most analyses focus on central tendency.
- 3. Some interesting action concentrated in the
tails e.g. Frenette, Green, and Milligan, CJE
(2007) Census data with tax adjustments imputed
overall inequality influenced by falls in
relative income at the bottom of the
distribution, not picked up by surveys.
45Some broader conclusions
- There is marked heterogeneity within the
University. - Different mechanisms to produce the same outcome
e.g. the University of Montreal.