Outline of key points:


During 1980's about 30% of the young doctors were women (where young means under 45) and only 10-15% of the older doctors were women. See:

historical data

Contrast this with ARCOM who has managed to attract enough women to make up over 50% of all its young doctors (those with under 10 years experience). Even historically, ARCOM has hired well in the sense that it has about 25% of its older doctors being women. So its hiring practice seems to be attractive enough to women to attract almost twice the typical number.

(See logistic graph from JMP for supporting data.)

  1. (1) But have these women been given comparable pay increases since they were hired?
  2. (2) Were they hired at equivalent salaries?
  3. (3) Is there other evidence of pay discrimination?
  4. (4) Is there promotion or hiring bias?

NOTE: They can't argue point 2 since there is no data on it. So innocent until proven guilty?)

Analysis in jmp

I'm working in jmp. Here is the data in that format.

Pay increase: (in dollars, in percent)

In 1985, Women were given pay increase that were $5513 less than men got. But, when controlling for field and experience women got pay raises within $200 of those of men. (In fact women were given larger increases when controling for field and experience.) These numbers aren't statistically meaningful due to issues of hetroscadasticity.

Using percentages, is better statistics. Raises were mostly between 5% and 14%. But, they were typically about .2% lower for women than for men (2*.1% se=.12). When controling for field, this is left basically unchanged (2 * .1%, se=.13). When years of experience is also controlled for, this stays statistically insignificant at .26% (2 * .13 se=.13).

So, it appears that over the year 84-85 there wasn't statistically significant differerence between the raises given to women and those given to men.

Current salary (log sal, outlier removed)

If there was a history of pay increase discrimination it might show up by having current pay be lower for women than men. Unfortunately, lots of other things can show up here also: years experience, field, previous discrimination done by the educational system of the 1950's, etc. So care will have to be taken to tease out these other effects.

Across the entire school women are paid about 40% less (2*19%, se=3%). So it is obvious that women and men are different. We didn't need regression to determine this fact though.

We will look at three sources of this pay difference: field of choice, years of experience and sex.

The most important of these three is the field each person chooses to enter. When this is controlled for the pay difference drops down to 20% (10.2% se=1.7%). Notice that the hospital pays surgeons twice the average (73%, se=3%), pediatrics about par (-4%, se=4%) and doctors in physiology almost 1/2 of typical (-48%, se=3.7%)

When years of experience is controled for, the pay difference for women is about 7% (2*3.4% se=1.7%). Obviously if we controlled for different things this number would keep changing. Rather than try every possibility, let's work with an analogy.

"Publish or perish." Its the montra of academia. But if you look for how much a paper affects your pay, it doesn't look like publishing is a good thing. Writing one more paper per year lowers your pay by 4% (4.4%, se=1%, controling for experience and field). But still most of the doctors write about 4 papers a year? Why are they cutting their salary by almost 20%?

Papers are a proxy for the type of medicine you do. It is a proxy for what you find interesting in life.

This data doesn't prove that women are getting 7% less pay than men are any more than it proves that publishing a paper decreases ones pay. Since both effects are statistically small, it is easy to believe that something else is the source of what is going on.

Personal expertise:

I have done research on the importance of affirmative action (see ``An Economic Argument for Affirmative Action,'' with R. Vohra, Rationality and Society, (1992), 176 - 188, with discussion by G. Loury, by D. Friedman, and by J. Heckman and T. Philipson.)

I will be using multiple regression for purposes of analysis. For the past 15 years I've been an expert on multiple regression. (See ``The Risk Inflation Criterion for Multiple Regression,'' with E. George, The Annals of Statistics, (1994), 1947 - 1975.) Multiple regression is the primary tool for investigating these sort of problems.
Last modified: Thu Dec 2 12:36:06 EST 2004