Tuesday, April 22, 2003

Twice-standardly deviant 

A colleague asked if I would comment on a salary equity study (click here and scroll to the bottom, please) that has been done for St. Cloud State. I confess at the outset the following: while I am an economist, labor economics isn't my forte; I have very incomplete information to work from (which is part of my complaint); and I am one of those types that views all statistics with serious skepticism (which I consider a good thing, but not everyone agrees.)

What bothered me as well as some friend is the following conclusion, found in slide 37 of this presentation (in .pdf):
  • Potential pay equity adjustments should be based on individual criteria rather than group membership criteria.
  • Pay equity issues should be addressed through an examination of disparities
    between actual and model predicted salaries (i.e., residuals).
    Using the variables identified in the total population model.

    Gender and ethnicity variables not included.

  • Any faculty member below a threshold (see next slide) will be eligible for a
    potential pay adjustment, if adjustments are recommended.
    We used the notion of a threshold (std error) due to the fact that there is a range of
    acceptable variation in the model not explained by the predictor variables. Therefore,
    adjustments to a predicted average salary are not recommended.

  • Potential pay equity adjustments will be scrutinized by a MnSCU/IFO salary
    review committee for data verification purposes.
Now there's something to like here, that people will be judged individually rather than members of a group. And the second bullet is standard operating procedure in these studies. But the last two? If someone lies outside some standard deviation banding, one, one-and-a-half, or two (to be bargained between union and administration) those below get paid. One colleague asked, what about the faculty who are at +2? Do they have their salaries frozen?

This would make sense, perhaps, if there was some evidence of inequity in pay by race or sex, but there's no evidence of that by 2002. (There was some evidence for 1997.) Over the entire MnSCU system, of the 72 people that lie outside -2 standard deviations, 48 are white males. So we are adjusting for pay equity on an individual basis, but without any understanding of how people got there; we are going to increase pay for people below the 1.5 or 2 standard deviation threshold without saying anything about the people above. That doesn't make much sense.

It would be nice to say more about the report, but the two reports lack much of the information needed to evaluate the quality of the studies. The consultants provide no evidence on whether the results are heteroskedastic or not, for instance (not at all uncommon, as I understand these types of studies to be). While they've examined proxy bias by checking the relationship between discipline and race or sex, they do no such check on the relationship between experience and race or sex. They claim there is no evidence of salary compression (for which we negotiated a huge increase in pay for people at 20 and 30 years of service in our last contract -- see section K), yet their results show $4560 additional salary received by those who are probationary versus those who are tenured. That result is consistent with the fact that faculty are hired in a competitive market but then are given pay increases based only on years of service -- no reference to market or merit. It could be consistent with something else, but there's no way to tell that from the reports, but it fits anecdotes on campus that the longer you're here, the more you are dragged down from what the market will bear. And of course only the good ones leave.

So I lack confidence in the study. I'm not convinced that anyone outside of the consultants have the statistical experience necessary to look at this. I'm not volunteering myself, as there are at least three members of my own department who are demonstrably more competent than me to do the job (though the decision should be based on comparative, not absolute advantage.) But before the university system starts doling out money on that basis, a closer look seems warranted.