My paper, Selling Students Short, released on March 20, 2013, examined teaching loads among tenured and tenure-track faculty. I relied on teaching load data from the U.S. Department of Education’s Data Analysis System (DAS). After publication of the paper, further investigation revealed that the teaching load data I used was incorrect.
Specifically, the 1987-88 DAS output for the “mean” value of the “Classes, number of” variable in 1987-88 is 3.6, but this is not the mean teaching load of the typical professor. Moreover, the method by which this value was calculated makes it incomparable to the mean teaching load of the typical professor in 2003-04, meaning that we simply cannot determine whether teaching loads have declined, stayed the same, or even increased between 1987-88 and 2003-04.
Here is what happened:
This is the DAS output table in question.
I interpreted the numbers in the “Classes, number of (Mean)” column to be the mean number of classes taught. This is the proper interpretation of the corresponding columns for the 1992-93, 1998-99 and 2003-04 surveys. But this is not the correct interpretation for the 1987-88 survey.
For the 1987-88 survey, responses for the number of classes taught question were numerically categorized, and the “Classes, number of (Mean)” column in the DAS screenshot above refers to the mean of the numerical categories, rather than the mean of classes taught, and these numerical categories do not match the number of classes taught. A professor who taught zero courses was assigned to numerical category one, and a professor who taught one course was assigned to numerical category two, etc. Since most professors were in a numerical category with a number equal to their actual teaching load plus one, the mean of the numerical categories is approximately equal to the mean teaching load plus one.
In addition, for the 1987-88 survey, professors who did not teach any classes did not answer the number of classes taught question, but did answer that question in the 2003-04 survey. In other words, the 1987-88 “mean” value does not contain any professors who taught zero classes, but the 2003-04 “mean” value does.
The bottom line is that changes in the surveys and the calculations by DAS mean that we cannot determine whether teaching loads for the typical professor declined, stayed the same, or increased. There are subsets of classes for a subset of professors meeting certain criteria where it may be possible to estimate the change in teaching loads by manually adjusting DAS figures, but this subset of classes and professors will not necessarily be representative of the change in teaching loads for the typical professor.
While reporting a “mean” value of a “Classes, number of” taught variable when it is not the mean number of classes taught could be considered misleading on the part of the U.S. Department of Education, the ultimate responsibility for this misunderstanding rests with me. I apologize for any confusion and inconvenience this may have caused. The only consolation I can offer is that, as this experience amply demonstrates, I will continue to faithfully follow the data wherever it may lead.