There is a lot of research on student evaluations, perhaps two-thousand articles. One meta analysis indicates that student ratings are affected much more by expressiveness than by content, while student achievement is affected more by content than by expressiveness. See 'Educational Seduction', by Abrami, Leventhal, and Perry, in Review of Educational Research. That does not mean there is no correlation between ratings and effectiveness, but it does mean we need to be cautious in making conclusions about perhaps subtle effects on effectiveness when our data are from ratings. I am not criticizing Benjamin for using ratings; indeed, I admire and applaud his effort. I would just prefer that the conclusions be stated without the logical leap. What is wrong with saying, "I found no correlation between writing output and student ratings." That would let the reader decide how much to conclude about teaching effectiveness.
As for Bill's market point, I am not arguing that these educated consumers we call students do not know what they want or that they do not know it when they see it. But I have little faith that they want what we educators think they need, as opposed to entertainment or bar passage. They could rationally choose to give up some small bit of lawyering effectiveness in the future for some entertainment now. I have even less faith that they want to learn what society needs them to learn for the good of all. I'd bet big money that I could raise my student ratings and lower the quality of my Property instruction by replacing future interests with more on landlord-tenant law.
The more important point that I wanted to make is that Benjamin seems to be looking for a simple relationship when a complex one is more likely. To carry my optimization point further, suppose that most faculty members have experimented with more or less research time, or in some other way have found a mix of preparation and publication that maximizes the value of their own efforts, in their own eyes of course. It seems entirely possible that asking them to do more writing would result in substantially lower quality teaching, otherwise they would have made that trade themselves. On the other hand, asking them to do substantially more class preparation would result in less writing without substantially advancing their teaching effectiveness. We should not be surprised to find that each faculty member reaches his or her optimum at a different mix of teaching and writing. This sets up what might be viewed as a restriction-of-range problem. It is possible that the data would not reveal any correlation for the group even though a complex relationship would indeed exist for each faculty member.
In short, it seems likely that writing and teaching are complements at low levels of writing and substitutes at high levels of writing, and determining whether any particular faculty member has increased his or her writing to the point that further increases will reduce teaching quality is a very tricky business.
Jeffrey E. Stake