This past spring, when I was teaching an advanced graduate statistics course at Texas Tech University on structural equation modeling, I wrote a tribute to the late Frank Andrews and Laura Klem (who is still active at UM), who taught the same course to me at UM in 1988 (April 13 entry, April archives).

Right now, in the Fall semester, I am teaching introductory statistics at the graduate level. Accordingly, I thought I'd say a few words about the professor I had for intro stats at Michigan during my first year of graduate school (1984-85), the late J.E. Keith Smith.

Keith was always very friendly with a quirky sense of humor, but he taught intro stats very rigorously, deriving formulas and attempting to document their theoretical background. In all candor, the material was complex and sometimes difficult to follow.

However, if you went to Keith's office with a specific data-analysis question (either while in his class or even several semesters after you'd had him), he was as clear as could be. He would instantly grasp the type of analysis you'd need to do given your research design, and his instructions for how to implement the analysis on the computer were easy to follow. I know that several faculty members and graduate students would consult Keith on various statistical and experimental-design questions and his advice was always valued.

(As an aside, I still have my textbook that I used in Keith's class, Statistics [3rd ed.], by William L. Hays. Hays, who passed away in 1995, had himself been a faculty member at Michigan until 1973 and finished his career at the University of Texas, Austin. My Hays book was bound so poorly that within a year of my purchasing the book, chunks of pages were falling out; I've had to scotch tape these pages back in over the years. Although I do not teach with the Hays book, I continue to refer back to it for formulas and explanations.)

In my second year of graduate school, I sat in on a categorical data analysis course Keith was teaching. Essentially, this latter course covered more advanced and sophisticated variations on the basic chi-square test. Given a table showing, for example, how many people fell into each of the six cells created by the combinations of gender (male/female) by party identification (Democrat/Republican/Independent), we would typically compare the frequencies in the table as a whole to what would be expected by chance. Keith was working toward facilitating tests of how the frequencies in one (or more) cell would compare to the frequencies of other cells.

When I remarked during one class that the technique he was showing looked very similar to contrasts in Analysis of Variance (ANOVA), which we had learned about in intro stats, he replied, "You've been with me long enough to know that eventually everything will look like a contrast!"

At the time, there was a famous categorical data analysis program called ECTA (Everyman's Contingency Table Analysis). Keith said he was working on a program called VECTA (Very Easy Contingency Table Analysis), which, Keith also noted, was how a New Yorker would pronounce "Vector."

Keith died in 2002. His obituary is available here (you have to scroll down once the page comes up).

To this day, I have a love of numbers and statistical analysis, as exemplified not just in my teaching of research methodology and statistics, but also in fun endeavors such as my hot hand website, which applies probability and statistics to the analysis of sports streaks. I think this represents, at least in part, a legacy of my having learned statistics from Keith.