Last week travails

Class,

I do get the sense that perhaps the course is way too loaded going into the last week. Keeping this in mind, am considering dropping Discriminant Analysis completely from the syllabus. Would rather spend the time on Logit and on how to use it better.

Better still, shall try to estimate the logit model on real data (ice cream dataset), if I can in time for class tomorrow.

I am convinced that Multinomial Logit (MNL) procedure simply does not run on SPSS. At first I thought maybe my simulated dataset was faulty (in Section A). But no, a different dataset for section B, fails to run on the same lines.

Its sad, because MNL is quite useful from a practical usage standpoint. Let me see if I can at least clogit (conditional logit) to work on R.

For sections A and B, because we couldn't get SPSS to run MNL, I'll take a short 10 minutes of lec 10 to estimate and demonstrate logit in general on R on the ice-cream dataset, if all goes well.

End of the day, I want the majority of the janta to understand the basics and building blocks of modeling - their single biggest take-away in terms of quant skills from the class. The ability to write models in the Y=f(X) form and know what they are doing and why.

I trust janta have learned the other take-aways (problem formulation, research design etc) well enough to confidently apply the same to their projects.

I also realize that I'd rather have people do fewer things if they can absorb and do them well than to overload information this late in the game.

Sudhir