[Bruce Weaver]

Hi Elise. I suspect that Nick's critique (in #2) would have included some of the points listed in this FAQ on step-wise selection.

Your output (in #13) shows another problem with completely algorithmic selection methods when there are categorical explanatory variables or product terms. At each step of the way, the model is lacking at least one term that should be included according to the usual rule about lower order terms that constitute a higher order term being included. I may have missed some, but here's what I came up with, looking at your output:

Step   Missing terms*
01     year, temp
02     year
03     year
04     year, weathercat, humidity, sqhumidity, weathercathumidity
05     year, weathercat, humidity, sqhumidity, weathercathumidity
06     year, weathercat, humidity, sqhumidity, weathercathumidity, windspeed
07     year, weathercat, humidity, sqhumidity, weathercathumidity, windspeed
08     year, weathercat, humidity, sqhumidity, weathercathumidity, windspeed
09     year, weathercat, humidity, sqhumidity, weathercathumidity, windspeed
10     year, weathercat, humidity, sqhumidity, weathercathumidity, windspeed
11     weathercat, humidity, sqhumidity, windspeed, weathercathumidity
12     weathercat, humidity, sqhumidity, windspeed
13     humidity, sqhumidity, windspeed
14     humidity, sqhumidity, windspeed
15     humidity, sqhumidity
16     sqhumidity

* Following the usual rule that all lower order constituents of a higher order
 product term must also be included in the model--e.g., if the a#b#c product
 term is included, one must also include a, b, c, a#b, a#c and b#c.

And finally, if your weathercat variable has more than 2 categories, you have another problem, because weather category would have to be represented by k-1 indicator variables (where k = number of categories), not by a single variable.

HTH.