When assumptions get violated.

noetsi

Fortran must die
#1
Not talking about violations of the known assumptions of a method. I have reasonable confidence on those. Plus I can always blaame spunky for them.

I normally run linear and logistic regression (and am starting to do time series regression) although I know other methods. One thing that always worries me is I run across articles that says you should not use this method in this situation or should use this approach to fix a problem (which often can not be done with the software I have). I am sure there are many errors I have not heard off (OS in particular gets slammed all the time).

So I have begun to worry about even running these methods (well any statistical method) :p since I could have missed something in the literature or not be able to do the correction suggested (it makes it worse that experts disagree on the right way for me a non-expert).

I was curious on other analyst views on this.
 
#2
I normally run linear and logistic regression
Your preaching to the choire. I heard once that there are basically two ways to run stats
1) Do a method so butt-simple that anyone who questions the result based on some assumption or other looks like they are being pedantic, which they probably are.
2) Do something so complicated that no one wants to say anything because they just want to get to lunch.

I do 1) these days, but I throw in some 2) depending on how the winds are blowing.