Is automatic linear modelling in SPSS a good or bad thing? In SPSS Version 19 there seems to be a new feature called Automatic Linear Modelling.  It creates a 'Model' (which is new to me) and the function seems to combine a number of the functions that is typically required for prediction model development.
The functionality seems incomplete with only a subset of prediction selection techniques and most notable it's missing Backwards step wise.
QUESTIONS


*

*Do people see this as good or evil?  

*And if 'good' then are there ways to decompose what it is doing? 

*Specifically how do I find the regression equation co-efficients when bagging or boosting? 


To me it seems to hides a lot of steps and I'm not exactly sure how it's creating what it presents.  So any pointers to tutorials or the like (as the SPSS documentation isn't great) is appreciated.
 A: I had a quick look at the IBM SPSS advertising material.
It sounds like it is part of a general move on the part of IBM/SPSS to get involved with predictive analytics.
Terms like automatic data preparation, boosting, bagging, and automated model selection are popular in data mining and predictive analytics communities.
In that sense you may see similarities with open source tools like Rattle and Weka.
You might find useful this article by John Maindonald introducing data mining.
In summary, if you have some combination of the following factors, then such tools may interest you:


*

*interested in building predictive models (as opposed to testing apriori hypotheses)

*you have lots of data

*you want some hand holding on the steps of data analysis

A: On tutorials:  SPSS provides a lot of material that is not mentioned in the Tutorial table of contents.  For many specialized topics, the way to access the tutorial is to go into the Help files for that topic and poke around until you see a link saying "Show me."  That'll take you to what is usually quite a good tutorial.
