0
$\begingroup$

I'm doing some clinical database research and in an effort to lessen the burden on our statistical staff, I started to look for different software solutions to get the analyses I need, and that's where BIDS (business intelligence development studio). BIDS allows queries to be run against a SQL Server and store the results of those queries in a table or view. That table or view is then consumed by BIDS and logistic and linear regressions as while as CART analyses can be done on them.
BIDS Version

The x axis that is cut off on the lift chart is 'overall population %' enter image description here

A mining accuracy chart of the CART enter image description here

I'm not quite sure how this works in other stats packages, but in BIDS you define a certain percentage of your data set to train the model and the rest of the set is compared against that model and the lift chart shows the improvement in identifying the outcome you desire vs. a guess. I'm only vaguely familiar with CART analyses in the first place, and don't know the first thing about R, but this same analysis was done in R with very similar results. The red portion of what looks like a health meter in a video game corresponds nearly identical to the analysis done in R. However, there are no p values in the BIDS version. Consider the node Max Total Poly Pharm >=7 and < 13. BIDS shows me (not present in the picture) value cases probability not present 2133 91.89

Is there any way to ascertain a p value from that. And does R use any of it's cases as training data for the model?

$\endgroup$
  • $\begingroup$ It might be useful to give more details about the R procedure. Which package did you use? $\endgroup$ – Gala Apr 26 '13 at 13:33
  • $\begingroup$ I was afraid of this. I didn't do the R analysis personally, and I wasn't aware that these analyses aren't package-agnostic. $\endgroup$ – wootscootinboogie Apr 26 '13 at 14:50
  • $\begingroup$ Well, R is more an ecosystem or statistical programming language than a centrally managed software package. CART is not included in R base install at all so it must have been carried out using some package or other. $\endgroup$ – Gala Apr 26 '13 at 15:05
  • $\begingroup$ Gotcha. Is it typical in your experience for a CART analysis to be trained be a subset of the data, and then have a comparison (the chart shown above) that shows how good of a predictor the model is? I'm a DBA and this whole stats realm has been thrust upon me so I figured I'll roll with the punches and learn what I can :) $\endgroup$ – wootscootinboogie Apr 26 '13 at 15:11
  • 1
    $\begingroup$ This might be useful: stackoverflow.com/questions/9666212/… $\endgroup$ – Gala Apr 27 '13 at 0:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.