1

Working on dissertation. Theory: in K-12 education putting more administrative authority in the state board of education is "better" that leaving it to the local boards.

DV: data from 43 states on % kids graduating high school within 4 years.

IVs: I have 37 different measures for the types of administrative authority- who is responsible for buying school books, setting school stop/start dates, etc.: 3 (state has complete control), 2 (shared/split) and 1 (locality has complete control).

So I fire up SPSS, data from 43 states on % kids graduating high school within 4 years as my DV, plunk in the 37 IVs, use "Enter" as my method (I've been told stepwise is evil, evil, evil) and...

Model Summary

R = .853

R Square = .727

Adjusted R Square = -.041

Std. Error of the Estimate = .148600607125323

This might be "good" if it means the predictors are useless and the hypothesis is wrong (or at least cannot be proven with this data).

It is "bad" if I am getting this because my model stinks. How can I determine which?

user39799
  • 71
  • 1
  • 7
  • You should mention the same question you posed to the [SPSS list-serve](http://spssx-discussion.1045642.n5.nabble.com/Negative-Adjusted-R-Square-is-a-quot-good-quot-thing-td5724399.html), as you are likely to get similar notes on over-fitting by the members here. – Andy W Feb 10 '14 at 03:24
  • Trying to cover all bases; this is my dissertation and the culmination of years of work and it is looking as if I have to abandon 2+ years of work and data collection, and my methodologist never even mentioned anything about over-fit, so not real happy now OK? – user39799 Feb 10 '14 at 03:30
  • The data collection is fine you don't need to throw anything out. You just need to figure out which of your IVs are most important. Forward step-wise regression is one way to do this, and it really isn't evil as you said. There are plenty of more sophisticated things you can try (like using LASSO or all subsets regression), but if you're in a pinch just do step-wise regression. Do a google search for "variable selection" and "multiple regression". – ahwillia Feb 10 '14 at 04:20
  • See also: http://stats.stackexchange.com/questions/34751/is-it-a-problem-to-get-a-negative-adjusted-r-squared – landroni Feb 21 '16 at 08:48

0 Answers0