1
$\begingroup$

I run an ordinal logistic regression model using both SPSS and Minitab. The dataset is exactly the same. The results are exactly the same, but in opposite directions. I have not manipulated default values of the two programs.

The result of SPSS is the correct one. I know this since when I analyze this dataset using Minitab, but using its linear regression (not ordinal logit), the Minitab itself gives something similar to SPSS. But Minitab's ordinal regression tends to consider a strange default. All betas in logistic regression of Minitab are negated.

Why is that?

$\endgroup$
  • 4
    $\begingroup$ Ummm... I guess someone at Minitab thought this was a good default? Every piece of software has some odd choices built in. $\endgroup$ – Peter Flom Jun 10 '13 at 18:19
  • 2
    $\begingroup$ See this answer for an explanation of the possible ways to write down the same model for ordinal logistic regression. $\endgroup$ – caracal Jun 10 '13 at 18:40
  • $\begingroup$ Caracal, your answer there was awesome. :) $\endgroup$ – Vic Jun 12 '13 at 17:32
3
$\begingroup$

I suspect that the models are specified differently between SPSS and Minitab. I think that SPSS "puts" a minus sign between the intercept and all the regression coefficients. This is ensures that, for positive coefficients, increases in X values lead to an increase of probability in the higher-numbered response categories. However, I wonder if Minitab "puts" a plus between the intercept and coefficients so increases in predictor values lead to an increase of probability in the lower-numbered response categories. Check documentation for each software package to verify this.

$\endgroup$
  • $\begingroup$ Thanks a lot dear Scott. I am wondering which one is the right one? Should that minus be there (I think the answer is yes, of course), but don't know why Minitab ignores it in the first place? I would check out the documentation and update later. $\endgroup$ – Vic Jun 10 '13 at 22:01
  • $\begingroup$ Both methods are "correct." Which one you choose will depend on which method you find to be easier to interpret. $\endgroup$ – Scott Millis Jun 11 '13 at 14:07
  • $\begingroup$ yes yes they are correct. I meant the one which is closer to common sense. Caracal's answer ( stats.stackexchange.com/a/38130/1909 ) made it clear that SPSS's approach is the good one (in which a positive beta indicates increases in X together with "increases" in Y), but the Minitab's approach is counterintuitive (increases in X accompanies decreases in Y). $\endgroup$ – Vic Jun 12 '13 at 17:31
  • $\begingroup$ Vic, Yes, I agree with you. $\endgroup$ – Scott Millis Jun 12 '13 at 17:56

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.