Not sure what program you are using but typically
Log_odds: appropriate beta coefficients are summed.
odds: exp(Log_odds)
predicted probability: odds /(1 + odds)
So. I have a binary outcome that is begin predicted by a race IV which has 4 categories.
here is my output:
LOGISTIC REGRESSION VARIABLES yesORno
/METHOD=ENTER race_eth
/CONTRAST (race_eth)=Simple(1)
/PRINT=GOODFIT CI(95)
/CRITERIA=PIN(0.05) POUT(0.10) ITERATE(20) CUT(0.5).
All racial categories are significant Betas follow:
Race1 comparison group
Race2 .175
Race3 .263
Race4 .808
Constant -.650
Here is the syntax I am using to create the predicted probabilities:
compute pred_logit2 = -.650 + .175*(race_eth=2).
compute pred_logit3 = -.650 + .263*(race_eth=3).
compute pred_logit4 = -.650 + .808*(race_eth=4).
compute dp1 = exp(pred_logit2).
compute dp2 = exp(pred_logit3).
compute dp3 = exp(pred_logit4).
compute dp4 = 1.
compute denom = dp1 +dp2+dp3+dp4.
compute pp1 = dp1/denom.
compute pp2 = dp2/denom.
compute pp3 = dp3/denom.
compute pp4 = dp4/denom.
execute.
HOWEVER,
the predicted probabilities I am getting go like this
Race1 38% comparison group
Race2 20%
Race3 21%
Race4 21%
And this feels like I am making an error some where, since all three of the non-comparison groups are reporting almost the exact same predicted probabilities.
Thoughts / help?
Last edited by jsayn; 07-09-2015 at 02:36 PM.
Not sure what program you are using but typically
Log_odds: appropriate beta coefficients are summed.
odds: exp(Log_odds)
predicted probability: odds /(1 + odds)
Last edited by Dason; 07-09-2015 at 04:02 PM. Reason: Fixed a parenthesis issue so that the predicted probability was actually correct.
Stop cowardice, ban guns!
jsayn (07-09-2015)
Really I just wanted another set of eyes on it. Like I said, I found it strange that all three of my categorical predictors were returning very similar predictive probabilities.
I appreciate the once over though. I don't wanna take these results up high just to make some bad recommendations.
Tweet |