Obsidian | Level 7

## Interpreting Analysis of Maximum Likelihood Estimates for predictor interactions in Proc Logistic

Hello,

I wondering what is the simplest way to interpret/explain the Analysis of Maximum Likelihood Estimates.

Can I use the estimate column, in the Analysis of Maximum Likelihood Estimates, as an OR?

For example,  if the estimate values for my interactions were 2.67 or .70,  can I say: Combo A is 2.67 times more likely... and Combo B is 70% more likely to...

proc logistic data=library.dataset plots(only)=(effect oddsratio) ;
format education edufmt. exposure yesnofmt.;
class education (ref="phd ") exposure(ref="No")/param=ref;
model depression (event="1") =exposure*education;

run;

I thought this interpretation was acceptable but now I'm not sure if it only applies when referring to the Odds Ratio Estimates and Profile-Likelihood Confidence Intervals table which look similar

1 ACCEPTED SOLUTION

Accepted Solutions
Super User

## Re: Interpreting Analysis of Maximum Likelihood Estimates for predictor interactions in Proc Logisti

I think you should post it at Statistic forum,since is about STAT:
https://communities.sas.com/t5/Statistical-Procedures/bd-p/statistical_procedures
Maybe @StatDave could give you a hand.

Here is my two cents.
I think you need EXP() to get OR=EXP(beta) .
And about interpret/explain of interaction effect, it is a bit of complicated.
Here is a url you can refer to:
https://support.sas.com/kb/38/384.html
2 REPLIES 2
Super User

## Re: Interpreting Analysis of Maximum Likelihood Estimates for predictor interactions in Proc Logisti

I think you should post it at Statistic forum,since is about STAT:
https://communities.sas.com/t5/Statistical-Procedures/bd-p/statistical_procedures
Maybe @StatDave could give you a hand.

Here is my two cents.
I think you need EXP() to get OR=EXP(beta) .
And about interpret/explain of interaction effect, it is a bit of complicated.
Here is a url you can refer to:
https://support.sas.com/kb/38/384.html
Obsidian | Level 7

## Re: Interpreting Analysis of Maximum Likelihood Estimates for predictor interactions in Proc Logisti

Thank you so much! This is helpful to know, I'll post it on that forum. Getting the exponential seems to work as well!
Discussion stats
• 2 replies
• 573 views
• 2 likes
• 2 in conversation