Hello,
I created different statistical models using demo data and need help interpreting/wording the odds ratio/output from Proc logistic simple terms. I have models with and without interactions. I appreciate any input as I am still learning to read output correctly. There are so many different ways to get Odds ratios that I worry about reading it incorrectly or am missing a step when it comes to inverted values.
Proc Logistic: Interaction Model Questions:
I used the exponentiated values from the Estimate value from the Analysis of Maximum Likelihood Estimates table, to calculate the Odds ratio (OR)
If I exponentiate , 0.2888 (for the interaction Race*SomeCollege Hispanic Yes and get 1.33,can I say that Hispanic women who attended some college were 33% more likely to have infants with low birthweight, compared to White women who did not attend college?
If I exponentiate an inverse value such as -0.2205, and get 0.802, do I substract from 1 and say: Hispanic women were ~20% less likely to have infants with low birthweight, compared to White women.
For the estimate table (Exp=1.9422), can I say, Black women who did not attend college were 94% more likely to have infants with low birthweight, compared to White women who did not attend college?
proc logistic data=Sashelp.Birthwgt plots(only)=(effect oddsratio) ; class race (ref="White")somecollege(ref="No")/param=ref; model LowBirthWgt (event="Yes")=somecollege|race/ clodds=pl selection=backward; estimate 'Black, no college' race 0 1 somecollege 0/exp; run;
Questions from a simple Logistic Model with No Interactions :
Using the Odds Ratio Estimates table- Can I say women who went to college, were 82% less likely to have low birth weight infants, than women who did not attend college? Or would it be, 18% because I have to subtract from 1?
proc logistic data=Sashelp.Birthwgt plots(only)=(effect oddsratio) ; class race (ref="White")somecollege(ref="No")/param=ref; model LowBirthWgt (event="Yes")=somecollege race/ clodds=pl selection=backward; run;
Lastly, I am using Proc glimmix for the first time.
Are the groups being compared with eachother both groups who attended some college?
For the odds ratio here, do I need to substract from 1 for inverted values? For example, if the OR is .802 is that comparison 80% less likely or or 20% less?
proc glimmix data=Sashelp.Birthwgt; class race (ref="White") somecollege(ref="No"); model LowBirthWgt (event="Yes") = somecollege*race/ dist=binary; lsmeans somecollege*race / slicediff=somecollege oddsratio ilink; run;
I apologize in advance for the long post, if it's better to break up the questions in the future, please advise. I learn best by seeing examples and annotated output, which I've had difficulty finding for this. Any recommendations on resources like this are welcome. Thank you for helping out a rookie!
... View more