topic Re: Interpretting OR of continuous variables in multilevel random intercept fixed effects model in Statistical Procedures

Interpretting OR of continuous variables in multilevel random intercept fixed effects model

rvdv — Mon, 15 May 2023 14:06:55 GMT

Hi all,

When I run a multilevel multivariable logistic regression model with random intercept and fixed effects, my 'Odds Ratio Estimates' table is presented. At the bottom of this table is the following: "Effects of continuous variables are evaluated as one unit offsets from the mean. The suboption AT changes the reference value and the suboption UNIT changes the offsets."

Now, I know that when presenting continuous variables in a logistic regression model, the OR means that for every one-unit increase in your continuous variable, the odds on the dependent variable increases by factor X.

However, the line in SAS talks about "one unit offsets from the mean."

Does anyone know what is meant by this? Why does this refer to the mean? And then how do you correctly interpret this OR?

Much appreciated for any insights

Re: Interpretting OR of continuous variables in multilevel random intercept fixed effects model

jiltao — Mon, 15 May 2023 14:28:42 GMT

It means one-unit increase in X from mean to mean+1.

Thanks,

Jill

Re: Interpretting OR of continuous variables in multilevel random intercept fixed effects model

rvdv — Mon, 15 May 2023 14:43:42 GMT

@jiltao

Thank you for your response!
I understand that it means one-unit increase in X from mean to mean+1.

However, what I am struggling with is understanding how to interpret it at lower numbers.

For example: suppose my mean age is 70, the OR shows the difference in odds between 70 years and 71 years, but what does that say about 69 years?

Thanks in advance,

Roos

Re: Interpretting OR of continuous variables in multilevel random intercept fixed effects model

StatDave — Mon, 15 May 2023 16:35:51 GMT

It depends on your model. If your continuous variable, X, is only included as a linear effect in the model (such as MODEL Y=X or CLASS A; MODEL Y=A X A*X) then the odds ratio is the same regardless of the value of X at which it is computed. That is because the change in X on the log odds scale is linear. But if your model involves X nonlinearly (such as MODEL Y=X X*X), then it will change.