Hello, I tried doing as you suggested (added a cubic term as a covariate), it did not help. I am using the linear mixed models procedure and I am not back translating. I attach the output in which I conduct the normal procedure first and then go on with the cubic term. The covariate is indeed not normal, I used LN function which normalized the distribution but it also gives the same combination of significant/insignificant results between tables.

Any other ideas? I'm pretty confused.

------------------------------

Karolina Ziembowicz

------------------------------

Original Message:

Sent: Mon November 08, 2021 04:06 PM

From: Jon Peck

Subject: Linear mixed models - interpretation of results

I am not sure which procedure you are running. Are you backtranslating table names from another language?

But if you get different significance results as a factor from as a covariate, that might suggest a nonlinearity, so you might want to try adding a quadratic term as a covariate or fitting a cubic spline, which you could create with the STATS SPLINES extension command.

--

Original Message:

Sent: 11/8/2021 11:50:00 AM

From: Karolina Ziembowicz

Subject: RE: Linear mixed models - interpretation of results

Ok, I get it. Thanks for the nice clarification. So - are you saying I should not worry that the estimate and t-test are insignificant in the "tests of fixed effects" table? Should I report the value of F and ignore the "estimates of fixed effects" table?

------------------------------

Karolina Ziembowicz

Original Message:

Sent: Mon November 08, 2021 11:24 AM

From: Jon Peck

Subject: Linear mixed models - interpretation of results

If you treat a continuous variable as a factor, you are giving it the number of degrees of freedom as the number of levels (ignoring the constant term). So that allows for a nonlinear relationship between the presumed factor and the dependent variable while as a covariate it would only pick up a linear relationship with one degree of freedom.

In other words, treating it as a factor is equivalent to creating a dummy variable for each level.

--

Original Message:

Sent: 11/8/2021 6:32:00 AM

From: Karolina Ziembowicz

Subject: Linear mixed models - interpretation of results

Hello, I have a mixed model wiith one categorical and one continuous predictor. When I input the continuous one as a fixed effect, both the "tests of fixed effects" and "estimates of fixed effects" table gives significant results for the continuous variable. When I input it as a covariate, suddenly there is a discrepancy between the tables - in "tests of fixed effects" it remains significant, while in "estimates of fixed effects" it becomes insignificant.

This is strange for me. If a continuous variable cannot be a fixed factor (as I understood from reading the web) - why do these results differ in both tables??

Also - there is one more difference between these two analyses - the categorical variable's estimate changes the sign. What statistics should I report to follow APA style?

------------------------------

Karolina Ziembowicz

------------------------------