nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Problem with Parameter Estimates in a Regression Interaction in mixed models with random effects

Dalia

Hi all, I am looking for a solution in understanding the interaction between ad category and compulsive buying score. I have used a linear mixed model with fixed effects and random effects statements ("Stimuli ID" and "Subject ID"). While in regression without random effects statements you have an option to use the Extended Estimates option to get the full parameter estimates table, there is no option of this in the case of mixed effects models. How can I see the regression coefficient and significance level of the category, which is hidden in my parameter estimates?

Accepted answers

Mark_Bailey

You see the parameter estimates for all but the 'last' level due to the way JMP parameterizes categorical effects in your linear model. The order for the levels is (generally) alphabetic or numeric. You can change this order and impose your own order by defining the Value Ordering column property. This way you can make another level that is not statistically significant or practically important the last level. Select the data column with the ad category, select Cols > Column Info, click Column Properties, select Value Ordering, and the select and move the last level up in the list of values.

All comments

Mark_Bailey

Dalia

Thank you so much, everything worked perfectly!

Dalia

I have one clarification question. Do I understand correctly that values in the parameter estimate table, such as p-value and coefficients are reported to be compared to the reference category? For instance, if I have three interaction Ad1*CBS, Ad2CBS and Ad3CBS, so the p-value in the parameter estimates table just shows that interaction Ad1*CBS is significantly different from the reference category Ad0CBS (which is not exposed on the parameter estimates level)? So if I recode by value ordering and get the values for my reference category, it does not say much more than this interaction is significantly different from the one that I set to 0 now?

If I want to understand, whether the CBS predicts the dependent variable within each of the categories, is that correct then that it is not enough to study the parameter estimates table and I need to run the dummy-coded interaction terms regression: http://www.restore.ac.uk/srme/www/fac/soc/wie/research-new/srme/modules/mod3/11/index.html ?

Sorry for my confusion, I usually work with categorical data and this case is still a bit puzzling me.

Thank you,

Dalia

Mark_Bailey

The tests of the parameter estimates are against the null hypothesis that the parameter is zero. That is separate and distinct from the model parameterization for categorical factors, which requires that the estimates sum to zero.