Linear Mixed Models

gjcampitelli

Hi JASP community,

I've just installed the 0.13 version and I am using the Mixed Models module.

Anybody can help me with the following issues?

The plots part does not show the variables to transfer to the plot boxes. It does not seem to be working.

Moreover, I would like to be able to do the following:

1. Analysis without random effects: Of course, this will be equivalent to linear model, but I would like to get the AIC, etc., in order to compare with models with random effects.
2. It seems random intercept is compulsory. That is, I cannot run a model with random slopes and no random intercept.

Of course, as always, grateful for your work on improving this great software.

Cheers,

Guillermo

František

Hi Guillermo,

I'm happy to hear that you are intersted in the new Mixed Models analysis.

The plots support only factor variables. I suspect that you have only continuous predictors in your model - these are not shown in the variable box.

You can fit a linear regression using the "Regressions" analysis but you are right that it does not provide fit statistics. The "Mixed Models" analysis provides fit statistics under the "Model summary" checkbox within the "Options" tab, however, it is not possible to fit models without random effects yet. We are currently working on adding this option. We will also incorporate the possibily to omit the random intercept.

Thanks for the suggestion, cheers,

František

gjcampitelli

Hi František,

Thanks for the clarification regarding the plots. Indeed, I used continuous variables. I think the "Visual Modelling" module allows to produce the graphs I wanted.

I tell you my philosophy for teaching multilevel modelling that may or may not be useful for the development of this module:

I first run a multilevel analysis with no fixed effect variable (only the random effect grouping factor). This is equivalent to a one-way ANOVA. I calculate the AIC or any other model fit measure. This is a very useful null model.

I also run another null model, which is using a fixed effect variable and no random effect grouping factor. This is equivalent to regression, but I want to calculate AIC.

So, I have two null models. More complex models are justified only if they provide better fit than these models.

Then I add the random effect grouping factor and run random intercept only, then random slope only, then both random intercept and random slope.

Thank you for your attention, and more importantly, for developing this module!

Best wishes,

Guillermo