Interpreting results from generalized linear mixed model

mercator88

I find the output from the Generalized Linear Mixed Model part of the Mixed Models module difficult to interpret. I'm trying to predict acute kidney injury ("AKI") by two fixed effects: study phase ("Phase," either 2 or 3) and mortality risk group ("Color," either green, yellow, or red) with intensive care units ("ICUID"). being modeled as a random effect. The link function is logit.

Here is my output:

I have a number of questions:

1. If the phase is either = 2 or 3, does phase(1) refer to Phase 2 as compared to Phase 3, or the opposite?
2. How does Jasp map green, yellow, red to Color (1) and Color (2)?
3. I'm unsure what the Note refers to as "contrast coding"?
4. When I look at the marginal means (not shown here), it doesn't look as if phase (1) is statistically different from phase (2). Yet the fixed effects term for phase is significant. I'm confused.

I greatly appreciate anyone who can shed some light on this for me.

Thanks!

Andrew

EJ

I've asked the experts!

Cheers,

E.J.

mercator88

Thanks! I'm at a standstill until I get guidance on the above.

František

Hi Andrew,

The contrast coding can be indeed a bit confusing. We added a more detailed explanation to the help file that summarizes what are the differences and reasons for using the sum contrasts (where the fixed effects are not readily interpretable vs. the dummy coding--default in R):

The analysis uses sum contrast encoding for categorical (nominal and ordinal) predictors (R uses dummy encoding by default). This scheme is used for better interpretability of models with interactions. However, the fixed and random effects estimates will differ from those obtained from R with default settings. We advise using the 'Estimated marginal means' section for obtaining mean estimates at individual factor levels. For comparing the mean estimates, use the contrasts option.

Let me know if this explains the issue!

Frantisek

mercator88

Frantisek,

I'm still a bit confused. The estimate I'm getting using the contrasts option is exactly the same as the marginal mean for Phase (1). Maybe I'm doing the contrasts wrong; I set the contrasts to 0 for Phase = 0 and 1 for Phase = 1.

I didn't know there was a help file. How do I access it?

František

Dear Andrew,

I'm slightly confused about your question now. When you have a factor with two levels, the sum contrast creates (-1, 1) coding for the levels (i.e., contr.sum(2)). In other words, the fixed effect estimate corresponds to half the difference. You can find more detailed information about sum contrasts e.g., here https://rpubs.com/monajhzhu/608609 (Linear Models with Sum Contrast).

The JASP help file can be opened by clicking the "i" icon:

Which opens the corresponding pop-up window.

Cheers,

Frantisek

mercator88

Frantisek,

Thanks for directing me to the Help pages.

I apologize for the confusion. My contrast was (0, 1) not (-1, 1), and here is what I got:

 

Estimate SE df Lower Upper z p†

Contrast 1 0.136 0.045 inf. 0.048 0.224 3.039 0.002

Is it correct to assume an odds ratio of exp(0.136)?

Thanks!!!

Andrew

František

Hi Andrew,

You can get OR from the contrasts only if you don't use the "Outcome scale" option. Otherwise, the contrast is a difference in the probabilities of those two outcomes.

Glad I could help,

Frantisek