CODEWITHSUNDEEP
routputlme4linguistics
Dakota J.
Dakota J.

Reputation: 1

Interpreting output in generalized linear mixed model

I'm trying to compare the effect of instruction to different groups at different testing times. I have the following variables:

  1. Independent Variables (Learner_Type: 3 conditions-LING, NOEX, TRAD; Testing_Time: 3 conditions-T0, T1, T2)
  2. Dependent Variable (Item_Score: a rating out of 7).

Here's the model I ran:

mod.04.esl.learner.time <- 
glmer(Item_Score ~ 1 + Learner_Type*Testing_Time + (1|Part_Number),
     data=x.ESL, family=binomial)
summary(mod.04.esl.learner.time)

I get the following FIXED EFFECTS output:

    Fixed effects:
                                Estimate Std. Error z value Pr(>|z|)    
(Intercept)                       3.3836     0.3013  11.229  < 2e-16 ***
Learner_TypeNOEX                  0.2421     0.5053   0.479  0.63187    
Learner_TypeTRAD                  0.2004     0.4673   0.429  0.66807    
Testing_TimeT1                    0.5309     0.1682   3.156  0.00160 ** 
Testing_TimeT2                    0.4456     0.1650   2.700  0.00692 ** 
Learner_TypeNOEX:Testing_TimeT1   0.1136     0.2997   0.379  0.70465    
Learner_TypeTRAD:Testing_TimeT1  -0.7340     0.2595  -2.829  0.00467 ** 
Learner_TypeNOEX:Testing_TimeT2  -0.3439     0.2755  -1.249  0.21181    
Learner_TypeTRAD:Testing_TimeT2  -0.4665     0.2621  -1.780  0.07513 .  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Now, from everything I've been reading, the results should be interpreted as all rows that fall below "intercept" are being compared to "intercept", with (in my case), (INTERCEPT) being equal to MEAN RATINGS for LING at T0 (because those come first in the alphabetical order???).

So, does that mean the following:

  1. group NOEX at T1 is not sig. different than LING at T0
  2. group TRAD at T1 is sig. different than LING at T0
  3. group NOEX at T2 is not sig. different than LING at T0
  4. group TRAD at T2 is not sig. different than LING at T0

When I try to look at the results based on an interaction plot, I get a different feeling:

AJT Overall Results for ESL Groups.

Any and all help is greatly appreciated!

Upvotes: 0

Views: 2389

Answers (1)

fujiu
fujiu

Reputation: 501

At first glance, your interpretation of the model output itself makes sense to me.

One reason you are getting strange results here might be because you could be fitting the wrong kind of model. As you have said, your dependent variable is a score that I assume could theoretically range from 0 to 7 (?), making it a continuous variable. However, you are specifying a generalized linear mixed effect model with the family argument set to 'binomial', which would require a binary dependent variable (0/1, "success"/"failure"). If that's the case, then lmer() instead of glmer() might be a better choice.

Upvotes: 1

Related Questions