Model Code

[pthane]

Hi @jvcasillas ,

Thanks again so much for your time today. Here is the code that you asked for me to share with you:

Mood_Production = glmer( Average ~ Token_Main_Lemma_Std + FofA_Std + DELE_Std + DELE:Token_Main_Lemma_Std + FofA_Std:Token_Main_Lemma_Std + (1+FofA_Std | Group_No) + (1 | Item), data = EPT_Mood_Data, family = "binomial", control = glmerControl(optimizer = "bobyqa", optCtrl = list(maxfun=3e5)))

Gracias de nuevo!

[jvcasillas]

full = avg ~ dele + aoa + token + freq + token:freq + token:aoa + aoa:freq nested1 = avg ~ dele + aoa + token + freq + token:freq + token:aoa nested2 = avg ~ dele + aoa + token + freq + token:freq nested3 = avg ~ dele + aoa + token + freq nested4 = avg ~ dele + aoa + token nested5 = avg ~ dele + aoa nested6 = avg ~ dele nested7 = avg ~ 1

anova(full, nested1, test = "Chisq") anova(nested1, nested2, test = "Chisq")

nested = avg ~ 1 nested = avg ~ dele nested = avg ~ dele + aoa nested = avg ~ dele + aoa + token nested = avg ~ dele + aoa + token + freq nested = avg ~ dele + aoa + token + freq + token:freq nested = avg ~ dele + aoa + token + freq + token:freq + token:aoa full = avg ~ dele + aoa + token + freq + token:freq + token:aoa + aoa:freq

[jvcasillas]

2018_casillas_simonet_jphon.pdf

ex. of describing and reporting nested model comparisons in a GLMM

[pthane]

Thanks, @jvcasillas . Just one question: what do I do once I run the ANOVA? In other words, I can code away, but I'm not exactly sure how to interpret what it is that I'm doing :)…

[jvcasillas]

Check this out: http://www.bodowinter.com/uploads/1/2/9/3/129362560/bw_lme_tutorial2.pdf

start on page 11

[pthane]

Okay, I see. He does indeed write very well…I have his book too, but this looks like something separate. So, I coded the nested models. I put the FofA:Token_Freq interaction at the beginning, because I am interested in this effect. The only anova that came back as significant was that for the model that did not have DELE. When I got to FofA:Token_Freq, I compared it against the "base" model of Average ~ 1. The resultant value was insignificant, suggesting that this variable was also not significant. Correct?

Also, I ran the composite model with task as a variable. It was significant, of course, but the FofA:Token_Freq interaction was not. Hence, I really don't gain anything by reporting the tasks separately or together, right? If that's the case, I'll follow Silvia's intuitions, although I agree with you that in the case you described, a composite model is just as good if not better for purposes of publication.

If this is indeed the case, I am going to move forward without the AoA, because frankly, I just don't think it matters. I can turn in a paper with null results (in production; the comprehension is still significant) and move on, and if Jen wants, I can report the aspect data.

Thanks a bunch! Ya veremos qué sucede. PT