Skip to content
Prev 18403 / 20628 Next

Struggling (probably problem in wetware not software) with multilevel logistic regression

(reposting my response since the first version failed to include the
mailing list among the recipients)

Here?s a non-statistician?s take:

You don?t need a multilevel model for this. Given that every respondent
contributes only one data point, no one?s idiosyncracies are
overrepresented, and we can expect them to cancel each other out. Your
question can be modeled using standard logistic regression.

Since you suspect *posItem* may influence the chance of omission, you
should include it as an additional fixed effect besides *negItem*.
Something like:

mymodel <- glm(missing ~ posItem + riskItem, family = binomial, data =
mydata)

It?ll be interesting to hear whether real statisticians disagree.

Best,

J

pe 1. toukok. 2020 klo 21.38 Chris Evans (chrishold at psyctc.org) kirjoitti: