# multi-response outcome: multivariate or univariate regression?

I have a dependent variable which represents participants' responses on 20 words, coded as 1 = correct , 0 = incorrect

Question: I am keeping the regression as univariate for now (as far as dependent variables).

But after more readings I am confused on whether I should transfer the words into 20 dependent variables to have a multivariate regression?

• If a participant's response on one word is affected by his/her responses on all or some of the other words, then you would be justified in analyzing simultaneously the responses to all words in the same model. But if the response to one word has nothing to do with the responses to any of the remaining words, then analyzing each word-specific response on its own seems reasonable. – Isabella Ghement Dec 14 '18 at 3:05
• Thank you Isabella. The participant's responses are not interdependent, but, to address my research hypothesis, the more words the better, that's why I have 20 words. I have a row for each word, hence, every participant with 20 responses, and normally the regression is considering the pretest posttest difference by each word ( as you explained in my former post). So multivariate does not apply here, what do you think plz? – Acer acer Dec 14 '18 at 9:05
• What is your research hypothesis? – Isabella Ghement Dec 14 '18 at 18:48
• For instance, Group A will demonstrate more vocabulary gains relative to Group B, with a significant interaction between these two groups and the control (C), i.e. no learning in control. – Acer acer Dec 14 '18 at 19:20
• There are many ways to analyze your data, but from what you indicated I wonder if you should use some sort of mixed effects binomial regression? In other words, consider the pre-test number of correct guesses (out of 20) and the post-test number of correct guesses (out of 20) as your outcome values - measured at two distinct occasions. – Isabella Ghement Dec 15 '18 at 17:31