3
$\begingroup$

I would like to find out why is it necessary and key to run the chi-square test before the multinomial logistic regression? I am reading someone's unpublished work and it says the Chi square test is initially performed to confirm which variables to run the multinomial logistic regression model, what does this imply and why is it necessary? Does it mean only those with a significant association are assessed further through multinomial or binary logistic (since I am using the two models)? What is the link between association (significant and non significant) and regression?

I understand the role of the chi-square test is to test the association between the dependent and the independent variables (socio-demographic variables in my case). I am not sure the link about the chi square and regression and why you need to perform the two to identify the factors that affect a phenomena under study? The two papers first did a chi square test before the multinomial logistic regression.

$\endgroup$

1 Answer 1

2
$\begingroup$

It isn't necessary. I don't think it's even recommended. This sounds like some sort of attempt to do bivariate screening, which is definitely not recommended.

Building a regression model (of any kind) is partly an art and partly a science, but this isn't the way to do it.

$\endgroup$
3
  • $\begingroup$ thanks for the feedback. I must admit however that I do not understand the response at all (beginner). I am trying to establish or identify factors that affect my study phenomena, by assessing socio demographic factors. Knowing the association through the chi-square is helpful in helping to answer this question. And so is regression analysis? But they all focus on the same thing, the relationship/association or what is the difference between the two and the role in my case? $\endgroup$ Commented Mar 25 at 17:24
  • 1
    $\begingroup$ Regression analysis sets one variable as the dependent variable and looks at how the other (independent) variables are associated with it. You can have many independent variables and you get regression coeficients, p value, confidence intervals, and so on. Chi square (usually) looks at only two variables and does not set one as dependent. $\endgroup$
    – Peter Flom
    Commented Mar 25 at 17:53
  • 1
    $\begingroup$ This clarified it all. I will do some further studies. $\endgroup$ Commented Mar 25 at 20:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.