So when using a biserial correlation you are able to test for an association between a continuous DV and categorical IV but can that be done with multiple IVs with a dichotomous DV similarly to a binary logistic regression or can you use the binary logistic regression to test for an association.
Logit models / logistic regression models are the standard tool for predicting a dichotomous outcome variable (1/0) using one or more independent variables.
If you want a bivariate test, with one outcome and just one independent variable, you could use a difference of proportions test or a difference of means test, depending on whether your independent variable is dichotomous or numeric.
Timothy Fraser would I be wrong if I used a binomial regression to test for an association between a dichotomous DV and multiple IVs in a similar way I would test for an association in a chi-squared or is it specifically to look at a predictive-probability association among the variables?
Logit/Binomial/logistic regression models (all the same thing) are the ideal way to test this kind of thing.
If you have a dichotomous outcome, like, did you vote or not (1/0), but you have multiple possible explanations (eg. person's political party, age, income, etc.), you can get the estimated log-odds of each independent variable on the outcome, controlling for all others.
You can assess the fit of your model using Pseuso-R2, which ranges from 0 to 1, where 0 mean a bad model and 1 means pretty perfect explanatory power. McFadden's R2 is often used. Nagelkerke's R2 works great too.
You can also use a logit model to produce predicted probabilities given a change in one variable, holding all others at their means/modes.
Based on the question, Timothy Fraser is quite correct. Logistic regression is the thing to do. Best, D. Booth
Use logit or probit regression but in general, logit/logistic regression is preferred as a standard analytical tool.
Logistic regression is the exactly appropriate test for that data
- Badges
- Science topic