Dear Professors and Researchers,
I am currently a third-year psychology undergraduate preparing for my graduate thesis. My proposed study investigates the impact of video game type (two groups) on prosocial behaviour, measured as a binary outcome (willing vs. unwilling to volunteer).
Initially, I planned to analyse the data using binary logistic regression. However, after researching, I found that the use of logistic regression for causal inference is debated—while some sources suggest it may allow for inferring causation, some against it.
I would greatly appreciate any guidance or recommendations regarding appropriate statistical approaches that can validly assess the causal impact of a categorical independent variable on a binary outcome, especially in the context of an experimental design. I have been stuck on this issue for over nine hours and would be truly grateful for your insights.
Thank you in advance for your time and support!
The question of whether you can infer causality has more to do with your research design (i.e., is it a randomized controlled trial with random assignment to the two experimental groups?) than with the statistical technique used to analyze potential associations in the data.
Unless you have covariates that you need to control, there is no reason to use logistic regression when you have a single, dichotomous predictor variable.. Instead, a t-test would be enough.
If I understand well, we do have frequencies and not actual measurements; that is, we know how many participants use 'type-1' video game and do not want to volunteer, we also know, how many of them use 'type-1" and do want to volunteer. If so, we cannot execute a t-test, but we can rely on Fischer's exact test to see if there is an association between the type of the video game and the intention to volunteer.
Christian Geiser Good Afternoon Dr.,
Thank you very much for your reply. I fully understand your point that causal inference depends primarily on research design (e.g., random assignment in a controlled trial) rather than the statistical method used.
In my study, there is an experimental manipulation, so according to my university’s guidelines, I would normally be expected to frame hypotheses in terms of an “effect.” However, because my dependent variable is binary (whether participants agree to volunteer for a future study by leaving their name and phone number), I analyzed the data using binary logistic regression. When reviewing past journals with similar manipulations, I noticed that many researchers phrased their hypotheses in terms of a “relationship” rather than an “effect,” which left me uncertain about what would be considered most appropriate in my case.
Thank you again for your helpful guidance. Wish you have a wonderful day.
David L Morgan David L Morgan
Good Afternoon Dr and Professor, Thank you very much for your advice. I will certainly take it into consideration. I am truly grateful for your guidance and support in helping me towards my graduation.
- Badges
- Science topic
- Similar topics
- Biological Science
- Neuroscience