can we statistically control confounding variables using tests other than ANCOVA or regression?
specially in variables that violate parametric assumptions, for example if i am studying the occurrence of GI bleed (DV) between two groups (IDV), and i think that use of aspirin (Confounding variable) before data collection may affect the results, can i examine the difference in GI bleed among the entire sample between aspirin users and nonusers and & if there was no statistically significant differences this might mean that its not a confounding variable?
Dear Elham,
Sure. In general, ANCOVA and regression belong to a broader concept of general linear model (GLM), which allow to control confounding variables: both categorical and linear (ordinal factors can be substituted with dummy variables). Moreover, nonlinear regression (e.g. logistic), survival analyses, etc. also allow for such adjusting.
In your specific situation - if I understand it correctly - you may include your groups as well as the use of aspirin (yes/no variable I guess) into two-way ANOVA (without interactions - if you assume the same effect of groups in both aspirin users and non-users). If the effect of aspirin use is non-significant you may conclude no effect of aspirin use in your final effect and exclude it from the model. However, in some situation you may a priori decide to adjust the model for "potential" confounders such as sex, age, other sociodemografic variables that may be of interest in your field.
Good luck!
Michal
To add to Micheal's comment there are forms of regression which make fewer assumptions and might suit you problem. I can't speak to your question but I recently used quantile regression in a area where a generalised linear model would not have been appropriate.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics