# which statistical t-test to use when having data from different demographies?

I would like to know whether Blood Pressure is significant in contributing to a diseaseHypo. I have data from three different continents(let's say A,B,C) and each with groups diseased and not_diseased. I have attempted to do t-test in R t.test() and wondering whether I would need to assume that there is a equal variance or not between diseased(diseased_A + diseased_B + diseased_C) and not_diseased(not_diseased_A + not_diseased_B + not_diseased_C) of combined population.

Any views in this scenario please ?

Thanks.

• Do you wonder if BP contributes to disease, or if BP differs by disease? These are not the same. – gung Jun 4 '15 at 12:53
• I would like to see whether BP contributes to the disease. Should I consider the sample data from different demographies separately or together, what are the impacts and what assumptions needs to be made ? What should I assume on the variance on the combined data, equal or unequal?? – Prradep Jun 4 '15 at 14:29
• If there might be differences among continents in the relation of Blood Pressure to diseaseHypo, you need to consider continent as a factor in your analysis: ANOVA instead of a simple t-test, or (if disease presence is clearly a yes/no classification) a logistic regression of diseaseHypo against Blood Pressure, continent, and any other variables of interest. – EdM Jun 4 '15 at 19:41
• Yeah, that's what my plan is. Before that, I want to choose variables from list of 100+ variables which have significant contribution to the disease. For this identification, I am doing t-test between each variable and disease status. Now, how do I consider the variance equal or unequal ? I have no prior information on variance. In another words, to use Welch test or student t-test ? – Prradep Jun 4 '15 at 19:55
• The big issue here is not which test to use; based on your latest comment, it's that choosing individual variables for later analysis in this way is not a good approach to this problem. There is a very large literature on the problem of variable selection; follow the feature-selection and similar tags on this web site, and study references like An Introduction to Statistical Learning. – EdM Jun 8 '15 at 13:52