First time poster...
Hoping someone can provide some insights into the following problem.
I am analyzing a large data set of subjects. A small fraction of this population exhibits a desirable behavior (say DB). And it can be measured on a continuous numerical scale, say 0 to 10.
Close to 99% this population exhibits no DB, so their DB values are zero. The rest, about 1% do have positive DB values. However, in any new data set of this population, it is not known which of the about 1% will exhibit positive DBs.
I randomly divided one such data set into two calling them baseline, and treatment. I treated the treatment dataset with a treatment and then measured DB for both treatment and baseline groups.
How can I evaluate if the treatment had any impact on -
1. the proportion of subjects exhibiting DB.
2. The level of DB exhibited by those who did respond positively.
I have tried the t-test and chi-squared tests for this (on the whole data set and not on just the ones that respond positively). However, since almost 99% of the values are zero, I am not too confident of the results.