Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Proportions - numerous samples

    Hello,

    I have following set-up in my experiment.

    - 1 control group
    - three treatment groups
    - Individuals in all treatments are asked to choose cost-efficient contracts. Therefore, the relevant/dependent variable is binary 1=success; 0=failure.

    1)
    Now I want to compare if "on average" the treatments exhibit different proportions with regards to successes/ correct choices, i.e. (number of successes/total number of choices in treatment)=proportion of successes.

    I also have some hypotheses, e.g. that the proportion in treatment 1 should be higher than in the control group etc. so 1-tailed tests should be run as well.
    --> I have looked in the forum and in the web, but I don't find a satisfactory answer on how to run a test that tells me whether the proportions of successes between the samples are statistically different. Moreover, before my experiment I ran the power/sampsi command in order to determine what sample size I need. However, my result proportions are closer together than I assumed and hence, if I run the power test now, my sample (for each treatment) is not large enough. So when I run sampsi p1 p2, n1 n2 onesided I get a maximum power of 0.56. What test can I use in stata to conduct one-sided tests to see if the proportions are different on a statistical level (and maybe control for my control variables)


    2)
    Moreover, I have several control variables, and I wanted to run a logit/probit in order to see, which of the factors have descriptive power with the binary variable being the dependent.
    - If I include the treatments as variables as well, i.e. dummy variable1 = 1 if control group, dummy variable2 = 1 if treatment 1; dummy variable3 =1 if treatment 2, etc. would this also be a test to see whether my treatment effects are significant?
    - Another approach I saw was using "margins", dydx; how I understand it, this should yield the same results as the normal logit/probit but ignore the constant terms.

    Thank you very much in advance for your help and best regards,
    Benjamin



  • #2
    Welcome to Statalist. This forum is for test messages. Please ask in the general forum. It would be useful if you could read the FAQ first, especially section 12.

    Comment

    Working...
    X