Sorry if this is a very basic question, but I don’t have much experience with statistics.

Say I have 2 groups. Each groups have about 1000 patients each. The first group is treated by a drug in an interval of 10 years with (patients can be treated in different years independently from each other) and the second group is an untreated control group. The control group and the treated group are measured in the same exact years.

The data in both groups is a score of how well they’re doing where each patient has a score for each year (the treated patients also have scores before their treatments).

I’m thinking doing maybe a student’s t-test to find out what the correlation is between the two groups. I’m thinking an independent two-sided t-test. Do you think this would be meaningful?

I’m also thinking about using a regressional model to maybe implement the treatment years.

Could there be any other statistical methods I could use?

I’m sorry if this question is not well-formulated enough, please feel free to ask me about anything additional.

  • $\begingroup$ For me sounds ok as long as the data follows a normal distribution. Otherwise use nonparametric statistics such as Mann–Whitney U or Wilcoxon rank sum tests. $\endgroup$
    – Roman
    May 30, 2018 at 8:02
  • $\begingroup$ I think your formulation is quite clear, but it would be clearer if you could add table showing how the data look like $\endgroup$
    – fabiob
    May 30, 2018 at 8:36
  • $\begingroup$ @Jimbou: Thanks for the suggestions, I'll have a look at them. $\endgroup$
    – WoeIs
    May 30, 2018 at 8:48
  • 1
    $\begingroup$ @fabiob: Here is a sample of the data where each row is a patient: i.imgur.com/xocr7A6.jpg $\endgroup$
    – WoeIs
    May 30, 2018 at 8:49
  • $\begingroup$ thanks! do I understand your data right that patient 14 receives a treatment in 2008 (or 2009)? what about participant 10? is she receiving treatment in or before 2005? your data do not seem to be normally distributed... there are many particippants full of zeros... or are these not receiving treatment (and because of this they feel bad)? $\endgroup$
    – fabiob
    May 30, 2018 at 9:00

1 Answer 1


Certainly an independent sample t-test is a start point. But considering the complexity of the data you have I think you can improve your model a bit. Here some suggestions.

I think you could analyse this kind of data within a full factorial mixed model with group and time as factors and a random intercept. In this case a significant time:group interaction would indicate that the treatment is effective.

However in your case the time factors will have 10 levels... so if this is too much you can either treat it as a continuous variable (and then maybe have also nonlinear terms - you should plot your data as a function of time and decide on that), or restructure your data, namely you average the timepoints corresponding to pre- and after treatment.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.