can I use ANOVA to compare number of days of asthma attacks per month followed for six months between two groups (group 1 and group 2). so we have a 2x6 table in rows we have the months (1st, 2nd, 3rd, .....sixth) and in column there are the groups (group 1 and group 2). each cell contain the total number of attacks of asthma per month (please see the attached table table

  • 1
    $\begingroup$ One problem is that the assumption of independence would not seem to be tenable. Secondly "days of asthma attacks" sounds like counts (each member of each group presumably having a 0/1 outcome on each day), which implies equality of variance is unlikely and by the look og it, you also have a problem with 0's (Red Zone/AAIG), so I don't think you can treat it as homoskedastic or normally distributed. $\endgroup$
    – Glen_b
    Commented Jan 22, 2016 at 9:52

1 Answer 1


Expanding on @glenb 's excellent comments:

1) Independence: It seems like you have the same people measured over multiple months. That would violate independence because the number of attacks that a person has in month 2 is related to the number he/she had in month 1.

2) Number of attacks is a count. You can't have 1/2 an attack or negative attacks. This isn't much of an issue when the number of attacks is large and varied, but it's a problem for the red zone.

3) 0's. Glen is right again that you have a problem here.

So, my answer is "no" and my suggestion is that you should probably use a nonlinear multilevel model with a zero inflated link (either Poisson or negative binomial).

  • $\begingroup$ thanks so much for your insight answer. could you please tell me how can I use a nonlinear multilevel model with a zero inflated link (either Poisson or negative binomial). to test the difference between the two study groups over the six months? $\endgroup$ Commented Jan 23, 2016 at 13:54
  • 1
    $\begingroup$ You would need a course or an entire text to learn those. $\endgroup$
    – Peter Flom
    Commented Jan 23, 2016 at 15:44

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.