In my analysis, I have 4 groups:

FLU-P (symptoms with virus)

FLU-N (symptoms without virus)

HCY-P (healthy with virus)

HCY-N (healthy without virus)

I am looking at the level of antibody (dichotomous: normal/ deficient) in these 4 groups....so I ran a Chi-square test (FLU-P vs FLU-N vs HCY-P vs HCY-N).

But I would like to do further comparisons, such as:

FLU-P & FLU-N vs HCY-N or


my 2 questions are: 1. am I allow to regroup groups & omit group like what I have just described in an analysis? 2. is chi-square still the right test to use? if not, any suggestion?


  • $\begingroup$ Have you thought about running a logistic regression? $\endgroup$ Commented Mar 10, 2015 at 5:18
  • $\begingroup$ There's really not a "the"; multiple different analyses might be appropriate, and "allowed" isn't really the right word either -- we're not the stats police. The question is more one of the properties a given analysis has, and whether that does what you want. $\endgroup$
    – Glen_b
    Commented Mar 10, 2015 at 7:06
  • $\begingroup$ Thanks StatsStudent. I am combining 2 of the groups and comparing this to only one of the remaining ones, so would doing this in logistic regression solve my query as to whether it's alright to omit one of the groups in the analysis. Thanks Glen_b for your comments. $\endgroup$
    – Marcy
    Commented Mar 16, 2015 at 0:08

1 Answer 1


If I understand your question correctly, your first analysis was a 4x2 contingency table with 4 independent groups of subjects who were rated on a dichotomous feature (normal vs. deficient antibody level). That means you are interested in whether being in one of these 4 groups is independent of having a normal or deficient antibody level. Your null hypothesis would be that the proportion of subjects with a deficient antibody level is equal across the 4 groups. In that case a Chi square test can be the right thing to do.

However, depending on your theory it might make sense to do stratification for the factor Flu vs. Healthy (or virus vs. no virus). Basically this means splitting the data into two 2x2 tables and thereby controlling for the confounding with that factor. Edit: This seems to be a nice online tool for that.

If your expected values are rather small, you might also want to have a look at Fisher's exact test instead of Chi square.

Whether you are "allowed" to regroup and omit groups really depends on the theory behind your analysis. I don't think I can confidently answer this for you.

  • $\begingroup$ Thanks Tobias. I shall look into the tool you mentioned. $\endgroup$
    – Marcy
    Commented Mar 16, 2015 at 0:12
  • $\begingroup$ If the answer was helpful feel free to upvote it, or even mark it as correct. :) $\endgroup$ Commented Mar 17, 2015 at 11:21

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.