I have gathered infectious disease incidence data of specific years and categorized that data based on e.g. gender and vaccination status. I test whether there is any difference between the adjacent groups (years) by using Pearson's chi-squared test.

But what test(s) do I use to check whether there is any statistical difference between e.g. gender based on the combined total of all observed years? For instance, if I have 10.000 total cases and 4.500 and 5.500 were male or female, respectively, how do I test whether women are statistically more affected than men? I have now used the binomial test but am not sure if that is correct. Could I use phi coefficient too?

Edit (to clarify what I'm asking here):

Underneath the total amount of men (4705), total amount of women (4531), and the combined total of all cases (9236). I wish to know whether one of the two (in this case men are slightly more frequent) are statistically more likely to acquire an infection. What statistical test do I use for that?

enter image description here

  • 1
    $\begingroup$ How did you get the data? The design of the study is important for analysis choice. $\endgroup$ Commented Dec 21, 2018 at 15:32
  • $\begingroup$ The data is partly from peer-reviewed articles and partly (the years that were not published about) from a data request. This request was honored by the national public health institute. All data that was provided and/or published about is derived from the national surveillance register. $\endgroup$
    – Jeroen
    Commented Dec 21, 2018 at 15:40
  • $\begingroup$ So this isn't cohort data? You just grabbed a bunch of data from articles and asked some institute for all the cases of this disease? $\endgroup$ Commented Dec 21, 2018 at 15:45
  • $\begingroup$ I didn't just grab it. The data that I requested is based on years in which incidence was relatively high, but not published on. The data that I got from the articles are outbreak reports that account for >98% of the cases in those years. Years that were not included in the study had an incidence of <50 per year. So no, it isn't cohort data. It's categorized incidence data and I am trying to review all cases and determine trends and characteristics of infected individuals. $\endgroup$
    – Jeroen
    Commented Dec 21, 2018 at 16:15
  • $\begingroup$ See my answer. I don't know enough about the study design to make a really solid recommendation, but linear regression may be a fine way to go. $\endgroup$ Commented Dec 21, 2018 at 16:16

1 Answer 1


Depending on the magnitude of the count data, linear regression might be the way to go.

If the counts are relatively small, a poisson regression might be the way to go.

I'd be careful about interpretation of the results. Unless you have designed the study to make a specific kind of inference, just grabbing data from papers and putting it together could provide a lot of bias.

  • $\begingroup$ I don't think you understand what I am asking here. Linear regression for determining statistical difference between two dichotomous values? I think you're confused because I used the word 'incidence'. The data that I have are cases that were infected by a particular disease and categorized by e.g. gender and vaccination status. The question is what test I should use to determine whether there is a difference between the amount of women affected by the disease compared to men. $\endgroup$
    – Jeroen
    Commented Dec 21, 2018 at 16:19
  • $\begingroup$ I stand by my answer. You should control for the effect of year and gender at the same time. Aggregating all the years together could lead to confounding. Could you post a sample, or facsimile, of the data? I very well might be wrong, but from my understanding, controlling for gender and time is preferable. $\endgroup$ Commented Dec 21, 2018 at 16:26
  • $\begingroup$ I tested whether there is any difference between the years by using the Chi-square test (found no difference). To control whether gender makes a significant change, I calculated phi from the same contingency table (found no difference) and performed a binomial test based on the combined totals that are shown in the original post above (again, no difference found). But I'm not sure whether the tests (phi and binomial test) are correct in this case... $\endgroup$
    – Jeroen
    Commented Dec 21, 2018 at 16:33
  • $\begingroup$ Do you know how many people did not get sick? If you did, you could just use a test of proportions. $\endgroup$ Commented Dec 21, 2018 at 16:36

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.