topic Re: Observational study: Large overall sample but with 70:30 group ratio – T-test or Wilcoxon Rank S in Statistical Procedures

Observational study: Large overall sample but with 70:30 group ratio – T-test or Wilcoxon Rank Sum?

PharmlyDoc — Thu, 24 Jun 2021 00:23:22 GMT

This is a retrospective observational study looking at a binary outcome (1=yes, 2=no).

Overall sample of 100,000 patients.

70% of patients are in the no group and 30% in the yes group. Am I still okay to use chi square tests to compare categorical variables as long as table cells contain more than 5 observations?

For univariable statistics comparing continuous variables between the two groups - for example, age in years - am I okay to use a two-sample T-test?

Also, my distribution for age is somewhat bimodal

Re: Observational study: Large overall sample but with 70:30 group ratio – T-test or Wilcoxon Rank S

Reeza — Thu, 24 Jun 2021 01:23:27 GMT

Since it's almost 2:1 why not do a case control matching instead?

Re: Observational study: Large overall sample but with 70:30 group ratio – T-test or Wilcoxon Rank S

Ksharp — Thu, 24 Jun 2021 12:03:50 GMT

If it was following NORMAL distribution ,then try ttest ,otherwise wilcoxon

Re: Observational study: Large overall sample but with 70:30 group ratio – T-test or Wilcoxon Rank S

SteveDenham — Thu, 24 Jun 2021 19:20:30 GMT

Or PROC FMM. That plot looks like a classic mixture of a normal and a gamma distribution.

SteveDenham

Re: Observational study: Large overall sample but with 70:30 group ratio – T-test or Wilcoxon Rank S

Ksharp — Fri, 25 Jun 2021 12:07:56 GMT

It can compare the difference of mean between two groups , like proc ttest ?

Re: Observational study: Large overall sample but with 70:30 group ratio – T-test or Wilcoxon Rank S

SteveDenham — Fri, 25 Jun 2021 12:48:15 GMT

No, but what it can do is a Bayesian analysis, where the mean, standard deviation and credible interval bounds of the posterior distributions are reported.

Granted that this approach may be using a bazooka to swat a fly...

SteveDenham