1

I am studying a variable in three sample sets, one from healthy controls ($N=14$) and two from cancer patients (one tumor and one non-malignant; $N=22\text{--}29$). It is easy for me to compare the controls vs tumor samples with independent t-test with Welch's correction. The tumor vs non-malignant groups (22 of which are matched from the same individual, 7 are not matched) can also be compared with each other with paired t-test or with a partially overlapping t-test (Partover in R, documentation here).

But is that the best way, or can I use a mixed model for the whole data set?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Christopher Fowler
  • 11
  • 1
  • See also [t-test-for-partially-paired-and-partially-unpaired-data](https://stats.stackexchange.com/questions/25941/t-test-for-partially-paired-and-partially-unpaired-data) and posts linked from that – kjetil b halvorsen Dec 17 '18 at 11:42
  • Why would you use a paried test? Do the two cancer datasets share some of their patients? – Kodiologist Dec 17 '18 at 15:52
  • Could you show in a more concrete manner how the subjects are distributed across the three groups? The overlap between _tumor_ and _non-malignant_, is that due to one group effectively being a subset of the other (some tumors are non-malignant)? Or are there subjects who have both malignant and non-malignant tumors? – AkselA Dec 17 '18 at 19:15
  • AkselA: thanks for the question and sorry not to be clearer. The tumour biopsies contain cancerous tissue ("tumour") as well as tissue that looks normal, but probably isn't ("non-malignant"). These are sampled separately when possible. Since the controls don't have cancer, they have neither "tumour" nor "non-malignant" samples. The cases with cancer have usually samples from both tumour and non-malignant regions of the biopsies, but sometimes only one of these two regions has been sampled. Hope that clarifies the issue. – Christopher Fowler Dec 18 '18 at 07:15

0 Answers0