1

We are polling respondents every month and asking a couple of questions about the number of times they eat certain foods. We have significant loss to follow up every month but generally retain 20% - 30%. For the participants who stay engaged and respond across the months, I understand we use the paired t-test.
My question is what test would be appropriate to use to compare between two months say for example comparing a june & july sample which contains some folks in june who dropped out in july, some folks who answered for both june and july, and some new randomly selected folks who were added in july? I've read through previous posts but cant get whether using the paired test is really appropriate. THANKS!

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
bolo
  • 11
  • 1
  • Do you truly want to know the effect for the contrast between arbitrary months? Aren't you rather interested in something like a general effect across time? – jona Aug 30 '15 at 15:30
  • 1
    This question is relevant: http://stats.stackexchange.com/questions/25941/t-test-for-partially-paired-and-partially-unpaired-data especially answer by Elvis – kjetil b halvorsen Aug 30 '15 at 15:32
  • Thanks Jona and Kjetil, my interest for now is in comparing month to month changes, june and july, july and august and so on. Thanks for the link to the previous post. As a stats novice--would it be advisable to use the unpaired t, it being the more conservative test? – bolo Aug 30 '15 at 16:25
  • No, you should not use the unpaired $t$ as that will lose information, and it assumes independence, which is not fulfilled! Try to use the method proposed in the answer by @Elvis in my link above. – kjetil b halvorsen Sep 01 '15 at 10:13

0 Answers0