1

In cohort studies of change over time (for instance change in repeatedly measured weight after a dietary intervention, using linear mixed models or GEE), what is the detriment of including individuals with only one data-point?

Does the answer change depending on whether GEE or GLMM is used?

bobmcpop
  • 1,283
  • +1 Not sure if coincidence or if you are following-up in a new thread for clarity's sake, but I am wondering this exact same question in the comments here: https://stats.stackexchange.com/questions/358231/puzzling-predicted-values-in-generalized-multilevel-model – Mark White Jul 24 '18 at 01:15
  • Also, shouldn't the title read more than one, per the question in the body of the post? – Mark White Jul 24 '18 at 02:45
  • What a coincidence, I’ll keep an eye out for your answer. And yes thanks for the correction – bobmcpop Jul 24 '18 at 11:04
  • That's kind of the definition of "repeated measures". Also, how would you look at change if you only have one measurement? – A. Donda Jul 24 '18 at 11:46
  • 1
    Repeated measures doesn’t equate, sadly, to participants having complete or valid measurements each time. The question is partly what to do with these individuals. I get that conceptually they shouldn’t be used in GLMM, but GEE is “population change.” Also I wondered more so the harm of including them, because many people are including single data points. – bobmcpop Jul 24 '18 at 12:22
  • 1
    @A.Donda imagine we send texts to people that are various questions. A lot of people only answer 1, but other people answer 2 or 3. How do we model the dependency between the answers of those people that answered 2 or 3 as repeated measures, when we still have lots of people that have single observations? – Mark White Jul 24 '18 at 12:46
  • Mark, I got some answers from here https://discourse.datamethods.org/t/what-are-the-problems-if-any-of-including-individuals-with-only-one-data-point-in-studies-of-change-over-time-in-repeated-measure-data/192/8 if you need more responses it may be worth asking there – bobmcpop Jul 26 '18 at 13:36
  • 1
    Thanks. I figured out what was happening in my multilevel model case. I will try to answer your question (as a way of making sure I understand the answer to my question) this evening! – Mark White Jul 26 '18 at 15:16

0 Answers0