Yes, I know it's a big design flaw, but I'm analyzing data from the same sample collected over time (change in mean scores), but data are not matched by ID. What are my options? Will using ANOVA with time as the grouping variable with dependent samples only effect power (creating last of power) or also other statistical estimates such as effect size or F? The between subjects effects are really within subjects, but the data can't be fixed. Any analysis and reporting advice? Thanks!
Asked
Active
Viewed 242 times
1
-
Need more information about data in order to help. – Carl Sep 24 '16 at 21:50
-
There are 3 time measurements, each 6 months apart, on one group that are not matched over time. So, time is categorical as pre-post-followup. The dependent variable is measured on a 5 point scale. – Lisa Sep 25 '16 at 18:38
-
See https://stats.stackexchange.com/questions/25941/t-test-for-partially-paired-and-partially-unpaired-data (and a lot of posts linked on that page) – kjetil b halvorsen Jun 30 '19 at 16:52
1 Answers
1
Difficult question. I would try simple things first, like the Wilcoxon-Mann-Whitney. However, caution because treatment of ties needs correction, see link Kolmogorov-Smirnov vs Mann-Whitney U When There Are Ties. That would allow one to check if {A,B} {B,C} and {A,C} suggest significant differences. Using ANOVA might be tricky, as doing regression on three factors, {A,B,C} that are not continuous variables might be hard to interpret.
