Paired comparison test for serially correlated data

[cassarch]

Hi,

I'm doing a paired data comparison (upwind vs. downwind), using the Wilcoxon signed rank test using hourly data.

The paired data sets are somewhat correlated because they were measured during the same hour. From text I have read, this appears to be permitted for paired data analysis. My problem is that the data from different hours may not be strictly independent (i.e. serially correlated). Is this a problem for paired comparison results, and if it is, how can I address it?

Many thanks!
Cassie

 

[terzi]

Hi cassarch,

As you mentioned, there is no trouble with the measurements done in the same hour. The paired test will handle that. Regarding the correlated data, that would require some further evaluation. Observations are never completely independent but if you think that there is a potential effect of time in your response variable, you may want to use that information to produce Blocks and account for that source of variation in your analysis. Maybe a Friedman test can be useful if that's your case.