0

Background

My friend did an observational experiment. She observed several colonies of animals for six months (one time per month) in the year 2014 and 2015. Some treatment to the environments of those animals were made in the year 2014 and in the year 2015, the environment was kept as natural as possible. She wants to test if there is a statistically significant difference between the sizes of colonies comparing the year 2014 and 2015.

The situation

The friend wants to use Wilcoxson's test for paired data to compare if differences in colony size among the year 2014 and 2015 are statistically significant. Each observation is the size of the colony in each month. In my opinion, the test is not appropriate as data is time series.

The plots of the results

enter image description here

The question

  1. Is Wilcoxon's test appropriate for this problem?
  2. If not, could you advise what are the appropriate methods (if possible, an example in R would be appreciated)?
  3. Are any methods of inferential statistics appropriate at all?

The data and R code for plots

library(data.table)
library(ggplot2)

col1 <- fread("
colony_size  year   month 
637          y2014  01    
609          y2014  02
582          y2014  03
902          y2014  04
945          y2014  05
954          y2014  06
640          y2015  01
615          y2015  02
585          y2015  03
922          y2015  04
1040         y2015  05
1097         y2015  06 ")



ggplot(col1, aes(month, colony_size, color = year)) +
    geom_line()  + geom_point() + theme_bw() + ggtitle("Colony 1")
GegznaV
  • 429
  • 1
  • 5
  • 17
  • The answer to your question should be found here https://stats.stackexchange.com/questions/78788/can-i-use-wilcoxon-signed-rank-test-to-test-pairs-of-data-in-a-time-series – tim May 24 '17 at 09:38
  • 1
    While the raw observations may quite dependent, the pair-differences might in some cases be considerably less dependent and then it might be reasonable. But this wouldn't be true for all situations and when there is strong serial dependence the signed rank test can be quite affected. At the least it would be useful to consider a diagnostic of serial dependence. – Glen_b May 24 '17 at 09:57

0 Answers0