Could I use the Mann-Kendall Trend test for a data set of monthly sales (counts)?


If your data are collected monthly over several years, what you might need is the seasonal Kendall test - a version of the Mann-Kendall test for detecting monotonic trends in seasonal data (where seasonal refers to monthly in your case). The seasonal Kendall test runs a separate Mann-Kendall trend test on each of months you have data for and computes an overall test statistic by adding up the monthly Mann-Kendall test statistics. Any monotonic trends present should all be in the same direction (either all up or all down). If the trend is up in some months and down in other months, the results of the seasonal Kendall test will be misleading. See https://pubs.usgs.gov/twri/twri4a3/pdf/chapter12.pdf.

If your data are collected monthly for a single year, then a Mann-Kendall test might be fine.

  • $\begingroup$ Can we use this, if the data was fb likes on a user post? instead of sales data $\endgroup$ – sherna 07 Apr 7 '18 at 4:00
  • $\begingroup$ The number of likes is a count variable, so you may need to use a different method to deal with it. You could log-transform the count variable first and then analyze it the same as you would sales, but I think you'll have zero likes to contend with. $\endgroup$ – Isabella Ghement Apr 7 '18 at 4:21
  • $\begingroup$ This is just to reduce the clutter in data, isn't it? I used with the count as it is, and it was not an issue. But, the graph shows higher count values. $\endgroup$ – sherna 07 Apr 7 '18 at 10:37
  • $\begingroup$ Usually count data are analyzed by special methods (e.g., Poisson regression), as they can have features such as over-dispersion, zero-inflation, etc. $\endgroup$ – Isabella Ghement Apr 7 '18 at 17:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.