0

I have a table where columns denote customer, row-supplier, cell- total amount of items, sold by supplier to customer. I have a zero hypothesis(H0) about random choise of customers and H1 hypothesis(customers perefer one supplier to others). I would like to calculate p-value for rejecting H0 and accepting H1 hypothesis. When I would like to exclude most prefered supplier and repeat this procedure.

The most dificult case for me is to compare one suppliers against all other with respect to block data. In case of two suppliers I could aplly Wilcoxcon paired sign-rank test. Is there a generalisation of such test for case of more then two connected samples?

A. Bykov
  • 1
  • 3

1 Answers1

1

Friedman test is what you are looking for, customer being your blocking factor. The ranks are calculated within each blocking factor, and only then compared to average ranks.

See ?friedman.testin R or good old Wikipedia at https://en.wikipedia.org/wiki/Friedman_test for more explanation.

Knarpie
  • 1,522
  • 9
  • 22
  • Friedman test is good to say that there are suppliers which are different(distingushable in positive or negative way by customers) from other suppliers. But what would I like to see: find top prefered suppliers and ensure that rest supliers indistingishable by customers(I can do the last thing by Friedman test) – A. Bykov Oct 27 '17 at 10:02
  • @A.Bykov Kindly explain "ensure that rest supliers indistingishable by customers"? – Knarpie Oct 27 '17 at 10:04
  • "ensure that rest supliers indistingishable by customers"- no statistical difference. Customers randomly choose betwen rest suppliers – A. Bykov Oct 27 '17 at 10:06
  • @A.Bykov Why do you want to ensure this? Even when your data still exhibits a weak signal of customer preference for them? – Knarpie Oct 27 '17 at 10:23
  • @Knarpie, sorry I didn't formulate my questions acurately enough. By "ensure" I ment comparing with zero hypothesis about random choise. I need to "ensure" to retrieve an important statistics for me in form of p-value. I also reformulate my task- see it in the first post. – A. Bykov Oct 27 '17 at 11:34
  • There is an answer here https://stats.stackexchange.com/questions/10309/friedmans-test-and-post-hoc-analysis?rq=1 which might help although as far as i can see it is about pair-wise contrasts whereas I think the OP wants a more general contrast – mdewey Oct 27 '17 at 13:08