How are deciles defined when you have very few records to compute it?
For ex: ntile(10) as defined by standard frameworks, over, say, [1,2,3,4] gives us [(1,1), (2,2), (3,3), (4,4)]
3
-
$\begingroup$ This isn't, perchance, because you are trying to implement the Hosmer-Lemeshow test, is it? $\endgroup$– AdamOCommented Apr 10, 2018 at 21:09
-
$\begingroup$ The Hyndeman and Fan paper gives a theoretical/historical overview of the 9 methods that Peter alludes to. $\endgroup$– AdamOCommented Apr 10, 2018 at 21:18
-
$\begingroup$ I don't understand how this is unclear at all. $\endgroup$– Peter FlomCommented Apr 11, 2018 at 12:07
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Personally, I'd say that deciles are silly if you have N < about 40 and utterly silly if N < 10. R offers 9 different variations of quantiles and deciles gives different results with different types for (1,2,3,4). E.g. for type 1 it gives:
1 1 2 2 2 3 3 4 4
but for type 3 it gives
1 1 1 2 2 2 3 3 4
is the third decile of this 1 or 2? Who knows? Who cares? Why ask? It's a case where taking deciles gives less information with more numbers!
answered Apr 10, 2018 at 21:02