0

In R the default contrasts used for ordered factors is Polynomial contrasts. However, this link says:

This type of coding system should be used only with an ordinal variable in which the levels are equally spaced.

In real life, we often have ordered factors, such as income low medium high, education, high school, bachelor, master, phd. BUT these levels almost never equally spaced.

Is it wrong for R to use Polynomial contrasts as default coding for ordered factors?

Haitao Du
  • 32,885
  • 17
  • 118
  • 213
  • 2
    The statement you quote makes little sense to me; if they're known to be equally spaced you'd have an *interval scale*, not ordered categories. Does it explain why it makes that statement? – Glen_b Apr 03 '17 at 23:13
  • @Glen_b it is from UCLA so I thought it is reliable. I also do not understand why they make this claim... Interestingly I found [this answer](http://stats.stackexchange.com/questions/105115/polynomial-contrasts-for-regression) is very helpful for my understanding. – Haitao Du Apr 04 '17 at 00:27
  • 1
    Depending on what you mean by reliable, it *might* be -- personally I'd rather not just accept a statement without really understanding it -- better to understand the reasoning that lay behind it if we can. The important thing to understand is what they're saying and on what basis they're saying it (which is why I was asking ... maybe there's a good reason for it). – Glen_b Apr 04 '17 at 00:46

0 Answers0