Is there any difference between the ways we can code a categorical character with more than 2 factors? I mean, besides the differences in the coding itself?
I've tried some tinkering with R to figure out if there is any issue, mainly between effect coding and dummy coding, and I found that residuals of a linear model were practically identical, but when I use the anova's functions, I got different sum of squares and significance values, but only if I use type III sum of squares (type I and II seems identical).
Can someone shed some light on this?
The residuals should be the same, but the parameter estimates will differ depending on your coding scheme. Here is a good basic description: http://talklab.psy.gla.ac.uk/tvw/catpred/
and here is a more detailed description of different coding schemes:
http://www.ats.ucla.edu/stat/r/library/contrast_coding.htm
http://stats.stackexchange.com/questions/23197/type-iii-sum-of-squares-from-sas-and-r
Might be useful.
The residuals should be the same, but the parameter estimates will differ depending on your coding scheme. Here is a good basic description: http://talklab.psy.gla.ac.uk/tvw/catpred/
and here is a more detailed description of different coding schemes:
http://www.ats.ucla.edu/stat/r/library/contrast_coding.htm
The similarity is because they satisfy assumption such as normal residuals for regression analysis including logistic or linear. The sum of squares vary because the the methods differ also the variables. Still dummy coding, arrays.. may be used or formats too so the factor levels work. Analysis will still work.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software