We are analysing data on a cluster randomised control trial. We have treated the clusters (classes, N = 8) as fixed effects, thus dummy coding to create 7 different variables. This is because we had non-convergance issues when running a MLM.
However, when we run the analysis (in SPSS or STATA), the analysis removes one of the classes (not always the same one, depending on the outcome), and the treatment variable. We are perplexed why this is happening, so if anyone has any suggestions and workarounds, they would be gratefully received.
you need k-1 levels so that estimates for each level are relative to the left-out level.
For example, if you're dealing with sex as a variable in a multiple regression, you'd code with something like (male=0, female=1), and a regression formula may look like this:
Y = B0 + B1X1 + B2Sex
Which means that the predicted value for a male is
Y = B0 + B1X1 + B2*0
= B0 + B1X1
but if it's female then the predicted value is
Y = B0 + B1X1 + B2*1
If you have three groups then the model might look like:
Y = B0 + B1X1 + B2Grp2 + B3Grp3
so the predicted value for group#1 is
Y = B0 + B1X1 + B2*0 + B3*0
for group#2 the value is
Y = B0 + B1X1 + B2*1 + B3*0
and for group#3
Y = B0 + B1X1 + B2*0 + B3*1
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running