|
|
Scaling and structural parameters |
|
Message/Author |
|
John Venz posted on Friday, April 29, 2016 - 8:32 am
|
|
|
Hi, I am running a simple SEM using MLR with 3 indicators, one latent variable and one observed endogenous variable. f1 BY i1 i2 i3; Ex ON f1; What I have discovered is that the p-value of the coefficient Ex ON f1 depends strongly on the chosen scale of f1, that is choosing instead of the above: f1 BY i1* i2 i3; f1 @1; Ex ON f1; gives a p value that .2 lower than the one in the model at the top. It is clear to me that the value of model parameters depends on the chosen scaling of latent variable but I thought until now that p values are more or less not influenced by scaling. So what is the explanation and what p value is right? (All 4 observed variables were standardized before analysis.) Thank you very much! |
|
|
For some reason, the slope estimate must have different sampling distributions in the two cases - one being more non-normal than the other. Try bootstrapping and look at the confidence intervals. |
|
Back to top |
|
|