Melvin C Y posted on Wednesday, May 09, 2012 - 10:13 am
I have three predictor variables, A B C, at level 2 on one outcome, Y. A and B have positive (beta is around 0.5 for both) but marginally significant effect on Y. But the standard errors are quite large which affected the significance of the estimates. Pvalue for A is 0.04 and B is 0.06. C is negative (beta around -0.40) but not significant with a large pvalue 0.50. L2 correlation for A and C is high at 0.73. And between 0.3 to 0.5 or the rest. Could the large standard errors be due to design effects? I have 100 clusters, but the average cluster size is 10.
When I constrained C to zero, the size of the estimates A and B remained quite close to before, but the pvalues went above 0.10. If it matters, I have a total of 1 L1 variable measured at both levels and 4 L2 variables plus 1 outcome.
Would be grateful for any advice you might have. Thank you.