Fixed effect and covariance parameter estimates (generalized linear mixed models)
- Still in the viewer for the model with after_t as a random effect, click the Covariance Parameters view thumbnail.
- From the Effect dropdown, select Block 1.
The order of parameters along the diagonal of the covariance matrix corresponds to the order of effects on the Random Effect Block dialog.
- UN(1,1) is the variance estimate for the random effect intercept term.
- UN(2,1) is the covariance between the intercept and after_t.
- UN(2,2) is the variance estimate for the after_t.
You can test the statistical significance of parameters in the covariance matrix with a simple Wald test; for example, for UN(2,2) compute a z statistic equal to the ratio of the estimate and its standard error, 0.187 / 0.022 = 8.5. A z statistic with a value of 8.5 has a p-value near 0, which suggests that including after_t as a random effect is necessary to account for heterogeneity among patients between the baseline and after treatment periods
- Now let's compare the fixed coefficients. Activate (double-click) the model object for the model with an intercept-only random effect.
- Click the Fixed Coefficients view thumbnail.
- From the Style dropdown of the Coefficients view, select Table.
- In the parameter estimates table, click the Coefficient cell.
This displays the standard error, t statistic, and confidence
interval.
The coefficient for after_t is statistically significant and its estimate of −0.086 implies that the expected seizure rate for a typical patient would be reduced by 1−exp(−0.086) = 8.2% after starting the trial on placebo. It's possible that there is a placebo effect, but we don't expect it to be that large.
- For comparison, back in the viewer for the model with after_t as a random effect, click the Fixed Coefficients view thumbnail.
- From the Style dropdown of the Coefficients view, select Table.
- In the parameter estimates table, click the Coefficient cell. This displays the standard error, t statistic, and confidence interval.
On the other hand, the after_t effect is not significant in the model with after_t as a random effect, which makes more intuitive sense. For this model, the expected seizure rate for a typical patient would reduce by 1−exp(−0.012−0.191) = 18.4% after taking the anticonvulsant, so the drug seems to have a modest effect.