Figure A3:

Validity and efficiency of Root 1 for the time-fitted random treatment case, for four different SWT designs and a variety of true ACC’s. The designs each have two sequences and K=5 observations per cluster per time period. Throughout, ${\sigma }_t^2=1$ and the balance of ${\tau }_t^2$ and ${\eta }_t^2$ is fixed at ${\eta }_t^2={\tau }_t^2/2$.