Table 3. Sample size required for detecting mortality (proportion), 20%-25% reduction, CEP vs TA/OR.
| Mortality Proportion in TA/OR and under null | With 25% change CEP vs RS | With 20% change CEP vs RS | ||||
| Proportion in treatment | Total N required if ICC=0 | Total N required if Cluster=200 ICC=.01 | Proportion in treatment | Total N required if ICC=0 | Total N required if Cluster=200 ICC=.01 | |
| .025 | .01875 | 17196 | 130050 | .020 | 27618 | NAa |
| .050 | .03750 | 8404 | 14580 | .040 | 13490 | 42240 |
| .075 | .05625 | 5474 | 7560 | .060 | 8782 | 15760 |
a. 80% power cannot be reached due to large design effects of clustering. For example, the sample sizes of 200,000 in one arm can only achieve 66% power to detect a difference between the group proportions of .005.
CEP, community engagement and planning; RS, resources for services; ICC, intracluster correlation coefficient.