Findings, implications and lessons learned

• The variation among dip-tank areas is greater than the variation among homesteads within dip-tank areas. This is generally true in cluster sampling - variation increases as one moves up the hierarchy.

  Thus, one can see in CS12Data2 (worksheet Cat_pop_inf) that the contribution of the variance from dip-tank areas to the standard error of the population mean (e.g. 22,928,938 for Lobamba) is greater than that for homesteads (5,448,984 for Lobamba).

  For Mayiwane, however, the larger proportion of dip-tank areas sampled has resulted in a smaller among dip-tank area variance (17,539,279) despite the larger variance that occurred within dip-tank areas (27,084,384).

• It can be seen how important it is, therefore, to ensure that there is adequate replication at the upper layers of a clustered hierarchy when planning a sampling frame.
  
  
  
  Source: Dave Elsworth