Findings, implications and lessons learned

There are many things of a statistical nature to learn from the foregoing that we shall note here.

• It is most important to record precisely how sampling was done. Any choice of sampling that is strictly non-random affects the ability to estimate population numbers without bias.

• We have obtained two population estimates, firstly, knowing the numbers of homesteads in all dip-tank areas and, secondly, knowing only those in dip-tank areas that were sampled. It follows that if we don't know how the dip-tank areas were selected we cannot say which method, if either, is correct.

• Sampling dip-tanks in such a survey is helped by taking into account the different agro-ecological zones, the proximity to urban areas and the livestock densities in order to form more homogeneous strata within which dip-tank areas can be sampled.

• Stratification has the further benefit of reducing standard errors as it puts similar types of dip-tank areas together, and hence reduces the among dip-tank area component of the variance.