Extrapolating results to other populations
(go to Outline)
Be careful if extrapolating results outside the sampling universe
Some survey managers may want to apply the results of a specific survey to a population which was not included in that survey, that is, a population which was not included in the sampling universe for that survey. This is called extrapolation. You must remember that the only population to which the results of a survey can be directly generalized is the population which made up the sampling universe for that survey.
Extrapolation may occur in two ways. The first consists of applying findings from one target group to another target group within the same population. For example, if the 2-week cumulative prevalence of diarrhoea was measured in children less than 5 years of age, you cannot apply these results to older children. The risk factors for diarrhoea may be completely different, leading to a very different level of diarrhoea in different age groups.
The second type of extrapolation consists of applying the survey results to populations not living in the area of the survey. The temptation to do this may be strong if security or poor transportation makes carrying out a survey very difficult. For example, if mortality is measured in a survey done in one province, instead of carrying out a separate survey in a neighbouring province, survey managers may wish to just assume that this neighbouring province has the same mortality rate. However, only in exceptional circumstances can this be done and only if there is compelling evidence that the factors contributing to health condition of interest are very similar in the surveyed and non-surveyed populations.
Democratic Republic of the Congo mortality survey 2000
Five mortality surveys were done in five different areas of eastern Democratic Republic of the Congo (DRC) in the year 2000. An overall estimate of excess mortality was then generated for all of rebel-controlled eastern DRC. The authors employed techniques to ensure that the final estimate was conservative, that is, that it did not overestimate mortality. First, they applied the survey findings to areas outside the sampling universe in three different ways and found similar estimates of excess deaths. Second, they pointed out that the major reason for not including areas in the surveys was insecurity due to violence. It is quite clear from this survey and others that mortality is higher in insecure, violent areas than in secure areas. Thus, the mortality rates applied from surveyed areas to unsurveyed areas probably underestimated mortality in the latter. Third, the authors explained in great detail how the extrapolation was done and why they made the assumptions they did. Click hereto look at the final report of this survey. The extrapolation is presented on pages 12-16.