Skip to main content

Data analysis

(go to Outline)

Is non-response calculated and described? Non-response can lead to substantial bias. Reports should present the procedures used to minimize non-response, the non-response rate, and some analysis of the reasons for non-response. If non-response is high, there should be an attempt to compare non-respondents to respondents to assess the level of potential bias.

Do the estimates of the main indicators and outcomes include confidence intervals? Without confidence intervals or some other measure of precision, the reader has no idea how certain to be that the true population value is close to the estimate from the survey.

Are the results clearly presented? A report in which the text is unclear implies that the writer either: a) has trouble writing, which is common and excusable (writing is a difficult thing to do), or b) has trouble thinking, which is also common but not at all excusable. A survey manager who has trouble thinking cannot analyze data appropriately, present results coherently, draw appropriate conclusions, nor formulate good recommendations.

Are the results presented in a standard manner? Over the years, a rough consensus has developed regarding the best way to present the results of a nutrition survey. Some of these consensus methods are described in a survey manual written for the World Food Programme. (Click here to open this manual)

Do the analyses answer the questions posed in the objectives or elsewhere? It is very easy to become distracted by interesting data analyses and lose sight of the reasons the survey was done. This is one reason why clearly written objectives are so important. (see pages on Survey goals and objectives)