Within-Household Sampling Conditioning on Household Size

Abstract

We review existing methods for sampling an adult from a household and highlight the strengths and weaknesses of these methods with respect to issues relevant for sampling in the Middle East and developing countries. We then develop a new, flexible within-household sampling scheme that takes full advantage of the fact that household size must be obtained to produce unbiased estimates. The proposed method randomly selects an adult conditional on the number of adults in the household. The method was applied in a national sample in Qatar and it proved it was effective in a country where large household sizes are common and the information required for other sampling methods is generally not known.

The list of methods for sampling a person from a household is long and diverse, dating back to a simple probability method proposed by Kish (1949). The Kish method requires a complete listing of all people in the household, which is believed to add to the length of the interview and increase the likelihood of refusals. Several creative methods have been developed to bypass the listing process to increase respondent cooperation and nearly all of these have concentrated on telephone surveys and ones conducted in the United States and Western Europe.

In this article, we propose a new method of within-household sampling that is not as intrusive as the Kish method, can be used in both telephone and face-to-face surveys, and is more appropriate for use in sampling in Middle East and developing countries. One of the key differences that affects the efficacy of the sampling is household size. Households in these countries are often much larger than in Western countries. In addition, information used to select the adult such as the birthday is not as commonly available in these countries as in the United States.

The other feature of our proposed approach is that it tailors the sampling method to the size of the household, which is not explicitly done in other methods. Most of the other within-household sampling methods collect information about the household size (number of adults in the household), since this information is needed to calculate the probability of selection and to compute unbiased estimates (Rizzo, Brick, & Park, 2004). However, these methods underutilize household size information in the development of the sampling method.1 In the next section, we review current methods in practice, along with the advantages and disadvantages.