Kish Selection Method

Most surveys of attitudes and opinions have two stages: The first is drawing a random sample of dwelling units, and the second is selecting one person within a dwelling unit to interview. Leslie Kish published a landmark article in the Journal of the American Statistical Association in 1949 that described a rigorous, almost pure probability method of sampling persons within households to be surveyed. He named two basic conditions: (1) There must be a known probability of inclusion (excluding zero) of each adult in the population, and (2) it must be a practical and efficient procedure to implement.

Although survey data collection at that time often was conducted in person, Kish's plan is suitable for telephone surveys as well. It is accomplished by the interviewer's listing the age and sex of the adult members of the household and their relationship to the head of household and then consulting a table to choose the correct respondent. The interviewer numbers all the males first from oldest to youngest and then all the females in order of decreasing age. Kish provided two different sets of tables. In his first example, the interviewer has six tables, labeled A to F, [Page 409]which indicate first the numbers of adults in households from 1 through 6 and then below, the number of the adult to be interviewed. Each table lists the target individual in systematic order that differs among the tables, and the first three tables are set up in such a way that males are more likely to be selected because males tend to be underrepresented in surveys. In his second example, used more frequently in research today than the first procedure, there are eight tables labeled A, Bl, B2, C, D, El, E2, and F. Tables A, C, D, and F will be used one sixth of the time, and the others will be used one twelfth of the time. These tables give equal chances of selection to individuals in households of 1, 2, 3, 4, and 6 adults. Dwelling units with 5 adults are overrepresented to “compensate” for the inability of this method to represent households of more than 6 adults, a very small proportion of the population. Kish estimated, at the time he was writing, that only 1 or 2 adults per 1,000 would not be represented, usually young females.

Here is one example of the Kish method question wording, used by Robert W. Oldendick and his colleagues:

Some researchers have introduced other methods because they believe that Kish's procedure takes too much time to enumerate household members, respondents may perceive it as intrusive, and it may increase refusal rates. In other words, they criticize it for potentially adding to nonresponse, although it can decrease within-unit noncoverage. The method may be more of a problem in larger, as opposed to smaller, households. Age is not asked in one-person households or in one male/one female units. Another criticism is that the tables may be outdated because of changes over time in the distribution of households or that the tables are inappropriate for some countries with different population patterns. Some surveyors have found that the Kish selection method, per se, is not very intrusive but that refusal rates tend to increase with any method that requires two consents—the first from the informant who answers the phone and the second from the designated respondent. Several researchers report that the Kish method is still a reliable and noninvasive method when interviewers are skilled and well trained in it. It is important that interviewers have faith in the method because they otherwise unconsciously may communicate negative attitudes toward a method, thereby decreasing the cooperation rate. Several comparative studies have included the Kish method with other methods on coverage, response, costs, and other characteristics.

...