Last-Birthday Selection

Survey researchers are usually concerned with choosing respondents within households after households are selected randomly. Within-unit coverage and non-response are key issues, so researchers want to select the correct respondent and gain his or her cooperation. Each of these goals has costs. One popular quasi-random compromise is the last-birthday (LB) method of selecting respondents from within a sampled household in random-digit dialing surveys. It circumvents the pitfalls of pure or nearly pure random methods by being relatively quick, easy, and likely to secure cooperation. Probability methods can involve a potentially lengthy and intrusive process of querying the informant (person who answers the phone) about all household members eligible to be interviewed before selecting the correct respondent from the resulting list.

An example of LB question wording is, In order to determine whom to interview, could you tell me, of the people who currently live in your household who are 18 or older—including yourself—who had the most recent birthday? I don't mean who is the youngest adult, but rather, who had the most recent birthday? If the respondent does not know all the birthdays, the following question can be asked: Of the ones you do know, who had the most recent birthday? Some researchers have sought to study the best wording to choose the correct respondent and secure cooperation.

The first published description of birthday methods tested a next-birthday (NB) method, assuming that the incidence of births is random and the first stage of selection in a two-step process. The researchers, Charles Salmon and John Nichols, considered this method to be the second stage of sampling, with all members of a household having an equal probability of being chosen. After implementation, however, the [Page 418]researchers recommended use of a last-birthday (or most recent birthday) procedure because some respondents were confused about whom to nominate. LB now is used more frequently than NB because it appears that informants understand it better or are more likely to know past birthdays than upcoming birthdays.

Several studies show that births are not necessarily distributed randomly across months in some countries and that LB also tends to produce distributions of birthdays skewed toward the months closely preceding the interviews in households of two or more adults. Skewed distributions should not result in bias, however, unless birthdays are connected to interview topics. Another drawback is that informants in large households, compared with those in small households, are less likely to know everyone's birthday. LB tends to overrepresent females but may represent blacks and younger males better than do other procedures. Females may be over-represented because they tend to answer the phone more frequently than males, and sometimes they may be “protecting” males from coming to the phone. Other times informants want to do the interview so they inaccurately report they had the last birthday.

It is important that informants understand the question because misunderstanding can lead to within-unit coverage problems, especially among the less educated, the foreign born, and larger households. Sometimes informants erroneously think that the interviewer wants the youngest household member. Interviewers also should ascertain the respondent's day and month of birthday to be sure they are speaking with the correct person. Stressing the need for scientific accuracy can improve the response rate. Research on the LB method indicates that the respondent has been nominated incorrectly between about 10% and 25% of the time. Intensive training of interviewers regarding the question and accuracy of respondent selection, especially in households of three or more persons, is likely to decrease the incidence of coverage mistakes. Some research on the LB technique shows that the inaccurately selected respondents did not contribute to biased results, whereas other research demonstrates some contribution to bias when certain demographic variables associated with incorrect respondent selection were related to specific substantive results.

...