Bounding

Bounding is a technique used in panel surveys to reduce the effect of telescoping on behavioral frequency reports. Telescoping is a memory error in the temporal placement of events; that is, an event is remembered, but the remembered date of the event is inaccurate. This uncertainty about the dates of events leads respondents to report events mistakenly as occurring earlier or later than they actually occurred. Bounding reduces telescoping errors in two ways. First, bounding takes advantage of the information collected earlier to eliminate the possibility that respondents report events that occurred outside a given reference period. Second, bounding provides a temporal reference point in respondents' memory, which helps them correctly place an event in relation to that reference point.

A number of specific bounding procedures have been discussed in the survey literature. The bounding interview procedure was first developed by John Neter and Joseph Waksberg in the 1960s in a study of recall of consumer expenditures (they call it “bounded recall”). The general methodology involves completing an initial unbounded interview in which respondents are asked to report events that occurred since a given date. In the subsequent bounded interviews, the interviewer tells the respondents the events that had been reported during the previous interview and then asks for additional events occurring since then. In other words, the information collected from each bounded interview is compared with information collected during previous interviews to ensure that the earlier reported events are not double counted.

For example, suppose panel respondents are interviewed first in June and then in July. The June interview is unbounded, where respondents are asked to report events that occurred in the previous month. The July interview is bounded. Interviewers would first inform respondents of the data they had provided in June and would then inquire about events that happened since then. Often the data from the initial unbounded interview are not used for estimation but [Page 68]are solely used as a means for reminding respondents in subsequent interviews about the behaviors that have already been reported.

Neter and Waksberg demonstrated in their study that bounding effectively reduced 40% of telescoping on expenditures and 15% on the number of home improvement jobs. This finding encourages panel or longitudinal surveys to employ the bounding technique to reduce the effect of telescoping. The National Crime and Victimization Study (NCVS) is one example. In its redesign, NCVS uses the first of its seven interviews to “bound” the later interviews. There is some evidence suggesting that this bounding technique reduces the likelihood of respondents reporting duplicate victimizations.

The bounding procedure proposed by Neter and Waksberg requires multiple interviews; thus, it is viable only for longitudinal or panel surveys. For onetime surveys, researchers have proposed bounding respondent memory by first asking about an earlier period and then about the more current period. For instance, within a single health interview, respondents are first asked about their health behavior in the previous calendar month and then asked about the same events in the current calendar month. One study shows that bounding within a single interview with two questions reduces reports by between 7% and 20% for health—related behaviors. It reduces telescoping by about 30% to 50% for trivial events, such as purchasing snacks.

...