Reference Period

The reference period is the time frame for which survey respondents are asked to report activities or experiences of interest. Many surveys intend to measure frequencies of events or instances within a given period of time; for example, How many times did you consult a medical practitioner during the last two months? or Think about the 2 weeks ending yesterday—have you cut down on any of the things you usually do about the house, at work, or in your free time because of illness or injury? Most of the time, the reference period starts at some point in the past and ends at the time of the survey. However, there are fixed reference periods as well—for example, a calendar year or a calendar quarter, depending on the design of the study. Whereas some ongoing surveys (rotating panels or ongoing cross-sectional surveys) aim to document change throughout the field period, others intend to measure the incidence or prevalence of certain events within a given period of time.

In addition to the relative position of the reference period in relation to the time of the interview, its length is of key interest for survey researchers. The length—for example, number of days, weeks, or months—affects the variance of the estimated frequency (prevalence) or the proportion of respondents who have experienced a certain event (incidence). The longer the reference period is, the more stable is the estimate in reducing the variance associated to the variable. Thus, considering a given level of precision, survey cost is reduced. However, the length of the reference period depends on the concept to be measured; in case of a high prevalence variable (e.g. number of restaurant visits) a relatively short reference period might be appropriate, whereas the same reference period would not be appropriate for events with a relatively low frequency of occurrence (e.g. crime victimization).

Whereas a lengthy reference period seems preferable in terms of the variance of the estimate, it has considerable drawbacks.

1. Measurements of past events are subject to recall error. Recall error consists of various components of which recall loss and telescoping are the predominant ones. Recall loss is due to respondents forgetting certain events or instances that actually happened within the reference period, which in turn reduces the estimate compared to the true value. By contrast, (forward) telescoping produces higher estimates because respondents accidentally place an event into the reference period although it actually happened before the starting point of the reference period. To compensate for recall loss, researchers make use of introductory questions that stimulate memory by asking for autobiographical events within the reference period or other landmark events. By contrast, telescoping may be dealt with by bounded recall. To reduce telescoping, respondents are asked in two or more consecutive interviews to report on the frequency of certain events. At the time of the second measurement, respondents are asked to report on the [Page 700]number of events or instances since the last interview, which serves as the outer “bound” for recall. Thus, events or instances reported in the first interview can be excluded from the response at the time of the second interview. This expensive procedure addresses the problem of respondents' tendency to “telescope” too distant experiences into the reference period.

...