Memory Bookmarking: Using In Situ Information to Promote Recall in Online Data Collection

Abstract

Understanding human behavior is the goal of many social sciences, but accurately measuring the everyday behaviors of individuals is difficult. To collect information about and from individuals, researchers usually use retrospective survey questions; that is, questions about events that occurred in the past. Although these retrospective self-reports are prevalent, completing them places a considerable burden on respondents. According to a number of validation studies, recall may also be inaccurate. These recall errors can undoubtedly contribute to error in population estimates derived from survey data, particularly if respondents predominantly under- or over-report their activities, thereby leading to biased estimates. In this dissertation, we propose and implement a new data collection methodology called memory bookmarking (MB). MB begins by collecting very minimal in situ information from respondents via texting. Texting for MB purposes does not require much input from respondents; it merely requires respondents to click on a web link to record their GPS location, take a single photo, or reply to a single text message at any signaled time. This in situ information is subsequently used as recall cues to prompt the respondents in a follow-up survey. We conducted an MB-based pilot study using textual cues and a follow-up time diary to demonstrate the feasibility of this two-step data collection framework (Chapter 2) and help us design the experiment and system for the main study. In the main study, we tested the ability of two kinds of text cues, photo and map cues, to promote recall. We found that such recall cues helped respondents to remember and report more cued events than they were able to during spontaneous recall without cues (Chapter 3). Moreover, the events reported using cues were more fine-grained than the events reported without cues. Map cues require the least effort on the part of the respondents but provide the least amount of information for recall compared to the other types of cues (text cues and photo cues). Subjectively, respondents in this study were most confident about their recall when photo cues were present. This study provides evidence that the MB approach works more effectively for individuals whose memory capability is relatively low. On the other hand, we found a neutral to negative effect on uncued events that were adjacent to cued events in the same time diary. Overall, we saw higher-quality reports of cued events (Chapter 3) but equivalent- or lower-quality reports of uncued events, making for a seemingly zero-sum game (Chapter 4). This make senses from the respondents’ perspective; it seems that they were only willing to put a fixed amount of effort into the survey task. If they put more effort into some subtasks, then they put less effort into others. The resulting net impact on the time diary depended on the sum of the effects on cued and uncued events, which varied by individual. For instance, respondents with low memory capability benefited the most with respect to cued events but they were not affected when recalling uncued events, which resulted in a net gain in the total number of reported events. Other respondents showed a negative impact using this approach, which may be because of the particular way that the web instrument was designed and implemented. We provide detailed recommendations and future directions for implementing the MB approach for the time diary application.