The evidence - affect measures
The length of reference period is likely to be of particular interest in relation to measures of recently experienced affect, where existing methods range from asking respondents about the intensity of feelings right now (e.g. in the experience sampling method, where respondents often report on their affect levels several times a day) to the presence or absence of emotions yesterday (e.g. the Gallup World Poll) or the frequency of positive and negative experiences over the past four weeks (e.g. the Diener and Biswas-Diener Scale of Positive and Negative Experience, 2009). Some examples of existing affect measures are provided in Annex A. Of course, to the extent that these measures are designed to capture different constructs, such as momentary mood versus more stable patterns of experience, differences in responses can be due to valid variance. The issue in these cases is to select the best measure for the construct of interest. However, the extent to which approaches differ also affects both the error sources and the comparability of the data, and thus needs to be considered when selecting measures and interpreting the findings.
The importance of reference periods to affect measures has been examined by Winkielman, Knauper and Schwarz (1998), who asked respondents how frequently they get angry within a 1-week reference period, and within a 1-year period. Consistent with the hypothesis that shorter reference periods prompt respondents to report more frequent experiences, and longer reference periods are assumed by respondents to pertain to rarer experiences, they found that fewer but more extreme episodes were reported within the 1-year period. Asking how often an individual experiences anger also presupposes that individuals will have experienced this emotion within the given timeframe, whereas prefixing this with a yes/no “were you ever angry during the last week...” question would send quite a different signal about the expected frequency of an experience. Arguably, the phrase get angry is simply too vague to have independent meaning - and thus the time frame associated with it provides respondents with information about how it should be interpreted. If the key concept of interest is better defined, the reference period may play less of a role in understanding meaning. However, particularly in the case of affect, the constructs of interest are intrinsically linked to the reference period - such that affect yesterday is a fundamentally different construct to affect in the last four weeks.
Whilst respondents might be expected to provide reasonably accurate recall of emotional experiences in end-of-day reports (e.g. Parkinson, Briner, Reynolds and Totterdell, 1995; Stone, 1995), quantifying the extent to which an emotion was experienced in terms of either frequency or intensity over a longer period may be more challenging. Cognitive testing by the UK’s ONS (2011c) indicated that some respondents find remembering their activates on the previous day quite difficult, and thus averaging the intensity of emotions over even just one day may be quite cognitively demanding. However, the ONS also reported that some respondents objected to questions that focus on affective experiences on a single day, raising concerns that this may be unrepresentative of how they usually feel.2 This has implications for both respondent motivation and accuracy of reporting, and suggests that interviewer briefing and the preamble provided for short-term affect questions will be particularly important (discussed in Chapter 3).
To better understand accuracy of recall for affect, Thomas and Diener (1990) compared small-scale experience sampling (Study 1) and daily diary (Study 2) data with retrospective accounts of mood from the same respondents across a period of three and six weeks respectively. In both studies, they found that retrospective reports of affect intensity were significantly higher than actual intensities experienced (for both positive and negative affect). Frequency judgements regarding the proportion (0-100%) of time when positive affect was stronger than negative affect proved to be more accurate, but the estimate was in this case significantly lower than the actual frequency. Perhaps most compelling, however, was evidence from the cross-individual correlations between retrospective reports and actual experiences. Although several of these correlations were highly significant, they suggest a considerable gap between experiences and retrospective reports.3
These findings, although limited, suggest that retrospective assessments over several weeks are likely to produce lower quality data than those over just 24 hours. Asking respondents to report on their affective experiences over long time-frames may still reveal something interesting about their overall level of subjective well-being (e.g. by tapping into dispositional components of affect), but may not be reliable as an accurate account of experienced affect or emotion. There is also tentative evidence that asking respondents to recall the frequency of their emotional experiences may produce more accurate reports than intensity judgements.