Wider survey context effects
Some concern has been raised that subjective well-being reports may be influenced by aspects of the wider survey context, such as the weather on the day of measurement, the day of the week that respondents were surveyed, and/or minor day-to-day events occurring immediately prior to the survey. While wider study context effects could be interpreted as a form of “satisficing” - where easily accessible information is used to help formulate responses, rather than necessarily the information that would lead to optimal answers - such effects could also indicate that respondents may simply find it difficult to distinguish between information sources when making subjective judgements or reporting subjective feelings.
Momentary mood is one particularly compelling information source that may interfere with longer-term evaluative judgements (Schwartz and Clore, 1983; Yardley and Rice, 1991) as well as retrospective recall processes (e.g. Bower, 1981; Clark and Teasdale, 1982; Herbert and Cohen, 1996). Retrospective judgements of emotional experience have also been found to display peak-end effects, whereby respondents give more weight to the most extreme experiences (the peaks) or to those experiences that have occurred most recently (the ends) (Redelmeier and Kahneman, 1996; Kahneman, 2003; Bolger, Davis and Rafaeli, 2003). This implies that, when making an assessment of subjective well-being, more recent experiences - such as those connected with very recent events - and more salient or extreme experiences - such as those that are subject to media focus - may disproportionately affect self-reports.
Two key measurement issues are at stake. The first is a threat to comparability: if different surveys are conducted in different contexts, does this limit the comparability of subjective well-being measures? And if so, what can be done to manage this? The second issue is the extent to which systematic context effects, particularly those capable of influencing a large proportion of respondents simultaneously, might drown out the impact of other significant and policy-relevant life circumstances that only affect a small number of respondents at any one time. This is perhaps more an issue of validity and data interpretation, rather than methodology, but there are nonetheless implications for the measurement approach adopted.
Day-to-day events. There is some evidence that short-term events can exert detectable effects on evaluations of subjective well-being. For example, Schwartz and Strack (2003) report some experimental manipulations in which finding a very small amount of money on a copy machine, spending time in a pleasant rather than unpleasant room, and watching a national football team win rather than lose a championship game served to increase reported happiness and satisfaction with life as a whole. However, much of this evidence comes from small-scale studies with students, sometimes involving an impressive level of stage management.
A recent analysis has, however, highlighted the impact that a major news event and seasonal holidays can have on national-level reports of subjective well-being. Deaton (2011) examined a three-year time series of 1 000 daily subjective well-being reports in a national US representative sample, yielding close to 1 million respondents overall. Although the analysis focused on the effects of the financial crisis that began in summer 2008, some specific bumps in the time series data associated with short-term events are highlighted. For example, St. Valentine’s Day produced a small one-day reduction in negative hedonic experiences (mean levels of worry and stress, physical pain and anger; and mean of not happy, not enjoying, not smiling and sad), and the Christmas holidays produced a much larger improvement in hedonic experience. There was also a sharp drop in overall life evaluations (measured using the Cantril Ladder) around the time of the collapse of Lehman Brothers in September 2008, which may have been due to respondents anticipating a potential change in their life circumstances as a result of this event.
Day of week. Day-of-week effects have been observed in large-scale national survey data. Taylor (2006) examined data from the 1992-2000 waves of the British Household Panel Survey (N = 8 301, contributing over 38 000 person-year observations) and found that both self-reported job satisfaction and subjective levels of mental distress systematically varied according to the day of the week when respondents were interviewed. In particular, controlling for a wide variety of other job-related, household and demographic determinants, men and women interviewed on Fridays and Saturdays reported significantly higher job satisfaction levels than those who were interviewed midweek, an effect that was particularly strong among full-time (as opposed to part-time) employees. In the case of mental distress, there were fewer significant effects, but employed women interviewed on Sundays reported significantly lower levels of mental well-being than those who were interviewed midweek (with an increase in stress levels of about 5% at the sample means).
Although the day of the week affected mean scores, Taylor found that the inclusion (or exclusion) of day-of-week controls did not alter observed relationships between job satisfaction or mental distress and job characteristics, education, demographics and employment status. However, the list of determinants investigated was not exhaustive, and there remains a risk that data collected on just one day of the week could systematically over- or under-estimate subjective well-being.
In another very large-scale data set, Helliwell and Wang (2011) found no day-of-week effects on life evaluations (Cantril Ladder), but a significant increase in happiness, enjoyment and laughter, and a significant decrease in worry, sadness and anger experienced on weekends and public holidays, relative to weekdays. This study examined data from 18 months of the Gallup Healthways Well-being daily telephone poll, in which 1 000 randomly-sampled adults in the United States are surveyed each day, yielding over half a million observations. Deaton (2011) reports the same pattern of results in a longer time series of the same data set, capturing nearly 1 million observations between January 2008 and December 2010.
Rather than calling into question the usefulness of subjective well-being data, the weekend effects observed by Helliwell and Wang are interpreted by the authors as evidence in favour of the validity of the measures used. In fact, one would expect that valid measures of momentary affect would vary according to the day of the week due to differences in patterns of activity, whereas life evaluations should remain more stable over time. Consistent with this, the strength of the weekend effect observed by Helliwell and Wang varied in predictable ways according to job characteristics, such as being stronger for full-time employees relative to the rest of the population. A central variable in explaining the weekend effects was the amount of time spent with friends or family - which was on average 7.1 hours on a weekend, compared to 5.4 hours on week-days - and this increased social time at weekends “raises average happiness by about 2%”.
Seasonal effects. Common folklore would certainly suggest a role for weather and climate in subjective well-being. This was demonstrated by Schkade and Kahneman (1998), who found that even though there were no significant differences between Californian and Midwestern US students in terms of their overall life satisfaction, respondents from both regions expected Californians to be more satisfied, and this expected difference was mediated by perceptions of climate-related aspects of life in the two regions.
The largest set of evidence about the role of seasonality in subjective well-being comes from clinically-depressed populations and those suffering from seasonal affective disorder (SAD); relatively less is known about the role of seasons on mood and positive mental states, such as life satisfaction, among normal population samples (Harmatz et al., 2000).
Harmatz et al. (2000) conducted a one-year longitudinal study in Massachusetts, US, taking repeated quarterly measures among a sample (N = 322) that excluded individuals positively assessed for SAD. Significant seasonal effects were present in the data, both on the Beck Depression Inventory, and on the emotion ratings for anger, hostility, irritability and anxiety scales, all of which followed the same seasonal pattern (highest in winter, lowest in summer, and spring and autumn somewhere in between these extremes). The general trend was evident in both male and female respondents, but was only significant for males in the cases of irritability and anxiety. Effect sizes were described by the authors as “relatively small” and “from a clinical perspective, these differences would be noted as mild mood fluctuations” (p. 349). However, mean score differences between summer and winter emotions for females were quite large, dropping at least one point on a 9-point scale between winter and summer.17
Seasonal patterns have also been detected among larger nationally-representative samples. Smith (1979) examined time trends in national subjective well-being data (N = 610 - 723 respondents per month) and found that reported happiness showed a seasonal pattern, with a 10 percentage-point range in the proportion of very happy respondents between the winter low and the spring high.18 A positive affect measure also showed the same seasonal pattern (with a time series correlation with overall happiness of r = 0.83). On the other hand, a life satisfaction measure included in the study remained constant throughout the 12 months, and Bradburn’s overall Affect Balance Scale also failed to show a seasonal effect. Curiously, negative affect actually dropped during the winter months, positively correlating (r = 0.66) with the happiness trend. This goes some way towards explaining the lack of a clear seasonal pattern in affect balance overall.
Smith (1979) also highlights the risk of drawing conclusions about seasonality based on only one year of data - pointing out that other context effects might be at play (e.g. the Arab oil embargo hit the US from October 1973 to March 1974, at the same time as the happiness study). Previous studies have also observed spring ups and winter downs
(Bradburn, 1969; Smith, 1979), but another national sample data set analysed by Smith, the US National Opinion Research Center’s General Social Survey in 1972 (N = 1 599), failed to show a spring upswing.
The potential existence of seasonal effects raises the possibility that climate differences may account for some of the differences in subjective well-being observed between countries when objective life circumstances are controlled. This question was examined by Redhanz and Maddison (2005), using World Values Survey data on self-reported evaluations of happiness (on a 1-4 point scale), across a panel of 67 countries. Ten different climate indicators were constructed and tested in three different models, controlling for a range of other anticipated determinants of happiness (such as GDP per capita, unemployment, life expectancy and literacy). Three of the ten climate indicators had a significant effect: higher mean temperatures in the coldest month were found to increase happiness; higher mean temperatures in the hottest month were found to decrease happiness; and more months with very little precipitation were found to decrease happiness.
Seasonal effects may also be produced by seasonal trends in some of the key determinants of subjective well-being, such as unemployment status. Cycles in work and study situations may also be of relevance, particularly with regard to momentary mood. Whilst some of these effects will be substantive (i.e. they reflect how respondents actually feel, rather than contributing error to the data), they will nonetheless potentially have implications for how mean scores might vary over the course of a year, and therefore how data should be collected and described.
Weather. Evidence regarding the effects of weather on subjective well-being is mixed, and appears to be contingent on a number of factors. It has been suggested, for example, that although sunny days can produce higher estimates of both affect and life evaluations than rainy days, respondents might be able to exclude this information from their judgements if their attention is brought to it. Specifically, Schwarz and Clore (1983) found that among a very small sample of student telephone interviewees, respondents reported higher levels of current happiness, overall happiness with life, and satisfaction with life on sunny days as compared to rainy days. However, when the interviewer drew respondents’ attention to the weather, there were no significant differences between evaluations on rainy versus sunny days.
The impact of cloud cover on life satisfaction has also been investigated. In two large-scale Canadian surveys (N = around 6 000 and 1 500) conducted over several months, Barrington-Leigh (2008) found that seven days of completely sunny weather more than doubled the chance of an individual reporting an extra point higher on a ten-point life satisfaction scale, as compared with a completely overcast week. This effect size was notable relative to other predictors examined.19 However, including or excluding weather conditions in statistical estimates of life satisfaction from a range of other determinants (such as health, trust and income) produced indistinguishable coefficients.
In contrast to the work of both Schwarz and Clore and Barrington-Leigh, other evidence indicates no consistent effect of weather on life evaluations. Lawless and Lucas (2011) examined a large-scale survey data from over 1.5 million people over 5 years, and found no evidence of an effect of weather (rain, temperature, or combined weather conditions such as cold and rainy) on life satisfaction. The one exception to this was a significant interaction effect for change in weather conditions. Specifically, a temperature drop during warm months produced a very small increase in life satisfaction.
Although one might expect weather to have a significant impact on momentary mood, there is limited evidence of this, and the pattern of results is somewhat complex. Connolly Pray (2011) examined Princeton Affect and Time Survey data collected in the US during summer months, and found that among the women (but not the men) in their sample, low temperatures increased happiness and reduced tiredness and stress,20 whereas higher temperatures reduced happiness. Despite observing a significant effect on life evaluations, Barrington-Leigh (2008) failed to find a significant relationship between cloud cover and a short-term affective measure of happiness. This makes Barrington-Leigh’s results on life evaluations somewhat difficult to interpret, given that mood would be the primary mechanism through which one might expect weather to contaminate life evaluations.
In another large sample, Denissen et al. (2008) conducted an online repeated-measures diary study in Germany (N = 1 233) and found that six different weather parameters accounted for very little variance in day-to-day positive affect, negative affect or tiredness.21 Importantly, however, there were individual differences in weather sensitivity, with the effects of hours of daylight varying the most between individuals - being, on average, 21 times greater than the average random effect of the other five weather variables. These individual differences in the relationship between weather and mood were not significantly associated with differences in age, gender or personality traits. It is possible that patterns of daily activity, such as the amount of time spent outside, could play a role (Keller et al., 2005).
Taken together, the results of Connolly Pray (2011), Lucas and Lawless (2011) and Barrington-Leigh (2008) tend to suggest that unusual weather events or shifts are most likely to have an impact on subjective well-being: living somewhere with year-round sunshine might mean a single day of sunshine has limited effects on activities, emotions or evaluations, whereas a rare day of sunshine during a long grey winter, or a rare day of cooler temperatures during a long hot summer, could have sufficient power to temporarily influence momentary affect and/or someone’s outlook on life more generally. This suggests that the impact of short-term weather is most likely to be a problem in subjective well-being data when surveys are conducted on a single day, or on very small number of days, and within a limited geographic region. If a wider range of days and seasons are sampled, the effects of weather should be less problematic for mean levels of data aggregated across the study period.