The evidence

It is difficult to isolate the impact of the survey source on responding. Most information on this effect therefore comes from studies where the survey source is experimentally manipulated. For example, Norenzayan and Schwarz (1999) asked respondents to provide causal attributions about mass murder cases - and they found that respondents were more likely to provide personality- or disposition-based explanations when the questionnaire letterhead was marked Institute for Personality Research and more social or situational explanations when it was marked Institute for Social Research.

In a further example, Smith et al. (2006, Study 1) found that the correlation between life satisfaction and a subsequent health satisfaction question was higher when the survey was introduced to respondents as being conducted by a university medical centre (focused on the quality of life of Parkinson’s disease patients) than when the survey was introduced as being conducted by the university in general (and focused on the quality of life of people in the eastern United States). The health satisfaction component of life satisfaction was much greater in the medical centre condition, accounting for three times as much variation in the life satisfaction measure (39.7% as opposed to 11.5%). Smith et al. liken this to the assimilation effects observed in studies of question order.

In national surveys that cover a very wide range of topics, any a priori assumptions that might be held about relationships between variables are likely to be obscured by the sheer breadth of the survey. Short, sharp, opinion-based surveys, on the other hand, might be more likely to be viewed by respondents as “hanging together”. So, for example, in the Gallup poll described by Deaton (2011), the questions around the “direction” of the country asked at the very beginning may have set the tone for the subjective well-being questions that followed, and respondents may have been conflating their own subjective well-being directly with their views on national well-being.

Knowing that a survey originates from a national statistical office is unlikely to give respondents many cues as to how they should respond to subjective well-being questions in particular - although knowing that the data is explicitly linked to the information needs of the government may introduce a risk that respondents will tailor their answers (consciously or otherwise) in order to send a message to those in power. It is not clear at present whether or how much of a threat this poses to subjective well-being data. Deaton (2011) noted the close relationship between subjective well-being measures and stock market movements between 2008 and 2010. When national-level data on subjective well-being become available for monitoring purposes, it will be interesting to see whether this follows the political cycle or other potential determinants of national mood. It will also be important to investigate the source of any differences in results between official and unofficial surveys.

< Prev   CONTENTS   Source   Next >