When economists study how people respond in various economic situations, they generally use one of two methods to measure behaviors. They observe people’s actions and use statistics to infer how people would have responded to different conditions, or, alternatively, they ask people to make the inference themselves by surveying them directly about what they did, or would have done in different conditions.
The first method is called “revealed preference” and has been a staple of social-science research for decades. But, because the second approach, “reported preference,” is often more straightforward, it has become more popular recently in measuring how people behave, and how they choose to save and consume in particular. However, MIT’s Jonathan A. Parker and University of Pennsylvania’s Nicholas S. Souleles find that people aren’t necessarily accurate when they recount how they spent their money.
Parker and Souleles analyzed two data sources to measure revealed preferences, then used these measured behaviors to evaluate the accuracy of reported preferences.
To do this comparison, Parker and Souleles picked a high-profile test case: the Economic Stimulus Act passed by US Congress and signed by President George W. Bush in February 2008. Under this act, designed to forestall a recession, the Internal Revenue Service disbursed $100 billion to 130 million taxpayers. Payments amounted to $600 for single filers and $1,200 for joint filers, with an additional $300 for every child who qualified for the child credit.
Looking at the data sets, one from Nielsen and one from the US Census Bureau, Parker and Souleles compared the share of the payments people actually spent with what share they said they spent.
Nielsen’s Consumer Panel allowed the researchers to see the purchasing behavior of 40,000–60,000 US households from a wide, geographically diverse range of retail outlets. The researchers merged these NCP data with data from another survey of NCP households that they conducted with Nielsen at the time the stimulus payments were distributed.
Meanwhile, the Consumer Expenditure Survey, conducted by the US Census Bureau for the Labor Department’s Bureau of Labor Statistics, uses interviews and diaries to record data on consumer behavior. Parker and Souleles worked with the BLS to add new questions to the survey about the 2008 stimulus payments during the months when households received the checks.
Using people’s spending behavior and differences in when they received their payments, the researchers calculate that the average respondent had a propensity to spend roughly half his payment in the CES and slightly less in the NCP. Among the same people, a third of households in the CES reported that they mostly spent the payments, versus one in five in the NCP. Notably, the households reporting that they mostly spent their payments were exactly the people identified statistically as big spenders, based on the increase in their spending when their payments arrived. Further, using a number of methods, including measures based on participants’ reports of how much they spent in response to the payments, the researchers find that the average spending rates—derived from the two methods, reported preference and revealed preference—were broadly similar.
However, the methods produced conflicting conclusions about which households spent more of their payments. Revealed preference data show that households with lower income or less liquidity spent more, while households with higher income or liquidity spent much less. But reported preference data didn’t reveal any such difference. Rather, survey respondents reported spending their payments at similar rates across different levels of income and liquidity.
“People who are persistently constrained and so spend money as it arrives may perceive and report ‘mostly saving’ when they spend more slowly than they typically do,” Parker and Souleles write. “We conclude that reported preference data do not reliably measure quantitative spending, but that they are highly informative.”