Consumption and Savings Balances of the Elderly: Experimental Evidence on Survey Response Bias

The unfolding bracket method for eliciting quantitative information in economic surveys is effective in reducing item non-response and outliers, but is vulnerable to bias induced by anchoring. To test for anchoring effects, this study introduced an experimental module in the AHEAD (Asset and Health Dynamics of the Oldest Old) panel in which the unfolding bracket (or gate) sequence varied by treatment. We find that there are strong anchoring effects in the two household variables studied, savings account balances and monthly consumption. We construct a model of anchoring that is successful in capturing much of the effects of anchoring. Thus, it appears that experimental variation in the unfolding bracket design combined with a suitable model of anchoring can be used to predict and undo the biases introduced by anchoring. These conclusions have important implications for the design of quantitative questions in current economic panels such as AHEAD and the Health and Retirement Survey (HRS), and for other applications that use gate designs, such as elicitation of willingness-to-pay by contingent valuation. _______________________________ 1 The first author is at S.U.N.Y. Stony Brook and the RAND Corporation, and the remaining authors are at the University of California, Berkeley. The first two authors are associates of the National Bureau of Economic Research. Correspondence should be directed to Daniel McFadden, Department of Economics, University of California, Berkeley CA 94707-3880, or [email protected]. A postscript version of this paper is posted at http://elsa.berkeley.edu/~mcfadden. This research was supported by a grant from the Institute on Aging of the National Institute of Health to the National Bureau of Economic Research. We are indebted to Danny Kahneman, whose remarks on anchoring motivated this research. CONSUMPTION AND SAVINGS BALANCES OF THE ELDERLY: EXPERIMENTAL EVIDENCE ON SURVEY RESPONSE BIAS Michael Hurd, Daniel McFadden, Harish Chand, Li Gan, Angela Merrill, Michael Roberts "Collecting bracket responses without varying the anchors is criminally negligent." Danny Kahneman, 1993 I. Psychometric Biases in Economic Survey Data A. The Need for Accurate Data. A prerequisite for understanding the economic behavior of the elderly, and the impacts of public policy on their health and wellbeing, is accurate data on key economic variables such as income, consumption, and assets, as well as expectations regarding future economic and demographic events such as major health costs, disabilities, and death. Standard practice is to elicit such information in economic surveys, relying on respondents’ statements regarding the variables in question. Economists are generally aware that stated responses are noisy. Item nonresponse is a common problem, and carefully done surveys are designed to minimize it. Well-designed analyses of economic survey data are careful about detecting implausible outliers, imputing missing values, and correcting for selection caused by dropping missing observations. Circumstances are recognized that tend to produce systematic biases in response, such as telescoping in recall of past events that arises from the psychophysical perception of time intervals, or overstatement of charitable contributions that arises from the incentive to project a positive selfimage. Nevertheless, economic studies are often too sanguine about the reliability of subject’s statements regarding objective economic data. B. Stated Versus Revealed Economic Data. For many economic variables, it is possible in principle to obtain the accounting or administrative records necessary to verify stated responses. For example, subjects may be asked to consult or provide copies of utility bills, bank statements, or income tax records, or to give permission for linking to medicare or social security records. In practice, this is rarely done, due to cost, the difficulty of obtaining compliance from the subjects, and privacy and disclosure issues surrounding government administrative records. In cases where direct comparisons of stated and revealed economic data are available, the results are sobering. For example, Poterba and Summers (1986) find