What Citizens Want: Determinants and Predictability of Economic Evaluations

• In Event: Measurement in Survey Research

Abstract

Conventional wisdom suggests that most people have a poor grasp of the objective strength of the economy, or that economic evaluations are endogenously induced by partisan biases. Contesting earlier dismissals of survey evaluations, this paper shows that subjective assessments of the economy are surprisingly informative about the true state of the economy in both developed countries and emerging markets. Machine learning approaches are shown to be a promising procedure for uncovering which economic outcomes citizens value by identifying the attributes of the economy that drive subjective economic sentiment.

Without pre-specifying which economic indicators "should" be salient in citizens' minds, agnostic methods (least absolute shrinkage and selection operator, conservative regression trees, and random forests) are used to predict economic evaluations in a large set of countries, and their out-of-sample performance is compared. While earlier research has emphasized the importance of income growth for overall satisfaction of the public, this paper shows that the state of the labor market is a better predictor of subjective economic evaluations than GDP growth. Regression trees indicate that low GDP growth does contain useful signal about the (perceived) state of the economy, but conditioned on particular circumstances. Knowing the GDP growth rate is valuable most of all in places where labor market outcomes for young workers are are especially poor.

Author

• Jan Zilinsky, New York University