Working Paper: NBER ID: w27860
Authors: Victor Stango; Jonathan Zinman
Abstract: Social scientists often consider temporal stability when assessing the usefulness of a construct and its measures, but whether behavioral biases display such stability is relatively unknown. We estimate stability for 25 biases, in a nationally representative sample, using repeated elicitations three years apart. Bias level indicators are largely stable in the aggregate and within-person. Within-person intertemporal rank correlations imply moderate stability and increase dramatically when using other biases as instrumental variables. Additional results reinforce three key inferences: biases are stable, accounting for classical measurement error in bias elicitation data is important, and eliciting multiple measures of multiple biases is valuable.
Keywords: behavioral biases; temporal stability; measurement error; consumer finance
JEL Codes: C36; C81; D90; E70
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| changes in financial conditions (G19) | changes in bias stability (C46) |
| subjective well-being (I31) | changes in bias stability (C46) |
| temporal stability (C41) | biases are largely stable at the aggregate level (D91) |
| within-person stability (C62) | biases are stable at the individual level (D91) |
| rank stability (C69) | individual biases do not fluctuate dramatically over time (D91) |
| measurement error (C20) | obscures true stability of biases (D91) |
| classical measurement error (C20) | accurate inferences about stability (C62) |