When a Belief Changes the Workweek
The Dutch experiment’s most direct result is a two-stage response: information brought some perceived effective marginal tax rates closer to the calculated rates, and the people who had initially overestimated those rates worked more hours. This chapter interprets that result as a mechanism rather than a slogan. It explains how large the belief correction was, why the response was asymmetric, how the study derives an implied compensated wage elasticity of about 0.3, and why the evidence supports a claim about work hours—not about tax rates, every kind of labor supply, or job preferences in general.
The correction was real, but incomplete
The researchers measure treatment effects separately for people who initially overestimated their effective marginal tax rate, underestimated it, or were already classified as about right. That separation follows the logic established earlier. If information is correcting a belief rather than exerting a common influence on everyone, the largest changes should appear among people who began with a material error, and the direction of any behavioral response should depend on the sign of that error.
The first outcome is accuracy. Among underestimators, absolute misperception fell by 3.7 percentage points from a baseline of 16.7 points. The reported p-value was 0.06. Among overestimators, absolute misperception fell by 6.0 points from a baseline of 43.7 points, with a p-value of 0.03. Respondents whose initial beliefs were about right showed no treatment change in misperception.
These figures say several things at once. The intervention moved beliefs in the intended direction, particularly for overestimators. The reduction was not a complete reset: a six-point improvement from a 43.7-point baseline still leaves a large average distance from the calculated rate. The underestimator estimate is smaller and less statistically distinguishable from zero under conventional thresholds. That should not be translated into “underestimators did not update.” It is more accurate to say that the study finds a smaller estimated correction with weaker statistical evidence for that subgroup.
A p-value is not a measure of practical importance, and it is not a switch that turns an estimate into a fact at 0.05. The overestimator result combines a sizeable estimated reduction with p = 0.03. The underestimator result combines a smaller estimated reduction with p = 0.06. The difference invites interpretation, but it does not prove that the two underlying mechanisms are fundamentally different. Subgroups can differ in baseline error, variation, sample composition, and responsiveness to a range-based message. The safest conclusion is that correction was more pronounced and clearer among overestimators in this study.
The asymmetry has an important starting point. Overestimators were not merely on the opposite side of a balanced error distribution; they were, on average, much farther from the calculated rate. Lower-income respondents were strongly overrepresented among them. A coarse information range can therefore produce a visible movement among people who have a large mistaken belief while producing little movement among people whose initial error is small or whose belief already falls within the communicated range.
The “about right” group provides a useful contrast. Their lack of treatment change in absolute misperception is consistent with there being less to correct. That result does not prove that the information had no effect on them in any psychological sense.