From a Tax Error to a Testable Prediction

The Dutch experiment now has a complete pre-test logic. A complicated schedule determines an actual marginal return; a respondent forms a belief about that return; the belief implies a perceived net wage; and the perceived wage can influence work hours only where a change in hours is feasible. This checkpoint connects those already-established ideas into predictions about overestimators, underestimators, and people whose beliefs are about right. It also clarifies why a randomized information comparison is needed before we call the relationship causal.

One chain, not four separate facts

The schedule is the starting condition, not the behavior itself. Taxes, credits, and allowances determine how much disposable income changes when gross income rises. That calculated change is the actual effective marginal return. The respondent may or may not know it. The gap between the calculated rate and the perceived rate is the initial belief error. If the respondent uses that perception when considering more work, the gap can alter the apparent payoff even though the official schedule remains unchanged.

The useful object for the decision is the perceived marginal net wage. Imagine two people looking at the same broad possibility of working an extra hour. One believes that almost all of the additional gross pay will be taken through tax and withdrawn benefits. The other believes that most of it will remain available. They may have similar preferences for leisure and similar jobs, yet the same hour presents different apparent returns. The first may decline an opportunity that would have been worthwhile under the actual schedule; the second may accept a low-return opportunity because it looks better than it is.

This does not turn the perceived net wage into the only determinant of hours. Work decisions also involve health, childcare, caregiving, study, employer schedules, and the value of time away from work. The experiment’s central claim is narrower: if a mistaken marginal-tax belief changes the perceived return, then correcting the belief may change a feasible hours choice. The word “feasible” carries much of the argument. A person who cannot add hours may have no behavioral margin on which a corrected calculation can operate.

The income pattern gives the chain a distributional direction. In the study’s baseline data, lower-income respondents were more likely on average to overestimate their marginal tax rate, while higher-income respondents tended to underestimate it slightly. An overestimate makes the perceived net return too low; an underestimate makes it too high. If the information intervention moves both groups toward the actual rate, the same correction can make extra work look more attractive to one group and less attractive to the other. Because the starting errors are not distributed evenly, the effects need not be distributed evenly either.

The chain also preserves an important distinction between accuracy and response. A person can hold a large error but have no interest in changing hours. A person can update a belief but remain unable to alter a schedule. A person can be classified as about right and consequently have little reason to update.