Literature review

This paper draws on several intersecting literatures: tax salience and misperception, the welfare costs of tax uncertainty, optimal taxation with behavioral agents, the elasticity of taxable income, tax complexity and compliance costs, and the economics of information provision. I survey each in turn, highlighting how prior work motivates both the theoretical framework and the empirical calibration strategy of this paper.

Tax salience and misperception¶

The foundational insight that the presentation and visibility of taxes shapes economic behavior comes from Chetty et al. (2009), who combined field experiments and quasi-experimental evidence to show that posting tax-inclusive prices at a grocery store reduced demand by approximately 8 percent, despite no change in actual tax liability. Their theoretical framework formalized the notion that agents respond not to actual tax rates but to perceived tax rates, and that the wedge between the two generates both behavioral distortions and welfare losses distinct from those arising under perfect information. This insight---that how taxes are communicated matters as much as what they are---provides the conceptual foundation for the present analysis of misperception-induced deadweight loss.

Finkelstein (2009) provides complementary evidence from electronic highway tolling: when tolls became invisible (deducted electronically rather than paid at a booth), toll rates rose, suggesting that reduced salience allows governments to levy higher taxes --- a pattern consistent with the prediction that misperception of rates has welfare consequences.

Building directly on the “schmeduling” concept introduced by Liebman & Zeckhauser (2004), who first documented that taxpayers mentally linearize progressive tax schedules by treating their marginal rate as if it were their average rate, Rees-Jones & Taubinsky (2020) provided the first rigorous empirical measurement of such misperception. Using linked survey and administrative data, they found that approximately 43 percent of taxpayers employ an “ironing” heuristic, collapsing the graduated rate structure into a single perceived rate. For my calibration, their estimates imply a standard deviation of tax rate misperception of approximately $\sigma \approx 0.12$ around the true marginal rate---the key parameter governing the magnitude of welfare losses in my model. The fact that this misperception is systematic rather than random (with ironing biasing perceived rates toward the average rate) means that approximately 43 percent of the working population makes labor supply decisions on the basis of incorrect price signals.

Gideon (2017) provides complementary survey evidence: only 33.7 percent of Americans correctly identify that their marginal tax rate exceeds their average tax rate. The remaining two-thirds either believe the rates are equal (consistent with the ironing heuristic) or report the marginal rate as lower---patterns that confirm the prevalence of misperception documented by Rees-Jones and Taubinsky in an independent sample. Feldman et al. (2016) offer additional evidence from the Child Tax Credit, showing that confusion about the structure of a single tax provision generates large behavioral distortions, reinforcing the view that misperception is not confined to the overall rate structure but pervades individual provisions as well. Ballard & Gupta (2018) find that 85 percent of Michigan taxpayers overstate their average federal income tax rate, with a mean overstatement of 11.6 percentage points, suggesting that the misperception problem may be even more severe than the ironing heuristic alone would imply.

Tax uncertainty and welfare¶

The welfare cost of tax policy uncertainty---uncertainty about what future tax rates will be---was first quantified by Skinner (1988), who developed a two-period model in which agents must commit to savings decisions before the resolution of uncertainty about future tax rates. Skinner estimated welfare losses of approximately 0.4 percent of GDP annually, establishing that tax uncertainty imposes first-order costs comparable to major tax distortions. This paper complements Skinner’s analysis by studying a distinct source of uncertainty: agents’ imperfect knowledge of their current marginal tax rates. While Skinner’s costs arise from genuine randomness in the policy environment, mine arise from complexity-induced misperception of the existing tax code. The two sources of welfare loss are additive: reducing misperception does not eliminate policy uncertainty, and stabilizing policy does not help agents who cannot parse the current schedule.

Alm (1988) extended the analysis of tax uncertainty to individual decision difficulty, demonstrating that when agents cannot easily determine their marginal tax rates, they make systematically suboptimal choices even absent policy uncertainty. Hassett & Metcalf (1999) examined how tax rate uncertainty affects investment decisions, finding that uncertainty can either stimulate or depress investment depending on the structure of adjustment costs and the degree of irreversibility---highlighting that uncertainty effects are not uniform across different margins of economic decision-making. More recently, Baker et al. (2016) constructed a comprehensive economic policy uncertainty index and showed that uncertainty shocks reduce investment and employment, with tax policy uncertainty accounting for a large share of the overall effect. Their macroeconomic evidence supports the theoretical prediction that tax uncertainty has first-order welfare consequences.

Optimal taxation with behavioral agents¶

The classical optimal income tax literature, inaugurated by Mirrlees (1971), assumes that agents have perfect information about tax schedules and optimize accordingly. Diamond & Saez (2011) provide a modern synthesis of optimal income taxation that retains this full-information assumption while incorporating empirical estimates of behavioral elasticities. Saez (2001) formalized the sufficient-statistics approach to optimal income taxation, showing that the optimal tax schedule depends on the elasticity of taxable income and the shape of the income distribution. The present paper applies an analogous sufficient-statistics logic: the welfare cost of misperception reduces to three observable quantities. Yet a growing body of work has begun to relax the perfect-information assumption in substantive ways.

Most directly relevant, Farhi & Gabaix (2020) develop a general framework for optimal taxation when agents exhibit behavioral biases, including misperception of tax rates. Their sufficient-statistics approach modifies the standard optimal tax formula by introducing a “behavioral wedge” that accounts for the divergence between actual and perceived marginal rates. In their notation, the optimal tax formula includes a correction term proportional to the covariance between the behavioral wedge and the social marginal welfare weight. When misperception is uncorrelated with income (as in the baseline specification here), their framework predicts that the optimal tax rate falls---consistent with the results below. The present paper contributes a specific quantification of this channel: using the Rees-Jones and Taubinsky estimates of $\sigma$, I derive a closed-form expression for the additional deadweight loss from misperception and show how it alters the optimal tax rate under a simple linear tax. The linear-tax setting permits closed-form solutions and transparent comparative statics, complementing the greater generality but reduced tractability of the Farhi-Gabaix framework. Gerritsen (2016) reaches a complementary conclusion from a different modeling approach, showing that when agents do not maximize well-being (due to misperception or other behavioral frictions), the optimal nonlinear tax schedule differs systematically from the Mirrlees benchmark. Chetty et al. (2009) provide a behavioral framework for evaluating when policymakers should correct versus exploit misperceptions, concluding that correction is generally welfare-improving when misperception is unrelated to the policy objective.

Elasticity of taxable income¶

The deadweight loss formula, $\mathbb{E}[\text{DWL}]/\text{earnings} = \tfrac{1}{2}\varepsilon\sigma^2/(1-\tau)$, is structurally analogous to the Harberger triangle expression for the standard DWL of taxation, but applied to the variance of perceived rates rather than the level of the tax rate. The empirical content of the formula depends critically on the labor supply elasticity $\varepsilon$.

Feldstein (1999) pioneered the elasticity-of-taxable-income (ETI) approach to measuring the efficiency cost of taxation, arguing that the ETI captures all behavioral margins of response---not just hours worked but also effort, occupation choice, tax avoidance, and evasion. Saez et al. (2012) provide a comprehensive review of ETI estimates, concluding that a central value of approximately 0.25 is appropriate for the U.S. income tax, though estimates range widely depending on the population and identification strategy. For my calibration, I draw on Chetty (2012), who reconciles the apparent conflict between small micro elasticities (from tax reform studies) and large macro elasticities (from cross-country comparisons) by showing that optimization frictions attenuate short-run responses. His preferred Frisch elasticity estimate of 0.5 on the extensive margin, combined with an intensive-margin elasticity near 0.33, informs my baseline $\varepsilon = 0.33$. Keane & Rogerson (2012) reach a similar conclusion through structural estimation, arguing that properly accounting for human capital, progressive taxation, and life-cycle dynamics yields macro-consistent elasticities from micro data.

Tax complexity and compliance costs¶

The welfare costs of tax complexity extend beyond misperception of rates to include direct compliance costs and reduced take-up of beneficial provisions. Slemrod (2005) documents that U.S. state income tax systems vary widely in complexity and that this complexity imposes real resource costs on taxpayers and administrators---costs that function as a hidden tax. Goldin (2018) shows that complexity in the Earned Income Tax Credit reduces take-up among eligible households, meaning that the intended redistribution is partially undone by the very mechanism designed to deliver it. Benzarti (2020) estimates compliance costs using revealed preferences---observing that taxpayers forgo substantial tax savings rather than itemize deductions---and finds costs larger than survey-based estimates suggest.

Chetty & Saez (2013) provides the closest empirical analog to the mechanism in my model. In a field experiment, they provided EITC recipients with information about how the credit’s phase-in and phase-out affect their marginal incentives. Recipients who received the information adjusted their earnings in subsequent years, demonstrating that misperception of marginal rates has real behavioral consequences that can be corrected through information provision. This result directly supports my conclusion that reducing $\sigma$ through better information generates welfare gains. Abeler & Jäger (2015) complement this finding with laboratory evidence showing that subjects respond less to tax incentives when the tax schedule is complex, even when they have access to all relevant information---suggesting that cognitive costs of processing complex schedules are themselves a source of effective misperception.

Information provision and economic decisions¶

A broader literature has established that correcting misperceptions generates first-order welfare improvements across many domains, including retirement savings Duflo & Saez, 2003, school choice Hastings & Weinstein, 2008, and returns to education Jensen, 2010. My analysis extends this logic to the tax domain, where misperception of the marginal tax rate directly distorts the perceived return to labor supply.