Left-Digit Bias: Theory and Empirical Evidence from a Tax Salience Experiment

summary

The gist

Demand often drops sharply at whole-dollar prices because consumers consider a good only if its price is below a heterogeneous attention threshold This paper presents a model of limited attention…

In short

This study uses a random utility model with attention-price thresholds to explain why demand drops sharply at whole-dollar prices—a phenomenon called left-digit bias. By analyzing experimental data on sales tax responses, the research shows that weak demand changes don't always mean low welfare costs, and it identifies how consumers' willingness to pay relates to their attention thresholds.

Key concepts

Random Utility Model with Attention-Price Thresholds (RUM–APT)
This model suggests consumers only consider a product if its price is below a specific 'attention-price threshold' (APT). The choice is determined by the minimum of the tax rate and this attention threshold, explaining demand drops at whole-dollar prices as a 'consider then choose' process.
Left-Digit Bias
This refers to the pattern where demand drops significantly when a price moves from one whole dollar to the next (e.g., from $10 to $11). The model explains this not just as price misperception but as consumers only considering goods whose prices fall below their personal attention threshold.
Debunching Probability
This measures the share of consumers whose purchase decision is affected by a tax, specifically those who are 'debunched' from a binding integer threshold. It quantifies how many consumers shift their consideration based on the tax rate, showing that larger taxes weakly reduce this mass.
Excess Burden Bounds
This analysis compares the welfare costs under two interpretations (RUM–APT and TRJ). It provides mathematical bounds showing that the excess burden under RUM–APT is greater than or equal to the excess burden under TRJ, illustrating that a weak demand response does not necessarily imply a small welfare cost.

Terminology used across episodes

This episode discusses

The paper

Left-Digit Bias: Theory and Empirical Evidence from a Tax Salience Experiment · Read on arXiv

Kaushil Patel

Demand often drops sharply at whole-dollar prices. To explain this phenomenon, known as left-digit bias, we develop a random utility model in which consumers consider a good only if its price is below a heterogeneous attention threshold. Masses of consumers with integer attention thresholds generate demand discontinuities. When a tax is introduced, the model predicts a weaker demand response compared to rational choice for left-digit biased consumers. Using data from a tax experiment, we find that the model matches behavioral moments and partially identify the distribution of willingness to pay relative to thresholds. A weak demand response need not imply a low excess burden of taxation and understanding the specific mechanism is important for welfare analysis.

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Left-Digit Bias: Theory and Empirical Evidence from a Tax Salience Experiment".

Warren: Here's the gist of the paper. Demand often drops sharply at whole-dollar prices because consumers consider a good only if its price is below a heterogeneous attention threshold This paper presents a model of limited attention that offers a novel explanation for left-digit bias Using rich data from an experiment studying how consumers respond to sales taxes, we test empirically the model’s behavioral implications and partially identify the relationship between consumers’ willingness to pay and attention Model Formalization The paper introduces a random utility model with attention-price thresholds (RUM–APT), where consumers consider a good only if its price is below an attention-price threshold (APT) The choice procedure is summarized via a posted-price cutoff, min(T, V) This model produces a demand discontinuity at an integer price if there is a mass of consumers with an attention threshold at that integer and willingness to pay greater than the threshold T p), where C = min(T, V) Empirical Identification and Moments The experiment of Taubinsky and Rees-Jones (TRJ) studies how consumers respond to sales taxes The model predicts that a weak demand response need not imply a low excess burden of taxation and understanding the specific mechanism is important for welfare analysis Population moments at finitely many tax rates allow partial identification of the overshoot distribution For example, between 43.8 and 52.2 percent of binding-threshold consumers have willingness to pay no more than 10 percent above their threshold Tax Response and Debunching Proposition 4.2 states that a tax can only reduce the mass at a binding integer threshold, and a larger tax weakly reduces more of that mass The share of debunching consumers is given by 1 − Jn(τ) / Jn(0) = P(V/T ≤ 1 + τ T = n, V > n) Proposition 4.4 shows that the tax response ψA(τ) is weakly increasing in τ and strictly increasing once τ > R − 1 The log threshold response measures the log cutoff change per unit of the tax rate Welfare Analysis and Bounds The paper compares the RUM–APT and TRJ welfare interpretations for a binding-threshold consumer The distribution of excess burden under RUM–APT first-order stochastically dominates the distribution of excess burden under TRJ Proposition 8.2 provides bounds on the mean excess burden, showing that 1 ≤ E

EBAPT(p): / E

EBTRJ(p): ≤ τp / (n − p) Identification Results Proposition 5.2 shows that the debunching probability identifies the share of consumers with valuation overshoot no greater than 1 + τ The mean log threshold response identifies the area under the CDF up to that point These restrictions jointly identify the distribution of log R, and equivalently R, for binding-threshold consumers Discussion and Welfare Implications The paper discusses how the model explains left-digit bias as a "consider then choose" explanation using natural attention thresholds The excess burden under RUM–APT is greater than under TRJ for consumers whose purchase is eliminated by the tax This illustrates why a weak demand response need not imply a small welfare cost Estimation Strategy The empirical strategy connects reported BDM cutoffs to the model by focusing on exact integer reports in untaxed Module 2 The estimate for gross debunching is derived using Assumption 6.3, which relates the observed persistence rates to structural debunching probability The common overshoot distribution is partially identified by jointly explaining the mean log tax response and gross debunching moments across experimental arms Quantitative Bounds The final results use the partially identified overshoot distributions to calculate welfare bounds in Section 8.4 The switching probability and mean excess burdens are computed by linear programming over candidate overshoot distributions These bounds substantially narrow the theoretical bound of 1–39.9 obtained from Proposition 8.2 The model permits both preference and attention heterogeneity and provides a new explanation for demand discontinuities at whole-dollar prices The identification result in Proposition 5.2 is limited in that it generally does not identify the joint distribution of the model primitives T and V This poses a challenge for welfare analysis References The paper cites numerous works related to consideration sets, two-stage choice procedures, and behavioral public economics,, and others.

Ruth: The references include works by Hauser and Wernerfelt (Hauser and Wernerfelt, 1990), Bernheim and Rangel (Bernheim and Rangel, 2009), Taubinsky and Rees-Jones (Taubinsky and Rees-Jones, 2018), Patel (Patel, 2025), List et al.

Warren: (List et al., 2023), Schindler and Kibarian (Schindler and Kibarian, 1996), Gabaix and Laibson (Gabaix and Laibson, 2006), and others.

Ruth: The references also include Ramsey (Ramsey, 1927) and Rubinstein and Salant (Rubinstein and Salant, 2012).

Title and authors: Ruth: Now we move into the title and authors of this paper, "Left-Digit Bias: Theory and Empirical Evidence from a Tax Salience Experiment." We’re going to unpack what that title actually tells us about the research.

Warren: It immediately points to two big things: the phenomenon known as left-digit bias, and that they are using an experiment involving sales taxes to test their theory Warren.

Iris: The title suggests a connection between how we perceive prices—that whole dollar price effect—and a specific economic mechanism tied to tax salience Iris.

Ruth: Basically, they’re saying that the sharp drops in demand at whole dollar prices aren't just random noise; they are linked to people only considering something if the price is below their attention threshold Ruth.

Warren: They are using the term "tax salience" because that's what drives their model—the tax makes people pay more attention to the price cutoff, and that’s why they test it under different tax scenarios Warren.

Iris: The authors are Kaushil Patel and they’ve built a random utility model in this paper to explain how limited attention causes these demand discontinuities Iris.

Ruth: So, the key takeaway is that the whole point of this paper is to link consumer psychology—specifically limited attention—to observable macroeconomic patterns like whole-dollar price demand drops Ruth.

Warren: It’s about taking a psychological concept and trying to ground it in empirical economic data through these tax experiments Warren.

Iris: The authors are testing if this model can actually match the observed behavioral moments, which is a big hurdle in this kind of identification Iris.

Ruth: They aren't just proposing a new model; they are using real data to test its behavioral implications and see how well it holds up under tax conditions Ruth.

Warren: It’s about empirical testing of the RUM–APT model in a setting where taxes create specific salience effects Warren.

The paper's summary: Ruth: So, what is the actual summary of the paper? Essentially, it lays out how this limited attention model predicts that when a tax is introduced, consumers react weakly compared to rational choice for people who are left-digit biased Ruth.

Warren: They then explain *why* they think this happens: because the model assumes a consumer only considers the good if its posted price is below their attention threshold and their willingness to pay Warren.

Iris: The paper summarizes how they test this by looking at two reported cutoffs for each consumer and good, depending on whether a sales tax is applied or not Iris.

Ruth: They then introduce the RUM–APT model where a consumer facing an ad valorem tax at rate tau chooses the good if its posted price is less than their attention threshold minus their willingness to pay divided by one plus tau Ruth.

Warren: That’s the formula, and they say this makes sharp predictions about these cutoffs, particularly under the condition that there is no sales tax Warren.

Iris: They also summarize that a consumer’s cutoff moving below their binding threshold is called debunching, and the model predicts that this mass at binding thresholds weakly decreases as the tax rate rises Iris.

Ruth: The summary points out that in Section five, they ask what can be learned about consumer preferences from these cutoff data, and they state that tax response and debunching are governed by the same distribution Ruth.

Warren: So, to put it simply, it’s a framework where you use tax experiment data to partially identify consumer willingness to pay relative to their attention thresholds Warren.

Iris: The summary confirms that they find that the model matches behavioral moments but also reveals how changes in reported cutoffs can reveal information about the underlying valuation structure Iris.

Ruth: The main thing is they show that understanding this specific mechanism is important for welfare analysis, even when demand response to taxes seems weak Ruth.

Warren: Because if you only look at the tax effect on demand, you miss the deeper structural reason why things are happening Warren.

The paper's improvements: Ruth: The authors suggest a few ways they can improve this work. They suggest focusing on how to make the identification of willingness to pay relative to attention thresholds even clearer Ruth.

Warren: Specifically, they suggest using observed changes in reported cutoffs under different tax regimes as a way to identify the no-tax willingness to pay of consumers who are moving their cutoff Warren.

Iris: That's a clever idea because it suggests that when you see a cutoff move under a known tax rate, you can infer the no-tax willingness to pay of that mover Iris.

Ruth: And they also suggest they should look at how the model predicts the movement of these cutoffs at noninteger cent prices Ruth.

Warren: That’s testing whether taxation actually generates some kind of "excess mass at a noninteger cent price," which would be predicted by their model Warren.

Iris: If they can rule out this relocation into other integer thresholds, that would strengthen the identification significantly by ruling out that possibility Iris.

Ruth: So, essentially, the suggested improvements are about pushing the identification to be more precise and checking for these subtle effects at noninteger prices Ruth.

Warren: It sounds like they want to ensure their model isn't just fitting data on integer points but is capturing the full behavior across all possible price points Warren.

Conclusion: Ruth: To wrap up, the paper "Left-Digit Bias: Theory and Empirical Evidence from a Tax Salience Experiment," the main implication is that this limited attention model provides a novel explanation for demand discontinuities at whole dollar prices Ruth.

Warren: It’s about moving beyond simple price effects by showing that consumers aren't just misperceiving the price but are actively choosing based on their attention threshold Warren.

Iris: The key finding is that even weak demand response to taxes doesn't automatically mean a low excess burden of taxation for those who buy things Iris.

Ruth: This means welfare analysis needs to focus on the specific behavioral mechanism driving the under-reaction, which is important because if you only look at tax effects on demand, you miss that structural reason Ruth.

Warren: In short, this paper gives us a framework to analyze how attention thresholds shape consumer choices in real economic settings Warren.

Iris: It’s a valuable contribution because it provides tools to test complex behavioral models against actual experimental data Iris.

More episodes

← Home