Behavioral Responses: Microeconomic Employment Effects

This post is part of a broader series explaining the modeling assumptions underlying our estimates. See here for other posts on this topic.

We produce several kinds of budget scores, ranging from conventional estimates (which do not account for substantive economic feedback) to dynamic estimates (which incorporate the effects of changes in the macroeconomy). Sometimes we produce budget estimates that fall somewhere between conventional and dynamic—estimates reflecting partial-equilibrium microeconomic responses.

One example of this kind of behavioral response is changes in labor supply. Traditionally, this kind of response is only included in dynamic scores: shifts in labor supply, after all, should feed through to the broader economy in general equilibrium, affecting wages, prices, and output. But it is often helpful to estimate the first-order, partial-equilibrium employment effects of a policy reform, especially when the policy change is targeted to a narrow subset of the overall population. This page describes our methodology for doing so in the context of tax policy reforms.

Tax policy affects the incentive to work through two channels. One is the income effect, wherein people may decide to work less in response to tax cuts because they can maintain the same standard of living despite working fewer hours. Economists generally believe income effects are small, especially for the range of policy proposals we typically consider. The other channel is the substitution effect: people may work less or quit working entirely if the return to working more falls due to an increase in marginal tax rates. In other words, taxes affect the cost-benefit calculation of working versus not working. The precise degree to which workers respond to changes in the return to work – the “participation elasticity” – is a matter of academic contention, but there is consensus that certain subgroups of workers are meaningfully sensitive. Parents and lower-income earners are generally thought to be more responsive than workers who are childless and/or high-income.

To incorporate labor supply changes in revenue estimates, we begin with the basic methodology laid out in Bastian (2023), a forthcoming National Tax Journal paper. The author assumes income effects are negligible but that substitution effects are nonzero and vary by income and demographic groups. Specifically, participation elasticities are assumed to be: 

• 0.4 for EITC-range unmarried working mothers
• 0.2 for all other mothers with AGI below $80,000.
• 0.05 for all others with AGI below $80,000.
• 0 for those with AGI above $80,000. 

Then, in our tax microsimulation model, for each worker, we compute the percent change in the “return to work,” defined as 1 minus the extensive marginal effective tax rate on realized earnings under the counterfactual policy change relative to current law. Finally, we apply participation elasticities to this quantity and aggregate by elasticity value to get the implied employment change per group. 

If the implied employment change for an elasticity group is negative, we choose simulated workers at random and set wages to zero such that post-adjustment employment matches implied labor force in expectation. Crucially, the probability of exiting the labor force is weighted in proportion to their contribution to the overall labor force change: a worker with a larger decline in the return to work has a higher probability of labor force exit than a worker with a smaller decline in the return to work.

If the implied employment change is instead positive, we add to the labor force by choosing simulated nonworkers in this elasticity group at random and assigning wages probabilistically based on the distribution of wages from workers who share demographic and economic characteristics.

The result is a simulated population which reflects the estimated new aggregate level of employment. As with all other behavioral feedback modules, we then re-calculate taxes using the post-adjustment simulated population, which generates the revenue feedback reflected in our final revenue estimates.

The code for this behavioral feedback module can be found here.