A welfare evaluation of the 1986 tax reform for married couples in the United States

As the EITC depends on family income, it is expected to stimulate the labor supply at the extensive margin but, in the phase-out region, it should reduce incentives at the intensive margins. Empirical studies find that the EITC expansions had heterogeneous effects on labor supply, depending on family composition. On the one hand, Dickert et al. (1995), Eissa et al. (1996), Ellwood (2000) and Meier and Rosenbaum (2000, 2001) show that the EITC reforms increased the workforce participation of single women with children, since it made work more attractive. On the other hand, Ellwood (2000) and Eissa and Hoynes (2004) find that the EITC expansion between 1984 and 1996 reduced the total family labor supply of married couples. Overall, the positive effect on labor force participation seems to be large enough to offset the negative impact at the intensive margin in working hours (Scholz 1996).

As an efficiency measure they use the excess burden of taxation, i.e., the sum of how much each individual would be willing to pay to get rid of all taxes and transfers.

In a similar fashion, Piketty and Saez (2007) analyze changes in the progressivity of the federal income tax over time but cannot disentangle policy changes from other factors.

Bargain et al. (2013), Carpantier and Sapata (2016) and Decoster and Haan (2015) also use well-being indexes that are similar to those which we employ here. They do it, however, with different purposes: while Bargain et al. (2013) focus on a cross-country comparison of well-being, the other two studies deal with the sensitivity of different well-being rankings (in the USA and Germany) to different normative principles.

If we run the same simulation for earnings larger than $200,000, we arrive at the same conclusion. The only difference is that the reduction in the taxes at federal and state level induced by TRA86 is increasing with earnings in absolute value.

A deeper analysis of the indexes presented here is in Fleurbaey (2008), Fleurbaey and Maniquet (2011) and Fleurbaey and Blanchet (2013).

A general normative characteristic of money-metric utilities is that they do not depend on the initial cardinalization \(U_i\) of individual preferences \(R_i\), as the same value \(m_i(.)\) would be obtained with any increasing transformation of \(U_i\) (Fleurbaey and Blanchet 2013). As a consequence, the comparison of individual situations does not depend on the initial cardinalization of \(U_i\).

As a matter of fact, relying on the social welfare functions proposed by Fleurbaey and Maniquet (2011), just knowing the properties of the tax function one could assess the goodness of a reform. Of course, as explained in Fleurbaey and Maniquet (2006), the assessment of a tax-benefit system depends on the value judgments pertaining to the treatment of preference heterogeneity and they specifically discuss the optimality of an EITC-type of scheme. Their result, however, crucially relies on the fact that they use a social welfare function of the maximin type. Hence, in order to assess a reform and to know which part of the tax function should be changed first is sufficient to know the earnings of the worst-off individual.

A similar sample selection can be found in the methodological contribution of van Soest et al. (2002), Eissa and Hoynes (2004), Bargain et al. (2013) and Decoster and Haan (2015).

See Dagsvik and Strøm (2006) for an evaluation of the performance of the Box–Cox functional form in analyzing labor supply behavior compared to less parametric (polynomial) specifications as studied by van Soest et al. (2002).

See van Soest et al. (2002) for a discussion on the advantages and limits of the discrete labor supply model with respect to the continuous one.

Leisure time enters the utility function through l, which is wife’s fraction of time not devoted to work. More precisely, \(l\equiv 1-h=1-H/2500\). Hence, l is equal to 1 if \(H=0\) and is equal to 0 if \(H=2500\).

It is worthy to remark that the parametric assumption of the random component \(\varvec{\varepsilon }\) is not restrictive. McFadden and Train (2000) showed indeed that under mild regularity conditions any discrete choice model derived from random utility maximization can be approximated by EV-I errors in combination with a nonparametric specification of unobserved individual characteristics determining preferences.

Note that \(v_y\) and \(v_l\) are independent if and only if \(p_1p_2=p_3p_4\) (van den Berg et al. 1994; van den Berg and Lindeboom 1998), making it easy to test for the independence of the unobserved components.

In the selection equation, we include also the number of children younger than 17, an indicator for the presence of children younger than 6 and the number of family members.

We used the Internet Taxsim (v9). See Feenberg and Coutts (1993) for an introduction to TAXSIM. Further information can be found in Internet at http://​www.​nber.​org/​taxsim/​. TAXSIM has already been used in the empirical studies of Eissa et al. (2008), Bargain et al. (2014) and Bargain et al. (2013).

We considered social security benefits as independent of wives’ working hours choice. In each year of our longitudinal dataset, less than 1% of the families received social security benefits. From 1984 to 1990, only 2.4% of the families (61 out of 2529) received social security benefits in at least one year in which they were in the sample. We instead did not include in the budget constraint food stamps and the Aid to Families with Dependent Children declared in the PSID, since they depend on family income (and therefore on wives’ working hours choice) but, differently from social security benefits, they are not handled by TAXSIM.

The PSID is produced and distributed by the Institute for Social Research, Survey Research Center, University of Michigan, Ann Arbor, MI.

Further information on the size of the sample used for the estimation of households preferences is provided in Appendix A.1 together with descriptive statistics.

We describe the simulation algorithm to obtain the figures in Table 2 in Appendix A.4.

Further details and labor supply elasticities for different demographic groups are in Table 14, Appendix A.3.

See also Bargain et al. (2016) for a recent survey on estimated labor supply elasticities in the USA and Western Europe.

Appendix A.5 describes the simulation algorithm to obtain the welfare metrics under the actual 1986 fiscal law and the counterfactual scenario with the 1988 fiscal rules.

See Table 13 in Appendix A.4.

\(8.3\%=[\exp (0.080)-1]\times 100\).

More details are provided in Appendix A.1.

The net income is deflated to 1990 prices using the consumer price index by the US Bureau of Labor Statistics and available at ftp://​ftp.​bls.​gov/​pub/​special.​requests/​cpi/​cpiai.​txt.

The notation in this section is inspired by the one in Wooldridge (2010, Sect. 19.9.2).

We do not report the estimation result of the model without unobserved heterogeneity for the sake of brevity. They are available from the authors upon request.

We also tried to include time dummies in the linear preference terms of leisure and net income. Since their coefficients were not jointly significant at the usual 5% statistical level and the goodness of fit did not show any relevant improvement, we removed the time dummies from the set of covariates to save in degrees in freedom.

Since the unobserved heterogeneity \({\mathbf {v}}_i\) is a random term at household level, at each replication of the simulation algorithm, each individual is assigned a new vector \((\widehat{v}_{yi},\widehat{v}_{li})\) of unobserved characteristics.