11.  Just Deserts

For an argument that the seventeenth century managed individually tailored lump-sum taxes, for example, in the 1660 Poll Tax, cf. Hahn (1973), 106.

Graaff (1957), 78.

Mitnik et al. (2015), 17–18.

This data is restricted, not classified, and thus requires significant investments of time, resources, and background verification and analysis to be accessed. Thus it is a future project.

“Labor part of farm income and business income, wages, bonuses, overtime, commissions, professional practice, labor part of income from roomers and boarders or business income.” Survey Research Center, Institute for Social Research, University of Michigan (2019).

Slottje et al. (2000), 350.

Pollin and Wicks-Lim (2013a,b). As such, our model would allow hyper-endowment individuals to keep some unmerited compensation. Nonetheless, over time, just as policy caused this shift during the Great Depression, one might sensibly predict that Just Deserts policy would be buttressed by policies to block the hyper-endowed from hiding their compensation.

Technically, social scientists would see such a model as a saturated model with a dummy variable for each individual.

For example, the standard federal income tax deduction for the blind may be viewed as such an expression with respect to responsibility. 26 U.S. Code § 63 - Taxable Income Defined (2000).

Nagel (1979), 35.

Please note that I implement a particular model for illustrative purposes only and do not suggest that this would be the actual “best” model for the economy of the United States for the calendar year 2016. It may turn out that frequentist methods are surpassed by Bayesian, black-box, or neural-net methods, it may turn out that generalized additive models perform better than linear models, it may turn out that multilevel models are superior to pooled models, and so on. The important thing is simply to compare one model with another and choose the “best” model.

Technically, we should call this a generalized linear model with an identity-link function, but this is the most popular, and well-understood statistical model available and is often called a linear model. As an agnostic proposal, I do not mean to suggest that this is the best model family for this problem, even less that this is the best model for this problem. I merely use a linear model as a first-order approximation to elucidate the proposal’s process and rules of selecting the best current annual model. Of course, given my brief discussion of decomposing birth-endowment into original resources and conversion factors, a multilevel linear model may outperform our first-order approximation. Similarly, should it turn out that Bayesian, machine learning black box, or any other method is “better” than frequentist statistics, then our adherence to the control principle demands we use such tools to make more accurate measurements of autonomous effort. Furthermore, note that some recent research suggests a more complicated model may underlie true earnings dynamics in the United States: earnings variance in the population is due to persistent factors (˜50%), transitory factors (˜30%), measurement error (˜20%), and a random walk component (˜?%). The transitory factors are serially correlated via a first-order autoregressive process (80%). Hyslop (2001); Mazumder (2001a); Although the PSID is too small, a much larger dataset of Canadian tax data has enabled researchers to further identify a random walk component, heterogeneity in age-earnings profiles, an age-based component to transitory variance, a time-varying permanent component. Baker and Solon (1999).

While it does put more pressure on measurement error, this focus on residuals follows standard econometric practice, cf. Bowles et al. (2001); Card and DiNardo (2002); Fields (2003); Katz and Autor (1999); Mazumder (2001b, 2003); Roemer (2002, 2013); Björklund et al. (2011).

Thus, as was argued in Chap. 2, we now formally and explicitly reject the contention of Dworkin and Fleurbaey that one’s subjective perceptions—identification with behaviors/outcomes—are related to descriptive responsibility. Dworkin (1981b); Fleurbaey (2008).

Administrative data will be far preferable to survey data when implementing the Just Deserts proposal. In the United States, this will require Internal Revenue Service or Social Security Administration (SSA) records. Cf. for a survey matched to SSA summary earnings records, Mazumder (2005).

Survey Research Center, Institute for Social Research, University of Michigan (2019).

Hill (1992), 4.

For administrative data, one might substitute for this step a filtering out of all individuals for whom one could not, or did not wish to, use multiple imputation.

Theoretical power comes at the cost of using very few years of mothers’ and fathers’ earliest lifetime compensation. It should be noted that the elasticity coefficient estimated from this brief number of years may exhibit attenuation bias of 10–40%. Once we have agreed that parents’ compensation is not caused by children’s compensation, we may improve our models even further. Cf. Mazumder (2001b). I would also like to have included neighborhood poverty rates based upon restricted-use PSID “geocodes,” but did not have access. Sharkey (2009).

Roemer leans, incorrectly in my view, toward more deserving rather than deserves more. Roemer (1998), 78.

The p-value of the studentized Breusch-Pagan test is 1.48e−24 prior to log transformation, 1.06e−01 afterward, and 1.32e−220 after including only the variables chosen by the stepwise AIC procedure on the log-transformed model.

One apparent weakness of our present data is that it measures gross labor income rather than net. Note that, “arguably, post-fisc [measures are] a more attractive objective, for it is a more proximate measure of consumption opportunities.” Roemer (2002), 6. I would argue, however, that this is the right measure as Just Deserts needs to occur prior to other taxes and transfers in order to measure autonomous effort.

Pace Roemer, note that this solves the problem of what units to use for autonomous effort. As our “best model” suggested, it cost identical autonomous effort to increase LC by 10%, regardless of Chance LC, this is a proper comparative unit of measure of autonomous effort. Had our “best model” suggested it cost identical autonomous effort to earn $1000 more in LC, regardless of Chance LC, then we would have used a unit (LC—Chance LC) instead. Roemer (2012), 179.

Although this does not occur in our three-person population, it does in our Panel Study of Income Dynamics sample.