I. Introduction

Legislation adopted in California in 2020 restricts the calculation of economic damages in personal injury and wrongful death cases such that the amount cannot be “reduced” based on a three-factor mix that includes race, ethnicity, or gender.¹ The goal of the legislation, presumably, is to address perceived injustices in which economic damages for members of protected classes are suppressed when race, ethnicity, and gender are taken into account. In this paper, we address the extent to which race-neutral, ethnicity-neutral, and gender-neutral approaches (“neutral models” or “neutral approaches”) correct for such perceived injustices by yielding a fair measure of damages that makes the plaintiff whole. The analysis is critical from a public policy standpoint because some members of a protected class might be adversely impacted by legislation that is intended to protect their individual right to be made whole through a fair and accurate calculation of economic damages.²

Litigation cases involving the expected earnings of minor children—where individual earnings have yet to be revealed—provide an ideal environment to examine the implications of public policies that attempt to mitigate systemic racial, ethnic, and gender discrimination in the calculation of economic damages. In this paper, we compare economic damage awards in cases involving minor children when race, ethnicity, and gender are taken into account and when race, ethnicity, and gender are not taken into account. Comparisons are done for specific types of individuals, based on demographic and economic attributes. The key questions then become: 1) are economic damages awards for members of protected classes (based on race, ethnicity, and gender) necessarily increased when race, ethnicity, and gender are removed from consideration, and 2) if economic damage awards are reduced, are individual members of protected classes meaningfully adversely impacted by neutral approaches?

The underlying question from a public policy standpoint is whether there are unintended consequences—adverse impacts to individual members of protected classes—from requiring forensic economists to use race, ethnicity, and gender-neutral modeling approaches when calculating economic damages? If so, should such requirements on forensic economists exist? The unintended effects create a tradeoff. The tradeoff is between 1) the aggregate economic benefits to the entire class from neutral modeling, and 2) the negative impact of reducing economic damages for some members of the protected class. Notably, this tradeoff extends beyond the issue of requiring an individual defendant to overcompensate an individual plaintiff to correct for a society-wide disparity, which raises the prospect of introducing another systemic injustice. This latter issue is not addressed in this paper.

Our paper is structured as follows. Section II provides background on recent legislation related to the removal of considering race, ethnicity, and gender in the calculation of economic damages awards. Section III presents our data and methodology for calculating economic damages with and without consideration of race, ethnicity, and gender. Section IV presents our findings, and Section V provides a discussion and conclusions.

II. Background

Prior to the introduction or adoption of any law dealing with discrimination in litigation, The Washington Post (Soffen, 2016a, 2016b) published a series of articles stating that a male child will receive a bigger award than a female child with the same injury based, apparently, on several anecdotes of verdict outcomes for children. The focus of The Washington Post article was on children because children lack earnings histories and, therefore, the forensic economist must use general statistical data to determine economic damages. Only anecdotes were provided in the article, however, and no statistical basis was provided for the claim that disparities in awards by gender were the product of systemic gender discrimination embedded in historical data sources.³

Shortly after The Washington Post articles, the House of Representatives and Senate introduced the Fair Calculation in Civil Damages Act on December 1, 2016.⁴ A flurry of press releases that followed, several citing The Washington Post’s articles, indicated the political importance of the issue. Senators Cory Booker (D-NJ) and Kirsten Gillibrand (D-NY), co-sponsors of the Senate version, justified the legislation stating, “troubling recent academic studies and news reports have revealed that in state and federal courtrooms across the United States, race, ethnicity and gender are used to calculate damages in civil court cases,” and that “white and male plaintiffs often receive larger compensatory awards than plaintiffs who are minorities or women.”⁵ The sponsors’ statements about “troubling recent academic studies” made no mention of what those academic studies are. The news reports were presumably referring to The Washington Post articles.

The House bill sponsored by Joe Kennedy III (D-MA) and Mia Love (R-UT) also included a press release on December 1, 2016.⁶ The press release referenced a 2009 survey by the National Association of Forensic Economics (NAFE) and claimed that, “44% of those responding reported they considered race and 92% consider gender when calculating potential lost earnings for injured children.”⁷ Neither the House or Senate bills passed and, following the failed attempt to address disparities at the national level, California passed §3361 and the bill became law on January 1, 2020.

Washington, D.C. also attempted to address the claimed disparity of earnings due to perceived discrimination with Bill 23-087. This bill was named the “Stormiyah Denson-Jackson Race and Gender Economic Damages Equality Amendment Act of 2020.” The language of the D.C. bill was similar to that of California §3361. In particular, the bill stated that its purpose is to “amend Title 16 of the District of Columbia Code in actions for personal injury or death caused by wrongful act, neglect, or default, to provide estimations, measures, or calculations of past, present, or future damages for lost earnings or impaired earning capacity may not be reduced based on race, ethnicity, or gender.” Bill 23-087 was not considered by legislators and was agendized.

Perhaps the most important issue in these legislative initiatives is that legislators, taken at their word, believe that racial, ethnic, and gender discrimination exists in economic damages awards made in civil litigation, and that the law should provide a remedy for the discrimination. Lawrence M. Spizman, Kevin E. Cahill and John Kane (2022) addressed this topic by examining the impact on individuals in protected classes from imposing race- and ethnicity-neutral models—that is, taking into account two factors (race and ethnicity) of the three-factor (race, ethnicity, and gender) mix. In particular, the authors examined earnings losses for a minor child by removing race and ethnicity from the ordered probit educational attainment model for the minor child. The race- and ethnicity-neutral model was then compared to the ordered probit educational attainment model of a minor child that included race and ethnicity. The authors found that lifetime earnings under the race-neutral and ethnicity-neutral models were not exclusively higher or lower than the models that took race into account.

This paper extends the research of Spizman, Cahill and Kane (2022) in two important ways. First, this paper takes into account all three factors (race, ethnicity, gender) included in recent legislation that require “neutral” calculations. Spizman, et al. focused on race and ethnicity. Second, this paper incorporates race, ethnicity, and gender-specific earnings amounts for each educational attainment category. We, therefore, apply race, ethnicity, and gender-specific values to both the educational attainment probabilities and the earnings amounts applied to those probabilities.

Gender is of particular interest because of the well-documented gender wage gap. Francine D. Blau and Lawrence M. Kahn (2017), for example, provide a comprehensive review of the literature on the gender wage gap, and the extent to which the gap reflects discrimination on the basis of gender. More recent studies have furthered the state of the literature by focusing on how the gender wage gap has changed over time and across cohorts. Blau, Kahn, Nikolai Boboshko and Matthew L. Comey (2021) document a substantial convergence in gender wage gap between 1981 and 2015, with a portion of the gap still unexplained by observable factors. William E. Even and David A. Macpherson (2020) examine whether differences in gender earnings are due to gender discrimination in labor markets or gender differences in preferences and find that most of the gender wage gap among recent cohorts can be explained by observable factors. Most recently, Vanessa Burbano, Nicolas Padilla, and Stephen Meier (2024) find that gender differences in preferences for meaning at work contribute to occupational segregation by gender.

Using a different approach, Patricia Cortés and Jessica Pan (2020) explore the impact of having children on females’ earnings relative to those of men, known as the “child penalty.” Cortés and Pan (2020) find that the child penalty five to 10 years after the birth of a first child can be as high as 40%. More generally, Amanda Barroso & Anna Brown (2021) discuss how mothers are twice as likely as fathers to experience negative impacts on their jobs and careers because they must take time off after the birth of a child. While these types of studies potentially identify a factor that explains gender differences in earnings, the research does not address whether discrimination might be embedded in family decisions or societal structures surrounding child rearing.

This research on the gender wage gap provides a strong justification to extend the work by Spizman, Cahill and Kane (2022) to include gender-neutral models, and to examine differences in earnings as well as educational attainment probabilities. The results, and the potential unintended consequences, can be used by policymakers when considering legislation to account for the impact of race, ethnicity, and gender in the estimation of earnings.

III. Data and Methods

Our estimation procedure for the calculation of lifetime earnings consists of two steps: 1) estimating educational attainment probabilities, and 2) estimating expected lifetime earnings for each educational attainment level. Both steps are done with and without taking into account race, ethnicity, or gender. Regarding the first step, estimated educational attainment is based on the Lawrence M. Spizman and John Kane model (2020). The Spizman-Kane model is a widely-used, ordered-probit approach to predict the probabilities of alternative levels of educational attainment for a minor child.

The Spizman-Kane model assigns individual probabilities to each educational attainment category, such as whether a student completes high school, attends college or receives a graduate degree.⁸ The model for the ordered probit specification is: $$Z_i=X_i\beta +{\mu }_i$$

The unobservable variable Z_i represents the benefits and/or costs of different levels of educational attainment. The vector X_i represents family background and demographic variables that influence Z_i. Since the variable Z_i is unobservable, an indicator variable (dummy variable) is used to show the actual educational level for everyone in the sample. It is assumed that individual i acquires:⁹

• less than a high school degree if $Z_i\le 0_{\mathit{1}}$

• GED if $0_{\mathit{1}}\hspace{0.17em}<\hspace{0.17em}{\text{Z}}_i\le 0_{\mathit{2}}$

• High school diploma if $0_{\mathit{2}}<Z_i\le 0_{\mathit{3}}$

• Associate’s degree if $0_{\mathit{3}}<Z_i\le 0_{\mathit{4}}$

• Bachelor’s degree if $0_{\mathit{4}}<Z_i\le 0_{\mathit{5}}$

• Master’s degree if $0_{\mathit{5}}<Z_i\le 0_{\mathit{6}}$

• Ph.D. degree if $0_{\mathit{6}}<Z_i\le 0_{\mathit{7}}$

• MD, JD, or DDS degree if $Z_i>0_{\mathit{7}}^1$¹⁰

The estimated coefficients from the ordered probit are then used to determine the probability of the minor child obtaining each different educational level as his or her highest level of educational attainment.¹¹

Since the first publication of Spizman Kane (1992), other studies using the model have verified the robustness of the original study as well as the data sources used (Spizman Kane (2001, 2020), Andrew M. Gill and Jack Foley (1996), Christopher Jespen and Lisa Jespen (2001), Christopher Bruce and Carmen Anderson (2006), John Kane, Lawrence M. Spizman, James D. Rodgers, Rick R. Gaskins (2010), John Kane, Lawrence M. Spizman and Don Donelson (2013), and Eric W. Christensen (2020)). The 2020 version of the Spizman-Kane model takes into account race, ethnicity, and gender as well as data from the most recent wave of the National Longitudinal Survey of Youth (NLSY). For the neutral model, we use a version of the Spizman-Kane model from Spizman, Cahill, and Kane (2022). The 2022 version of the model differs from the original Spizman-Kane ordered probit model in that it provides estimates that exclude race and ethnicity (race/ethnicity-neutral model), with gender as a dichotomous independent variable. In both models, the levels of educational attainment are as follows: 1) less than high school diploma, 2) GED, 3) high school diploma, 4) associate degree, 5) BA or BS degree, 6) master’s degree, and 7) PhD degree or professional degree.

Once the educational probabilities are estimated for each educational attainment category using the Spizman-Kane model, the next step in the calculation is to estimate expected lifetime earnings by race, ethnicity, and gender for each of the educational attainment categories. In order to determine those earnings we obtain data on earnings from the 5-year Public Use Microdata Sample (PUMS) of the American Community Survey (ACS), a dataset containing information on approximately a 5% sample of the total United States population (U.S. Census Bureau, 2022).¹² We use the ACS data to calculate earnings in 2019 (a 1% sample) by gender (male, female), race (Black, White, Other), and ethnicity (Hispanic, not Hispanic), for a total of 12 combinations. For each combination, we use total person’s earnings for each educational attainment level (specified above) and take the average within 10-year age categories (18-24, 25-34, 35-44, 45-54, 65-74, 75-84). We then conduct a linear extrapolation between each of the age categories to obtain an estimate of earnings at each age. Estimated lifetime earnings for each level of educational attainment is therefore the summation of earnings through worklife expectancy.

We make two important assumptions about the calculation of lifetime earnings for simplicity and expository purposes. First, by summing across all ages we implicitly assume a zero net discount rate. This approach provides a straightforward way to estimate lifetime earnings for each educational attainment level by race, ethnicity, and gender categories. We also do not adjust earnings by a scaling factor to account for fringe benefits. While such adjustments are necessary and no doubt important in practice when quantifying economic damages for an individual child, a more sophisticated approach involving a wage growth rate and discount rate or a precise fringe benefit rate would be unlikely to impact our conclusions in any meaningful way.¹³ The reason is that earnings under both the neutral and non-neutral models would be subject to the same adjustments, and a comparison between the two models is the focus of this paper.

Second, estimated lifetime earnings of a child are calculated using worklife expectancy (Gary R. Skoog, James E. Ciecka, and Kurt V. Krueger, 2019). Worklife expectancy is available by educational attainment levels and by gender, but not by race or ethnicity. As such, we do not have race- or ethnicity-adjusted worklife expectancies; instead, we use gender-specific worklife expectancies for each educational attainment category, assuming the individual is initially active. Unlike the net discount rate and fringe benefit issues identified above, the inability to adjust worklife expectancy by race and ethnicity could potentially impact the conclusions of our analysis. The reason is that the race-neutral model would presumably produce different estimated lifetime earnings amounts for a child than the race-specific earnings amounts. The discrepancy between the two would be a function of how large the worklife expectancy gap is by race and ethnicity.

The total estimated lifetime earnings of a child is calculated by summing the product of the estimated lifetime earnings amounts for each educational attainment category and the probability of achieving that level of educational attainment. We perform this calculation for the 12 combinations of gender, race, and ethnicity using both the neutral (not stratified by race and ethnicity) and non-neutral models (stratified by race and ethnicity). We then compare the estimated lifetime earnings amounts of a child under the neutral model and the non-neutral model. We do this calculation for different combinations of the right-hand side variables in the Spizman-Kane (2020) model: region (urban, rural), parent’s educational attainment, both biological parents in household, mother’s age at first birth, religion (raised), number of siblings, and family income. Note, there are several variables that influence the probability of different educational levels. In total, we examine approximately 3,000 combinations of right-hand side variables for females, with more than 1,000 each for White, Black, and Hispanic individuals. We then run the same number of calculations for males. In addition to various combinations of the categorical right-hand side variables (e.g., region, parents’ education, religion), we also examine specific values of the continuous variables (e.g., family income of $45,000, mother’s educational attainment of 12 years).

Finally, we assess whether excluding race, ethnicity, and gender results in a gain or loss in estimated lifetime earnings of a child. A gain in estimated lifetime earnings implies that the neutral model yields a higher amount than the non-neutral model that includes race, ethnicity, and gender. A loss implies the opposite.

IV. Results

The results of our calculations with and without race and ethnicity included in the model are shown in Tables 1 and 2.¹⁴ Each table shows results from two models: Model 1 includes the income to poverty ratio; Model 2 does not. In each table we present 1,599 combinations of the explanatory variables—533 each for White, Hispanic, and Black individuals—for each gender and each model.

Table 1Change in Income From Race-Based to Race-Neutral Model Associated With Various Combinations of Right-Hand Side Variables, by Gender and Model Specification (Group 1 of 2)

View Fullscreen

Table 2Change in Income From Race-Based to Race-Neutral Model Associated With Various Combinations of Right-Hand Side Variables, by Gender and Model Specification (group 2 of 2)

View Fullscreen

Table 1 uses the following values for the explanatory variables: region (urban, rural, neither), family income ($45,000 and $95,000), mother’s and father’s educational attainment (11 years, 11 years), (12 years, 12 years), (14 years, 16 years), (16 years, 12 years), mother’s age at first birth (age 19, age 28), and two parents at home. For simplicity, all observations have one sibling, two parents at home, and combinations of religion designated as Baptist, Catholic, Jewish and none.

The results in Table 2 are based on the following values for three of the explanatory variables: family income ($75,000 and $125,000),¹⁵ mother’s and father’s educational attainment (12 years, 16 years), (14 years, 14 years), (16 years, 14 years), (16 years, 16 years), and mother’s age at first birth (age 22, age 25). All other variables are as in Table 1. The purpose of Table 2 is to provide additional combinations to assess how lifetime earnings of a child vary with a race-, ethnicity-, and gender-neutral approach.

The numbers in each cell of Tables 1 and 2 correspond to the number of combinations where the difference in lifetime earnings of a child between the two models falls into that difference in lifetime earnings category. For example, row 1 in Table 1 contains the results for White males using Model 1. Instances where the difference in lifetime earnings is less than 0.1% (i.e., the absolute value of the difference in lifetime earnings between the neutral model is less than one tenth of 1% of lifetime earnings under the non-neutral model) are shown in the “Neutral” column. Then gains and losses in lifetime earnings are shown in increments: 0.1% to 0.9%, 1.0% to 2.9%, 3.0% to 4.9%, 5.0 to 9.9%, and 10% or higher.

Among females in Table 1, lifetime earnings under the neutral model are not exclusively higher or lower than that of the non-neutral model, with the exception of White Hispanic females, who by and large experience a gain under the neutral approach. For example, among White non-Hispanic females using Model 1, the use of the neutral model lowers lifetime earnings for 56% (=298/533) of the combinations and increases lifetime earnings for 41% (=219/533) (Table 1). When Model 2 is used, the corresponding percentages are nearly the same (53% and 44%, respectively).

Among Black non-Hispanic females, the use of the neutral model lowers lifetime earnings for 75% (=402/533) of the combinations and among Black Hispanic females the use of the neutral model lowers lifetime earnings for all of the combinations in both Model 1 and Model 2. This finding for Black females is a result of the large positive coefficient on the Black variable in the race/ethnicity-based Spizman-Kane model for females (see Spizman and Kane, 2020). In summary, the use of a neutral model benefits White females for roughly one half of the combinations, reduces estimated lifetime earnings for Black non-Hispanic females for roughly three-quarters of the combinations, and reduces estimated lifetime earnings for Black Hispanic females for all of the combinations.

The results are quite different for males. Most notably, for White males, the use of the neutral model results in a reduction in lifetime earnings for all combinations. For Hispanic and Black males, the neutral models result in an increase in lifetime earnings for nearly all combinations. Therefore, for the men, the findings suggest that the use of neutral model appears to be in line with the objective of the policy.

V. Discussion

Presumably, the goal of policy initiatives to remove race, ethnicity, and gender from consideration in the calculation of lifetime earnings of a child is to increase the fairness of damages awards (i.e., prevent a reduction in damages awards on the basis of membership in a protected class). Our analyses show that removing race, ethnicity, and gender from consideration does not fulfill this policy goal for females. In many instances, members of protected classes would receive lower damages awards and in other instances members of non-protected classes would receive higher damages awards if neutral modeling were used. The reason behind this result has to do with the fact that children within protected classes (and non-protected classes, too) are diverse, exhibiting differences in family composition, income, parental educational attainment, urban or rural location, religious preference, and other factors.

In contrast to the findings for females, the neutral models do, for the right-hand side combinations considered in this analysis, uniformly reduce damages awards for White, non-Hispanic males and uniformly increase damages awards for Black males, both Hispanic and non-Hispanic. Therefore, the issue of unintended consequences that we found among females does not appear to be the case among men. We note that this finding differs from Spizman, Cahill, and Kane (2022) which focused on educational attainment probabilities only. Therefore, this discrepancy by gender is the result of lower earnings amounts for Black and Hispanic males relative to White males within educational attainment categories.

As to the general public policy issues that motivate this study, one clear implication of the analysis is that, among females, children belonging to protected classes will likely experience a decline in economic damages if neutral models are used instead of non-neutral models. For females, in particular, the legislation designed to protect members of the class to which they belong adversely impacts the individual, and necessarily prevents the individual from being made whole, for the sole reason that the individual is a member of a protected class. Viewed this way, requirements for neutral modeling appear to introduce another injustice.

We note that the purpose of our analysis is to examine whether children belonging to a protected class might be adversely impacted by neutral models. We, therefore, do not comment on perhaps a larger public policy issue associated with restricting economic damages calculations to neutral models only. To the extent that disparities by race, ethnicity, and gender are not due to society-wide discrimination, failure to take such differences into account can lead to over and undercompensating individual plaintiffs and can lead to over and under compensation by individual defendants. As such, by requiring the use of neutral models only—i.e., forcing economists to not consider non-neutral models—a new set of systemic injustices would be introduced into economic damages calculations. The empirical challenge then becomes whether the new systemic injustices outweigh past systemic injustices on a societal level. The challenge is much deeper than empirical as well and involves ethical and moral judgements about the extent to which individuals, through no fault of their own, are adversely impacted because of their membership in a favored or unfavored societal group.

Finally, we note that cross examination during the litigation process can expose injustices stemming from calculations of economic damages that are discriminatory. It is against this benchmark—in which discrimination can be exposed through the litigation process—that policymakers should consider any uniform policy that restricts how economic damages calculations are conducted, and the potential for adverse impacts to individual members of protected groups, as documented in this paper.