Work-Limiting Disability and Intergenerational Economic Mobility

Abstract

Objective:

To examine whether work-limiting disability may modify intergenerational economic mobility in the United States.

Methods:

Using the Panel Study of Income Dynamics, common metrics of intergenerational mobility are estimated by parent work-limiting disability. These include rank slope coefficients capturing persistence of socioeconomic status and absolute upward economic mobility capturing expected child outcomes.

Results:

Parent-child pairs with work-limiting disability experience five to twelve percentiles lower absolute economic mobility at the 25^th percentile of parent income. More severe and/or chronic conditions have larger disparities and higher parent income is associated with smaller disparities. Women may experience larger mobility differences, while non-Hispanic black children may face a higher likelihood of parents experiencing work limitations.

Conclusions:

Work-limiting disability appears to modify children’s economic opportunity. This contributes to the understanding of disparate access to opportunity in the United States while also identifying economic disadvantages associated with disability for subsequent generations.

1. Introduction

Although the United States (U.S.) has been characterized as a land of opportunity, estimates in intergenerational economic mobility suggest economic opportunity may be low (Mazumder, 2005; Solon, 1992, 2002). Within the U.S. though, significant heterogeneity in economic mobility particularly in terms of geography and race/ethnicity is present (Chetty et al., 2018; Hertz, 2005; Mazumder, 2014; Chetty et al., 2014). In spite of a robust association between health and socioeconomic status (Case et al., 2005; Conti et al., 2010; Hokayem & Ziliak, 2014), the intersection of health and intergenerational economic mobility has received less attention. This study contributes to this literature by examining parent health as a moderating characteristic of children’s economic opportunities. Specifically, this paper focuses on health conditions that limit a parent’s ability to participate in the labor market, or work-limiting disability, as the moderating characteristic.

Becker and Tomes (1979) modeled the persistence of permanent incomes across generations, and suggested a shock to permanent income in one generation could persist to a second, even if the second generation did not experience below-average luck or endowment. Solon (2004, 2014) advanced these theoretical underpinnings by justifying the common practice estimating intergenerational elasticity of income (IGE) with a log-log specification, which captures the relation of parent and child incomes. Importantly, these models rely on utility-maximizing parents who are subject to a budget constraint consisting of their own consumption and investment in the next generation (Becker & Tomes, 1979; Solon, 2004, 2014).

With this model in mind, a health condition that limits parents’ engagement in the labor market could be associated with less economic mobility in the next generation. Literature suggests persons with work-limiting disabilities may experience marked economic disadvantages persisting years after onset (Charles, 2003; Jolly, 2013; Meyer & Mok, 2019). For persons with chronic and severe work-limiting conditions, Meyer and Mok (2019) found a precipitous drop in employment, and estimated post-transfer income around 28% lower ten years after work-limitation onset relative to experiences more than five years preceding onset (Meyer & Mok, 2019). Poverty in particular is higher for persons with disabilities (Brucker et al., 2015; She & Livermore, 2007), and persists across generations (Rodgers, 1995). In addition to lower income and earnings, persons with disabilities typically have higher medical expenditures (Mitra et al., 2009), which could increase parent consumption, at the expense of investment in the subsequent generation. From the theoretical framework, work-limiting disability could therefore constitute a negative shock to parent permanent income exasperated by a more constrained budget.

Second-generation decisions such as educational attainment could feasibly be related to this constrained budget. For example, Manoli and Turner (2018) found higher tax refunds in low-income families had a positive impact on high school seniors’ college enrollment the following semester – suggesting relaxing budget constraints could increase educational attainment. More broadly, children of parents in poorer health are more likely to drop out of high school, and conditional on transitioning to college, less likely to complete their degree (Boardman et al., 2012). Children of parents with disabilities in particular are also at higher risk of attaining lower grades while in school (Miles et al., 2011), and research identifies a negative association between child’s educational attainment and the number of years a household head reports work-limiting disability (Haveman & Wolfe, 1994).

Outside of the budget constraint, caregiving responsibilities could play a role in reducing economic outcomes for children of work-limited parents. One study suggests some women who both work and participate in caregiving forgo approximately $5,000 annually from a combination of reduced hours and wages (Van Houtven et al., 2013). Since young children can also participate in caregiving (Hunt et al., 2005; Miles et al., 2011), such responsibilities could reasonably factor into their educational decision. Apart from these labor market outcomes, caregiving could also deteriorate a caregiver’s health (Kohl et al., 2019), buttressing an argument for intergenerational health transmission. A nascent literature demonstrates that there is indeed an intergenerational health linkage, whether it be from early in the lifecourse with intergenerational persistence of low birth weight (Currie & Moretti, 2007), to other conditions such as asthma, chronic migraine, diabetes, or hay fever (Thompson, 2014). More generally, self-reported general health status (Halliday et al., 2018) exhibits an intergenerational relationship.

Given documented disadvantages of disability, intergenerational spillovers of ill health, and a theoretical framework to support differentiated mobility, this study hypothesizes that children of parents experiencing work-limiting disability have less economic mobility relative to peers of parents without work limitations and similar socioeconomic status. Approximately one in five male household heads experience a work-limiting disability before the age of 30 (Meyer & Mok, 2019). This suggests parents with work-limiting disability are not a negligible portion of the population, and examining parent health as a modifying characteristic in intergenerational economic mobility could help understand heterogeneous access to economic opportunity in the U.S.

This analysis suggests children of parents with work-limiting disability experience less economic mobility. On average, these children achieve an earnings rank that is five to twelve percentiles lower than children of parents without work limitations. The observed gap in economic mobility tends to be larger as the severity and/or duration of a parent’s condition increases, but declines as parent income rises. Examining patterns by race-ethnicity and gender reveals women tend to disproportionately experience these gaps in economic mobility, while non-Hispanic black children more often have parents reporting health conditions that limit labor force participation.

2. Methods and Data

2.1. Estimating Intergenerational Economic Mobility

Intergenerational economic mobility is estimated from an Ordinary Least Squares (OLS) regression of parent economic status on child’s status, controlling for age as in equation (1).

$${Rank}_c={\beta }_0+{\beta }_1{Rank}_p+\sum {\beta }_i{X}_i+\epsilon$$ (1)

In this study, age controls (i.e. ∑β_iX_i) take a quadratic form in both generations, and all children are also observed at approximately the same age as adults (i.e. at least twice between age 30 and 33). Parent and child status is measured in percentile ranks of their respective distributions (Rank_p and Rank_c) following Chetty et al. (2014, 2018) due to desirable linearity and subpopulation comparison properties (Chetty et al., 2014; Hertz, 2005). One metric of interest is the slope coefficient, β₁ in equation (1), which captures the persistence of economic status between parents and children. The concept of absolute mobility from Chetty et al. (2014, 2018) is also of interest, capturing the expected percentile rank of a child conditional on his/her parent’s percentile rank as in equation (2).

$$E({Rank}_c|{Rank}_p=x)={\beta }_0+{\beta }_1(x)$$ (2)

Of particular interest in this analysis is a difference in absolute economic mobility between children from similar socioeconomic backgrounds but with differing parent work limitations. Therefore, a gap comparing economic mobility of children whose parents report no work limiting disability relative to children whose parents experience work limiting disability is the main parameter of interest in this analysis as in equation (3)¹.

$$Gap={(E({Rank}_c|{Rank}_p=x))}_{nolimit}-{(E({Rank}_c|{Rank}_p=x))}_{mostlimit}$$ (3)

This analysis first examines the mobility gap associated with parent work limitations over a range of condition severities and/or durations at set points in the income distribution. Particular attention is placed on results at the 25^th percentile of parent income, which Chetty et al. (2014) define as absolute upward mobility². In this study, as will be demonstrated more concretely in the results section, mobility gaps are concentrated in the lower portion of the parent income distribution, supporting the 25^th percentile as a reasonable summary measure for absolute upward mobility. Second, the mobility gap calculated by equation (3) is also estimated throughout the parent income distribution when holding parent work limitations constant. Statistical significance is assessed at the five-percent level, and reported estimates and standard errors incorporate the data’s complex survey (i.e. cluster and strata) design and longitudinal weights unless otherwise noted.

2.2. Data

The Panel Study of Income Dynamics (PSID) is used to complete this analysis due to its panel structure over both generations with detailed income, earnings, and health data (PSID, 2018). The main sample includes 3,687 parent-child pairs from the 1959 – 1984 birth cohorts who were resident with biological or adoptive parent(s) and part of the original Survey Research Center (SRC) or Survey of Economic Opportunity (SEO). Parents are identified as the observed Head and Partner of the family unit where the child resided at age 15 (the reference wave)³. On average, children are co-resident with their identified parents for approximately 10 waves. Birth cohorts are pooled for sample size assuming similar mobility (Lee & Solon, 2009).

Each parent’s family-unit income⁴ is averaged over the three waves up to, and including, the reference wave to partially address attenuation bias (Solon, 1992). For example, children from the 1959 birth cohort observe each parent’s family-unit income in the 1972–1974 waves while children from the 1984 birth cohort observe each parent’s income in the 1996–1999 waves. Since income reports reference the previous year, the child is consistently around 12 – 14 years old at the time parent income is observed. If two parents are present in the reference wave (approximately 75% of the sample), the average income of observed parents is used. Child individual labor market earnings are observed at least twice when the child is between ages 30 and 33, regardless of birth cohort⁵. Therefore, depending on birth cohort, observed individual earnings are recorded in waves 1990 – 2017. In determining income or earnings rank, parents are ranked among parents with children in a particular birth cohort and children are ranked among their birth cohort as in Chetty et al. (2014, 2018). Due to including the SEO, an oversample of low-income households, ranks are generated from comparable distributions of the March Supplement of the Current Population Survey using IPUMS data⁶ (Flood et al., 2018). Life-cycle bias (Haider & Solon, 2006) is mitigated by controlling for both generations’ ages in quadratic form. Sensitivity analyses relax and address these assumptions in Table 3. Specifically, using the child’s family income, accounting for longer observation windows, and varying the definition of parent health are discussed in the Appendix.

Table 3:

Sensitivity of Absolute Mobility Estimates – E(Rank_c|Rank_p = 25)

	Sample Size								Minimum Threshold
Subpopulation	Total	No Limit	Any Limit	≥ 20%	≥ 40%	≥ 60%	No Limit	Any Limit	≥ 20%	≥ 40%	≥ 60%
Main Sample	3,687	2,776	911	721	442	292	45.23	40.34*	40.26*	38.21*	35.67**
Child Income	4,693	3,471	1222	988	604	400	42.83	38.55+	37.99*	35.87**	34.32**
Males	1,825	1,397	428	336	211	129	51.71	51.12	50.77	49.87	48.16
Females	1,862	1,379	483	385	231	163	37.48	30.5**	30.94**	26.41**	25.47**
non-Hispanic White	2,057	1,608	449	333	196	120	47.13	40.48*	39.7*	39.56	38.6
non-Hispanic Black	1,347	956	391	329	209	148	37.31	38.77	39.09	38.25	35
Minority	1,629	1,167	462	388	246	172	41.89	38.68	38.88	36.22	32.79*
Cross Section Weights	2,486	1,831	655	483	279	159	46.21	41.02+	40.54+	39.22	36.47*
Custom Weights	3,301	2,485	816	649	395	265	45.73	41.18+	41.54+	40.17+	38.29*
Unweighted, No SEO	2,293	1,801	492	360	206	121	45.25	40.84+	40.29+	39.39+	35.49*
PSID Ranks	3,608	2,721	887	700	430	284	44.66	40.29*	40.18*	37.82*	34.91**
7yr Parent Income	3,576	2,691	885	702	431	286	44.06	39.48*	39.54+	38.13+	35.03**
6yrs Work Limit	2,815	1,936	879	475	248	157	46.06	41.65+	41.34	38.96+	37.76*
Average Work Limit	3,687	2,776	911	610	341	214	45.23	40.34*	40.45*	36.54**	33.15**
Self-Reported Health	1,904	679	1225	706	236	70	50.13	41.86+	41.01*	42.05	47.97
Two Parents	2,760	2,112	648	504	310	195	46.12	40.5+	39.87+	39.4+	39.12
Observe children age 15/16	3,544	2,676	868	685	419	277	45.31	40.08*	40.04*	37.97*	36.3*
Children only	3,580	2,733	847	664	405	269	45.85	39.95*	39.98*	38.91*	36.5*
Remove Work-Limited Children	2,650	2,039	611	470	277	175	48.14	44.33	43.55	41.3+	38.92*
Remove Children in Ill Health	2,570	1,983	587	460	262	156	48.94	45.05	44.01	45.2	41.34
Birth cohorts 1969–1984	2,120	1,558	562	407	231	125	45.82	40.47+	40.42+	39.95	37.37
HS or less	1,697	1,197	500	426	277	199	44.71	39.61+	39+	38.17	38.23
Some College	1,990	1,579	411	295	165	93	46.52	41.32	41.66	35.76+	26.26**
College Degree	1,027	846	181	128	69	36	49.74	44.4	46.61	38.41	35.1
Birth Cohort Fixed Effects	3,687	2,776	911	721	442	292	43.57	38.33*	38.39*	36.52*	34.13**

Source: Author’s calculations using PSID core data and IPUMS-CPS. Estimates adjusted for PSID complex survey design. Main sample estimates are included in row 1 for comparison.

^**p < 0.01

^*p < 0.05

⁺p < 0.1

2.3. Identifying Work-limiting Disability

The presence and degree of work-limiting disability for each parent is determined from two or three questions in the PSID. Respondents report whether he/she/partner has a “physical or nervous condition that limits the type of work or the amount of work you can do.” If the respondent indicates the presence of a work limitation, follow-up questions are asked to determine the severity of the condition (i.e. “does this condition keep you from doing some types of work?” and “for work you can do, how much does it limit the amount of work you can do…”), and respondents can indicate a condition limits work “not at all”, “just a little”, “somewhat”, “a lot”, or that they “can do nothing.”

Each wave the respondent is categorized as having either no, mild, moderate, or severe work limitations. Those with no work limitation report “no” to the first question, while those with any (mild, moderate, or severe) answer the first question in the affirmative. Mild conditions correspond to reports of conditions limiting work “not at all” or “just a little”, moderate conditions correspond to conditions which limit work “somewhat”, and severe reports indicate the condition limits work “a lot” or the individual “can do nothing.” Conditions are assigned zero, one, two, or three points respectively each wave when the child is 12–14 years old. This corresponds to the same time frame when parent income is observed⁷. Points are summed for each parent over the three waves and divided by nine (the maximum possible score) to produce an index ranging from zero to one that captures the severity and duration of the parent’s work limitation. Higher index values correspond to more chronic and/or severe work-limiting disability. In two-parent households the highest observed work limitation index is used.

Parent-child pairs are stratified into three mutually exclusive groups representing parents with no limitations, some limitations, and the most limitations. The main comparison is between the first group (no limitations) and last group (most limitations), ignoring the group with some limitations, so that there is a sharper distinction of parent experiences with work limiting disability. A threshold delineating membership in the most limited category then becomes key to the analysis. For example, if the threshold is set to one percent, operationally this is a comparison of economic mobility for children of parents with any work limitation (all non-zero index values are above one percent) with no work limitation. If the threshold were set at one hundred percent, this would compare only children of parents with consistently severe work limitations with children of parents with no limitations⁸.

Within this analysis the threshold varies, but focuses on two thresholds. A low threshold of 20% groups all work limitation reports together in the most-limited category except parents who have just a single mild report (index = 11%). A high threshold of 60% only includes parents with two severe reports (out of three waves) or all three waves with moderate or severe work limitations in the most limited category. These thresholds are hereafter referred to as the low or high thresholds respectively, and can be conceptualized as progressively more chronic and/or severe groupings of otherwise heterogeneous parent experiences of work-limiting disability.

3. Results

The sample’s descriptive statistics align with expectations from previous research with respect to parent characteristics (see Table 1). In particular, parents without work limitations on average have higher income and corresponding rank. Approximately 23% of children in the sample have at least one parent reporting a work-limiting disability, and among these children the average of their parents’ work limitation index is about 44% (e.g. two of three waves with moderate limitations or a mild and a severe report). In the second generation, children of parents with work limiting disability also tend to have lower individual earnings and family income. These children however also seem more likely to experience a work-limiting disability or ill health relative to children of parents who never report a work-limiting disability.

Table 1:

Descriptive Statistics for the main sample

	Full Sample	Non-Limited	Any Limitation	Index ≥ 20%	Index ≥ 40%	Index ≥ 60%
Parent Family Income	$95,952	$102,050	$75,683**	$69,962**	$63,848**	$58,105**
	(2,869)	(2,955)	(4,103)	(3,840)	(4,488)	(4,488)
Parent Family Income Rank	58.91	62.28	47.73**	44.91**	41.77**	38.60**
	(1.215)	(1.051)	(2.248)	(2.324)	(2.866)	(3.105)
Parent Work Limit Index	0.101	0	0.436**	0.533**	0.706**	0.822**
	(0.00856)	(0)	(0.0213)	(0.0189)	(0.0184)	(0.0181)
Parent Age	41.54	41.00	43.33**	43.86**	44.30**	45.49**
	(0.167)	(0.181)	(0.388)	(0.483)	(0.467)	(0.493)
Child’s Earnings (ages 30–33)	$51,428	$53,330	$45,102**	$43,256**	$40,962**	$38,437**
	(1,225)	(1,484)	(1,758)	(2,298)	(2,810)	(3,677)
Child’s Earnings Rank (ages 30–33)	54.08	55.81	48.33**	46.96**	45.01**	41.88**
	(0.882)	(0.997)	(1.346)	(1.569)	(2.077)	(2.600)
Child’s Family Income (ages 30–33)	$87,242	$90,781	$75,479**	$72,920**	$70,294**	$65,110**
	(2,419)	(2,848)	(2,562)	(2,883)	(3,879)	(4,751)
Child’s Family Income Rank (ages 30–33)	54.95	56.52	49.73**	48.25**	46.17**	42.61**
	(1.021)	(1.095)	(1.525)	(1.615)	(2.150)	(2.683)
Percent Children Ever with Work Limit	0.278	0.268	0.312+	0.324*	0.355**	0.364**
	(0.00892)	(0.00967)	(0.0201)	(0.0247)	(0.0282)	(0.0322)
Percent Children Ever in Ill Health	0.251	0.237	0.300*	0.309*	0.371**	0.448**
	(0.0126)	(0.0132)	(0.0243)	(0.0268)	(0.0369)	(0.0439)
Percent Children Married (ages 30–33)	0.580	0.589	0.550+	0.542*	0.531+	0.490*
	(0.0111)	(0.0124)	(0.0171)	(0.0192)	(0.0289)	(0.0392)
Observations	3687	2776	911	721	442	292

Source: Author’s calculations using PSID core data and IPUMS-CPS. “Any limit” includes all children with parent(s) ever reporting work limitations while the child is aged 12–14. Index values above a particular threshold (i.e. 20, 40, or 60%) comprise the subsample of children whose parents’ work limitation index meets or exceeds that threshold. All income and earnings reports are in 2016 U.S. Dollars. Standard errors in parenthesis and estimates adjusted for PSID complex survey design. Adjusted Wald test for group mean comparison indicates statistical significance:

^**p < 0.01

^*p < 0.05

⁺p < 0.1 relative to reference category (No Limit).

Table 2 gives an overview of estimated mobility metrics for the sample. On average, the rank slope in the sample is 0.342. However, children of parents without work-limiting disabilities have an estimated slope coefficient that is 0.314 relative to children of parents with any work-limiting disability where persistence is higher with a slope coefficient of 0.383. Although the rank slope (i.e. persistence of parent income rank) is estimated to be 22% higher for children whose parents report work limitations, the difference is not statistically different. In terms of absolute mobility, this analysis estimates that children of parents at the 25^th percentile in the income distribution without work-limiting disability on average can expect to reach near the 45^th percentile of their individual earnings distribution as adults. Children whose parents experience work-limiting disability however, are consistently estimated to have lower absolute upward economic mobility. The magnitude of the gap in outcomes however depends on severity and duration of the work-limiting condition and parent socioeconomic status.

Table 2:

Regression Coefficients and Absolute Upward Mobility

				Minimum Threshold
	Full Sample	No Limit	Any Limit	≥ 20%	≥ 40%	≥ 60%
Slope	0.342	0.314	0.383	0.379	0.363	0.381
	(0.0171)	(0.0225)	(0.0384)	(0.0466)	(0.0623)	(0.0828)
Intercept	34.70	37.38	30.76	30.78	29.14	26.15
	(1.551)	(2.055)	(2.365)	(2.310)	(3.620)	(4.104)
E(Rankc|Rankp = 25)	43.27	45.23	40.34*	40.26*	38.21*	35.67**
E(Rankc|Rankp = 50)	51.83	53.07	49.93+	49.74	47.28*	45.19*
E(Rankc|Rankp = 75)	60.39	60.91	59.51	59.22	56.34	54.71
N	3,687	2,776	911	721	442	292

Source: Author’s calculations using PSID core data and IPUMS-CPS. Slope and intercept estimates are from equation (2), while E(Rank_c| Rank_p = x) are estimates from equation (3) taken from the 25^th, 50^th, and 75^th percentiles. “Any limit” includes all children with parent(s) ever reporting work limitations while the child is aged 12–14. Index values above a particular threshold (i.e. 20, 40, or 60%) comprise the subsample of children whose parents’ work limitation index meets or exceeds that threshold. Standard errors in parenthesis and estimates adjusted for PSID complex survey design.

^**p < 0.01

^*p < 0.

⁺p < 0.1 relative to reference category (No Limit)

Allowing the threshold that separates any condition from “the most” limiting conditions to range from near zero to one hundred percent reveals a gap in absolute upward mobility that is consistently positive and increasing over most the interval (Figure 1). Figure 1 shows that this gap is estimated to be between five and twelve percentiles⁹. Recall that as the threshold increases, the resulting grouping of “most” limited parents becomes less heterogeneous and represents conditions that are more chronic and/or severe. Lower thresholds allow larger sample sizes within “the most” limited category, but more variation in parent experiences. For example, the low threshold (i.e. 20%) maintains a sample size with 721 parent-child pairs classified as “most” limited; however, the only pattern of work-limiting disability not contained in the most limited category are parents with a single mild report (N = 190). The gap in upward mobility at this low threshold is around five percentiles. Meanwhile, a high threshold (i.e. 60%) contains 292 parent-child pairs classified as “most” limited but provides a sharper distinction with less heterogeneity of parent experiences. The observed gap in upward mobility at this high threshold is near ten percentiles.

Figure 1:

Gap in Absolute Upward Mobility over a Range of Thresholds Delineating the Most Limited

Fig 1 shows an increasing gap in absolute upward mobility at the 25^th percentile of parent rank as the parent work limitation index threshold increases. Positive gaps throughout are indicative of lower absolute mobility for parent-child pairs where the parent reports a work-limiting disability (i.e. E(Rank_c | Rank_p = 25) _{most limit} < E(Rank_c | Rank_p = 25)_{no limit}).

Source: Author’s calculations using PSID core data and IPUMS-CPS.

Figure 2 examines the gap in absolute upward mobility over the parent income distribution holding the threshold from Figure 1 constant at either the low (20%) or high (60%) threshold. Regardless of the threshold examined, the absolute economic mobility gap declines as parent income rises. Figure 2 highlights that upward mobility gaps are concentrated among parent-child pairs of lower socioeconomic status (i.e. parent family income). In fact, the diminishing gap in upward mobility becomes statistically indistinguishable from zero around the bottom third and half of parent income for the low and high threshold respectively. This could reflect income offering some protection against mobility disadvantages associated with work limitations. However, it also reflects lower relative mobility (i.e. higher rank slope coefficients, or more persistence of economic status) for parent-child pairs with the most limiting conditions, although these differences are not statistically significant.

Figure 2:

Gap in Absolute Upward Mobility over the Parent Income Distribution

Fig. 2 shows the gap in child’s absolute upward mobility by parent work-limiting disability over the parent income distribution. Solid (dashed) lines correspond to estimated gaps that are statistically significant at the 5% (10%) level, while the dotted line outlines the estimated gap for the statistically insignificant portion of the parent income distribution. Negative slopes indicate a decreasing gap over the parent income distribution. Positive gaps throughout most of the parent income distribution are indicative of lower absolute mobility for parent-child pairs where the parent reports a work-limiting disability.

Source: Author’s calculations using PSID core data and IPUMS-CPS.

Partitioning the sample by gender suggests women experience decreased economic opportunity associated with parents’ work-limiting disability more so than men (Figure 3). According to these estimates, sons do not experience statistically lower economic mobility relative to their parents’ work limitations. Meanwhile, daughters experience lower economic mobility when parents face a work-limiting disability that increases with the duration/severity of a parent’s condition as observed in the full sample. Statistically significant gaps are present for daughters at a low threshold in the bottom 42% of parent income, and at a high threshold in the bottom 79% of parent income.

Figure 3:

Gap in Absolute Upward Mobility by Child Gender

Fig. 3 stratifies the analysis by child gender. Sons do not experience statistically lower absolute mobility. Daughters experience lower mobility throughout most of the distribution of parent work-limitation index, and the gap is almost always estimated to exceed the gap for sons while also being statistically significant at the 5% level. Results are robust to various weights, including using an unweighted sample excluding the SEO. Solid (dashed) lines correspond to estimated gaps that are statistically significant at the 5% (10%) level, while the dotted line outlines the estimated gap for the statistically insignificant estimates.

Source: Author’s calculations using PSID core data and IPUMS-CPS.

Similarly partitioning the sample by the child’s self-reported race/ethnicity reveals non-Hispanic white children face a larger upward economic mobility gap associated with parent disability relative to non-Hispanic black peers. This does not suggest any advantage in economic mobility for non-Hispanic black children. Rather, conditional on a parent experiencing any work-limiting disability, absolute economic mobility for non-Hispanic white children more closely resembles the experience of non-Hispanic black peers, resulting in a larger mobility gap for non-Hispanic white children based on parent work-limiting disability (Figure 4 and Appendix Figure A1). These results however rely on small samples. As demonstrated in Figure 4 and Table 3, examining all minority children (approximately 83% are non-Hispanic black) reveals an upward mobility gap at high thresholds that is more comparable to that observed in non-Hispanic white children. These sensitivities are further addressed in the discussion below.

Figure 4:

Gap in Absolute Upward Mobility by Child Race/Ethnicity

Fig. 4 stratifies the analysis by child race/ethnicity. Results suggest non-Hispanic white children experience a drop in economic mobility associated with parent work-limitation status at low thresholds and parent income rank, while non-Hispanic black children do not. All minorities (i.e. any race/ethnicity that does not identify as non-Hispanic white) experience gaps in absolute mobility at higher thresholds and lower parent income. Note however that sample sizes are small and results are sensitive to the weight (i.e. longitudinal versus cross-sectional). Results should be interpreted with caution. Solid (dashed) lines correspond to estimated gaps that are statistically significant at the 5% (10%) level, while the dotted line outlines the estimated gap for the statistically insignificant estimates.Source: Author’s calculations using PSID core data and IPUMS-CPS.

4. Discussion

These results highlight children whose parents report work limitations experience lower economic mobility. More chronic and/or severe conditions are associated with larger gaps (Figure 1), suggesting diminishing economic mobility for children of parents with more chronic and/or severe health conditions that limit their ability to actively engage in the labor market. However, gaps in absolute upward mobility decline as parent family income rises, indicating high parent income could potentially be protective of economic mobility disadvantages associated with work-limiting disability. Table 3 demonstrates that these results are stable to a variety of sensitivity analyses including observing parent income or work limitations over longer periods, observing parent health, birth cohort fixed effects, and removing potentially influential subpopulations. For example, the main results incorporate all family structures, and do not adjust for children who also experience work limitations or ill health. Removing these subpopulations (i.e. single-parent homes or children who also experience work-limitations) does not affect the main conclusions (Table 3). Measuring child outcomes earlier or later in the lifecourse is also addressed in Appendix Table A3, and again results are robust.

Substituting child family-unit income for individual earnings yields consistent results. As demonstrated in Table 3 and Appendix Figure A2 though, predicted income ranks are typically one to two percentiles lower across groups in this analysis. While child’s family-income is more analogous to the parent income measure, there are a couple of key drawbacks to its use in this context. Briefly, PSID does not observe health conditions of parents-in-law, adding unobserved heterogeneity into estimation. Additionally, family-unit income incorporates marital patterns, and children of parents with more chronic and/or severe work-limiting conditions are less likely to be married in their early 30s (see Table 1). Further discussion of this analysis is found in Appendix section 1.

The main analysis suggests women face larger mobility gaps associated with parent disability. This could be indicative gender differences in individual earnings or reflective of women being less likely to experience healthy aging (McLaughlin et al., 2012). For example, in this sample, approximately 31% of daughters ever experience work limitations while only 25% of sons do. Another alternative could be related to caregiving responsibilities. If women are more likely to participate in caregiving (Laditka & Laditka, 2000; Van Houtven et al., 2013), the concentration of gaps in upward economic mobility among daughters could also plausibly be related to such responsibilities that surface early when parents experience work limitations.

Results from this analysis also suggest non-Hispanic whites experience greater mobility gaps related to parent work-limiting disability, which could be related to non-Hispanic blacks experiencing relatively low upward mobility regardless of parent work-limitation status (Chetty et al., 2018). However, as noted above, results rely on small samples and exhibit sensitivity. For example, examining all minority children reveals a particularly large gap in upward mobility at high thresholds (see Figure 4). This is in stark contrast to results for non-Hispanic black children, even though they compose 83% of the minority sample. Chetty et al. (2018) suggests Hispanic and Asian American children experience higher upward absolute mobility than non-Hispanic black children, and therefore results should be interpreted with caution pending further research. It is interesting to note that non-Hispanic black children are likely to have parents experience work limitations, and when they do have limitations, the conditions are typically more chronic and/or severe, although this does not adjust for other observable characteristics, including socioeconomic status (see Appendix Table A1). While results of race/ethnicity stratified mobility are sensitive, collectively these observations could be related to the documented black-white economic mobility gap in the U.S. (Chetty et al., 2018; Hertz, 2005; Mazumder, 2014), and further analysis should seek to examine this issue more fully.

It should also be observed that the estimated gap in upward mobility does not consistently increase as the threshold delineating “the most” limited category increases. If more chronic and/or severe conditions limit economic opportunity more, the gap should increase throughout the threshold in Figure 1. Rather, at very high threshold levels, the gap declines (see Figure 1). There are a variety of explanations that could produce this observation. First, there could be a ceiling gap such that child individual earnings rank have some minimum value regardless of parent work-limitation experience leading to a maximum gap. Second, since Figure 1 does not rule out a continually increasing gap, smaller sample sizes in the “most limited” category at larger thresholds may obscure an increasing gap. However, another explanation could be related to public policy and transfer income. Social Security Disability Insurance (SSDI) and Supplemental Security Income (SSI) in particular are targeted towards persons with chronic and severe conditions that limit their ability to actively engage in the labor market. Previous research from Canada suggests children may have improved cognitive and non-cognitive outcomes due to disability benefit receipt (Chen et al., 2015). However, the relatively low joint incidence of chronic and severe work-limiting disability with SSDI/SSI among parents in this sample is prohibitively small, and this particular topic is left to future research.

Other threats to identification in this analysis include selection, attrition and underreporting of transfer income. Attrition is more common in the PSID among parents in the lowest quartile whose children also end up in the lowest quartile (Schoeni & Wiemers, 2015), and sample selection follows a similar pattern here, with relatively high observed ranks in both generations. Additional examination of the issue in the Appendix suggests children in this sample with higher socioeconomic status are more likely to have observed parent characteristics, and children of parents with work limitations are more likely to attrite. The pattern of attrition seems progressively stronger for more chronic and/or severe limitations. However, there is not sufficient evidence to suggest that these patterns affect this study’s main conclusions (see the Appendix for a full discussion). Underreporting of transfer income is also present in PSID (Meyer et al., 2015), which could underestimate income ranks particularly for parents with health conditions if they are also more likely to receive transfers such as SSDI/SSI. Similarly, if work-limiting disability precedes observed lower income, then the latent human capital of the parent may be underestimated by income. In either of these two latter cases, the analysis would underestimate of the gap in absolute upward economic mobility. In fact, the gap in absolute upward mobility is largest among parents with postsecondary education (Table 3).

This analysis suggests children of work-limited parents experience less intergenerational economic mobility. The persistence of socioeconomic status may be a little stickier for children of parents with work-limiting disabilities (i.e. the slope coefficient is higher); however, it is not statistically significant in this sample. The main observed difference in intergenerational economic mobility for children of parents with work-limiting disabilities is in terms of upward mobility from the 25^th percentile of parent income rank. The estimated five to twelve percentile gap in upward mobility means children of parents with work-limiting disabilities experience less economic opportunity and that disadvantages associated with disability could spill over to a subsequent generation. The expected rank of a child whose parents exhibit significant work limitations is around the 36^th percentile, meaning these children are more similar to children of parents at the bottom of parent income who do not report limitations (see Table 2 estimates). This represents a large gap in economic mobility, and suggests that parent health may be an important modifier to observed heterogeneity in intergenerational economic mobility in the United States.

Appendix to: Work-Limiting Disability and Intergenerational Mobility

1. Technical Details on Key Examined Subpopulations in Sensitivity Analyses

A variety of sensitivity analyses are in order to support the main conclusion of observed lower economic mobility for children whose parents experience work limitations. This section outlines the key analyses reported below in Table A3.

Child Income

Child family-unit income is substituted for child individual earnings in the first sensitivity analysis. The purpose is two-fold. First, it allows for a significantly larger sample (N = 4,693) as children who are not engaged in the labor market at ages 30–33 remain in this sample. This provides a glimpse of what could be happening with a larger sample and is more analogous to the measurement of parent economic status.

However, there are a few notable issues with family-unit income. First, PSID does not observe the work-limitation status of a focal child’s parents-in-law. Therefore, the focal child’s partner’s parents may experience unobserved work-limitations. It is expected that observing just two (of a potential four) parents’ work-limitation characteristics for married children yields the minimum value of (unobserved) maximum work limitations for the full four-parent set. At each categorization of parent work-limitations then, unobserved work limitations may be higher than observed limitations, putting downward pressure on upward mobility estimates. To the degree which this is equally present across all classifications of work-limited parents, little bias should result in estimated mobility gaps. This hypothesis is supported by observing lower upward mobility estimates across all work-limitation groups in Appendix Table A3. Additionally, family-unit income necessarily incorporates marital patterns, and children of more chronic and/or severe work-limited parents in this sample are less likely to be married (see Table 1) at the time of observation. So particularly for more chronic and/or severe work limitation groups, their upward mobility estimates then may not reflect as much downward bias (also evidenced in Table A3). Figure A2 highlights that in spite of these challenges, the estimated gap using child family income instead of child earnings looks very similar.

Weights

The main analysis uses PSID provided longitudinal weights due to inclusion of the SEO. However, the SEO is known to have sampling irregularities (Brown 1996). In unweighted analyses that exclude the SEO, robust errors are used in place of the PSID complex survey design. As demonstrated in Table A3 below, the magnitudes of main results are remarkably robust to this exercise; however, sample size is significantly reduced. In particular, since work limitations are more prevalent at lower incomes (and the SEO by extension), excluding the SEO yields particularly small samples for chronic and/or severe work limitations. At the same time, it is noted that PSID longitudinal weights are not necessarily comparable across all waves (see Noncomparability Notes of longitudinal weight variables in PSID technical documentation) (PSID 2018), yet children’s outcomes are observed from 1990–2017. This non-comparability across waves prompts a check using cross-section weights instead of longitudinal weights. Results are robust even though these weights are available only from 1997–2017, meaning birth cohorts 1959–1965 cannot be observed.

Another source of concern related to weighting is differential attrition. As noted in the main text, lower socioeconomic status individuals are more likely to attrite (Fitzgerald 2011; Fitzgerald, Gottschalk, & Moffitt 1998; Schoeni & Wiemers 2015). Figure A3 highlights significant attrition over time in the PSID¹. Conditional on a child being observed as a head or wife at age 24–27 in the PSID, just six years later (when children are age 30–33 – the age of main sample estimates), approximately 20% of children of parents without work limitations have already left the sample. An even higher portion of children of parents with work limitations (approximately 25%) have left (see Figure A3). PSID longitudinal weights adjust for attrition (Berglund & Heeringa 2015), and intergenerational persistence estimates generally do not exhibit attrition bias (Fitzgerald et al. 1998). Nevertheless, the application to parent work limitations as a modifying characteristic of intergenerational economic mobility warrants consideration.

Following Fitzgerald et al. (1998), an examination of potential attrition bias is considered. Birth cohorts 1957–1976 are eligible to be observed as adult children between the ages of 24 and 41 and form the base sample. Intergenerational economic mobility from equation (1) using family-unit income² is estimated for those who do not attrite at ages 26–29, 28–31, …, 39–41. As highlighted in Figure A4, there is some indication of downward pressure on upward economic mobility estimates particularly for children of parents with more chronic and/or severe work limitations. Slightly lower upward mobility estimates among those who remain in sample at older ages though are not statistically distinguishable from the initial point estimate.

Attrition however could also occur before age 24. A customized weight that adjusts the child’s longitudinal weight for the inverse probability of observing parent characteristics (controlling for child gender, race/ethnicity, and geographic region) is also incorporated³. Particularly for the non-limited sample, results are very close. Even for low threshold levels, there is little difference in results as highlighted in Table A3 below. At higher thresholds (work limitation index ≥ 40% or ≥ 60%), results suggest absolute upward mobility may be two to two and a half percentiles higher (with correspondingly smaller mobility gaps), which coincides well with estimates in Figure A4. This analysis therefore concludes that while differential attrition may be present, and may slightly overestimate mobility gaps, it does not pose a significant threat and results remain robust.

Longer Observation Windows

Parent income and work limitations are observed only for three waves, which may attenuate results (Solon 1992). However, it is noted that similar biases present across all groups should not meaningfully affect estimated gaps in mobility. Additionally, in terms of work-limitations, restricting observation to smaller time frames raises the probability of not observing a parent’s work-limitation that actually exists prior to observation. Increasing the number of waves required for observation decreases the available sample size, yet in both cases, results are similar across these modifications.

Varying the Definition of Parent Health or Disability

The main analysis assigns work-limitation status based on the highest observed parent index in two-parent homes. An alternative method would assign the average of two parents. It should be noted however, that an average of parent work limitations at 60% is more severe than a parent maximum index of 60%, as both parents must report limitations (or be a single-parent home) in the former construction. Consistent with this hypothesis and the overall results of lower economic mobility for more chronic and/or severe work limitations, the average parent work limitation construction reveals a slightly larger mobility gap for children of work limited parents.

This analysis also examines parent self-reported health status⁴ as an alternative to work-limitations due to concerns on work-limitation validity (Black, Johnston, & Suziedelyte 2017; Hale 2001). In this case the “no limit” reference category corresponds to parents who are on average in very good to excellent self-reported health. Sample size is significantly reduced with this partition as self-reported health was not asked in PSID until 1984. With the exception of the high threshold category (with only 70 observations), point estimates consistently indicate decreased economic mobility for children whose parents are in worse health.

2. Additional Tables and Figures

Table A1:

Descriptive Statistics by Gender and Race/Ethnicity

					| Any Work Limitation
	Males	Females	nH White	nH Black	nH White	nH Black
Parent Family Income Rank	59.74	57.96	63.68	33.39**	53.83	26.61**
	(1.308)	(1.433)	(0.965)	(2.043)	(2.176)	(2.551)
Parent Work Limit Index	0.0975	0.105	0.0904	0.162*	0.410	0.536*
	(0.0104)	(0.00863)	(0.00737)	(0.0302)	(0.0169)	(0.0575)
Percent Parents with Work Limits	0.223	0.241	0.221	0.302*	1	1
	(0.0146)	(0.0149)	(0.0149)	(0.0311)	(0)	(0)
Parent Age	41.52	41.56	41.74	40.56	42.85	45.00
	(0.162)	(0.268)	(0.197)	(0.631)	(0.506)	(1.079)
Child’s Earnings Rank (ages 30–33)	61.31	45.73**	56.52	41.75**	51.50	39.44**
	(1.070)	(1.054)	(0.885)	(1.561)	(1.572)	(2.282)
Percent Children Ever with Work Limit	0.246	0.315**	0.278	0.290	0.313	0.268
	(0.0113)	(0.0152)	(0.0116)	(0.0209)	(0.0273)	(0.0281)
Percent Children Ever in Ill Health	0.243	0.261	0.230	0.412**	0.268	0.457**
	(0.0158)	(0.0148)	(0.0140)	(0.0274)	(0.0287)	(0.0410)
Percent Children Married (ages 30–33)	0.594	0.563+	0.633	0.313**	0.613	0.332**
	(0.0130)	(0.0155)	(0.0118)	(0.0211)	(0.0192)	(0.0323)
Observations	1,825	1,862	2,057	1,347	449	391

Source: Author’s calculations using PSID core data and IPUMS-CPS. Males and non-Hispanic (nH) white subpopulations serve as reference categories. Standard errors in parenthesis and estimates adjusted for PSID complex survey design. Adjusted Wald test for group mean comparison indicates statistical significance:

^**p < 0.01

^*p < 0.05

⁺p < 0.1 relative to reference category (Males and nH white).

Table A2:

Results using a Categorical Partition of Work Limitations

		Limited
	No Limit	Temporary	Chronic, Not Severe	Chronic, Severe
Rank Slope	0.314	0.410	0.388	0.348
(Rankc|Rankp=25)	45.23	39.90+	41.96	38.04*
(Rankc|Rankp=50)	53.07	50.15	51.66	46.73+
(Rankc|Rankp=75)	60.91	60.40	61.36	55.42
N	2,776	334	356	221

Source: Author’s calculations using PSID core data and IPUMS-CPS. Standard errors in parenthesis and estimates adjusted for PSID complex survey design. Work limitations are observed over three waves. Temporary conditions correspond to a single work-limitation report of any severity. Chronic conditions consist of reported limitations in at least two waves. If reported conditions are severe in two waves, the parent is categorized as chronic and severe. If a severe condition is reported at most once, the chronic condition is classified as not severe. The partition does not allow the flexibility of the index, but suggests similar patterns.

^**p < 0.01

^*p < 0.05

⁺p < 0.1 relative to reference category (non-limited).

Table A3:

Results from Different Points in the Child’s Lifecycle

				Minimum Threshold
	Age	No Limit	Any Limit	≥ 20%	≥ 40%	≥ 60%
Main Sample (N = 3,687)	30 – 33	45.23	40.34*	40.26*	38.21*	35.67**
Earnings	28 – 31	45.68	43.48	41.93	41.35	41.42
Earnings	30 – 33	47.55	42.28	41.58	42.25	42.07
Earnings	32 – 35	48.09	41.47+	40.86+	40.97*	38.73*
Earnings	34 – 37	47.50	41.97+	41.5+	42.41	40.19+
N		1450	460	353	204	136
Child Income (N = 4,693)	30 – 33	42.83	38.55+	37.99*	35.87**	34.32**
Income	28 – 31	42.52	39.47	37.87+	34.05**	32.79**
Income	30 – 33	43.10	38.36+	37.49*	35.33**	34.83**
Income	32 – 35	42.73	37.8*	36.56**	34.19**	31.48**
Income	34 – 37	41.90	37.34*	35.95**	34.6**	30.32**
N		2201	764	614	366	243

Source: Author’s calculations using PSID core data and IPUMS-CPS. Standard errors in parenthesis and estimates adjusted for PSID complex survey design. Estimates from ages 28 – 37 include only children who are observed at all age ranges (i.e. ages 28 – 37).

^**p < 0.01

^*p < 0.05

⁺p < 0.1 relative to reference category (non-limited).

Figure A1:

Observed Upward Mobility by Race/Ethnicity over a Range of Thresholds Delineating the Most Limited

Fig. A1 illustrates upward mobility by race/ethnicity over a range of thresholds delineating the most limited. “nH” abbreviates non-Hispanic. Other race/ethnicities excluded from analysis due to small sample sizes. Non-limited subpopulation estimates are represented by dashed lines. Solid lines represent upward mobility point estimates for limited parent-child pairs as the threshold increases from near zero percent to one hundred percent. Dotted lines represent point estimates when the subpopulation of most limited parent-child pairs within a particular race/ethnicity category drops below 75. Point estimates when sample size decreases below 50 observations are omitted. Gap estimates from Figure 4 in the main text represent the difference between the non-limited (dashed line) and limited (solid line) for each race/ethnicity. Note the gap for non-Hispanic black is never statistically significant, and the gap for non-Hispanic white is only significant at low thresholds.

Source: Author’s calculations using PSID core data and IPUMS-CPS.

Figure A2:

Gap in Expected Mobility over a Range of Thresholds Delineating the Most Limited – Family Unit Income

Fig. A2 contrasts with Fig. 1 in that the child’s economic status is measured by family-unit income as opposed to individual earnings. Sample size is 27% larger.

Source: Author’s calculations using PSID core data and IPUMS-CPS.

Figure A3:

Attrition by Work Limitations

Fig. A3 highlights significant attrition over time in PSID. Attrition among adult children from birth cohorts 1957–1976 are pictured, and attrition is most pronounced for children whose parents experience more chronic and/or severe work limitations. The dashed red line indicates the age of observation in the main sample estimates.

Source: Author’s calculations using PSID core data and IPUMS-CPS.

Figure A4:

Attrition Association to Mobility Estimates

Fig. A4 highlights the minimal association of differential attrition with absolute upward mobility estimates. Dotted black lines are the baseline estimates of intergenerational economic mobility (using child family-unit income) for each subpopulation among observed children in the 1957–1976 birth cohorts at age 24–27. The red solid line depicts estimates for non-attriting children of the initial 1957–1976 cohort observed at age 24–27 who remain at each age of observation (corresponding to one wave later in the PSID). 95% confidence intervals of the non-attriting sample shaded in gray.

Source: Author’s calculations using PSID core data and IPUMS-CPS.