Do Stronger Employment Discrimination Protections Decrease Reliance on Social Security Disability Insurance? Evidence from the U.S. Social Security Reforms

Abstract

The United States Social Security Amendments of 1983 increased the full retirement age and penalties for retiring before that age. This increased Social Security Disability Insurance (SSDI) applications by making SSDI relatively more generous. We explore if state disability and age discrimination laws moderated these spillovers, using variation whereby many state laws are broader or stronger than federal law. We estimate the effects of these laws on SSDI applications and receipt using a difference-in-differences approach. We find that a broader definition of disability, where only a medically diagnosed condition is required to be covered under state law, along with being able to sue for more damages under state disability discrimination law, are both associated with a significant reduction in induced SSDI applications and receipts. We also find some evidence that some features of state disability discrimination laws are also associated with increased employment, especially for women. While we find some positive association between age discrimination laws and employment effects, we do not find any moderating effect of age discrimination laws on SSDI.

The Social Security Amendments of 1983 (SSA1983) made many significant cuts to the Social Security program in the United States. Notably, SSA1983 increased the full retirement age (FRA), the age at which individuals could retire with full Social Security benefits. The FRA increased gradually from 65 to 67 for cohorts born from 1938 or later, based on a graduated scale that increased for cohorts born later. SSA1983 also increased the penalty of claiming Old Age Survivor Insurance (OASI) benefits before the FRA for cohorts born 1938 or later (see Table 1).

Table 1.

The effect of the social security amendments of 1983 (SSA1983) on OASI benefits for primary earners, by age at retirement and year of birth.

Year of birth	Full benefits retirement age (FRA)	Benefit as a percentage of primary insurance amount (PIA) received if retiring at… (%)
		Age 67	FRA	Age 65	Age 62
1934–1937* and earlier	65	106–113	100	100.0	80.0
1938*	65 + 2 months	111.9	100	98.9	79.2
1939*	65 + 4 months	111.7	100	97.8	78.3
1940*	65 + 6 months	110.5	100	96.7	77.5
1941*	65 + 8 months	110.0	100	95.6	76.7
1942*	65 + 10 months	108.8	100	94.4	75.8
1943–1948*	66	108.0	100	93.3	75.0
1949–1951**	66	108.0	100	93.3	75.0
1952–1954	66	108.0	100	93.3	75.0
1955	66 + 2 months	106.7	100	92.2	74.2
1956	66 + 4 months	105.3	100	91.1	73.3
1957	66 + 6 months	104.0	100	90.0	72.5
1958	66 + 8 months	102.7	100	88.9	71.7
1959	66 + 10 months	101.3	100	87.8	70.8
1960 and later	67	100.0	100	86.7	70.0

NOTE:

^*indicates cohorts included in our aggregated SSDI count data sample and our HRS data sample;

^**indicates cohorts that are only included in our aggregated SSDI data sample. For the HRS sample, we include those born between 1931–1948.

The aggregated SSDI dataset uses data from 1992–2013, the HRS dataset uses data from 1992–2014, and the HRS-Form-831 merged dataset uses years 1992–2012.

SOURCE: Social Security’s Retirement Planner, at https://www.ssa.gov/oact/ProgData/ar_drc.html (accessed August 1, 2018).

SSA1983 was intended to increase employment and delay OASI benefits claiming to improve the program’s financial solvency in the long run. Under SSA1983, retiring at the previous FRA of 65 or earlier meant a cut in benefits, forcing individuals to choose between reduced benefits or attempting to work longer to delay retirement. Labor economic theory suggests that the SSA1983 did increase labor supply,¹ and Neumark and Song (2013), Behaghel and Blau (2012), and Song and Manchester (2007) document this empirically.

However, part of the reduction in spending for OASI benefits in the short run through these reforms is offset by the spillover effects onto the Social Security Disability Insurance (SSDI) program. SSDI and OASI benefits are substitutes for many older individuals,² making SSDI relatively more generous after the cut imposed by the SSA1983 at all ages before reaching the FRA for the affected cohorts. Li and Maestas (2008) document these spillovers, finding that an average four months’ increase in the FRA for those born between 1938 and 1940 increases the SSDI application rate by 0.04–0.30 percentage points, a moderate increase relative to the application rate of 4.3 percent for the affected cohort. They find a much stronger effect, ranging from a 0.22 to a 0.89 percentage point increase, for individuals at or above age 62 and those with work-limiting health problems. Duggan, Singleton, and Song (2007) find that 0.6 percent of men and 0.9 percent of women between the ages of 45 and 64 enrolled in the SSDI program in 2005 due to the relative induced generosity of the SSDI program.

This paper explores whether stronger disability and age discrimination laws moderated the spillover effects of SSA1983 on SSDI application, SSDI receipt, and employment. The intention behind disability and age discrimination laws is to improve the employment of individuals with disabilities and older workers by reducing employment discrimination. Disability discrimination laws have the added requirement that employers provide “reasonable accommodations” to workers with disabilities. These laws could increase hiring and reduce terminations, thus reducing the incentive to apply for SSDI, softening the spillover effect of SSA1983 on SSDI.

To explore how disability and age discrimination laws affect spillovers to SSDI, we focus on the differences in state discrimination laws. These laws are often stronger or broader than the federal Americans with Disabilities Act of 1990 (ADA) (Long 2004; Neumark, Song, and Button 2017) and the federal Age Discrimination in Employment Act of 1967 (ADEA). Many state laws are stronger or broader than the federal ADA or ADEA in three primary ways:

1. by lowering the burden of proof, relative to the ADA, for plaintiffs to establish under the law that they have a disability (“medical definition of disability”);

2. by covering employers with fewer employees by discrimination laws, relative to the higher employee requirements under the ADA (15 employees) or the ADEA (20) for discrimination laws to apply (“lower firm size (disability or age)”); and

3. by allowing plaintiffs to sue for more damages than allowed by the ADA or the ADEA (“larger damages (disability),” “larger damages (age)”).

Using a difference-in-differences (DD) regression analysis, we estimate the moderating effects of these stronger and broader discrimination laws on SSA1983-induced spillovers to SSDI application, SSDI receipt, and employment. We compare cohorts affected (born 1938–1951) and unaffected (1931–1937) by SSA1983 in states with and without discrimination law features, controlling for age. If, for example, SSDI application rates are higher for those affected by the SSA1983 (as we and others document), but this effect is weaker in states with stronger or broader discrimination laws, this suggests that these laws decrease spillovers to SSDI and likely reduce barriers to employment.

To quantify these effects in our DD, we use three sources of data: (1) aggregated counts of the universe of SSDI applications and receipts by year, state, age cohort, and sex in the United States, (2) the Health and Retirement Study (HRS) to quantify impacts on employment, and (3) the HRS matched with Form-831 Social Security Administration (SSA) disability records to quantify impacts on SSDI applications and receipts. While (1) provides a more precise estimation of effects on SSDI applications and receipts by leveraging the universe of records rather than a sample, (3) allows us to explore heterogeneous effects on SSDI applications and receipts by finer age cohorts and disability status.

This paper provides several contributions to the literature. First, the effect of disability and age discrimination laws on labor market outcomes is still not yet clear in either the theoretical or the empirical literature.³ Second, SSA1983 provides a notable and exogenous source of variation to identify the effects of these laws on labor market outcomes and SSDI application and receipt. SSA1983 affects older workers uniformly in a way that is not correlated with existing state laws.

Our DD approach of leveraging this exogenous “shock” avoids bias from the plausibly endogenous adoption of discrimination laws.⁴ Our assumption to identify an unbiased causal effect of state laws is that the age-specific trend difference between affected and unaffected cohorts does not vary in a way that correlates with existing state discrimination law. We believe that the likelihood that this assumption is violated is very small.⁵

Our use of SSA1983 as an exogenous source of variation also allows us to avoid bias from pre-existing differences in trends between treatment and control groups that are not sufficiently similar.⁶ For our study, we use more similar cohorts, those affected (cohorts born 1938–1951) and unaffected (1931–1937) by SSA1983, controlling for age. We then go further by comparing the affected and unaffected cohorts across states by existing state law, thus controlling for any differences in trends between these two groups.

Another unique contribution of our paper is that we know very little about how disability and age discrimination laws affect the application and receipt of SSDI.⁷ Our paper provides evidence of the effectiveness of important labor demand-side policies that could reduce the spillover effects of the Social Security reforms and otherwise reduce employment barriers and help alleviate strain on the Social Security Trust Fund. This is critical because the combined OASI and DI funds are projected to be exhausted by 2035 as the population continues to age, leading to increased withdrawals from the fund in the form of a higher caseload for OASI and SSDI (Board of Trustees 2017, 2020).

BACKGROUND ON SOCIAL SECURITY PROGRAMS

Social Security Amendments of 1983 (SSA1983)

Since 1975, the Old-Age, Survivors, and Disability Insurance (OASDI) program expenditures exceeded the existing law’s revenues. Without any legislative changes to the program, it would have been impossible to pay the OASDI cash benefits on time, beginning in July 1983. To deal primarily with the short-and long-term financial challenges faced by the OASDI program, the SSA1983 was signed into law on April 20, 1983. This law made sweeping changes in Social Security coverage, financing, and benefit structure. The changes included an increase in the payroll tax rate,⁸ expansion of the program to some federal government employees,⁹ and an increase in the actuarial adjustment factors beyond the FRA, increasing the benefit of retiring after the FRA (Svahn and Ross 1983).¹⁰ Perhaps the most significant change was a maximum of a two-year increase in the FRA and a corresponding increase from 20–30 percent in the penalty for claiming OASI benefits at the early retirement age of 62.

The first cohort affected by the SSA1983 was born in 1938. Their FRA increased from 65 years to 65 years and two months, and they reached this new FRA in 2003. Therefore, the SSA1983 had minimal effects on labor supply until later, such as starting in 2000 when the first affected cohort, born in 1938, reached the early retirement age of 62.

As shown in Table 1, these reductions in OASI benefits’ generosity were phased in gradually and occurred in two main stages. Beyond the 1938 cohort, who faced a two month increase in the FRA, the FRA then increased two months for each additional birth cohort year until reaching age 66 for those born in 1943. For individuals born between 1943 and 1954 (inclusive), the FRA remained 66 years until again increasing in two-month increments each birth year for the 1955–1960 cohorts. Along with this change, the proportion of full benefits that individuals could receive at the early retirement age of 62 fell from 80 percent for those born in 1937 to 75 percent for those born between 1943 and 1954 and to 70 percent for those born in 1960 or later.¹¹ While these amendments changed the OASI benefit structure, they did not change the benefit structure of the SSDI program, leading SSDI to be relatively more generous for some affected older workers.

Social Security Disability Insurance

The SSDI program is a social insurance program for workers with disabilities, with eligibility conditioned on previous sufficient employment in jobs covered by Social Security.¹² It is a part of the more extensive OASDI program of Social Security. The SSDI program defines disability as the “inability to engage in substantial gainful activity (SGA) by reason of any medically determinable physical or mental impairment(s) which can be expected to result in death, or which has lasted or can be expected to last for a continuous period of not less than 12 months.”¹³ An activity is considered “substantial” if it involves significant physical or mental exertion and “gainful” if performed for pay or profit. SSA implements the definition of SGA by setting an earnings threshold, which is adjusted over time. For example, in 2018, the earnings threshold was $1,180 per month for nonblind individuals and $1,970 per month for blind individuals.¹⁴ Anyone earning over the threshold is considered engaging in SGA and is therefore disqualified from participating in the SSDI program.

A summary of the SSDI application process is as follows. Individuals apply for SSDI benefits at their local field office, which screens out those not currently insured (have insufficient work history covered by Social Security) or engaging in SGA. These are labeled as “technical denials” and do not receive further review. The remaining applications are forwarded to a state Disability Determination Services (DDS) office, where cases are assigned randomly to disability examiners for review. The disability examiner determines disability eligibility using rules and medical and vocational criteria laid out in the code of federal regulations. The rejected applicants at the DDS level are then entitled to a series of appeals.¹⁵ The applicants can bring in new information to each level of appeals to make their case stronger. Because the appeal process can take months or years, some applicants who appeal to the DDS rejection may simultaneously file a new application to get final adjudication.

In 2016, approximately 8.8 million individuals with disabilities received SSDI benefits.¹⁶ Among these beneficiaries, 4.5 million were men, and 4.3 million were women. The average age of beneficiaries was 54 years. Of the population ages 18–64, 4.7 percent received SSDI benefits in 2016. The average cash benefit received was $1,171 in 2016 ($1,293 for men, $1,043 for women), for a total of $11.3 billion across all beneficiaries. On average, two-thirds of the SSDI applicants are denied benefits at the Disability Determination Services (DDS) office. In 2015 alone, 2.4 million individuals applied for SSDI, and 1.5 million were denied at the DDS level. However, a fraction of denied applicants appeal their initial denial and get approved for the benefits.

LITERATURE REVIEW

Social Security Programs

The Social Security Act and the regulations implementing it set up universal criteria to determine the disability status of someone who applies for SSDI benefits. Historically, not only are there variations in the application rate for SSDI across states¹⁷ and over time, but there also are similar variations in the receipt rate of SSDI (McVicar 2006; Stapleton et al. 1998). Much of the variations in application and receipt rates can be explained by economic, health, and demographic factors (Coe et al. 2011; Gruber and Kubik 1997; Strand 2002). We contribute to this literature by incorporating state discrimination laws in explaining the variations in SSDI application and receipt rates across states.

Both OASI and SSDI’s cash benefits are calculated based on earnings history (see Online Appendix B for more details). The main difference between the OASI and SSDI benefit calculations is that actuarial adjustment factors apply only to OASI when individuals between age 62 and the FRA claim the benefit. The SSA1983 increased the FRA for cohorts born after 1937 and the penalty of claiming OASI benefits earlier than the FRA. Individuals may apply for SSDI benefits up to their FRA, and there is no penalty in claiming the SSDI benefits earlier than the FRA. The fact that SSDI benefits, unlike OASI benefits, are not subject to actuarial reduction makes SSDI application quite valuable to individuals who think that they are disabled enough to qualify for SSDI.

The present value of OASI benefits at any given age is considerably lower for cohorts affected by the SSA1983 compared to those unaffected. The reduction in benefits is also greater for the cohorts with a higher FRA, making SSDI relatively more generous than OASI for younger cohorts insured under both programs (Duggan, Singleton, and Song 2007). Theoretically, this relative generosity provides greater incentives for people who would have claimed the OASI benefits earlier than the FRA to apply for and possibly enroll in the SSDI program.

The incentive to apply for SSDI is higher for cohorts born in more recent years and for those who are closer to the age of claiming OASI benefits (see Table 1). Consequently, this incentive is proportionately greater at any given age for workers born on or after 1938, depending on their FRA, compared to people born on or before 1937. As a result, the incentive to apply for SSDI is proportionately greater for workers with higher FRA at any age before their FRA.

Effects of Employment Discrimination Laws

There are three lines of related literature on the effects of discrimination laws on labor market outcomes. First, there is literature on the effect of disability discrimination laws on labor market outcomes, and sometimes SSDI claiming, for individuals with disabilities. Second, there is more extensive literature on the effect of age discrimination laws on labor market outcomes and retirement for older workers. Third, there is a small but growing literature on the effect of disability discrimination laws on older workers’ labor market outcomes. Also related are discrimination laws covering other groups summarized by Button (2018).

The research on the effects of discrimination laws on labor market outcomes generally does not come to a clear conclusion. Theory suggests mixed impacts, where discrimination laws could reduce terminations but reduce hiring (Bloch 1994). For disability, laws could also improve workplace accommodations, increasing productivity and tenure of employees. However, the extra costs imposed by reasonable accommodations may strengthen the hiring disincentivize (Acemoglu and Angrist 2001). The empirical literature also finds mixed impacts, although most studies find effects that lean positive or non-negative. We summarize each of these types of literature in depth in Online Appendix A and provide summary tables for all the studies (Online Appendix Tables A1, A2, and A3).

DATA

Disability Discrimination Laws

We use a database of state disability discrimination laws first created by Neumark, Song, and Button (2017) and then updated by Neumark et al. (2019). The creation of this database required extensive background research on statutes in states and D.C.,¹⁸ and their histories (primarily found through Westlaw), acts that amended these statutes (primarily found using Hein Online), and many other sources (e.g., case law, secondary sources, law journal articles, state offices). See Neumark et al. (2019), especially its online appendix, for a more detailed discussion of the coding of these laws.

This database includes three dimensions of how state disability discrimination laws are broader or stronger than the federal ADA: “medical definition of disability,” “lower firm size (disability),” and “larger damages (disability).” Table 2 shows how these characteristics of disability discrimination laws vary by state and how we code these laws into indicator variables. First, and most importantly, some states have a lower burden of proof to establish a disability than under the definition of disability in the ADA, leading to increased coverage of disability discrimination law in those states. These “medical definition” states define individuals with a medically diagnosed condition to be disabled under the law, regardless of whether their condition “substantially limits” a “major life activity.” Second, the ADA applies only to firms with at least 15 employees, but state laws often cover firms with fewer workers “lower firm size (disability).” Third, many states also allow for plaintiffs to sue for more damages than the ADA, leading state laws to have more bite than the ADA “larger damages (disability)” (Button, Armour, and Hollands 2017a,b; Neumark, Song, and Button 2017; Neumark et al. 2019). Below we summarize this legal variation, and we refer the reader to Neumark et al. (2019) for more information.

Table 2.

State disability and age discrimination laws, 1992–2000.

	Disability discrimination laws			Age discrimination laws
State	Minimum firm size	Larger damages than ADA	Medical definition	Minimum firm size	Larger damages than ADEA
Alabama	No law	No law	No law	20	No
Alaska	1	Yes	No	1	No/Yes (changed 1997)
Arizona	15	No	No	15	No
Arkansas	9	No (same as ADA)	No	No law	No law
California	5	Yes (uncapped)	No (“limits” as of 2001)	5	Yes
Colorado	1	No (same as ADA)	No	1	No
Connecticut	3	No (unclear)	Yes	3	No
Delaware	4	No (same as ADA)	No	4	Yes
District of Columbia	1	Yes (uncapped)	No	1	Yes
Florida	15	No (punitive capped at $100k)	No	15	Yes
Georgia	15	No	No	1	No
Hawaii	1	Yes (uncapped)	No	1	Yes
Idaho	5	No (punitive capped at $10k)	No	5	Yes
Illinois	1	No	Yes	15	Yes
Indiana	15	No	No	1	No
Iowa	4	No	No	4	Yes
Kansas	4	No (no punitive damages, damages capped at $2k)	No	4	Yes
Kentucky	15	No	No	8	Yes
Louisiana	15	No	No	8/20 (changed 1997)	Yes
Maine	1	Yes	No	1	Yes
Maryland	15	No (same as ADA)	No	15	Yes
Massachusetts	6	Yes (uncapped)	No	6	Yes
Michigan	1	No	No	1	Yes
Minnesota	1	No (punitive capped at $25k)	No (“materially limits”)	1	Yes
Mississippi	No law	No law	No law	No law	No law
Missouri	6	Yes (uncapped)	No	6	Yes
Montana	1	No	No	1	Yes
Nebraska	15	No	No	25	No
Nevada	15	No	No	15	No
New Hampshire	6	No	No	6	Yes
New Jersey	1	Yes (uncapped)	Yes	1	Yes
New Mexico	4	No	No	4	Yes
New York	4	No	Yes	4	Yes
North Carolina	15	No	No	15	No
North Dakota	1	No (no damages)	No	1	No
Ohio	4	Yes (uncapped)	No	4	Yes
Oklahoma	15	No	No	15	No
Oregon	6	Yes (uncapped)	No	1	Yes
Pennsylvania	4	No	No	4	No
Rhode Island	4	Yes (uncapped)	No	4	Yes
South Carolina	15	No (same as ADA)	No	15	No
South Dakota	1	No	No	No law	No law
Tennessee	8	No	No	8	Yes
Texas	15	No (same as ADA)	No	15	No/Yes (changed 1993)
Utah	15	No	No	15	No
Vermont	1	Yes (uncapped)	No	1	No/Yes (changed 1999)
Virginia	1	No	No	1	No
Washington	8	No	No	8	Yes
West Virginia	12	Yes (uncapped)	No	12	No
Wisconsin	1	No	No	1	No
Wyoming	2	No	No	2	No

NOTE: State laws cover 1992–2000. For the states listed as “Yes” under “Larger damages than ADA,” but not uncapped, details are as follows: Alaska—uncapped compensatory damages, punitive damages capped above ADA levels; Maine—exceeds ADA cap for firms of 201+ employees. For states listed as “No” under “Larger damages than ADA,” unless otherwise noted, the “No” is because punitive damages are not allowed. See Neumark et al. (2019), and especially the online appendix to that paper, for more information on the laws.

SOURCE: Age discrimination laws are from Neumark and Song (2013) and disability discrimination laws are from Neumark et al. (2019).

Medical definition of disability

The federal disability discrimination in employment law, Title I of the Americans with Disabilities Act of 1990 (effective July 1992), is significantly less broad in its coverage compared to employment discrimination laws covering other groups.¹⁹ This stems from a narrow definition of disability. For plaintiffs to establish that they have a disability under the ADA, and to thus even have a prima facie case that can go forward, they must meet one of the three definitions of disability in the ADA:

“The term ‘disability’ means, with respect to an individual, (A) a physical or mental impairment that substantially limits one or more major life activities²⁰ of such individual; (B) a record of such an impairment; or (C) being regarded as having such an impairment.” (42 U.S. Code §12102 (1))

The difficulty for plaintiffs was proving that the condition was at the level of “substantial.” The burden of proof for “substantially limits” was high, leading plaintiffs to have their cases effectively thrown out because they did not meet the standard (Burgdorf 1997; Colker 1999).

Four states (Connecticut, Illinois, New Jersey, and New York) have a broader definition of disability whereby individuals are considered disabled under state law if they have a diagnosed medical condition.²¹ This is irrespective of if the condition “substantially limits” a “major life activity,” thus significantly lowering the burden of proof for individuals in these states to be covered by state disability discrimination law (Long 2004; Neumark, Song, and Button 2017).

Lower firm size minimum (disability)

The ADA covers firms with at least 15 employees. As Table 2 shows, 34 states as of 1992 had disability discrimination laws with firm size minimums lower than 10, 25 states that are lower than 5, and 15 states that cover all employees regardless of firm size. The smaller firm size matters since older workers are more likely to work at smaller firms.²² Following previous studies (Neumark, Song, and Button 2017; Neumark et al. 2019), in our analysis, we create an indicator variable equal to one if the firm size is less than ten, zero otherwise, and compare states with a firm size of less than ten versus greater than ten.

Larger damages (disability)

There is also variation in damages available under state law. Many states allow more damages than the capped sum of compensatory and punitive damages allowed under the federal ADA, which range from $50,000-$300,000, depending on the firm’s size.²³ Following Neumark, Song, and Button (2017) and Neumark et al. (2019), we deem states to have “larger damages (disability)” if they have damage caps that exceed the ADA, or allow uncapped punitive damages, as punitive damages are likely to drive large judgments. As shown in Table 2, 13 states, as of 1992, allow larger damages than the ADA.

Age Discrimination Laws

We use the age discrimination laws as determined and coded by Neumark and Song (2013). While there are several dimensions of state laws, the critical two that the literature focuses on (Neumark and Button 2014; Neumark and Song 2013; Neumark et al. 2019) are the availability of damages and the firm size minimum. The idea behind both is like their counterparts for disability, but they differ in realization.

Lower firm size minimum (age)

Firm size operates similarly, except that the federal ADEA covers only firms with at least 20 employees, while the ADA covers firms with at least 15 employees. As shown in Table 2, there are 46 states with a lower firm size than the ADEA, 36 states with a firm size less than 15, 35 less than 10, 26 less than 5, and 16 states that cover all firms. The firm size for age and disability tends to match since both often are in the same statute, but this is not the case for nine states as of 1992.²⁴

Larger damages (age)

Damages available are different for age laws. Instead of caps on compensatory and punitive damages, like in the ADA, the ADEA does not allow compensatory or punitive damages. It allows for back pay and benefits—double this amount (“liquid damages”) if there is a willful violation (Neumark and Song 2013). We follow Neumark and Song (2013) and deem any state that allows compensatory or punitive damages as having “larger damages (age).” As shown in Table 2, 27 states have larger damages for age discrimination law as of 1992. More states have larger damages for age than for disability because of the restrictive way that damages are treated in the ADEA. Because of this, the overlap for damages for age and disability under state laws is much lower than it is for firm size.²⁵

Aggregated SSDI Application and Receipt Data

We use aggregated SSDI application and receipt data from the Form-831 disability records, generated at state Disability Determination Services (DDS) offices and provided by the Social Security Administration.²⁶ This SSA data includes the universe of the yearly total number of applications and receipts of SSDI for each state by sex and age cohorts (age 45–49, 50–54, 55–59, and 60–64) from 1992 to 2013. Using the year and age cohort information, we categorize individuals as either being affected by SSA1983 or unaffected by SSA1983.²⁷

State-by-Year Demographic and Economic Control Variables

We use the Current Population Survey (CPS) Annual Social and Economic Supplement (ASEC) (Flood et al. 2018) to get population estimates by sex, state, and age cohort from 1992–2013. We generate SSDI application and receipt rates by dividing the number of SSDI applications and receipts, respectively, by the sex and age cohort-specific population in each state from 1992–2013. We also use the CPS ASEC to generate these state- and year-specific population estimates by age cohort and sex. We also use the CPS ASEC to estimate several socio-economic and demographic control variables.²⁸

The Health and Retirement Study (HRS) and Form-831 Social Security Administration Disability Records

To complement the aggregated SSDI data, we also use data from the Health and Retirement Study (HRS). The HRS is a nationally representative longitudinal household survey that interviews Americans over the age of 50 every two years. We use twelve waves of data from 1992–2014. We primarily use the RAND HRS version P data (Bugliari et al. 2016). We complement this data with additional variables from different HRS modules, downloaded directly from the HRS website.²⁹

The HRS also has data on employment, which allows us to test the mechanisms of how state discrimination laws could impact spillovers onto SSDI. We can see if stronger or broader laws increase (decrease) employment and if this is associated with a decrease (increase) in SSDI application or receipt. Given that the HRS data is at the individual level (rather than the state-by-age cohort-by-sex-by-year level as with the aggregated SSDI data), we can also use the HRS data to better explore heterogeneity by demographic factors, namely by disability status. However, this is at the cost of significantly reduced precision since the aggregated SSDI data includes the universe of SSDI applications and receipts, while the HRS data is just a population-representative sample.

To analyze SSDI application and receipt using the HRS data, we combine the public HRS data files with the SSA’s administrative Form-831 disability records. The Form-831 data allows us to determine if and when HRS respondents applied for or received SSDI, allowing us to make similar SSDI application and receipt rate variables as with the aggregated SSDI data. Our combined HRS-Form-831 data covers eleven waves of data from 1992–2012. For more information on this Form-831 data, see Online Appendix B.³⁰

We code individuals in the HRS as having a disability using two different measures. This is important since there are numerous types of disabilities and different classifications have pros and cons.³¹ First, we use the work-limiting measure of disability. Individuals have a work-limiting disability if they answer “yes” to having an impairment or health problem that limits the kind or amount of paid work. Second, we use the Activities of Daily Life (ADL) limitations measure of disability. Under this measure, we consider individuals to have a disability if they report one or more ADL limitations.

Summary Statistics

Table 3 presents summary statistics for the aggregated SSDI application and receipt data, and Online Appendix Table B1 presents summary statistics for the associated CPS ASEC controls. Across all states and D.C. (Column [1] of Table 3), the SSDI application rate is 0.0139 (1.39 percent per year), and the SSDI receipt rate is 0.00796 (0.796 percent per year). Notably, SSDI application rates are lower in states with a medical definition of disability (1.27 percent, Column [2]). For SSDI receipt, measured as receipts per capita, this is reversed with somewhat higher receipt rates in medical definition states compared to states without any broader or stronger disability discrimination laws (0.796 percent versus 0.783 percent). States also differ in the control variables (Online Appendix Table B1), which correlate with SSDI program use according to the previous literature (Coe et al. 2011; Gruber and Kubik 1997; Strand 2002). Online Appendix B2 presents a similar table to Table 3 but using the HRS matched with Form-831 disability records.

Table 3.

Summary statistics – aggregated SSDI application and receipt count data, by state discrimination law features.

	(1)	(2)	(3)	(4)	(5)	(6)
	All states	Medical definition of disability	Larger damages (disability)	Firm size < 10 (disability)	Larger damages (age)	Firm size < 10 (age)
SSDI application rate	0.0139	0.0127	0.0137	0.0131	0.0136	0.0134
SSDI receipt rate	0.00796	0.00818	0.00815	0.00792	0.00783	0.00796
N	2,754	270	810	1,782	1,620	1,836

NOTE: These are per year rates for 1992–2013. Observations are at the state, year, sex, and age cohort (55 to 59 or 60 to 64). We weight these means using population estimates for each age group and state from the American Community Survey.

SOURCE: Authors’ calculations from aggregated Social Security Disability Insurance statistics, graciously provided by Alexander Strand, matched with the age and disability discrimination law data presented in Table 2. These rates are calculated by dividing the counts by population estimates from the Current Population Survey.

Table 4 presents SSDI application and receipt rates by sex for the two age groups we use in our primary analysis in the aggregated SSDI application and receipt data: ages 55 to 59 and 60 to 64. Men have higher SSDI application and SSDI receipt rates. SSDI application (receipt) rates are 1.60 (0.948) percent for men ages 55 to 59 and 1.66 (1.05) percent for men ages 60 to 64, compared to 1.25 (0.607) percent and 1.07 (0.570) percent for women. For men, these SSDI application and receipt rates increase somewhat by age. Conversely, SSDI application rates decrease significantly with age for women, with only a small decrease in the SSDI receipt rates.

Table 4.

Summary statistics – aggregated SSDI application and receipt count data, by age and sex.

	(1)	(2)	(3)	(4)
	Ages 55 to 69		Ages 60 to 64
	Women	Men	Women	Men
SSDI application rate	0.0125	0.0160	0.0107	0.0166
SSDI receipt rate	0.00607	0.00948	0.00570	0.0105
N	561	561	816	816

NOTE and SOURCE: See Table 3.

Table 5 presents summary statistics for any employment by work-limiting or ADL limitation status. Online Appendix Table B3 presents disability rates, using both the work-limited and any ADL limitation measures. For the entire sample, their employment rate is 61.1 percent. For those with a work-limiting disability (23.7 percent of the sample), their employment rate is 27.3 percent. Those with any ADL limitation are a much larger group (63.0 percent of the sample) and their employment rate is 55.0 percent.

Table 5.

Summary statistics – HRS data, by disability status.

	(1)	(2)	(3)	(4)	(5)
	Entire Sample	No Work-Limiting Disability	Work-Limiting Disability	No ADL Limitation	ADL Limitation
Any employment	0.611	0.713	0.273	0.717	0.550
N	41,472	31,527	9,945	14,957	26,515

NOTE: Any employment equals one if individuals have a full or part-time job or are partially retired. We restrict the sample to ages 55 to the FRA, born between 1931–1948, and have worked at least five years. We exclude individuals from Puerto Rico and those with missing state identifiers. Summary statistics are weighted by HRS person-level weights.

SOURCE: This data is from the public HRS data for 1992–2014. Author’s calculations using public Health and Retirement Study data for 1992–2014.

Table 6 presents SSDI application and receipt rates by disability status from the HRS data matched with Form-831 SSA records. The SSDI application (receipt) rate is 0.975 percent (0.563 percent) for every two years for the entire sample. This varies significantly by disability status. For those with (without) a work-limiting disability, their SSDI application rate is 2.45 percent (0.531 percent). This difference is even more considerable for SSDI receipt: 1.62 percent versus 0.243 percent, which is not surprising since those who report disabilities would have a stronger case for receiving SSDI. The differences in SSDI application and receipt rates by ADL limitation are similar but less pronounced.

Table 6.

Summary statistics – HRS merged with Form-831 data, disability status.

	(1)	(2)	(3)	(4)	(5)
	Entire Sample	No Work-Limiting Disability	Work-Limiting Disability	No ADL Limitation	ADL Limitation
Applied for SSDI	0.00975	0.00531	0.0245	0.00326	0.0135
Received SSDI	0.00563	0.00243	0.0162	0.00123	0.00819
N	41,794	31,767	10,027	15,048	26,746

NOTE: See note to Table 5. These are two-year average rates. This data is from the public HRS data merged with Form-831 data for 1992–2012.

SOURCE: Author’s calculations using Health and Retirement Study data, matched with Social Security Administration Form-831 disability records, restricted-access state identifiers, and the age and disability discrimination law data presented in Table 2. We access this data virtually through the MiCDA enclave.

METHODOLOGY

Aggregated SSDI Application and Receipt Data

We first use the aggregated SSDI application and receipt counts by state, year, age cohort, and sex. We exploit state variation in the degree of disability (age) discrimination protections relative to the ADA (ADEA) to identify the moderating effects of stronger or broader discrimination laws on the spillover effects of SSA1983 for those of any sex of ages 55 to 64. Using a difference-in-differences (DD) framework, we estimate the effects of stronger and broader discrimination laws on SSDI application and receipt, based on the differences in SSDI application and receipt of cohorts affected by SSA1983 and unaffected cohorts observed at the same age across states. Our regression equation is:

$${\mathit{y}}_{\mathit{a}\mathit{s}\mathit{t}}=\mathit{\gamma }{\mathit{S}\mathit{S}\mathit{A}1983}_{\mathit{a}\mathit{t}}+{\mathit{S}\mathit{S}\mathit{A}\mathbf{1983}}_{\mathit{\text{at}}}*{\mathit{L}\mathit{A}\mathit{W}\mathit{S}}_{\mathit{s}}\mathit{\beta }+{\mathit{\text{State}}}_{\mathit{s}}\mathit{\theta }+{\mathit{\text{AgeCohort}}}_{\mathit{a}}\mathit{\omega }+{\mathit{\text{Time}}}_{\mathit{t}}\mathit{\psi }+{\mathit{X}}_{\mathit{s}\mathit{t}}\mathit{\delta }+{\mathit{\epsilon }}_{\mathit{a}\mathit{s}\mathit{t}}$$ [1]

where, y_ast represents the two outcome variables defined as the proportion of the working-age population in age-cohort a living in state s in year t that applied for SSDI or received SSDI. Since receipt of SSDI occurs on average 15 months after application, we lag the SSDI receipt variable one year (Autor et al. 2015).³²

SSA1983 is an indicator variable that takes the value of 1 for cohorts born after 1937 (affected by SSA1983) and takes 0 otherwise (unaffected). X_st includes control variables at the state-year level: labor force participation rates, unemployment rates, the proportion of the working-age population that has completed high school, the proportion with a work-limiting disability, and per capita real disposable personal income. We include age-cohort fixed effects (AgeCohort), state fixed effects (State), year fixed effects (Time). We cluster our standard errors at the state level.

LAWS is a set of indicator variables for each stronger or broader age or disability discrimination law feature under state law. More specifically, these are indicator variable for if the state uses the medical definition of disability under disability discrimination law (“medical definition of disability”), provides larger potential damages for disability discrimination compared to the ADA (“larger damages (disability)”), provides larger potential damages for age discrimination compared to the ADEA (“larger damages (age)”), and if the minimum firm size (number of employees) for either age or disability discrimination law to apply is fewer than 10 (“Firm size < 10 (disability or age)”).³³

β captures the DD estimate, which represents the change in the outcome variable for the cohorts affected by the SSA1983 living in states with stronger or broader discrimination laws compared to both the unaffected cohorts and the affected cohorts living in control states without stronger or broader discrimination laws. Our key identifying assumption is that the trends in the unobserved factors affecting the SSDI application and receipt of different cohorts do not systematically vary across states based on their discrimination laws.

We estimate results for both women and men separately and pool them together in the same sample. This analysis by sex is critical since age discrimination varies by sex, as does the effectiveness of age (and perhaps disability) discrimination laws.³⁴

Health and Retirement Study Data

Average Spillovers of SSA1983 to SSDI and Employment

Before investigating state discrimination laws, we first estimate the average spillovers to SSDI applications and receipt and effects on employment induced by SSA1983. We start by estimating effects on the entire sample (ages 55 to the FRA), then by age cohort (ages 55 to 61, ages 62 to 64, and ages 65 to the FRA³⁵), disability status, sex, and intersections of these groups. The general structure of these regressions, using the example of estimating effects by disability status for each age cohort, is:

$${\mathit{y}}_{\mathit{i}\mathit{a}\mathit{s}}={\mathit{S}\mathit{S}\mathit{A}1983}_{\mathit{i}}*{\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_1+{{\mathit{\text{Disabled}}}_{\mathit{i}}*\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_2+{\mathit{S}\mathit{S}\mathit{A}1983}_{\mathit{i}}*{\mathit{\text{Disabled}}}_{\mathit{i}}*\mathit{\text{AgeCohor}}{\mathit{t}}_{\mathit{i}}{\mathit{\tau }}_3+{\mathit{\text{State}}}_{\mathit{s}}\mathit{\theta }+{\mathit{\text{Age}}}_{\mathit{a}}\mathit{\omega }+{\mathit{X}}_{\mathit{i}\mathit{a}\mathit{s}}\mathit{\delta }+{\mathit{\epsilon }}_{\mathit{i}\mathit{a}\mathit{s}}$$ [2]

where, y_ias represents the three outcome variables defined as individual i who is age a living in state s that applied for SSDI, received SSDI, or was employed. AgeCohort is a set of mutually exclusive age cohort indicators for ages 55 to 61, ages 62 to 64, ages 65 up to the FRA.³⁶ Disabled is an indicator variable for having a disability. We measure this in two ways: (1) having a health condition that limits work (“work-limiting disability”) or (2) having a limitation to an activity of daily living (“ADL limitation”). SSA1983 is defined in the previous section.

τ₁ captures the effect of SSA1983 by age cohort for those without disabilities, compared to similar individuals who were not affected by SSA1983. τ₂ captures average differences in outcomes by age cohort between those with and without disabilities, for those unaffected by SSA1983. Finally, τ₃ captures the differential effect of SSA1983 for those with disabilities for each age cohort.

X_ias includes control variables at the individual level: race, sex, marital status, number of years worked, retiree health insurance status, health insurance status, and self-reported health status. We include fixed effects for each age in years (Age) and state fixed effects (State).

We cluster our standard errors at the state level. We weight our regressions by population to make our estimates nationally representative, rather than implicitly weighting less populous states the same as populous states.³⁷

Moderating Effects of Discrimination Laws on the Effects of SSA1983

We then analyze the HRS data, and the HRS data merged with the SSA Form-831 data. To explore all this, we follow a similar DD framework as with the aggregated SSDI data, estimating how existing state laws moderated the effects of SSA1983 on SSDI application, SSDI receipt, and employment.

We first start by estimating the moderating effects of laws by age cohort and sex. Our regression model is like equation [2] above but replaces the interactions with disability status with interactions with state discrimination laws (LAWS):

$${\mathit{y}}_{\mathit{i}\mathit{a}\mathit{s}}={\mathit{S}\mathit{S}\mathit{A}1983}_{\mathit{i}}*{\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_1+{{\mathit{\text{LAWS}}}_{\mathit{s}}*\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_2+\mathit{S}\mathit{S}\mathit{A}{\mathbf{1983}}_{\mathit{i}}*{\mathit{\text{LAWS}}}_{\mathit{s}}*\mathit{\text{AgeCohor}}{\mathit{t}}_{\mathit{i}}\mathit{\beta }+{\mathit{\text{State}}}_{\mathit{s}}\mathit{\theta }+{\mathit{\text{Age}}}_{\mathit{a}}\mathit{\omega }+{\mathit{X}}_{\mathit{i}\mathit{a}\mathit{s}}\mathit{\delta }+{\mathit{\epsilon }}_{\mathit{i}\mathit{a}\mathit{s}}$$ [3]

where LAWS is defined in the previous section. We again cluster our standard errors at the state level. β captures the central coefficient of interest, which is if the effects of SSA1983 differed by existing state laws.

We then estimate the moderating effects of laws by age cohort, sex, and disability status. This makes our analysis a difference-in-difference-in-differences (DDD), with three dimensions (affected and unaffected by SSA1983, with and without a disability, and living in or not living in a state with stronger or broader discrimination laws). Our regression model is:

$${\mathit{y}}_{\mathit{i}\mathit{a}\mathit{s}}={\mathit{S}\mathit{S}\mathit{A}1983}_{\mathit{i}}*{\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_1+{{\mathit{\text{Disabled}}}_{\mathit{i}}*\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_2+{{\mathit{\text{LAWS}}}_{\mathit{s}}*\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_3+{\mathit{S}\mathit{S}\mathit{A}1983}_{\mathit{i}}*{\mathit{\text{Disabled}}}_{\mathit{i}}*\mathit{\text{AgeCohor}}{\mathit{t}}_{\mathit{i}}{\mathit{\tau }}_4+{\mathit{\text{LAWS}}}_{\mathit{s}}*\mathit{\text{Disable}}{\mathit{d}}_{\mathit{i}}*{\mathit{\text{AgeCohort}}}_{\mathit{i}}{\mathit{\tau }}_5+\mathit{S}\mathit{S}\mathit{A}{1983}_{\mathit{i}}*{\mathit{\text{LAWS}}}_{\mathit{s}}*\mathit{\text{AgeCohor}}{\mathit{t}}_{\mathit{i}}{\mathit{\tau }}_6+\mathit{S}\mathit{S}\mathit{A}{\mathbf{1983}}_{\mathit{i}}*{\mathit{\text{LAWS}}}_{\mathit{s}}*{\mathit{\text{Disabled}}}_{\mathit{i}}*\mathit{\text{AgeCohor}}{\mathit{t}}_{\mathit{i}}\mathit{\beta }+{\mathit{\text{State}}}_{\mathit{s}}\mathit{\theta }+{\mathit{\text{Age}}}_{\mathit{a}}\mathit{\omega }+{\mathit{X}}_{\mathit{i}\mathit{a}\mathit{s}}\mathit{\delta }+{\mathit{\epsilon }}_{\mathit{i}\mathit{a}\mathit{s}}$$ [4]

β in this regression captures the main coefficient of interest, which is the difference between two DD estimates: (a) the moderating effect of state laws on SSA1983 by age cohort for those with a disability compared to (b) the moderating effect of state laws on SSA1983 by age cohort for this without a disability. That is, β captures if the moderating effects of existing state laws differ for those with disabilities.

RESULTS

Results Using Aggregated SSDI Application and Receipt Count Data

Table 7 presents the results for SSDI application and receipt rates by sex for age cohorts 55 to 59 and 60 to 64. The first row estimates the effects on those in states without any broader or stronger features of disability or age discrimination laws, i.e., the ADA and ADEA provide the strongest and broadest coverage. The following four rows of estimates show the differential spillover for states with that legal feature. Thus, the total estimated spillover for, say, states with the medical definition of disability only would be the sum of the coefficient on SSA1983 plus the coefficient on SSA1983 * medical definition of disablity.

Table 7.

Interactive effects of state discrimination laws and SSA1983-driven spillovers, aggregated SSDI count data.

	(1)	(2)	(3)	(4)	(5)	(6)
Outcome:	SSDI Application Rate			SSDI Receipt Rate
Sex:	All	Female	Male	All	Female	Male
SSA1983	0.00163*** (0.000567)	0.00150** (0.000632)	0.00176** (0.000783)	0.000674 (0.000443)	0.000439 (0.000384)	0.000909 (0.000629)
SSA1983 x …
Medical definition of disability	−0.00162** (0.000789)	−0.00132** (0.000518)	−0.00191 (0.00120)	−0.00107 (0.000641)	−0.000576* (0.000341)	−0.00157 (0.00102)
Larger damages (disability)	−0.000908* (0.000463)	−0.00143*** (0.000430)	−0.000385 (0.000625)	−0.000496 (0.000452)	−0.000661** (0.000300)	−0.000332 (0.000661)
Larger damages (age)	−0.000401 (0.000491)	−0.0000428 (0.000427)	−0.000760 (0.000727)	0.0000975 (0.000540)	0.000162 (0.000419)	0.0000327 (0.000730)
Firm size < 10 (disability or age)	0.000101 (0.000439)	0.0000984 (0.000425)	0.000104 (0.000633)	0.0000428 (0.000485)	0.000148 (0.000384)	−0.0000619 (0.000662)
N	2,744	1,372	1,372	2,744	1,372	1,372

NOTE: The mean SSDI application (receipt) rate is 0.0139 (0.00796). Standard errors, in parentheses, are clustered at the state level. Estimates come from equation [1]. Significantly different from zero at 1-percent level (***), 5 percent level (**), or 10 percent level (*). Age cohorts 55–59 and 60–64 are included in all the regressions. The models include age-cohort fixed effects, state fixed effects, and year fixed effects. We weight these regressions by using population estimates for each age group and state from the American Community Survey. All the models also include variables that vary across states and over time: labor force participation rates, unemployment rates, the proportion of African-American/black, the proportion of Hispanic, the proportion of working-age population completed high school degree, the proportion of working-age population reported work-limiting disability, proportion of population 50 or over, faction of the population works in agriculture, mining, and construction industry, and per capita real disposable personal income.

SOURCE: See the source for Table 3.

We start by discussing the effects on the SSDI application rate. For the medical definition of disability, we find that there are effectively no spillovers to SSDI applications in states with a medical definition of disability. The coefficient estimate for the interaction between SSA1983 and the medical definition is −0.00162 (standard error 0.000789, significant at the 5 percent level), which is like the estimate of the spillover in control states (0.00163). This effect is similar by sex, with the estimates being slightly larger for men but more precise for women.

Larger damages under disability discrimination law are also associated with a decrease in the SSDI application rate, but in this case, only for women. For women, the coefficient on the interaction between SSA1983 and larger damages (disability) is −0.00143 (significant at the 1 percent level) and nearly the same size as spillover in control states (0.00150). This suggests, again, that for women, the spillovers to SSDI applications were concentrated in states without the medical definition of disability or larger damages for disability.

We do not find any relationship between larger damages for age discrimination or a lower firm size minimum for age or disability discrimination laws. These estimates are small (ranging from −0.000760 to 0.000104) and never statistically significant.

The effects of state disability and age discrimination laws on SSDI receipt rates are similar, although the estimates are less precise and therefore less statistically significant. The only two estimates that are statistically significant are for women. These are the medical definition of disability (−0.000576, significant at the 10 percent level) and larger damages for disability (−0.000661, significant at the 5 percent level). Again, we do not find any relationship between larger damages for age discrimination or a lower firm size minimum.

Robustness checks

We conduct several robustness checks, considering other plausible specifications. We present these results fully in Online Appendix D. Here, we summarize whether our results vary based on the following:

1. Re-defining the “medical definition of disability” to include Minnesota, which also has a broader (but not medical) definition of disability during the timespan of our sample (Online Appendix Table D1).

2. Removing the weights by the population of the state of age 50 and older so that the results are now unweighted (Online Appendix Table D2).

3. Dropping 2009 and 2010 from the analysis to remove possible effects from the Great Recession (Online Appendix Table D3).³⁸

4. Removing the CPS ASEC control variables (Online Appendix Tables D4).

5. Including ages 50 to 54, in addition to default ages 55 to 59 and 60 to 64.

Online Appendix Table D6 summarizes the results of these robustness checks and the results in Table 7 for our two law variables with significant results: the medical definition of disability and larger damages under disability discrimination law. Online Appendix Table D6 shows that our results are very robust across specifications, except they are only somewhat robust to choosing now to weight the results by state population. The results for women only, which are always statistically significant, remain significant in all cases except for one case when we do not weight by state-by-age cohort population.³⁹ The results for men remain insignificant except, again, for the unweighted results for the medical definition of disability, where there is a statistically significant decrease in the SSDI application rate at the 1 percent level.

Results Using Health and Retirement Study and Form-831 SSA Data

Average Spillovers of SSA1983 to SSDI and Effects on Employment

We start by estimating the average spillover effects of SSA1983 (equation [2]), ignoring, for now, the possible moderating effects of existing state discrimination laws. Given that the effects of SSA1983 vary by age (e.g., Neumark and Song 2013), disability status, and sex, we estimate separate effects of SSA1983 for all combinations of sex (women, men, pooled), age group (ages 55 to 61, ages 62 to 64, ages 65 to the FRA), and disability status (with and without either a work-limiting disability or an ADL limitation).⁴⁰

Spillovers to SSDI applications

Table 8 presents the results for SSDI applications. The first three rows estimate the effects on those of three age groups without disabilities. The next three rows estimate the differential effects for those with disabilities. The total effect on those with disabilities is the sum of the two coefficients (SSA1983 plus SSA1983 x disability).

Table 8.

Effects of SSA1983 on SSDI applications by age cohorts and disability, HRS merged with Form-831 data.

	(1)	(2)		(3)	(4)	(5)	(6)
Disability measure:	Work-Limiting Disability				ADL Limitation
Sex:	All	Female	Male		All	Female	Male
SSA1983 x
55 ≤ Age < 62	0.00157 (0.00129)	0.00110 (0.00178)		0.00239 (0.00168)	−0.00108 (0.00134)	−0.00187 (0.00170)	0.000110 (0.00188)
62 ≤ Age < 65	0.00000232 (0.00103)	0.000431 (0.00160)		0.000644 (0.00136)	0.00136 (0.00151)	0.00103 (0.000963)	0.00298 (0.00290)
65 ≤ Age < FRA	−0.000530 (0.00243)	0.00318 (0.00197)		−0.00445 (0.00425)	0.000389 (0.00251)	0.00412*** (0.00130)	−0.00267 (0.00467)
SSA1983 x disability x …
55 ≤ Age < 62	0.0147*** (0.00509)	0.0115* (0.00680)		0.0200** (0.00909)	0.0137*** (0.00336)	0.0107** (0.00413)	0.0152** (0.00598)
62 ≤ Age < 65	0.0106* (0.00566)	0.00910 (0.00675)		0.0136 (0.0108)	0.00505 (0.00349)	0.00374 (0.00330)	0.00330 (0.00756)
65 ≤ Age < FRA	−0.00168 (0.00264)	−0.00792*** (0.00282)		0.0114** (0.00561)	0.000379 (0.00250)	−0.00297 (0.00255)	0.00252 (0.00502)
N	41,723	23,329		18,394	41,723	23,329	18,394

NOTE: See the notes to Table 6. Estimates come from Equation [2]. Standard errors, in parentheses, are clustered at the state level. We weight our regressions using HRS person-level weights. Significantly different from zero at 1-percent level (***), 5-percent level (**), or 10-percent level (*). The mean rate for SSDI application rate is 0.00975.

SOURCE: See the source for Table 6.

Table 8 shows that most of the spillovers are concentrated in those with disabilities. For the work-limited measure of disability (columns (1) to (3), first three rows), there are no statistically significant spillovers onto SSDI applications for those without a work-limiting disability. For those without an ADL limitation (columns (4) to (6), first three rows), the only statistically significant result is that women from ages 65 to the FRA without an ADL limitation had an increase in SSDI applications due to SSA1983.

For those with disabilities, most of the spillovers to SSDI applications are concentrated in those ages 55 to 61. Spillovers occur for both men and women but are larger for men and slightly larger for those with a work-limiting disability compared to an ADL limitation.⁴¹ For the pooled sample that includes women and men, those ages 55 to 61 with a work-limiting disability (ADL limitation) have an increase in SSDI applications of 0.0147 (0.0137), statistically significant at the 1 percent level. These spillovers are large relative to the rate of SSDI application for individuals with disabilities in general.⁴²

For those 62 or older with a disability, generally, the spillovers to SSDI applications decline with age, especially for women. We find a few statistically significant estimates for those with a work-limiting disability only. For women and men combined with work-limiting disabilities of ages 62 to 64, we only find a marginally significant increase in spillovers – an increase of 0.016. For those with work-limiting disabilities of ages 65 to the FRA, spillovers are negative for women but positive for men.

Spillovers to SSDI receipts

Table 9 presents the results for SSDI receipt. Here there are fewer statistically significant spillovers, although most of the estimates are positive. The weak evidence of spillovers to SSDI receipt is again concentrated in those ages 55 to 61, although there is no apparent difference between those with and without a disability. The only group with both a statistically significant increase in SSDI applications and SSDI receipt is men with a work-limiting disability ages 65 to the FRA.

Table 9.

Effects of SSA1983 on SSDI receipts by age cohorts and disability, HRS merged with Form-831 data.

	(1)	(2)	(3)	(4)	(5)	(6)
Disability measure:	Work-Limiting Disability			ADL Limitation
Sex:	All	Female	Male	All	Female	Male
SSA1983 x
55 ≤ Age < 62	0.00181 (0.00121)	0.00243* (0.00139)	0.00123 (0.00183)	0.00218 (0.00134)	0.00181 (0.00194)	0.00291** (0.00142)
62 ≤ Age < 65	−0.000181 (0.000578)	−0.000513 (0.000867)	0.000443 (0.000521)	0.000184 (0.000242)	0.0000253 (0.000437)	0.000559* (0.000302)
65 ≤ Age < FRA	0.000250 (0.000487)	0.0000622 (0.000643)	0.000274 (0.000644)	0.00107** (0.000483)	0.00105 (0.000774)	0.00109 (0.000663)
SSA1983 x disability x …
55 ≤ Age < 62	0.00313 (0.00310)	0.00190 (0.00568)	0.00554 (0.00462)	0.00190 (0.00262)	0.00320 (0.00336)	−0.000167 (0.00402)
62 ≤ Age < 65	0.00106 (0.00236)	0.00437* (0.00252)	−0.00156 (0.00338)	0.00119 (0.000913)	0.00251 (0.00155)	−0.000519 (0.00208)
65 ≤ Age < FRA	−0.000642 (0.00142)	−0.00187 (0.00170)	0.00400* (0.00204)	−0.000684 (0.000741)	−0.00124 (0.00126)	0.000285 (0.00153)
N	38,260	21,592	16,668	38,260	21,592	16,668

NOTE: See the notes to Table 6. Estimates come from equation [2]. Standard errors, in parentheses, are clustered at the state level. We weight our regressions using HRS person-level weights. Significantly different from zero at 1-percent level (***), 5-percent level (**), or 10-percent level (*). The mean rate for SSDI receipt rate is 0.00563.

SOURCE: See the source for Table 6.

Overall, across Tables 9 and 10, it appears that while younger cohorts (primarily ages 55 to 61) with disabilities experienced the largest spillovers onto SSDI application, there is little evidence that this translated in a straightforward way to SSDI receipt. This is not unusual for two reasons. First, we have less statistical power to estimate effects on SSDI receipt compared to SSDI applications since only a portion of applications lead to receipts. Second, those induced to apply to SSDI due to the relative cut in Social Security retirement benefits may be doing so for economic reasons rather than for reasons related to the severity of their disability. This suggests that their SSDI applications may have been less likely to have been approved anyways.

Table 10.

Effects of SSA1983 on any employment by age cohorts and disability, HRS data.

	(1)	(2)	(3)	(4)	(5)	(6)
Disability measure:	Work-Limiting Disability			ADL Limitation
Sex:	All	Female	Male	All	Female	Male
SSA1983 x
55 ≤ Age < 62	−0.0297*** (0.00788)	−0.0220* (0.0119)	−0.0302*** (0.00955)	−0.0242* (0.0120)	−0.00929 (0.0195)	−0.0301** (0.0134)
62 ≤ Age < 65	0.00171 (0.00916)	0.0176 (0.0128)	−0.0170 (0.0138)	0.0217 (0.0173)	0.0525* (0.0286)	0.00318 (0.0241)
65 ≤ Age < FRA	0.0795* (0.0444)	0.0728 (0.0561)	0.0849 (0.0540)	0.0829 (0.0530)	0.104 (0.0668)	0.0695 (0.0609)
SSA1983 x disability x …
55 ≤ Age < 62	0.161* (0.0805)	0.140* (0.0826)	0.104 (0.210)	−0.0307 (0.0504)	−0.00425 (0.0597)	−0.0884 (0.103)
62 ≤ Age < 65	0.158* (0.0807)	0.136* (0.0804)	0.0901 (0.213)	−0.0265 (0.0598)	−0.0142 (0.0659)	−0.0902 (0.117)
65 ≤ Age < FRA	−0.0828 (0.0859)	−0.126 (0.0837)	−0.0865 (0.216)	−0.0988 (0.0602)	−0.122* (0.0672)	−0.101 (0.0998)
N	41,750	23,355	18,395	41,750	23,355	18,395

NOTE: See the notes to Table 5. Estimates come from equation [2]. Standard errors, in parentheses, are clustered at the state level. We weight our regressions using HRS person-level weights. Significantly different from zero at 1-percent level (***), 5-percent level (**), or 10-percent level (*). The mean rate for any employment is 0.611.

SOURCE: See the source notes for Table 5.

Effects on employment

Table 10 presents the results for any employment. Again, the employment effects differ significantly by age and disability status, with most of the effects again occurring for those ages 55 to 61. Those without disabilities of ages 55 to 61 faced negative employment effects, experiencing a 0.9 to 3.0 percentage point decrease in employment, statistically significant in most cases. These effects were larger and more statistically significant for men. Those without disabilities of ages 62 to 64 and 65 to the FRA generally may have experienced positive effects on employment. However, these estimates are generally very imprecise, and only two estimates are significant at the 10 percent level.

For those with work-limiting disabilities, the differential effect on employment is positive for those ages 55 to 61 and ages 62 to 64, although these estimates are imprecise. The differences in spillovers between those with and without work-limiting disabilities for these age groups are only significant at the 10 percent level for women and in the pooled sample. There is no apparent differential effect by disability status, using the ADL limitation measure of disability. Overall, this evidence suggests negative spillovers to employment for those ages 55 to 61 without a disability, and perhaps positive spillovers for those ages 55 to 64 with a work-limiting disability.

Moderating Effects of Laws on SSA1983 Spillovers

We now move to our primary analysis of the HRS data: estimating possible moderating effects of state discrimination laws on SSA1983. We first present estimates by sex and age group in Table 11 (following equation [3]), and then we again estimate our results by sex, age group, and disability status (following equation [4]).⁴³ We present these results in Table 12 (SSDI applications) and Online Appendix Tables H1 (SSDI receipts) and H2 (any employment). Given the number of interactions in these regression models, we only present the coefficient estimates that show the differential effects of SSA1983 on individuals with disabilities, the group that faced the most spillovers to SSDI application and receipt.⁴⁴ For Tables 11 and 12, we also calculate p-values that correct for the false discovery rate using the Simes procedure.⁴⁵

Table 11.

Interactive effects of state discrimination laws and SSA1983-driven spillovers by age cohorts, HRS data.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Outcome:	SSDI Application			SSDI Receipt			Any Employment
Sex:	All	Female	Male	All	Female	Male	All	Female	Male
SSA1983 x …
55 ≤ Age < 62	0.00425 (0.00306)	0.000559 (0.00314)	0.00860* (0.00497)	0.00476* (0.00255)	0.00149 0.00288	0.00842 (0.00517)	−0.0368** (0.0177)	−0.0431** (0.0190)	−0.0259 (0.0217)
62 ≤ Age < 65	0.00132 (0.00305)	0.00533 (0.00478)	−0.00362 (0.00269)	−0.000801 (0.000875)	−0.000886 (0.00112)	−0.000500 (0.00112)	−0.0142 (0.0110)	−0.000523 (0.0177)	−0.0311 (0.0211)
65 ≤ Age < FRA	−0.00348 (0.00305)	−0.00151 (0.00216)	−0.00610 (0.00498)	−0.000102 (0.000902)	−0.00176 (0.00129)	0.00162 (0.00207)	−0.0652 (0.0638)	−0.123 (0.0819)	−0.0566 (0.0928)
SSA1983 x …
Medical defn. of disability x …
55 ≤ Age < 62	−0.00304 (0.00306)	−0.00306 (0.00387)	−0.00306 (0.00433)	−0.00321 (0.00253)	−0.00374 (0.00294)	−0.00303 (0.00351)	−0.0120 (0.0213)	−0.0152 (0.0174)	−0.0130 (0.0257)
62 ≤ Age < 65	−0.00191 (0.00567)	−0.00804 (0.00880)	0.00508 (0.00451)	−0.00290* (0.00167)	−0.00636** (0.00290)	0.00113 (0.00147)	−0.00212 (0.0195)	−0.00708 (0.0324)	0.0155 (0.0369)
65 ≤ Age < FRA	0.00168 (0.00301)	−0.0000524 (0.00225)	0.00251 (0.00577)	0.000199 (0.000960)	0.00230** (0.00114)	−0.000883 (0.00182)	0.0389 (0.0780)	0.0857 (0.0990)	−0.0455 (0.178)
Larger damages (disability) x …
55 ≤ Age < 62	0.000542 (0.00294)	0.00301 (0.00296)	−0.00227 (0.00506)	−0.00122 (0.00229)	−0.000228 (0.00271)	−0.00347 (0.00518)	0.0227 (0.0190)	0.0439** (0.0187)	−0.00280 (0.0238)
62 ≤ Age < 65	0.000856 (0.00269)	0.00171 (0.00402)	−0.00108 (0.00394)	−0.000438 (0.000870)	−0.000493 (0.00179)	−0.000729 (0.00183)	0.000945 (0.0164)	−0.0218 (0.0226)	0.0205 (0.0338)
65 ≤ Age < FRA	0.00113 (0.00279)	−0.000840 (0.00218)	0.00427 (0.00579)	−0.00181* (0.000980)	−0.00159 (0.00145)	−0.00198 (0.00160)	0.144* (0.0802)	0.177* (0.0922)	0.0263 (0.129)
Larger damages (age) x …
55 ≤ Age < 62	−0.00298 (0.00300)	−0.00252 (0.00349)	−0.00379 (0.00473)	−0.00360 (0.00224)	−0.00257 (0.00268)	−0.00449 (0.00448)	−0.00820 (0.0182)	0.00431 (0.0198)	−0.0234 (0.0212)
62 ≤ Age < 65	0.0000971 (0.00359)	0.00125 (0.00500)	−0.000852 (0.00384)	0.00208 (0.00127)	0.00473*** (0.00173)	−0.00115 (0.00161)	0.0147 (0.0151)	0.00686 (0.0255)	0.0301 (0.0250)
65 ≤ Age < FRA	0.00566 (0.00536)	0.00208 (0.00238)	0.00975 (0.0100)	0.000450 (0.000829)	0.000637 (0.00119)	−0.000704 (0.00157)	0.0559 (0.0758)	0.0602 (0.0850)	0.101 (0.119)
Firm size < 10 (dis. or age) x …
55 ≤ Age < 62	0.00272 (0.00329)	0.00651* (0.00386)	−0.00181 (0.00479)	0.000968 (0.00231)	0.00480* (0.00251)	−0.00320 (0.00473)	0.00877 (0.0176)	0.00244 (0.0194)	0.0216 (0.0205)
62 ≤ Age < 65	−0.000244 (0.00328)	−0.00393 (0.00431)	0.00514 (0.00392)	−0.0000888 (0.00127)	−0.000700 (0.00159)	0.000839 (0.00162)	0.0265* (0.0145)	0.0372* (0.0213)	0.00605 (0.0252)
65 ≤ Age < FRA	0.00839 (0.0110)	0.0176 (0.0155)	−0.00929 (0.00606)	−0.00103 (0.00362)	−0.00262 (0.00412)	−0.000450 (0.00488)	0.166** (0.0752)	0.312** (0.118)	−0.111 (0.110)
N	41,723	23,329	18,394	38,260	21,592	16,668	41,750	23,355	18,395

NOTE: See the notes to Tables 5 and 6. Estimates come from equation [3]. Standard errors, in parentheses, are clustered at the state level. We weight our regressions using HRS person-level weights. Significantly different from zero at 1-percent level (***), 5-percent level (**), or 10-percent level (*). We also calculate p-values using the Simes procedure, which controls for the false discovery rate, and none of the estimates in this table are statistically significant after that correction. The mean rate for SSDI applications for the entire sample is 0.00975, the mean rate for receiving SSDI is 0.00563, and the mean rate of any employment is 0.611.

SOURCE: See the source notes to Tables 5 and 6.

Table 12.

Interactive effects on SSDI applications of state discrimination laws and SSA1983-driven spillovers by age cohorts, HRS merged with Form-831 data.

	(1)	(2)	(3)	(4)	(5)	(6)
Disability measure:	Work-Limiting Disability			ADL Limitation
Sex:	All	Female	Male	All	Female	Male
SSA1983 x disability x …
55 ≤ Age < 62	0.00987 (0.0113)	0.00382 (0.0142)	0.0192 (0.0168)	0.0132** (0.00576)	0.0120* (0.00673)	0.0165* (0.00873)
62 ≤ Age < 65	0.0127 (0.00931)	0.0104 (0.0146)	0.0120 (0.0113)	0.00472 (0.00490)	0.00311 (0.00775)	0.00454 (0.00627)
65 ≤ Age < FRA	−0.00576** (0.00254)	−0.00786** (0.00362)	−0.00459 (0.00390)	−0.00497** (0.00236)	−0.00508 (0.00362)	−0.00783** (0.00297)
SSA1983 x disability x …
Medical definition of disability x …
55 ≤ Age < 62	0.00601 (0.0119)	0.0129 (0.0216)	0.0119 (0.00970)	−0.00515 (0.0101)	−0.000629 (0.00915)	−0.00359 (0.0149)
62 ≤ Age < 65	−0.0155 (0.0153)	−0.0149 (0.0250)	−0.00305 (0.0136)	−0.0178** (0.00795)	−0.0138 (0.0119)	−0.0148 (0.0143)
65 ≤ Age < FRA	−0.00627 (0.00525)	0.00644 (0.00580)	−0.00670*** (0.00218)	−0.0133*** (0.00381)	−0.00579 (0.00370)	−0.0167* (0.00992)
Larger damages (disability) x …
55 ≤ Age < 62	0.0101 (0.0122)	0.00367 (0.0152)	0.0189 (0.0161)	0.00354 (0.00723)	0.0112 (0.00755)	−0.00279 (0.0113)
62 ≤ Age < 65	0.00238 (0.0104)	0.0101 (0.0145)	−0.0108 (0.0140)	−0.00181 (0.00526)	0.00132 (0.00628)	−0.00447 (0.0103)
65 ≤ Age < FRA	−0.0000688 (0.00398)	−0.000521 (0.00476)	0.00547* (0.00286)	−0.00532 (0.00356)	−0.00132 (0.00368)	−0.00556 (0.00750)
Larger damages (age) x …
55 ≤ Age < 62	−0.00926 (0.0104)	0.00324 (0.0122)	−0.0179 (0.0160)	−0.00590 (0.00691)	−0.00170 (0.00724)	−0.0159 (0.0125)
62 ≤ Age < 65	−0.0170 (0.0113)	−0.00847 (0.0128)	−0.0190 (0.0207)	−0.00399 (0.00767)	0.00942 (0.00760)	−0.0262* (0.0141)
65 ≤ Age < FRA	0.00154 (0.00584)	0.00847 (0.00554)	0.000828 (0.00877)	0.000206 (0.00516)	0.00297 (0.00576)	−0.00473 (0.00898)
Firm size < 10 (dis. or age) x …
55 ≤ Age < 62	0.00856 (0.0120)	0.00563 (0.0149)	0.0177 (0.0209)	0.00513 (0.00841)	−0.00338 (0.00821)	0.00966 (0.0152)
62 ≤ Age < 65	0.0136 (0.0119)	0.00651 (0.0147)	0.0345 (0.0219)	0.00797 (0.00888)	−0.00390 (0.00686)	0.0223 (0.0164)
65 ≤ Age < FRA	0.00487 (0.00691)	−0.00636 (0.00550)	0.0317** (0.0144)	0.0119* (0.00663)	0.00368 (0.00557)	0.0199 (0.0121)
N	41,723	23,329	18,394	41,723	23,329	18,394

NOTE: See the notes to Table 6. Estimates come from equation [4]. We present the coefficients for the interactions with SSA1983 * disability only. Standard errors, in parentheses, are clustered at the state level. We weight our regressions using HRS person-level weights. Significantly different from zero at 1-percent level (***), 5-percent level (**), or 10-percent level (*). The mean rate for SSDI application (receipt) rate for the entire sample is 0.00975 (0.00563). We also calculate p-values using the Simes procedure, which controls for the false discovery rate. The only statistically significant estimate that maintains significance after applying Simes is the −0.00670 coefficient in column (3) for ages 65 to the FRA in states with the medical definition of disability (significant at 10 percent level).

SOURCE: See the source notes to Table 6.

Moderating effects on SSDI applications

Table 11, columns (1) to (3), presents the moderating effects of existing state disability and age discrimination law features on SSA1983 for SSDI applications by age cohort and sex only. Here we find virtually no moderating effects of laws.

Table 12 extends Table 11 by adding interactions with disability status. We find no moderating effects across all state law features for those ages 55 to 61 or ages 62 to 64 except in two out of 48 cases⁴⁶ (and only without the Simes correction). Most of the effects we see are decreases in SSDI applications for men in some cases for the pooled sample for ages 65 to the FRA in states with the medical definition of disability. For men ages 65 to the FRA, we find that the medical definition of disability is associated with a differential 0.00670 percentage point decrease (significant at the 1 percent level, or 10 percent level after applying a Simes correction) in the SSDI application rate for those with work-limiting disabilities and a 0.0167 percentage point decrease (significant at the 10 percent level, not significant under Simes) for those with ADL limitations.

There are few statistically significant effects for features of state law other than the medical definition of disability. Only one estimate is significant for larger damages for disability discrimination – but only at the 10 percent level, and without the Simes correction. This is similar for larger damages for age discrimination. State laws applying to a firm size less than ten are associated with an increase in SSDI applications in two cases out of 18 – for those ages 65 to the FRA only – but again only when not applying the Simes correction.

Moderating effects on SSDI receipts

Table 11, columns (4) to (6), presents the moderating effects of existing state disability and age discrimination law features on SSA1983 for SSDI receipts by age cohort and sex only. Here we find virtually no moderating effects of laws on men but some effects for women (and the pooled sample). These effects vary in sign, so there is no clear trend.⁴⁷

Online Appendix Table H1 presents the moderating effects of laws for SSDI receipts. Here we find few statistically significant estimates, with eight estimates out of 72 being statistically significant at the 10 percent level with no apparent pattern. Notably, the decrease in SSDI applications that we saw in Table 12 associated with the medical definition of disability, primarily for men ages 65 to the FRA with a disability, does not seem to materialize into a corresponding decrease in SSDI receipts. This mirrors earlier results where the general spillovers to SSDI applications (Table 8) did not generally materialize into spillovers to SSDI receipts (Table 9). This also mirrors the results from the aggregated SSDI count data, where statistically significant effects on SSDI applications were more common than statistically significant effects on SSDI receipt.

Moderating effects on employment

Table 11, columns (7) to (9), present the moderating effects of existing state laws by age group, without including interactive effects with disability status. Here we find positive effects on employment for larger damages for disability and lower firm size for age or disability. We do not find any association between state laws and employment for the medical definition of disability or larger damages for age discrimination regardless of sex. We also find no statistically significant effects on men only for any legal features. For larger damages for disability, we find statistically significant increases in employment for women ages 55 to 61 and 65 to the FRA. For a firm size minimum less than 10, we find statistically significant increases in employment for those ages 62 to 64 and ages 65 to the FRA. None of these estimates are significant, however, after applying the Simes procedure.

Online Appendix Table H2 presents the moderating effects of laws by disability status for any employment. Like SSDI receipts, we find few effects – only five estimates out of 72 are statistically significant at the 10 percent level, again with no clear pattern. Point estimates and standard errors are quite large, making this analysis somewhat inconclusive, except to say that there is no evidence that the moderating effects of state disability and age discrimination laws differ by disability status. However, they do seem to differ by sex and age.

DISCUSSION

SSDI Application and Receipt

We start by discussing and contextualizing our primary results: the moderating effects of the medical definition of disability and larger damages for disability on SSDI applications (and sometimes SSDI receipt). In the aggregated SSDI count data, we find a strong and robust association between the medical definition of disability and larger damages for disability and reduced SSDI applications for women. The magnitude of the moderation effect is to the point that, in many cases, there appears to be no estimated spillover to SSDI applications in states with the medical definition of disability or larger damages for disability.

Our analysis of the HRS and Form-831 data only somewhat confirms these results. We replicate the spillovers onto SSDI applications and receipts from SSA1983 that we observed with the aggregated SSDI data, and as confirmed in prior work. However, in the HRS data matched with Form-831 data, we find few (no) moderating effects of the medical definition of disability (larger damages for disability) on SSDI application or receipt for the same age range (ages 55 to 61 or ages 62 to 64 in the HRS sample).⁴⁸ This could be due to the HRS data having insufficient statistical power compared to the SSDI count data.

However, the HRS data matched with Form-831 data does find that the medical definition of disability is associated with a decrease in SSDI applications for men ages 65 to the FRA with disabilities (and those ages 62 to 64 with ADL limitations in the pooled sample).⁴⁹ So, the two data sources do not find entirely inconsistent results as they both point in the direction of the medical definition of disability being associated with a decrease in SSDI applications.

Across both data sets, we do not find any clear associations between larger damages for age discrimination or a lower firm size for age or disability discrimination and SSDI applications or receipts.

Employment

Our results from the aggregated SSDI count data that discrimination laws were associated with reduced SSDI applications and receipts suggest that the decreased spillovers to SSDI occurred because of greater access to employment. However, our analysis of the HRS data generally does not confirm this. While we sometimes find that discrimination laws are associated with increased employment, these are not cases where there is a corresponding statistically significant decrease in SSDI applications and receipts.

There are two possible interpretations of why we do not find corresponding employment effects where we find the largest and most robust decreases in SSDI applications or receipts. First, there could be no moderating effect of laws on employment at all, and the decrease in SSDI applications and receipts is attributable to some other unclear mechanism. Second, our analysis with the HRS data is underpowered, and thus it is inconclusive as to what extent discrimination laws moderate the employment effects of SSA1983. We think the latter interpretation is more realistic, especially given that our estimated effects on employment are not precise. We hope that future research can continue to explore this, especially given the relative statistical power limitations in the HRS data.

CONCLUSION

In this paper, we study state disability and age discrimination laws that are broader or stronger than the federal Americans with Disabilities Act or the Age Discrimination in Employment Act to see how they affect Social Security Disability Insurance (SSDI) application, SSDI receipt, and employment for older people. To do this, we leverage the Social Security Amendments Act of 1983 (SSA1983) which gradually increased the full retirement age from 65 to 67 and increased the penalty to claiming Social Security benefits early. We find that this effective cut to Social Security retirement benefits increased SSDI application and receipts primarily for those ages 55 to 61, especially those with disabilities. We then quantify to what extent state disability and age discrimination laws moderated the effects of SSA1983 on SSDI applications and receipts. We apply a difference-in-differences regression methodology to SSDI application and receipt counts, HRS data, and HRS data matched with Form-831 SSDI records.

We find that some broader or stronger features of state disability discrimination laws are associated with reductions in SSDI applications and sometimes SSDI receipts. Our most striking evidence comes from the medical definition of disability and larger damages under state disability discrimination law. We find strong and robust evidence in the aggregated SSDI count data that these features of state disability discrimination law reduce spillovers onto SSDI applications and receipts. These strong and robust results are not replicated for ages 55 to 64 in the less precise HRS data matched with Form-831 disability records. However, we find evidence of similar reductions in SSDI applications associated with the medical definition of disability but for men ages 65 to the full benefits retirement age with a disability.

Regardless of the sample or data set, we rarely find effects of state age discrimination laws on SSDI application or receipt. While prior work (especially Neumark and Song 2013) establishes that these laws are often associated with improved labor market outcomes and delayed retirement, we generally do not find that this translates into impacts on SSDI application and receipt.

We find only some moderating effects of state discrimination laws on employment. We find some evidence that larger damages for disability discrimination and a lower firm size minimum for age or disability discrimination are associated with increased employment,⁵⁰ especially for those aged 65 to the full benefits retirement age, regardless of disability status. We do not find any clear associations between the medical definition of disability and larger damages for age discrimination and employment.⁵¹

Overall, our results suggest that in addition to broader or stronger features of state disability discrimination laws sometimes having positive effects on employment outcomes, they also occasionally reduce reliance on SSDI. This is important for several reasons. First, reducing reliance on SSDI reduces pressure on the Social Security Trust Fund, which is forecasted to be exhausted by 2035 (Board of Trustees 2017, 2020). Reducing pressure on this trust fund could minimize or prevent future cuts to Social Security programs. Second, SSDI applications are very costly, especially for unsuccessful applicants (who disproportionately may be applying for economic reasons, such as due to shocks like SSA1983). SSDI applications lead to significant employment and earnings losses in the short and long run (Autor et al. 2015; Khan 2018; von Wachter, Song, and Manchester 2011). So, even if discrimination laws only reduce unsuccessful SSDI applications, this is hugely beneficial on its own.⁵²