I. Introduction

Although the U.S. Census Bureau defines disability as “a long-lasting physical, mental, or emotional condition,”¹ it also states that collecting data about individuals with disabilities is challenging because disability is a “complex, multi-dimensional concept” that is “difficult” to measure in a “survey format.”² This paper examines disability as measured and reported in three Census tsurveys: the American Community Survey (ACS), the Current Population Survey (CPS), and the Survey on Income Program Participation (SIPP). We focus on whether person-specific disability, as measured by the Census, is a permanent or a transitory condition.

We begin with the evolution of the Census’ recent measurement of disability. Survey data results regarding the incidence of Census-measured disability are shown both cross-sectionally and longitudinally. It is found from longitudinal CPS and SIPP responses that Census-measured disability is largely transitory. That finding empirically confirms the longstanding theoretical objections to the use of Census cross-section disability data along with a permanent disability assumption in worklife expectancy models; hence, any worklife expectancy model which assumes that Census disability measures are permanent conditions is misspecified and empirically invalid. Instead of using the permanent disability assumption, we use actual longitudinal disability transition probabilities and life table analysis to quantify the lifetime duration of disability as measured by the Census in total life years and working life years. We show that the realistic feature of disability transition dramatically lowers the effect of disability on worklife expectancy.

II. The Six Census Disability Questions

On March 13, 1998, President Clinton issued Executive Order 13078³ which established the National Task Force on the Employment of Adults with Disabilities. Its purpose, per §1(c) was “to create a coordinated and aggressive national policy to bring adults with disabilities into gainful employment at a rate that is as close as possible to that of the general adult population” and “to support the goals articulated in the findings and purpose section of the Americans with Disabilities Act of 1990.” Regarding federally collected disability statistics, the Executive Order stated at §2(f):

The Bureau of Labor Statistics of the Department of Labor and the Census Bureau of the Department of Commerce, in cooperation with the Departments of Education and Health and Human Services, the National Council on Disability, and the President's Committee on the Employment of People with Disabilities shall design and implement a statistically reliable and accurate method to measure the employment rate of adults with disabilities as soon as possible, but no later than the date of termination of the Task Force (September 2002). Data derived from this methodology shall be published on as frequent a basis as possible.

To accomplish the Order, the task force established the Employment Rate Measurement Methodology Work Group (ERMM) which was represented by 17 federal agencies.

From 2000 through 2007, the questionnaires used in the decennial “Census 2000” and the ACS had included six disability topics, covered in three questions, each with two subparts. The first question asked about persons five years and older was (a) whether they had any “long-lasting conditions” such as blindness, deafness, or severe vision or hearing impairment, and then (b) whether they had a condition that substantially limits their basic physical activities. The second question asked about persons five years and older was whether they had any long-lasting condition that gave them difficulty in (a) learning, remembering, or concentrating, or (b) dressing, bathing, or getting around inside the home. The third question for persons ages 16 and older asked was whether they had any long-lasting condition giving them difficulty (a) going outside their home alone to shop or visit a doctor’s office, or (b) working at a job or business.

Amidst disability researchers’ dissatisfaction with the three disability questions, the OMB Interagency Committee for the ACS initiated an ACS Subcommittee on Disability Measurement. From January through March of 2006, tests were conducted of new and modified survey disability questions and content. A resulting report titled the 2006 ACS Content Test produced changes to the ACS disability topics resulting in six 2008 ACS disability questions (which we label as the “Census Six”).⁴ Also for 2008, the Census Six disability questions were added to topical modules of SIPP’s longitudinal waves 4, 7, and 10. Concurrently, the Bureau of Labor Statistics (BLS)-led ERMM Work Group had developed its own disability questions. The questions were tested in the National Co-Morbidity Survey (NCS) and then placed in the February 2006 CPS (McMenamin-Miller-Polivka (2006), p. 4). Despite that work, beginning in June, 2008, the BLS instead chose to incorporate the Census Six questions into the back end of the CPS.

The Census Six are the disability questions whose responses are analyzed in this paper. The wording of the Census Six disability questions, by survey, is shown in Appendix A. It is common to refer to the Census Six disability conditions as hearing, vision, cognitive, ambulatory, self-care, and independent living. The six questions have binary yes/no outcomes, so there is only one way for a respondent to have no disability (that is to answer no to each disability question). Conversely, there are 2⁶–1  =  63 ways to have one or more disability, which in this paper is labeled the “portmanteau Census Six” definition.

The publication of CPS Census Six disability micro-data began in February 2009 with the issuance of labor force data for January 2009. Beginning in February, 2010, information on the employment status of persons by portmanteau Census Six disability status was included as part of the BLS’ monthly Employment Situation news release in table A-6. Additional disability information is available in a BLS annual news release focusing on the employment status of persons with a disability.⁵

III. CPS: Not in the Labor Force Because Disabled and Unable to Work

The main purpose of the CPS is to determine the size of the U.S. labor force and its characteristics, the chief of which are the unemployment rate and various other measures of labor under-utilization. It also provides many variables relating to employment and earnings, and its data files include abundant demographic data that economists use to study labor market conditions in the U.S. population and in sub-populations. One of the long-running features of the CPS is its recording of disability as the reason why a person is not in the labor force (NILF). During the labor force participation section of the CPS, the interviewer does not specifically probe for disability but he or she is allowed to classify somebody that is NILF and “Disabled” or “Unable to work” when the respondent volunteers such information.⁶ Whenever a respondent gives “disability” or “illness” as the reason why he or she did not work or look for work during the last week, in order for the respondent to skip out of the remaining labor force questions he or she must respond that:

Disability must be so severe that it completely incapacitates the individual and prevents him/her from doing any kind of work for at least the next 6 months (not just the type of work of the last job).

Thus, a truck driver who is unable to drive a truck because of a heart condition might be able to do less strenuous work (for example, an office job as a personnel clerk). Likewise, do not assume that persons reported on Social Security Disability are completely incapacitated. They may be able to do some kind of work.

For unable to work to be used, the conditions listed under disabled must be met; that is, the person's medical condition prevents him/her from doing any kind of work, not just the type of work at his/her last job, for the next 6 months. (CPS Interviewing Manual, January 2013, page B3-3.)

After the CPS respondent completes the labor force section of the interview, all respondents are asked the Census Six questions—persons earlier in the interview responding that they are NILF due to disability are not singled out for the Census Six questions. So, in addition to finding persons with Census Six disabilities, the CPS is able to identify persons who report that they have a disability which prevents them from working for at least the next six months. Unlike the portmanteau definition which allows one to participate in the labor force while disabled, this CPS-specific disability classification permits no labor force participation due to disability lasting at least for six months following survey response. Consequently such persons cannot be employed, so that their employment and unemployment rates are not even defined. This paper also addresses the Census Six disabilities of persons stating in the CPS that they have work-preventing disabilities.

IV. Data

The U.S. Census Bureau provides public microdata access to the ACS, CPS, and SIPP.⁷ Data on the responses to the Census Six questions were added to the public ACS and CPS microdata beginning in 2009. Both the CPS and SIPP do not survey people living in institutions such as correctional, residential nursing, and mental health care facilities and those stationed in armed forces facilities. Since the ACS includes institutional populations, for comparability we deleted from the ACS microdata all persons living in institutions. Our analysis focuses on persons ages 18 and older. Using the ACS 5-year 2009-2013 public use microdata samples (PUMS), we found 11,443,524 person cross-section observations. Using the 2009 to 2014 CPS outgoing rotations,⁸ we found 1,834,646 person cross-section observations and 599,740 unique matched-person longitudinal observations.⁹ Using the 2008 SIPP wave 4 (September to December 2009), wave 7 (September to December 2010), and wave 10 (September to December 2011), we found 92,311 unique person cross-section observations with 50,753 unique wave 4/7 longitudinal matches and 48,167 unique wave 7/10 longitudinal matches. The combined unique person SIPP longitudinal matches were 58,602.¹⁰

V. Disability Prevalence (Cross-sectional Disability)

The prevalence of disability within the Census Six disability conditions is shown in Table 1. The top section of the table, “All adult persons in households,” shows the cross-sectionally estimated percentages of the age 18 & over U.S. non-institutional population of persons having the indicated disability (and possibly other disabilities) and, in the last row, as being portmanteau Census Six disabled. The first column of data, “ACS American FactFinder” are estimates taken from the Census Internet site relating to the Census’ tabulation of the entire ACS survey (not just the microdata sample) from 2009 to 2013 for the non-institutional population.¹¹ The remaining columns show the disability percentages extracted from the public use microdata by survey. The microdata tabulation of the ACS PUMS produces nearly identical estimates to the Census’ reporting based on all ACS data.

Table 1 Disability Prevalence by Census Survey, by Type of Disability

View Fullscreen

Table 2 Percentage of persons with the indicated disability, by number of reported disabilities, by Census survey Ages 18 & over in the U.S. non-inst. Population ACS data are for 2009-13; CPS data are for 2009-2014; SIPP data are for 2009-2011

View Fullscreen

Table 3 Presence of a “Census Six” disability by main activity, Current Population Survey Ages 18 and over in the 2009-2014 U.S. non-inst. population

View Fullscreen

Table 4 Disability Transitions, CPS and SIPP Persons in the U.S. non-institutional population ages 18 and over CPS data are for 2009-2014; SIPP data are for 2009-2011

View Fullscreen

Table 4 (Continued)

View Fullscreen

Table 5 Reference person disability transitions, CPS and SIPP Persons in the U.S. non-institutional population ages 18 & over CPS data are for 2009-2014; SIPP data are for 2009-2011

View Fullscreen

Table 5 (Continued)

View Fullscreen

Table 6 Main activity transitions, Current Population Survey Ages 25 to 61 in the 2009-2014 U.S. non-inst. Population Percent transitioning from an activity at age x to an activity at age x+1

View Fullscreen

Table 6 (Continued) Percent in an Activity at Age x+1 that Transited from an Activity at Age x

View Fullscreen

Table 7 Remaining lifetime years, by gender and starting disability condition Current Population Survey, 2009 to 2014; U.S. Life Tables, 2010 Any "Census Six" Disability

View Fullscreen

Table 7 (Continued)

View Fullscreen

Table 7 (Continued)

View Fullscreen

Table 7 (Continued)

View Fullscreen

Table 7 (Continued)

View Fullscreen

Table 7 (Continued)

View Fullscreen

Table 7 (Continued)

View Fullscreen

Table 8 Remaining lifetime years, by main activity and gender Current Population Survey, 2009 to 2014; U.S. Life Tables, 2010 Starting the age in the labor force

View Fullscreen

Table 8 (continued)

View Fullscreen

Table 9 Remaining Lifetime Years in or not in the Labor Force, by Gender and Disability Status Current Population Survey, 2009 to 2014; U.S. Life Tables, 2010 Beginning Age State: All Persons Any “Census Six” Disability

View Fullscreen

Table 9 (continued)

View Fullscreen

Table 9 (continued)

View Fullscreen

From Table 1, ambulatory disabilities have the highest prevalence followed by independent living, cognitive, hearing, self-care, and then vision disabilities. The ACS and SIPP estimates are closer to each other than the CPS estimates. The CPS disability rate is likely the lowest because the CPS question format does not record disability information by household person name, but it relies on the recollection of the respondent for the entire household as to whether “any” household member has a disability (see Appendix A).

From the microdata, the percentage of the U.S. population that did not answer at least one of the Census Six disability questions was 5.1% in the ACS, 3.8% in the CPS, and 10.7% in the SIPP. For data records not containing answers, the Census imputes an answer using the distribution of actual answers. To remove imputation and the reporting of household members’ disability status by another household member who is not the Census respondent, in the bottom part of Table 1 we show the percentage of survey reference persons reporting on their own non-imputed disabilities.¹² Disability prevalence increases slightly for reference persons as opposed to all persons. That result is predicted because young adults (who are more likely to be non-disabled) who live with their parents are not usually survey reference persons. Overall, disability rates in the entire microdata as opposed to the non-imputed reference person microdata are not substantively different.

From Table 2, for the persons having a Census Six disability, around half report having just one of the disabilities (50.2% in the ACS, 53.6% in the CPS, and 51.1% in the SIPP). From the ACS, 22.4% of the disabled population has two disabilities, 14.0% have three disabilities, 8.5% have four disabilities and 4.9% have five or more disabilities. The number of disabilities per disabled person are proportionally nearly the same across the three surveys.

For persons who are not in the labor force, the CPS probes to find out the respondent’s main activity. The CPS divides the NILF population into those persons who are unable to work at any job in the next six months due to a disability (“work disabled” below), who have a short-term illness, who are in school, who are taking care of house or family, who are retired, or who are doing something else/other. In Table 3, we cross tabulate the persons who are in these various NILF situations along with the employed and unemployed persons against the presence of any of the Census Six disabilities. In the upper sub-table capturing adults ages 18 and over, of all employed persons, 3.5% of them are portmanteau disabled.¹³ The NILF group with the highest frequency of portmanteau disabled is those persons unable to work for the next six months due to disability. Of those work-disabled persons, 63.8% state that they have at least one Census Six disability but 36.2% do not have a Census Six disability. We find this surprising because the six disability questions, though not meant to identify people with individual disabilities, were meant to capture all disabled persons when taken together, as our portmanteau definition does. Restricting the population to the predominant working years of ages 25 to 61 has little effect on the percentage of disabled employed persons (3.1% of persons ages 25 to 61 versus 3.5% of all adults) and those persons unable to work for the next six months due to a disability (61.3% of persons ages 25 to 61 having a Census Six disability versus 63.8% of all adults). The question remains: why are 36.2% of persons ages 18 and over too disabled to work in the next six months but are not picked up by any of the six disability questions?

Table 3 may be used to calculate additional economic quantities of interest. The labor force participation percentage for persons aged 25-61 is 34.0% for the portmanteau Census Six disabled and 83.4% of the non-portmanteau Census Six disabled. We need to keep in mind, however, that these are cross tabulations, and are not to be given a causal interpretation, such as that disability causes the lowered participation rate. While undoubtedly disability can cause an inability to work, research has pointed out that (a) some people decide not to work by claiming to be disabled, and (b) some people who begin receiving disability benefits fail to reject those benefits (and return to work) after recovering from their disabling condition.¹⁴

VI. Transitions Into and Out of Disability (Longitudinal Disability)

Disability transition can be computed by measuring individual persons’ reported disability conditions at one-year-apart intervals. In tables 4 and 5, we show the accumulation of Census Six disability transitions in the U.S. non-institutional population for persons ages 18 and over. All adult persons in the U.S. are shown in Table 4 and all household reference persons without imputed disability data are shown in Table 5. From Table 4 using CPS (SIPP) data, we see that 4.9% (7.9%) of the population without any Census Six disability at age x becomes portmanteau disabled by age x+1; and, 40.5% (36.4%) of the portmanteau Census Six disabled population at age x no longer report having any Census Six disability by age x + 1. The acquisition of Census Six disability is higher in the SIPP than in the CPS and the departure of disability is generally, and in three of the six individual cases, higher in the CPS than in the SIPP. Persons acquire ambulatory disabilities at the highest rate followed by independent living, cognitive, hearing, self-care and then vision disabilities. Vision disabilities are lost at the highest rate followed by cognitive, self-care, hearing, independent living, and then ambulatory disabilities. Table 5 shows disability transitions reported by the reference person who speak for themselves to the Census taker. We find that the annual incidence of portmanteau disability for reference persons rises slightly to 5.9% (9.0%) and the loss of portmanteau disability falls slightly to 39.6% (35.1%). By disability, reference persons’ own non-imputed responses about their disability status is negligibly different from that for their household members.

Given the significant loss in Census Six disability status age-to-age, it might be argued that the CPS and SIPP questions are inappropriately worded, creating a bias towards temporary conditions being reported. The Census Bureau attempts to minimize such bias by guiding respondents to report only long-term disabilities. For example, the CPS Interviewing Manual at page C3-38 indicates that

… the goal of these questions is to identify persons with long-term disabilities, but the questions do not specify a minimum length of time that the disability needs to last. Respondents who ask should be informed that these questions are not intended to include persons with disabilities that are expected to be temporary, or last for only a few months.

During a November 2000 statistical policy meeting,¹⁵ John McNeil, then a Census disability specialist, suggested that the disability questions in the Survey of Income and Program Participation (SIPP), long thought to be the “gold standard,” were not reliable. McNeil (2000) had presented a paper “Employment, Earnings and Disability” at the Western Economic Association meetings in which he concluded that there were “problems in using SIPP to measure changes in the employment rate of individuals with disabilities. These problems have to do with … an apparent lack of reliability for many individual measures of disability status …” McNeil presented tables which cross-classified the time 1 and time 2 responses (one year later) of individuals for various classification categories. He reported (p. 4)

that the great amount of change between time 1 statuses and time 2 statuses suggests a substantial reliability problem. For example, of the 461,000 (weighted) individuals who were classified as unable to see the words and letters in ordinary newspaper print in time 1, only 117,000 had the same classification in time 2.

Also at the later 2000 conference, Hale presented SIPP data showing continuations from time 1 to time 2 in the data for 78.6% of those with a work disability, 61.7% of those with wheelchair use, 38.3% of those with speaking conditions, 25.2% of those with seeing conditions, 20.7% of those with hearing conditions, 47.6% of those with walking conditions, and 49.1% of those with handling money conditions. Subtracting these non-transit probabilities from 1 shows very high transition probabilities out of these transition states.

McNeil and Hale have not been the only persons writing about the reliability of specific disability survey questioning when focusing on longitudinal responses in the SIPP. A February 2006 Cornell University report¹⁶ found the same prevalence of year-to-year changes in specific disability responses. The Cornell report discounts looking at specific disability questions for the impact of disability on employment, and instead refers to questions regarding “work limitations.” However, the Cornell report and others have found that work limitations come-and-go as does the impact of work limitations on employment. From the Cornell study of the 2001 SIPP,¹⁷ regarding “work limitations” without reference to employment status, of persons ages 25 to 61, 75.6% of persons with work limitations in Wave 5 of the 2001 SIPP had work limitations one year later, but 24.4% did not; of all persons ages 25 to 61 without a work limitation in Wave 5, 3.2% had a work limitation one year later.

From the 2009-2014 CPS, 83.4% of those between ages 25 to 61 who were NILF persons not working due to disability maintain the same NILF-disabled status year-to-year but 16.6% transit out of NILF disability (see Table 6, top sub-table). From the Cornell study of the 2001 SIPP, with reference to employment, of the persons ages 25 to 61 with a work limitation and employed at Wave 5, 79% ( = 22%/28% in their data) were employed one year later so that 21% transitioned out of employment as compared to 8% ( = 6.4%/82%) of the population without work limitations who transit from employed to not employed over the course of the next year. Of the persons ages 25 to 61 with a work limitation at Wave 5 and not employed, 8.4% become employed over the course of the next year and 91.6% remain not employed. From both the CPS and SIPP, whether measured by specific disability questioning or unmentioned disability that prevents work altogether, or limits work somehow, there is no empirical support for the proposition or assumption that disability or non-disability is a permanent condition. This result, in combination with earlier results showing coherence across the three surveys, strongly suggests that the disability population status as measured by the cross-sectional ACS survey, which contains the same Census Six disability questions independently asked of all respondents, would also exhibit such transitory year-to-year probabilities.

VII. Disability Life Tables

We have found that from one year to the next, respondents to Census surveys transit into and out of disability. In this section, we use the age-specific transitions into and out of disability as computed from the CPS to measure (a) the expected portion of life that the U.S. population spends Census disabled and Census non-disabled, and (b) how many extra Census disabled years are incurred by those persons starting an age as Census disabled when compared with those who start as Census non-disabled. To answer these questions, we adopt the workhorse Markov (or increment-decrement) model¹⁸ as our methodological vehicle. That model permits the analysis of time spent in any number of states over a lifetime.

To construct Census Six disability increment-decrement life tables, we divide the population into two groups: those persons with a Census Six disability (“D” for “disabled” in the notation below) and those persons without a Census Six disability (“H” for “healthy” in the notation below). The Markov model computes the probability of living persons

• staying without a disability, H, from ages x to x+1; this probability is ;

• transiting from H at age x to being disabled, D, at age x+1; this probability is ;

• transiting from D at age x to H at age x+1; this probability is ; or,

• staying D at ages x and x+1; this probability is .

The sum of the probabilities of staying/moving from a beginning living state at age x to living at age x+1 is equal to one. For example +  =  1 and +  =  1.¹⁹ Recall from Table 4, the matching proportions in the CPS U.S. population for any Census Six disability were 94.9% + 5.1%  =  100% and 41.0% + 59.0%  =  100%, respectively.

Using the increment-decrement model, transition probabilities computed from the CPS data,²⁰ and a U.S. Life Table,²¹ we can work through the usual Markov increment-decrement recursions, starting with persons either without a specific disability or specifically disabled, at each beginning age in the U.S. population. By Census Six disability and various ages, we show disability life tables in Table 7. In the first sub-table, for the persons beginning age 20 without any Census Six disability, males (females) spend 5.74 (6.41) remaining life expectancy years with a disability and 51.35 (55.33) years without a disability. For the persons beginning age 20 with a Census Six disability, males (females) spend 7.12 (7.68) remaining life expectancy years with a disability and 49.98 (54.05) years without a disability. The “penalty years” or additional years lived in the disabled state for persons starting at age 20 with a Census Six disability are therefore 1.38 for males and 1.27 for females. The rank order of the years spent in the specific disability when starting non-disabled generally matches the rank order of disability acquisition: ambulatory disabilities have the longest duration followed by independent living, hearing, cognitive, self-care and then vision disabilities. However the largest “penalty years” do not follow this pattern; independent living slightly tops ambulatory followed by self-care, cognitive, hearing, and then vision disabilities. In the same manner, the additional tables perform the same calculations for the individual six disability definitions. The finding is that the penalty years in the disabled state is always under two years (and sometimes under one year), regardless of the Census Six disability chosen.

While we refine these life table calculations so as to incorporate the generally accepted definition of worklife expectancy in the next two sections, we point out here an important application of Table 7. Consider two 30-year-old males, one without an ambulatory disability and one with an ambulatory disability (e.g. a plaintiff in a personal injury lawsuit before and after an accident). Applying the life table model, Table 7 indicates that a person will spend only 1.32 more years in the ambulatory disabled state after the accident. Any economic statistic measured over this small period will therefore be small. If one measures the probability of being employed without disability as 80% and with that disability as 30%, then the difference in the number of years employed is years.

VIII. Disability Worklife Tables

We have empirically found that a person starting an age disabled does not mean that the person will always be disabled throughout his or her life expectancy, or anything close to it. The next question to address is the effect of disability on worklife expectancy. Again, the Markov model is well suited to measure that effect.

IX. Working Life of the Census Six Disabled

Historically before the Census Six questions existed, the SIPP showed that disability as measured in a Census survey format was a transitory event. The problem with the SIPP disability data in constructing full worklife tables has been its inadequate sample size to permit the estimation of an increment-decrement worklife table delineated by both labor force participation and disability. Since 2009, the Census and BLS have incorporated the Census Six disabilities into the CPS, which produces sufficient sample data to allow the computation of a labor force/disability increment-decrement life table as in Table 9. We have shown that transitions out of disability as measured by the Census disability definitions are frequent. In the forensic economic literature, several articles have appeared criticizing worklife models which attempt to split labor force participation by disability state at beginning worklife table age x without accounting for the transitions in-or-out of disability at ages x + n.²⁴ The worklife models presented by Gamboa and Gibson (2006, 2010, 2015) and by Gamboa in previous years as far back as 1987 all assume that the disability transition probabilities age-to-age are fixed at 0%—the disability and non-disability statuses are unrealistically assumed (i.e., forced) to be permanent. Since the probability of being employed in the disabled state is lower than in the non-disabled state, a large number of years of a disability worklife penalty, or lowering due to an initial disability status, is claimed by those static models. When population transitions are accounted for, as we have seen, the effect of Census Six disability on worklife expectancy is low.

Using a Markov worklife expectancy model, from Table 9, consider a formerly non-disabled male who, on his 35^th birthday, as a result of an accident, becomes Census Six disabled. From the second and third sub tables of Table 9, he will experience 25.52 – 24.03  =  1.49 fewer active years without a disability and will also experience 1.94 – 1.02  =  0.92 more years active in the disabled state. Thus his total worklife decreases by 1.49 – 0.92  =  0.57 years. Because working in a disabled state may be worth less than working in a non-disabled state, it may be misleading to perform the subtraction. The injury diminution of 1.49 years of non-disabled worklife is an overstatement of the effect of disability because the extra 0.92 disabled worklife years are worth something. Evaluating those 0.92 years at 50% of a non-disabled year would result in a 1.49 – 50%×0.92  =  1.03 year worklife reduction. This effect is minor, and is usually swamped by any pre-trial time out of the labor force. At age 55, the previous calculation would be 9.96 – 8.38  =  1.58 fewer nondisabled years and 1.54 – 0.68  =  0.86 more disabled years, leading to 1.58 – 50%×0.86  =  1.15 year reduction.

This paper, in Table 9, presents the first worklife expectancy estimates incorporating disability and its transition probabilities, the generally accepted definition of worklife, and the current generally accepted best practice methodology—the Markov model. Just because something can be computed, however, doesn’t mean that it should be computed and used. To be clear about this latter point, we did not include education level, which is an important factor in most worklife tables. Our calculations in Table 9 are averages over all educational levels. Clearly incorporating education would not change the main result that incorporating transition probabilities into and out of disability results in small average worklife reductions due to disability. We cannot advocate the quantitative use of the worklife expectancy reductions presented in the previous two paragraphs in practice for additional methodological reasons. There is a great deal of heterogeneity in the disabled population, which this analysis, and the Gamboa-Gibson methodology, ignores completely. For example, the Census ambulatory disability definition treats persons having problems walking due to foot bunions equally with quadriplegics. Meaningful disability worklife expectancy statistics for particular persons would require much more thorough evaluations of disability. As we have seen in this paper, neither the Census Six disability questions nor the CPS disabled and NILF definitions of disability provide any clear indication of the reason or severity of disability. Therefore, assuming that any specific person’s own disability conditions affect his or her worklife expectancy at the population rate of labor force participation of all persons with a disability is invalid and unreliable.²⁵

X. Conclusion

The measurement of disability is a complex matter. Over their lifetime, persons acquire disability and then either stay disabled or recover. We have investigated how persons report their transitions into and out of disability as now uniformly measured by the U.S. Census in its ACS, CPS, and SIPP population surveys. In addition to routinely finding age-to-age transitions into disability, we have found large age-to-age transitions out of disability—a result consistent with earlier findings for different Census survey disability questioning. A major implication of these findings is that during prime working years there is a small loss of worklife expectancy for those starting an age with a Census-measured disability as opposed to starting at age non-disabled. Our general conclusions are: (1) that Census survey data regarding disability conditions, if used to study the impact of disability on worklife expectancy, produces small reductions, and (2) worklife disability tables operated under the assumption of “once-disabled, always disabled,” produce large and spurious reductions in worklife expectancy. These conclusions re-enforce many other well-known objections to any model of worklife expectancy which employs (or forces) the once disabled/always disabled assumption into worklife expectancy models, most notably those produced by the Gamboa/Gibson.