Earnings Among People With Spinal Cord Injury

Purpose 

The purpose of this study was to identify differences in conditional and unconditional work-related actual earnings after SCI attributable to biographic, injury, educational, and, of particular importance, work-related factors. To accomplish this, we developed the 2-part predictive model using biographic, injury-related, educational, and work-related factors as predictors. The 3 parts of the model include factors associated with gainful employment (ie, working, not working), conditional earnings (ie, earnings among those who are employed), and unconditional earnings (ie, earnings among all participants, allocating $0 for those who are unemployed). The knowledge of employment status is necessary to compute conditional and unconditional earnings. Because conditional earnings reflect only the actual earnings of those who are employed, they are calculated only on this subset of individuals. However, those who are unemployed are allocated $0 in terms of unconditional earnings. It is important to have both concepts because unconditional earnings are heavily influenced by the employment rate (ie, portion of people with $0 earnings), whereas conditional earnings are an indicator of quality of employment among those employed only. For instance, factors such as geographic location may be significantly related to the percentage of people employed but unrelated to conditional earnings (ie, no differences among those who are employed). Unconditional earnings would still be higher for the geographic location with the higher employment rate because fewer people would have $0 earnings.

Hypotheses 

Because this study builds on earlier work, the hypotheses reflect only analyses that are unique to the current study based on the consideration of new additional variables (ie, no hypothesis related to biographic, injury, or educational factors). The following hypotheses are related to conditional earnings because differences in unconditional earnings may reflect either probability of gainful employment or earnings level: (1) conditional earnings will be significantly related to the ability to transition quickly into the workforce, including working at injury, return to preinjury job, and return to preinjury employer (even if at a different job); (2) conditional earnings will be significantly higher with indicators of work intensity after SCI including hours a week spent working, percentage of time after SCI onset gainfully employed, number of jobs held, and years working; and (3) type of employer will be significantly related to conditional earnings, with greater earnings among those working in private industry.

Participants 

Study participants were selected from rehabilitation hospitals in the midwestern and the southeastern United States. The 3 inclusion criteria were (1) traumatic SCI, (2) at least 1 year duration of SCI, and (3) a minimum age of 18 years at the time of first participation in the study. There were 1543 respondents from an initial sample of 2010 cases (77%).⁎ All participants who were past the traditional retirement age of 65 at the time of completion of the survey were excluded from the analyses, decreasing the sample by 146 cases to a working sample of 1397. To have complete data for the analyses, we eliminated all cases with any missing data, including the parts of the equation that apply only to those who were employed. This reduced the working sample to 1296 cases. Most participants were white (78%), and 74% were men. Fifty-two percent of the participants reported cervical injuries, and 23% were ambulatory. The primary etiology of injury was a motor vehicle crash (55%) followed by falls or flying objects (14%), sporting injuries (13%), and acts of violence (12%). The average age of the participants at the time of injury onset was 27.2 years, whereas the average age at the time of the study was 42.2 years. An average of 15.1 years had passed since injury onset. The average number of years of education was 13.7. Approximately one third of the final sample was gainfully employed at the time of the study (n=430).⁎

Procedures 

After receiving approval from the institutional review board, participants were sent preliminary letters describing the study and alerting them that materials would be forthcoming. An initial packet of materials was mailed to participants approximately 4 to 6 weeks later, followed by a second set of materials to all nonrespondents and a follow-up telephone call. A third mailing was used for participants who had misplaced or discarded materials but consented to participate by telephone and requested an additional set of materials. Updated addresses were also requested from the U.S. Post Office for those who had moved recently. Participants were offered $25 compensation for their participation in the study.†

Instruments 

A single item was used to measure earnings: “What is your annual income from your salary only (if you are working)?” (item presented as it appeared in the questionnaire). Earnings were grouped into the same 8 categories as used in the Behavioral Risk Factor Surveillance System by the Centers for Disease Control and Prevention: (1) less than $10,000, (2) $10,000 to $14,999, (3) $15,000 to $19,999, (4) $20,000 to $24,999, (5) $25,000 to $34,999, (6) $35,000 to $49,999, (7) $50,000 to $74,999, or (8) $75,000 or more. The employment cost index from the Department of Labor was used to inflate all dollar values from 2003 to present values (2007).

The Life Situation Questionnaire was used to assess employment outcomes.8 It was first developed in 1973 to elicit mostly objective information on a wide range of long-term post-SCI outcomes, including recent medical history, employment and education, social activities, and self-rated adjustment and life satisfaction.10 A revised version of this measure was administered to study participants and included biographic, educational, or employment outcomes relevant to the current study.8

Preliminary Analysis 

To assess selective participation, we compared biographic and injury characteristics and 2 outcome measures (employment status, earnings) as a function of response status. Data were only available on 760 subjects who participated during the previous longitudinal follow-up. Of these, 598 (79%) also responded in the current study. The others were lost because of mortality, inability to locate, and refusal to respond. The chi-square statistic and t tests were used to compare the respondents (n=598) and nonrespondents (n=162).

Variables Selection 

Several factors were considered when selecting variables for the regression analyses. For the first part of the equation, prediction of employment status, there is a clear body of research that has identified the most prominent biographic, injury, and educational predictors. These factors were used in the preliminary study of earnings and included sex, race, age at injury, years since injury onset, injury severity, and educational level.8 We maintained all these variables, with the exception that we used chronologic age as a single indicator of aging rather than the combination of age at injury onset and years lived postinjury, in order to build as closely as possible on the earlier work. We chose chronologic age because it is a linear combination of these 2 variables (ie, chronologic age = age at injury onset + years lived postinjury) and is compatible with the selection of additional predictor variables for the second part of the equation (ie, those that apply only to those employed), which are of much greater interest and have more policy relevance. Specifically, we could not have included years lived since injury and percentage of time since injury gainfully employed due to colinearity, and the latter variable is of much greater interest conceptually. Race was dichotomized as white versus minority in the current analysis, whereas Krause and Terza8 grouped all nonblacks together. We considered eliminating all nonblack minorities for clarity of analyses but chose not to in order to maintain the maximal number of participants. Three categories were used for both age (<35, 35–49, ≥50) and years of education (<12, 13–15, ≥16).‡ Two variables were used to measure injury severity/injury level (cervical, noncervical) and ambulatory status (ambulatory, nonambulatory). We chose these categories over the more traditional category breakdown according to American Spinal Injury Association grade and injury level2 to maximize the power of the analyses given the sample size, consistency with the previous manuscript on earnings, and ease and accuracy of measurement when using self-report. We also added 2 work-related variables as predictors of employment status in the analysis of the first part of the equation (ie, employment status) including (1) return to the preinjury job and (2) return to a different job within the same company. Last, because participants come from 2 different geographic regions, we added this variable into the equation (Midwest, Southeast).

We also measured more diverse vocational variables for the second part of the equation of conditional earnings, as the sample size was sufficient for these analyses. These variables included (1) total number of years working since SCI onset, (2) hours a week spent working, (3) number of jobs since SCI onset, (4) portion of time since SCI onset gainfully employed,§ and (5) type of employer (government, private, self employed/family business, other). These are in addition to the aforementioned predictors from the first part of the equation, including the 3 employment-related variables (work status at injury, return to the preinjury job, return to a different job within the same company).

Data Analysis 

We refer to biographic and injury-related characteristics and employment status at the time of injury as disparities variables and call the earnings differential for which each is accountable its attributable difference. The disparities variables are those over which policy makers have little or no direct control. We also examined several variables whose effect on earnings may be subject to policy influence including education and work factors. We refer to these as policy variables and to their potential impact on earnings as the policy effects. We tested our hypotheses within a regression framework that is designed to control for confounders.

In formulating the regression model, we took account of 2 unique features of the survey sample. First, the post-SCI earnings regression model should take explicit account of its 2 natural components: (1) whether the person has found post-SCI employment and (2) the amount he/she earns if employed. With this in mind, we implemented a variant of the 2-part model of Duan et al,11 the same model used by Krause and Terza.8 The 2-part formulation allowed us to measure separate and distinct attributable differences and policy effects for each variable with regard to (1) the probability of post-SCI employment (first part of the 2-part model), (2) post-SCI earnings conditional on being employed (second part of the 2-part model), and (3) post-SCI expected earnings (overall unconditional earnings).⁎

Employment Status 

Results from the first part of the model (table 4) indicate that nearly all factors selected were significantly related to employment status. A statistically significant attributable difference in the probability of employment was found for the following disparities variables: race, sex, age, and injury severity. It is noteworthy that those over the age of 50 were significantly less likely to be employed than those under the age of 35, although there were no significant differences between those in the 35- to 49-age group and those under 35. Among the policy variables, education and returning to work postinjury manifested significant policy effects vis-à-vis the probability of being currently employed. Those with 16 or more years of education had the highest probability of employment, although those who reported 13 to 15 years of education also had a significantly higher probability of employment than those with 12 or less years of education. Those who returned to their preinjury job or to a different job with the same company were also more likely to be employed at the time of follow-up, although this is expected because they, by definition, returned to work postinjury. Overall, those with the highest probability of being employed were white, male, had noncervical injuries, were ambulatory, were under the age of 35, had completed 16 or more years of education, had at least 1 job since injury, and were identified from the midwestern United States rather than the southeastern United States.

Table 4. Attributable Difference and Policy Effects for Probability of Employment (Part 1 of 2-Part Model)

	Disparities Variable	Attributable Difference	t	P
	Race	.13	4.73	<.01
	Age at the study (35–49y)	−.03	−1.07	.29
	Age (≥50y)	−.11	−4.40	<.01
	Sex	.08	3.06	<.01
	Severity (=1, cervical)	−.06	−2.82	<.01
	Ambulatory status	.13	4.50	<.01

	Policy Variable	Policy Effect	t	P
	Some education beyond high school (13–15y)	.16	5.59	<.01
	College	.21	6.16	<.01
	Return to same job	.27	6.07	<.01
	Return to different job, same company	.22	4.14	<.01
	State of residence (=1 if Southwest, =0 if Midwest)	−.19	−5.79	<.01

Conditional Earnings 

The second part of the model (table 5) identifies attributable difference and policy effects for conditional earnings (pertaining only to the employed) for all disparities and policy variables involved in the first part of the model. In addition, the second part of the model includes several policy variables that apply only to those who were gainfully employed (eg, type of employer).

Table 5. Attributable Difference and Policy Effects for Probability of Conditional Earnings (Part 2 of 2-Part Model)

	Disparities Variable	Attributable Differences	t	P
	Race	13,508	2.88	<.01
	Age at the study (35–49y)	6,087	1.47	.14
	Age (≥50y)	−7,413	−1.70	.09
	Sex	11,317	3.17	<.01
	Severity (=1, cervical)	−5,780	−1.71	.09
	Ambulatory status	4,170	0.98	.33

	Policy Variable	Policy Effects	t	P
	Some education beyond high school (13–15y)	6,725	1.51	.13
	College	21,751	5.41	<.01
	Return to same job	6,257	1.06	.29
	Return to different job, same company	−1,898	−0.31	.76
	State of residence (=1 if Georgia, =0 if Minnesota)	5,755	1.43	.15
	No. of jobs since SCI⁎	1,588	1.68	.09
	Total years worked since SCI†	1,027	3.16	<.01
	Percentage time since SCI working‡	1,132	2.21	.03
	Employed by the government	24,106	3.09	<.01
	Employed by private company	15,061	3.13	<.01
	Employed by other	16,514	1.86	.06

⁎ Change from 3 to 2. † Change from 13 to 14. ‡ Change from 70% to 75%.

In terms of biographic variables, only sex and race were significantly associated with attributable differences in conditional earnings because men made an estimated $11,317 more than women and whites made an estimated $13,508 more than nonwhites (all figures have been converted to 2007 dollars). Neither age nor the 2 indicators of severity of SCI (level of injury, ambulatory status) were significantly related to conditional earnings.

Several of the policy variables had significant impacts on conditional earnings. Although there were no systematic differences between those with 13 to 15 years of education and those with 12 or less years of education, the completion of 16 or more years of education was associated with $21,751 greater conditional earnings. Compared with those who reported being either self-employed or working in the family business, the earnings of those who work for the government were $24,106 higher. Similarly, those who work in private industry, either for profit or nonprofit, earn $15,061 more than those self-employed or working in the family business. The number of years having worked since injury and the portion of time spent working postinjury were also significantly associated with higher earnings. An additional year worked postinjury is associated with a $1027 increase in earnings, and a 5% increment in the percentage of time worked since injury would account for $1132 of additional earnings.†

Several variables were not significantly related to conditional earnings including return to preinjury employer (same job or different job), total number of jobs, geographic location, and employed in a company other than government, private, or self-employed/family business.

Unconditional Earnings 

In part 3 (table 6), substantially more variables were significantly related to unconditional earnings, as would be expected given that the combined 2-part model incorporates both employment status and conditional earnings among those employed. The dollar amounts attributable to significant disparities variables were generally less than for conditional earnings. Significantly greater unconditional earnings were observed for men ($6105), whites ($8383), and ambulatory participants ($6311); lower conditional earnings were related to age of greater than 50 years (–$6293) and with having a cervical injury (–$4218). Among the biographic and injury related variables, only age between 35 and 49 was nonsignificant.

Table 6. Attributable Difference and Policy Effects for Probability of Unconditional Earnings (Parts 1 and 2 of 2-Part Model Combined)

	Disparities Variable	Attributable Differences	t	P
	Race	8,383	4.20	<.01
	Age at the study (35–49y)	861	0.40	.69
	Age (≥50y)	−6,293	−3.08	<.01
	Sex	6,105	3.31	<.01
	Severity	−4,218	−2.29	.02
	Ambulatory status	6,311	2.57	.01

	Policy Variable	Policy Effects	t	P
	Some education beyond high school (13–15y)	7,856	3.62	<.01
	College	15,432	5.18	<.01
	Return to same job	12,425	3.15	<.01
	Return to different job, same company	6,890	1.70	.09
	State of residence (=1 if Georgia, =0 if Minnesota)	−4,833	−2.00	.05
	No. of jobs since SCI⁎	495	2.90	<.01
	Total years worked since SCI†	325	5.33	<.01
	Percentage time since SCI working‡	355	3.76	<.01
	Employed by the government	7,562	5.23	<.01
	Employed by private company	4,666	5.34	<.01
	Employed by other	5,150	3.21	<.01

⁎ Change from 3 to 2. † Change from 13 to 14. ‡ Change from 70% to 75%.

Among the policy variables, return to the same company (but not the same job) was not significantly related to unconditional earnings. Participants with 13 to 15 years of education and those with a minimum of 16 years of education reported higher unconditional earnings than participants who had completed 12 years or less education ($7856, $15,432, respectively). Although returning to a different job within the same company was not significantly related to unconditional earnings, returning to the same job after SCI onset was associated with $12,425 greater unconditional earnings. Higher unconditional earnings were also obtained among those working for the government ($7562), in the private sector ($4666), or for other employment situations ($5150) when compared with those who are either self-employed or in the family business. The number of postinjury jobs, total years employed since injury, and the percentage of time gainfully employed since injury were all associated with greater unconditional earnings. Each additional job postinjury was associated with an increment of $495 in earnings, an extra year worked was associated with an increment of $325, and a 5% increase in the proportion of years spent on the job since injury was associated with an incremental increase of $355 in earnings.

Study Implications 

There is perhaps no greater challenge in rehabilitation than eliminating disparities in outcomes between people with SCI. Our results suggest that women and minorities experience poor vocational outcomes that go beyond return to work and extend to differential earnings among those who are working. These differences cannot be attributed to variations in education, likelihood of return to a preinjury employer, type of employer, or intensity of employment (eg, hours spent working a week), each of which is frequently invoked to explain inequities in outcomes. We simply cannot determine from our research why these disparities occur, other than they do not appear to occur as a result of educational and vocational factors under study and likely relate to factors underlying similar trends in the general population.

Several variables in the current study have direct policy implications. First, because education has consistently been linked with a greater probability of return to work and has been linked to higher earnings, education must continue to be a primary focus of policies that promote opportunities for people with SCI and other disabling conditions. Although having 13 to 15 years of education was associated with a greater probability of return to work, it was not associated with higher earnings, suggesting the need for a minimum of 16 years of education to maximize earnings potential. Because the intensity of employment factors was significantly related with earnings, except for the number of jobs, policy should also be directed not only at returning individuals to work but also at doing so expeditiously and focusing on job retention. Stated simply, the longer one stays in the workforce, the greater their earnings will be. Last, it is clear that opportunities for work in the government are important to people with SCI. Employment policies and government agencies should be routinely reviewed and enhanced whenever possible to create better opportunities.

Study Limitations 

There are several limitations of the study. First, all data are self-report. We do not expect substantial differences between objective records and self-report, but we anticipate some deviations related to retrospective recall. Second, data were obtained from a cross-sectional design, so we cannot assume any observed relationships between earnings and aging variables imply that an individual's earnings would change as he/she ages or as time postinjury advances (this would require a longitudinal design). Third, the earnings data were categoric in nature; it would have been preferable to have continuous data. However, categories are frequently used for reporting to minimize error related to self-report, and this is a common practice in the establishment of large government databases, such as the Behavioral Risk Factor Surveillance System after which the categories are modeled. Our data analytic techniques were based on statistical methods specially designed to handle categoric data. Fourth, although there was a substantial number of black participants, the representation among other minority groups was limited, and they were grouped with blacks within the category of minority participants for data analytic purposes. We decided to do this rather than eliminate them altogether because their participation contributes to our overall understanding of earnings. We do not know how earnings are impacted among these other minority groups, although previous research7 clearly suggests that Hispanic and other minority participants experience disparities in outcomes compared with whites. Future research would be substantially enhanced by the inclusion of more diverse racial-ethnic groups. Fifth, as with all studies, we were limited in terms of our inclusion and definitions of predictor and outcome variables. For instance, we did not have data on preinjury earnings, which would have been a powerful predictor (data on preinjury earnings would have allowed us to estimate earnings losses caused by the injury). Furthermore, because a portion of people with SCI also have dual diagnoses of traumatic brain injury,13 it is possible that brain injury contributed to the findings (because we do not have data on this variable, we could not include it as a predictor). Similarly, our outcome measure was restricted to gainful earnings in the past 12 months, exclusive of analysis of benefits such as health insurance. Sixth, there was some evidence of selective response by age at injury and previous employment status, and this may have some unknown effect on the findings. Last, the conditional earnings analysis applies only to actual earnings. It would be erroneous to assume that the latent (potential) earnings of a currently unemployed individual with given observable characteristics would be equal to the actual earnings of an employed individual with the same characteristics.14 For instance, we cannot assume that if we are successful in raising the employment rate among those with cervical injuries that their earnings would be the same as the actual earnings of those employed in the current analysis because there are likely variables not included in the study that systematically differentiate between these employed and unemployed groups.

Future Research 

The most significant need for future research is to address policy relevant factors and earnings. The next step would be to look at job accommodations and other environmental factors. These are parameters that have long been emphasized as having great importance to the lives of people with disabling conditions, yet their importance has not been evaluated in relation to earnings.

Another consideration is the relationship between the type of employer and earnings. Why do those who work in government jobs fare better in terms of earnings? One might expect that those in private industry would do better, yet they did not in the current study. What are the characteristics of the jobs that have resulted in the highest earnings?

A highly significant research question relates to disparities in earnings as a function of sex and race. Why do they occur? This is one of the puzzling findings from our research because we statistically controlled for educational and vocational factors.