Abstract
This study examines demographic correlates of the onset of work limiting health conditions in the United States using retrospective histories from the 2004 Survey of Income and Program Participation (SIPP). Because the SIPP data are based on respondent recall, comparisons are first made to information from the 2004 National Health Interview Survey (NHIS). Reports of work limiting health in the SIPP and the NHIS correspond closely and little evidence of recall bias is found in the retrospective SIPP reports.
Analysis of the SIPP data reveals that most onsets of work limiting health conditions occur for persons younger than 16 or older than 46. However, onset age varies greatly by condition. Conditions like deafness, cerebral palsy, and intellectual disabilities tend to show up early in life while others like back problems and heart conditions usually arrive much later in life. Logistic regressions relating common demographic factors including age, gender, race, Hispanic origin, marital status and educational level to the onset of health limitations show that most measures are highly predictive. However, we find that some specific conditions, like cancer, the presence of tumors, and paralysis, appear to be primarily related to aging particularly among married individuals.
Introduction
From multiple disciplinary perspectives, the timing of the onset of health problems in the U.S. population that limit work is of importance. In the U.S., more than some other industrial societies, social benefits including both Social Security Retirement and Disability Insurance benefits are contingent on prior work. In gauging population health, work limitations have been shown to be correlated with medical diagnoses of disability [1, 2]. And from the perspective of life course studies, the onset of work related health problems would be expected to be predictive of or move in tandem with other events such as divorce or reduced labor market activity [3, 4, 5, 6, 7, 8]. While other research has investigated the relationship of demographic correlates with alternative measures of health, often focusing on an individual measure or single conditions, this research provides a similar investigation of the onset of any work limiting health condition but also the comprehensive set of individual underlying conditions.
As a related matter, there are alternative information sources that might be used in this study. One possibility would be the National Health Interview Survey (NHIS) that asks individuals about the timing of onset of different work limiting health conditions.1 However, a deficit of those data is that they would only provide information as of the date of interview regarding characteristics of individuals that would not be directly related to health conditions that initially occurred much earlier. As an alternative, retrospective modules are contained in each panel of the Survey of Income and Program Participation (SIPP) that provide reports on the timing of each respondent’s marital history, educational attainment, and the onset of health problems.2 These data offer the advantage of allowing researchers to use contemporaneous measures of these time specific demographic characteristics in the analysis as well as other static measures of factors like race and ethnicity.
Since the NHIS was designed specifically for health related studies, prior to examining the demographic correlates of the onset of health limitations, reports of work limitations obtained from the NHIS are first compared to those found in the SIPP for all work limitations as well as reports of individual conditions. Also, the analysis considers whether there is evidence of recall bias in the SIPP by comparing the distribution of reports of the onset of work limiting health conditions that occurred recently relative to more distant reports by age of occurrence.
On average, reports of the age at which any work limitation began correspond closely between the SIPP and the NHIS. Additionally, calculations of the incidence of the majority of health limiting conditions using data from the SIPP correspond well with those obtained from the NHIS when the reporting categories match across the two surveys. There is also little evidence of recall bias in the SIPP data when comparing reports of recent onsets of work limitations to those that occurred long ago.
When logistic regression models are used to examine demographic correlates of the onset of any work limiting health condition using the SIPP data, most of the demographic characteristics available are predictive of onset based on standard tests of statistical significance. However, as a contrast to the prior literature that has most often examined gross measures of health or specific high incidence conditions, there are a number of work limitations that are not related to basic demographic measures other than age, particularly among married individuals: these include cancer, deafness or serious trouble hearing, paralysis, thyroid, tumor cyst or growth, and carpal tunnel syndrome. From a demographic perspective, their occurrence appears to be largely exogenous conditional on age. The arrival of these less predictable health conditions may potentially be usefully leveraged in other studies to examine subsequent behavioral changes in response to their arrival among individuals and their households.
The paper proceeds with a brief review of the relevant literature followed by a description of the data used in the analysis. The comparisons of the incidence of the onset of work limiting health conditions using the SIPP and NHIS follow along with the analysis of possible recall bias in the SIPP. Then estimates of logistic regressions using the SIPP data for the onset of any work limiting health condition are provided along with additional estimates for the 30 different health limitations reported in the SIPP. A discussion of the implications of the results concludes the analysis.3
Literature review
In this analysis, we seek to document the timing of the onset of work limiting health conditions, differential timing of onset for different demographic groups, and the relationship of contemporaneously observed demographic covariates to those events. The prior literature has looked at a diverse set of measures of health often in isolation. For the greatest part, the literature has emphasized gross measures of health and their relationship to demographic predictors. Studies that have examined specific high incidence conditions leave an unclear picture regarding a more complete set of disaggregated conditions. This paper considers a gross measure of health not often considered by other researchers but also the underlying comprehensive set of conditions that comprise it. In part, our focus here is on identifying conditions that are not as well predicted by demographic covariates.
A diverse set of studies has examined the incidence of aggregate summary measures of health or disability, often in relation to demographic or behavioral predictors. Examples of the most commonly considered measures available in social scientific and health surveys can be seen in studies of mortality rates [9, 10, 11, 12, 13, 14], difficulties performing either Activities of Daily Living (ADL) or Instrumental Activities of Daily Living (IADL) [15, 16, 17], better self-assessed physical health [8, 17], and work limiting health problems [6, 18]. However, for the greatest part, these studies have not considered the underlying composition of the outcomes studied.
Closest to this study and exemplary of this issue are the recent studies of Couch et al. [18] and Tamborini et al. [6] that examine whether divorce is predictive of subsequent reports of health limiting work problems. For men [18] their analysis shows that not controlling for selection into divorce by less healthy individuals, divorce is predictive of self-reports of work disability by men but not after controlling for selection. They also show that those who do not remarry after divorce have much higher rates of reported work disability. For women [6] divorce is found to be predictive of later self-reported work limitations regardless of whether attempts are made to control for selection of the less healthy into divorce. Marital status and gender are shown to be differentially predictive of health problems that limit work in these studies; however, neither analysis examines whether underlying specific conditions might contribute to these findings.
Work limiting health problems are conceptualized as being driven by the interaction of an underlying health condition and the social and physical environment [19]. The interest in work related limitations is often motivated by its relationship with labor market behavior; however, the existence of a health problem is antecedent and work disability is influenced by personal characteristics and social context. Some disabling conditions, such as cardio-vascular disease or malignant neoplasms, would be predictive of mortality. Thus, characteristics that have been shown to be related to underlying diseases, functioning limitations, accommodations, and mortality are relevant to this analysis.
In studies of mortality, marriage or the presence of other adults is widely reported as predictive of better health for both men and women [9, 11, 12, 13, 14, 20, 21]. These effects with respect to mortality as well as on other measures of health are often argued to arise through a social monitoring role within marriage that reduces poor health behaviors [10] particularly for men. In these studies of mortality, other demographic factors are also predictive of worse outcomes. Being male is related to shorter survival times (greater mortality) due to work related stress and poor health behaviors [9, 10]. Other demographic measures related to disadvantage in access to resources are also associated with greater mortality including being Black, having lower education, and the absence of children which helps gauge the extent of a person’s social supports [9, 11, 21]. The correlates that are found to be predictive of mortality are largely consistent with measures we are able to incorporate into this study including gender, marital status, race, Hispanic origin, education and number of children.
Other summary outcomes examined in the literature include functional limitations as measured by difficulty in performing ADL and IADL and also self-assessed health. Marriage is found to be associated with reduced difficulty in performing ADL and IADL while being Black and having low education is associated with reports of more problems [15, 16, 17, 22]. With respect to self-assessed health, divorce is found to be predictive of worse reports for men and women [8, 17]. Being Black, having a low education, and age are related to worse health outcomes as well [17]. These factors that are predictive of difficulty performing daily activities and self-assessed health overlap with those found to be predictive of survival or mortality and are consistent with those included in this study.
Other researchers have focused on specific underlying conditions, most often in isolated studies. While too numerous to catalogue, some examples include demographic studies of the incidence of accidental injuries [23], work based injuries [21, 24], mental health disorders [17, 25, 26], back problems [25, 27], musculoskeletal pain [28], and cardio-vascular disease [29]. These studies are motivated by the relatively large incidences of these conditions and also their relationship to the claiming of private health insurance benefits from workplace injuries as well as claiming of Disability Insurance benefits (due to mental health, back problems and cardio-vascular disease). While it is fair to say that the determinants that appear to be associated with other outcomes such as mortality are also generally found to be predictive in these contexts, in this study we seek to simultaneously examine a broader set of conditions. In part, we are interested in learning if there are some conditions for which common demographic controls that proxy for health behaviors and disadvantage are not as highly predictive.
Here, we temporally align available measures of demographic factors drawn from the prior literature that we expect to be correlated with the onset of work limitations with the timing of that event. We descriptively examine empirical patterns of the onset of any work limiting health condition and also provide a more detailed perspective by examining the 30 separate conditions respondents to the SIPP can report in the analysis. We also provide multivariate estimates of the relationship between the onset of work limiting health conditions and our demographic controls making use of logistic regression. The data allow us to control for the most common demographic factors reported as predictive of health status in prior studies including measures of age, gender, marital status, education, race, Hispanic origin, educational attainment and number of children.
Data
Comparison between the SIPP and NHIS data
This paper examines the evolution of individuals’ health conditions by age and other demographic characteristics over time. Two sources of survey data are available for use in the research: the Survey of Income and Program Participation (SIPP) and the National Health Interview Survey (NHIS). The SIPP is a series of national panels, providing comprehensive information about individuals’ and households’ income and participation in government transfer programs in the United States. The NHIS is a cross-sectional survey that provides information on individuals’ health status, effects of illness and disability and health care utilization in the United States since 1957.
The SIPP data appear to be more suitable for the purpose of this research, since it contains retrospective modules in the second wave regarding the timing of the onset of work related health limitations and also similar retrospective surveys of the timing of educational attainment and changes in relationship status in addition to information on other time invariant covariates. This provides the opportunity to relate reports of some variables found to be important in the prior literature such as marriage, divorce, and changes in educational attainment contemporaneously to reports of the onset of health problems that limit work. The NHIS also provides recall data on the onset of health related work limitations but not for any of the other demographic correlates we seek to make use of in this analysis; however, the NHIS is much more commonly used in health related research. Thus, we begin this analysis with a comparison of the retrospective reports of health related work limitations across the two data sources.
The duration of each individual SIPP panel varies from 2.5 to 4 years. Within the panel, there are sequenced interviews over 4-month recall periods, called waves. Here, we make use of data from the SIPP 2004 Panel and correspondingly, make comparisons to 2004 NHIS data.
In Wave 2 of the SIPP 2004 Panel (from June 2004 to September 2004), “work disability history” information is collected in a topical module. The questionnaire lists 30 health conditions and asks individuals to mark every applicable condition that leads to their work limitation and to mark the main condition among those. It also asks the year when the person’s work limitation began. Combining these reports of the timing of work limitations with individuals’ ages, it is possible to observe the age at which individuals report their health related work limitation began, i.e. the age of onset for each health condition, and the main health condition reported as a work limitation for each age group. However, some ambiguity exists since the survey asks the beginning year of work limitation but does not specify the beginning year of each individual health condition a person can identify as being associated with their work limitation. Thus, we will make use of the reported year of onset in two ways. First, we will assign the year of onset only to the main condition but then we will alternatively assign the year of onset to all reported conditions. We will compare the measures of health related work limitations in the SIPP coding their timing in these two different ways to the NHIS data to see which corresponds more closely.
The timing of the onset of health problems in the NHIS is coded specifically for each individual condition. The 2004 NHIS lists 36 health conditions and asks specifically “What conditions or health problems cause limitations”. For each individual condition or health problem participants in the survey can specify different onset dates. Again, using individuals’ self-reported ages, we are able to determine the age of onset for each work related health limit. Since the NHIS reports provide more specific dating information for the onset of health conditions than is available in the SIPP, this also provides further motivation for a comparison between the two data sources.
SIPP 2004 Wave 2, mark all conditions that cause work limitation (%)
SIPP 2004 Wave 2, mark all conditions that cause work limitation (%)
Note: All persons 18 to 67 years old. All statistics are weighted proportions except the sample numbers, proportions are shown in percentage (%), observations
SIPP 2004 Wave 2, mark the main condition that causes work limitation (%)
Note: All persons 18 to 67 years old. All statistics are weighted proportions except the sample numbers, proportions are shown in percentage (%), sample size
NHIS 2004, mark all conditions that cause work limitation (%)
Note: All persons 18 to 67 years old. All statistics are weighted proportions except the sample numbers, proportions are shown in percentage (%), sample size
The comparisons we make between the SIPP and NHIS are based on tabulations of the age of onset of specific conditions responsible for individuals’ self-reported work limitations in the population. The set of health conditions that the SIPP and NHIS take into consideration are not exactly the same. In the SIPP, there are 30 listed conditions but 36 in the NHIS. There are nineteen overlapping conditions that can be found in either survey: alcohol or drug problem or disorder, arthritis or rheumatism, back or spine problems, blindness or vision problems, broken bone or fracture, cancer, cerebral palsy, deafness or serious trouble hearing, diabetes, head or spinal cord injury, heart trouble, high blood pressure, lung or respiratory trouble, mental or emotional conditions, intellectual disability, missing limbs or foot or hand or finger, stomach trouble, stroke, and tumor cyst or growth. However, there are 10 specific conditions in the SIPP (except for “other”) that do not correspond directly with the NHIS. Similarly, there are 15 conditions (except for “other impair or problem (1)” and “other impair or problem (2)”) listed in the NHIS that do not correspond to the detailed conditions listed in the SIPP. Here, we make comparisons across conditions the two surveys have in common.
The SIPP questionnaire for the work limitation questions includes all persons 16 to 67 years old with a health condition that limits the kind or amount of work they can do, while the NHIS includes all persons 18 years of age and older who have at least one limitation. So the possible onset ages of any health condition are from 0 to 67 years old for SIPP; while there is no upper limit for NHIS.
To make the tabulations across surveys comparable, we take the 19 health conditions that can be found in both the SIPP and NHIS and divide individuals who are 18 to 67 years old and have work a limitation into 11 onset age groups: younger than 16, 16–20, 21–25, 26–30, 31–35, 36–40, 41–45, 46–50, 51–55, 56–60, and 61–67. We apply relevant survey weights to both the SIPP and NHIS data and tabulate the proportions of people in the total population 18 to 67 years old who have those 19 health conditions based on these 11 onset age groups; the weighted statistics are shown in Tables 1–3.
Tables 1 and 2 are based on the Wave 2 sample of the SIPP 2004 Panel. Table 1 is based on responses to the question “Which of these conditions cause your work limitation, mark all that applies”, so each person reporting a work limitation may have more than one health problem. Table 2 is based on the question “Which condition is the main cause for work limitation”, so each person with work limitation is only assigned one main condition. Table 3 is constructed from the NHIS 2004 survey. Since it asks individuals to report all conditions responsible for their work limitation, each individual may have more than one health problem. In short, Tables 1 and 3 are about the arrival of all conditions in the relevant population and are from the SIPP and NHIS respectively; Table 2 is about the main condition and is based on the SIPP.
As can be seen in all three tables, people suffer from work limitations most often because of back or spine problems, arthritis or rheumatism, mental or emotional conditions, heart trouble, diabetes and high blood pressure. Some health problems, such as cerebral palsy, deafness, and intellectual disabilities, are usually found to first occur among the youngest group (
SIPP 2004 Wave 2, mark all conditions that cause work limitation within 5 years (%)
Note: All statistics are weighted proportions except the sample numbers, proportions are shown in percentage (%), observations
SIPP 2004 Wave 2, mark all conditions that cause work limitation (%)
Note: All statistics are weighted except the sample numbers, proportions are shown in percentage (%), observations
We conduct t-tests across the marginal distributions of the tables, examining incidence by age and alternatively by condition. This allows us to compare whether the SIPP and NHIS provide the same average onset age for the examined health conditions in combination, and whether they provide the same proportion of individuals experiencing the individual tabulated health conditions. We make use of a 0.05 level of significance in conducting the tests.
The asterisks in the margins of Table 1 show whether the null hypothesis that the marginal probabilities from the SIPP estimates and corresponding NHIS estimates are equal is rejected. We reject the null hypothesis that the mean difference equals zero for three out of nineteen health conditions: alcohol or drug problem, broken bone or fracture, and head or spinal cord injury. We are unable to reject the null hypothesis of equality in the probability of experiencing one of the tabulated health limitations for any of the onset age groups. Thus, of the 30 tests conducted, only three were rejected. We conclude there is a close correspondence between the marginal SIPP tabulations in Table 1 and the NHIS reports in Table 3 with respect to the age of onset of any health problem that limits work across the categories considered and also a reasonably close correspondence in the reported incidence of specific health limitations. We would note that the three categories where reports of conditions appear to differ across the SIPP and NHIS are not among those found in our multi-variate analysis to not be closely related to the demographic covariates examined.
On the other hand, the asterisks in Table 2 show that statistically significant differences exist between those SIPP tabulations and comparable results in Table 3 for the NHIS data for the occurrence of nine out of nineteen specific health conditions and for one out of eleven onset age groups. These ten statistically significant differences between the marginal probabilities of events by age or condition are many more than are found in Table 1. Thus, based on these tests of statistical significance across different uses of the data in the SIPP relative to reports in the NHIS (and related t-statistics), the data based on the occurrence of conditions in the population from the SIPP in Table 1 most closely resemble tabulated data from the NHIS.
Weighted descriptive statistics by 30 health conditions, SIPP 2004 Wave 2
Note: All statistics are weighted except the sample numbers, proportions are shown in percentage (%).
We are also concerned about the possibility of recall bias in the SIPP since we will make use of retrospective reports of life events in the study. One might think that more recent reports of the onset of health limitations would be more accurate than onsets that occurred many years ago. To investigate this issue, we tabulate proportions of those whose initial onsets of different health limitations occurred within the past five years by age and condition; then we conduct a t-test comparing those recent reports of conditions by age with those for the entire sample in the SIPP regardless of how long ago they were stated to have initially occurred. If the proportions of reports by age and condition are not significantly different, this would be evidence that is consistent with a lack of recall bias.
Table 4 shows the proportions of people who report any onset of a work limiting health problem within the last five years stratified by age and type of limitation. Table 5 is the corresponding table based on all reports. Notice that now there are only 10 onset age groups from “16–20” to “61–67”. This is because the SIPP did not collect the exact beginning year if the person had a work limitation that first occurred before age 16. So for the onset age group “
Asterisks in the margins of Table 4 show results of T-tests for the null hypothesis that the differences between the cumulated table entries for the margins of occurrence by age group or occurrence of the condition relative to Table 5 are zero. The
Based on the preceding analysis, we will make use of all individuals’ reports in Wave 2 of the 2004 SIPP data of any health condition that causes work limitations in the subsequent analysis. These data bear close resemblance to those found in the NHIS and there does not appear to be any evidence that recall of distant events suffer from strong bias related to timing when compared to the full sample of the SIPP.
Some of the most commonly occurring conditions and their average rate of occurrence among those that experience a health condition independent of age as shown in Table 5 include Back or Spine Problems (3.30 percent), Arthritis or Rheumatism (1.56 percent), Heart Trouble (1.31 percent), Mental or Emotional Conditions (1.13 percent), and Diabetes (0.93 percent). These conditions are reflective of a range of functioning and mobility issues related to back and joint problems, circulatory issues and mental health.
Although disability is most often thought of as likely to first occur at later ages, some conditions also arrive at early ages at rates similar to the average across all ages. Those conditions that are shown in Table 5 to have a somewhat high rate of onset at early ages, prior to age 26, include reporting an alcohol or drug problem, a broken bone or fracture, cerebral palsy, epilepsy or seizures, head or spinal cord injuries, a learning disability, or an intellectual disability. Some of these conditions, such as intellectual or learning disabilities, may well be present from birth.
The arrival of many other conditions shown in Table 5 does appear to be related to age. Considering the first age at which the rate of occurrence is greater than the average at all ages, back or spine problems tend to first occur after age 30. A wide range of other problems are first observed with rates of occurrence greater than average after age 40 including arthritis or rheumatism, blindness or vision problems, cancer, carpel tunnel syndrome, diabetes, heart trouble, strokes, respiratory problems, and high blood pressure. Many of these conditions would be related to each other and might occur simultaneously such as high blood pressure, heart trouble and respiratory problems.
Descriptive statistics for estimation sample
We report the weighted descriptive population statistics of the variables we constructed from the SIPP: age, onset age, race, Hispanic origin, education level, marital status and number of children for the 30 detailed health conditions available.
Weighted descriptive statistics by 30 health conditions for men, SIPP 2004 Wave 2
Weighted descriptive statistics by 30 health conditions for men, SIPP 2004 Wave 2
Note: All statistics are weighted except the sample numbers, proportions are shown in percentage (%).
The tabulations of these variables are reflective of characteristics at the age at which the conditions occur with the exception of the number of children which is reported at the time of the survey. Tables 6–8 contain information for the whole sample, men and women respectively. As can be seen in Table 6, the greatest proportions of people experience work limitations because of back or spine problems (1.88 percent), arthritis (0.85 percent), mental or emotional conditions (0.81 percent), heart trouble (0.74 percent) and diabetes (0.53 percent). As can also be seen in Tables 7 and 8, those five health conditions are also the most common ones for both men and women although the prevalence by gender differs a bit. The average age in the sample is 36 while the average age of onset for some specific conditions (e.g. arthritis, cancer, diabetes, heart trouble, high blood pressure, stroke and thyroid trouble) is well beyond age 50 as shown in the top row of Table 6. The proportion of people in the population experiencing limitations has a high early life occurrence. More than 1 percent of the population experiences the onset of a work limiting disability by age 16. Beyond that time, the likelihood of the onset of a work limiting health condition increases steadily through the mid-50s where the odds of onset are 0.91 (see the right margin of the table).
There are, however, some health conditions, such as intellectual disabilities, learning disabilities and cerebral palsy where the mean age of onset is relatively low and a large proportion of those who experience those conditions have very young onset ages. This can be seen by examining the row of the tables for the age category of “
Considering the demographic covariates, categorical variables for four racial categories are created for the estimations. In our sample, the proportions shown in the right hand column of Table 6 are: White (80.56 percent), Black (12.54 percent), Asian (3.42 percent) and other (3.48 percent). We also create a variable for Hispanic origin from the SIPP data where 14.14 percent of the sample reports being Spanish, Hispanic or Latino.
Weighted descriptive statistics by 30 health conditions for women, SIPP 2004 Wave 2
Note: All statistics are weighted except the sample numbers, proportions are shown in percentage (%).
We also created variables based on education level, marital status and number of children. The proportions in these categories can also be found in the right-hand column of Table 6. The highest education level that most people have achieved is a high school diploma (21.62 percent) or some college (25.85 percent); and most people are either married with their spouse present (41.12 percent) or never have been married (43.48 percent) due to their young age at the time of the survey. For women, the SIPP asks about the number of children they ever gave birth to and the answer ranges from zero to six. For men, the SIPP asks about the number of children they are the biological father of and the answer ranges from zero to five. 52.24 percent report having children and this may also affect a person’s likelihood of experiencing the onset of some health conditions. In addition, the SIPP collects information on education level and number of children only for individuals older than 15. Thus, 21.06 percent of the weighted population do not have available information on education and number of children, as is shown in the right-hand column of Table 6.
Tables 7 and 8 allow us to descriptively compare men and women. In the survey, there are 49,951 males and 53,877 females that make up 48.88 percent and 51.12 percent of the weighted population respectively as shown in Table 6. There are 13 conditions experienced by a higher proportion of men than women: an alcohol or drug problem, AIDS, blindness, broken bone, deafness, head injury, heart trouble, hernia, learning disability, intellectual disability, missing limbs, paralysis, and stiff foot/hand. The average age for men in the SIPP (with or without health conditions) is 34.99, younger than that of women 37.12; however, for one third of the health conditions considered, the average ages for men at the point of onset of those health conditions are older than for women. This includes alcohol or drug problems, AIDS, head injury, heart trouble, learning disability, intellectual disability, multiple sclerosis, paralysis and stroke. So, on average, men experience some health limits at different stages in life than women. On the other hand, in comparison to women, there are higher proportions of men that experience back problems or a broken bone during their younger ages, but lower percentages of men have these problems at older ages.
Following the prevalent method in the related literature, we use logistic regression models to conduct multi-variate analyses of the relationship between the onset of health conditions that limit work and the available demographic covariates. The dependent variable is a categorical indicator of whether the individual reports any health condition that causes a work limitation,
Logistic regression models (estimated odds ratios) of any health conditions for men and women, 2004 SIPP
Logistic regression models (estimated odds ratios) of any health conditions for men and women, 2004 SIPP
Note: Exponentiated coefficients; Standard errors in parentheses. **
Logistic regression models (estimated odds ratios) of 30 specific health conditions for men, 2004 SIPP
Note: Exponentiated coefficients. **
In our main regression model, the relative probability of the onset of any health condition for individual
The probability that
Here, “Whites”, “non-Hispanic”, “less than high school”, ”married”, and”no child” are the reference groups and thus not included in the Eq. (1). Based on Eq. (1), the probability of having a health condition (probability that
Alternatively, the odds of having a health condition (odds of
The estimated coefficients cannot be interpreted as a change in the dependent variable for a one unit change in the independent variable. Instead, we have to translate the parameter estimates using the exponential function to obtain an odds ratio. For example, the odds ratio for the first term
Considering the information on education level and number of children are only available for individuals older than 15, we exclude 23,252 observations from the total 103,828 observations when we do the logistic regressions.
The logistic regression results relating observed demographic characteristics to experiencing the onset of any work related health condition are contained in Table 9. In Model 1, we include only age as a control. In Model 2, we add available measures of fixed characteristics (race and Hispanic origin). In Model 3, we additionally include controls for the time-varying socio-demographic variables (education level and marital status). Model 4 adds to Model 3 a control for the number of children.
In all four models for men, we find strong and statistically significant results indicating that all of the included covariates except for the number of children are related to the onset of any health condition that limits work. For women, we also obtain statistically significant results for virtually all of the estimates across the four models in Table 9 against the null hypothesis except for the variables that indicate a woman was never married and also the variables capturing the number of children.
In all of the alternative models estimated for men and women, the estimated odds ratios for age indicate that for each additional year of life, there is about a 1.1 to 2.4 percentage point increase in the likelihood of experiencing the onset of a work limiting health problem. The odds for men range from 1.4 to 2.4 percent, somewhat higher than for women, 1.1 to 1.4 percent.
Logistic regression models (estimated odds ratios) of 30 specific health conditions for women, 2004 SIPP
Logistic regression models (estimated odds ratios) of 30 specific health conditions for women, 2004 SIPP
Note: Exponentiated coefficients. **
Logistic regression models (estimated odds ratios) of cancer
Note: Exponentiated coefficients; Standard errors in parentheses. **
Logistic regression models (estimated odds ratios) of carpal tunnel syndrome
Note: Exponentiated coefficients; Standard errors in parentheses. **
Considering the race and ethnicity indicators, Asians appear to be less likely to report a health problem that limits work. The odds ratios of 0.476, 0.585 and 0.584 in Models 2–4 for men in Table 9 indicates that the odds of reporting any health condition is 41.5 (i.e.
Compared to Whites, Blacks are consistently found to be more likely to have work limiting health conditions. The odds ratios in Models 2–4 are 1.757, 1.442 and 1.432 for Black men, indicating that they are more likely to experience the onset of work limiting health conditions but the somewhat lower estimate in Model 4 indicates that education, marital status and number of children (variables added to Model 4) moderate this relationship. Similarly, for Black women, the odds ratio reported from Model 2 is 1.582 while the estimate from Model 4 with all covariates included is somewhat smaller, 1.375. However, both estimates indicate that Black women face much higher odds of experiencing the onset of a health limitation.
Probabilities of reporting health conditions are lower for Hispanic men than for non-Hispanics; as can be seen in Models 2–4. The estimated odds ratios from those models are 0.828, 0.689 and 0.685, indicating that Hispanic men are 31.5 to 17.2 percent less likely than White men to experience the onset of a work limiting health condition. Among Hispanic women, the odds of experiencing the onset of a work limiting health condition are also lower than among White women. The estimates range from 0.922 in Model 2 to 0.760 in Model 4 indicate the odds of experiencing onset of a health limitation are 7.8 to 34 percent lower. The odds are more sharply lower once the full set of covariates is included in Model 4 suggesting that conditional on education and marital status, Hispanic women experience much better health than white women.
Logistic regression models (estimated odds ratios) of deafness or serious trouble hearing
Note: Exponentiated coefficients; Standard errors in parentheses. **
Logistic regression models (estimated odds ratios) of paralysis
Note: Exponentiated coefficients; Standard errors in parentheses. **
Logistic regression models (estimated odds ratios) of thyroid trouble
Note: Exponentiated coefficients; Standard errors in parentheses. **
Logistic regression models (estimated odds ratios) of tumor cyst or growth
Note: Exponentiated coefficients; Standard errors in parentheses. **
People with higher education levels have lower odds of experiencing the onset of a work limiting health condition. The categorical variables for educational level are included in Models 3 and 4 for the sample of men and women. Among men in Models 3 and 4, every educational category above the omitted group with less than a high school education experiences a reduced odds of experiencing the onset of a disability and the odds progressively decline with higher levels of education. Among women, in Models 3 and 4, a very similar pattern is observed.
Marital status indicators are included in Models 3 and 4 for men and women. Divorced men and women have a much higher probability of the onset of a work limiting health condition compared to married individuals in the sample. For men, the divorced have a higher probability (odds ratio of 2.297 and 2.309) of experiencing onset of a health limitation while the odds ratio for widowers (0.532 and 0.536) is less than one. For women, the odds ratio for those who are divorced in Models 3 and 4 (2.225 and 2.220) and widows (0.601 and 0.600) are similar to those of men and the probability of the onset of a work limiting health condition is lower for widowed than married women; however, there are no significant estimates associated with being a never married woman. The categorical variables for the number of children included in Model 4 for men and women do not have a statistically significant effect on the probability of onset of a work limiting health condition.
The demographic covariates are highly related to the onset of any work limiting health condition, but we are also interested in whether this pattern is observed for individual disaggregated conditions. We make use of categorical indicators of whether an individual experiences each of the individual 30 health conditions as dependent variables, and run a set of logistic regression models for each condition on all available covariates for men and women separately making use of Model 4. The regression results for the samples of men and women are contained in Tables 10 and 11 respectively.
We find significant relationships between the demographic covariates and the existence of health limitations for most of the 30 health conditions as was observed for the aggregated outcome of any health limitation. The importance of the covariates in predicting the onset of most of the specific disabilities is similar to that of all health limitations. However, for some specific health conditions, many of the demographic controls are not statistically related to the health condition at conventional levels for both men and women. There are six particular conditions where we find that fewer of the demographic covariates are predictive of the specific work limiting health problem other than age, particularly for men. These six conditions include cancer, carpal tunnel syndrome, deafness or serious trouble hearing, paralysis, thyroid, and tumor cyst or growth.
To examine these patterns more closely in Tables 12 through 17 we present additional estimates for each condition varying the sets of covariates used in the analysis as was done in Table 9 for the report of the onset of any health problem that limits work for both men and women. In those tables, it can be seen that for men, the only observed demographic characteristic that is significantly related to having cancer (Table 12) or a tumor (Table 17) is divorce, other than age. Thus, among samples of married men, the arrival of these conditions controlling for age is not predictable in this sample based on the other observed characteristics. Similarly, the only covariate predictive of paralysis (Table 15) other than age and divorce is being a member of the residual race category. None of the covariates are predictive of carpal tunnel syndrome (Table 13). Only higher education and the residual race category, other than age, are predictive of deafness (Table 14) or thyroid (Table 16) problems (respectively) that limit the type or amount of work that men do.
For women, similarly, the only significant correlate of cancer (Table 12) or paralysis (Table 15) other than age is being never married or divorced respectively. Thus, among samples of married women, these conditions may be observed as arriving randomly controlling for age. For tumors or growths, none of the covariates are statistically significant. For thyroid problems (Table 16), being Black or in the residual race categories appear to be significantly related to experiencing related work limitations. Widowed women and those with high levels of education (Master or higher) also appear to be more likely to experience hearing problems (Table 14). While the demographic covariates examined are somewhat more related to these six conditions for women than for men, there are far fewer covariates related to these conditions than for others. Among married white women, these conditions also seem to arrive at random conditional on the covariates available for the analysis.
The prior literature examining the relationship of demographic factors to health conditions has often focused on either single outcomes such as mortality and self-assessed health or single diagnoses such as mental health problems or back pain. Examination of these health outcomes was important in establishing which groups in society and which proxies for lack of access to resources were most strongly related to worse health outcomes. Nonetheless, a more detailed examination of underlying conditions is useful in helping to understand differentiated outcomes in the population.
In this paper, we make use of the topical module on work disability history from Wave 2 of the 2004 SIPP to describe the onset of any health problem that limits work as well as an underlying set of 30 specific conditions across different demographic groups. We find that most people, men and women, experience work limitations because of back or spine problems, arthritis, mental or emotional conditions, heart trouble and diabetes. People of different races and Hispanic origin also have differing incidences of health conditions. Asians and Hispanics both experience fewer work limiting health conditions.
Most people report the onset of work limiting health conditions when they are younger than 16 or beyond the age of 46. The most common ages for first experiencing different work limiting health problems varies depending on the specific condition being examined. This type of observation cannot be drawn from only examining the onset of any health problem that limits work. Arthritis, cancer, diabetes, heart trouble, high blood pressure, stroke and thyroid trouble occur in later stages of the lives of the sample members while intellectual disabilities, learning disabilities and cerebral palsy occur earlier.
These differential outcomes in health experiences also appear to vary by gender. Men and women face unequal risks of having different health conditions and the conditions occur in different stages of their lives. For example, men are more likely to experience alcohol and drug as well as heart problems and these issues tend to show up at later ages than among women.
Beyond descriptive patterns, a multi-variate analysis based on logistic regression was carried out to examine the relationship between demographic covariates commonly used in prior research (age, gender, race, Hispanic origin, marital status and education level) and the odds of experiencing both any work limiting health condition and also the 30 detailed conditions in the 2004 SIPP. With respect to the onset of any health limitation and the vast majority of the 30 individual health problems, the demographic covariates are also found here to be strongly predictive of onset. However, there is a subset of conditions that are not highly related to these demographic covariates. In particular, there appear to be some conditions that among married populations, controlling for age and one or two additional covariates, would be seen as arriving unexpectedly; cancer, carpal tunnel syndrome, deafness or serious trouble hearing, paralysis, thyroid, and tumor cyst or growth.
The potential usefulness of this finding is that adjustment of individuals and their families to the onset of a work limiting condition may be differentiated by whether a health problem arrives in a less predictable manner. For example, if alcohol abuse is fairly predictable and has been preceded by warnings, then its interference with work might precipitate being fired from a job or a marital dissolution. Alternatively, the arrival of an unpredictable cancer or paralysis might be less likely to precipitate divorce.
The findings of this paper are largely consistent with the prior literature while providing a more detailed and nuanced picture of variation in underlying health conditions. Nonetheless, there are some concerns about the analysis that we have tried to address within the paper. We compare the retrospective reports of health problems from the SIPP as well as alternative methods of data handling to differently collected data in the NHIS and make use of the method of data handling in the SIPP that gives the closest observed correspondence across the two sources of information. We also examine whether recent reports of work limiting health problems in the SIPP differ from those obtained in the entire sample and do not find evidence that would be indicative of recall bias.
It would be worthwhile to replicate the findings of this paper using detailed measures of underlying health conditions from other data sources. Also, the retrospective SIPP data do not contain measures of specific health related behaviors such as exercise that might be predictive of future health. The data also lack measures of household resources and access to health care over time that would more directly reflect individuals’ abilities to engage in preventive health care. Extending these analyses in contexts where additional measures such as these are available would also be useful.
Footnotes
Detailed information on the NHIS data can be found at this link:
Detailed information on the SIPP data can be found at this link:
The econometric models used in this study were implemented using STATA/SE 12.0 on an Intel Core i5-2520M CPU laptop running at 2.50 GHz with 16 GB of memory, running Windows 10 64-bit operating system.