Critical Perspectives on Racial and Ethnic Differences in Health in Late Life (2004) / Chapter Skim
Currently Skimming:
6 Selection Processes in the Study of Racial and Ethnic Differentials in Adult Health and Mortality
6 Selection Processes in the Study of Racial and Ethnic Differentials in Adult Health and Mortality
Pages 171-226
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 171... ...
estimates of effects of membership in a racial or ethnic group that occur due to the existence of intervening mechanisms. We do not undertake this task with the presumption that selection processes are the only or even the most important mechanisms that generate observed racial and ethnic disparities in health and mortality.
|
From page 172... ...
CONCEPTUAL CLARIFICATION: NATURE OF SELECTION PROCESSES Conceptualization and Examples The observed association between an individual's social class or position and health status and mortality risks can be due to two different processes. The first is one whereby influences on health and mortality result from the action of characteristics intrinsic to the social position.
|
From page 173... ...
In statistical jargon, accounting for selection is necessary to obtain consistent estimates of the effects of social class on health status and mortality. Second, except in the case when traits relevant for both health and social stratification are allocated at random, mechanisms involving health selection are part of the overall process whereby social stratification generates health and mortality inequalities and thus should also be a focus of study for researchers interested in the genesis of health and mortality differentials.
|
From page 174... ...
These relations contribute to the observational correlation between social position and mortality, but are not selection effects as defined earlier. Instead, they should be genuinely attributable to the occupancy of the position.
|
From page 175... ...
. But this may overlook the fact that Japanese migrants to California were not a random sample of the Japanese population, either in terms of social class or in terms of region of residence or their own ethnicity.
|
From page 176... ...
. Finally, the "Hispanic paradox" in the United States refers to the fact that Mexican and some non-Mexican Hispanics experience similar or better health status and lower adult mortality rates, and their infants are born at higher weights than African Americans and non-Hispanic whites (Palloni and Morenoff, 2001)
|
From page 177... ...
. Because the relevant processes may be spread out over a lifetime, involve long time lags, and are mediated by a number of intervening mechanisms, the resulting effects are referred to in the epidemiological literature with the rather unfortunate label of "indirect selection effects." This is apparently to distinguish from "direct selection effects," which are more akin to reverse causality (to be reviewed)
|
From page 178... ...
. Another possibility we will explore as an example of reverse causality is that at older ages, there is substantial return migration to Mexico by Mexican immigrants with poor health status.
|
From page 179... ...
This type of selection effect is referred to in the literature as heterogeneity. Reverse Causality, `Drift' or Social Mobility Through Ill Health There are situations in which severe health limitations and impairments constrain individuals affected by them to occupy a much narrower range of occupations or social positions than the general population.
|
From page 180... ...
Like the healthy migrant and healthy worker effect, the observable association does not reflect the influence of characteristics or traits of social positions on the health status or mortality risks of individuals. In the epidemiological literature, these relations and their observational features are referred to as "direct selection effects" or "drift" (West, 1991)
|
From page 181... ...
Although in the more Social class Health status Traits FIGURE 6-1a Relations involved in health selection effects.
|
From page 182... ...
, we will refer to it as an example of reverse causality. Antecedent Social class health status of destination Social class of origin FIGURE 6-1b Relations involved in reverse causation.
|
From page 183... ...
The Hispanic paradox, a possible outcome of both health selection effects and reverse causality, is a key phenomenon in the assessment of differentials in health and mortality for Hispanics in general, but also for other ethnic groups heavily affected by migration inflows. According to recent estimates, the Hispanic population aged 65 and over represents about 4.1 percent of the total U.S.
|
From page 184... ...
. More recently, the issue of health selection effects and some types of reverse causality have resurfaced, but have also become more contentious and controversial, often pitting polarized positions against each other, from those who believe that virtually all social class differentials in
|
From page 185... ...
, in which selection processes (mainly health selection effects and reverse causality) are considered as one of four possible mechanisms producing social class differentials in mortality.
|
From page 186... ...
Bringing Selection Processes Back In The tenor of the dispute about health selection effects, particularly in the context of British studies, has distracted researchers from the key issues on which all parties in the dispute should agree (Blane et al., 1993; West,
|
From page 187... ...
Similarly, West (1991) cogently argues for a framework that, while directing attention to social class attributes, should also include consideration of selection processes (mostly health selection effects and reverse causality)
|
From page 188... ...
First, we examine the so-called Hispanic paradox and assess the claim that it may be the result of health selection effects and reverse causality. Second, we discuss the importance of heterogeneity in the case of black and white mortality disparities in the United States.
|
From page 189... ...
But it could also reflect the impacts of health selection and reverse causality. Health selection through migration can occur as individuals who reach the United States and become more or less established residents are more likely to be drawn from a population that is less frail than the one that does not migrate.
|
From page 190... ...
. Magnitude of Biases: Health Selection Effects We first calculate the magnitude and direction of biases that may affect estimates of adult mortality and health status when there is health selection of migrants.
|
From page 191... ...
This means that even under a mild selection regime, the observed mortality ratios will be consistent with a "Hispanic" advantage that becomes diluted at older ages. These results suggest a strategy, albeit precarious, to identify health selection effects.
|
From page 192... ...
This exercise is suggestive and illustrates the potential magnitude of health selection effects. Its results are also consistent with the fact that life expectancy at adult ages (over 45)
|
From page 193... ...
The most important prediction from this simple representation is that migrants will be selected on health, and that health selection will increase if the costs of migration increase and if skill prices in the area of origin are lower relative to skill prices at destination. The answer to the original query about 8o is now straightforward: the higher the migration costs and the wider the skill price disparities, the lower the value of 8o will be of origin relative to those at destination.
|
From page 194... ...
If so, the comparison between those remaining in the United States and the non-Hispanic population will lead to incorrect inferences. To assess empirically the magnitude of effects due to reverse causality, we perform an empirical exercise using the so-called National Health Interview Survey-National Death Index (NHIS-NDI)
|
From page 195... ...
Though modest, these differences are somewhat paradoxical. Although the magnitude of the differences favoring Hispanics, particularly Mexicans and other Hispanics, is admittedly large, it falls well within the bounds of what would be expected via health selection (as discussed
|
From page 196... ...
Yet, this does not prove the case for health selection effects. It merely raises a red flag.
|
From page 197... ...
. Comparisons of mortality rates for two ethnic groups or social classes yield discrepancies that attain maximum values at adult ages and decline steadily thereafter (Manton et al., 1979; Nam and Okay, 1977; Strehler, 1977; Vaupel et al., 1979)
|
From page 198... ...
With suitable modifications, similar procedures can be employed to examine the convergence of migrants' mortality rates by duration of stay. A Multivariate Parametric Approach Let us focus on the black-white disparity in the United States.
|
From page 199... ...
If there is convergence driven by a more severe mortality regime early in life among blacks, we should see a deceleration in the rate of increase of mortality rates for blacks, but much less so for whites. Thus, the first test is to verify that the slope for the Gompertz function is lower for black males than it is for white males.
|
From page 200... ...
With an important caveat, the main conclusion from this exercise is that, as suggested initially by Figure 6-3, there are no signs of even weak convergence of black and white mortality patterns. There is no empirical TABLE 62 Hazard Models for Mortality of Males in the NLSM, 1966 1991 Parameter (std error)
|
From page 201... ...
applies various procedures to the study of mortality disparities between individuals with different genotypes. One can adapt one of these strategies to assess approximately the effects of heterogeneity on estimates of black-white mortality differentials.
|
From page 202... ...
, the ratios would be much flatter than observed, indicating that disparities are larger than those observed. The main point that these figures illustrate is that inferences about mortality disparities that rely on observed ratios of death rates are subject to a great deal of uncertainty unless we know more about the parameters of the distribution of unmeasured traits on which selection is occurring.
|
From page 203... ...
Health Selection Effects Associated with Social Class The case of the Hispanic paradox examined earlier illustrates the importance of at least two types of selection, health selection effects ("healthy migrant effect") and reverse causality ("return migration effects" or "salmon bias")
|
From page 204... ...
Even when interethnic or interrace mobility is unimportant or impossible, health selection effects implicating membership in social classes within each racial or ethnic group are relevant. The reason is that attribution of health or mortality effects to a racial or ethnic group normally requires controls for a number of confounding factors, including social class or indicators of social class such as income education.
|
From page 205... ...
We now turn our attention to an assessment of the potential magnitude of health selection effects proper, that is, processes through which individuals are allocated to various social classes by virtue of traits acquired early in life that affect both risk of accession to the social positions and health status and mortality. In an insightful but overlooked account of the misleading inferences that the presence of health selection may produce, Stern (1983)
|
From page 206... ...
We also assume that inheritance of health status leads to immutable adult health status and that no improvement or deterioration will take place by virtue of membership in a particular social class. These assumptions are confining, but they enable us to perform calculations that are revealing of the magnitude of effects attributable to health selection effects.
|
From page 207... ...
; Social 0 0.7 0 0.3 H3: 0.5 0.5 0 0 net .8 0 good by health-dependent status represent and higher 1 0 0 0 0.5 0.5 0 0 1.8 01 00 00 (poor expanded = inheritance; = health class, = columns class Population M2 H2 and F2 and M2: health social low no class rows 0 0 0 1 0 0 0 1 0 and growing Calculate mobility; H2: social represent to by F2: 0.05 0 0.95 0 0 0 1 0 0 matrices values columns) inheritance; Matrices 0 0.95 0 0.05 matrices 4-by-4 0 1 0 0 0 100 010 001 (and population; are matrices health-dependent health rows status 6-3a 1 0 0 0 1 0 0 0 reproductive 1 0 0 0 two limited matrices Mobility = perfect Health = Net stationary = first *
|
From page 208... ...
What Mechanisms or Processes Produce Health Selection into Social Classes? Having illustrated that race comparisons can be affected by health selection into social classes, the task remains of identifying those processes that result in health selection effects.
|
From page 209... ...
These results are all from steady state distributions.
|
From page 210... ...
In his paper on health selection effects, Stern (1983) drew attention to the possibility that health selection effects could be responsible for artificially inflated effects of social class and, furthermore, that these effects were more pervasive than those more commonly attributable to reverse causality.
|
From page 211... ...
. In summary, the evidence gathered for the various intervening mechanisms or processes is fragile, and the case for health selection effects operating through individual traits shaped early in life remains a suggestive but elusive hypothesis.
|
From page 212... ...
will we increase our ability to assess the actual importance of intervening processes that are potential generators of health selection effects. Reverse Causality Reverse causality has been examined with some detail by epidemiological studies of the relation between occupation and mortality.
|
From page 213... ...
and that of time-dependent assets must be considerably attenuated once we control for health status at the onset of the study. Note that it is likely to be the case that if there are any health selection effects that manifest themselves late in the life of individuals, a control for health status will also reduce their impact.
|
From page 214... ...
Fourth and finally, the estimated effects of variable assets are considerably reduced, though they do not disappear, when controlling for self-reported health status. The fact that effects of time-varying assets are at least 40 percent higher than those associated with fixed assets is an important sign of either health selection effects or of reverse causality of the type discussed by Smith (1999)
|
From page 215... ...
But because it does not address the issue of whether health early in life influences economic status of individuals during the early part of their labor market experience, it is simply not pertinent for health selection effects of the type investigated before. In summary, although it is too early to make strong statements about the exact nature of these relations, the estimated effects obtained so far are important and, at the very least, suggest that overlooking the direct influence of health on wealth in late life may result in misinterpretations of the nature of the relation between socioeconomic position and health status.
|
From page 216... ...
has shown that the range of patterns for mortality gradients generated by different health selection processes is much broader than what researchers have speculated. Although Goldman's simulations were tailored to deal with the problem of marriage selection, her results are equally valid for social class gradients.
|
From page 217... ...
The identification problem could be minimized if the information available is appropriate. The data demands to deal with health selection effects addressed in this chapter are high because the researcher requires information over the life course of individuals.
|
From page 218... ...
The first is that estimates of parameters are somewhat sensitive to the specification of distributional properties of the unmeasured traits. Recall that selection through survival is attributable to individual traits that increase (decrease)
|
From page 219... ...
. Another example is the estimation of unmeasured heterogeneity using spouse data and its application to the assessment of mortality differentials by social class (Mare and Palloni, 1986)
|
From page 220... ...
To simplify terminology and unless otherwise stated, we use the term "unmeasured heterogeneity" to refer to selection processes through survival only. However, research that requires the formulation of multistate hazard models must face up to both "heterogeneity" problems.
|
From page 221... ...
11. Arguments analogous to these can be invoked to justify attention to health selection via reverse causality, or processes whereby health status directly impacts the risk of social mobility.
|
From page 222... ...
. Social class mortality differentials: Arti fact, selection or life circumstances?
|
From page 223... ...
. Childhood living conditions, health status, and social mobility: A con tribution to the health selection debate.
|
From page 224... ...
. Social classes, inequalities and health dispari ties: The intervening role of early health status.
|
From page 225... ...
. Do health selection effects last?
|
From page 226... ...
. Rethinking the health selection explanation for health inequalities.
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.