Introduction
Subjective health and well-being are of considerable interest in population health
research. Subjective assessments offer unique scope to capture latent health concerns
that cannot be directly (or cost-effectively) captured through objective assessments
(Jylhä, 2009). These assessments are sometimes more reliable predictors of old-age
mortality than standard clinical biomarkers (Idler & Benyamini, 1997). At the same
time, greater caution is warranted while studying subjective health and well-being
in a cross-cultural setting (Jürges, 2007, Jylhä et al., 1998, McDowell, 2006, Schwarz
et al., 2010). Glei (2017) succinctly elicit these concerns while examining self-reported
physical limitation in United States and three other countries having similar life-expectancy
(England, Taiwan and Costa Rica). It is noted that the absolute population-level prevalence
of self-reported physical limitations varies across the four US-based surveys (HRS,
MIDUS, NHIS and NHANES) and thus disallows a robust cross-country comparison. Such
variability can have non-trivial implications for comparisons as one can arrive at
contradictory conclusions regarding the nature of associations. For instance, MIDUS
suggests that American men report walking limitations much earlier than Costa Ricans
whereas HRS based assessment indicates that the American men are more advantaged.
Glei (2017) set up an interesting hypothesis regarding the association of subjective
health indicators with macro- or country-level indicators such as life expectancy
or per capita incomes. Although, data limitations in terms of country units for analysis
can prohibit robust country-level inferences but it outlines an important agenda for
further research. For example, Fig. 1 presents the country-level association of percentage
bad and very-bad self-rated health with per capita gross national income and life
expectancy at birth. There is an inverse association of per capita incomes and life-expectancies
with percentage of population (age 25 and over) reporting bad or very bad health.
The association appears consistent because high-income countries are also likely to
have higher life expectancy at birth. Also, the fact that self-rated health and mortality
share a significant relationship lends credence to this association (Idler & Benyamini,
1997). But the scatter also depicts considerable heterogeneity and reporting of bad
or very-bad health has large variations at lower per capita incomes and at higher
life expectancies.
Fig. 1
Association of percentage self-rated bad/very bad health (among aged 25 and over)
with per capita GNI and life-expectancy at birth across 69 countries in 2002. Note:
Data for the self-rated health status is based on World Health Survey (2002) and is
sourced from Subramanian, Huijts, and Avendano (2010) whereas data for GNI per capita
and life expectancy at birth are from World Development Indicators database. Linear
trendline is also presented for GNI per capita (p < 0.01) and life expectancy at birth
(p < 0.10).
Fig. 1
The question raised by Glei (2017) is embedded in this fundamental concern of heterogeneity
in self-rated health outcomes and makes it important to understand how much of these
variations are truly systematic, can be standardized (by age, education, income) or
can be associated with variations in language, culture and categorical ordering of
the question. In this commentary, we highlight some of these intricacies involved
in robust cross-cultural comparative research. We present a list of key concerns that
desire specific attention. Given the increasing relevance of subjective health research,
we also call for systematic reviews on methodologies and comparability of findings
in cross-comparative research.
Desiderata for cross comparative research
With increasing cross-cultural and cross-national analysis of health and well-being,
it is mandatory to work toward strategies to ensure sound comparisons. While there
is no gold-standard in comparative subjective health assessments, nevertheless it
is worth reiterating certain tenets and requirements (Table 1). To begin, a clear
comprehension of research question is elementary to outline the merit and the unit
of analysis (Ragin, 1982, Buil et al., 2012). Further, from a methods perspective,
the subjective indicator of interest should enjoy a shared level of confidentiality
across contexts. For instance, a general question on self-rated health may be more
comparable across contexts than subjective assessments of intimate partner violence
or sexual and reproductive health outcomes (Pallitto et al., 2013, Vyas and Watts,
2009). Robust comparisons in such cases may warrant further sensitivity adjustments
for survey-related or contextual factors. Similarly, high correspondence between survey
instruments and appropriate linguistic translation across contexts is a necessity
for such assessments (Buil et al., 2012, McDowell, 2006, Schwarz, 1999, Schwarz et
al., 2010). Under variable contexts, statistical approaches such as vignettes-based
adjustments and cut-point shift are among a few alternatives recommended to facilitate
robust comparisons (Jürges, 2007, Lindeboom and Van Doorslaer, 2004, Salomon et al.,
2004).
Table 1
Critical concerns in comparative analysis of subjective health and well-being.
Table 1
Concern
Meaning
Relevance
Desirability
The purpose (question) of temporal, national or cultural comparison of subjective
health status should be outlined with a clear motivation from a research and/or policy
perspective.
In general, there should be uniformity in the conceptualization of outcomes, unit
of analysis and/or explanatory processes to arrive at meaningful comparisons. For
instance, from a health perspective, a direct comparison of self-rated health between
rich and poor countries has to be well-thought to understand explanatory processes
that are common to these contexts. Analysis of social capital and self-rated health
is an example of such research.
Confidentiality
The indicator used for the comparative analysis should display a similar degree of
confidentiality or privacy level across comparison groups.
Despite use of uniform survey design, questions and interview protocols, it is plausible
that the response to subjective assessment can vary depending on the level of comfortability
shared by the respondent in general as well as particularly during the survey. Certain
indicators that require greater degree of confidentiality and privacy are more likely
to be biased because of systematic differences in the social environment. Subjective
assessment of intimate partner violence or sexual and reproductive health are good
examples.
Harmony
The survey and instrument design across contexts should correspond well in terms of
sequencing, ordering, coding and composition of the survey questions, in general,
and subjective questions, in particular.
Sequencing and ordering of the questions have considerable influence on the reported
outcomes. These can be due to reasons associated with attention and interest of the
respondents as well as fatigue factor that may lead to biased reporting. Any comparative
analysis should aim to present a careful review of aspects related to sequencing,
ordering, coding and composition of the survey questions along with its potential
implications for analytical inferences.
Transferability
The use of wording and coding structures should display transferability from a sociocultural
perspective and not merely reflect language translation.
Translation loss can be severe in comparative analysis across contexts that have very
different linguistic and cultural outlook. A clear identification of the presence
of such problem and its potential implications should be presented as a limitation
or adjusted using statistical analysis. Adjustments using vignettes-approach or cut-point
shift is often used under such circumstances.
Replicability
Estimates of subjective health should be consistent across replications within the
same population or context.
A high degree of conformity between survey based estimates from the same population
enhances its validity as a reference estimate for cross-cultural or cross-country
comparisons. In case of variations, the best possible survey source should be identified
by reflecting upon other critical concerns described here.
Sensitivity
The transformation of qualitative or categorical subjective health information for
quantitative analysis should be tested for sensitivity.
Application of quantitative techniques for analysis of subjective health information
involves certain assumptions for recoding of qualitative or categorical information.
A binary coding of ordered or multi-categorical variables to facilitate a logistic
regression is perhaps the most common example. Sensitivity analysis is one reasonable
approach to verify the implications of such recoding practices.
Consistency
The information used for comparative analysis should be verified for internal consistency
with other subjective information within the surveys.
A high internal consistency of information implies greater validity of reported information
and can facilitate better comparative analysis across contexts and situations. There
are standard statistical tools such as reliability coefficients to aid such analysis.
Objectivity
Greater correspondence between subjective health status and objective assessments
of health is necessary as it can provide more direct understanding of health among
the subjects or else it may be capturing certain other (unintended) aspect of health
and well-being.
While indicators of subjective health status and objective health assessments may
not always be in confirmation but still one can expect significant overlaps. Thus,
a greater divergence between subjective and objective health indicators may imply
that the subjective health indicator is potentially capturing certain other influences
on health status reporting. Any variability in subjective and objective assessments
across surveys and contexts therefore will require appropriate interpretation and
understanding of the research question.
Credibility
The surveys and indicators used for comparisons should have negligible limitations
and greater scope for generalizability of the findings.
Comparative analysis should be aware of the potential limitations of the respective
surveys and findings and its broader implications for generalizability of cross-cultural
or cross-country analysis and its inferences. In particular, huge limitations may
reduce credibility of the findings and can be contested with data that has more advantages
than limitations.
Completeness
It is critical that all comparative analysis should justify its empirical merits based
on the above defined concerns.
Cross-country comparisons should be based on sound understanding of the research problem,
nature of data, survey design, and information comparability.
Furthermore, wherever feasible, it is important to verify the consistency of the estimates
across multiple data sources for the same population. Quantitative analysis based
on post-survey re-coding of ordered and categorical subjective health information
should be tested for sensitivity bias. Similarly, it should be mandatory for studies
to report reliability coefficients of relevant indicators (Cortina, 1993, Webb et
al., 2006). The accuracy of interpretation can be further enhanced if one can ascertain
the objective health information embedded within the subjective indicators. This is
linked to what Sen (2002) refers to as an internal view of health whereby one can
expect overestimation of positive self-rated health in deprived and uninformed settings.
Empirical studies have confirmed such tendencies and also suggest that the deviations
can be sensitive to the nature of the health indicator being assessed (Innerd et al.,
2015, Johnston et al., 2009, Mosca et al., 2013). A simple bivariate comparison of
subjective and objective health indicators can shed some preliminary insights on these
issues with corrections for endogeneity bias necessary in some contexts (Jones & Wildman,
2008). Finally, limitations of the survey, indicator and data should be explicitly
discussed for generalizability and credibility of the findings. Comparative analysis
that effectively address these concerns can allow more meaningful research inferences
(Buil et al., 2012). In particular, well-thought research desirability backed by harmonious
data and robust sensitivity checks can be a good starting point for such comparative
research.
Concluding remarks
Glei (2017) draw attention towards an important methodological concern in comparative
assessments of subjective health and well-being. This implies that researchers and
organizations engaged in routine assessments of subjective health status also have
a shared responsibility to report standard quality check parameters, particularly
about consistency and reliability of subjective data and indicators. In fact, with
increasing cross-cultural research, systematic reviews on various domains of subjective
health are necessary to develop comprehensive methodological guidelines for robust
comparisons.
Conflict of interest
None.
Funding support
None.