Income and economic exclusion: do they measure the same concept?

In most developed countries, glaring health inequalities exist that reflect, but are not reducible to, lifestyle and health behaviours, and that bear a strong relationship with socioeconomic position [1‐3]. This situation has been deemed of such concern to researchers and policymakers alike that limiting these inequalities has recently been put at the forefront of the policy agenda by the World Health Organisation [4]. In Canada for instance, the Government of Quebec passed in 2002 An Act to combat poverty and social exclusion (R.S.Q., c. L-7). This was the first legislation of its kind to be passed in North America and it was received positively by the public health community [5]. Starting in 2008, many other Canadian provinces introduced poverty reduction strategies or action plans [6]. There will be a need for monitoring and evaluating the impact of such policies and strategies on health. Acknowledging the existence of a social gradient in health (that goes beyond the dichotomy of the richest and the poorest), we deemed it important to assess different measures of economic position.

Socioeconomic position is a multidimensional concept, most commonly operationalised in health inequalities research by education, income, and employment-based social class or social status [7]. While some early research attempted to establish which of these indicators was the most stable or strongest predictor of health, the consensus is now that, although correlated, these measures are not interchangeable [8, 9]. Indeed, it appears that these indicators do not capture the same constructs of material resources, power and prestige [8].

Moreover, each of these dimensions of socioeconomic position has been subjected to studies of the internal consistency of their various indicators. For instance, material resources have been captured through numerous indicators, of which income is only one facet, which may not accurately capture the extent of material deprivation in the population. In this paper, we examine the extent to which a direct, "gold-standard" measure of income overlaps with other measures of economic deprivation or economic exclusion in Canada.

Limitations to income measurement are rarely thoughtfully acknowledged in health research [8]. First, income can be measured in various ways (e.g. through self-report or tax file linkage; using individual or household measures; pre or post-tax; as main source of income, or through all possible sources of income; not to mention all possible transformations and use of percentiles). This contributes to study findings heterogeneity, as different measurement can introduce unmeasured confounding: in a dose-response perspective, decreases in income should lead to poorer health, but if income is measured as individual income from salary, the relationship may be weakened by the fact that household income and wealth can mitigate this negative impact. Furthermore, income is typically a sensitive question in surveys, which leads to high levels of missing cases that are not randomly distributed in the population, and thus often biased at the extremes of the distribution [10]. Finally, and perhaps even more importantly, income might not accurately reflect the material conditions of individuals and households. Indeed, the provision of in-kind social services such as housing assistance and food stamps cannot be captured through income measures [11, 12].

Alternative indicators attempt to more explicitly measure the effective economic hardship or the level of 'well-being' instead of implicitly assuming it from income. For instance, economic hardship and material deprivation [13] are conceptually more proximal measures of effects of poverty than income. Economic hardship measures mainly focuses on the extent of deprivation in possessing goods, accessing services and engaging in certain activities [14, 15]. Indicators of economic hardship include cutbacks on monthly expenditures, inability to pay bills, difficulty in meeting basic needs such as food and shelter. Material deprivation measures are assessing the inability to afford some basic goods, lifestyle or opportunities "to participate in a way identified generally as appropriate in [a given] community" [13]. Interestingly, while these indicators are undoubtedly conceptually linked to income, a number of previous studies have shown that these constructs do not overlap perfectly [16‐19].

Moreover, while there is a wealth of research linking income to health, studies linking material deprivation [20‐25] and economic hardship [26‐31] to health are scant, particularly in the Canadian context.

Thus, in this study, we take advantage of the very rich information and innovative measurement of the Canadian Household Panel Survey - Pilot (CHPS-Pilot) [32] to create an index of economic exclusion based on validated questionnaires of economic hardship and material deprivation, and examine its association with health in Canada.

More specifically, our main study objective will be to determine the extent to which income and an economic exclusion index are overlapping measurements of the same concept. We will pursue this objective by answering the following two research questions:

In our sample of 1, 588 respondents, there were more women (52.1%) than men, and more non-immigrants (79.0%) than immigrants (Table 4). Almost two third of respondents had an educational level higher than high school (63.1%), were employed (77.1%), and were in a couple relationship (74.0%). Regarding health status, 11.1% rated their health status as fair or poor, while 45.7% reported at least one chronic or long-term condition. The 975 households interviewed were composed of single individuals (23.1%), couples without children (27.2%), couples with children (35.2%), and single parents (11.3%).

The means of the three indicators used to create the index were 0.082 (0.165) for the MTD_ratio, 0.081 (0.169) for the FHR_ratio and 1.124 (0.961) for the FP_score. Correlations between these three indicators vary from 0.48 to 0.65 (Table 5). By construction, the mean of the index of economic exclusion is 0, and thus, the higher the index of economic exclusion, the higher the exclusion (positive values) while the lower the index, the lower the exclusion (negative values). Finally, the mean individual and household log-incomes were respectively 1.420 (0.675) and 1.973 (0.696).

As expected, we see that economic exclusion is negatively correlated with both individual and household log-income. While the correlation between economic exclusion and income is relatively low (under 0.4), it was significant in both cases, and slightly stronger for household income, which is conceptually congruent (given that the index was also measured as a household characteristic). These correlation coefficients provide the first indication that income and a more explicit measurement of economic exclusion are not perfectly overlapping constructs.

In Table 6 a positive score of economic exclusion indicates a greater exclusion than the mean, while a negative score indicates lower-than-average exclusion. We can see that exclusion decreased with increasing age and education. It was also lower among the employed, and interestingly among those out of the labor force as well, although the unemployed reported higher levels of exclusion. Similarly, lone parents and immigrants reported significantly higher levels of exclusion. Finally, while exclusion was higher among those who reported poor health, the difference was less significant with those who reported chronic condition.

As mentioned above, the index of economic exclusion we have created was inversely associated with income (either individual or household). However, the negative correlation was low and we can observe some additional distinguishing features. First, while log-income increased with age until age 55 (it then decreased), exclusion 'continuously' decreased with age. Moreover, while males have higher income than women, there was no significant difference regarding economic exclusion. Finally, those with chronic conditions earned significantly less money than those without chronic conditions, although they did not report more economic exclusion.

Nested multilevel logistic regressions have been estimated to contrast the impact of economic exclusion and income on the probability of being in poor health. Table 7 present estimates of odds ratios (OR) and 95% confidence interval (CI) for fixed effects, as well as the variance of the random intercept and the Median Odds Ratio (MOR) at the household level.

We observed that self-rated health varied significantly between households as the empty model (M0) indicates a variance of the intercept of 1.367 and an MOR of 3.05. Moreover, the likelihood-ratio test of this empty model indicated the between-household variance was not equal to zero (p = 0.004): this confirmed the need to use multilevel methods rather than a standard logistic regression.

After adding individual characteristics (M1), the variance was reduced to 0.640 and the MOR to 2.14. Respondents with low levels of education were more likely to report being in poor health. Those OLF were more likely to be in poor health than those employed. These effects persisted in subsequent models, remaining significant even if the associations were weakened. However, age, gender, migration status and the province of residence were not significantly associated with health status. The estimates of other variables were unaffected by the presence or absence of the province of residence. This model indicates that the likelihood of being in poor health increased with increasing economic exclusion (OR = 1.66, 95%CI = [1.35-2.05]).

Models 2 introduced income, measured at the individual (M2a) and household (M2b) levels. As with economic exclusion, report of poor health decrease with increasing individual (OR = 0.46, 95%CI = [0.32-0.66]) and household (OR = 0.492, 95%CI = [0.34-0.70]) income.

Finally, when combining both economic exclusion and individual income (M3a) or household income (M3b), effects of all these variables slightly decreased in magnitude but were still statistically significant (OR being respectively 0.55 [0.39-0.80] for individual income and 1.52 [1.24-1.87] for economic exclusion in the first model, and 0.65 [0.45-0.95] household income and 1.51 [1.22-1.87] for economic exclusion in the second one). Here, Model M3a with both individual income and economic exclusion seems to best explain the variation between clusters and offers the best fit for the data.

Contrasting the index of economic exclusion and income significantly contributes to our understanding of social inequalities in health in Canada. Our analyses indeed show that, while economic exclusion is inversely correlated with both individual and household income, these are not perfectly overlapping constructs. Indeed, not only are these indicators weakly correlated, but they also point to slightly different sociodemographic groups at risk of low income and economic exclusion. Moreover, the respective associations with health are of comparable significance, but when these income and economic exclusion indicators are included together in the same model, they point to independent and cumulative, not redundant effects. In sum, economic exclusion index measures additional aspects that are not completely addressed with income, such as the effective hardship or level of economic 'well-being'. These findings also confirm the validity of the tools and the index in the CPHS-pilot.

Previous work done at Statistics Canada [38] shows that items as well as the level of analysis (household or individual) may differ between countries. However, using a ratio of the number of items experienced to the total number of questions, we were able to create comparable information despite survey and country idiosyncrasies. We argue that this is a fruitful strategy first because it allows us to keep respondents with missing information within the present study (only 23 household with missing information on all items were not considered), and second, because it improves the comparability with other international panel surveys for future analyses.

These results reinforce previous literature suggesting that economic exclusion does not exactly measure the same concept as income does [17]. Yet, while the positive impact of income (both individual and household) on health has been extensively described in the literature, publication on the impact of material deprivation or economic exclusion on health are scant. Previous studies did find that wealth has a positive effect on health [40‐43], while material deprivation [20‐25] and economic hardship [26‐31] have negative effect on self rated health. However, the operationalisation of material deprivation often consisted of income and employment status, or, at best, household goods. Thus, these effects conflate the impact of income with more explicit measures of economic exclusion. We found only few studies comparing income and other measures of economic position in the literature and all found that the impact on health status of (monetary) wealth [40] or economic hardship using a single indicator [30] differed than the one from income. Our study goes further using a more complex and multidimensional index of economic exclusion that is easily replicable in other surveys.

In terms of limitations, we ought to restate that the CHPS-pilot is not a nationally representative survey. Moreover, it is cross-sectional, which restricts the causal implications of our research. Some studies analyzing longitudinal data have shown a dose response relationship between poverty trajectories and health, those persistently poor having worse health outcomes than the non-poor or transient [29, 30, 44]. However, one advantage of our data is that the income information is taken directly from the tax records for a substantial proportion of our sample. Therefore, while the index of economic exclusion may be proned to self-reported bias, it is unlikely that the income tax report for the previous financial year suffers from the same limitations. Of course, this temporal lag may also explain the weak correlation between income and economic exclusion: indeed, it is possible that the latter reflects more current circumstances that did not prevail a year ago. This possibility points of course to the need for longitudinal data examining these processes in a dynamic way. Another explanation that could explain the weak correlation (but not the cumulative effect) between income and economic exclusion is that we had only access to income information before tax, and not after tax. This is a limitation (common to many surveys) in that it does indeed not accurately reflect disposable income. Finally, these processes should be examined not only with income, but also with low income cut-off measures to get closer to the estimation of poverty status.

Objective measures such as income have intuitive appeal. However, we would argue that income is only a rough indicator of lack of economic resources, which limits its usefulness for public policy. Even though concepts such as wealth, material deprivation or economic hardship might better assess this lack of economic resources, it seems crucial to take into account the source of stress stemming from perceived inadequacy of one's available economic resources [45, 46]. Although income is associated with adverse events, this association is often mediated by variety of psychological, social and environmental factors [26, 47, 48]. There is therefore a growing interest to look at more complex and extend concepts and social factors to understand and explain health inequalities. The Commission on Social Determinants of Health and more specifically its Social Exclusion Knowledge Network recently defined a list of social determinants leading to health inequalities [4, 49]. Among them, they highlighted social exclusion and developed a parsimonious and comprehensive framework composed of four dimensions: economic, social, political and cultural. However, much remains to be done to operationalise these four dimensions adequately and comparably between surveys and countries. One of the main strengths of this framework is its flexibility for identifying exclusion in one dimension (e.g. economically excluded) but maybe not in another (e.g. socially included). Thus, it proposes that it is the sum of these dimensions that leads to greater levels of social exclusion. Another advantage is the depiction of inclusion to exclusion as a continuum rather than a dichotomous scale. This work highlights the need and the importance of measuring multiple dimensions of economic exclusion in panel surveys, and suggests that we might learn even more by examining the four dimensions of social exclusion proposed by the Social Exclusion Knowledge Network. Indeed, the extension of the measures available in the CHPS-Pilot to the four dimensions of economic, cultural, social and political exclusion could highlight fruitful avenues for future public health, social and intersectorial policies.

The contribution of this paper was to explicitly distinguish, both conceptually and empirically, between income and economic exclusion, one of the main dimensions of social exclusion. Thus, we show that, in the Canadian context, different groups can be at risk of lower income and of economic exclusion. Moreover, while an inverse correlation of economic exclusion with income exists, it is weak, and suggests that these variables measure different concepts. The independence of these two measures demonstrates construct validity of our index.

For instance, associations differed by gender and age. While men tended to report both higher individual income, there was no significant gender difference in the index of economic exclusion. This difference might highlight the increasing empowerment of women in Canada despite persistent income discrepancies. While income increases up to a certain age and then decreases, economic exclusion appeared to linearly decrease with increasing age, probably as the result of accumulation of goods. Thus, even though they report lower incomes, women and people over 55 report a good level of economic integration. These differences may be also due to the positive impact of social transfers and social policies more generally in Canada. Social policies could help people with low income face economic hardship in their daily life. As an additional example, those who reported a chronic condition had lower incomes, but did not appear to face significantly higher economic exclusion. As argued by Gannon et al. [50], this could be a consequence of accessibility policies favoring the inclusion of disabled individuals.

However, despite these comforting findings, we did also identify groups that were vulnerable to both lower income and economic exclusion, and already known to be at risk of social exclusion [51, 52]: those with poor self-rated health (though not necessarily - yet- diagnosed with a chronic condition), the unemployed and immigrants. This observation highlights the need for a better understanding of the interplay between income, social policies and the risk of economic exclusion.