A comparison of health and socioeconomic gradients in health between the United States and Canada

Abstract

Data from the early 2000s indicated worse overall health and larger socioeconomic (SES) health inequalities in the U.S. than in Canada. Yet, sociopolitical contexts, health levels, and SES-health inequalities have changed in both countries during the intervening two decades. Drawing on new data, we update the comparison of health levels and SES-health gradients between the two countries.

Analyses, focused on self-rated health, are based on two complementary sets of data sources: Resilience and Recovery (RR) data, a harmonized U.S.-Canada survey of social conditions collected in 2020 (N = 3743); and a pair of leading nationally representative health data sources from each country: the National Health Interview Surveys (NHIS, N = 104,027) and the Canadian Community Health Survey (CCHS, N = 97,605), both collected in 2017–2018. Health levels and disparities, net of demographic and socioeconomic covariates, were estimated using modified Poisson models for relative comparisons; descriptives and predicted levels of fair/poor health show the comparisons from absolute perspective.

Both data sources show that U.S. adults continue to have significantly worse health than Canadians; the disadvantage may be due to SES differences between the two populations. However, the two data sources yield conflicting findings on SES-health inequalities: the RR data indicate no difference between the two countries in socioeconomic health gradients, while the NHIS/CCHS data show a significantly steeper gradient in the U.S. than in Canada for both education and income.

Canadian adults continue to report better health than their U.S. peers, but it is unclear whether health inequalities remain smaller as well. We discuss potential reasons for the conflicting findings and call for a large new cross-national data collection, which will enable scholars and policymakers to better understand health and wellbeing in the U.S. and Canadian contexts.

Comparing health levels and gradients between countries is a powerful lens that can reveal how social context shapes the wellbeing of populations. The comparative lens is particularly fruitful for peer countries such as the United States (the U.S.) and Canada because they are similar in numerous structural and cultural aspects, yet different in others. For instance, both are high-income countries classified as Liberal Economies (Esping-Andersen, 1990; Hall and Soskice, 2001; Huber and Stephens, 2001), with racially diverse populations that predominantly use the English language. At the same time, the two countries differ in social support and welfare orientation toward their citizens (Siddiqi et al., 2013b), with particularly marked differences in educational and health care systems (Aspalter, 2017), both integral to population health.

Two decades ago, the U.S. Centers for Disease Control and Prevention (CDC) and Statistics Canada (2004) jointly developed an important and unique survey, the Joint Canada/United States Survey of Health (JCUSH), with the goal of facilitating the comparison of health and wellbeing between the two countries. The dataset engendered a surge of studies that examined varied dimensions of health and health inequality comparisons between the two countries. By and large, this JCUSH-based body of research concluded that Canadians had better, or at least no worse, health than U.S. adults across a range of measures such as disability, mobility limitations, health-related quality of life, chronic conditions, cognitive impairment, obesity, as well as self-rated health (Altman and Gulley, 2009; Feeny et al., 2010; Huguet et al., 2008; Lasser et al., 2006; McGrail et al., 2009; Prus, 2011; Prus et al., 2010; Sanmartin et al., 2006; Siddiqi et al., 2013b, 2015).

More specifically, the studies found that the health of Canadians was better overall because low-socioeconomic status (SES) Americans had significantly worse health than their low-SES Canadian peers. That is, the SES-health inequalities were greater in the U.S. than in Canada, a difference noted across a range of SES measures and health outcomes. For instance, educational inequalities were larger in the U.S. in self-rated health (Prus, 2011), health-related quality of life (McGrail et al., 2009), and obesity (Siddiqi et al., 2015), as were income inequalities in self-rated health and physical limitations (Sanmartin et al., 2006), health-related quality of life (Huguet et al., 2008; McGrail et al., 2009), oral health (Farmer et al., 2016; Mejia et al., 2018), and obesity (Siddiqi et al., 2015). Some studies suggested comparable gradients with respect to education (Yamashita and Kunkel, 2015) or even steeper income gradients in self-rated health in Canada (Prus, 2011) but these comparisons were only described, not statistically tested. Thus, exceptions notwithstanding, the bulk of the evidence pointed to larger SES inequalities in the U.S. in the early 2000s, suggesting a socio-political context that failed to protect the vulnerable from health-eroding stressors of disadvantage, at least when compared to the prevailing context in Canada.

The profusion of studies that used the 2002/2003 JCUSH data highlights a pent-up demand for a better understanding of health and health inequality comparisons between the U.S. and Canada. Over the past 20 years since the release of JCUSH, no comparable data collection has been undertaken by the CDC, Statistics Canada, or other agencies. In the absence of cross-national data, some scholars used separate U.S. and Canadian datasets for indirect comparison (Farmer et al., 2016; Lebrun and LaVeist, 2011; Pylypchuk and Sarpong, 2013; Ramraj et al., 2016). These studies employed the National Health Interview Survey (NHIS) and the Canadian Community Health Survey (CCHS) as the leading sources of information on population health in each country. While this approach has notable strengths, such as the large sample sizes and excellent representativeness, it also has important weaknesses, such as measurement differences between the independent data collection efforts (Ólafsdóttir and Beckfield, 2020). Moreover, the literature petered out over time, perhaps as a function of unavailable data; we are not aware of any comparative studies with data collected more recently than 2010. As a result, existing knowledge is based on data collected two decades ago in the JCUSH, or slightly more recent data from independent U.S. and Canadian health surveys.

Our study brings the comparative U.S.-Canada research up to date. Drawing on multiple data sources, we compare health levels and health inequalities between the U.S. and Canada. We capture health using self-rated health (SRH) as a well-validated measure of population health used widely in both countries (Santos-Lozada, 2022; Veenstra and Vanzella-Yang, 2020), as well as in comparative research more broadly (Krueger et al., 2015). Further, we examine two measures of socioeconomic status–educational attainment and household income– to obtain a broader evidence base. We also present all comparisons from both absolute and relative perspectives. Our study thus provides a comprehensive, up-to-date portrait of the health levels and inequalities in the two neighboring countries that reflect how their contexts shape their residents’ health. We conclude with an in-depth critical assessment of the available data, highlighting the urgent need for a new large joint health survey in the two neighboring countries.

1. Methods

1.1. Data

The analyses draw on three data sources. The first one is a cross-sectional, cross-national U.S./Canada online survey “Recovery and Resilience: COVID-19” ([RR], Stephenson et al., 2020). The RR dataset was designed in May–June 2020 by a large interdisciplinary team of investigators at The University of Western Ontario and administered by Leger Opinion Market Research Company in August 2020. The respondents were recruited from a large pool of about 500,000 adults geographically distributed across the U.S. and Canada, who were participating in an ongoing Leger Opinion Panel. Most of the selected respondents received a single email invitation to participate. Sampling weights provided by Leger render the sample representative of the respective populations with respect to age, sex, and region of residence. The response rate was 17% for Canada and 25% for the U.S.; we discuss the implications of the response rate and the nature of the underlying Leger Panel in the discussion section. The dataset and documentation are available at https://dataverse.scholarsportal.info/dataset.xhtml?persistentId=doi:10.5683/SP3/5QHKJE. Informed consent was obtained from all participants involved in the data collection, and all procedures were in accordance with the ethical standards of the institutional and national research committees, as certified by the approval of the Ethics Board of the University of Western Ontario (Project ID 116046).

Second, we used a pair of surveys, the National Health Interview Survey (NHIS) for the U.S. and the Canadian Community Health Survey (CCHS) for Canada. The NHIS and CCHS are both large, nationally representative, repeated cross-sectional surveys, administered by each country’s federal statistical agency (the U.S. Census Bureau and Statistics Canada, respectively). These data are the leading sources of information about each country’s population health and health determinants. We accessed the NHIS data via the IPUMS-harmonized version, the appropriate source given that we are using information pooled from two survey years. Detailed information on the sampling design of the NHIS and the IPUMS is available elsewhere (IPUMS, n.d.; National Center for Health Statistics, 2017). We accessed the CCHS data via the Odesi data portal (https://search1.odesi.ca/#/); full documentation is available there as well (Statistics Canada, 2022). We used the 2017–2018 waves from both sources because those are the most recent publicly available CCHS data at the time of writing (March 2022).

We included adults 25 years and older. This age threshold was chosen to capture adults with completed education and living in independent households who, thus, can report their own household’s income. Adults with valid self-rated health (SRH) were included in the analyses (99.0% in the RR data, 99.9% in the NHIS data, and 99.8% in the CCHS data provided valid SRH).

Missingness was zero or almost zero on most covariates; a notable exception was income in NHIS, which was missing for 12% of respondents. We conducted complete-case analysis with the NHIS/CCHS. For the RR data, given its modest sample size, we chose to impute missing values; we weighed the loss of statistical power due to dropped cases against potential violations of the underlying assumptions for imputation (Mukaka et al., 2016; Sullivan et al., 2017). These analytic decisions yielded a sample size of N = 3743 for the RR data, divided approximately equally between U.S. and Canadian respondents, N = 104,027 for NHIS, and N = 97,605 for the CCHS data.

1.2. Variables

The outcome variable is self-rated health (SRH). SRH was assessed identically in all data sources with the standard question: “In general, would you say your health is excellent, very good, good, fair, or poor?” We dichotomized SRH as excellent, very good, or good, versus fair or poor, following precedent and enabling comparisons with prior studies (Prus, 2011; Sanmartin et al., 2006).

The main predictors of interest were country of residence (U.S. versus Canada), as well as the two measures of socioeconomic status: education and household income. In the RR data, these were measured identically in both countries. For education, this included six categories: less than high school, high school diploma, some higher education but no degree, AA degree or vocational/technical subbaccalaureate credential, BA degree, and MA or higher degree. Household income was assessed with two prompts. Respondents were first asked to report their pre-tax 2019 total household income, to the nearest thousand dollars. We used this continuous income measure for the 71% of respondents who answered it. Those who did not answer this item were then asked: “We don’t need the exact amount; does your household income fall into one of these broad categories?” Eight categories were displayed: $0, $1–30,000, $30,001–60,000, $60,001–90,000, $90,001–110,000, $110,001–150,000, $150,001–200,000, and more than $200,000. We used the midpoints of these eight income categories and combined them with the continuous income measure. We adjusted for the difference between the currencies using Purchasing Power Parity (Organisation for Economic Co-operation and Development (OECD), 2021; Statistics Canada, 2020) and converted the Canadian incomes to be equal to the U. S. incomes (the Canadian incomes were divided by 1.206).

The CCHS, alas, included only 3 levels of education: less than high school, high school, and any postsecondary education. While the NHIS included a more detailed measure of education, we had to collapse it into the same 3 levels as in the Canadian data for comparability. The CCHS household income variable was available in the following 5 levels: 0–19 k, 20–39 k, 40–59 k, 60–79 k, and 80 k+. The NHIS income was available in the following 4 levels: 0–34 k, 35–74 k, 75–99 k, and 100 k+. Not only are these very coarse categories, but the lower income groups include relatively few respondents while the top ones are much larger, and the thresholds are difficult to line up between the two studies. We chose the following categories: low income (0–19 k and 20–39 k in CCHS and 0–34 k in NHIS), middle (40–59 k and 60–79 k in CCHS and 35–74 k in NHIS), and high (80 k in CCHS, and 75–99 k and 100 k in NHIS). We did not adjust for the currency exchange between USD and CAD because the three broad income levels with different thresholds would be difficult to adjust meaningfully. In the discussion section, we explain how these coarse levels may impact findings and their interpretation.

Covariates included four demographic variables conceptualized as potential confounders of the relationship between the country of residence and/or socioeconomic status, and self-rated health: age (in five 10-year categories from 25 to 34 [reference] to 65+), gender (male as reference), immigrant status (native-born as reference versus immigrant), and race/ethnicity. In the RR data, race/ethnicity (hereafter referred to as race) was categorized as White (reference), Black, Hispanic, Asian, Native American or Indigenous, and ‘Other’. Alas, the CCHS only categorized race as white versus visible minority and Indigenous, constraining us to a dichotomy of white (reference) versus not white in the NHIS/CCHS-based analytic steps. In addition, we also controlled for two basic demographic characteristics: marital status (married/partnered as reference, previously married, never married) and parental status (has children living in the household versus does not [reference]). Although marital and parental status are not confounders of the education-health association, they may jointly be conceptualized as such for the household income-health association. They also help determine the household type and thus economic demands on the household income. Models without these two variables yielded sub-stantively the same results as those shown below.

1.3. Analysis

Our analytic approach follows four steps. The first two compare the health levels between the U.S. and Canada, while the latter two compare health inequalities (gradients). The steps also jointly show the comparisons in both absolute and relative terms.

First, descriptive statistics show the distribution of fair/poor SRH and all covariates in the target populations across the U.S. and Canada, using both the RR data and the NHIS/CCHS data (Table 1). Table 1 also indicates whether the distributions differ significantly between the two countries, using a design-based F-test, a corrected weighted Pearson chi square statistic appropriate with survey weights.

Table 1.

Characteristics of the target populations in the U.S. and Canada.

	Recovery and Resilience (RR) data			NHIS and CCHS data
	U.S.	Canada	Difference	U.S.	Canada	Difference
Self-rated health			0.029			0.000
Excellent/very good/good	0.79	0.83		0.86	0.87
Fair/poor	0.21	0.17		0.14	0.12
Household income (RR data)			0.000
0–29.9 k	0.31	0.21
30–59.9 k	0.27	0.27
60–89.9 k	0.17	0.27
90–149.9 k	0.18	0.20
150+	0.07	0.05
Household income (NHIS/CCHS)						0.000
0–34 k NHIS, 0–39 k CCHS				0.25	0.19
35–74 k NHIS, 40–49 k CCHS				0.29	0.28
75 k + NHIS, 80 k + CCHS				0.45	0.53
Education (RR data)			0.000
Less than high school	0.03	0.03
High school	0.17	0.12
Some college	0.22	0.17
AA/VTE credential	0.11	0.22
BA	0.28	0.30
MA+	0.19	0.16
Education (NHIS/CCHS)						0.000
Less than high school				0.14	0.12
High school, no postsecondary				0.38	0.21
Any postsecondary schooling				0.48	0.67
Age			0.204			0.511
25–34	0.21	0.18		0.20	0.20
35–44	0.18	0.18		0.19	0.18
45–54	0.18	0.20		0.19	0.19
55–64	0.19	0.20		0.19	0.20
65+	0.23	0.24		0.23	0.23
Female	0.44	0.52	0.000	0.52	0.51	0.021
Non-white				0.22	0.22	0.259
Race/ethnicity			0.000
White	0.77	0.78
Black	0.09	0.02
Hispanic	0.05	0.01
Asian	0.04	0.12
Other	0.04	0.07
Immigrant	0.07	0.20	0.000	0.19	0.28	0.000
Marital status			0.006			0.000
Married/	0.59	0.64		0.59	0.69
Previously married	0.15	0.14		0.22	0.14
Never married	0.26	0.22		0.19	0.17
Has children at home	0.58	0.58	0.897	0.42	0.35	0.000
N	1893	1850		104,027	97,605

Weighted proportions shown, as well as p-values from F-tests of differences in the distributions of all variables between the U.S. and Canada.

Source: Recovery and Resilience (RR) data, N = 3,743, and National Health Interview Survey/Canadian Community Health Survey (NHIS/CCHS, 2017–2018), N = 201,632.

Second, we compare health levels between the U.S. and Canada using the RR and NHIS/CCHS data (Table 2) by estimating Poisson models with robust standard errors (modified Poisson) for fair/poor SRH as a function of the country of residence, net of different covariate blocks. Modified Poisson models are preferred over logistic models because they yield prevalence ratios, which are more easily understood and communicated, especially to non-specialists (Barros and Hirakata, 2003; Tamhane et al., 2016). A footnote to Table 2 lists the covariates used for each of the Models 1–6, starting with bivariate estimation in Model 1 to the fully adjusted Model 6.

Table 2.

Prevalence ratios for fair/poor SRH in the U.S. compared to Canada.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
RR	1.17*	1.19*	1.16*	1.19*	1.07	1.09
	1.02,1.35	1.04,1.38	1.00,1.35	1.02,1.38	0.93,1.24	0.94,1.27
NHIS/CCHS	1.11***	1.12***	1.09***	1.00	1.01	0.96*
	1.07,1.16	1.08,1.16	1.05,1.14	0.96,1.04	0.97,1.05	0.92,1.00

^*p < .05

^**p < .01

^***p < 001.

Source: Recovery and Resilience (RR) data, N = 3743, and National Health Interview Survey/Canadian Community Health Survey (NHIS/CCHS, 2017–2018), N = 201,632.

The table displays prevalence ratios for fair/poor health (versus excellent/very good/good) in the U.S. compared with Canada, as exponentiated coefficients and their 95% confidence intervals from six separate weighted Poisson models with robust variance for each data source. The covariates in each Model are: Model 1 is bivariate (only a dummy for the U.S. relative to Canada, no controls).

Model 2 controls for age and gender.

Model 3 controls for age, gender, race, immigrant, marital, and parental status.

Model 4 controls for age, gender, race, immigrant, marital, parental status, and education.

Model 5 controls for age, gender, race, immigrant, marital, parental status, and household income.

Model 6 controls for age, gender, race, immigrant, marital, parental status, education, and household income

Next, we compare health inequalities (gradients) in the U.S. and Canada, providing the answers in relative and absolute terms. In the third analytic step, we test the equality of the gradients by estimating interaction models of fair/poor SRH (Table 3a for RR data and 3b for NHIS/CCHS data). These models include an indicator of SES (education or income), an indicator for country (Canada), an interaction between SES and country, as well as age, gender, race, and immigrant status. The interaction terms offer formal tests of differences in SES-health inequalities between the two countries. In the fourth and final step, we calculate average predicted probabilities of fair/poor SRH in each country at each level of education and income, respectively, as well as the 95% confidence intervals of the predicted probabilities. We display the results in Fig. 1 for the RR data and Fig. 2for the NHIS/CCHS data.

Table 3a.

Prevalence ratios for fair/poor SRH in the U.S. compared to Canada estimated with interaction terms between SES and country using the Recovery and Resilience (RR) data.

	Prev. ratio	95% CI	Prev. ratio	95% CI
Education (postgraduate degree = ref)
Less than high school	1.91	0.94,3.86
High school	2.95***	2.00,4.35
Some postsecondary	2.50***	1.70,3.69
AA (subbaccalaureate degree)	2.69***	1.75,4.14
BA degree	1.45	0.96,2.21
Interactions
Less than high school * Canada	1.62	0.67,3.88
High school * Canada	0.76	0.43,1.33
Some postsecondary * Canada	0.82	0.47,1.41
AA (subbaccalaureate) * Canada	0.69	0.39,1.22
BA degree * Canada	0.78	0.44,1.40
Household income (150 k+ = ref)
0–29.9 k			4.56***	2.24,9.31
30–59.9 k			2.63**	1.27,5.45
60–89.9 k			1.70	0.79,3.66
90–149.9 k			1.10	0.49,2.51
Interactions
0–29.9 k * Canada			1.19	0.41,3.47
30–59.9 k * Canada			1.32	0.45,3.91
60–89.9 k * Canada			1.13	0.37,3.49
90–149.9 k * Canada			1.52	0.47,4.94

^*p < .05

^**p < .01

^***p < 001.

Source: Recovery and Resilience (RR) data, N = 3743.

The table displays prevalence ratios for fair/poor health (versus excellent/very good/good) in the U.S. compared with Canada, as exponentiated coefficients and their 95% confidence intervals from weighted Poisson models with robust variance. In addition to the main and interaction effects shown, both models also include an indicator for Canada, age, gender, race, and immigrant status. Models that only control for age and gender, or models that also control for marital status and children, yield comparable results.

Table 3b.

Prevalence ratios for fair/poor SRH in the U.S. compared to Canada estimated with interaction terms between SES and country using the NHIS/CCHS data.

	Prev. ratio	95% CI	Prev. ratio	95% CI
Education (any postsecondary = ref)
Less than high school	3.57***	3.33,3.83
High school	1.92***	1.79,2.05
Interactions
LHS * Canada	0.66***	0.60,0.73
HS * Canada	0.76***	0.69,0.84
Income (highest category = ref)
Income (lowest category)			4.48***	4.12,4.87
Income (middle category)			2.41***	2.20,2.64
Interactions
Lowest category * Canada			0.59***	0.53,0.66
Middle category * Canada			0.67***	0.60,0.76

^*p < .05

^**p < .01

^***p < 001.

Source: National Health Interview Survey/Canadian Community Health Survey (NHIS/CCHS, 2017–2018), N = 201,632.

The table displays prevalence ratios for fair/poor health (versus excellent/very good/good) in the U.S. compared with Canada, as exponentiated coefficients and their 95% confidence intervals from weighted Poisson models with robust variance. In addition to the main and interaction effects shown, both models also include an indicator for Canada, age, gender, race, and immigrant status. Models that only control for age and gender, or models that also control for marital status and children, yield comparable results. For income, the lowest category is 0–34 k in NHIS and 0–39 k in CCHS; the middle category is 35–74 k in NHIS, and 40–49 k in CCHS; the highest category is 75 k+ in NHIS, and 80 k+ in CCHS.

Fig. 1.

Predicted prevalence of poor/fair SRH by socioeconomic indicators in the U.S. and Canada using the Recovery and Resilience (RR) data.

Source: Recovery and Resilience (RR) data, N = 3,743.

Average predicted probabilities for fair/poor SRH calculated from weighted Poisson models with robust variance shown in Table 3. The models include age, gender, race, and immigrant-status-adjusted regression models with interaction terms between an indicator for Canada and each SES measure.

Fig. 2.

Predicted prevalence of poor/fair SRH by socioeconomic indicators in the U.S. and Canada (NHIS/CCHS data).

Source: National Health Interview Survey/Canadian Community Health Survey (NHIS/CCHS 2017–2018), N = 201,632.

Average predicted probabilities for fair/poor SRH calculated from weighted Poisson models with robust variance shown in Table 3b. The models include age, gender, race, and immigrant-status-adjusted regression models with interaction terms between an indicator for Canada and each SES measure, estimated using the pooled U.S. and Canadian samples.

2. Results

Table 1 summarizes and compares the distribution of the outcome and the characteristics of the target U.S. and Canadian populations (i.e., weighted sample descriptives) based on the three data sources. Overall, unfavorable SRH is significantly higher in the U.S. than in Canada (4 percentage points higher in the RR data, p = .029, and 2 percentage points higher in the NHIS/CCHS data, p < .001). The U.S. and Canadian populations also differ in the distributions of both socioeconomic covariates. Household income distributions differ between the two countries, in the direction of higher incomes in Canada: fewer Canadian compared with U.S. respondents report the lowest income in both RR and NHIS/CCHS data, and more Canadians report high incomes in the NHIS/CCHS. As we explained above, the NHIS/CCHS income categories do not match up well, meaning the comparison must be viewed with caution. For education, the NHIS/CCHS shows higher schooling for Canadians (p < .001). However, the highest category available in CCHS was “any post-secondary education,” which is too broad for meaningful comparisons: 67% of Canadians and nearly half of Americans fall into this category. Moreover, as the RR data show, the distribution of the actual post-secondary attainment differs in complex ways between the two countries: for instance, more Americans than Canadians completed no degree (“some college”), while more Canadians than Americans earned a subbaccalaureate credential. With respect to demographics, the two populations do not differ in the distribution of age, but do differ in the distributions of gender, immigrant status, and marital status. Racial composition differs in the RR data where major non-white groups can be disaggregated; alas, the NHIS/CCHS comparison obscures potential differences between the two countries because the CCHS effectively only allows us to examine dichotomized race and the proportion of white respondents is the same (78%) in both countries. Jointly, the descriptives thus point to worse health and economic wellbeing in the U.S. population when compared to Canadians.

The next question is whether the worse SRH in the U.S. may be linked to the differences in socioeconomic characteristics. Table 2 summarizes the answer in relative terms. The unadjusted prevalence of fair/poor SRH is significantly higher in the U.S. compared to Canada (17% higher based on the RR data, p < .05; 11% higher based on the NHIS/CCHS data, p < .001). Controlling for demographics (age and gender in Model 2, and age, gender, race, immigrant, marital, and parental status in Model 3) has relatively little impact on the U.S./Canada prevalence ratios. Controlling for education in Model 4 does not attenuate the U.S. disadvantage in the RR data (PR = 1.19, p < .05); but in the NHIS/CCHS data, the U.S./Canada difference becomes null (PR = 1.00). Controlling for income in Model 5 renders the U.S. SRH disadvantage statistically null using both data sources (PR = 1.07 with RR data; PR = 1.01 with NHIS/CCHS data, both p-values>.05). Finally, net of all covariates in Model 6, the U.S. and Canada have either statistically comparable SRH (based on the RR data), or the U.S. has a 4% lower prevalence of fair/poor SRH (PR = 0.96, p < .05). That is, household income and perhaps educational attainment differences appear to account for the higher prevalence of fair/poor SRH in the U.S.

Thus far, the analyses focused on health levels in the two populations. The next step was to compare health inequalities (gradients). We present the findings in relative terms using interaction models, and in absolute terms by visualizing the average predicted probabilities of fair/poor SRH by each level of education and household income. The findings in this analytic step differ between the RR and the NHIS/CCHS data.

The RR data (results in Table 3a) indicate no statistically significant differences in the educational or income gradients between the U.S. and Canada: none of the 9 interaction terms is statistically significant. The direction of the effects suggests a steeper gradient in Canada for education but a flatter gradient for household income, compared with the U. S. In contrast, the results using the NHIS/CCHS data strongly and consistently indicate steeper gradients in the U.S. with respect to both education and income: all four interaction terms shown in Table 3b are statistically significant and substantively large.

Finally, we visualize the inequalities in each country by plotting the average predicted probabilities of fair/poor health at each level of education and income, respectively. Fig. 1 shows results from the RR data: the graphs show very similar SRH at each income level. For education, the patterns are more complex but the confidence intervals overlap at each schooling level and overall do not tend toward a steeper gradient in either country. Fig. 2 shows parallel findings from the NHIS/CCHS data. The patterns corroborate the relative results summarized in Table 3b: the Figure highlights the steeper gradients, that is, greater health inequalities, in the U.S. compared with Canada with respect to both education and income.

3. Discussion

Comparing health levels and health inequalities between countries can help reveal how sociopolitical contexts impact population health and wellbeing. Twenty years ago, the Joint Canada-United States Health Survey (JCUSH) engendered a body of research that revealed worse health and larger health inequalities in the United States that in Canada, suggesting that Canada’s social welfare orientation or other contextual factors may have protected its most vulnerable adults, at least when compared to the U.S. During the intervening two decades, however, both countries experienced changes in their sociopolitical contexts, including growth in socioeconomic inequalities (Bor et al., 2017; Osberg, 2018) and, correspondingly, health inequalities (Bushnik et al., 2020). Despite these changes, the literature lacks up-to-date comparisons in health or SES-health inequalities between the U.S. and Canada, or how they changed over the past twenty years.

As such, we updated the comparative literature with analyses using complementary data sources: the 2020 Recovery and Resilience (RR) data, a U.S./Canada survey developed specifically for a comparative analysis of the two countries’ contexts, and a pair of the largest leading sources of health information in each country: the National Health Interview Survey (NHIS) in the U.S. and the Canadian Community Health Survey (CCHS), both from 2017/2018. Using self-rated health as a well-validated measure of population health (Bombak, 2013; Schnittker and Bacak, 2014), we found a continued Canadian advantage in health levels, but our findings on health inequalities diverged between the two data sources.

The proportion of the population who reported their health to be only fair or poor, as opposed to excellent, very good, or good, was significantly higher among American respondents, by either 2 or 4 percentage points (pp), depending on the data source. While this difference may appear modest, even the lower estimate of 2 pp translates to 4.5 million additional Americans aged 25 and older in bad health, compared to how many there would be if U.S. health were equivalent to that of their Canadian counterparts (estimate based on 225 million U.S. adults aged 25 and older, U.S. Census Bureau, 2020). Our findings corroborate prior estimates derived from the JCUSH data, which reported 2.7 pp higher prevalence of fair/poor health in the U.S. among adults 18 and older as compared to Canada (Prus, 2011; Sanmartin et al., 2006). This suggests that the well-documented health advantage in Canada over the U.S. has persisted over time.

The U.S. health disadvantage was largely due to SES differences between the two populations. Based on both sources of data, if the distribution of socioeconomic characteristics were comparable between the two countries, their general health would be comparable as well. In fact, the NHIS/CCHS analysis even suggested that net of education and income, the prevalence of fair/poor health is marginally higher in Canada. This is an intriguing finding in its own right, and would benefit from additional investigation in the future.

Our analysis of health inequalities (gradients) yielded conflicting findings across the two sources of data. The RR data showed effectively identical inequalities in the U.S. and Canadian populations. Whether for education or income, and whether analyzed in relative or absolute terms, there was no discernible difference in health inequalities between the two countries: the prevalence of fair/poor health were statistically the same at every SES level in both contexts. The NHIS/CCHS data, in contrast, yielded results similar to those from twenty years ago: a significantly and substantially steeper socioeconomic health gradient in the U.S. It appears that the steeper U.S. gradient in more recent times derives from both ends of the SES spectrum: high-SES U.S. adults had better health than their high-SES Canadian peers, and low-SES U.S. adults had worse health than low-SES Canadians. This pattern corroborates some prior work (Huguet et al., 2008), although others found that the U.S.-Canada difference derives primarily from large differences at the low end of the SES spectrum (Siddiqi et al., 2015).

The contradictory findings are likely a function of the weaknesses of each data source, alongside their strengths. The RR data was collected recently, in 2020, with the explicit purpose of cross-national comparisons and thus harmonized across both countries, and included detailed SES measures necessary for describing the inequalities adequately. However, the samples in both countries skewed toward the more advantaged: The respondents needed to have the ability and willingness to complete the online questionnaire during the challenging early months of the COVID-19 pandemic. The data thus undersampled rural, non-white, low-SES individuals in both countries. It is possible that the selection into either the underlying Leger Opinion Panel respondent pool and/or into our survey yielded samples that were more alike between the two countries than the target populations. This concern is alleviated with the NHIS/CCHS data, which are closely representative of the U.S. and Canadian populations, respectively. These data, however, have their own weaknesses. For one, the most recent public-use CCHS data available at the time of this writing were collected in 2017–2018, or nearly half a decade ago. If the SES-health gradients have been converging over time, as suggested by the RR results as well as other studies (Baker et al., 2019), this five-year lag could yield larger differences than we would observe with up-to-date data. More importantly, the data were not harmonized since the two surveys were never designed to be comparable, meaning the SES categories were difficult to align meaningfully. This issue is exacerbated by the exceedingly coarse categorization of educational attainment in CCHS and income in both datasets. The top educational attainment category in CCHS, ‘any post-secondary education,’ comprises two thirds of Canadians and half of Americans. Thus, any estimates must be viewed with caution because this gross category obscures tremendous variability within. In fact, the major health threshold in both countries is now college completion versus not (Case and Deaton, 2021; Zajacova and Jehn, 2020; Zajacova and Lawrence, 2021), but the NHIS/CCHS data aggregates subbaccalaureate attainment and college completion into a single category. Similarly coarse categories are for household income, allowing only three income groups where, moreover, the top income category comprises nearly half of U.S. adults and over half of Canadians.

The similarity of SES-health gradients we identified with the RR data is new and contrary to the sizeable prior literature. While, as we stated above, the results may be a function of non-representative samples, it is also possible that the findings are a true reflection of population health trends, and particularly the impact of conditions experienced during the first summer of the COVID-19 pandemic. In other words, perhaps the stressors of COVID-19 impacted the gradients in both countries such that, under the resulting extreme stress and economic upheaval, the two populations converged toward similar levels of SES-health gradients. This explanation is not extremely likely, however, as it would require that the impact of the pandemic blunt the inequalities in the U.S., a finding that is contrary to all COVID-related research to date (Dorn et al., 2020; Perry et al., 2021).

There are additional caveats to keep in mind for our study. The self-rated health measure is one of the leading health indicators used in population research thanks to its high reliability, as well as concurrent and predictive validity (Kananen et al., 2021; Zajacova et al., 2017). At the same time, it isn’t fully understood how respondents generate the health rating (Au and Johnston, 2014) and different groups of respondents may form health judgments in distinct ways (Layes et al., 2012). The adverse conditions during the COVID-19 pandemic, further, could also influence health judgements. All these factors could potentially impact the comparison of health levels or SES-health inequalities.

Finally, as with any observational study, it is important to keep in mind that residual confounding might influence the results. For instance, while we included a range of demographic covariates, each was measured and modelled out of necessity in relatively crude terms, which may distort their role in the analyses. As an example, the impact of race may vary by immigrant generation, which we were unable to examine. It is also possible that we excluded important variables entirely, such as family wealth, because they were unavailable in our datasets. These and other examples represent possible sources of residual confounding, which should be considered when interpreting our study findings.

Despite these limitations, the analysis contributed to our understanding of health and health inequality differences between the U.S. and Canada, or at the very least has given us reason to question whether earlier cross-national health differences still hold. The data corroborated older findings with respect to health levels: the Canadian health advantage persisted, at least with respect to self-rated health. The second question, which pertained to the comparison of health inequalities (gradients), yielded inconsistent results. Based on the RR data, inequalities have converged between the two countries. Based on the NHIS/CCHS analysis, however, the U.S. still has steeper SES-health gradients. Interestingly, this appears to be a product of both worse health at the low end of the SES distribution, and better health at the high end, in the U.S. versus in Canada.

The contradictory findings on a very important question highlight the urgent need for an updated data collection similar to the JCUSH dataset. We urge the Centers for Disease Control and Statistics Canada to plan a new data collection. This new harmonized data should include a detailed education and income categorization, which should adequately capture the population distributions, with detailed postsecondary levels and finer income categories at the upper tail. In addition, the data should include occupation as an additional measure of SES, as well as a range of health outcomes in addition to SRH, such as chronic pain, mental health, chronic conditions, functional limitations, and disability. If these new data corroborated the RR findings of similar gradients, Canada would have to reckon with the impact that neoliberal encroachment may have had on its population health (Siddiqi et al., 2013a), while both countries need to explore political and policy avenues to improve the health of their residents and decrease health inequalities. We hope our findings fuel not only the necessary new data collection, but also, more broadly, debates about health, economic, and social policies, which are particularly critical as both countries plan for recovery from the COVID-19 pandemic.