1 Introduction

This report investigates the association between socioeconomic and demographic characteristics and healthy behaviours among the Canadian population, drawing on data from the 2022 Canadian Community Health Survey (CCHS). The study begins by detailing the conceptual and methodological framework for constructing the Healthy Behaviour Index (HBI), which serves as a composite proxy for overall health-related behaviours. A descriptive analysis of the index is then presented, followed by the application of econometric and statistical models to examine the underlying relationships.

2 Healthy Behaviour Index (HBI): Construction, Rationale, and Limitations

This analysis develops a Healthy Behaviour Index (HBI) to represent overall health-related behaviours among the Canadian population using data from the 2022 Canadian Community Health Survey (CCHS). Previous literature typically constructs the HBI or similar health behaviour scores using four indicators: (1) smoking status, (2) drinking status, (3) physical activity, and (4) fruit and vegetable consumption (Poortinga 2007; Schuit et al. 2002). However, the 2022 CCHS did not include modules on physical activity or fruit and vegetable consumption. Consequently, this study substitutes two alternative indicators—flu shot status and body mass index (BMI)—alongside smoking and drinking status. Details of these four indicators are presented in Table 1.

2.1 Rationale for Substitution

Flu shot status was chosen as a proxy for preventive health behaviour, reflecting adherence to public health recommendations (Centers for Disease Control and Prevention 2023). BMI, while not a direct behaviour, is widely used in epidemiological research as an outcome strongly associated with diet and physical activity patterns (World Health Organization 2021). These substitutions ensure the index captures multiple dimensions of health-related behaviour—risk avoidance (smoking, drinking), preventive care (vaccination), and lifestyle outcomes (BMI).

2.2 Method of Construction

Each indicator is dichotomized into healthy (=1) or unhealthy (=0) behaviour. The HBI is computed as the sum of these binary indicators, producing a composite score from 0 to 3. For interpretability, the HBI is categorized into three ordinal groups:
- 1 = poor healthy behaviour (score = 0–1)
- 2 = fair healthy behaviour (score = 2)
- 3 = good healthy behaviour (score = 3-4)

Thus, the constructed HBI provides a simplified ordinal measure of overall healthy behaviour and facilitates both descriptive analysis and regression modeling.

2.3 Limitations

  • Reduced behavioural coverage: Substituting flu shot and BMI alters the behavioural domains compared to traditional indices, limiting comparability across studies (Poortinga 2007).
  • Proxy bias: BMI conflates body composition and lifestyle factors, potentially misclassifying individuals (e.g., athletes vs. sedentary individuals) (World Health Organization 2021).
  • Self-reported data: Smoking, drinking, and vaccination data are subject to recall and social desirability bias.
  • Ordinal categorization: Collapsing scores into three categories may obscure finer behavioural differences.

Despite these limitations, the adapted HBI provides a practical and conceptually valid measure of health-related behaviour using available survey data.

Table 1: Indicators of the Healthy Behaviours Index or Score, negative and positive health behaviours
Health indicator Negative health behaviour Positive health behaviour
Smoking status Current daily or occasional smoker Current non-smoker
Drinking status Heavy drinker in the past 12 months1 Not a heavy drinker in the past 12 months
Flu shot status No seasonal flu vaccine in the past 12 months Seasonal flu vaccine in the past 12 months
Body Mass Index (BMI) Overweight / Obese — Class I, II, III Underweight / Normal weight2
Note:
1 Heavy drinkers are men who consumed 5 or more drinks per occasion, at least once a month in the past year. Women are heavy drinkers if they consumed 4 or more drinks per occasion, at least once a month in the past year. Source: https://www150.statcan.gc.ca/n1/pub/82-625-x/2018001/article/54975-eng.htm; https://www150.statcan.gc.ca/n1/pub/82-229-x/2009001/deter/int3-eng.htm
2 Physical activity and fruit/vegetable consumption were not included in the 2022 CCHS survey for provinces; therefore, flu shot status and BMI are used as alternative healthy behaviour indicators. See: https://academic.oup.com/abm/article/54/10/783/5828146; https://stacks.cdc.gov/view/cdc/207738/cdc_207738_DS1.pdf

Figure 1 illustrates the provincial distribution of the Healthy Behaviour Index (HBI) across Canada, based on the composite measure described earlier. The analysis shows that British Columbia (58%), Quebec (56%), and Ontario (54%) have the highest prevalence of healthy behaviours, suggesting stronger adherence to recommended lifestyle practices in these regions. In contrast, Yukon (39%), Newfoundland and Labrador (45%), and Prince Edward Island (47%) report the lowest proportions, indicating potential disparities in health-promoting practices. Notably, these same provinces—Yukon (20%), PEI (19%), and Newfoundland and Labrador (14%)—also exhibit the highest prevalence of poor health behaviours, pointing to a more polarized distribution of health practices within these populations.

These provincial variations reflect well-established socioeconomic determinants of health. Nationwide, higher-income and urban populations display greater engagement in behaviours such as regular physical activity, healthy eating, and preventive care, compared to lower-income or rural populations (Statistics Canada 2024). Moreover, socioeconomic status—including income, education, and employment—strongly predicts key health behaviours such as smoking, physical inactivity, and alcohol consumption in Canada (Chai, Tan, and Dong 2024). Similarly, unhealthy behaviours like smoking, poor diet, and physical inactivity, particularly among socioeconomically disadvantaged Ontarians, contribute substantially to healthcare expenditures and reflect socioeconomic barriers to healthy lifestyles (Manuel et al. 2019).

Overall, the findings reveal a pronounced geographic and socioeconomic gradient in healthy behaviour across Canada, underscoring the pivotal role of structural determinants—such as income, education, and access to health resources—in enabling or constraining health-promoting choices.

*Future research: Does Healthy behavior automatically ensure good health - compare rural vs urban?

Provincial distribution of HBI in Canada

Figure 1: Provincial distribution of HBI in Canada

3 Empirical Strategy for Investigating Socioeconomic Influences on Health Behaviors

I estimate the association between socioeconomic and demographic characteristics and healthy behaviour using an ordered logit regression model. This approach is appropriate because the dependent variable—the Healthy Behaviour Index (HBI)—is an ordinal measure with three ordered categories, ranging from 1 (poor behaviour) to 3 (good behaviour). The general specification of the ordered logit model applied in this analysis is as follows: Assume Y be HBI ordinal outcome with J (three) categories. Then \({P(Y \leq j)}\) is the cumulative probability of Y less than or equal to a specific category j=1, 2 (J-1). The odds of being less than or equal to a particular category j can be defined as \(\frac{P(Y \leq j)}{P(Y > j)}\) for j=1, 2 (J-1). The log odds is also known as the logit, can be written as \(\log \left( \frac{P(Y \leq j)}{P(Y > j)}\right)\) = \(logit [P(Y \leq j)]\). The ordered logit regression model can be written as

check this: https://stats.oarc.ucla.edu/other/mult-pkg/faq/ologit/ https://bookdown.org/chua/ber642_advanced_regression/ordinal-logistic-regression.html

\[\begin{equation} logit [P(Y \leq j)] = \beta_0 + \beta_1 X_1 +.....+ \beta_k X_k \tag{1} \end{equation}\]

For simplicity of interpretation of the results, we estimate the probability of objerving outcome j obtained from the logit model. The probability of observing outcome i corresponds to the probability that the estimated linear function, plus random error of the regression, is within the range of the cutpoints estimated for the outcome:

\[\begin{equation} Pr(outcome_j=i) = Pr(c_{i-1} < \beta_1 X_{1j} +.....+ \beta_k X_{kj} + u_j \leq c_i) \tag{2} \end{equation}\]

and the associated marginal effect is

\[\begin{equation} \displaystyle \frac{\partial Pr(outcome_j=i)}{\partial X_k} =\frac{\partial Pr(c_{i-1} < \beta_1 X_{1j} +.....+ \beta_k X_{kj} + u_j \leq c_i)} {\partial X_k} \tag{3} \end{equation}\]

Where:

The coefficients (( \(\beta\) )) in equation (1) show the ordered log-odds (logit) regression coefficients, which can be interpreted as the change in the ordered log-odds scale of the dependent or outcome variable from a one-unit increase in the explanatory variable or predictor, holding every other variable in the model are constant.

The marginal effect in equation (3) can be interpreted as one unit increases in raises the probability of being towards healthy behaviour from poor behaviour by some percentage.

4 Results and Discussion

4.1 Interpretation of regression coefficients

Results of the regression are shown in Table 2. The table reports the estimates of both coefficients (log-odds ratio) and odds ratio along with respective standard errors (S.E) and 95% confidence intervals. The Standard interpretation of the ordered logit coefficient is that for a small change in the predictor, the response variable level is expected to change by its respective regression coefficient in the ordered log-odds scale while the other variables in the model are held constant. Interpretation of log-odds are not intuitive because they are logarithms of ratios. What matters most is the direction of the coefficients. Below I provide interpretation of coefficients in terms of direction (increase or decrease) of changes.

  • Sex or Gender

The model shows that males have a log‑odds coefficient of –0.410 relative to females, indicating that men are substantially less likely to fall into a healthier behaviour category. This negative association is both sizeable and highly significant (p < 0.001), suggesting a robust difference between sexes. The direction and magnitude of the estimate are consistent with patterns often observed in health behaviour research, where men tend to engage less in preventive practices or adopt riskier lifestyle habits, which may explain their lower predicted likelihood of healthier behaviour in this model.

  • Age group

The estimates show a clear age‑related gradient in healthier behaviour. Compared with adults aged 12–34, those aged 35–49 have a positive log‑odds coefficient of 0.421, indicating moderately higher likelihood of being in a healthier behaviour category. This likelihood increases further for the 50–64 group, with a coefficient of 0.665, and becomes especially pronounced among adults 65 and older, who show a large coefficient of 1.497. All three effects are highly significant (p < 0.001), highlighting a robust and consistent pattern: older age groups are progressively more likely to exhibit healthier behavioural profiles, potentially reflecting greater health awareness, stability in lifestyle routines, or accumulated health management experience.

  • Education

The results indicate that respondents with higher levels of education have increased log‑odds of being in a healthier behaviour category compared to those with less than secondary schooling. Individuals with secondary education already show a modest but positive increase in the likelihood of reporting healthier behaviours. This association becomes substantially stronger among those with a university education, who exhibit the highest positive shift in log‑odds relative to the reference group. These findings suggest a clear gradient in health behaviours that improves with higher educational attainment. All estimated coefficients are statistically significant at the 1% level, reinforcing the robustness of these associations.

  • Perceived Mental Health

The findings suggest a clear pattern in which better self‑rated mental health corresponds to healthier behaviour outcomes. Respondents who report their mental health as good show a slight increase in the likelihood of being in a higher health‑behaviour category compared with those experiencing poor or fair mental health. This relationship becomes more pronounced among individuals who describe their mental health as very good or excellent, who display the most notable improvement in log‑odds relative to the reference group. Overall, the results point to a positive link between mental well‑being and healthier behavioural patterns, with the coefficient for the very good/excellent group reaching statistical significance at the 1% level.

  • Household income

The income results reveal a gradual upward trend in healthier behaviour as household income increases. Compared with individuals earning below $20,000, those in the $20,000–$39,999 range show virtually no difference in log‑odds. However, once income reaches $40,000 or more, a clearer positive pattern emerges. Respondents earning between $40,000 and $79,999 display modest increases in the likelihood of healthier behaviours, while those with incomes above $80,000 exhibit the largest positive shift relative to the reference group. Taken together, these findings point to a progressive association between higher income and healthier behavioural outcomes, with the effects for the top three income categories reaching statistical significance.

  • Food Security

Patterns in food security show a clear gradient in health behaviour outcomes. Individuals who are marginally food insecure have noticeably higher log‑odds of being in a healthier behaviour category compared with those who are severely food insecure, and this effect is statistically significant. For those experiencing moderate insecurity, the increase is smaller and not statistically distinguishable from the severely insecure group. In contrast, respondents who are fully food secure demonstrate the highest positive shift in log‑odds, indicating a substantially greater likelihood of reporting healthier behaviours. Overall, the results highlight that greater food security is linked to better health‑related behaviour, with the strongest evidence observed at the secure end of the spectrum.

  • Perceived Social Support

The association between social connection and healthier behaviour does not follow a simple linear pattern. Compared with individuals who report very weak social ties, those with somewhat weak connections show a notable and statistically significant increase in the likelihood of engaging in healthier behaviours. This suggests that even modest improvements in social connection—perhaps enough to reduce feelings of isolation without introducing new obligations—may support healthier routines. In contrast, respondents who describe their connections as somewhat strong display only a slight, non‑significant rise in log‑odds, indicating that moving from “weak” to “somewhat strong” ties may not meaningfully shift behavioural patterns. Interestingly, individuals with very strong social connections exhibit a small but statistically significant decrease relative to the reference group. One possible explanation is that dense or demanding social networks may introduce stress, competing priorities, or social obligations that crowd out time for health‑supportive behaviours. Taken together, the results point to a more complex relationship in which moderate levels of social connection appear beneficial, while both very weak and very strong connections may be associated with less favourable health behaviour patterns.

Overall, the ordered logit results point to a clear pattern: older adults are progressively more likely to fall into healthier behaviour categories than the youngest group; males are less likely than females; and higher educational attainment is associated with better outcomes, with the strongest association at the university level. Income shows little difference at the lowest two brackets but shifts positive from the mid-range upward, with the strongest associations at the highest incomes. Greater food security is linked to healthier behaviours, with clear advantages for those who are secure and a smaller, non‑significant difference for those who are moderately insecure. Better self‑rated mental health aligns with healthier behaviours, especially at the very good/excellent level, whereas “good” alone shows only a minimal, non‑significant improvement. Social connection does not follow a simple gradient: somewhat weak ties are positively associated, somewhat strong ties show little difference, and very strong ties are slightly less favourable compared with very weak ties.

Table 2: Ordered Logit Estimates of Socioeconomic and Demographic Factors Associated with Healthy Behavior Index)
Item β (log-odds) SE (β) 95% CI (β) OR 95% CI (OR) p Sig.
Predictors (β and OR)
sex: Male vs Female -0.410 0.015 -0.441 – -0.380 0.66 0.64 – 0.68 <0.001 ***
age_group: 35-49 vs 12-34 0.421 0.025 0.373 – 0.469 1.52 1.45 – 1.60 <0.001 ***
age_group: 50-64 vs 12-34 0.665 0.023 0.619 – 0.711 1.94 1.86 – 2.04 <0.001 ***
age_group: 65 years & older vs 12-34 1.497 0.024 1.450 – 1.543 4.47 4.26 – 4.68 <0.001 ***
education: Secondary vs Less than secondary 0.192 0.035 0.124 – 0.260 1.21 1.13 – 1.30 <0.001 ***
education: University vs Less than secondary 0.558 0.033 0.494 – 0.622 1.75 1.64 – 1.86 <0.001 ***
mental_health: Good vs Poor/Fair 0.050 0.026 -0.001 – 0.100 1.05 1.00 – 1.11 0.053
mental_health: Very good/Excellent vs Poor/Fair 0.124 0.025 0.074 – 0.173 1.13 1.08 – 1.19 <0.001 ***
income: $20,000-$39,999 vs 0-$20,000 -0.011 0.051 -0.110 – 0.088 0.99 0.90 – 1.09 0.823
income: $40,000-$59,999 vs 0-$20,000 0.100 0.050 0.002 – 0.198 1.10 1.00 – 1.22 0.046
income: $60,000-$79,999 vs 0-$20,000 0.163 0.051 0.064 – 0.262 1.18 1.07 – 1.30 0.001 **
income: >$80,000 vs 0-$20,000 0.224 0.048 0.131 – 0.317 1.25 1.14 – 1.37 <0.001 ***
Immigrant: Immigrant vs Non-immigrant 0.238 0.020 0.198 – 0.278 1.27 1.22 – 1.32 <0.001 ***
Food_security: Marginally insecure vs Severely insecure 0.187 0.054 0.082 – 0.293 1.21 1.08 – 1.34 <0.001 ***
Food_security: Moderately insecure vs Severely insecure 0.075 0.049 -0.021 – 0.170 1.08 0.98 – 1.19 0.126
Food_security: Secure vs Severely insecure 0.439 0.041 0.358 – 0.519 1.55 1.43 – 1.68 <0.001 ***
Social_connection: Somewhat strong vs Very weak 0.046 0.030 -0.013 – 0.105 1.05 0.99 – 1.11 0.126
Social_connection: Somewhat weak vs Very weak 0.151 0.031 0.090 – 0.212 1.16 1.09 – 1.24 <0.001 ***
Social_connection: Very strong vs Very weak -0.072 0.034 -0.138 – -0.005 0.93 0.87 – 1.00 0.036
Ordered logit thresholds (cutpoints)
1|2 1.045 0.081 0.886 – 1.204
2|3 2.702 0.082 2.542 – 2.862
Selected model statistics
Number of observations 60766
LRT χ² (Full vs Null) 7670.394
McFadden R² 0.058
* Predictors: β are log-odds (logit) coefficients from MASS::polr; SE and CI are on the logit scale.
† Odds ratios (OR) = exp(β); their CIs are exp(CI on the logit scale).
‡ Thresholds: cutpoints on the latent logit scale that partition adjacent HBI categories (no ORs).
§ LRT χ² compares the full model vs an intercept-only (thresholds-only) null model.
Provinces (GEOGPRV) were included as controls but omitted from the predictors section.

4.2 Significance of overall model and Threshold value

The overall model results in Table 2 show that the predictors included in the model do a meaningful job of explaining why some people show poorer healthy behaviour while others show better healthy behaviour. The likelihood‑ratio test demonstrates that the full model performs much better than a model with no predictors at all, and although the McFadden R² is modest—as is typical for health‑behaviour research—it still indicates that the model captures important differences in people’s behaviour patterns.

The cutpoints show where the model draws the boundaries between different levels of healthy behaviour—for example, the point where someone is more likely to be classified as moderate rather than poor, or good rather than moderate. These thresholds sit on an underlying scale that represents a person’s overall tendency toward healthy behaviour. Because the second cutpoint is much higher than the first, it indicates that moving from moderate to good healthy behaviour requires a bigger shift on this underlying scale than moving from poor to moderate. In other words, the model treats the leap into the highest category of healthy behaviour as a more substantial step.

4.3 Predicted probabilities and Marginal Effects

Figures 3 to 21 show predicted probabilities and average marginal effects (AMEs)/changes in predicted probabilities for each category of HBI to changes in a number socioeconomic/demographic variables as we used in our regression models (see table 1 for the relevant variables).

5 Conclusion and Recommendation

As for the support of public health programs or initiatives that can meaningfully increase the number of healthy non-users of the B.C. public health care system over the next two to five years, we would like to the make following recommendations: i. Need to increase awareness about poor healthy behavior ii. Focus on programs that can improve the mental health condition of the people iii. Invest more in education iv. Increase income generating activities

Appendix

A.1 Healthy Behavior index (HBI) and Socioeconomic characteristics

A.1.1 Education level

A.1.2 Age

A.1.3 Mental health

A.1.4 Income

A.1.5 Food security

A.1.6 Social connection

References

Centers for Disease Control and Prevention. 2023. “Influenza Vaccination Coverage.” https://www.cdc.gov/flu/fluvaxview/index.htm.
Chai, X., Y. Tan, and Y. Dong. 2024. “An Investigation into Social Determinants of Health Lifestyles of Canadians: A Nationwide Cross-Sectional Study on Smoking, Physical Activity, and Alcohol Consumption.” BMC Public Health 24: 2080. https://doi.org/10.1186/s12889-024-19427-4.
Manuel, D. G., C. Bennett, R. Perez, A. S. Wilton, A. R. Dass, A. Laporte, and D. A. Henry. 2019. “Burden of Health Behaviours and Socioeconomic Position on Health Care Expenditure in Ontario.” F1000Research 8: 303. https://doi.org/10.12688/f1000research.18205.1.
Poortinga, W. 2007. “The Prevalence and Clustering of Four Major Lifestyle Risk Factors in an English Adult Population.” Preventive Medicine 44 (2): 124–28.
Schuit, AJ et al. 2002. “Clustering of Lifestyle Risk Factors in a General Adult Population.” Preventive Medicine 35 (3): 219–24.
Statistics Canada. 2024. “Health Determinants: Health Behaviours and Socioeconomic Status.” https://www150.statcan.gc.ca/n1/pub/82-570-x/2024001/section1-eng.htm.
World Health Organization. 2021. “Body Mass Index Classification.” https://www.who.int/news-room/fact-sheets/detail/obesity-and-overweight.