SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

The objective of our study was to examine sociodemographic and behavioral variables underlying the geographic variation of obesity in Canada. We aimed to quantify the share of regional variation in average BMI attributable to commonly cited determinants of obesity and the remaining share, which is attributable to the idiosyncrasies of the regional environment (“regional effects”). Using data from the Canadian Community Health Survey (CCHS) (2004), ordinary least squares (OLS) regression, and Blinder–Oaxaca decomposition to decompose the difference in mean BMI between regions, we quantify two parts of the difference: a share explained by different levels of the covariates and a share explained by those covariates having different effects on BMI in the different regions, using the Atlantic provinces as the reference group. We observed that some differences (e.g., average BMI for males in Quebec compared to the Atlantic provinces) are mostly explained by the different levels of socio-demographic and behavioral covariates, while others (e.g., average BMI for females in Quebec compared to the Atlantic provinces) are mostly explained by the different effects of the covariates on BMI. In the latter scenario, even if covariates were made to be identical in the different regions, the difference in average BMI would persist. Thus, targeting covariates in different regions through plans like physical activity or nutrition policy, income equalization, or education subsidies will have ambiguous effects for addressing disparate obesity levels, being plausible policy options in some regions but less so in others. Future research and policy would benefit from identifying these region-specific attributes that have local implications for BMI.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

Obesity (often defined as a BMI ≥30 kg/m2) is a major public health concern that has been growing more prevalent in Canada and elsewhere since the 1950s but particularly during recent decades (1,2,3). Between 1985 and 1998 (4) the prevalence of obesity among adults more than doubled, from 5.6% to 14.8% (these values are for self-reported height and weight), and further increased to 23.1% by 2004 (5). From a population health perspective (6), obesity's association with various chronic diseases (e.g., cancer (7), leukemia (8), type 2 diabetes (9), dementia (10), arthritis (11), hypertension (12), and disability related to cardiovascular disease (13)) draws attention to the determinants of obesity: individual-level risk factors (e.g., physiology, behaviors) interacting with population-level regional determinants of health (e.g., built environment, societal norms), which may be driving the increase in obesity prevalence by affecting the population as a whole.

In Canada, there exists regional variation in obesity (4,14,15,16,17) such that Canadians from the Atlantic provinces (i.e., residents of Prince Edward Island, New Brunswick, Nova Scotia, and Newfoundland and Labrador) have been reported to have a higher average BMI than Western Canadians (i.e., residents of Manitoba, Saskatchewan, Alberta, and British Columbia.) More specifically, Shields and Tjepkema (16) showed that the proportion of obese people was lowest in British Columbia (19% obese), higher in the Prairies (the average of Manitoba, Saskatchewan, and Alberta being 28% obese), then dropped in Ontario and Quebec (Ontario 23% obese, Quebec 22% obese), and increased again in the Atlantic provinces (the average being 28.5%.) The overall average for obesity reported in Canada was 23%.

Geographical or spatial analysis of prevalence, while an important tool for public health surveillance (15) and a key starting point for studying obesity, needs to be followed by an investigation of the possible reasons underlying geographic variation in health outcomes such as obesity. The difference in prevalence between provinces insinuates that there could be possible determinants of obesity so widespread (e.g., applying to entire provinces) in society that they fail to register as risks to individuals (6). These determinants might include the socio-cultural environment, the physical environment, or economic policy in a region (18) and are by definition widespread if acting on a regional level. Therefore, estimates of causality or correlation within regions could either miss these important regional risks or over-state minor but heterogeneous ones.

Several factors may contribute to regional variation in obesity. For example, in an attempt to explain the regional variation in obesity across Canada described above, Vanasse et al.(14) estimated fruit and vegetable consumption and physical activity patterns by province. These authors detected substantial regional variation, reporting that 33.5% of British Columbians had low levels of leisure-time physical activity (i.e., energy expenditure of <1.5 kcal per kilogram of body weight per day) compared to 71% of Newfoundlanders, and 46.3% of British Columbians ate fewer than five servings of fruits and vegetables per day compared to 79.8% of Newfoundlanders. Although these patterns correspond with geographic variation in obesity, it is unlikely that these two variables are sufficient to fully explain geographic variation in obesity; other variables are likely at play. Further, the findings by Vanasse et al. could be interpreted to mean that interventions should focus on facilitating behavior change, when in fact the impact of even highly effective behavior change interventions might differ between regions.

Differences in BMI between geographic regions may also be explained, in part, by differences in socioeconomic variables across the regions. Research exists to show that, on average, there is an increased risk of obesity associated with lower education and different levels of income (19,20,21,22,23). Thus, a relatively lower obesity rate in one region may reflect different educational and income attainment profiles in that region if education and income have a homogenous effect across regions. On the other hand, education and income may have heterogeneous effects across regions, in which case region-specific attributes may have a role in the BMI differential by interacting with individual characteristics like household income and educational attainment to produce different outcomes. Region-specific variables may be conceptualized as “ecological” variables to which everyone in a region is exposed; examples include regional variables from the economic domain (e.g., childcare policy and food prices), socio-cultural domain (e.g., attitudes toward physical activity), and physical domain (e.g., the built environment and access to healthy food vendors.)

The objective of this paper was to examine potential sources of geographic variation of obesity in Canada, by analyzing BMI among Canadian working-age men and women in relation to various socio-demographic and behavioral variables across different regions of Canada. Policymakers may find it valuable to know whether the variation reflects a differential distribution of variables (e.g., more people are getting a higher education in one region than another) or whether something else is going on that causes the effect of a variable to manifest differently (e.g., having a higher education has a different (in direction or magnitude) impact on BMI in one region than in another (this phenomenon will henceforth be referred to as the return to education, for BMI)). Quantifying the contribution of these two separate yet simultaneously occurring effects on BMI will help predict the magnitude of impact of possible interventions. To separate these two sources of regional differences in BMI we use a Blinder–Oaxaca decomposition (24,25), which partitions the BMI differences between regions into “a share attributed to endowments” (i.e., variation in BMI that is a function of variation in the attributes of residents in the different regions), and “a share attributed to coefficients” (i.e., variation in BMI that is a function of variation in the returns to the attributes considered). This approach enables identification of potential policy impacts.

Methods and Procedures

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

Data source and study design

This study used a cross-sectional design with data from Statistics Canada's Canadian Community Health Survey (CCHS), Cycle 2.2 (public use version). Details are available at www.statcan.gc.ca. The CCHS was designed to collect health and demographic information from people of all ages living in Canada, with the exclusion of people living on Crown lands or Aboriginal People's Reserves, institutional residents, full-time members of the Canadian forces, and people living in the Territories or remote areas of the provinces. The cycle attained a national response rate of 76.5% (26). Our subsample of interest was adults of working-age, which we defined as 20–64 years.

We used the following variables: BMI (kg/m2 based on height and weight which were measured as part of a home visit), sex, age in groups of 5 years, marital status (single, including divorced and never married; or married, including married and common law), highest level of education attained (less than secondary, graduated secondary school, some postsecondary, or graduated postsecondary), employment status (employed or not), total household income levels ($0–$14,999, $15,000–$29,999, $30,000–$49,999, $50,000–$79,999, $80,000 and up), smoking status (nonsmoker, occasional smoker, daily smoker), alcohol intake (nondrinker, occasional drinker, daily drinker), physical activity levels (active, moderate, or inactive based on an index constructed by Statistics Canada using daily energy expenditure), fruit and vegetable consumption (number per day), and whether or not the individual was born in Canada. Canada was split into five regions for this analysis: British Columbia, the Prairies (Alberta, Saskatchewan, and Manitoba), Ontario, Quebec, and the Atlantic provinces (Nova Scotia, New Brunswick, Prince Edward Island, and Newfoundland). These regions are familiar regional breakdowns of Canada and provinces within the grouped regions are similar in their obesity and overweight profiles (16).

Statistical analysis

First, we analyzed the conditional mean of BMI using ordinary least squares (OLS) regression with Huber–White standard errors. More specifically, we regressed BMI on the “socio-demographic” variables of whether or not the individual was born in Canada, employment status, marital status, level of education, level of income, and age group, along with the “behavioral” variables of smoking, alcohol intake, physical activity, and fruit and vegetable consumption, for each geographic region and sex separately.

Next, we decomposed the differences in mean BMI between geographic regions to explore the underlying differences between regions. By using a Blinder–Oaxaca decomposition (24,25), we were able to break down the difference between regions into (i) the difference in the covariates by region (e.g., does Ontario have higher levels of income than the Atlantic provinces, which in turn contributes to the difference in BMI between the two regions?) and (ii) the difference in the returns to those covariates (e.g., does income have a larger effect on BMI in Ontario than in the Atlantic provinces?). We used the Atlantic provinces as a reference group because that region had the highest average BMI. Thus, it provides a compelling counterfactual comparison: by looking at the region with the highest BMI we can see how all the other regions would look if they had the Atlantic region's properties. For example, if Ontario and the Atlantic provinces were made to have the same proportions of people with identical levels of income, would the average difference in BMI between these regions still exist? This can be referred to as the difference in BMI attributed to the different endowments of income in the Ontario and Atlantic regions. If a difference still exists after the difference in endowments has been accounted for, then it would be considered a difference attributable to different returns to the endowments in the two regions, or “a difference in coefficients”.

Finally, for regional comparisons with statistically significant differences in coefficients (i.e., returns to covariates) we investigated which variables most influenced the overall coefficients effect by examining the disaggregated Blinder–Oaxaca output. Regional comparisons with statistically significant differences in the levels of the covariates were also investigated.

All data were analyzed using Stata SE ver 10.1. Standard Stata analytic commands were used, with the exception of the Blinder–Oaxaca mean decomposition, for which we used a Stata command by Jann (27). All analyses incorporated the appropriate sampling weight as directed by Statistics Canada.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

Analyses are based on a sample of individuals with complete data on all study variables (n = 7,494). Of those individuals age 20–64 with complete data on BMI (n = 8,519), 12.03% (n = 1,025) were excluded due to missing data on covariates. Relative to included participants, excluded participants were (at a significance level of P < 0.05) younger, had lower BMI, were less often employed, earned less, were less likely to smoke daily, and were less likely to be daily drinkers. Table 1 shows the average BMI and obesity rates for each province of Canada and the whole country in aggregate.

Table 1.  Summary statistics for observations with nonmissing variables
inline image

OLS estimates

The OLS output for the five regions is presented in Table 2. (A Chow test was conducted to determine whether or not the OLS coefficients could be combined into a single model for all regions. The test's result suggested such a combination would obscure important differences between the regions, thus Table 2 reports each region separately.) Several statistically significant effects were observed in several regions. For example, for males in Quebec, the Prairies, and Ontario, and for females in the Atlantic region, British Columbia, and the Prairies, being born outside of Canada was inversely associated with BMI. Income had a positive association with BMI for men and women in the Atlantic provinces, and for women in the Prairies. In general, a positive association between age and BMI was observed for women. Education was negatively associated with BMI for women in Quebec and British Columbia, and for men in British Columbia and Ontario. These varying estimates of the covariate effects indicate how different the determinants of BMI can be by region. Such differences set the stage for the next step (the Blinder–Oaxaca decompositions) wherein we further investigate these different effects.

Table 2.  OLS regression estimates for the five regions, disaggregated by region
inline image

Blinder–Oaxaca decompositions

The results of the Blinder–Oaxaca decompositions are presented in Table 3. The values in the table refer to the region in question being compared to the Atlantic provinces. These estimates are reported from the region in question's point of view (e.g., “If British Columbia had the same level of endowments as the Atlantic provinces then BMI in British Columbia would change by this value” or “If British Columbia had the same returns to the variables as the Atlantic provinces then BMI in British Columbia would change by this value.”)

Table 3.  Oaxaca decompositions using the Atlantic provinces as the reference group
inline image

For each two-way comparison, the following information is reported: difference (the predicted difference in mean measured BMI between the two regions given the covariates specified in the model (Atlantic provinces-Region X (the region indicated in the row))), endowments (how would BMI in Region X change if Region X had the Atlantic provinces' level of the covariates in the model), coefficients (how would BMI in Region X change if they had the Atlantic provinces' returns to the covariates in the model), interaction (the term that is used to balance the equation of the predicted difference; this is often referred to as the “unexplained difference” between the Atlantic provinces and Region X, since the correction of both coefficients and endowments levels need not reduce the difference to zero), and the number of observations on which the two-way comparison is based.

Statistically significant findings for the males appeared only in the Quebec–Atlantic provinces comparison. If males in Quebec had the same endowments as males in the Atlantic provinces, Quebec's expected average BMI would be 0.76 units higher than its current value. The overall difference between the regions is 1.06 BMI units, so the statistically significant estimate of the effects of the endowments means that the overall difference between the regions can be mostly explained by different levels of the covariates included in the model. The share of the difference attributable to the different returns to the covariates between regions was not statistically significant for any regional comparison.

For females, most regions exhibit statistically significant differences when compared to the Atlantic provinces. Females in British Columbia have an overall average BMI that is 1.60 units less than the average predicted BMI for women in the Atlantic provinces, and it is almost completely explained by the endowments (different levels of covariates) between the two regions, which account for 1.46 units of the difference. Females in Quebec have a statistically significant overall difference in BMI from women in the Atlantic provinces, with the females in Quebec having a predicted average BMI 1.50 units lighter than females in the Atlantic provinces. Unlike the British Columbia–Atlantic comparison, the Quebec–Atlantic difference for females is entirely explained by the share of the difference attributable to the difference in the coefficients (i.e., different returns to the covariates.) Finally, the females in Ontario have a statistically significant difference in predicted BMI from the females in the Atlantic provinces, with Ontario being 1.49 BMI units lighter. The overall difference is attributable to both the difference in the endowments (levels of the covariates) (0.91 BMI units) and the coefficients (returns to the covariates) (1.27 BMI units).

For females, in both the Quebec–Atlantic comparison and the Ontario–Atlantic comparison, the coefficients (returns to covariates) dominate as determinants of BMI differences between the regions. For these two comparisons, we investigated which variables most influenced the overall coefficients effect by looking at the disaggregated Blinder–Oaxaca output, as shown in Table 4. For females in Ontario, higher income has a different return to earning high income in the Atlantic provinces, with Ontario females in the $50,000–$79,000 range being 1.16 BMI units lighter than their Atlantic province counterparts. Similarly, those Ontario women in the $80,000 and up income category were 1.26 BMI units lighter than their Atlantic province counterparts. Eating fruits and vegetables has a statistically significant and large effect on the coefficients estimate for Ontario women compared to Atlantic women, with the same fruit and vegetable designation resulting in the average BMI in Ontario being 2.91 units lower than in the Atlantic provinces. Finally, Ontario women who had an “active” level of physical activity were 0.49 BMI units heavier than women in the Atlantic provinces that had the same level of physical activity. For females in Quebec, the estimated contribution of fruit and vegetable consumption was statistically significant and large such that comparable levels of fruit and vegetable consumption in the two regions results in an average BMI 3.8 units lower in Quebec than in the Atlantic provinces.

Table 4.  Disaggregated results of select Blinder-Oaxaca decompositions
inline image

We also compared the disaggregated Blinder–Oaxaca output for those comparisons with statistically significant contributions from the different endowments (data not shown). For females in British Columbia, the majority of the statistically significant difference with females from the Atlantic provinces was attributed to the different proportions of those born outside of Canada. The average BMI of females in British Columbia would be 1.08 BMI units higher if the mix of people born in Canada were the same for both regions. For males in Quebec, the only statistically significant difference with males from the Atlantic provinces was attributed to the different levels of consumption of fruits and vegetables. Males in Quebec would be 0.27 BMI units heavier on average if fruit and vegetable consumption among males was the same in both regions. For the Ontario–Atlantic comparison for women, none of the covariates were found to contribute to the statistically significant endowments effect, in the disaggregated analysis.

Finally, due to the right-skewed nature of the BMI variable we considered models with the dependent variable log-transformed. We reran the Blinder–Oaxaca analysis with the natural log of the measured BMI as the dependent variable (data not shown) and the results were qualitatively similar. The only differences were (i) the overall difference in mean BMI for females in the Prairies vs. the Atlantic provinces became statistically significant, and (ii) the endowments (levels of covariates) became a statistically significant contributor to the difference in mean BMI between females in Quebec and the Atlantic provinces, yet the magnitude of the effect was approximately half that of the coefficients, which remained the same. Due to the more natural interpretation of BMI units (vs. the natural log of BMI units) coupled with the substantively similar findings between the two formats, we opted to present the findings based on the untransformed BMI variable.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

The OLS regressions showed that the different regions of Canada exhibit different associations between BMI and covariates but we receive additional information about regional differences by applying a Blinder–Oaxaca decomposition. According to the Blinder–Oaxaca decompositions on the same model estimated for the OLS regression, the relative contributions of both levels of covariates and returns to covariates differ between regions.

The difference between males in Quebec and the Atlantic provinces is due to the statistically significant endowments (i.e., levels of covariates.) This indicates that correction of the different levels of covariates in the Atlantic provinces may be a viable policy option toward obesity prevention for males, if bringing the Atlantic provinces' average male BMI in line with those of Quebec males is the goal of policy. For females in other regions, such as Ontario or Quebec compared to the Atlantic provinces, correcting levels of covariates is not necessarily the most effective policy target, because it is the different returns to the covariates—rather than the levels—that are important. The scenarios in which the differences in average BMI are primarily attributable to different returns to covariates draw attention to factors that may be present in one province but not in another, and impact BMI outcomes in combination with personal characteristics. An example would be females in Ontario vs. the Atlantic provinces, a comparison for which the different levels of the covariates are not able to wholly explain the difference in average BMI. The significant contribution of the coefficients (returns to the covariates) in this example indicates that there is some sort of region-specific effect that differs between the regions and causes the covariates to have different effects in the two regions, resulting in a higher average BMI in the Atlantic provinces.

Regional effects are unmeasured in our analysis and therefore we cannot state with certainty what these might be. However, we can look to other sources for insight. Taking the comparison between females in Quebec and females in the Atlantic provinces as an example, one plausible contributor is regional differences in government social policy. In 1997, Quebec instituted a subsidy that made the cost of daycare $5/day (no Atlantic province had funding available in this fashion) and this subsidy was available to a large portion of the population (28). Given that formal child care arrangements from ages 3–5 have been linked with a reduced risk of obesity among children at ages 6–12 (29) and considering the evidence linking childhood obesity with obesity in adulthood (30,31), this is one example of an age-related social policy that could be contributing to regional differences. If children in Quebec had a better starting position than children in the Atlantic provinces through policy, then we might expect those advantages to carry through to adulthood, and thus mitigate the effect of factors like physical activity on BMI in the Atlantic provinces compared to Quebec.

Food prices are another factor that could affect the cross-regional obesity rates and manifest as different returns to behavioral and socioeconomic variables. Residents of the different provinces are subject to different food prices, and are also subject to different provincial levels of income. There is evidence that in at least one Canadian province it would not be possible for a family earning minimum wage or seniors on social assistance to afford a nutritious basket of food monthly (32,33), and through such a mechanism, the relationship between socioeconomic circumstances and BMI could differ between regions. To explore, post hoc, the plausibility of regional differences in food price in contributing to regional differences in BMI we compiled a simple table (Figure 1) that shows the average personal income after tax by region and consumer price indexes for food for the year 2004 (to correspond with the CCHS data used in this paper). The data for the table in Figure 1 comes from Statistics Canada's database of national statistics CANSIM (Canadian Socio-economic Information Management System) and more information can be found at http:cansim2.statcan.gc.ca. One can see that there is substantial variation between the regions, reflecting a difference in the ability of regional incomes to match the regional prices of food. The Atlantic provinces have the lowest purchasing power for fruits and vegetables and food bought in stores, and relatively speaking, food bought in a restaurant is the cheapest alternative. Such a price differential, and the associated incentive that it gives the regional population, could be a regional effect as the other regions may be able to afford both more money toward food and more money toward healthy food options, therefore potentially relying less on cheaper processed food (34) or restaurant foods that are generally served in larger portions and encourage overeating (35). Different food environments could potentially impact the protective effects of other covariates on BMI, such as income, as the population in these regions simply cannot make the same choices as they could in a different food environment (e.g., income in the Atlantic provinces has a different meaning, relative to BMI, than income in other regions, such that it places constraints on food purchases to a larger extent in the Atlantic region.)

image

Figure 1. Purchasing power for food sources by region. Data is from CANSIM (Canadian Socio-economic Information Management System) table 2020602 and table 326-0015. Data is available at http:cansim2.statcan.gc.ca. BC, British Columbia.

Download figure to PowerPoint

The results of the disaggregated Blinder–Oaxaca decomposition analysis for different coefficients provides some support for these proposed causes of the contribution of the coefficients to some regional differences in women. The income effect observed when comparing Ontario and the Atlantic provinces means that despite a similar and relatively high level of personal financial resources in the two regions, there are different circumstances that define how the women in the Atlantic provinces can use this income compared to those in Ontario, which in turn have implications for BMI. Thus, the food environment (or environment in general) could be quite different between regions such that a protective effect of higher income for BMI is region-specific: stronger in Ontario than in the Atlantic provinces.

The case of fruits and vegetables having a different effect in both Ontario and Quebec when compared to the Atlantic provinces is less straightforward. One possibility is that the quality of fruits and vegetables consumed differs by region. For example, fruit servings in the Atlantic provinces may consist of more canned fruits (due to a climate less conducive to growing a variety of fruits or geographical distance affecting the efficiency of transporting perishables to the region), which are often packed in syrup, adding to the calorie count, compared to fresh fruits. Another plausible explanation is that consumption of other foods varies regionally, and differentially offsets the impact of fruit and vegetable consumption. For example, if high levels of consumption of fruits and vegetables in the Atlantic region are associated with higher consumption of food overall (including less healthy foods), perhaps reflecting dietary social norms, then we would observe different returns to the consumption of fruits and vegetables. Finally, it is possible that the variable “Fruit and Vegetable consumption” is not actually capturing just that, and regionally there is variation in what it is capturing. However, this measure has been shown to be a valid indicator of diet quality (36). The difference in the coefficients attributable to fruit and vegetable consumption is intriguing and further research should interrogate what fruit and vegetable consumption entails in different regions, and investigate what factors are associated with fruit and vegetable consumption in the Atlantic provinces that are causing these less favorable returns.

Taken altogether, our results indicate that for females, there is interesting regional variation in BMI which manifests among those living in Ontario and Quebec compared with the Atlantic region. For males, on the other hand, there is less average regional variation in BMI and what does exist can be readily explained by different levels of covariates between the different regions. These gender differences raise the interesting possibility, for pursuit in future research, that males may be more resilient than females to regional effects.

Our study has some limitations. One limitation is that the inclusion of behavioral variables may introduce statistical bias into the study. Specifically, inclusion of these variables introduces endogeneity into the interpretation of the model (possible reverse causation from the dependent variable to the independent variables). Endogeneity contributes to the impossibility of distinguishing causality, for example, do people who exercise have a lower BMI through exercise, or are people who have a lower BMI more likely to participate in exercise, because they find it easier or more enjoyable? Notwithstanding this problem, we judged the inclusion of behavioral variables in our models to be important, considering the demonstrated association of these variables with obesity (37).

A second limitation is that any omitted variables that have an effect on BMI will have their share of the difference erroneously attributed to differences in coefficients (i.e., the returns to the covaries). Variables such as race/ethnicity could be considered important omitted variables, since people of East Asian descent are known to have a lower BMI than the general population of Canada, and there are regional differences in the proportion of the population of East Asian descent (e.g., British Columbia has a larger immigrant proportion from East Asia than the other regions in our analysis) (5,38,39). Our analysis identified situations in which the endowments (i.e., covariates) were statistically significant even in the face of this limitation, so it suggests that sizable contributions to BMI were captured in our choice of variables. A third limitation is that the explanatory power of the models overall was low in some cases, with values of R2 ranging from 0.06 to 0.24 for the OLS models. Also, it is important to note that the population distribution of BMI is not Gaussian but, rather, right-skewed (1). Although we ascertained that the results presented here are not unduly affected by the right skew, we acknowledge that our approach (which focuses on the mean) neglects the potential policy-relevant possibility that effects and returns are different at higher percentiles of the distribution (amongst the heaviest individuals). Future research would need to accommodate the right skew by using techniques such as quantile regression to explicitly model such issues (23) in the decomposition context.

Acknowledgmant

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

D.J.D. is supported by a Population Health Intervention Research Centre (PHIRC) Traineeship within the Population Health Intervention Research Network (PHIRNET). L.M. is supported by a Population Health Investigator Award and Establishment Grant from Alberta Innovates-Health Solutions (formerly the Alberta Heritage Foundation for Medical Research). We thank Dr J.C. Herbert Emery for helpful comments on an earlier version of this manuscript.

REFERENCES

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. SUPPLEMENTARY MATERIAL
  8. Acknowledgmant
  9. DISCLOSURE
  10. REFERENCES
  11. Supporting Information

supporting Information

FilenameFormatSizeDescription
oby_1460_sm_oby2010339x1.doc44KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.