Sage Journals: Discover world-class research

Abstract

Background: Empirical evidence on the relationship between consumption of fruits and vegetables and body weight is inconclusive. Previous studies mostly use linear regression methods to study the correlates of the conditional mean of body mass index (BMI). This approach may be less informative if the association between fruit and vegetable consumption and BMI significantly varies across the BMI distribution. Objective: The association between fruit and vegetable consumption and the BMI is examined using quantile regression. Methods: A nationally representative sample of 11,818 individuals from the Canadian Community Health Survey (2004) is used. A quantile regression model is estimated to account for the potential heterogeneous association between fruit and vegetable intake and BMI at different points of the conditional BMI distribution. The analyses are stratified by gender. Results: The multivariate analyses reveal that the association between fruit and vegetable intake and BMI is negative and statistically significant for both males and females; however, this association varies across the conditional quantiles of the BMI distribution. In particular, the estimates are larger for individuals at the higher quantiles of the distribution. The ordinary least squares (OLS) model overstates (understates) the association between FV intake and BMI at the lower (higher) half of the conditional BMI distribution. Conclusion: Findings of the standard models that assume uniform response across different quantiles of BMI distribution may be misleading. The findings of this paper suggest that increasing the intake of fruits and vegetables may be an effective dietary strategy to control weight and mitigate the risk of obesity.

Keywords

fruits and vegetables BMI eating behavior quantile regression nutrition

Introduction

The dramatic rise in obesity prevalence and its well documented adverse effects have become a challenging issue for policy makers and academics over the past 2 decades. An individual is classified as obese when the body mass index (BMI) equals 30 or more (BMI is calculated as weight in kilograms divided by height in meters squared). Over 1 billion individuals worldwide are overweight, with approximately 300 million obese.¹ Obesity is a precursor of many chronic diseases (eg, cardiovascular disease, type 2 diabetes, hypertension, liver disease, as well as certain types of cancer).² In addition, obesity may cause psychological disorders through societal prejudice and discrimination against obese individuals.³ Moreover, the economic cost attributable to overweight and obesity is substantial.^4–6 For example, a recent study estimates that the total economic cost of overweight and obesity in the United States is $270 billion yearly and the cost in Canada is $30 billion yearly.⁶ There has been remarkable increase in the prevalence of obesity in Canada, the adult obesity prevalence rate almost doubled from 13.8% in 1978 to 23.1% in 2004.⁷ This has been confirmed by an increase in the average BMI of Canadian adults from 25.2 in 1994 to 26.5 in 2008.

Several studies have argued that technological innovation may have contributed to increase body weight through a reduction in food prices, as well as the promotion of sedentary behaviors.^8–11 The World Health Organization¹² together with empirical studies has linked individual’s diet and nutrition behavior including the consumption of fruits and vegetables (FV) to the global rise in obesity (for a comprehensive review, see Rolls et al and Tohill et al).¹³,¹⁴ The health benefits of adequate consumption of FV daily (5 servings or a minimum of 400 grams) are enormous.¹²,¹⁵ Inadequate consumption of FV has been linked to approximately 2.7 million deaths per year, 19% of gastrointestinal cancers, 31% of ischemic heart diseases, and 11% of strokes.¹²,¹⁶

The rising obesity rate in Canada has been accompanied by increasingly poor eating behavior among Canadians.¹⁷ According to Health Canada,¹⁸ Canadian eating habits do not fully meet Canada’s food guide to healthy eating. A significant proportion of the Canadian population aged 12 and older reported consuming FV less than 5 times per day.¹⁹ For example, during the period 2003–2010, approximately half of Canadian females and more than 60% of males consumed FV less than 5 times per day (see Figure 1). It is assumed that the frequency of FV consumption is equal to serving, hence, consuming FV less than 5 times per day is below the recommended level.²⁰ Whereas in the United States, more than two-thirds of adults consume fruits less than 2 times per day and three-quarters consume vegetables less than 3 times per day in 2009.²¹

Figure 1.

Percentage of males and females aged 12 or older reporting that they consumed fruits and vegetables at least 5 times daily in Canada from 2003 to 2010.

High intake of FV may help in reducing the risk of obesity because most FV are high in water and fiber content and low in fat content.¹³ Thus, adding FV to the diet enhances satiety, reduces feelings of deprivation and hunger, and reduces energy intake.¹³ There is mixed empirical evidence about the association between FV intake and body weight in both clinical¹³ and epidemiologic studies.¹⁴ In some studies, body weight is negatively associated with the intake of FV,²²,²³ while other studies find no significant association.^24–26 For example, using a sample of children and adolescents in the United States, Field et al²⁴ find that the intake of FV or juice is not related to changes in BMI during 3 years of follow-up. In a prospective cohort study among middle-aged women, He et al²² find that the higher the consumption of FV over time, the lower the risk of obesity and weight gain. Using a sample of participants in the Baltimore Longitudinal Study of Aging, Newby et al²³ find that consuming a diet rich in FV and low in fat, dairy, whole grains, meat, fast food, and soda is associated with smaller gains in BMI and waist circumference.

Recently, several studies have examined the indirect effect of FV on BMI through its prices. For example, using repeated cross-sectional US data, Auld and Powell¹⁰ find that the prices of FV are positively associated with adolescents BMI. They also find that a decrease in the relative price of FV (a proxy for low energy-dense foods) tend to reduce BMI if the price per calorie of less energy-dense foods is lower than those of high energy dense food. Sturm and Datar²⁷ find that lower real prices for FV predict a significantly lower gain in BMI between kindergarten and third grade. Some studies report that there are gender differences in eating patterns and report how these affect body weight.²⁸,²⁹ Baker and Wardle³⁰ find that men consume fewer servings of FV daily than women. They attributed this to poorer nutrition knowledge of men relative to women: they find that men are less likely to know the healthy recommendations for FV intake, and the benefits of FV consumption for disease prevention.

The objective of this study is to examine the association between FV intake and body weight along different points of the BMI distribution using data from the Canadian Community Health Survey (CCHS). The key contribution of this study is twofold. First, most of the previous studies report only the bivariate association between the intake of FV and BMI, without controlling for confounding factors like socio-demographic and lifestyle (such as physical activity and smoking status) which have been shown to be important determinants of individual BMI.¹⁴ Second, previous multivariate studies mostly use linear regression methods to examine the correlates of the conditional mean of BMI. This approach may be less informative if the association between the intake of FV and BMI significantly varies across the BMI distribution. Moreover, logistic regression treats observations that exceed a particular cut off level equally. For example, 2 individuals with a BMI of 40 and 30 are equally classified as being obese, notwithstanding the intensity of obesity for the first person is higher. This leads to a statistical loss of information that may be relevant for intervention measures. Individuals may respond differently to the factors causing obesity, depending on their location in the BMI distribution.

Accordingly, this study uses a quantile regression framework to characterize the heterogeneous association across the different quantiles of the BMI distribution. This is relevant to the nutrition and obesity literature where attention is given to certain segments of the BMI distributions. For example, individuals in the upper quantiles of the BMI distribution, both obese and overweight, are of more interest to policies aimed at reducing obesity. Standard linear regressions, like ordinary least squares (OLS), estimate the effect of different covariates on the conditional mean of the BMI. This average effect may over or under estimate the influence of the covariates at different points across the BMI distribution and, hence, may lead to misleading policy inferences.

Data

This study uses data from the Statistics Canada 2004 Canadian Community Health Survey (CCHS) cycle 2.2. CCHS is a nationally representative cross-sectional survey of the Canadian population and it collects vital information on health related behavior, as well as corresponding economic and social-demographic variables. The survey excludes those living on Indian Reserves and Crown Lands, institutional residents, full-time members of the Canadian forces, and residents of certain remote regions. 45 889 households were selected to participate in Cycle 2.2 of the CCHS. However, a national response rate of 76.5% was achieved. Data were collected in person and approximately 7% of respondents had their first 24-hour dietary recall interview completed over the telephone. The nutrition questionnaire of the CCHS consists of 2 components: general health and 24-Hour Dietary Recall. The general health component has information about socio-demographic characteristics of respondents, their height and weight, physical activity, and chronic health conditions. The 24-hour dietary recall component has information about all the food and beverages a respondent consumed during the 24 hours preceding the interview. A second dietary recall interview was conducted 3 to 10 days after the initial interview. Unlike previous cycles of the CCHS, Cycle 2.2 has the merit that the BMI is measured rather than self reported. This was done personally by the interviewer. For more information about the CCHS, see Statistics Canada.³¹ We restrict the sample to those aged 14–65 years, and after excluding missing observations, the sample includes 11 818 individuals. Seniors tend to have a low BMI due to ageing rather than dietary choice. The eating behavior of children is largely affected by their parental background. Accordingly, we restricted our sample to those aged 14–65 so as to minimize factors that may bias our results.

The BMI, which is the dependent variable, is derived from the measured anthropometric information (height and weight) available in the CCHS. The BMI is calculated as body weight in kilograms divided by height in meters squared. The merit of using the 2004 CCHS cycle 2.2 is that the BMI is based on a respondent’s actual (measured) weight and height. This study follows the standard in the literature by using a set of covariates that has been shown to be potential determinants of the BMI. The independent variable of interest is an individual’s FV consumption. This variable indicates the total number of times per day the respondent consumes FV. Other individual socio-demographic and lifestyle variables are also included in the analyses. Age is represented by 3 categories: 14–30 (reference group), 31–50, and 51–65. Gender is captured by a dummy variable (male = 0, female = 1). Marital status is represented by 3 dummy variables: married, separated and single (reference group). Four dummy variables represent an individual’s educational attainment: less than secondary, secondary, some post secondary (reference group), and post secondary. Household income is represented by 4 dummy variables: less than $30,000 (reference group), $30,000 to $49,999, $50,000 to $79,999, and $80,000 or more. A dummy variable indicating individual social interaction (sense of belonging to a local community) is included (strong =1, weak = 0). Individual physical activity level is represented by 3 categories: active, moderate, and inactive (reference group). This classification is based on the total daily energy expenditure values (kcal/kg/day) on leisure-time physical activities. The daily energy expenditure for each activity is measured using the frequency, duration per session and the metabolic energy cost of the activity. An individual is classified as physically active if the total daily energy expenditure is greater than 3, as moderately active if the total daily energy expenditure is greater than 1.5 and less than 3 and inactive otherwise. For more information see (statistics Canada, 2005). Smoking status is classified as: never smoker (reference group), current smoker, and former smoker. Immigration status is captured by a dummy variable (immigrant = 1, non-immigrant = 0). Provincial or regional effects are captured in 5 categories: Ontario, British Colombia, Atlantic (comprising New Brunswick, Prince Edward Island, Nova Scotia and Newfoundland and Labrador), Western (Alberta, Saskatchewan and Manitoba) with Quebec as the reference group.

Method

Economists have developed economic models to explain how individuals engage in different consumption behaviors. Individuals maximize their utility subject to income, time and other resource constraints (eg, Lakdawalla and Philipson and Auld and Powell).⁹,¹⁰ One of those behaviors is food consumption. These models are based on the Becker and Murphy³² rational addiction (RA) framework which has become the canonical model of analysis. Consumers in this model make optimal choices on what to consume. Borrowing from the behavioral economics literature, Ruhm³³ adds to the traditional economic model by allowing for the possibility that individual weight outcomes could also be determined by biological and environmental cues. These cues can subvert the decision part of the brain which may lead to sub-optimal choices. In his model, advances in food engineering by producers may have contributed to the difficulty of resisting food cravings.

To examine how BMI is associated with the frequency of FV consumption across the BMI distribution, we estimate the following quantile regression³⁴,³⁵:

Q_{B M I} (τ | f v_{i j}, X_{i j}) = α (τ) + β (τ) f v_{i j} + X_{i j} θ (τ)

(1)

where i and j denote individual and province of residence. BMI denotes an individual’s Body Mass Index which is derived from measured height and weight. fv denotes the frequency of fruit and vegetable consumption and X is a vector of control variables. τ represents quantile, $Q_{B M I} (τ | f v_{i j}, X_{i j})$ is the τ th conditional BMI quantile function. Re-estimating equation (1) by varying the quantile, τ captures the heterogeneous association between BMI and fv along the different points in the conditional BMI distribution.

Results

Summary statistics of the variables used in the analyses are reported in Table 1. The mean BMI is 26.5, which indicates that, on average, the study population is slightly overweight. The average number of FV servings per day is approximately 4, and this is below the recommended number of 5 times per day. In terms of educational level completed, approximately 45% have completed 1 or more post secondary education, whereas 21% have less than secondary education. A large percentage (54%) of the sample is physically inactive. Approximately 53% are female and 21% are immigrants.

Table 1.

Summary Statistics

Variables	Mean	SD
BMI	26.50	5.73
Fruits &Vegetables	4.20	2.14
Age
Age 14-30	0.33	0.47
Age 31-50	0.42	0.49
Age 51-65	0.25	0.43
Gender
Male	0.47	0.50
Female	0.52	0.50
Marital status
Single	0.32	0.46
Married	0.58	0.49
Separated	0.10	0.28
Educational attainment
Less secondary education	0.21	0.41
Secondary education	0.18	0.38
Some post secondary	0.10	0.30
Post secondary	0.50	0.50
Income Level
Income level(less than30)	0.17	0.38
Income level(30-49)	0.21	0.40
Income level(50-79)	0.27	0.44
Income level(>=80)	0.27	0.44
Social interaction
Strong	0.61	0.48
Weak	0.38	0.48
Physical activity
Active	0.20	0.40
Moderate	0.26	0.44
Inactive	0.54	0.50
Smoking status
Current smoker	0.26	0.44
Former smoker	0.23	0.42
Never smoker	0.51	0.50
Immigration status
Immigrants	0.21	0.41
Non immigrants	0.78	0.41
Province of residence
Quebec	0.27	0.44
Ontario	0.34	0.47
British Columbia	0.13	0.34
Atlantic provinces	0.08	0.27
Western provinces	0.17	0.38
N	11818

The statistics are weighted using the CCHS sampling weights.

The BMI quantile regression and the OLS estimates for the full sample are reported in Table 2 for some selected quantiles between the 10th and 90th BMI distribution. In addition, Figure 2 displays the OLS and quantile regression estimates for the BMI determinants over the entire BMI distribution. The conditional mean estimate of BMI shows a negative relationship between FV and BMI. The quantile regression enables us to examine the heterogeneous responses of individual’s BMI to the model covariates at different tails of the BMI distribution.

Table 2.

OLS and Quantile Regression Results for BMI Determinants at Selected Quantiles for the Whole Sample

		Quantile regression estimates
	OLS	(10)	(30)	(50)	(70)	(90)
Fruit & vegetables	−0.183***	−0.110***	−0.113***	−0.182***	−0.235***	−0.292***
	(0.050)	(0.026)	(0.029)	(0.039)	(0.052)	(0.078)
Age
Age 31-50	2.024***	1.430***	1.579***	1.758***	1.643***	1.984***
	(0.330)	(0.140)	(0.166)	(0.225)	(0.326)	(0.539)
Age 51-65	2.889***	2.200***	2.513***	2.657***	2.692***	3.411***
	(0.352)	(0.153)	(0.183)	(0.240)	(0.336)	(0.563)
Gender
Female	−0.489**	−1.275***	−1.557***	−1.180***	−0.403*	0.982***
	(0.213)	(0.105)	(0.127)	(0.161)	(0.206)	(0.307)
Marital status
Married	0.831***	0.703***	1.208***	1.378***	0.940***	0.630
	(0.317)	(0.127)	(0.159)	(0.221)	(0.326)	(0.493)
separated	1.013**	1.083***	1.362***	1.296***	1.512***	1.404**
	(0.444)	(0.168)	(0.242)	(0.317)	(0.443)	(0.681)
Educational attainment
Less secondary	−0.187	−0.792***	−0.567***	−0.441*	0.046	0.645
	(0.356)	(0.164)	(0.199)	(0.255)	(0.329)	(0.492)
Secondary	0.171	0.287	−0.133	−0.580**	0.153	1.599***
	(0.427)	(0.196)	(0.228)	(0.291)	(0.380)	(0.542)
Post secondary	−0.407	0.066	−0.342*	−0.475*	−0.094	−0.135
	(0.342)	(0.170)	(0.199)	(0.254)	(0.325)	(0.459)
Income Level
Income 30-49	0.319	0.492***	0.081	0.441*	0.828***	0.558
	(0.265)	(0.124)	(0.178)	(0.226)	(0.291)	(0.396)
Income 50-79	0.385	0.144	0.116	0.420*	0.646**	0.792*
	(0.275)	(0.136)	(0.174)	(0.217)	(0.277)	(0.411)
Income 80 and above	0.203	0.174	0.108	0.208	0.409	0.347
	(0.343)	(0.149)	(0.182)	(0.225)	(0.286)	(0.474)
Social interaction
Strong	0.169	0.215**	0.096	0.137	0.393*	−0.380
	(0.229)	(0.109)	(0.128)	(0.162)	(0.211)	(0.323)
Physical activity
Active	−1.216***	0.175	−0.303*	−0.798***	−1.688***	−2.678***
	(0.248)	(0.120)	(0.155)	(0.195)	(0.266)	(0.415)
Moderate	−0.707***	0.182	−0.186	−0.454**	−0.894***	−1.681***
	(0.264)	(0.136)	(0.155)	(0.194)	(0.248)	(0.369)
Smoking status
Current smoker	−0.086	−0.273**	−0.164	−0.432**	−0.168	0.452
	(0.257)	(0.110)	(0.143)	(0.186)	(0.246)	(0.397)
Former smoker	0.880**	0.502***	0.649***	0.812***	0.757***	1.185**
	(0.350)	(0.151)	(0.173)	(0.214)	(0.276)	(0.461)
Immigration status
Immigrant	−1.122***	−0.209	−0.903***	−1.082***	−1.313***	−2.253***
	(0.396)	(0.154)	(0.184)	(0.235)	(0.310)	(0.546)
Province of residence
Ontario	0.341	−0.075	0.210	0.526**	0.419	0.921*
	(0.362)	(0.151)	(0.184)	(0.235)	(0.314)	(0.501)
British Columbia	0.108	0.071	0.064	0.030	0.051	0.838
	(0.383)	(0.172)	(0.217)	(0.281)	(0.368)	(0.565)
Atlantic	0.956***	0.514***	0.833***	0.743***	0.834***	2.150***
	(0.332)	(0.152)	(0.182)	(0.236)	(0.308)	(0.487)
Western	0.835**	0.496***	0.523***	0.752***	0.727**	1.944***
	(0.337)	(0.131)	(0.178)	(0.229)	(0.306)	(0.489)
Constant	25.39***	19.86***	22.59***	24.72***	27.09***	31.46***
	(0.451)	(0.239)	(0.291)	(0.368)	(0.457)	(0.653)
Observations	11,784

Robust standard errors are in parentheses. *** P < .01, ** P < .05, * P < .1. The estimates are population weighted using the CCHS sampling weights.

Figure 2.

OLS and quantile regression estimates for BMI determinants: whole sample

Although the results reveal that the frequency of FV intake has a negative and statistically significant association with BMI, the coefficient of FV varies across quantiles of the conditional BMI distribution. In particular, the FV coefficient increases in size for individuals at higher points of the conditional BMI distribution. For example, the coefficient of FV at the 90th quantile is almost 3 times the estimate at the 30th quantile, suggesting that an increase in the intake of FV may be an effective dietary strategy to control weight and reduce the risk of obesity especially for the overweight. The FV estimate at the median (50th quantile) is equal to the OLS estimate.

In terms of other control variables, the results show substantial differences across the quantiles of the BMI distribution (see Figure 2). Age has a positive relationship with BMI; those that are older (51 to 65, 31 to 50 years old) have higher BMI than the reference group (14 to 30 years old). The female coefficient is negative and statistically significant, indicating that females have less BMI compared to males. At the 90th quantile, the female estimate changes sign to positive. This means that at the 90th percentile, females’ BMI are higher than males. In general, the variables: married, separated and former smokers have positive and statistically significant association with BMI. Also, physically active (active and moderate), current smoker, and immigrant variables have a negative association with BMI. The association between socio-economic status (education and income) and BMI is less clear both in terms of sign and size of estimates.

The BMI quantile regression results for males are reported in Table 3 while females are reported in Table 4. The estimates for FV for both males and females reveal a similar pattern to the full population estimates shown in Table 2. The association between the frequency of fruit and vegetable intake and BMI is negative and statistically significant. Other covariates are largely similar to the full sample results with the exception of the income variables. For males, individuals in a high income household have higher BMI than those in low income households. The income variables for females have a negative effect on BMI in most of the quantiles and only a few of the estimates are statistically significant.

Table 3.

OLS and Quantile Regression Results for the BMI Determinants at Selected Quantiles for Males

		Quantile regression estimates
	OLS	(10)	(30)	(50)	(70)	(90)
Fruit & vegetables	−0.165***	−0.109**	−0.089*	−0.177***	−0.162***	−0.238***
	(0.061)	(0.051)	(0.051)	(0.051)	(0.034)	(0.051)
Age
Age 31-50	1.932***	1.286***	1.947***	1.552***	1.812***	1.942***
	(0.425)	(0.222)	(0.272)	(0.290)	(0.237)	(0.447)
Age 51-65	2.844***	1.883***	2.399***	2.361***	2.804***	3.844***
	(0.467)	(0.261)	(0.313)	(0.321)	(0.251)	(0.551)
Marital status
Married	0.687	1.247***	1.099***	1.337***	0.478*	0.0611
	(0.425)	(0.203)	(0.273)	(0.293)	(0.251)	(0.462)
Separated	0.294	0.826**	0.362	0.502	0.240	0.613
	(0.655)	(0.365)	(0.387)	(0.440)	(0.346)	(0.669)
Educational attainment
Less secondary	0.051	−0.877**	−0.792**	−0.148	0.224	1.674***
	(0.429)	(0.344)	(0.350)	(0.331)	(0.229)	(0.406)
Secondary	0.881*	0.931**	0.352	0.513	0.835***	1.342***
	(0.455)	(0.374)	(0.399)	(0.382)	(0.272)	(0.403)
Post secondary	0.315	0.299	0.326	0.419	0.557**	−0.329
	(0.379)	(0.360)	(0.349)	(0.338)	(0.225)	(0.358)
Income level
Income 30-49	0.996***	1.007***	0.933***	0.859***	1.089***	0.726**
	(0.337)	(0.258)	(0.321)	(0.303)	(0.212)	(0.355)
Income 50-79	1.282***	0.585**	1.068***	1.250***	1.328***	1.919***
	(0.386)	(0.281)	(0.319)	(0.281)	(0.202)	(0.355)
Income 80 and above	1.191***	1.093***	1.269***	0.863***	1.117***	1.146***
	(0.370)	(0.307)	(0.335)	(0.289)	(0.203)	(0.364)
Social interaction
Strong	0.008	0.183	0.107	0.274	0.278*	−1.056***
	(0.257)	(0.191)	(0.214)	(0.205)	(0.148)	(0.255)
Physical activity
Active	−0.648**	0.033	0.148	−0.402	−1.078***	−1.918***
	(0.329)	(0.224)	(0.245)	(0.250)	(0.183)	(0.346)
Moderate	−0.801***	−0.081	−0.325	−0.343	−0.946***	−2.157***
	(0.295)	(0.251)	(0.272)	(0.246)	(0.171)	(0.272)
Smoking status
Current smoker	−0.224	−0.468**	−0.377	−0.594**	−0.540***	0.843***
	(0.326)	(0.232)	(0.256)	(0.241)	(0.170)	(0.313)
Former smoker	1.102***	0.861***	1.427***	1.245***	1.212***	−0.264
	(0.317)	(0.274)	(0.293)	(0.272)	(0.192)	(0.295)
Immigration status
Immigrant	−1.385***	−0.678**	−1.252***	−1.369***	−1.440***	−1.946***
	(0.385)	(0.333)	(0.334)	(0.296)	(0.193)	(0.356)
Province of residence
Ontario	0.644*	−0.194	0.437	0.798***	0.309	1.320***
	(0.367)	(0.291)	(0.327)	(0.303)	(0.211)	(0.365)
British Columbia	0.347	−0.117	0.086	0.533	−0.295	0.638
	(0.414)	(0.340)	(0.374)	(0.357)	(0.247)	(0.395)
Atlantic	0.784*	0.149	0.582*	0.817***	0.246	1.464***
	(0.402)	(0.321)	(0.326)	(0.305)	(0.217)	(0.382)
Western	1.246***	0.340	0.673**	1.255***	1.134***	2.433***
	(0.378)	(0.253)	(0.316)	(0.293)	(0.209)	(0.372)
Constant	24.12***	19.45***	21.19***	23.22***	26.13***	31.26***
	(0.527)	(0.449)	(0.476)	(0.465)	(0.306)	(0.548)
Observations	5,358

Robust standard errors are in parentheses. *** P < .01, ** P < .05, * P < .1. The estimates are population weighted using the CCHS sampling weights.

Table 4.

OLS and Quantile Regression Results for the BMI Determinants at Selected Quantiles for Females

		Quantile regression estimates
	OLS	(10)	(30)	(50)	(70)	(90)
Fruit & vegetables	−0.177**	−0.109***	−0.180***	−0.151**	−0.211***	−0.279**
	(0.074)	(0.032)	(0.034)	(0.059)	(0.067)	(0.111)
Age
Age 31-50	2.146***	1.123***	1.487***	1.898***	1.585***	4.050***
	(0.479)	(0.210)	(0.213)	(0.343)	(0.424)	(0.711)
Age 51-65	2.810***	1.961***	2.405***	2.692***	2.859***	4.674***
	(0.491)	(0.212)	(0.228)	(0.356)	(0.426)	(0.733)
Marital status
Married	1.011**	0.309*	1.255***	1.480***	1.766***	0.023
	(0.455)	(0.188)	(0.201)	(0.320)	(0.382)	(0.709)
Separated	1.357**	0.932***	1.543***	1.625***	2.868***	0.175
	(0.613)	(0.225)	(0.280)	(0.451)	(0.510)	(0.914)
Educational attainment
Less secondary	−0.330	−0.226	−0.207	−0.358	0.277	−0.910
	(0.556)	(0.232)	(0.239)	(0.376)	(0.458)	(0.691)
Secondary	−0.349	0.571**	−0.363	−1.281***	−0.393	0.572
	(0.642)	(0.237)	(0.273)	(0.423)	(0.502)	(0.822)
Post secondary	−0.930*	0.039	−0.529**	−1.160***	−0.781*	−1.142
	(0.564)	(0.229)	(0.237)	(0.366)	(0.441)	(0.711)
Income level
Income 30-49	−0.134	0.011	−0.530**	0.002	0.267	−0.687
	(0.393)	(0.161)	(0.209)	(0.334)	(0.377)	(0.545)
Income 50-79	−0.286	−0.140	−0.352*	−0.391	−0.061	−0.981*
	(0.388)	(0.174)	(0.199)	(0.330)	(0.353)	(0.571)
Income 80 and above	−0.552	−0.399**	−0.801***	−0.577*	−0.522	−0.970
	(0.559)	(0.173)	(0.209)	(0.346)	(0.392)	(0.766)
Social interaction
Strong	0.347	0.366***	0.064	0.026	0.489*	0.583
	(0.372)	(0.135)	(0.160)	(0.252)	(0.287)	(0.480)
Physical activity
Active	−1.754***	0.192	−0.282	−1.195***	−2.246***	−3.975***
	(0.359)	(0.173)	(0.206)	(0.321)	(0.322)	(0.566)
Moderate	−0.622	0.248*	0.090	−0.515*	−0.941***	−0.787
	(0.416)	(0.148)	(0.185)	(0.296)	(0.342)	(0.566)
Smoking status
Current smoker	0.112	0.085	−0.134	0.114	0.970***	0.048
	(0.392)	(0.142)	(0.173)	(0.285)	(0.327)	(0.554)
Former smoker	0.727	0.375**	0.009	0.180	0.928**	1.693**
	(0.625)	(0.172)	(0.208)	(0.335)	(0.390)	(0.835)
Immigration status
Immigrant	−0.910	−0.149	−0.792***	−0.652*	−0.956**	−2.199**
	(0.665)	(0.186)	(0.219)	(0.375)	(0.451)	(0.873)
Province of residence
Ontario	0.186	0.067	−0.118	0.588	0.304	0.024
	(0.592)	(0.180)	(0.216)	(0.366)	(0.434)	(0.786)
British Columbia	−0.096	0.362	−0.297	0.265	0.231	0.379
	(0.610)	(0.251)	(0.265)	(0.438)	(0.507)	(0.896)
Atlantic	1.137**	0.860***	0.746***	1.210***	1.619***	2.638***
	(0.515)	(0.187)	(0.210)	(0.365)	(0.425)	(0.759)
Western	0.504	0.681***	0.077	0.568	0.276	1.366*
	(0.545)	(0.175)	(0.210)	(0.359)	(0.426)	(0.778)
Constant	25.58***	18.76***	22.33***	24.16***	26.54***	33.62***
	(0.708)	(0.316)	(0.327)	(0.542)	(0.612)	(0.981)
Observations	6,426

Robust standard errors are in parentheses. *** P < .01, ** P < .05, * P < .1. The estimates are population weighted using the CCHS sampling weights.

Discussion

It has been reported that worldwide, 1 in 3 and 1 in 9 adults are overweight and obese respectively.³⁶ Several studies link obesity and excess weight to the eating behavior of individuals, which includes the consumption of fruits and vegetables (FV).¹³,¹⁴ The health benefits of consuming FV are numerous.¹²,¹⁵

Evidence from the clinical and epidemiological literature on the relationship between the intake of FV and body weight is inconclusive. In this study, we examine the association between the consumption of FV and the body mass index (BMI) using data from the Canadian Community Health Survey. Based on the unconditional estimates, we find that the daily average number of FV servings among individuals in our sample is approximately 4, which is below the recommended number of 5 servings per day. Results, from the OLS baseline model, show that the conditional mean of the BMI is negatively and significantly associated with FV consumption. We use a quantile regression to characterize the effect of FV consumption on the entire BMI distribution. We find that the association between FV intake and BMI is negative and statistically significant. Quantile regression shows that this association varies significantly across the conditional BMI distribution. In particular, the effect of FV increases in size for individuals at higher points of the conditional BMI distribution. The estimates for both males and females reveal similar patterns to the full population estimates that FV intake is negatively and significantly associated with BMI. The OLS model overstates (understates) the effect of FV intake on the BMI at the lower (higher) half of the conditional BMI distribution. This suggests that conclusions from standard models (eg, OLS) that assume uniform response across different quantiles of the BMI distribution may be misleading.

Results for the other BMI determinants are comparable to previous studies. Socioeconomic status (SES), as usually measured by income and education level, largely affects the dietary choices of individuals.³⁷ The level of income affects the amount of financial resources available for healthy and nutritious food, and also the time devoted to physical activity.³⁸ Educational attainment affects nutritional knowledge and awareness about the benefits of physical activity. Several studies show that people with higher SES have healthier, nutritionally more balanced diets and are more physically active than those with lower SES.³⁹,⁴⁰ Existing literature mostly documents a negative association between SES and the BMI among females in developed countries, however, this association is less consistent among males.⁴¹ In line with previous studies (eg, McLaren and Sánchez-Vaznaugh et al),⁴²,⁴³ we find a negative SES gradient in BMI among females, and a relatively strong positive income gradient among males.

Results for both males and females show that smoking status significantly affects the BMI. In particular, we find that smokers have lower BMI, while former smokers have a higher BMI, compared to those who never smoked. This is consistent with the general belief that smoking cessation is usually associated with an increase in the BMI.⁴⁴ For example, using a prospective study, Munafò et al⁴⁴ find that the BMI of never and former smokers is on average 1.6 kg/m2 higher than the BMI of current smokers. The authors also find an average increase in BMI of 1.6 kg/m2 due to smoking cessation. It has been reported that smoking suppresses the appetite,⁴⁵ where smokers may have higher metabolic rates than non-smokers and, hence, smoking may be used to control weight.⁴⁶

We find that immigrants have lower BMI than natives. This is in line with the findings of an early study on differences in obesity prevalence among US immigrants and natives.⁴⁷ The authors find that immigrants in the United States in general have lower BMI than natives, but these differences decreases overtime due to acculturation and the influence of the US lifestyle. Results also show that the BMI increases with age and this is consistent with a previous study by Baum and Ruhm,⁴⁸ who predicted an annual increase in the BMI of 0.12 kg/m². Since physical activity affects the expenditure side of the energy balance equation, it is well established that regular physical activity is an important determinant of bodyweight, people who are physically active being less likely to be obese.⁴⁹ Our results are consistent with this evidence.

This study has some strength. First, the BMI used is based on measured height and weight rather than the frequently used self-reported measure. It has been documented that individuals tend to over-report their height and under-report their weight, which may have implications in terms of the consistency of estimated parameters. Second, many of the previous studies report only the bivariate association between the intake of FV and BMI which could lead to misleading conclusion about the true association.¹³ Moreover, previous multivariate studies mostly estimate the effect of FV on the conditional mean of BMI/likelihood of being obese using standard linear/binary response regressions. Results from these estimation methods assume that the effect of the explanatory variables is the same at different parts of the BMI distribution. However, nutrition promotion and weight management policies give more attention to individuals at certain segments of the BMI distribution.

The current study has some limitations. First, the cross-sectional design of the analysis limits the ability to infer causality. Second, due to data limitation, the intake of FV is based on the number of times per day an individual consumes FV rather than the quantity consumed. Third, we did not control for the form in which FV are consumed. FV in their natural physical shape are low in energy density and have higher satiety effects, while they become more energy dense when fried, served with high-calorie sauces or dried.¹³ Fourth, there may be omitted variable bias due to unobserved individual characteristics like preferences.

Conclusion

From the public policy perspective, the findings of this paper suggest that policies aimed at increasing the intake of FV may help to control weight and mitigate the risk of obesity. The multivariate analyses showed that conclusions from the standard models that assume uniform response across different quantiles of BMI distribution may be misleading. Accordingly, understanding how the association between FV and BMI depends on individuals’ location on the BMI distribution may help in implementing intervention measures that target the most vulnerable groups (overweight and obese).

Footnotes

Acknowledgements

We are thankful to 2 anonymous referees of this journal as well as Gordon Fisher and Ian Irvine for their valuable comments and suggestions.

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

The authors received no financial support for this study.

Bios

Sunday Azagba is a PhD candidate in his final year in the Department of Economics, Concordia University, Montreal, Canada working in the area of Public Health, Health Economics and Applied Econometrics.

Mesbah Fathy Sharaf is a PhD Candidate in the Department of Economics, Concordia University, Canada. His field of specialization is Health Economics, Labor Economics and Applied Econometrics.

References

WHO. 10 Facts on Obesity. http://www.who.int/features/factfiles/obesity/facts/en/index.html. Accessed April 27, 2011.

. Obesity Epidemiology. Cambridge: Oxford University Press; 2008.

Wadden

Woenble

Stumkard

Anderson

. Psychosocial Consequences of Obesity and Weight Loss. Obesity Handbook. New York: Guilford Press; 2002;144-169.

Katzmarzyk

Janssen

. The economic costs associated with physical inactivity and obesity in Canada: an update. Can J Appl Physiol 2004;29:90-115.

Finkelstein

Ruhm

Kosa

. Economic causes and consequences of obesity. Annu Rev Public Health 2005;26:239-257.

Society of Actuaries. Obesity and its Relation to Mortality and Morbidity Costs. http://www.soa.org/files/pdf/research-2011-obesity-relation-mortality.pdf. Accessed September 20, 2011.

Tjepkema

. Adult obesity. Health Rep 2006;17:9-25.

Bleich

Cutler

Murray

Adams

. Why is the developed world obese? Annu Rev Public Health 2008;29:273-295.

Lakdawalla

Philipson

. The growth of obesity and technological change. Econ Hum Biol 2009;7:283-293.

10.

Auld

Powell

. Economics of food energy density and adolescent body weight. Economica 2009;76:719-740.

11.

Cutler

Glaeser

Shapiro

. Why have Americans become more obese? J Econ Perspect 2003;17:93-118.

12.

WHO. Diet, Nutrition, and the Prevention of Chronic Diseases: Report of a Joint WHO/FAO Expert Consultation. Technical Report Series 916. Geneva: WHO; 2003.

13.

Rolls

Ello-Martin

Tohill

. What can intervention studies tell us about the relationship between fruit and vegetable consumption and weight management? Nutr Rev 2004;62:1-17.

14.

Tohill

Seymour

Serdula

Kettel-Khan

Rolls

. What epidemiologic studies tell us about the relationship between fruit and vegetable consumption and body weight? Nutr Rev 2004;62:365-374.

15.

Bazzano

. The high cost of not consuming fruits and vegetables. J Am Dietet Assoc 2006;106:1364-1368.

16.

WHO. The World Health Report, Reducing Risks, Promoting Healthy Lives. Geneva: WHO; 2002.

17.

Azagba

Sharaf

. Eating behavior and obesity in Canada: evidence from panel data. J Prim Care Community Health 2011. doi:10.1177/215013191141744510.1177/2150131911417445

18.

Health Canada. Canadian Nutrient File. http://www.hc-sc.gc.ca/fn-an/surveill/nutrition/commun/art-nutr-eng.php.

19.

Statistics Canada. Fruit and Vegetable Consumption, 2010 http://www.statcan.gc.ca/pub/82-625-x/2011001/article/11461-eng.htm . Accessed Month day, year.

20.

Riediger

Moghadasian

. Patterns of fruit and vegetable consumption and the influence of sex, age and socio-demographic factors among Canadian elderly. J Am Coll Nutr 2008;27:306-313.

21.

Center for Disease Control and Prevention. State-Specific Trends in Fruit and Vegetable Consumption among Adults-United States, 2000-2009. MMWR Morb Mortal Wkly Rep 2010;59:1125-1130.

22.

Colditz

Manson

Willett

Liu

. Changes in intake of fruits and vegetables in relation to risk of obesity and weight gain among middle-aged women. Int J Obes 2004;28:1569-1574.

23.

Newby

Muller

Hallfrisch

Qiao

Andres

Tucker

. Dietary patterns and changes in body mass index and waist circumference in adults. Am J Clin Nutr 2003;77: 1417-1425.

24.

Field

Gillman

Rosner

Rockett

Colditz

. Association between fruit and vegetable intake and change in body mass index among a large sample of children and adolescents in the United States. Int J Obes 2003;27:821-826.

25.

Smith-Warner

Elmer

Tharp

. Increasing vegetable and fruit intake: randomized intervention and monitoring in an at-risk population. Cancer Epidemiol Biomarkers Prev 2000;9:307-317.

26.

Patterson

Block

Rosenberger

Pee

Kahle

. Fruit and vegetables in the American diet: data from the NHANES II survey. Am J Public Health 1990;80: 1443-1449.

27.

Sturm

Datar

. Body mass index in elementary school children, metropolitan area food prices and food outlet density. Public Health 2005;119:1059-1068.

28.

Wirfalt

Hedblad

Gullberg

. Food patterns and components of the metabolic syndrome in men and women: a cross-sectional study within the Malmo Diet and Cancer cohort. Am J Epidemiol 2001;154:1150-1159.

29.

Wirfalt

Jeffery

. Using cluster analysis to examine dietary patterns: nutrient intakes, gender, and weight status differ across food pattern clusters. J Am Diet Assoc 1997;97: 272-279.

30.

Baker

Wardle

. Sex differences in fruit and vegetable intake in older adults. Appetite 2003;40:269-275.

31.

Statistics Canada. Canadian Community Health Survey (CCHS) Cycle 2.2 (Nutrition) 2004 Public Use Microdata File (PUMF) Derived and Grouped Variable Specifications. 2005. http://prod.library.utoronto.ca/datalib/codebooks/cstdli/cchs/cycle2_2/derive_e.pdf. Accessed August 20, 2011.

32.

Becker

Murphy

. Theory of rational addiction. J Polit Econ 1988;96:675-700.

33.

Ruhm

. Understanding Overeating and Obesity. NBER Working Paper No. 16149. Cambridge, MA: National Bureau of Economic Research; 2010.

34.

Koenker

Bassett

. Regression quantiles. Econometrica 1978;46:33-50.

35.

Koenker

Hallock

. Quantile regression. J Econ Perspect 2001;15:143-156.

36.

Anand

Yusuf

. Stemming the global tsunami of cardiovascular disease. Lancet 2011;377:529-532.

37.

Smith

Baghurst

. Public health implications of dietary differences between social status and occupational groups. J Epidemiol Community Health 1992;46:409-416.

38.

Yoon

Park

. Socioeconomic status in relation to obesity and abdominal obesity in Korean adults: a focus on sex differences. Obesity 2006;14:909-919.

39.

Azagba

Sharaf

. Disparities in the frequency of fruit and vegetable consumption by socio-demographic and lifestyle characteristics in Canada. Nutr J 2011;10:18.

40.

Lim

Taylor

, Factors associated with physical activity among older people-a population based study. Prev Med 2005;40:33-40.

41.

Sobal

Stunkard

. Socioeconomic status and obesity: a review of the literature. Psychol Bull 1989;105: 260-275.

42.

McLaren

. Socioeconomic status and obesity. Epidemiol Rev 2007;29:29-48.

43.

Sánchez-Vaznaugh

Kawachi

Subramanian

Sanchez

Acevedo-Garcia

. Do socioeconomic gradients in body mass index vary by race/ethnicity, gender, and birthplace? Am J Epidemiol 2009;169:1102-1112.

44.

Munafò

Tilling

Ben-Shlomo

. Smoking status and body mass index: a longitudinal study. Nicotine Tob Res 2009;11:765-771.

45.

Talmage

Role

. Nicotinic receptor-mediated effects on appetite and food intake. J Neurobiol 2002;53:618-632.

46.

Kane

Konu

. Nicotine, body weight and potential implications in the treatment of obesity. Curr Top Med Chem 2003;3:899-919.

47.

Goel

McCarthy

Phillips

Wee

. Obesity among US immigrant subgroups by duration of residence. JAMA 2004;292:2860-2867.

48.

Baum

Jr Ruhm

. Age, socioeconomic status and obesity growth. J Health Econ 2009;28:635-648.

49.

Jakicic

. The effect of physical activity on body weight. Obesity 2009;17:S34-S38.

Fruit and Vegetable Consumption and Body Mass Index

Abstract

Keywords

Introduction

Data

Method

Results

Discussion

Conclusion

Footnotes

Acknowledgements

Bios

References