Socio-demographic attributes.

Data on age, gender (male/female), geographic location (urban/rural) and socio-economic status (SES) were collected. The latter was assessed using a method developed by the Integrated Social Indicator System for Ecuador [26], based on World Bank recommendations to develop household surveys in LMICs [27]. This method measures poverty using the “Unsatisfied Basic Needs” criteria and classifies a household as poor when it lacks access to one or more basic needs (such as education, health, nutrition, housing, urban services and employment opportunities). Using this method, participants were classified into two groups: “Poor” and “Better off”.

Anthropometric measurements.

Anthropometric measurements were carried out in duplicate by two trained researchers. Adolescents wore light clothing but no shoes during the measurements. Height was measured to the nearest 0.1 cm with a portable stadiometer (model PORTROD, Health O Meter, USA) and body weight to the nearest 0.1 kg using a digital calibrated balance (model SECA 803, Seca GmbH & CO, Hamburg, Germany). Adolescents were then classified into age- and sex-specific Body Mass Index (BMI) categories (underweight, healthy weight, overweight and obese) according to the International Obesity Task Force criteria [28, 29].

Dietary behavior.

Food intake was measured using two interview-administered 24h dietary recalls on a randomly selected weekday and weekend day. Local household measures (cups, bowls, etc.) were calibrated and used by the trained interviewers to quantify the amount of food consumed. A food composition database was compiled using databases from the US (USDA, 2012), Mexico (INNSZ, 1999), Central America (INCAP/OPS, 2012) and Peru (CENAN/INS, 2008). When detailed information on the ingredients and/or cooking methods of a recipe was unavailable, recipes were prepared in triplicate by local volunteers. The ingredients used, and their weights, were measured and averaged to obtain a final estimate for the recipe. For locally processed and pre-packed food items, food labels were used to obtain the food composition.

Data for the four components of dietary behavior were extracted from both 24h recalls. Fruit and vegetable intake were examined separately and combined. Sugary drinks included all soft drinks, fizzy drinks, energy drinks, and juices with added sugar. Breakfast was defined as a meal consumed between 5:00–7:00 am or 5:00–8:00 am for adolescents in schools with a morning or afternoon schedule, respectively. Unhealthy snacks were defined as foods rich in sodium, fat or sugar (e.g. sweets, salty snacks, and any other packaged food) eaten as a morning, afternoon or evening refreshment. Sugary drinks and fruit and/or vegetable intake were calculated as the total average daily intake (g/day) over both days to best represent habitual intake. Breakfast and intake from unhealthy snacks were expressed as a percentage of daily energy intake averaged over both days (E %/day).

Assessment of individual and environmental factors influencing dietary practices.

The conceptual framework including key individual and environmental factors for dietary behavior is illustrated in Fig 1 [21]. A self-administered questionnaire was used to quantify each factor. As no culturally appropriate and validated psychometric scales to measure these factors existed, a questionnaire was developed using qualitative data from this population [21], relevant literature [30], and the expertise of the research team. The questionnaire was piloted for understanding and readability with a group of school-going adolescents (11–15 y old) not included in this study using cognitive interviewing (a qualitative process encompassing two main techniques: think aloud interviewing and verbal probing) [31]. As part of this pretesting, the questionnaire was administered twice with a four week interval. Both single and multiple items were used to measure factors (i.e. constructs) in the framework; socio-cultural changes and lack of self-control were not measured. Items in the questionnaire were measured using 5-point interval scales. Items were recoded into the same direction so that higher construct scores corresponded to the most favorable conditions for healthy dietary practices (e.g. a high score on perceived barriers indicates fewer barriers to eat healthily). Sum scores were calculated for each construct. The outcome variables were left unchanged.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Conceptual framework for healthy dietary behavior in an Ecuadorian population.

https://doi.org/10.1371/journal.pone.0157744.g001

Construct validity analyses.

A comprehensive assessment of each scale’s quality was performed [32]. Item distribution and variation were examined using descriptive analyses. Internal consistency of each construct was examined using Cronbach’s alpha; values of alpha > 0.50 were considered acceptable as i) it was a newly developed questionnaire and ii) some constructs included only a few items [33]. Repeatability (test-retest) of the questionnaire was examined using the ICC to assess absolute agreement between single items or the sum scores of the constructs; values of ICC > 0.30 were considered to be acceptable.

Structural Equation Modeling.

Structural Equation Modeling (SEM) was conducted to statistically test the inter-relationships of constructs and their relationship with the four components of dietary behavior in our participant population [34]. SEM is a multivariate technique that allows for the modeling of a series of hypothesized relationships simultaneously. It combines aspects of factor analysis and multiple regression and allows for the inclusion of observed and unobserved (latent) variables (i.e., theoretical constructs) to determine whether the hypothesized associations are consistent with data of the participant population [35, 36]. Prior to modeling the relationship between latent variables, a measurement model was evaluated for each component of dietary behavior. This step involves a confirmatory factor analysis to confirm the relationship between the latent variables (constructs) and their indicator variables (items). The following step, i.e. the testing of the structural model, estimates the strength of the relationships between these latent variables. It also allows for examining the direct and indirect effects among the constructs in the model. Data were examined prior to modeling to ensure they met assumptions of performing a SEM and analyzed using the robust maximum likelihood procedure in LISREL 8.72 [37]. Using multi-level SEM with a small number of clusters (< 100; in our study: 34) and low ICC (<0.25, in our study: < 0.10) has been shown to produce inaccurate results. So even though a cluster sampling design was used, we applied regular SEM [38].

First, correlation coefficients were calculated between the variables of interest. All correlations were <0.70, thus multi-collinearity was not a concern in the present data [39]. Path coefficients were then estimated and the general fit of the model was assessed for each component of dietary behavior. To evaluate the goodness of fit of the model, the χ²-value together with degrees of freedom were calculated, as well as four other indices: the root mean square error of approximation (RMSEA), the normed fit index (NFI), the non-normed fit index (NNFI) and the comparative fit index (CFI). The χ²-value has traditionally been used to test the hypothesis that the relationships suggested in the model provide a plausible explanation of the data, i.e. how well the proposed model structure fits the structure in the observed data set. It is however sensitive to sample size; a large sample size increases the power to reject the models. Other fit indexes have been proposed to compensate for this problem. The RMSEA is a measure of discrepancy between the true population model and the hypothesized model with unknown but optimally chosen parameter estimates. In other words, it favors a more parsimonious model with fewer parameters. It is relatively insensitive to sample size, since it is a population-based index [40]. The CFI and NNFI both compare the fit of the hypothesized model to that of a baseline or null model, where all parameters are assumed to be independent. Values below 0.08 for RMSEA [41] and above 0.90 for NFI, NNFI and CFI [35] indicate an acceptable fit of the data to the hypothesized model. One requirement for using SEM for model testing is complete data with no missing values. To minimize exclusion of observations from the analyses, imputation of missing data values of constructs was performed using the expectation-maximization algorithm (1000 iterations) for those included in SEM analysis [42]. Separate SEM analyses were conducted for each of the dietary behaviors.

Goodness-of-fit of the models.

The confirmatory factor loadings were significant at the 0.05 level and ranged from 0.40 to 0.86. The overall fit of the measurement model was adequate (χ²(271) = 503,72, p< 0.001; RMSEA = 0.034; NNFI = 0.94; CFI = 0.95, GFI = 0.95). Using these same fit criteria, the overall fit of the structural models predicting fruit, vegetable, fruit and vegetables, unhealthy snacking, sugary drinks, and breakfast intake was also adequate (See Fig 3A–3D for details).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

(A) Individual and environmental influences on sugary drink consumption. Only statistically significant paths at P < 0.05 are shown. Goodness of Fit-statistics: χ²(303) = 630.18, p< 0.001; RMSEA = 0.038; NNFI = 0.92; CFI = 0.93, GFI = 0.93. HE: Healthy Eating. (B) Individual and environmental influences on breakfast consumption. Only statistically significant paths at P < 0.05 are shown. Goodness of Fit-statistics: χ²(303) = 630.50, p< 0.001; RMSEA = 0.038; NNFI = 0.92; CFI = 0.93, GFI = 0.94. HE: Healthy Eating. (C) Individual and environmental influences on fruit, vegetable and fruit and vegetable consumption. Only statistically significant paths at P < 0.05 are shown. Goodness of Fit-statistics: χ²(303) = 623.69, p< 0.001; RMSEA = 0.038; NNFI = 0.92; CFI = 0.93, GFI = 0.94. ^a significant pathway for vegetable intake only. HE: Healthy Eating(D) Individual and environmental influences on unhealthy snacking. Only statistically significant paths at P < 0.05 are shown. Goodness of Fit-statistics: χ²(303) = 619.19, p< 0.001; RMSEA = 0.037; NNFI = 0.92; CFI = 0.93, GFI = 0.94. HE: Healthy Eating.

https://doi.org/10.1371/journal.pone.0157744.g003

Interrelationships of the constructs.

The majority of the hypothesized inter-relationships of individual and environmental factors were confirmed and appeared to be similar for each of the dietary behaviors (See Fig 3A–3D for details). A strong and expected relationship in the model was the link between greater accessibility to healthy food in the environment and fewer perceived barriers to eating healthily. Greater accessibility was also associated with a better understanding of healthy food, meaning that greater access to healthy foods was related to perceiving healthy foods more in function of their nutritional value rather than their added chemicals (e.g. colorants, chemicals, etc.). Greater accessibility was also inversely associated with higher support from the school to eat healthily, i.e. if healthy food is more accessible, there is less support from the school. Having less permissive (i.e. stricter) parents was directly associated with higher self-efficacy, fewer perceived barriers and stronger habits for healthy eating. The relationship between parental permissiveness and stronger habits for healthy eating was the strongest in the model, suggesting that adolescents with more permissive parents had stronger unhealthy eating habits. Furthermore, greater school support was associated with higher perceived benefits of eating healthily and negatively associated with stronger habits for healthy eating.

Sugary drink intake.

Lower parental permissiveness was inversely associated with sugary drink intake, indicating that adolescents of more permissive parents consumed more sugary drinks. None of the individual factors were related to sugary drink intake (Fig 3A).

Breakfast consumption.

School support to healthful eating predicted breakfast consumption, meaning that a more supportive environment at school to eat healthily increased the consumption of breakfast in adolescents. Additionally, the perceived accessibility to healthy foods in the environment was associated with breakfast consumption indirectly through its influence on school support (Fig 3B). This suggests that a lower accessibility to healthy food is related to a more supportive school environment to eat healthily, which is in its turn positively associated with breakfast consumption.

Fruit, vegetable, and fruit and vegetable intake.

In contrast to both previous models, none of the environmental factors were directly associated with fruit, vegetable or fruit and vegetable intake (Fig 3C). Adolescents perceiving strong benefits of eating healthily ate more vegetables, though significant, the strength of this association was rather low. This association was not significant for fruit intake or fruit and vegetable intake combined. The indirect relationship between greater school support and higher vegetable intake operated through perceived benefits and in its turn, greater school support to eating healthily was influenced by lower accessibility to healthy food. Thus, adolescents’ perception of a greater accessibility to healthy food in the environment indirectly increased their vegetable intake.

Unhealthy snacking.

Greater perceived benefits of eating healthily were directly associated with more unhealthy snacking, and the indirect relationship between higher accessibility to healthy food and unhealthy snacking was facilitated by greater school support and greater perceived benefits (Fig 3D). The estimated associations indicate that lower accessibility to healthy food is related to more school support to eat healthily and a better understanding of the importance of healthy eating (perceived benefits). Surprisingly, the latter was associated with a higher intake of unhealthy snacks. Furthermore, a better understanding of healthy food was inversely and directly associated with the consumption of unhealthy snacks. This better understanding of healthy food mediated the relationship between the perceived accessibility to healthy foods and consuming unhealthy snacks. Thus, adolescents reporting low access to healthy foods were less likely to believe that the quality of healthy foods is associated with nutritional value, which may explain their higher consumption of unhealthy snacks.