This exploratory CA model illustrates how a robust mathematical modelling approach can be applied to analyze non-linear relationships associated with high-school student’s social interactions and eating behaviour. Our results suggest that when a positive environmental influence is introduced to individuals, whether manifest in inter-personal relationships and friendship ties, school health promotion initiatives, positive parental influence, reduction of competitive foods in school, or other environmental and structural changes, students will show a higher propensity to choose healthy eating habits.
The simulation illustrating the gradual increase of purchasing power and positive environmental influence represents hypothetical scenarios of the potential dynamic processes that interplay with adolescent eating behaviour and environmental influences. This simulation shows that when positive and negative environmental influences are set to the same level of P=0.01, and P
is gradually increased, students bring both healthy and unhealthy foods remains constant, and students purchasing healthy and unhealthy foods steadily increases. This suggests that adolescent students will purchase both healthy and unhealthy food when present, and when the population’s overall purchasing power permits it. Though past studies show support for students purchasing unhealthy food when it is readily available [37, 38] evidence supporting increased student purchasing of healthy foods is mixed [39–41]. As the positive social influence parameter was increased by ten percent to P=0.011, students previously purchasing unhealthy foods, transitioned to a bringing healthy state. Such a change in the system suggests that the positive social influence can encourage students to bring healthy foods from home. An example of this social phenomenon can be seen from a study that implemented a school health promotion strategy that paired up school-aged students from different grades, and allotting daily class time for instruction on healthy-living and nutrition . Students in the intervention group showed an increase in healthy-living knowledge, behaviour and attitude, and lower blood pressure measurements than the control group. It is conceivable that a change in eating behaviour may take shape through other health-oriented activities, such as physical activity . The social influence parameter may also represent the influence of other elements in the adolescent’s exogenous and endogenous environments. Social influence could also be in the form of school food program policy changes, or neighbourhood level influences such as increased accessibility to healthy foods. Future simulations may also be tailored to include multiple environmental parameters, often used in dynamic systems models, such as agent-based models.
Isolating parameters of models allowed us to experiment with different pathways of eating influence (Figures 3a, 3b, 3c). When the parameter representing transition rules between healthy and unhealthy states was increased, this change was exerted on neighbouring cells and their states changed accordingly. Our simulations in Figures 3a, 3b and 3c illustrate this effect by setting all parameters equal, except for HU and running the model over 1000 time steps. As time increased, eating behaviours clustered based on the individual’s state of healthy or unhealthy. As the healthy/uhealthy parameter gradually increased, clustering behaviours persisted, and the gap between the healthy and unhealthy populations widened. Results from the simulation show resemblance to other studies that examined the effect of social behaviour on physical activity , and obesity [23, 24, 44]. The social network behaviour observed in these studies posited that individuals exhibiting similar interests, behaviours, or traits, will ultimately form groups with others of the same distinction. The parameter settings for this simulation were specifically tailored to illustrate robustness of the model. Conversely, parameter settings could be altered to reflect an increase of influence from all individuals in the bring/purchase category.
Phase diagrams are another way to monitor how individual processes interact with each other in dynamic models. As with the previous simulations in diagrams (Figure 3a, 3b, 3c), the phase diagram isolated parameters to show how the system interacted under individual parameter permutations. Our phase diagram simulation suggests that at a certain point, the positive environmental influence ceases to insulate healthy students from adopting unhealthy behaviours, and the negative environmental influence increases the amount of unhealthy student behaviour exponentially. The use of phase diagrams in CA modelling has been used elsewhere in the literature [35, 36], and is supported as an effective utility for understanding global behaviours in dynamic models.
School-aged children and adolescents are a population who view the environment through a highly impressionable lens. The complexities of the environments they interact with interdependently affect the way their behaviours are formed and continuously adapted. Forming a better understanding of how environments interact with one another requires a holistic interpretation of individual pathways and mechanisms as dynamically linked parts of a whole. Our model suggests that, as individuals who exhibit similar behaviours socially interact, their ties grow stronger and produce a pulling effect on proximal actors in their social environment. When a change to the population’s environment takes place through the increase in the environmental parameter, the model suggests individuals will react to the respective change and their behaviours will change according to their social proximity and initial state. As research in social behaviour and health increases, the need for richer datasets will prove ever important if we are to build a better understanding of the underlying mechanisms. Illustrative of the potential for mathematical modelling as a means to answer these questions is the study from Bahr and colleagues , which simulated interventions in the population by implementing different parameter settings. Similarly, our model applies a dynamic systems modelling approach to examine a social interaction and eating behaviour, two pathways that are suspected to influence individuals consumption of food. The parameters of the model may represent a number of variables in the environment. Likewise, the environmental parameters we used in this example were meant to represent a myriad of factors occurring at multiple levels. While we know this to be true of obesogenic environments, we acknowledge that such an assumption is not likely to hold true for all populations. Rather, the intended use of the social influence parameter should be to replicate different environmental factors or interventions as single influences affecting the entire population. Examining how the population is affected by the newly introduced influences will aid with optimization of parameters and addition of more complexity. This model should serve as an impetus to refine data collection methods to be more suitable for models of this nature.
Our model does have limitations. This is the initial stage of model development, and as every developing model requires, real data is needed in order to calibrate parameter settings. Model assumptions are commonplace in mathematical modelling and as such, our model strived for simplicity in our assumptions in order to illustrate the robustness of this approach. Future implementations may include different parameters that emulate environmental influences. For instance, time one spends in school is not the only socializing that individuals partake in, and individuals may strengthen bonds in extracurricular school and non-school activities. Thus, having a more complete picture of an individual’s socialization restraints and enablers, as well as their existing social network, can help calibrate the model to a more realistic depiction ) Individual characteristics were not included for each individual as real data was not accessible to us. It is likely that eating norms and socialization vary along ethnic lines. Socio-economic status may also be a predictor of an individual’s food quality brought to school, or the individual’s ability to purchase food at school. As well, an individual’s weight status will also explain part of the eating behaviours inherent in an individual. Incorporating both contextual, and individual determinants in to model development will greatly enhance calibration of future models.