plex systems models. The information gathered to inform decision makers in those cases would necessarily include both empirical data and models that identify intended effects, dependencies, and potential adverse effects. For these reasons, answers to “Why” questions may require a critique of the model used or the underlying context.
Estimates of the effectiveness or impact of an intervention within a well-defined population or setting are essential for assessing its potential to address a public health problem such as obesity. To this end, the evidence should be synthesized for each potential intervention to answer three questions. First, what is the broader context for the intervention? This question should be addressed using a systems perspective. Second, what does the evidence say about the effects of the intervention? Finally, what will be the overall public health impact of the intervention? Decision makers may need to weigh highly effective interventions that reach only a modest number of people against less effective interventions that reach the entire population at risk; they may decide to bring promising interventions to scale even in the face of incomplete information.
Reducing the proportion of the population that is overweight or obese is the ultimate goal of all obesity prevention activities; reducing individual caloric intake and/or increasing individual physical activity among at-risk persons are intermediate outcomes required to achieve this goal. Direct measurement of obesity reduction may be infeasible, or may require the passage of time or the simultaneous adoption of multiple policy and program innovations. As a result, most research studies and program evaluations focus on the intermediate outcomes, based on their logical link to the prevention of obesity.
Program evaluations and other evidence synthesis methods (e.g., a mixed-method realist review, discussed below) typically require evaluation designs that incorporate testable theories, technically referred to as logic models. The simplest logic model involves the direct chain of causation of an individual intervention. Predicting the effect of a population-level behavioral intervention on the incidence of obesity should take into account three causal links. First, does the intervention affect behavior? Second, does the change in behavior—presumably a reduction of caloric intake or an increase in physical activity—lead to the desired outcome? Third, does the change in behavior lead to other outcomes, and are they adverse or favorable? At best, the causal links that define each of these relationships are imperfect, which means that predicting the effect of an intervention will always be an inexact science.
Multiple interventions often are required to achieve improvements in population health, and studying the effects of individual interventions will be inadequate to determine their overall effectiveness. As discussed in Chapter 4, the Foresight Group