Limits...
The theory of reasoned action as parallel constraint satisfaction: towards a dynamic computational model of health behavior.

Orr MG, Thrush R, Plaut DC - PLoS ONE (2013)

Bottom Line: The reasoned action approach, although ubiquitous in health behavior theory (e.g., Theory of Reasoned Action/Planned Behavior), does not adequately address two key dynamical aspects of health behavior: learning and the effect of immediate social context (i.e., social influence).In two simulations, we show that constraint satisfaction can simultaneously incorporate the effects of past experience (via learning) with the effects of immediate social context to yield behavioral intention, i.e., intention is dynamically constructed from both an individual's pre-existing belief structure and the beliefs of others in the individual's social context.As the first known computational model of health behavior, it represents a significant advance in theory towards understanding the dynamics of health behavior.

View Article: PubMed Central - PubMed

Affiliation: Department of Epidemiology, Columbia University, New York, New York, USA. Mo2259@columbia.edu

ABSTRACT
The reasoned action approach, although ubiquitous in health behavior theory (e.g., Theory of Reasoned Action/Planned Behavior), does not adequately address two key dynamical aspects of health behavior: learning and the effect of immediate social context (i.e., social influence). To remedy this, we put forth a computational implementation of the Theory of Reasoned Action (TRA) using artificial-neural networks. Our model re-conceptualized behavioral intention as arising from a dynamic constraint satisfaction mechanism among a set of beliefs. In two simulations, we show that constraint satisfaction can simultaneously incorporate the effects of past experience (via learning) with the effects of immediate social context to yield behavioral intention, i.e., intention is dynamically constructed from both an individual's pre-existing belief structure and the beliefs of others in the individual's social context. In a third simulation, we illustrate the predictive ability of the model with respect to empirically derived behavioral intention. As the first known computational model of health behavior, it represents a significant advance in theory towards understanding the dynamics of health behavior. Furthermore, our approach may inform the development of population-level agent-based models of health behavior that aim to incorporate psychological theory into models of population dynamics.

Show MeSH
The valence structure in the empirical input sets used to generate the internal processing constraints for Simulation II.Each horizontal line represents the proportion of respondents (in the respective input set) that had a positive or negative valence (the proportion negative is, by definition, 1 minus the proportion positive). Black represents the F10V input set; red and green, F12V and F12NV, respectively. The proportion positive is captured from the zero-midpoint on the x-axis towards the left (see the dotted black vertical line); negative is to the right.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3643950&req=5

pone-0062490-g002: The valence structure in the empirical input sets used to generate the internal processing constraints for Simulation II.Each horizontal line represents the proportion of respondents (in the respective input set) that had a positive or negative valence (the proportion negative is, by definition, 1 minus the proportion positive). Black represents the F10V input set; red and green, F12V and F12NV, respectively. The proportion positive is captured from the zero-midpoint on the x-axis towards the left (see the dotted black vertical line); negative is to the right.

Mentions: Figure 2 represents the F10V, F12V and F12NV input sets separately, showing the proportion of respondents that had a positive and negative valence for each of the 14 TRA constructs. By considering these proportions as frequencies in the input to the system, we can understand to a rough approximation what the system learned when exposed to the F10V input set. In particular it learned: 1) inhibitory constraints between constructs where one is frequent and the other is not, 2) excitatory constraints between constructs where both are frequent, and 3) no constraint between constructs that are both infrequent. For, example, it should learn an inhibitory constraint between the “feel good” construct and the “get an STD” construct in the intend bank of units in the system. At a more aggregate level, we can understand what the system learned by illustrating the average exposure to valence from the inputs. The mean valence, across beliefs, was 0.29 positive for the F10V input set–it leaned heavily towards the negative valence (to not intend). For comparison, this statistic was 0.36 and 0.42 for the F12V and F12NV input sets, respectively, indicating an increase in positive beliefs about sexual behavior from 10th to 12th grade and between 12th grade virgins and non-virgins.


The theory of reasoned action as parallel constraint satisfaction: towards a dynamic computational model of health behavior.

Orr MG, Thrush R, Plaut DC - PLoS ONE (2013)

The valence structure in the empirical input sets used to generate the internal processing constraints for Simulation II.Each horizontal line represents the proportion of respondents (in the respective input set) that had a positive or negative valence (the proportion negative is, by definition, 1 minus the proportion positive). Black represents the F10V input set; red and green, F12V and F12NV, respectively. The proportion positive is captured from the zero-midpoint on the x-axis towards the left (see the dotted black vertical line); negative is to the right.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3643950&req=5

pone-0062490-g002: The valence structure in the empirical input sets used to generate the internal processing constraints for Simulation II.Each horizontal line represents the proportion of respondents (in the respective input set) that had a positive or negative valence (the proportion negative is, by definition, 1 minus the proportion positive). Black represents the F10V input set; red and green, F12V and F12NV, respectively. The proportion positive is captured from the zero-midpoint on the x-axis towards the left (see the dotted black vertical line); negative is to the right.
Mentions: Figure 2 represents the F10V, F12V and F12NV input sets separately, showing the proportion of respondents that had a positive and negative valence for each of the 14 TRA constructs. By considering these proportions as frequencies in the input to the system, we can understand to a rough approximation what the system learned when exposed to the F10V input set. In particular it learned: 1) inhibitory constraints between constructs where one is frequent and the other is not, 2) excitatory constraints between constructs where both are frequent, and 3) no constraint between constructs that are both infrequent. For, example, it should learn an inhibitory constraint between the “feel good” construct and the “get an STD” construct in the intend bank of units in the system. At a more aggregate level, we can understand what the system learned by illustrating the average exposure to valence from the inputs. The mean valence, across beliefs, was 0.29 positive for the F10V input set–it leaned heavily towards the negative valence (to not intend). For comparison, this statistic was 0.36 and 0.42 for the F12V and F12NV input sets, respectively, indicating an increase in positive beliefs about sexual behavior from 10th to 12th grade and between 12th grade virgins and non-virgins.

Bottom Line: The reasoned action approach, although ubiquitous in health behavior theory (e.g., Theory of Reasoned Action/Planned Behavior), does not adequately address two key dynamical aspects of health behavior: learning and the effect of immediate social context (i.e., social influence).In two simulations, we show that constraint satisfaction can simultaneously incorporate the effects of past experience (via learning) with the effects of immediate social context to yield behavioral intention, i.e., intention is dynamically constructed from both an individual's pre-existing belief structure and the beliefs of others in the individual's social context.As the first known computational model of health behavior, it represents a significant advance in theory towards understanding the dynamics of health behavior.

View Article: PubMed Central - PubMed

Affiliation: Department of Epidemiology, Columbia University, New York, New York, USA. Mo2259@columbia.edu

ABSTRACT
The reasoned action approach, although ubiquitous in health behavior theory (e.g., Theory of Reasoned Action/Planned Behavior), does not adequately address two key dynamical aspects of health behavior: learning and the effect of immediate social context (i.e., social influence). To remedy this, we put forth a computational implementation of the Theory of Reasoned Action (TRA) using artificial-neural networks. Our model re-conceptualized behavioral intention as arising from a dynamic constraint satisfaction mechanism among a set of beliefs. In two simulations, we show that constraint satisfaction can simultaneously incorporate the effects of past experience (via learning) with the effects of immediate social context to yield behavioral intention, i.e., intention is dynamically constructed from both an individual's pre-existing belief structure and the beliefs of others in the individual's social context. In a third simulation, we illustrate the predictive ability of the model with respect to empirically derived behavioral intention. As the first known computational model of health behavior, it represents a significant advance in theory towards understanding the dynamics of health behavior. Furthermore, our approach may inform the development of population-level agent-based models of health behavior that aim to incorporate psychological theory into models of population dynamics.

Show MeSH