Connectionism and Behavioral Clusters: Differential Patterns in Predicting Expectations to Engage in Health Behaviors

Connectionism and Behavioral Clusters: Differential Patterns in Predicting Expectations to Engage... Abstract Background The traditional approach to health behavior research uses a single model to explain one behavior at a time. However, health behaviors are interrelated and different factors predict certain behaviors better than others. Purpose To conceptualize groups of health behaviors as memory events that elicit various beliefs. A connectionist approach was used to examine patterns of construct activation related to expectations to engage in health behavior clusters. Methods A sample of lay people (N = 1,709) indicated their expectations to perform behaviors representing four clusters (Risk Avoidance, Nutrition & Exercise, Health Maintenance, and General Well-Being) and rated them on 14 constructs obtained from health behavior literature. Results Expectations to engage in all behavioral clusters were significantly and positively associated with “frequency of performance,” “perceived behavioral control,” and “anticipated regret,” and negatively associated with “effort.” However, each behavioral cluster was also predicted by activation of a unique pattern of predictors. Conclusions A connectionist approach can be useful for understanding how different patterns of constructs relate to specific outcomes. The findings provide a rationale for lay people’s cognitive schema of health behaviors, with each behavioral cluster possessing characteristics associated with distinct predictors of expectations to engage in it. These unique activation patterns point to factors that may be particularly significant for health interventions targeting different clusters of health behaviors. Connectionism, Health behavior change models, Multiple health behaviors, Behavioral expectations, Behavioral clusters, Health behavior taxonomy Introduction Several theories have been developed in order to understand and predict health-related behaviors [1] that affect morbidity and premature mortality [2]. The theories involve a variety of social-cognitive factors that explain individual differences in engagement in health behaviors ranging from exercise and fruit and vegetable consumption to smoking cessation and HIV prevention (e.g., [3, 4]). Traditionally, a specific behavior is examined in relation to a particular theory [5, 6], on the assumption that health behaviors share similar predictive paths. However, specific factors were found to be more associated with certain health behaviors than with others. For example, “perceived behavioral control” was a stronger predictor of physical activity than of illness detection behaviors and safe sex [7]. The significance of the specific combination of behavior and predictor underscores the following two major limitations of the traditional approach. The first limitation is the restricted number of predictors: using constructs derived from only one model to predict health behaviors precludes detection of factors not included in that particular model that might explain the behavior. For example, “anticipated regret” was found to predict intention and behavior beyond what the components of the Theory of Planned Behavior could explain [8, 9]. In addition, specific theories might be especially appropriate for predicting certain behaviors [10]. For example, dieting was predicted better by the Theory of Planned Behavior than by a modified Health Belief Model, while the latter predicted intentions for testicular self-examination better than the former [11, 12]. Thus, the explanatory utility of a model depends not only on its components, but also on the targeted behavior, raising questions about whether any given theory is fully generic in accounting for all health behaviors [13]. The second major limitation of the traditional approach is its focus on a single behavior [14]. There is evidence that lay people perceive similarities among health behaviors [15, 16] based on such factors as the extent to which they require effort or are habitual [15, 17]. Since several behaviors can be successfully targeted in health interventions (e.g., ref. [18]), maximizing health benefits and reducing costs, it might be more efficient to address multiple behaviors simultaneously [19]. Moreover, given the known associations among health behaviors, such as smoking and alcohol consumption [20], focusing on a single behavior can narrow the scope of the modification introduced during the process of behavior change. In sum, the “one behavior - one theory” approach does not address the interrelations between health behaviors, and its adequacy might depend on the congruence between the specific behavior and the particular theoretical model. The current study presents an alternative approach that combines the principles of connectionism [21] with the Health Behavior Taxonomy [15] and uses key social-cognitive factors to elucidate distinct predictive patterns of health behavior clusters. Connectionism According to the connectionist approach, cognitive phenomena are viewed as interactions of neuron-like units (or nodes) in a brain-like system [22], and human information processing is modeled as a network of many simple interconnected processing units that generate an output [23]. Nodes are grouped into layers, and knowledge is encoded as a set of connection weights between them, which constitute a single representation. The connection weights—the strength of the connection between nodes in the network—are stored in memory and modify themselves with experience, as a function of associative learning [24]. The context in which the association between units occurs is also likely to be encoded and affect connection weights. Although various types of networks exist, the current study utilized a simple conceptual network with one input layer and one output layer [21]. Activation of a single unit has no fixed meaning independent of the pattern of activation occurring with it across many units [25]. Within this theoretical framework, we propose that the various constructs included in different health behavior models (mostly attitudes and beliefs) reflect mental events or context-related memories, in which relevant beliefs are activated when cued by an applicable situation [21, 26]. In other words, we assumed that any health behavior under investigation provides the context that evokes an array of perceptions and beliefs relating to it. Thus, we focused on a network of associations between several constructs and a known target (health behavior). Since cognitions related to specific health behaviors can consist of signals from a large number of constructs [27], we propose that each particular behavior is associated with a unique pattern of construct activation that represents a specific network of connections encoded in memory. The connectionist approach has already been applied to predicting intentions to exercise using behavioral, normative, and control beliefs derived from the Theory of Planned Behavior [28]. The approach has also been suggested as a unifying framework that can accommodate expectancy-value and automaticity for clarifying the impact of attitudes on health behaviors [26]. We conceptualized the interactions between health behaviors and their predicting constructs in a similar way, with constructs representing the input layer and behavioral expectations representing the output layer (Fig. 1). In order to consider a broad activation pattern, a comprehensive list of relevant constructs was assembled. Key constructs were obtained from review papers of models related to health behaviors [29–32], supplemented by constructs from empirical studies [5, 33]. The list was reviewed by three Professors of Health Psychology [17]. The constructs were “perceived behavioral control” (or “self-efficacy”), “negative affect,” “positive affect,” “impact on health,” “perceived social support,” “non-health rewards,” “anticipated regret,” and “frequency of performance.” The theories from which the constructs were obtained were the Health Action Process Approach [34], Protection Motivation Theory [35], the Health Belief Model [36], and regular and extended versions of the Theory of Planned Behavior [9, 37]. Additional constructs were taken from empirical findings that differentiated between health behaviors: “effort,” “public-private,” “observable impacts,” “outcome timeframe,” “pleasure deprivation,” and “habit” [17, 38]. Fig. 1 View largeDownload slide Diagram of a two-layer network, with constructs as input units receiving activation from other networks, and clusters of expectations to engage in health behavior as output units. Dashed lines indicate low activation signals. Fig. 1 View largeDownload slide Diagram of a two-layer network, with constructs as input units receiving activation from other networks, and clusters of expectations to engage in health behavior as output units. Dashed lines indicate low activation signals. The Health Behavior Taxonomy Despite the growing recognition that targeting multiple behavioral risk factors in behavior change interventions will improve their impact on population health [39], it remains unclear which health behaviors should be targeted in one intervention. According to the connectionist approach, if a number of representations are similar in important respects, a prototypical representation will ignore their differences while preserving their common characteristics [21]. Once a representation of the general characteristics of a category has been formed, it will be elicited by a particular set of cues that bring about a common pattern of network activation [27]. The Health Behavior Taxonomy [15] provides a conceptual framework for this approach: it defines clusters of health behaviors with similar cognitive representations (Fig. 2). The Health Behavior Taxonomy was developed by utilizing a comprehensive list of 45 health behaviors, drawn from laypeople, health promoters, and a literature review, and judged as most important by a sample of laypeople. By using a card sorting technique, a representative sample generated clusters of health behaviors based on their perceived similarities. The findings revealed a cognitive schema of health behaviors, represented by a reliable taxonomy (r2monotonic = .93). The structure was then replicated using a different large sample (reliability between samples: rspearman = .95) and was also found highly stable across age, gender, education, and income (rspearman = .97 between subgroup matrices [15]). Thus, the taxonomy enables the examination and comparison of connections between a limited number of clusters that represent meaningfully-grouped behaviors and cognitive constructs that may explain and predict them. The Health Behavior Taxonomy is comprised of four main clusters: Risk Avoidance, such as avoiding smoking; Health Maintenance, such as regular teeth brushing; Nutrition & Exercise, such as fruit and vegetable consumption; and General Well-being, such as sedentary behaviors. The 45 behaviors on which the taxonomy is based encompass a wide range of health behaviors that have been used in various studies (e.g., refs. [6, 40]). They include behaviors with well-documented health effects (e.g., fruit and vegetable consumption, [41]), as well as “distal health behaviors,” such as participating in social activities [42]. Fig. 2 View largeDownload slide The Health Behavior Taxonomy (adapted from ref. [15]). Behavior abbreviations were used; full descriptions can be found in Supplementary Material 1. Fig. 2 View largeDownload slide The Health Behavior Taxonomy (adapted from ref. [15]). Behavior abbreviations were used; full descriptions can be found in Supplementary Material 1. Expectations Versus Intentions Many theoretical approaches outline precursors of behavioral intentions, considered to be the closest antecedent of behavior [35, 37]. However, empirical studies report an intention-behavior gap [43] whereby intentions represent motivation but do not necessarily translate into actual behavior. Whether or not behavior is performed at a particular point in time may reflect situational and other uncontrollable factors besides motivation, such as physical or resource limitations. Thus, in the current investigation, the outcome is the expectations to engage in behaviors, which are assumed to encompass intentions and other factors related to their realization [44, 45]. Expectations combine willingness with perceived ability to perform a behavior at a particular time, and were found to be more predictive of behavior than intentions [46]. Hypotheses In line with the above considerations, four hypotheses regarding specific connections between psychological constructs and clusters of health behaviors were formulated based on the following reasoning: 1. Since “perceived behavioral control” had high predictive value in previous theories and studies (e.g., refs. [7, 47, 48]), it was expected to be a consistent element across activation patterns, positively associated with expectations to engage in all behavioral clusters. 2. “Effort,” which is related to ego depletion [49], was also predicted to be a consistent element across activation patterns, negatively associated with expectations to engage in all behavioral clusters [7]. 3. “Anticipated regret” was hypothesized to be specifically significant in the activation pattern of expectations to engage in Risk Avoidance behaviors, due to self-blame and the immediate harsh conceivable consequences of nonadherence [50]. 4. “Social support” was expected to be part of the activation pattern of expectations to engage in behaviors related to General Well-being (e.g., activities to maintain interpersonal relationships), since the cluster includes several “distal health behaviors” that entail interactions with other people [42]. Methods Participants The participants, who took part in a larger research project on perceptions of health behaviors, were recruited by an online panel of a large survey company to represent the general population [17]. The final sample consisted of 1,709 participants who reported having no chronic illness: 1,009 women (59%) and 700 men (41%); average age 37.31 years (SD = 11.00); and average income 2.83 (SD = 1.30) on a scale from 1 (much below average) to 5 (much above average). Fifty participants (3%) had primary school or partial secondary school education, 398 (23%) had high school education, 348 (20%) had some form of higher education, 666 (39%) had a bachelor’s degree, and 246 (14%) had a master’s degree or higher (1 unreported). Measures Perceptions of health behaviors questionnaire [17] Twelve behaviors representing all four clusters of the Health Behavior Taxonomy [15] were randomly selected for each participant from the 45 behaviors that were used to develop the taxonomy [17]. Participants were asked to rate each behavior on seven-point Likert-type scales representing 14 different constructs (e.g., “effort”) presented in random order (Table 1). The behavior selection procedure was intended to avoid participation overload (rating 45 × 14 scales). Table 1 Constructs associated with health behaviours Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) View Large Table 1 Constructs associated with health behaviours Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) View Large Behavioral expectations The question: “How likely are you to perform the following behavior (or keep performing it)?” was presented regarding each of the 12 behaviors of the Perceptions of Health Behaviors Questionnaire. Answers were given on a scale ranging from very unlikely [1] to very likely [7]. Demographic items Participants were asked to indicate their gender, age, income, and education. Procedure The study was approved by the university Institutional Review Board. To reduce the risk of participant fatigue and increase validity, the number of behaviors included in the Perceptions of Health Behaviors Questionnaire was limited to 12 (out of 45). A small pilot study revealed that the roughly 100 rating scales were clear and easy to complete in 10–15 min. In addition, 206 respondents were excluded from the analysis since it took them less than 15 s to judge 12 behaviors, thus their answers were suspected of being invalid [17]. Data were collected online via a designated website (Demographic Items were recorded at registration to the online panel). The study was presented as dealing with behaviors affecting health. After receiving informed consent, the Perceptions ofhealth behaviors and the Behavioral expectations Questionnaires were administered consecutively. Respondents were paid a nominal sum of $2 for their participation. Statistical Analysis For each participant, average ratings of the constructs and behavioral expectations were calculated using the behaviors in each cluster. The items representing each cluster were not necessarily identical for all participants due to the random selection process. Afterwards, scores were averaged across participants, resulting in 15 scores for each cluster representing its mean ratings on the 14 constructs and behavioral expectations. Repeated measures analysis of variance was used to test differences between expectations to engage in the four main clusters of health behaviors: Health Maintenance, Nutrition & Exercise, Risk Avoidance, and General Well-being. Neural networks employ an array of possible functions intended to find the best mathematical solution to connect inputs and outputs, frequently resulting in extremely complex equations. Subsequently, linear approximations are often presented in the results. A widely used algorithm in neural network analysis is Backpropagation, which consists of a training set that includes many examples of inputs and their desired outputs [51]. The weights between inputs and outputs of the network to be trained are initially set to random values and then members of the training set are repeatedly exposed and compared to the network. Consequently, the weights are slightly adjusted in the direction that would bring the output values of the network closer to the values of the desired output. After many repetitions of this process, the network learns to produce a result that corresponds to the desired output for each input, which may then be generalized to inputs and outputs that were not in the training set [51]. The current analyses were conducted in MATLAB using the Neural Network Fitting Tool. The networks were trained by the Levenberg-Marquardt backpropagation algorithm and the sample was divided into three subsets, corresponding to the different phases of the process: Training, Validation, and Testing. The Training subset computes a gradient and updates the network weights and biases, so that the network is adjusted according to its error. The Validation subset measures the network’s generalization and stops the training when the generalization stops to improve. The error of the validation set is monitored during the training process, and the network weights and biases are saved at the point of minimum error, before it begins to rise due to data over-fitting. The Testing subset provides a completely independent measure of network performance and generalization. Once the network development has been complete, it is possible to use the entire sample to evaluate the performance of the modeled network [52]. We used 70% of the sample for training, 15% for validation, and 15% for testing. In addition, the software uses an algorithm of multilayer networks that includes a layer of hidden neurons, which are activated by nodes of preceding layers and send signals to succeeding layers [26]. No hidden layers were assumed in the current study, thus all inputs were expected to affect a single node in the hidden layer. However, we also examined a solution driven by the Theory of Planned Behavior that defined three nodes in the hidden layer [28] as well as the possibility of separate nodes in the hidden layer for each individual construct, resulting in a total of 14 nodes (see Supplementary Material 2). Since the goal of the current study was to demonstrate the feasibility and utility of a connectionist approach to researchers in the field of health behavior [53], the more common Regression Model is primarily presented instead of a Neural Network implementation (e.g., Parallel Distributed Processing, [54]). Thus, although the function of the units in the network is often assumed to be sigmoidal [21], the strengths of the links between nodes were represented by the regression equation coefficients, which provide a lower bound approximation to the actual activation strength [55]. Moreover, when a full Neural Network analysis was performed, it was not found superior to Regression Models in terms of explained variance of expectations to engage in behavioral clusters (see Supplementary Material 2). In addition, multicollinearity could have posed a computational and logical problem in our study, due to the large number of predicting variables and their possible overlap in explained variance. Thus, Statistical (stepwise) Multiple Regression, a “model building” procedure [56], was performed separately for each cluster, with the 14 constructs of the Perceptions of Health Behaviors Questionnaire as predictors. This method was used to enable empirical selection of a set of variables that can best explain engagement expectations in each cluster. Only predictors contributing at least 1% to the total explained variance of the model, that is, part of the activation pattern, are presented in the results. Results The correlations between demographic variables and expectations to engage in behavioral clusters were negligible (see Supplementary Material 3). Expectations to engage in the various health behavior clusters differed significantly (F[3,5124] = 217.79, p < .0001, partial η2 = .11). According to Bonferroni post-hoc tests (p < .0001), expectations to engage in the Risk Avoidance cluster were significantly higher (M = 5.97, SD = 1.05) than the Health Maintenance cluster (M = 5.80, SD = 0.99), which in turn was significantly higher than the General Well-being cluster (M = 5.51, SD = 1.10), which was higher than Nutrition & Exercise (M = 5.33, SD = 1.34). Pearson correlations between constructs and behavioral expectations for the main cluster of behaviors can be found in Supplementary Material 4. The patterns of activation related to each behavioral cluster were manifested in the regression equation coefficients, which indicate the strength of the association between the input and output nodes (Table 2). As predicted, expectations to engage in all clusters of health behaviors were positively associated with “perceived behavioral control” and negatively with “effort” (hypotheses 1 and 2, respectively). In addition, all clusters were predicted primarily by “frequency of performance” as well as by “anticipated regret” (Table 2). Table 2 Regression models representing patterns of activation for expectations to engage in health behavior clusters using 14 constructsa Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 R2 Accumulative explained variance. β Standardized coefficients for predictors contributing at least 1% to the total explained variance. In all models, R2 changes and coefficients are significant at p < .001. aA table enabling easier comparison of coefficients of specific predictors across clusters appears in Supplementary material 5. View Large Table 2 Regression models representing patterns of activation for expectations to engage in health behavior clusters using 14 constructsa Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 R2 Accumulative explained variance. β Standardized coefficients for predictors contributing at least 1% to the total explained variance. In all models, R2 changes and coefficients are significant at p < .001. aA table enabling easier comparison of coefficients of specific predictors across clusters appears in Supplementary material 5. View Large Furthermore, each cluster displayed a distinct activation pattern, which included not only the aforementioned constructs, but also an array of specific predictors with varying connection strengths. The explanatory power of each construct represents its specific contribution to the activation pattern. Expectations to engage in the Risk Avoidance cluster were explained primarily by “frequency of performance” (21% of the variance), followed by “perceived behavioral control” (an additional 8%), and, as predicted by hypothesis 3, by “anticipated regret” (an additional 4%). Expectations to engage in Nutrition & Exercise behaviors were predicted mainly by “frequency of performance” (15% explained variance), followed by “effort” (an additional 10%) and “positive affect” (an additional 5%). Expectations to engage in Health Maintenance behaviors were explained by “frequency of performance” (21% of the variance), followed by “impact on health” (an additional 6%) and “effort” (an additional 5%). Finally, expectations to engage in the General Well-being cluster were also predicted primarily by “frequency of performance” (19%), followed by “perceived behavioral control” (an additional 10%), and “effort” (an additional 3%). Hypothesis 4 was also confirmed: “social support” significantly explained variance related to expectations to engage in General Well-being (Table 2). In general, expectations to engage in the Risk Avoidance cluster were best explained by the constructs measured in the current study (41% explained variance using multiple regression, 45% using neural networks), followed by Health Maintenance and General Well-being (both with 37% explained variance using multiple regression, 38% using neural networks), and Nutrition & Exercise (34% explained variance using multiple regression, 37% using neural networks). These findings are in line with previous findings [57], and an example of the results of the different stages of the neural network process and the linear approximation for the Risk Avoidance behavioral cluster is presented in Table 3. Table 3 Explained variance for each step of the neural network analysis, for the overall model, and linear approximations regarding the Risk Avoidance behavioral cluster Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 View Large Table 3 Explained variance for each step of the neural network analysis, for the overall model, and linear approximations regarding the Risk Avoidance behavioral cluster Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 View Large Discussion This study serves as proof of concept for the usefulness of a connectionist approach to the investigation of health behaviors, in line with previous research [26, 28]. The goal of the connectionist approach is to empirically find an optimal set of weights for a network. In order to accomplish that, sigmoidal functions are used to create neuron-like connections that simulate the human brain, consequently increasing accuracy of predictions. The current study modeled the unique mental processes involved in expectations to engage in health behaviors. The findings add to the current knowledge by uncovering the activation patterns of cognitive constructs related to clusters of health behaviors, reflecting relevant beliefs associated with them [21, 26]. In other words, the findings delineate the differential effects of predictors of behavioral expectations, which signify the beliefs and perceptions attached to each particular group of health behaviors and disclose the unique activation patterns characterizing each behavioral cluster. Moreover, in line with the connectionist assumption regarding the network’s ability to generalize [27], the current findings provide support for the utility of addressing clusters of related behaviors [15], since the health behaviors in each cluster possess schema-consistent information [58]. In addition, the findings indicate that expectations to engage in health behavior clusters can be predicted by unique patterns of theoretical constructs. This underscores the importance of considering similarities and differences among health behaviors beyond previous studies that had shown the existence of cognitive health behavior clusters and described the nature of differences among them [15–17]. Expectations to engage in Risk Avoidance behaviors were best explained by the constructs measured in the current study. Similar findings were reported for these behaviors compared with physical activity, dietary behaviors, and abstinence [7]. We speculate that behaviors such as smoking might be better predicted due to a combination of stronger learned stimulus-response association and habit-generated impulses [59]. It is important to note that a connectionist approach maintains that the activation of a single unit, which may contribute to different representations [60], should be interpreted in the context of a pattern that symbolizes a semantically meaningful mental state [25]. Subsequently, for example, whether or not “positive affect” is a significant predictor would be best understood when the entire construct pattern is considered. However, for reasons of clarity and simplicity, the following discussion will deconstruct the activation patterns and explore the roles of individual constructs that were found relevant for all behavioral expectations, as well as of those with unique significance for specific clusters. Common Explanatory Factors Expectations to engage in all clusters were predicted by higher levels of “frequency of performance”, “perceived behavioral control” and “anticipated regret,” and lower levels of “effort.” Frequency perceptions (how frequently behaviors should be performed) were the primary predictor of expectations to engage in all clusters of health behaviors, consistent with previous findings that past behavior had the highest predictive value for exercise [61], and that “frequency of performance” was a key characteristic of health behaviors in general [5]. It may represent a cue-response relationship component [62], as well as habitual automaticity and strength [63], signifying the influence of frequent performance on future behavior. The fact that “perceived behavioral control” was also a significant predictor of expectations to engage in all behavioral clusters underscores its importance and is in line with the majority of theories of health behaviors [1]. However, “effort” and “anticipated regret,” which were also found to predict expectations to engage in all health behavior clusters, are absent from most health behavior models. The degree of effort required to perform a behavior can be partly analogous to “perceived barriers” in the Health Belief Model [36] and to “self-efficacy” [64]. Considering its consistent role as a predictor of expectations and behavior [65], “effort” may add significant explanatory power to health behavior models. “Anticipated regret” has already been shown to have a unique contribution over and above the components of the Theory of Planned Behavior [8, 66], and the current findings support the utility of including this construct in health behavior theories. Together, all four constructs play a key role in the activation patterns of expectations to engage in health behaviors. However, their relative strengths vary according to relevance to the predicted behavioral cluster. For example, “anticipated regret” was a stronger predictor of the Risk Avoidance cluster (e.g., wearing a seatbelt) than the other clusters. This may result from the higher emotionality associated with the immediate grave outcomes of failing to avoid risks, reflected also in the high ratings of “anticipated regret” for Risk Avoidance behaviors compared to the other clusters. Thus, the strong motivation to engage in these health behaviors [67] can be due to the wish to avoid both the physical and emotional consequences of nonadherence [50]. In addition, “perceived behavioral control” was a stronger predictor of expectations to engage in General Well-being behaviors (e.g., spending quality time with one’s partner) compared to the other clusters. This may illustrate the perceived challenge many people face in balancing activities related to work, family, and personal life [68] that affects health and well-being [69]. Finally, the Nutrition & Exercise cluster was more strongly predicted by “effort” than by “perceived behavioral control,” which may reflect the perceived difficulty entailed in overcoming external temptations and controlling one’s diet [43]. Unique Explanatory Factors In addition to the different predictive weights of the common explanatory factors, distinct units in the activation pattern of expectations to engage in specific behavioral clusters were uncovered, corresponding to their different features and purposes. For example, “positive affect” was an important predictor of expectations to engage in Nutrition & Exercise behaviors, but not in Risk Avoidance or General Well-being. This is consistent with previous findings of stronger associations between attitudes and intentions to engage in dietary behaviors compared to risky health behaviors [7]. Further investigation is needed to determine the processes by which food-related emotional experiences develop into this association. In addition, behaviors related to Health Maintenance (e.g., attending medical check-ups) were more strongly predicted by cognitive appraisals regarding their “impact on health” compared to the other behavioral clusters. This can represent a medicalized danger-control approach [70], in which medical attention is sought as a reaction to health threats. It is also noteworthy that perceptions of “impact on health” were associated with all clusters except Nutrition & Exercise. This cluster was found highest on “impact on health” while displaying the lowest within-cluster variance compared to other clusters [17]—a finding that may reflect a consensus among lay people regarding the major influence of nutritional habits on health, which might have attenuated its predictive power here. Implications The connectionist approach may be helpful in discovering meaningful patterns that can facilitate performance expectations and engagement in health behaviors. The results reveal common and unique factors that could be relevant to interventions targeting different clusters of health behaviors. Of special interest is the finding that perceived “frequency of performance” was the primary predictor of expectations to engage in all behavioral clusters. This underscores the key role of habit formation in lay people’s cognitions and supports defining it as a target in health interventions [71]. “Effort” appears particularly important with regard to Nutrition & Exercise behaviors. Consequently, interventions aimed at modifying dietary habits would benefit from reducing effort levels, such as providing easier access to healthy foods [72]. Finally, since perceptions regarding the “impact on health” do not predict nutrition behaviors, the introduction of such information in interventions may be superfluous. In sum, the findings of the present study may help conceptualize how to design health research and interventions by providing a framework for addressing patterns of associations between cognitive constructs and clusters of interrelated health behaviors. Limitations The current study serves as proof of concept of the utility of connectionism in the field of health behavior theorizing and research. It is only a first step in mapping the activation patterns related to expectations to engage in clusters of health behaviors, which should be further examined with regard to actual engagement in behaviors, in different contexts, cultures, and populations [26, 73]. For example, the fear of injury among patients with chronic low back pain [74] may amplify the role of “anticipated regret” in the activation pattern related to exercise, due to strong associations between certain movements and physical pain. It is important to note that the results obtained by neural network analysis and regression analysis were comparable, although the former method was found to be superior in a previous study [28]. A possible explanation for this discrepancy might be related to measurement: the predictors in the current study were measured by single items, whereas the advantages embodied in the complexity of the neural network method may require assessing theoretical constructs using multiple items. Consequently, future studies employing neural networks should consider examining constructs related to health behaviors using multiple indicators. In addition, utilizing more complex network analyses that incorporate hidden layers corresponding to the constructs measured by multiple indicators may increase prediction accuracy for both unique and common variability [28], since they would enable the generation of a correct output pattern in case more than a simple reconstruction of the input pattern is required [25]. The use of an empirical approach such as connectionism to study health behaviors is limited since the predictors that emerge as significant depend on the variables included in the study and their multicollinearity. However, the selection of variables from health behavior theories increases the confidence in the predictors. Further, the risk of spuriousness is reduced since multiple regression has the advantage of controlling for the variance explained by the other variables when calculating each predictor’s coefficient. In general, statistical regression is considered a model-building procedure rather than a model-testing technique [56]. Future research on health behaviors should continue to integrate empirical and theoretical efforts by applying the connectionist approach to theoretically-driven constructs and modifying theories according to empirically-driven findings. Although numerous theoretical constructs were included in the current study, further attempts should be made to examine the contribution of additional predictors related to health behaviors such as dispositional optimism [75]. These efforts might be guided by theoretical considerations as well as particular behavioral circumstances or populations under investigation. As is common in the health behavior literature, the current analysis was interpreted to be indicative of a causal relationship (constructs predict expectations) despite the use of correlational data (e.g., ref. [76]). However, to determine causality, the effects of interventions on the modification of specific behaviors must be examined in randomized controlled trials. In addition, although results of online studies were found to be equivalent to those obtained using other methods of data collection [77], it is important to replicate the current finding by using additional samples and methods. The use of behavioral expectations as an outcome in the current study was intended to partly overcome the intentions-behavior gap [43]. However, since expectations can fluctuate according to various moderators [78] and may be partly behavior dependent [43], comparing them with actual engagement in behavioral clusters is the next required step. This would allow examining whether or not the actual behaviors display similar patterns to those uncovered in this study. Future studies should also explore whether self-regulatory mechanisms that predict behavior [79], such as action control [80], moderate activation patterns associated with specific health behavior clusters. Conclusions Multiple health behavior research is at an early stage [81] and is considered one of the most important challenges for behavioral medicine [39]. The potential benefits of addressing multiple behaviors simultaneously in health interventions have led to a sharp increase in studies reporting them [82]. While the optimal way to design such interventions remains unknown, using lay people’s cognitive schema of health behaviors as a lead for intervention development seems promising [15]. The connectionist approach applied in the current study suggests re-examining the relationships between theoretical components of behavioral change and desired outcomes. It offers a new perspective on the diverse effects of predictive constructs as well as on possible moderators and barriers to engagement in clusters of health behaviors. It also emphasizes patterns of activation rather than individual constructs and highlights common and unique explanatory factors. The differential importance of specific constructs for distinct behavioral clusters draws attention to the essential features of those clusters and to the need to consider them when planning health prevention and promotion programs. Supplementary Material Supplementary material is available at Annals of Behavioral Medicine online. Acknowledgments This research was funded by the Israel Science Foundation, grant number 257/11. Compliance with Ethical Standards Statements The study was approved by the Institutional Review Board of Tel Aviv University. Informed consent was given prior to participation in the study. G. Nudelman and S. Shiloh developed the study concept and design. G. Nudelman developed the study website and performed the data analysis. G. Nudelman interpreted the results and wrote the first draft of the manuscript, and S. Shiloh provided critical revisions. Both authors approved the final version of the manuscript for submission. Conflict of Interest We declare adherence to ethical standards and that there is no actual or potential conflicts of interest with the Israel Science Foundation, which sponsored the research. References 1. Conner M , Norman P. Predicting Health Behaviour . UK : McGraw-Hill Education ; 2005 . 2. Loef M , Walach H . The combined effects of healthy lifestyle behaviors on all cause mortality: a systematic review and meta-analysis . Prev Med . 2012 ; 55 ( 3 ): 163 – 170 . doi:10.1016/j.ypmed.2012.06.017 Google Scholar CrossRef Search ADS PubMed 3. Godinho CA , Alvarez MJ , Lima ML , Schwarzer R . Health messages to promote fruit and vegetable consumption at different stages: a match-mismatch design . Psychol Health . 2015 ; 30 ( 12 ): 1410 – 1432 . doi:10.1080/08870446.2015.1054827 Google Scholar CrossRef Search ADS PubMed 4. Fisher WA , Fisher JD , Shuper PA . Social psychology and the fight against AIDS: an information-motivation-behavioral skills model for the prediction and promotion of health behavior change . Adv Exp Soc Psychol . 2014 ; 50 : 105 – 193 . doi:10.1016/B978-0-12-800284-1.00003-5 Google Scholar CrossRef Search ADS 5. McEachan RR , Lawton RJ , Conner M . Classifying health-related behaviours: exploring similarities and differences amongst behaviours . Br J Health Psychol . 2010 ; 15 ( Pt 2 ): 347 – 366 . doi:10.1348/135910709X466487 Google Scholar CrossRef Search ADS PubMed 6. Sniehotta FF , Presseau J , Allan J , Araújo-Soares V . “You can’t always get what you want”: a novel research paradigm to explore the relationship between multiple intentions and behaviours . Appl Psychol Health Well Being . 2016 ; 8 ( 2 ): 258 – 275 . doi:10.1111/aphw.12071 Google Scholar CrossRef Search ADS PubMed 7. McEachan RRC , Conner M , Taylor NJ , Lawton RJ . Prospective prediction of health-related behaviours with the theory of planned behaviour: a meta-analysis . Health Psychol Rev . 2011 ; 5 : 97 – 144 . doi:10.1080/17437199.2010.521684 Google Scholar CrossRef Search ADS 8. Sandberg T , Conner M . Anticipated regret as an additional predictor in the theory of planned behaviour: a meta-analysis . Br J Soc Psychol . 2008 ; 47 ( Pt 4 ): 589 – 606 . doi:10.1348/014466607X258704 Google Scholar CrossRef Search ADS PubMed 9. Sandberg T , Hutter R , Richetin J , Conner M . Testing the role of action and inaction anticipated regret on intentions and behaviour . Br J Soc Psychol . 2016 ; 55 ( 3 ): 407 – 425 . doi:10.1111/bjso.12141 Google Scholar CrossRef Search ADS PubMed 10. Garcia K , Mann T . From ‘I Wish’ to ‘I Will’: social-cognitive predictors of behavioral intentions . J Health Psychol . 2003 ; 8 ( 3 ): 347 – 360 . 10.1177/13591053030083005 Google Scholar PubMed 11. McClenahan C , Shevlin M , Adamson G , Bennett C , O’Neill B . Testicular self-examination: a test of the health belief model and the theory of planned behaviour . Health Educ Res . 2007 ; 22 ( 2 ): 272 – 284 . doi:10.1093/her/cyl076 Google Scholar CrossRef Search ADS PubMed 12. Nejad LM , Wertheim EH , Greenwood K . Comparison of the health belief model and the theory of planned behavior in the prediction of dieting and fasting behavior . E J Appl Psychol . 2005 ; 1 : 63 – 74 . doi:10.7790/ejap.v1i1.10 Google Scholar CrossRef Search ADS 13. Noar SM , Zimmerman RS . Health Behavior Theory and cumulative knowledge regarding health behaviors: are we moving in the right direction ? Health Educ Res . 2005 ; 20 ( 3 ): 275 – 290 . doi:10.1093/her/cyg113 Google Scholar CrossRef Search ADS PubMed 14. Fleig L , Ngo J , Roman B et al. Beyond single behaviour theory: adding cross-behaviour cognitions to the health action process approach . Br J Health Psychol . 2015 ; 20 ( 4 ): 824 – 841 . doi:10.1111/bjhp.12144 Google Scholar CrossRef Search ADS PubMed 15. Nudelman G , Shiloh S . Mapping health behaviors: constructing and validating a common-sense taxonomy of health behaviors . Soc Sci Med . 2015 ; 146 : 1 – 10 . doi:10.1016/j.socscimed.2015.10.004 Google Scholar CrossRef Search ADS PubMed 16. Vickers RR Jr , Conway TL , Hervig LK . Demonstration of replicable dimensions of health behaviors . Prev Med . 1990 ; 19 ( 4 ): 377 – 401 . doi:10.1016/0091-7435(90)90037-K Google Scholar CrossRef Search ADS PubMed 17. Nudelman G , Shiloh S . Understanding behavioural clusters: Identifying differences between clusters of health behaviours on key constructs . Psychol Health . 2016 ; 31 ( 12 ): 1375 – 1390 . doi:10.1080/08870446.2016.1208822 Google Scholar CrossRef Search ADS PubMed 18. Prochaska JO , Butterworth S , Redding CA et al. Initial efficacy of MI, TTM tailoring and HRI’s with multiple behaviors for employee health promotion . Prev Med . 2008 ; 46 ( 3 ): 226 – 231 . doi:10.1016/j.ypmed.2007.11.007 Google Scholar CrossRef Search ADS PubMed 19. Prochaska JJ , Prochaska JO . A Review of multiple health behavior change interventions for primary prevention . Am J Lifestyle Med . 2011 ; 5 : (3) 208 – 221 . doi: 10.1177/1559827610391883 Google Scholar CrossRef Search ADS 20. Hofstetter H , Dusseldorp E , van Empelen P , Paulussen TW . A primer on the use of cluster analysis or factor analysis to assess co-occurrence of risk behaviors . Prev Med . 2014 ; 67 : 141 – 146 . doi:10.1016/j.ypmed.2014.07.007 Google Scholar CrossRef Search ADS PubMed 21. Smith ER . What do connectionism and social psychology offer each other ? J Pers Soc Psychol . 1996 ; 70 ( 5 ): 893 – 912 . doi:10.1037/0022-3514.70.5.893 Google Scholar CrossRef Search ADS PubMed 22. Nelson R . Expanding the role of connectionism in SLA theory . Lang Learn . 2013 ; 63 : 1 – 33 . doi:10.1111/j.1467-9922.2012.00699.x Google Scholar CrossRef Search ADS 23. Joanisse MF , McClelland JL . Connectionist perspectives on language learning, representation and processing . Wiley Interdiscip Rev Cogn Sci . 2015 ; 6 ( 3 ): 235 – 247 . doi:10.1002/wcs.1340 Google Scholar CrossRef Search ADS PubMed 24. Olson MA , Fazio RH . Implicit attitude formation through classical conditioning . Psychol Sci . 2001 ; 12 ( 5 ): 413 – 417 . doi:10.1111/1467-9280.00376 Google Scholar CrossRef Search ADS PubMed 25. McClelland JL , Rumelhart DE. The PDP Research Group: Parallel Distributed Processing: Explorations in the Microstructure of Cognition . Cambridge, MA : MIT Press ; 1986 . 26. Lowe R , Norman P . Attitudinal approaches to health behavior: integrating expectancy-value and automaticity accounts . Soc Personal Psychol Compass . 2013 ; 7 : 572 – 584 . doi:10.1111/spc3.12046 Google Scholar CrossRef Search ADS 27. Smith ER . Distributed connectionist models in social psychology . Soc Personal Psychol Compass . 2009 ; 3 : 64 – 76 . doi:10.1111/j.1751-9004.2008.00160.x Google Scholar CrossRef Search ADS 28. Lowe R , Bennett P , Walker I , Milne S , Bozionelos G . A connectionist implementation of the theory of planned behavior: association of beliefs with exercise intention . Health Psychol . 2003 ; 22 ( 5 ): 464 – 470 . doi:10.1037/0278-6133.22.5.464 Google Scholar CrossRef Search ADS PubMed 29. Armitage CJ , Conner M . Social cognition models and health behaviour: a structured review . Psychol Health . 2000 ; 15 : 173 – 189 . doi:10.1080/08870440008400299 Google Scholar CrossRef Search ADS 30. Noar SM , Chabot M , Zimmerman RS . Applying health behavior theory to multiple behavior change: considerations and approaches . Prev Med . 2008 ; 46 ( 3 ): 275 – 280 . doi:10.1016/j.ypmed.2007.08.001 Google Scholar CrossRef Search ADS PubMed 31. Glanz K , Bishop DB . The role of behavioral science theory in development and implementation of public health interventions . Annu Rev Public Health . 2010 ; 31 : 399 – 418 . doi:10.1146/annurev.publhealth.012809.103604 Google Scholar CrossRef Search ADS PubMed 32. Sutton S . Health behavior: psychosocial theories . In: Smelser NJ , Baltes PB , eds. International Encyclopedia of the Social and Behavioral Sciences . Oxford, UK : Elsevier ; 2001 : 6499 – 6506 . 33. Kiviniemi MT , Voss-Humke AM , Seifert AL . How do I feel about the behavior? The interplay of affective associations with behaviors and cognitive beliefs as influences on physical activity behavior . Health Psychol . 2007 ; 26 ( 2 ): 152 – 158 . doi:10.1037/0278-6133.26.2.152 Google Scholar CrossRef Search ADS PubMed 34. Schwarzer R . Modeling health behavior change: how to predict and modify the adoption and maintenance of health behaviors . Appl Psychol . 2008 ; 57 : 1 – 29 . doi:10.1111/j.1464-0597.2007.00325.x 35. Rogers RW , ed. Cognitive and Physiological Processes in Fear Appeals and Attitude Change: A Revised Theory of Protection Motivation . London, England : Guildford ; 1983 . 36. Janz NK , Becker MH . The Health Belief Model: a decade later . Health Educ Q . 1984 ; 11 ( 1 ): 1 – 47 . doi:10.1177/109019818401100101 Google Scholar CrossRef Search ADS PubMed 37. Ajzen I . The theory of planned behavior . Organ Behav Hum Decis Process . 1991 ; 50 : 179 – 211 . doi:10.1016/0749-5978(91)90020-T Google Scholar CrossRef Search ADS 38. Shiloh S , Peretz G , Iss R , Kiedan R . Recovery attributions: explicit endorsement of biomedical factors and implicit dominance of psycho-social factors . J Behav Med . 2007 ; 30 ( 3 ): 243 – 251 . doi:10.1007/s10865-007-9100-0 Google Scholar CrossRef Search ADS PubMed 39. Evers KE , Quintiliani LM . Advances in multiple health behavior change research . Transl Behav Med . 2013 ; 3 ( 1 ): 59 – 61 . doi:10.1007/s13142-013-0198-z Google Scholar CrossRef Search ADS PubMed 40. Bergkvist L , Rossiter JR . The predictive validity of multiple-item versus single-item measures of the same constructs . J Mark Res . 2007 ; 44 : 175 – 184 . doi:10.1509/jmkr.44.2.175 Google Scholar CrossRef Search ADS 41. Wang X , Ouyang Y , Liu J et al. Fruit and vegetable consumption and mortality from all causes, cardiovascular disease, and cancer: systematic review and dose-response meta-analysis of prospective cohort studies . BMJ . 2014 ; 349 : g4490 . doi:10.1136/bmj.g4490 Google Scholar CrossRef Search ADS PubMed 42. Ziegelmann JP , Knoll N . Future directions in the study of health behavior among older adults . Gerontology . 2015 ; 61 ( 5 ): 469 – 476 . doi:10.1159/000369857 Google Scholar CrossRef Search ADS PubMed 43. Sheeran P . Intention—behavior relations: a conceptual and empirical review . Eur Rev Soc Psychol . 2002 ; 12 : 1 – 36 . doi:10.1080/14792772143000003 Google Scholar CrossRef Search ADS 44. Sheppard BH , Hartwick J , Warshaw PR . The theory of reasoned action: A meta-analysis of past research with recommendations for modifications and future research . J Consum Res . 1988 ; 15 ( 3 ) 325 – 343 . doi:10.1086/209170 Google Scholar CrossRef Search ADS 45. Venkatesh V , Brown SA , Maruping LM , Bala H . Predicting different conceptualizations of system use: the competing roles of behavioral intention, facilitating conditions, and behavioral expectation . MIS Q . 2008 ; 33 : 483 – 502 . doi:10.1016/j.chb.2013.07.018 Google Scholar CrossRef Search ADS 46. Armitage CJ , Norman P , Alganem S , Conner M . Expectations are more predictive of behavior than behavioral intentions: evidence from two prospective studies . Ann Behav Med . 2015 ; 49 ( 2 ): 239 – 246 . doi:10.1007/s12160-014-9653-4 Google Scholar CrossRef Search ADS PubMed 47. Cooke R , French DP . How well do the theory of reasoned action and theory of planned behaviour predict intentions and attendance at screening programmes? A meta-analysis . Psychol Health . 2008 ; 23 ( 7 ): 745 – 765 . doi:10.1080/08870440701544437 Google Scholar CrossRef Search ADS PubMed 48. Topa G , Moriano JA . Theory of planned behavior and smoking: meta-analysis and SEM model . Subst Abuse Rehabil . 2010 ; 1 : 23 – 33 . doi:10.2147/SAR.S15168 Google Scholar CrossRef Search ADS PubMed 49. Baumeister RF , Vohs KD , Tice DM . The strength model of self-control . Curr Dir Psychol Sci . 2007 ; 16 : 351 – 355 . doi:10.1111/j.1467-8721.2007.00534.x Google Scholar CrossRef Search ADS 50. Connolly T , Zeelenberg M . Regret in decision making . Curr Dir Psychol Sci . 2002 ; 11 : 212 – 216 . doi:10.1111/1467-8721.00203 Google Scholar CrossRef Search ADS 51. Garson J . Connectionism . In: Zalta EN , ed. The Stanford Encyclopedia of Philosophy . Stanford University ; 2016 . 52. MathWorks . Divide Data for Optimal Neural Network Training. Retrieved November 7, 2017 from https://www.mathworks.com/help/nnet/ug/divide-data-for-optimal-neural-network-training.html. 53. Yermolayeva Y , Rakison DH . Connectionist modeling of developmental changes in infancy: approaches, challenges, and contributions . Psychol Bull . 2014 ; 140 ( 1 ): 224 – 255 . doi:10.1037/a0032150 Google Scholar CrossRef Search ADS PubMed 54. McClelland JL. Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises . CA : Palo Alto, Stanford University Press ; 2015 . 55. Rawlings JO , Pantula SG , Dickey DA. Applied Regression Analysis: A Research Tool . New York, NY : Springer-Verlag New York Inc .; 2001 . 56. Tabachnick BG , Fidell LS. Using Multivariate Statistics . 6th ed . Boston : Pearson ; 2013 . 57. Armitage CJ , Conner M . Efficacy of the theory of planned behaviour: a meta-analytic review . Br J Soc Psychol . 2001 ; 40 ( Pt 4 ): 471 – 499 . doi:10.1348/014466601164939 Google Scholar CrossRef Search ADS PubMed 58. McClelland JL . Incorporating rapid neocortical learning of new schema-consistent information into complementary learning systems theory . J Exp Psychol Gen . 2013 ; 142 ( 4 ): 1190 – 1210 . doi:10.1037/a0033812 Google Scholar CrossRef Search ADS PubMed 59. Gardner B . A review and analysis of the use of ‘habit’ in understanding, predicting and influencing health-related behaviour . Health Psychol Rev . 2015 ; 9 : 277 – 295 . doi:10.1080/17437199.2013.876238 Google Scholar CrossRef Search ADS PubMed 60. Bowers JS . Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand . Cogn Psychol . 2002 ; 45 ( 3 ): 413 – 445 . doi:10.1016/S0010-0285(02)00506-6 Google Scholar CrossRef Search ADS PubMed 61. Bozionelos G , Bennett P . The theory of planned behaviour as predictor of exercise: the moderating influence of beliefs and personality variables . J Health Psychol . 1999 ; 4 ( 4 ): 517 – 529 . doi:10.1177/135910539900400406 Google Scholar CrossRef Search ADS PubMed 62. Orbell S , Verplanken B . The automatic component of habit in health behavior: habit as cue-contingent automaticity . Health Psychol . 2010 ; 29 ( 4 ): 374 – 383 . doi:10.1037/a0019596 Google Scholar CrossRef Search ADS PubMed 63. Ouellette JA , Wood W . Habit and intention in everyday life: the multiple processes by which past behavior predicts future behavior . Psychol Bull . 1998 ; 124 : 54 – 74 . doi:10.1037/0033-2909.124.1.54 Google Scholar CrossRef Search ADS 64. Bandura A . Self-efficacy mechanism in human agency . Am Psychol . 1982 ; 37 : 122 – 147 . doi:10.1037/0003-066X.37.2.122 Google Scholar CrossRef Search ADS 65. Carpenter CJ . A meta-analysis of the effectiveness of health belief model variables in predicting behavior . Health Commun . 2010 ; 25 ( 8 ): 661 – 669 . doi:10.1080/10410236.2010.521906 Google Scholar CrossRef Search ADS PubMed 66. Cooke R , Sniehotta F , Schüz B . Predicting binge-drinking behaviour using an extended TPB: examining the impact of anticipated regret and descriptive norms . Alcohol Alcohol . 2007 ; 42 ( 2 ): 84 – 91 . doi:10.1093/alcalc/agl115 Google Scholar CrossRef Search ADS PubMed 67. Koch EJ . How does anticipated regret influence health and safety decisions? A literature review . Basic Appl Soc Psych . 2014 ; 36 : 397 – 412 . doi:10.1080/01973533.2014.935379 Google Scholar CrossRef Search ADS 68. Tausig M , Fenwick R . Unbinding time: alternate work schedules and work-life balance . J Fam Econ Issues . 2001 ; 22 : 101 – 119 . doi:10.1023/A:1016626028720 Google Scholar CrossRef Search ADS 69. Lunau T , Bambra C , Eikemo TA , van der Wel KA , Dragano N . A balancing act? Work-life balance, health and well-being in European welfare states . Eur J Public Health . 2014 ; 24 ( 3 ): 422 – 427 . doi:10.1093/eurpub/cku010 Google Scholar CrossRef Search ADS PubMed 70. Leventhal H , Brissette I , Leventhal EA . The common sense models of self-regulation of health & illness . In: Cameron LD , Leventhal H , eds. The Self-Regulation of Health and Illness Behaviour . London : Routledge ; 2003 : 42 – 61 . 71. Gardner B , Lally P , Wardle J . Making health habitual: the psychology of ‘habit-formation’ and general practice . Br J Gen Pract . 2012 ; 62 ( 605 ): 664 – 666 . doi:10.3399/bjgp12X659466 Google Scholar CrossRef Search ADS PubMed 72. Matson-Koffman DM , Brownstein JN , Neiner JA , Greaney ML . A site-specific literature review of policy and environmental interventions that promote physical activity and nutrition for cardiovascular health: what works ? Am J Health Promot . 2005 ; 19 ( 3 ): 167 – 193 . doi:10.4278/0890-1171-19.3.167 Google Scholar CrossRef Search ADS PubMed 73. Stringhini S , Dugravot A , Shipley M et al. Health behaviours, socioeconomic status, and mortality: further analyses of the British Whitehall II and the French GAZEL prospective cohorts . Plos Med . 2011 ; 8 ( 2 ): e1000419 . doi:10.1371/journal.pmed.1000419 Google Scholar CrossRef Search ADS PubMed 74. de Jong JR , Vlaeyen JW , Onghena P , Goossens ME , Geilen M , Mulder H . Fear of movement/(re)injury in chronic low back pain: education or exposure in vivo as mediator to fear reduction ? Clin J Pain . 2005 ; 21 ( 1 ): 9 – 17; discussion 69 . doi:10.1016/S0304-3959(03)00255-0 Google Scholar CrossRef Search ADS PubMed 75. Steptoe A , Wright C , Kunz-Ebrecht SR , Iliffe S . Dispositional optimism and health behaviour in community-dwelling older people: associations with healthy ageing . Br J Health Psychol . 2006 ; 11 ( Pt 1 ): 71 – 84 . doi:10.1348/135910705X42850 Google Scholar CrossRef Search ADS PubMed 76. Muzaffar H , Chapman-Novakofski K , Castelli DM , Scherer JA . The HOT (Healthy Outcome for Teens) project. Using a web-based medium to influence attitude, subjective norm, perceived behavioral control and intention for obesity and type 2 diabetes prevention . Appetite . 2014 ; 72 : 82 – 89 . doi:10.1016/j.appet.2013.09.024 Google Scholar CrossRef Search ADS PubMed 77. Weigold A , Weigold IK , Russell EJ . Examination of the equivalence of self-report survey-based paper-and-pencil and internet data collection methods . Psychol Methods . 2013 ; 18 ( 1 ): 53 – 70 . doi:10.1037/a0031607 Google Scholar CrossRef Search ADS PubMed 78. Webb TL , Sheeran P . Does changing behavioral intentions engender behavior change? A meta-analysis of the experimental evidence . Psychol Bull . 2006 ; 132 ( 2 ): 249 – 268 . doi:10.1037/0033-2909.132.2.249 Google Scholar CrossRef Search ADS PubMed 79. de Bruin M , Sheeran P , Kok G et al. Self-regulatory processes mediate the intention-behavior relation for adherence and exercise behaviors . Health Psychol . 2012 ; 31 ( 6 ): 695 – 703 . doi:10.1037/a0027425 Google Scholar CrossRef Search ADS PubMed 80. Sniehotta FF , Nagy G , Scholz U , Schwarzer R . The role of action control in implementing intentions during the first weeks of behaviour change . Br J Soc Psychol . 2006 ; 45 ( Pt 1 ): 87 – 106 . doi:10.1348/014466605X62460 Google Scholar CrossRef Search ADS PubMed 81. Prochaska JJ , Nigg CR , Spring B , Velicer WF , Prochaska JO . The benefits and challenges of multiple health behavior change in research and in practice . Prev Med . 2010 ; 50 ( 1–2 ): 26 – 29 . doi:10.1016/j.ypmed.2009.11.009 Google Scholar CrossRef Search ADS PubMed 82. King K , Meader N , Wright K et al. Characteristics of interventions targeting multiple lifestyle risk behaviours in adult populations: a systematic scoping review . Plos One . 2015 ; 10 ( 1 ): e0117015 . doi:10.1371/journal.pone.0117015 Google Scholar CrossRef Search ADS PubMed © Society of Behavioral Medicine 2018. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Annals of Behavioral Medicine Oxford University Press

Connectionism and Behavioral Clusters: Differential Patterns in Predicting Expectations to Engage in Health Behaviors

Loading next page...
 
/lp/ou_press/connectionism-and-behavioral-clusters-differential-patterns-in-FGUa5K7Kty
Copyright
© Society of Behavioral Medicine 2018. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
0883-6612
eISSN
1532-4796
D.O.I.
10.1093/abm/kax063
Publisher site
See Article on Publisher Site

Abstract

Abstract Background The traditional approach to health behavior research uses a single model to explain one behavior at a time. However, health behaviors are interrelated and different factors predict certain behaviors better than others. Purpose To conceptualize groups of health behaviors as memory events that elicit various beliefs. A connectionist approach was used to examine patterns of construct activation related to expectations to engage in health behavior clusters. Methods A sample of lay people (N = 1,709) indicated their expectations to perform behaviors representing four clusters (Risk Avoidance, Nutrition & Exercise, Health Maintenance, and General Well-Being) and rated them on 14 constructs obtained from health behavior literature. Results Expectations to engage in all behavioral clusters were significantly and positively associated with “frequency of performance,” “perceived behavioral control,” and “anticipated regret,” and negatively associated with “effort.” However, each behavioral cluster was also predicted by activation of a unique pattern of predictors. Conclusions A connectionist approach can be useful for understanding how different patterns of constructs relate to specific outcomes. The findings provide a rationale for lay people’s cognitive schema of health behaviors, with each behavioral cluster possessing characteristics associated with distinct predictors of expectations to engage in it. These unique activation patterns point to factors that may be particularly significant for health interventions targeting different clusters of health behaviors. Connectionism, Health behavior change models, Multiple health behaviors, Behavioral expectations, Behavioral clusters, Health behavior taxonomy Introduction Several theories have been developed in order to understand and predict health-related behaviors [1] that affect morbidity and premature mortality [2]. The theories involve a variety of social-cognitive factors that explain individual differences in engagement in health behaviors ranging from exercise and fruit and vegetable consumption to smoking cessation and HIV prevention (e.g., [3, 4]). Traditionally, a specific behavior is examined in relation to a particular theory [5, 6], on the assumption that health behaviors share similar predictive paths. However, specific factors were found to be more associated with certain health behaviors than with others. For example, “perceived behavioral control” was a stronger predictor of physical activity than of illness detection behaviors and safe sex [7]. The significance of the specific combination of behavior and predictor underscores the following two major limitations of the traditional approach. The first limitation is the restricted number of predictors: using constructs derived from only one model to predict health behaviors precludes detection of factors not included in that particular model that might explain the behavior. For example, “anticipated regret” was found to predict intention and behavior beyond what the components of the Theory of Planned Behavior could explain [8, 9]. In addition, specific theories might be especially appropriate for predicting certain behaviors [10]. For example, dieting was predicted better by the Theory of Planned Behavior than by a modified Health Belief Model, while the latter predicted intentions for testicular self-examination better than the former [11, 12]. Thus, the explanatory utility of a model depends not only on its components, but also on the targeted behavior, raising questions about whether any given theory is fully generic in accounting for all health behaviors [13]. The second major limitation of the traditional approach is its focus on a single behavior [14]. There is evidence that lay people perceive similarities among health behaviors [15, 16] based on such factors as the extent to which they require effort or are habitual [15, 17]. Since several behaviors can be successfully targeted in health interventions (e.g., ref. [18]), maximizing health benefits and reducing costs, it might be more efficient to address multiple behaviors simultaneously [19]. Moreover, given the known associations among health behaviors, such as smoking and alcohol consumption [20], focusing on a single behavior can narrow the scope of the modification introduced during the process of behavior change. In sum, the “one behavior - one theory” approach does not address the interrelations between health behaviors, and its adequacy might depend on the congruence between the specific behavior and the particular theoretical model. The current study presents an alternative approach that combines the principles of connectionism [21] with the Health Behavior Taxonomy [15] and uses key social-cognitive factors to elucidate distinct predictive patterns of health behavior clusters. Connectionism According to the connectionist approach, cognitive phenomena are viewed as interactions of neuron-like units (or nodes) in a brain-like system [22], and human information processing is modeled as a network of many simple interconnected processing units that generate an output [23]. Nodes are grouped into layers, and knowledge is encoded as a set of connection weights between them, which constitute a single representation. The connection weights—the strength of the connection between nodes in the network—are stored in memory and modify themselves with experience, as a function of associative learning [24]. The context in which the association between units occurs is also likely to be encoded and affect connection weights. Although various types of networks exist, the current study utilized a simple conceptual network with one input layer and one output layer [21]. Activation of a single unit has no fixed meaning independent of the pattern of activation occurring with it across many units [25]. Within this theoretical framework, we propose that the various constructs included in different health behavior models (mostly attitudes and beliefs) reflect mental events or context-related memories, in which relevant beliefs are activated when cued by an applicable situation [21, 26]. In other words, we assumed that any health behavior under investigation provides the context that evokes an array of perceptions and beliefs relating to it. Thus, we focused on a network of associations between several constructs and a known target (health behavior). Since cognitions related to specific health behaviors can consist of signals from a large number of constructs [27], we propose that each particular behavior is associated with a unique pattern of construct activation that represents a specific network of connections encoded in memory. The connectionist approach has already been applied to predicting intentions to exercise using behavioral, normative, and control beliefs derived from the Theory of Planned Behavior [28]. The approach has also been suggested as a unifying framework that can accommodate expectancy-value and automaticity for clarifying the impact of attitudes on health behaviors [26]. We conceptualized the interactions between health behaviors and their predicting constructs in a similar way, with constructs representing the input layer and behavioral expectations representing the output layer (Fig. 1). In order to consider a broad activation pattern, a comprehensive list of relevant constructs was assembled. Key constructs were obtained from review papers of models related to health behaviors [29–32], supplemented by constructs from empirical studies [5, 33]. The list was reviewed by three Professors of Health Psychology [17]. The constructs were “perceived behavioral control” (or “self-efficacy”), “negative affect,” “positive affect,” “impact on health,” “perceived social support,” “non-health rewards,” “anticipated regret,” and “frequency of performance.” The theories from which the constructs were obtained were the Health Action Process Approach [34], Protection Motivation Theory [35], the Health Belief Model [36], and regular and extended versions of the Theory of Planned Behavior [9, 37]. Additional constructs were taken from empirical findings that differentiated between health behaviors: “effort,” “public-private,” “observable impacts,” “outcome timeframe,” “pleasure deprivation,” and “habit” [17, 38]. Fig. 1 View largeDownload slide Diagram of a two-layer network, with constructs as input units receiving activation from other networks, and clusters of expectations to engage in health behavior as output units. Dashed lines indicate low activation signals. Fig. 1 View largeDownload slide Diagram of a two-layer network, with constructs as input units receiving activation from other networks, and clusters of expectations to engage in health behavior as output units. Dashed lines indicate low activation signals. The Health Behavior Taxonomy Despite the growing recognition that targeting multiple behavioral risk factors in behavior change interventions will improve their impact on population health [39], it remains unclear which health behaviors should be targeted in one intervention. According to the connectionist approach, if a number of representations are similar in important respects, a prototypical representation will ignore their differences while preserving their common characteristics [21]. Once a representation of the general characteristics of a category has been formed, it will be elicited by a particular set of cues that bring about a common pattern of network activation [27]. The Health Behavior Taxonomy [15] provides a conceptual framework for this approach: it defines clusters of health behaviors with similar cognitive representations (Fig. 2). The Health Behavior Taxonomy was developed by utilizing a comprehensive list of 45 health behaviors, drawn from laypeople, health promoters, and a literature review, and judged as most important by a sample of laypeople. By using a card sorting technique, a representative sample generated clusters of health behaviors based on their perceived similarities. The findings revealed a cognitive schema of health behaviors, represented by a reliable taxonomy (r2monotonic = .93). The structure was then replicated using a different large sample (reliability between samples: rspearman = .95) and was also found highly stable across age, gender, education, and income (rspearman = .97 between subgroup matrices [15]). Thus, the taxonomy enables the examination and comparison of connections between a limited number of clusters that represent meaningfully-grouped behaviors and cognitive constructs that may explain and predict them. The Health Behavior Taxonomy is comprised of four main clusters: Risk Avoidance, such as avoiding smoking; Health Maintenance, such as regular teeth brushing; Nutrition & Exercise, such as fruit and vegetable consumption; and General Well-being, such as sedentary behaviors. The 45 behaviors on which the taxonomy is based encompass a wide range of health behaviors that have been used in various studies (e.g., refs. [6, 40]). They include behaviors with well-documented health effects (e.g., fruit and vegetable consumption, [41]), as well as “distal health behaviors,” such as participating in social activities [42]. Fig. 2 View largeDownload slide The Health Behavior Taxonomy (adapted from ref. [15]). Behavior abbreviations were used; full descriptions can be found in Supplementary Material 1. Fig. 2 View largeDownload slide The Health Behavior Taxonomy (adapted from ref. [15]). Behavior abbreviations were used; full descriptions can be found in Supplementary Material 1. Expectations Versus Intentions Many theoretical approaches outline precursors of behavioral intentions, considered to be the closest antecedent of behavior [35, 37]. However, empirical studies report an intention-behavior gap [43] whereby intentions represent motivation but do not necessarily translate into actual behavior. Whether or not behavior is performed at a particular point in time may reflect situational and other uncontrollable factors besides motivation, such as physical or resource limitations. Thus, in the current investigation, the outcome is the expectations to engage in behaviors, which are assumed to encompass intentions and other factors related to their realization [44, 45]. Expectations combine willingness with perceived ability to perform a behavior at a particular time, and were found to be more predictive of behavior than intentions [46]. Hypotheses In line with the above considerations, four hypotheses regarding specific connections between psychological constructs and clusters of health behaviors were formulated based on the following reasoning: 1. Since “perceived behavioral control” had high predictive value in previous theories and studies (e.g., refs. [7, 47, 48]), it was expected to be a consistent element across activation patterns, positively associated with expectations to engage in all behavioral clusters. 2. “Effort,” which is related to ego depletion [49], was also predicted to be a consistent element across activation patterns, negatively associated with expectations to engage in all behavioral clusters [7]. 3. “Anticipated regret” was hypothesized to be specifically significant in the activation pattern of expectations to engage in Risk Avoidance behaviors, due to self-blame and the immediate harsh conceivable consequences of nonadherence [50]. 4. “Social support” was expected to be part of the activation pattern of expectations to engage in behaviors related to General Well-being (e.g., activities to maintain interpersonal relationships), since the cluster includes several “distal health behaviors” that entail interactions with other people [42]. Methods Participants The participants, who took part in a larger research project on perceptions of health behaviors, were recruited by an online panel of a large survey company to represent the general population [17]. The final sample consisted of 1,709 participants who reported having no chronic illness: 1,009 women (59%) and 700 men (41%); average age 37.31 years (SD = 11.00); and average income 2.83 (SD = 1.30) on a scale from 1 (much below average) to 5 (much above average). Fifty participants (3%) had primary school or partial secondary school education, 398 (23%) had high school education, 348 (20%) had some form of higher education, 666 (39%) had a bachelor’s degree, and 246 (14%) had a master’s degree or higher (1 unreported). Measures Perceptions of health behaviors questionnaire [17] Twelve behaviors representing all four clusters of the Health Behavior Taxonomy [15] were randomly selected for each participant from the 45 behaviors that were used to develop the taxonomy [17]. Participants were asked to rate each behavior on seven-point Likert-type scales representing 14 different constructs (e.g., “effort”) presented in random order (Table 1). The behavior selection procedure was intended to avoid participation overload (rating 45 × 14 scales). Table 1 Constructs associated with health behaviours Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) View Large Table 1 Constructs associated with health behaviours Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) Construct Question Scale Effort What level of effort is usually needed to perform the behavior? Little effort (1) – much effort (7) Frequency How frequently does the behavior need to be performed? Very infrequently (1) – very frequently (7) Habit To what degree is the behavior a habit? Not a habit at all (1) – a strong habit (7) Impact on Health What is the level of impact of the behavior on health? No impact (1) – high impact (7) Public-private To what degree is the behavior performed alone or with others? Performed alone (1) – performed with others (7) Observable Impact How noticeable is the behavior’s influence on health? Not noticeable (1) – very easily noticeable (7) Outcome Timeframe Are the results of the behavior immediate or long term? Immediate results (1) – long term (7) Perceived Behavioral Control / Efficacy To what degree do people control the performance of the behavior? No control (1) – full control (7) Pleasure Deprivation To what extent does the performance of the behavior deprive one of pleasure? Does not deprive one of pleasure (1) – deprives one of pleasure (7) Non-health Rewards To what degree do people get rewards from the behavior that are not related to health? (e.g., related to looks or psychosocial factors) To a minimal degree (1) – to a high degree (7) Social Support How much support does a person performing the behavior receive from others close to him/her? Very little support (1) – a lot of support (7) Anticipated Regret (if not engaged) To what extent would people feel regret if they did not perform the behavior? No regret at all (1) – high level of regret (7) Negative Affect To what degree does thinking about performing the behavior raise negative emotions in people? Does not raise negative emotions (1) – raises negative emotions to a high degree (7) Positive Affect To what degree does thinking about performing the behavior raise positive emotions in people? Does not raise positive emotions (1) – raises positive emotions to a high degree (7) View Large Behavioral expectations The question: “How likely are you to perform the following behavior (or keep performing it)?” was presented regarding each of the 12 behaviors of the Perceptions of Health Behaviors Questionnaire. Answers were given on a scale ranging from very unlikely [1] to very likely [7]. Demographic items Participants were asked to indicate their gender, age, income, and education. Procedure The study was approved by the university Institutional Review Board. To reduce the risk of participant fatigue and increase validity, the number of behaviors included in the Perceptions of Health Behaviors Questionnaire was limited to 12 (out of 45). A small pilot study revealed that the roughly 100 rating scales were clear and easy to complete in 10–15 min. In addition, 206 respondents were excluded from the analysis since it took them less than 15 s to judge 12 behaviors, thus their answers were suspected of being invalid [17]. Data were collected online via a designated website (Demographic Items were recorded at registration to the online panel). The study was presented as dealing with behaviors affecting health. After receiving informed consent, the Perceptions ofhealth behaviors and the Behavioral expectations Questionnaires were administered consecutively. Respondents were paid a nominal sum of $2 for their participation. Statistical Analysis For each participant, average ratings of the constructs and behavioral expectations were calculated using the behaviors in each cluster. The items representing each cluster were not necessarily identical for all participants due to the random selection process. Afterwards, scores were averaged across participants, resulting in 15 scores for each cluster representing its mean ratings on the 14 constructs and behavioral expectations. Repeated measures analysis of variance was used to test differences between expectations to engage in the four main clusters of health behaviors: Health Maintenance, Nutrition & Exercise, Risk Avoidance, and General Well-being. Neural networks employ an array of possible functions intended to find the best mathematical solution to connect inputs and outputs, frequently resulting in extremely complex equations. Subsequently, linear approximations are often presented in the results. A widely used algorithm in neural network analysis is Backpropagation, which consists of a training set that includes many examples of inputs and their desired outputs [51]. The weights between inputs and outputs of the network to be trained are initially set to random values and then members of the training set are repeatedly exposed and compared to the network. Consequently, the weights are slightly adjusted in the direction that would bring the output values of the network closer to the values of the desired output. After many repetitions of this process, the network learns to produce a result that corresponds to the desired output for each input, which may then be generalized to inputs and outputs that were not in the training set [51]. The current analyses were conducted in MATLAB using the Neural Network Fitting Tool. The networks were trained by the Levenberg-Marquardt backpropagation algorithm and the sample was divided into three subsets, corresponding to the different phases of the process: Training, Validation, and Testing. The Training subset computes a gradient and updates the network weights and biases, so that the network is adjusted according to its error. The Validation subset measures the network’s generalization and stops the training when the generalization stops to improve. The error of the validation set is monitored during the training process, and the network weights and biases are saved at the point of minimum error, before it begins to rise due to data over-fitting. The Testing subset provides a completely independent measure of network performance and generalization. Once the network development has been complete, it is possible to use the entire sample to evaluate the performance of the modeled network [52]. We used 70% of the sample for training, 15% for validation, and 15% for testing. In addition, the software uses an algorithm of multilayer networks that includes a layer of hidden neurons, which are activated by nodes of preceding layers and send signals to succeeding layers [26]. No hidden layers were assumed in the current study, thus all inputs were expected to affect a single node in the hidden layer. However, we also examined a solution driven by the Theory of Planned Behavior that defined three nodes in the hidden layer [28] as well as the possibility of separate nodes in the hidden layer for each individual construct, resulting in a total of 14 nodes (see Supplementary Material 2). Since the goal of the current study was to demonstrate the feasibility and utility of a connectionist approach to researchers in the field of health behavior [53], the more common Regression Model is primarily presented instead of a Neural Network implementation (e.g., Parallel Distributed Processing, [54]). Thus, although the function of the units in the network is often assumed to be sigmoidal [21], the strengths of the links between nodes were represented by the regression equation coefficients, which provide a lower bound approximation to the actual activation strength [55]. Moreover, when a full Neural Network analysis was performed, it was not found superior to Regression Models in terms of explained variance of expectations to engage in behavioral clusters (see Supplementary Material 2). In addition, multicollinearity could have posed a computational and logical problem in our study, due to the large number of predicting variables and their possible overlap in explained variance. Thus, Statistical (stepwise) Multiple Regression, a “model building” procedure [56], was performed separately for each cluster, with the 14 constructs of the Perceptions of Health Behaviors Questionnaire as predictors. This method was used to enable empirical selection of a set of variables that can best explain engagement expectations in each cluster. Only predictors contributing at least 1% to the total explained variance of the model, that is, part of the activation pattern, are presented in the results. Results The correlations between demographic variables and expectations to engage in behavioral clusters were negligible (see Supplementary Material 3). Expectations to engage in the various health behavior clusters differed significantly (F[3,5124] = 217.79, p < .0001, partial η2 = .11). According to Bonferroni post-hoc tests (p < .0001), expectations to engage in the Risk Avoidance cluster were significantly higher (M = 5.97, SD = 1.05) than the Health Maintenance cluster (M = 5.80, SD = 0.99), which in turn was significantly higher than the General Well-being cluster (M = 5.51, SD = 1.10), which was higher than Nutrition & Exercise (M = 5.33, SD = 1.34). Pearson correlations between constructs and behavioral expectations for the main cluster of behaviors can be found in Supplementary Material 4. The patterns of activation related to each behavioral cluster were manifested in the regression equation coefficients, which indicate the strength of the association between the input and output nodes (Table 2). As predicted, expectations to engage in all clusters of health behaviors were positively associated with “perceived behavioral control” and negatively with “effort” (hypotheses 1 and 2, respectively). In addition, all clusters were predicted primarily by “frequency of performance” as well as by “anticipated regret” (Table 2). Table 2 Regression models representing patterns of activation for expectations to engage in health behavior clusters using 14 constructsa Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 R2 Accumulative explained variance. β Standardized coefficients for predictors contributing at least 1% to the total explained variance. In all models, R2 changes and coefficients are significant at p < .001. aA table enabling easier comparison of coefficients of specific predictors across clusters appears in Supplementary material 5. View Large Table 2 Regression models representing patterns of activation for expectations to engage in health behavior clusters using 14 constructsa Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 Risk Avoidance Nutrition & Exercise Health Maintenance General Well-being Predictors β R2 Predictors β R2 Predictors β R2 Predictors β R2 Frequency 0.23 .21 Frequency 0.24 .15 Frequency 0.24 .21 Frequency 0.23 .19 Perceived control 0.16 .29 Effort −0.29 .25 Impact on health 0.18 .27 Perceived control 0.22 .29 Anticipated regret 0.15 .33 Positive affect 0.18 .30 Effort −0.20 .32 Effort −0.19 .31 Prevents enjoyment −0.15 .37 Anticipated regret 0.17 .32 Perceived control 0.15 .34 Impact on health 0.14 .34 Observable impact 0.13 .39 Perceived control 0.14 .34 Anticipated regret 0.13 .36 Social support 0.13 .36 Effort −0.13 .40 Positive affect 0.11 .37 Anticipated regret 0.11 .37 Impact on health 0.14 .41 R2 Accumulative explained variance. β Standardized coefficients for predictors contributing at least 1% to the total explained variance. In all models, R2 changes and coefficients are significant at p < .001. aA table enabling easier comparison of coefficients of specific predictors across clusters appears in Supplementary material 5. View Large Furthermore, each cluster displayed a distinct activation pattern, which included not only the aforementioned constructs, but also an array of specific predictors with varying connection strengths. The explanatory power of each construct represents its specific contribution to the activation pattern. Expectations to engage in the Risk Avoidance cluster were explained primarily by “frequency of performance” (21% of the variance), followed by “perceived behavioral control” (an additional 8%), and, as predicted by hypothesis 3, by “anticipated regret” (an additional 4%). Expectations to engage in Nutrition & Exercise behaviors were predicted mainly by “frequency of performance” (15% explained variance), followed by “effort” (an additional 10%) and “positive affect” (an additional 5%). Expectations to engage in Health Maintenance behaviors were explained by “frequency of performance” (21% of the variance), followed by “impact on health” (an additional 6%) and “effort” (an additional 5%). Finally, expectations to engage in the General Well-being cluster were also predicted primarily by “frequency of performance” (19%), followed by “perceived behavioral control” (an additional 10%), and “effort” (an additional 3%). Hypothesis 4 was also confirmed: “social support” significantly explained variance related to expectations to engage in General Well-being (Table 2). In general, expectations to engage in the Risk Avoidance cluster were best explained by the constructs measured in the current study (41% explained variance using multiple regression, 45% using neural networks), followed by Health Maintenance and General Well-being (both with 37% explained variance using multiple regression, 38% using neural networks), and Nutrition & Exercise (34% explained variance using multiple regression, 37% using neural networks). These findings are in line with previous findings [57], and an example of the results of the different stages of the neural network process and the linear approximation for the Risk Avoidance behavioral cluster is presented in Table 3. Table 3 Explained variance for each step of the neural network analysis, for the overall model, and linear approximations regarding the Risk Avoidance behavioral cluster Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 View Large Table 3 Explained variance for each step of the neural network analysis, for the overall model, and linear approximations regarding the Risk Avoidance behavioral cluster Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 Hidden neurons Training Validation Test Overall Equation for overall (≈) 1 0.47 0.31 0.48 0.45 0.40*X+3.6 3 0.48 0.34 0.43 0.45 0.44*X+3.4 14 0.38 0.23 0.33 0.35 0.47*X+3.2 View Large Discussion This study serves as proof of concept for the usefulness of a connectionist approach to the investigation of health behaviors, in line with previous research [26, 28]. The goal of the connectionist approach is to empirically find an optimal set of weights for a network. In order to accomplish that, sigmoidal functions are used to create neuron-like connections that simulate the human brain, consequently increasing accuracy of predictions. The current study modeled the unique mental processes involved in expectations to engage in health behaviors. The findings add to the current knowledge by uncovering the activation patterns of cognitive constructs related to clusters of health behaviors, reflecting relevant beliefs associated with them [21, 26]. In other words, the findings delineate the differential effects of predictors of behavioral expectations, which signify the beliefs and perceptions attached to each particular group of health behaviors and disclose the unique activation patterns characterizing each behavioral cluster. Moreover, in line with the connectionist assumption regarding the network’s ability to generalize [27], the current findings provide support for the utility of addressing clusters of related behaviors [15], since the health behaviors in each cluster possess schema-consistent information [58]. In addition, the findings indicate that expectations to engage in health behavior clusters can be predicted by unique patterns of theoretical constructs. This underscores the importance of considering similarities and differences among health behaviors beyond previous studies that had shown the existence of cognitive health behavior clusters and described the nature of differences among them [15–17]. Expectations to engage in Risk Avoidance behaviors were best explained by the constructs measured in the current study. Similar findings were reported for these behaviors compared with physical activity, dietary behaviors, and abstinence [7]. We speculate that behaviors such as smoking might be better predicted due to a combination of stronger learned stimulus-response association and habit-generated impulses [59]. It is important to note that a connectionist approach maintains that the activation of a single unit, which may contribute to different representations [60], should be interpreted in the context of a pattern that symbolizes a semantically meaningful mental state [25]. Subsequently, for example, whether or not “positive affect” is a significant predictor would be best understood when the entire construct pattern is considered. However, for reasons of clarity and simplicity, the following discussion will deconstruct the activation patterns and explore the roles of individual constructs that were found relevant for all behavioral expectations, as well as of those with unique significance for specific clusters. Common Explanatory Factors Expectations to engage in all clusters were predicted by higher levels of “frequency of performance”, “perceived behavioral control” and “anticipated regret,” and lower levels of “effort.” Frequency perceptions (how frequently behaviors should be performed) were the primary predictor of expectations to engage in all clusters of health behaviors, consistent with previous findings that past behavior had the highest predictive value for exercise [61], and that “frequency of performance” was a key characteristic of health behaviors in general [5]. It may represent a cue-response relationship component [62], as well as habitual automaticity and strength [63], signifying the influence of frequent performance on future behavior. The fact that “perceived behavioral control” was also a significant predictor of expectations to engage in all behavioral clusters underscores its importance and is in line with the majority of theories of health behaviors [1]. However, “effort” and “anticipated regret,” which were also found to predict expectations to engage in all health behavior clusters, are absent from most health behavior models. The degree of effort required to perform a behavior can be partly analogous to “perceived barriers” in the Health Belief Model [36] and to “self-efficacy” [64]. Considering its consistent role as a predictor of expectations and behavior [65], “effort” may add significant explanatory power to health behavior models. “Anticipated regret” has already been shown to have a unique contribution over and above the components of the Theory of Planned Behavior [8, 66], and the current findings support the utility of including this construct in health behavior theories. Together, all four constructs play a key role in the activation patterns of expectations to engage in health behaviors. However, their relative strengths vary according to relevance to the predicted behavioral cluster. For example, “anticipated regret” was a stronger predictor of the Risk Avoidance cluster (e.g., wearing a seatbelt) than the other clusters. This may result from the higher emotionality associated with the immediate grave outcomes of failing to avoid risks, reflected also in the high ratings of “anticipated regret” for Risk Avoidance behaviors compared to the other clusters. Thus, the strong motivation to engage in these health behaviors [67] can be due to the wish to avoid both the physical and emotional consequences of nonadherence [50]. In addition, “perceived behavioral control” was a stronger predictor of expectations to engage in General Well-being behaviors (e.g., spending quality time with one’s partner) compared to the other clusters. This may illustrate the perceived challenge many people face in balancing activities related to work, family, and personal life [68] that affects health and well-being [69]. Finally, the Nutrition & Exercise cluster was more strongly predicted by “effort” than by “perceived behavioral control,” which may reflect the perceived difficulty entailed in overcoming external temptations and controlling one’s diet [43]. Unique Explanatory Factors In addition to the different predictive weights of the common explanatory factors, distinct units in the activation pattern of expectations to engage in specific behavioral clusters were uncovered, corresponding to their different features and purposes. For example, “positive affect” was an important predictor of expectations to engage in Nutrition & Exercise behaviors, but not in Risk Avoidance or General Well-being. This is consistent with previous findings of stronger associations between attitudes and intentions to engage in dietary behaviors compared to risky health behaviors [7]. Further investigation is needed to determine the processes by which food-related emotional experiences develop into this association. In addition, behaviors related to Health Maintenance (e.g., attending medical check-ups) were more strongly predicted by cognitive appraisals regarding their “impact on health” compared to the other behavioral clusters. This can represent a medicalized danger-control approach [70], in which medical attention is sought as a reaction to health threats. It is also noteworthy that perceptions of “impact on health” were associated with all clusters except Nutrition & Exercise. This cluster was found highest on “impact on health” while displaying the lowest within-cluster variance compared to other clusters [17]—a finding that may reflect a consensus among lay people regarding the major influence of nutritional habits on health, which might have attenuated its predictive power here. Implications The connectionist approach may be helpful in discovering meaningful patterns that can facilitate performance expectations and engagement in health behaviors. The results reveal common and unique factors that could be relevant to interventions targeting different clusters of health behaviors. Of special interest is the finding that perceived “frequency of performance” was the primary predictor of expectations to engage in all behavioral clusters. This underscores the key role of habit formation in lay people’s cognitions and supports defining it as a target in health interventions [71]. “Effort” appears particularly important with regard to Nutrition & Exercise behaviors. Consequently, interventions aimed at modifying dietary habits would benefit from reducing effort levels, such as providing easier access to healthy foods [72]. Finally, since perceptions regarding the “impact on health” do not predict nutrition behaviors, the introduction of such information in interventions may be superfluous. In sum, the findings of the present study may help conceptualize how to design health research and interventions by providing a framework for addressing patterns of associations between cognitive constructs and clusters of interrelated health behaviors. Limitations The current study serves as proof of concept of the utility of connectionism in the field of health behavior theorizing and research. It is only a first step in mapping the activation patterns related to expectations to engage in clusters of health behaviors, which should be further examined with regard to actual engagement in behaviors, in different contexts, cultures, and populations [26, 73]. For example, the fear of injury among patients with chronic low back pain [74] may amplify the role of “anticipated regret” in the activation pattern related to exercise, due to strong associations between certain movements and physical pain. It is important to note that the results obtained by neural network analysis and regression analysis were comparable, although the former method was found to be superior in a previous study [28]. A possible explanation for this discrepancy might be related to measurement: the predictors in the current study were measured by single items, whereas the advantages embodied in the complexity of the neural network method may require assessing theoretical constructs using multiple items. Consequently, future studies employing neural networks should consider examining constructs related to health behaviors using multiple indicators. In addition, utilizing more complex network analyses that incorporate hidden layers corresponding to the constructs measured by multiple indicators may increase prediction accuracy for both unique and common variability [28], since they would enable the generation of a correct output pattern in case more than a simple reconstruction of the input pattern is required [25]. The use of an empirical approach such as connectionism to study health behaviors is limited since the predictors that emerge as significant depend on the variables included in the study and their multicollinearity. However, the selection of variables from health behavior theories increases the confidence in the predictors. Further, the risk of spuriousness is reduced since multiple regression has the advantage of controlling for the variance explained by the other variables when calculating each predictor’s coefficient. In general, statistical regression is considered a model-building procedure rather than a model-testing technique [56]. Future research on health behaviors should continue to integrate empirical and theoretical efforts by applying the connectionist approach to theoretically-driven constructs and modifying theories according to empirically-driven findings. Although numerous theoretical constructs were included in the current study, further attempts should be made to examine the contribution of additional predictors related to health behaviors such as dispositional optimism [75]. These efforts might be guided by theoretical considerations as well as particular behavioral circumstances or populations under investigation. As is common in the health behavior literature, the current analysis was interpreted to be indicative of a causal relationship (constructs predict expectations) despite the use of correlational data (e.g., ref. [76]). However, to determine causality, the effects of interventions on the modification of specific behaviors must be examined in randomized controlled trials. In addition, although results of online studies were found to be equivalent to those obtained using other methods of data collection [77], it is important to replicate the current finding by using additional samples and methods. The use of behavioral expectations as an outcome in the current study was intended to partly overcome the intentions-behavior gap [43]. However, since expectations can fluctuate according to various moderators [78] and may be partly behavior dependent [43], comparing them with actual engagement in behavioral clusters is the next required step. This would allow examining whether or not the actual behaviors display similar patterns to those uncovered in this study. Future studies should also explore whether self-regulatory mechanisms that predict behavior [79], such as action control [80], moderate activation patterns associated with specific health behavior clusters. Conclusions Multiple health behavior research is at an early stage [81] and is considered one of the most important challenges for behavioral medicine [39]. The potential benefits of addressing multiple behaviors simultaneously in health interventions have led to a sharp increase in studies reporting them [82]. While the optimal way to design such interventions remains unknown, using lay people’s cognitive schema of health behaviors as a lead for intervention development seems promising [15]. The connectionist approach applied in the current study suggests re-examining the relationships between theoretical components of behavioral change and desired outcomes. It offers a new perspective on the diverse effects of predictive constructs as well as on possible moderators and barriers to engagement in clusters of health behaviors. It also emphasizes patterns of activation rather than individual constructs and highlights common and unique explanatory factors. The differential importance of specific constructs for distinct behavioral clusters draws attention to the essential features of those clusters and to the need to consider them when planning health prevention and promotion programs. Supplementary Material Supplementary material is available at Annals of Behavioral Medicine online. Acknowledgments This research was funded by the Israel Science Foundation, grant number 257/11. Compliance with Ethical Standards Statements The study was approved by the Institutional Review Board of Tel Aviv University. Informed consent was given prior to participation in the study. G. Nudelman and S. Shiloh developed the study concept and design. G. Nudelman developed the study website and performed the data analysis. G. Nudelman interpreted the results and wrote the first draft of the manuscript, and S. Shiloh provided critical revisions. Both authors approved the final version of the manuscript for submission. Conflict of Interest We declare adherence to ethical standards and that there is no actual or potential conflicts of interest with the Israel Science Foundation, which sponsored the research. References 1. Conner M , Norman P. Predicting Health Behaviour . UK : McGraw-Hill Education ; 2005 . 2. Loef M , Walach H . The combined effects of healthy lifestyle behaviors on all cause mortality: a systematic review and meta-analysis . Prev Med . 2012 ; 55 ( 3 ): 163 – 170 . doi:10.1016/j.ypmed.2012.06.017 Google Scholar CrossRef Search ADS PubMed 3. Godinho CA , Alvarez MJ , Lima ML , Schwarzer R . Health messages to promote fruit and vegetable consumption at different stages: a match-mismatch design . Psychol Health . 2015 ; 30 ( 12 ): 1410 – 1432 . doi:10.1080/08870446.2015.1054827 Google Scholar CrossRef Search ADS PubMed 4. Fisher WA , Fisher JD , Shuper PA . Social psychology and the fight against AIDS: an information-motivation-behavioral skills model for the prediction and promotion of health behavior change . Adv Exp Soc Psychol . 2014 ; 50 : 105 – 193 . doi:10.1016/B978-0-12-800284-1.00003-5 Google Scholar CrossRef Search ADS 5. McEachan RR , Lawton RJ , Conner M . Classifying health-related behaviours: exploring similarities and differences amongst behaviours . Br J Health Psychol . 2010 ; 15 ( Pt 2 ): 347 – 366 . doi:10.1348/135910709X466487 Google Scholar CrossRef Search ADS PubMed 6. Sniehotta FF , Presseau J , Allan J , Araújo-Soares V . “You can’t always get what you want”: a novel research paradigm to explore the relationship between multiple intentions and behaviours . Appl Psychol Health Well Being . 2016 ; 8 ( 2 ): 258 – 275 . doi:10.1111/aphw.12071 Google Scholar CrossRef Search ADS PubMed 7. McEachan RRC , Conner M , Taylor NJ , Lawton RJ . Prospective prediction of health-related behaviours with the theory of planned behaviour: a meta-analysis . Health Psychol Rev . 2011 ; 5 : 97 – 144 . doi:10.1080/17437199.2010.521684 Google Scholar CrossRef Search ADS 8. Sandberg T , Conner M . Anticipated regret as an additional predictor in the theory of planned behaviour: a meta-analysis . Br J Soc Psychol . 2008 ; 47 ( Pt 4 ): 589 – 606 . doi:10.1348/014466607X258704 Google Scholar CrossRef Search ADS PubMed 9. Sandberg T , Hutter R , Richetin J , Conner M . Testing the role of action and inaction anticipated regret on intentions and behaviour . Br J Soc Psychol . 2016 ; 55 ( 3 ): 407 – 425 . doi:10.1111/bjso.12141 Google Scholar CrossRef Search ADS PubMed 10. Garcia K , Mann T . From ‘I Wish’ to ‘I Will’: social-cognitive predictors of behavioral intentions . J Health Psychol . 2003 ; 8 ( 3 ): 347 – 360 . 10.1177/13591053030083005 Google Scholar PubMed 11. McClenahan C , Shevlin M , Adamson G , Bennett C , O’Neill B . Testicular self-examination: a test of the health belief model and the theory of planned behaviour . Health Educ Res . 2007 ; 22 ( 2 ): 272 – 284 . doi:10.1093/her/cyl076 Google Scholar CrossRef Search ADS PubMed 12. Nejad LM , Wertheim EH , Greenwood K . Comparison of the health belief model and the theory of planned behavior in the prediction of dieting and fasting behavior . E J Appl Psychol . 2005 ; 1 : 63 – 74 . doi:10.7790/ejap.v1i1.10 Google Scholar CrossRef Search ADS 13. Noar SM , Zimmerman RS . Health Behavior Theory and cumulative knowledge regarding health behaviors: are we moving in the right direction ? Health Educ Res . 2005 ; 20 ( 3 ): 275 – 290 . doi:10.1093/her/cyg113 Google Scholar CrossRef Search ADS PubMed 14. Fleig L , Ngo J , Roman B et al. Beyond single behaviour theory: adding cross-behaviour cognitions to the health action process approach . Br J Health Psychol . 2015 ; 20 ( 4 ): 824 – 841 . doi:10.1111/bjhp.12144 Google Scholar CrossRef Search ADS PubMed 15. Nudelman G , Shiloh S . Mapping health behaviors: constructing and validating a common-sense taxonomy of health behaviors . Soc Sci Med . 2015 ; 146 : 1 – 10 . doi:10.1016/j.socscimed.2015.10.004 Google Scholar CrossRef Search ADS PubMed 16. Vickers RR Jr , Conway TL , Hervig LK . Demonstration of replicable dimensions of health behaviors . Prev Med . 1990 ; 19 ( 4 ): 377 – 401 . doi:10.1016/0091-7435(90)90037-K Google Scholar CrossRef Search ADS PubMed 17. Nudelman G , Shiloh S . Understanding behavioural clusters: Identifying differences between clusters of health behaviours on key constructs . Psychol Health . 2016 ; 31 ( 12 ): 1375 – 1390 . doi:10.1080/08870446.2016.1208822 Google Scholar CrossRef Search ADS PubMed 18. Prochaska JO , Butterworth S , Redding CA et al. Initial efficacy of MI, TTM tailoring and HRI’s with multiple behaviors for employee health promotion . Prev Med . 2008 ; 46 ( 3 ): 226 – 231 . doi:10.1016/j.ypmed.2007.11.007 Google Scholar CrossRef Search ADS PubMed 19. Prochaska JJ , Prochaska JO . A Review of multiple health behavior change interventions for primary prevention . Am J Lifestyle Med . 2011 ; 5 : (3) 208 – 221 . doi: 10.1177/1559827610391883 Google Scholar CrossRef Search ADS 20. Hofstetter H , Dusseldorp E , van Empelen P , Paulussen TW . A primer on the use of cluster analysis or factor analysis to assess co-occurrence of risk behaviors . Prev Med . 2014 ; 67 : 141 – 146 . doi:10.1016/j.ypmed.2014.07.007 Google Scholar CrossRef Search ADS PubMed 21. Smith ER . What do connectionism and social psychology offer each other ? J Pers Soc Psychol . 1996 ; 70 ( 5 ): 893 – 912 . doi:10.1037/0022-3514.70.5.893 Google Scholar CrossRef Search ADS PubMed 22. Nelson R . Expanding the role of connectionism in SLA theory . Lang Learn . 2013 ; 63 : 1 – 33 . doi:10.1111/j.1467-9922.2012.00699.x Google Scholar CrossRef Search ADS 23. Joanisse MF , McClelland JL . Connectionist perspectives on language learning, representation and processing . Wiley Interdiscip Rev Cogn Sci . 2015 ; 6 ( 3 ): 235 – 247 . doi:10.1002/wcs.1340 Google Scholar CrossRef Search ADS PubMed 24. Olson MA , Fazio RH . Implicit attitude formation through classical conditioning . Psychol Sci . 2001 ; 12 ( 5 ): 413 – 417 . doi:10.1111/1467-9280.00376 Google Scholar CrossRef Search ADS PubMed 25. McClelland JL , Rumelhart DE. The PDP Research Group: Parallel Distributed Processing: Explorations in the Microstructure of Cognition . Cambridge, MA : MIT Press ; 1986 . 26. Lowe R , Norman P . Attitudinal approaches to health behavior: integrating expectancy-value and automaticity accounts . Soc Personal Psychol Compass . 2013 ; 7 : 572 – 584 . doi:10.1111/spc3.12046 Google Scholar CrossRef Search ADS 27. Smith ER . Distributed connectionist models in social psychology . Soc Personal Psychol Compass . 2009 ; 3 : 64 – 76 . doi:10.1111/j.1751-9004.2008.00160.x Google Scholar CrossRef Search ADS 28. Lowe R , Bennett P , Walker I , Milne S , Bozionelos G . A connectionist implementation of the theory of planned behavior: association of beliefs with exercise intention . Health Psychol . 2003 ; 22 ( 5 ): 464 – 470 . doi:10.1037/0278-6133.22.5.464 Google Scholar CrossRef Search ADS PubMed 29. Armitage CJ , Conner M . Social cognition models and health behaviour: a structured review . Psychol Health . 2000 ; 15 : 173 – 189 . doi:10.1080/08870440008400299 Google Scholar CrossRef Search ADS 30. Noar SM , Chabot M , Zimmerman RS . Applying health behavior theory to multiple behavior change: considerations and approaches . Prev Med . 2008 ; 46 ( 3 ): 275 – 280 . doi:10.1016/j.ypmed.2007.08.001 Google Scholar CrossRef Search ADS PubMed 31. Glanz K , Bishop DB . The role of behavioral science theory in development and implementation of public health interventions . Annu Rev Public Health . 2010 ; 31 : 399 – 418 . doi:10.1146/annurev.publhealth.012809.103604 Google Scholar CrossRef Search ADS PubMed 32. Sutton S . Health behavior: psychosocial theories . In: Smelser NJ , Baltes PB , eds. International Encyclopedia of the Social and Behavioral Sciences . Oxford, UK : Elsevier ; 2001 : 6499 – 6506 . 33. Kiviniemi MT , Voss-Humke AM , Seifert AL . How do I feel about the behavior? The interplay of affective associations with behaviors and cognitive beliefs as influences on physical activity behavior . Health Psychol . 2007 ; 26 ( 2 ): 152 – 158 . doi:10.1037/0278-6133.26.2.152 Google Scholar CrossRef Search ADS PubMed 34. Schwarzer R . Modeling health behavior change: how to predict and modify the adoption and maintenance of health behaviors . Appl Psychol . 2008 ; 57 : 1 – 29 . doi:10.1111/j.1464-0597.2007.00325.x 35. Rogers RW , ed. Cognitive and Physiological Processes in Fear Appeals and Attitude Change: A Revised Theory of Protection Motivation . London, England : Guildford ; 1983 . 36. Janz NK , Becker MH . The Health Belief Model: a decade later . Health Educ Q . 1984 ; 11 ( 1 ): 1 – 47 . doi:10.1177/109019818401100101 Google Scholar CrossRef Search ADS PubMed 37. Ajzen I . The theory of planned behavior . Organ Behav Hum Decis Process . 1991 ; 50 : 179 – 211 . doi:10.1016/0749-5978(91)90020-T Google Scholar CrossRef Search ADS 38. Shiloh S , Peretz G , Iss R , Kiedan R . Recovery attributions: explicit endorsement of biomedical factors and implicit dominance of psycho-social factors . J Behav Med . 2007 ; 30 ( 3 ): 243 – 251 . doi:10.1007/s10865-007-9100-0 Google Scholar CrossRef Search ADS PubMed 39. Evers KE , Quintiliani LM . Advances in multiple health behavior change research . Transl Behav Med . 2013 ; 3 ( 1 ): 59 – 61 . doi:10.1007/s13142-013-0198-z Google Scholar CrossRef Search ADS PubMed 40. Bergkvist L , Rossiter JR . The predictive validity of multiple-item versus single-item measures of the same constructs . J Mark Res . 2007 ; 44 : 175 – 184 . doi:10.1509/jmkr.44.2.175 Google Scholar CrossRef Search ADS 41. Wang X , Ouyang Y , Liu J et al. Fruit and vegetable consumption and mortality from all causes, cardiovascular disease, and cancer: systematic review and dose-response meta-analysis of prospective cohort studies . BMJ . 2014 ; 349 : g4490 . doi:10.1136/bmj.g4490 Google Scholar CrossRef Search ADS PubMed 42. Ziegelmann JP , Knoll N . Future directions in the study of health behavior among older adults . Gerontology . 2015 ; 61 ( 5 ): 469 – 476 . doi:10.1159/000369857 Google Scholar CrossRef Search ADS PubMed 43. Sheeran P . Intention—behavior relations: a conceptual and empirical review . Eur Rev Soc Psychol . 2002 ; 12 : 1 – 36 . doi:10.1080/14792772143000003 Google Scholar CrossRef Search ADS 44. Sheppard BH , Hartwick J , Warshaw PR . The theory of reasoned action: A meta-analysis of past research with recommendations for modifications and future research . J Consum Res . 1988 ; 15 ( 3 ) 325 – 343 . doi:10.1086/209170 Google Scholar CrossRef Search ADS 45. Venkatesh V , Brown SA , Maruping LM , Bala H . Predicting different conceptualizations of system use: the competing roles of behavioral intention, facilitating conditions, and behavioral expectation . MIS Q . 2008 ; 33 : 483 – 502 . doi:10.1016/j.chb.2013.07.018 Google Scholar CrossRef Search ADS 46. Armitage CJ , Norman P , Alganem S , Conner M . Expectations are more predictive of behavior than behavioral intentions: evidence from two prospective studies . Ann Behav Med . 2015 ; 49 ( 2 ): 239 – 246 . doi:10.1007/s12160-014-9653-4 Google Scholar CrossRef Search ADS PubMed 47. Cooke R , French DP . How well do the theory of reasoned action and theory of planned behaviour predict intentions and attendance at screening programmes? A meta-analysis . Psychol Health . 2008 ; 23 ( 7 ): 745 – 765 . doi:10.1080/08870440701544437 Google Scholar CrossRef Search ADS PubMed 48. Topa G , Moriano JA . Theory of planned behavior and smoking: meta-analysis and SEM model . Subst Abuse Rehabil . 2010 ; 1 : 23 – 33 . doi:10.2147/SAR.S15168 Google Scholar CrossRef Search ADS PubMed 49. Baumeister RF , Vohs KD , Tice DM . The strength model of self-control . Curr Dir Psychol Sci . 2007 ; 16 : 351 – 355 . doi:10.1111/j.1467-8721.2007.00534.x Google Scholar CrossRef Search ADS 50. Connolly T , Zeelenberg M . Regret in decision making . Curr Dir Psychol Sci . 2002 ; 11 : 212 – 216 . doi:10.1111/1467-8721.00203 Google Scholar CrossRef Search ADS 51. Garson J . Connectionism . In: Zalta EN , ed. The Stanford Encyclopedia of Philosophy . Stanford University ; 2016 . 52. MathWorks . Divide Data for Optimal Neural Network Training. Retrieved November 7, 2017 from https://www.mathworks.com/help/nnet/ug/divide-data-for-optimal-neural-network-training.html. 53. Yermolayeva Y , Rakison DH . Connectionist modeling of developmental changes in infancy: approaches, challenges, and contributions . Psychol Bull . 2014 ; 140 ( 1 ): 224 – 255 . doi:10.1037/a0032150 Google Scholar CrossRef Search ADS PubMed 54. McClelland JL. Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises . CA : Palo Alto, Stanford University Press ; 2015 . 55. Rawlings JO , Pantula SG , Dickey DA. Applied Regression Analysis: A Research Tool . New York, NY : Springer-Verlag New York Inc .; 2001 . 56. Tabachnick BG , Fidell LS. Using Multivariate Statistics . 6th ed . Boston : Pearson ; 2013 . 57. Armitage CJ , Conner M . Efficacy of the theory of planned behaviour: a meta-analytic review . Br J Soc Psychol . 2001 ; 40 ( Pt 4 ): 471 – 499 . doi:10.1348/014466601164939 Google Scholar CrossRef Search ADS PubMed 58. McClelland JL . Incorporating rapid neocortical learning of new schema-consistent information into complementary learning systems theory . J Exp Psychol Gen . 2013 ; 142 ( 4 ): 1190 – 1210 . doi:10.1037/a0033812 Google Scholar CrossRef Search ADS PubMed 59. Gardner B . A review and analysis of the use of ‘habit’ in understanding, predicting and influencing health-related behaviour . Health Psychol Rev . 2015 ; 9 : 277 – 295 . doi:10.1080/17437199.2013.876238 Google Scholar CrossRef Search ADS PubMed 60. Bowers JS . Challenging the widespread assumption that connectionism and distributed representations go hand-in-hand . Cogn Psychol . 2002 ; 45 ( 3 ): 413 – 445 . doi:10.1016/S0010-0285(02)00506-6 Google Scholar CrossRef Search ADS PubMed 61. Bozionelos G , Bennett P . The theory of planned behaviour as predictor of exercise: the moderating influence of beliefs and personality variables . J Health Psychol . 1999 ; 4 ( 4 ): 517 – 529 . doi:10.1177/135910539900400406 Google Scholar CrossRef Search ADS PubMed 62. Orbell S , Verplanken B . The automatic component of habit in health behavior: habit as cue-contingent automaticity . Health Psychol . 2010 ; 29 ( 4 ): 374 – 383 . doi:10.1037/a0019596 Google Scholar CrossRef Search ADS PubMed 63. Ouellette JA , Wood W . Habit and intention in everyday life: the multiple processes by which past behavior predicts future behavior . Psychol Bull . 1998 ; 124 : 54 – 74 . doi:10.1037/0033-2909.124.1.54 Google Scholar CrossRef Search ADS 64. Bandura A . Self-efficacy mechanism in human agency . Am Psychol . 1982 ; 37 : 122 – 147 . doi:10.1037/0003-066X.37.2.122 Google Scholar CrossRef Search ADS 65. Carpenter CJ . A meta-analysis of the effectiveness of health belief model variables in predicting behavior . Health Commun . 2010 ; 25 ( 8 ): 661 – 669 . doi:10.1080/10410236.2010.521906 Google Scholar CrossRef Search ADS PubMed 66. Cooke R , Sniehotta F , Schüz B . Predicting binge-drinking behaviour using an extended TPB: examining the impact of anticipated regret and descriptive norms . Alcohol Alcohol . 2007 ; 42 ( 2 ): 84 – 91 . doi:10.1093/alcalc/agl115 Google Scholar CrossRef Search ADS PubMed 67. Koch EJ . How does anticipated regret influence health and safety decisions? A literature review . Basic Appl Soc Psych . 2014 ; 36 : 397 – 412 . doi:10.1080/01973533.2014.935379 Google Scholar CrossRef Search ADS 68. Tausig M , Fenwick R . Unbinding time: alternate work schedules and work-life balance . J Fam Econ Issues . 2001 ; 22 : 101 – 119 . doi:10.1023/A:1016626028720 Google Scholar CrossRef Search ADS 69. Lunau T , Bambra C , Eikemo TA , van der Wel KA , Dragano N . A balancing act? Work-life balance, health and well-being in European welfare states . Eur J Public Health . 2014 ; 24 ( 3 ): 422 – 427 . doi:10.1093/eurpub/cku010 Google Scholar CrossRef Search ADS PubMed 70. Leventhal H , Brissette I , Leventhal EA . The common sense models of self-regulation of health & illness . In: Cameron LD , Leventhal H , eds. The Self-Regulation of Health and Illness Behaviour . London : Routledge ; 2003 : 42 – 61 . 71. Gardner B , Lally P , Wardle J . Making health habitual: the psychology of ‘habit-formation’ and general practice . Br J Gen Pract . 2012 ; 62 ( 605 ): 664 – 666 . doi:10.3399/bjgp12X659466 Google Scholar CrossRef Search ADS PubMed 72. Matson-Koffman DM , Brownstein JN , Neiner JA , Greaney ML . A site-specific literature review of policy and environmental interventions that promote physical activity and nutrition for cardiovascular health: what works ? Am J Health Promot . 2005 ; 19 ( 3 ): 167 – 193 . doi:10.4278/0890-1171-19.3.167 Google Scholar CrossRef Search ADS PubMed 73. Stringhini S , Dugravot A , Shipley M et al. Health behaviours, socioeconomic status, and mortality: further analyses of the British Whitehall II and the French GAZEL prospective cohorts . Plos Med . 2011 ; 8 ( 2 ): e1000419 . doi:10.1371/journal.pmed.1000419 Google Scholar CrossRef Search ADS PubMed 74. de Jong JR , Vlaeyen JW , Onghena P , Goossens ME , Geilen M , Mulder H . Fear of movement/(re)injury in chronic low back pain: education or exposure in vivo as mediator to fear reduction ? Clin J Pain . 2005 ; 21 ( 1 ): 9 – 17; discussion 69 . doi:10.1016/S0304-3959(03)00255-0 Google Scholar CrossRef Search ADS PubMed 75. Steptoe A , Wright C , Kunz-Ebrecht SR , Iliffe S . Dispositional optimism and health behaviour in community-dwelling older people: associations with healthy ageing . Br J Health Psychol . 2006 ; 11 ( Pt 1 ): 71 – 84 . doi:10.1348/135910705X42850 Google Scholar CrossRef Search ADS PubMed 76. Muzaffar H , Chapman-Novakofski K , Castelli DM , Scherer JA . The HOT (Healthy Outcome for Teens) project. Using a web-based medium to influence attitude, subjective norm, perceived behavioral control and intention for obesity and type 2 diabetes prevention . Appetite . 2014 ; 72 : 82 – 89 . doi:10.1016/j.appet.2013.09.024 Google Scholar CrossRef Search ADS PubMed 77. Weigold A , Weigold IK , Russell EJ . Examination of the equivalence of self-report survey-based paper-and-pencil and internet data collection methods . Psychol Methods . 2013 ; 18 ( 1 ): 53 – 70 . doi:10.1037/a0031607 Google Scholar CrossRef Search ADS PubMed 78. Webb TL , Sheeran P . Does changing behavioral intentions engender behavior change? A meta-analysis of the experimental evidence . Psychol Bull . 2006 ; 132 ( 2 ): 249 – 268 . doi:10.1037/0033-2909.132.2.249 Google Scholar CrossRef Search ADS PubMed 79. de Bruin M , Sheeran P , Kok G et al. Self-regulatory processes mediate the intention-behavior relation for adherence and exercise behaviors . Health Psychol . 2012 ; 31 ( 6 ): 695 – 703 . doi:10.1037/a0027425 Google Scholar CrossRef Search ADS PubMed 80. Sniehotta FF , Nagy G , Scholz U , Schwarzer R . The role of action control in implementing intentions during the first weeks of behaviour change . Br J Soc Psychol . 2006 ; 45 ( Pt 1 ): 87 – 106 . doi:10.1348/014466605X62460 Google Scholar CrossRef Search ADS PubMed 81. Prochaska JJ , Nigg CR , Spring B , Velicer WF , Prochaska JO . The benefits and challenges of multiple health behavior change in research and in practice . Prev Med . 2010 ; 50 ( 1–2 ): 26 – 29 . doi:10.1016/j.ypmed.2009.11.009 Google Scholar CrossRef Search ADS PubMed 82. King K , Meader N , Wright K et al. Characteristics of interventions targeting multiple lifestyle risk behaviours in adult populations: a systematic scoping review . Plos One . 2015 ; 10 ( 1 ): e0117015 . doi:10.1371/journal.pone.0117015 Google Scholar CrossRef Search ADS PubMed © Society of Behavioral Medicine 2018. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com

Journal

Annals of Behavioral MedicineOxford University Press

Published: Feb 2, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off