Redish et al. treat addictions as maladaptive, repeated responses that arise through vulnerabilities in the interaction between a slow-learning, goal-independent habit system and a fast-learning, goal-oriented planning system. Although we applaud Redish et al.'s attempt to locate addiction within a “unified theory of decision making in the mammalian brain” (target article, Abstract), their approach ultimately traces vulnerabilities in the habit and planning systems to the direct pharmacological effects of addictive substances. In doing so, they neglect substantial evidence that the interplay between these behavioral control systems also is affected by a range of psychological and environmental factors that are commonplace elements of daily life and normal decision making (e.g., Reason Reference Reason1990; Wood & Neal Reference Wood and Neal2007). We argue that a unified theory of addiction must integrate these normal (i.e., non-pharmacologically induced) psychological processes. In support, we (1) review data showing that the role of habit- and planning-based behavioral control varies dynamically in everyday life and not simply in instances in which pharmacological agents have disrupted normal control, and (2) discuss specific implications for a subset of the vulnerabilities advanced in the target article.

As we have elaborated elsewhere, the distinction between habit-based and planning/goal-based behavioral control is studied across a range of subdisciplines within psychology and neuroscience (Neal et al. Reference Neal, Wood and Quinn2006; Wood & Neal Reference Wood and Neal2007). In laboratory settings, the two systems can be operationalized by relatively direct tests (e.g., presence versus absence of reinforcer devaluation effects in animals; Dickinson et al. Reference Dickinson, Balleine, Watt, Gonzalez and Boakes1995) or by manipulating the procedures of behavioral training (e.g., observational versus feedback-based learning in humans; Poldrack et al. Reference Poldrack, Clark, Paré-Blagoev, Shohamy, Moyano, Myers and Gluck2001). In real-world settings, the strength of the planning component is quantified by self-reported intentions to engage in the behavior in the future, whereas strength of the habit component is quantified by the frequency and context stability of past performance (see Neal & Wood, in press). The relative contribution of planning versus habit-based control can then be determined by regressing future behavior on each of these measures to determine which predictor is the strongest, and hence, which system primarily is controlling the behavior.

Real-world behavior prediction studies using this method typically yield statistical interactions in which strong habits trump strong intentions (i.e., intentions cease to predict behavior when habits are strong). This interaction has emerged in models predicting purchasing fast food, watching TV, driving the car, taking the bus, recycling, donating blood, exercising, and reading the newspaper (Danner et al., in press; Ferguson & Bibby Reference Ferguson and Bibby2002; Ji & Wood Reference Ji and Wood2007; Ouellette & Wood Reference Ouellette and Wood1998, Study 2; Verplanken et al. Reference Verplanken, Aarts, van Knippenberg and Moonen1998; Wood et al. Reference Wood, Tam and Guerrero Witt2005). Supporting the extrapolation from everyday habits to addictions, similar patterns have emerged in studies of smoking cessation. For example, in Baldwin et al.'s (Reference Baldwin, Rothman, Hertel, Linde, Jeffery, Finch and Lando2006) study, participants' initial success at quitting was predicted by planning-related constructs, whereas over the long term, only the number of months that participants had been quit predicted whether they continued not to smoke.

The behavior prediction data have important implications for Redish et al.'s model because they suggest the habit-based component of behavioral control often dominates actual behavior even in the absence of any dysfunction associated with a pharmacological agent. In such cases, we presumably must look to normal psychological processes for an explanation. A key question then becomes, to what extent do such normal psychological processes also contribute to addictions, for which pharmacologically induced dysfunction likely is a factor? In the remainder of this commentary, we consider this question by exploring several purely psychological factors that may contribute to Vulnerabilities 4, 5, 7, and 8, which Redish and colleagues trace instead to pharmacologically induced dysfunction.

Vulnerabilities 4 and 5 involve, respectively, overvaluation in the planning system and incorrect search in stimulus-action-outcome evaluation. These biases are attributed to the direct effects of addictive substances on neural systems underlying the valuation and search of action outcomes. However, there is an additional, pervasive information processing mechanism that may also contribute to this vulnerability. That is, people often rely on their past behavior to infer their preferences (Ouellette & Wood Reference Ouellette and Wood1998). Repeated actions lead to especially strong preference inferences (e.g., “I smoke a lot so I must really like it”). People tend to make such inferences when they are uncertain about the actual causes of their behavior (Bem Reference Bem and Berkowitz1972), as is often the case with habits. Thus, Vulnerabilities 4 and 5 plausibly may be induced or enhanced by a pervasive tendency to infer preferences from past behavior, leading to overvaluation or incorrect search for previously experienced outcomes.

Vulnerability 7, involving overvaluation in the habit system, also is traced in Redish et al.'s model to bottom-up effects of drugs on neurotransmitter systems. However, evidence from everyday habits indicates that higher-level decision-making processes involving switching costs also can lock people into repeating past actions and choices. That is, the familiarity of an action decreases decision costs and thus increases the relative cognitive switching costs for alternative actions (e.g., Murray & Häubl Reference Murray and Häubl2007). Of course, switching costs may well be encoded as overvaluation of habits in the dopaminergic system that Redish et al. describe in relation to Vulnerability 7 (see Niv et al. Reference Niv, Daw, Joel and Dayan2007). Our point is simply that overvaluation in the habit system need not solely be the product of drug-induced dysfunction but also may be the product of normal decision-making processes involving switching costs.

Vulnerability 8 addresses factors that disrupt the planning system and thereby reduce its scope to correct a misguided habit system. Although Redish et al. trace such disruptions to the pharmacological effects of drugs such as amphetamines and alcohol, our own model of habits predicts that the effortful inhibition of habits in favor of planned behavior is impaired by reductions in psychological self-control that are commonplace in daily life (Wood & Neal Reference Wood and Neal2007). This prediction builds on substantial evidence that effortful self-control draws on a single domain-general psychological resource that is temporarily depleted with use (Muraven & Baumeister Reference Muraven and Baumeister2000). Using field experiments, we have shown that reduced self-control impairs people's capacity to implement planned behavior and perpetuates habits in the real world (Neal et al. Reference Neal, Pascoe and Wood2008). Thus, the balance between habit and planning systems is dynamically responsive not solely to pharmacological agents but also to the many factors that deplete self-control in daily life (e.g., emotion regulation, complex decision making).

In summary, studies of everyday behavior repetition largely align with Redish et al.'s core premise regarding interacting behavioral decision systems corresponding to habit- and planning-based control. However, such studies also show that the relative influence of each system depends on a range of everyday psychological processes, including normal inferences, switching costs of decisions, and self-control capacity (Wood & Neal Reference Wood and Neal2007). A truly unified model of addiction must consider the contribution of these processes alongside pharmacologically induced dysfunction.