Approach, avoidance, and their conflict: the problem of anchoring

To understand the neurobiology of individual differences in approach and avoidance behavior, we must anchor constructs at the behavioral level to the long-term global sensitivities of the neural systems that give rise to the observed stable patterns of behavior. We will argue that this requires not only appropriate data at both the neural and behavioral levels but also appropriate account to be taken of interactions at the intervening level of the conceptual nervous system (Hebb, 1949; Gray, 1975). In particular, in accounting for approach and avoidance behavior we must include consideration of the distinction between valuation and motivation (Corr and McNaughton, 2012), of interactions between the approach system and the avoidance system (Gray and Smith, 1969), and of their interaction with a distinct additional system that is activated by approach-avoidance conflict (Gray, 1977; summarized in Corr, 2013). But first we need to ask why would we expect there to be traits linked to global approach and avoidance systems? Simple animals (with little or no brain) can produce approach and avoidance behavior (toward benefits and ultimately reproduction; and away from dangers and ultimately failure to reproduce) via multiple independent rules of thumb (Krebs et al., 1983). But we can expect more complex brains to have largely integrated these simple elements into systems more generally dedicated to approach or avoidance “because this is how [a few] genes can build a complex system that will produce appropriate but flexible behavior to increase fitness. . . . Rather than just pre-programmed movements such as tropisms and taxes, . . . if the genes are efficiently to control behavior . . . they must specify the goals for action” (Rolls, 2000, pp. 183, 190). Together with the evolution of general approach and avoidance systems that are not tied to any specific motivating stimulus (reinforcer), we would expect evolution of the long-term adaptive control of their overall sensitivity to adequate inputs. Such stable sensitivity would be the neurobiological basis of approach and avoidance personality traits. Determining the appropriate neurobiological measure for the sensitivity of a highly evolved approach or avoidance system is not simple. These systems have hierarchically organized neural levels with processing ranging from “quick and dirty” to “slow and sophisticated” for both perception (LeDoux, 1994) and action (Graeff, 1994, 2010). Sensitivity to input determines which level of the system is activated and so sensitivity cannot reside in any one of the modules within the system (McNaughton and Corr, 2004). The source of any sensitivity must, therefore, be identified independently—in essence requiring at least a preliminary surface level description of traits. Existing theories of personality provide a number of competing surface level, lexically-derived, systems with trait measures that relate to approach and avoidance either indirectly via constructs such as Extraversion and Neuroticism (Eysenck, 1957) or directly via constructs such as Harm Avoidance (Cloninger et al., 1993). Each system is stable, with links to mental disorder (Strelau and Zawadzki, 2011; Gomez et al., 2012; Mullins-Sweatt and Lengel, 2012; Trull, 2012) and brain structure (Gardini et al., 2009; DeYoung et al., 2010). But even when starting with approach and avoidance as primary constructs, they are derived “top-down” from pools of lexically-chosen questionnaire items (Carver and White, 1994; Elliot and Thrash, 2010) not from biological anchors. They also depend on factor analysis, which determines the number of dimensions, but not location of trait axes of the personality “space” that items occupy (Lykken, 1971; Corr and McNaughton, 2008). It is little more than an act of faith to believe that the causal structure of personality is isomorphic with its lexical factor structure. So, even if we knew for certain that there were only two dimensions within a particular measured personality space, one questionnaire system could have a single simple trait anxiety dimension (orthogonal to, say, impulsiveness) that was a combination of neuroticism and introversion in another (Gray, 1970)—the two systems differing only on which items from an original pool were used to create scales. Factor analytically derived trait measures can also easily meet the criterion of having “simple structure” (in the sense that a set of items loads highly on only one factor so factors can be clearly identified by unique item loadings) while implying improbable causation (Lykken, 1971). Further, not only is there no reason to suppose that biologically accurate scales should have simple structure but also current scale systems, even though designed to have this, often do not (DeYoung, 2006, 2010).