Externalizing psychopathology and behavioral disinhibition: working memory mediates signal discriminability and reinforcement moderates response bias in approach-avoidance learning

Abstract

Research has suggested that reduced working memory capacity plays a key role in disinhibited patterns of behavior associated with externalizing psychopathology. In this study, participants (N = 365) completed 2 versions of a go/no-go mixed-incentive learning task that differed in the relative frequency of monetary rewards and punishments for correct and incorrect active-approach responses, respectively. Using separate structural equation models for conventional (hit and false alarm rates) and signal detection theory (signal discriminability and response bias) performance indices, distinct roles for working memory capacity and changes in payoff structure were found. Specifically, results showed that (a) working memory capacity mediated the effects of externalizing psychopathology on false alarms and discriminability of go versus no-go signals; (b) these effects were not moderated by the relative frequency of monetary rewards and punishments; (c) the relative frequency of monetary rewards and punishments moderated the effects of externalizing psychopathology on hits and response bias for go versus no-go responses; and (d) these effects were not mediated by working memory capacity. The findings implicate distinct roles for reduced working memory capacity and poorly modulated active approach and passive avoidance in the link between externalizing psychopathology and behavioral disinhibition.

Figure 1

Signal Detection Theory model of Go/No-Go discrimination. Hypothetical No-Go (S−) and Go (S+) signal distributions are assumed to be equal-variance normally distributed along a single “No-Go” (R−) and “Go” (R+) response decision axis. As demonstrated in Macmillan and Creelman, (1990), the SDT model parameters signal discriminability (d'), decisional criterion (C), and response bias (β) can be calculated by applying the inverse normal density function transformation (IDF: Φ⁻¹) to the conditional probability of hits (pHT = R+|S+) and false alarms (pFA = R+|S−). These conditional probabilities were calculated by dividing the observed number of hits and false alarms by 28, which was the maximum number possible hits and false alarms. The d' parameter is calculated by taking the difference of the IDF transformed pHT and pFA quantities. The C parameter is calculated by taking the negative of one half of the sum of the IDF transformed pHT and pFA quantities; it is then used as an intermediate quantity in calculating the log-likelihood ratio criterion. The log-likelihood ratio criterion or response bias measure log(β) is calculated by taking the product of d' and C parameters.

Figure 2

Effects of latent externalizing problems (EXT) factor on discriminability (d') (panel A) and response bias (log(β)) (panel B)measures of Go/No-Go performance for the high-reward/low-punishment (HRLP, circles) and low-reward/high-punishment (LRHP, triangles) conditions. Linear regression model fits for the (HRLP, solid lines) and (LRHP, hashed lines) shown with 95% confidence intervals (CI).

Figure 3

Confirmatory factor analyses of competing two-factor (panel A) and common variance (panel B) measurement models of simple and complex working memory span task performance. Latent factors: EWMC = executive working memory capacity (EWMC), STMC = short-term memory capacity; EAC = executive attention capacity; COM = common variance; EXT = externalizing psychopathology. Manifest variables: ACT = Auditory Consonant Trigram; OWS = Operation Word Span; DFS = Digits Forward Span; DBS = Digits Backward Span; LNS = Letter Number Sequencing; ALC = Alcohol ; DRG = Illicit Drugs; AAB = Adult antisocial Behavior. All standardized regression weights and correlation coefficients are significant at p < .001.

Figure 4

Structural models used to test for meditation and moderation effects of latent working memory factors (EWMC – executive working memory capacity and STMC = short-term memory capacity) and experimental condition (HRLP = high-reward/low-punishment and LRHP = low-reward/high punishment), respectively, on the association between externalizing psychopathology (EXT) and separate measures of Go/No-Go performance: pFA = false alarms (panel A), pHT = hits (panel B), d' = discriminability (panel C), and β = response bias (panel D). Model parameters were freely estimated. Unmediated effects of latent EXT on Go/No-Go measures are shown in parentheses and corresponding path drawn with dotted lines. All path coefficients represent standardized regression weights with significant paths bolded (p < .05, †p < .07). Squared multiple correlations are the italicized quantities located behind latent and manifest variables. A) latent EXT effects on pFA are mediated by latent EWMC in both HRLP and LRHP conditions and these indirect effects are not moderated by condition. B) latent EXT effects on pHT are not mediated by latent working memory factors and these direct effects are moderated by condition. C) latent EXT effects on d' are mediated by both latent EWMC and latent STMC in both HRLP and LRHP conditions and these indirect effects are not moderated by condition. D) latent EXT effects on β are not mediated by latent working memory factors and these direct effects are moderated by condition. Multivariate kurtosis was .63, 3.27, −.51, .70 for models A–D, respectively.