The Influence of Threat and Aversive Motivation on Conflict Processing in the Stroop Task

Ming-Ray Liao; Laurent Grégoire; Brian A Anderson

doi:10.3758/s13414-020-02072-5

. Author manuscript; available in PMC: 2021 Aug 1.

Published in final edited form as: Atten Percept Psychophys. 2020 Aug;82(6):2802–2813. doi: 10.3758/s13414-020-02072-5

The Influence of Threat and Aversive Motivation on Conflict Processing in the Stroop Task

Ming-Ray Liao ¹, Laurent Grégoire ¹, Brian A Anderson ¹

PMCID: PMC7428066 NIHMSID: NIHMS1609467 PMID: 32627132

Abstract

Motivated attention can be driven by the desire to maximize gains or escape punishment. In the Stroop task, when rewards can be obtained by responding quickly to certain colors, corresponding color-words are prioritized and produce enhanced interference, suggesting transfer of an attentional bias from color-hues to color-words. In the present study, we replicated this transfer effect using reward and conducted a parallel experiment exchanging the prospect of reward (appetitive motivation) with the opportunity to avert punishment (aversive motivation). Participants were required to identify the color (hue) of color-words and received electric shocks for responses to particular hues that were slow or incorrect. Shock-related words similarly impaired color-ink naming performance. In contrast to prior studies with reward, however, responding to hues associated with shock was also impaired, with threat producing an increase in error rate that ironically resulted in more frequent shocks. Our results suggest that aversive and appetitive motivation affect attention to task-relevant information differently, although each produces a common bias in automatic stimulus processing presumably driven by valence.

Keywords: Stroop, reinforcement learning, motivation, punishment, aversive conditioning

Introduction

Punishment, along with reward and other extrinsic motivators, are an important determinant of human and animal behavior (Maren, 2001; Schultz, 2004; Thorndike, 1911). Threat detection is especially important to prevent and avoid harm, and aversive outcomes have been posited to heighten sensitivity to negative information (Vuilleumier, 2005, for a review). Reward has been shown to facilitate both visual search and cognitive control, possibly through increased neural representation of reward-associated stimuli via biased competition (Anderson, 2019; Bijleveld, Custers, & Aarts, 2010; Knutson, Adams, Fong, & Hommer, 2001). Consequently, task-irrelevant reward-associated stimuli can impair task performance (Anderson, 2016, for a review). Some studies have shown punishment to have similar effects to reward (Anderson & Britton, in press; Kim & Anderson, in press; Nissens, Failing, & Theeuwes, 2017; Schmidt, Belopolsky, & Theeuwes, 2015, 2017; Van Damme et al., 2004). For example, Van Damme et al. (2004) reported that visual targets were detected faster when their location was predicted by a cue conditioned with a pain signal, relative to a neutral cue. When presented as a distractor in a visual search task, a stimulus previously conditioned with an aversive electrical stimulation impairs performance compared to a neutral stimulus, independent of perceptual salience (Schmidt et al., 2015). Thus, the effect of punishment on information processing can be conceptualized as being similar to reward, where both types of outcome serve as motivational inputs to attention and cognition (Engelmann & Pessoa, 2007). How extrinsic motivators shape behavior in the presence of conflicting information, as in the Stroop task or other conflict paradigms, is less understood.

The classic Stroop task entails the presentation of stimuli comprising two dimensions, word and color (Stroop, 1935; see MacLeod, 1991, for a review). Participants are required to identify the ink color (hue) of color-words while ignoring their meaning. Performance is usually impaired when the two dimensions are incongruent (e.g., the word “red” presented in blue ink) and improved when the two dimensions are congruent (e.g., the word “blue” presented in blue ink), relative to a neutral situation (e.g., a non-word presented in blue ink). It has been proposed that automatic reading of the irrelevant dimension (i.e., word) interferes with processing of the relevant dimension (i.e, ink color; Carter & Van Veen, 2007; Cohen, Dunbar, & McClelland, 1990). Krebs, Boehler, and Woldorff (2010) investigated the effect of appetitive motivation in the Stroop task by associating two ink colors with the opportunity to earn a monetary reward for fast and accurate responses, and found improvements in color-naming for reward-associated hues. They also found that there were greater performance deficits for reward-related words (appetitive association) compared to words unrelated to reward, suggesting transfer of a motivated attentional bias from one feature dimension (hue) to another (word meaning).

Prospect theory proposes that humans are more sensitive to losses than to reward (Kahneman & Tversky, 1979; but see Ert & Erev, 2013; Lejarraga et al., 2019; Yechiam & Hochman, 2013a, 2013b; Yechiam, 2019). Loss aversion explains why people are more willing to take risks to avoid a loss than to procure a gain (Schindler & Pfattheicher, 2016), and how penalties are more effective than reward in motivating people (Gächter et al., 2009). However, Wentura, Müller, and Rothermund (2014) showed that there were no differences in the magnitude of attentional capture by reward- and loss-associated distractors (see also Grégoire et al., in press). Carsten, Hoofs, Boehler, and Krebs (2019) tested predictions arising from prospect theory by modifying the paradigm from Krebs et al. (2010) to compare how rewards and losses facilitated task performance by having one hue associated with reward and the other associated with loss. While both rewards and losses improved Stroop task performance, participants were slower in loss trials compared to reward trials. The authors suggested that the valence of the association may have modulated responses such that positive valence facilitated approach behavior, while negative valence had a contradictory mapping with avoidance behavior that had to be overcome (Guitart-Masip, Duzel, Dolan, & Dayan, 2014). In other words, loss is potentially inducing both appetitive and aversive associations, which differently affect attention.

The present study follows up on Krebs et al.’s (2010) and Carsten et al.’s (2019) studies, but we exchanged the prospect of reward with the opportunity to avert punishment as administered with electric shocks. Monetary loss may have been more motivating with respect to negative reinforcement, especially in the context of other available gains, leading to a focus on maximizing overall value and the results seen in Carsten et al. (2019). We therefore opted for a punisher that is inherently aversive and would serve as a more potent unconditioned stimulus in the context of associative learning. In a color-word Stroop task, two ink colors were associated with shock, while two other ink colors were never followed by shock. Participants were asked to identify the ink color ignoring the semantic meaning of the word. For ink colors associated with shock, slow responses and errors resulted in the delivery of shock. Much like rewarding fast responses, this manipulation provided (in this case, aversive) motivation to maximize performance on trials with a shock-associated ink color. Although prior research has investigated how the emotional content of the irrelevant dimension interferes with task performance (Algom, Chajut, & Lev, 2004; Fox, Russo, Bowles, & Dutton, 2001; Fox, Russo, & Dutton, 2002; Stormark, Nordby, & Hugdahl, 1995; Yiend & Mathews, 2001), few studies have applied that information to the relevant dimension (color hue) via associative learning.

If the prospect of procuring a reward, averting a loss, and averting an aversive physical punisher have similar motivational influences on attention, we should observe the same effects that Krebs et al. (2010) and Carsten et al. (2019) did – a general facilitation of performance (faster responses and increased accuracy) for shock-associated colors. In addition, the attentional bias for shock-associated colors should spillover to corresponding words, impeding performance on incongruent trials with shock-related words. Alternatively, punishment associations might have a generally detrimental effect on performance, similar to emotionally charged words (e.g. Algom et al., 2004; Williams, Mathews, & MacLeod, 1996).

EXPERIMENT 1

Method

Participants

Thirty-six participants, between the ages of 18 and 35 inclusive, were recruited from the Texas A&M University community. All reported normal or corrected-to-normal visual acuity, normal color vision, and provided written informed consent. Data from two participants were excluded from the analyses due to low proportion of correct responses. Specifically, the error and miss rates exceeded three standard deviations from the group mean. The final sample included 34 participants (24 females), with a mean age of 19.62 years (SD = 1.92). Using the smallest effect size for the main effect of reward on either the relevant (hue) or irrelevant (word) stimulus dimension in Krebs et al. (2010) yielded power β > 0.90 with α = 0.05 for the obtained sample size (G*Power 3.1). Participants were compensated with $10 USD or course credit. All procedures were approved by the Texas A&M University Institutional Review Board and conformed with the principles outlined in the Declaration of Helsinki.

Apparatus

A Dell OptiPlex equipped with the Matlab software and Psychophysics Toolbox extensions (Brainard, 1997) was used to present the stimuli on a Dell P2717H monitor. Electric shocks were administered to participants’ left forearms via paired electrodes (EL500, BioPac Systems, Goleta, CA, USA) linked to an isolated linear stimulator (STMISOLA, BioPac Systems) set to constant current and controlled by MATLAB scripts.

Stimuli

The stimuli consisted of five color words: red, green, blue, yellow, and brown. The words appeared in lowercase font (Arial 40) and were presented in four prototypical ink colors: red, green, blue, and yellow. The semantic meaning of a given word could be congruent (e.g. “red” written in red ink) or incongruent (e.g. “red” written in green ink) with regards to the ink color. Furthermore, trials consisting of words with no response mapping (e.g. “brown” written in red ink) were intermixed to provide a neutral category (Figure 1). It is important to note that the word “brown” is neutral with respect to valence in that it is neither associated with shock nor the absence of shock (safe response) as a reported color, although it is not neutral with respect to Stroop congruity (even though it does not have an explicit response mapping, the word is still incongruent with the ink color). It is in this sense of valence that we use the label “neutral,” consistent with Krebs et al. (2010). Of interest in the present study is the comparison of words and ink colors that differ with respect to their relationship to shock, and in this context “neutral” words merely provide an opportunity to subtract out any effect of the ink color on responses when assessing the influence of the valence of the words. A complete breakdown of how all of the different color/word combinations differ with respect to outcomes (shock), Stroop congruity, and response mapping is provided in Supplemental Table 1.

Figure 1. — Stimuli and experimental conditions (counterbalanced across participants). A subset of ink colors was associated with the potential of shock (potential-shock; e.g., red and green), while the remaining ink colors were not (no-shock; e.g., blue and yellow). The word meaning (irrelevant dimension) could be congruent, incongruent shock-unrelated, incongruent shock-related, or neutral with regard to the valence of the ink color.

Procedure

Throughout the experiment, a grey fixation square (visual angle 0.5°) was maintained at the center of a black background screen. Each trial began with the fixation square for a duration that varied randomly between 1200 and 1600 ms, followed by the Stroop stimulus (vertical visual angle between 1.3° and 1.5°, horizontal visual angle between 2.6° and 5.6°) on top of the square for 1500 ms or until response. On punished trials, a 2-ms shock was administered 1000 ms after the Stroop stimulus (Figure 2).

Figure 2 – — Sequence of events in the Color-Stroop task. The fixation square remained in the middle of the screen for the entire task. Each trial began with the fixation square for a random duration between 1200 and 1600 ms, followed by the stimulus on top of the square for 1500 ms or until response. When a shock is delivered like in A), the fixation square remained for 1000 ms before the shock is administered for 2 ms. Otherwise, the trial ends like in B).

Participants were required to report the ink color of words as fast and accurately as possible while ignoring their semantic meaning. Participants were instructed to manually enter their responses using a standard (QWERTY layout) keyboard with their left and right index and middle fingers on the ‘z’, ‘x’, ‘n’ and ‘m’ keys. Each key was mapped onto one of four colors. Color-key assignments were pseudo-randomized across participants such that there would always be one color associated with shock for each hand. This was to avoid participants associating one hand with punishment.

Before the Stroop task, participants had eight trials to learn the color-key pairing with colored squares (visual angle 3.0°). A colored square remained on the screen until the correct key for that color was pressed. Next, participants completed eight practice trials of the Stroop task with no time limit and feedback messages if they pressed an incorrect key, after which there were forty more practice trials with no feedback and a 1500 ms time limit. Participants were then connected to the shock device and shock intensity was adjusted individually to be “unpleasant but not painful” (Grégoire & Greening, 2019, 2020; Kim & Anderson, 2019; Murty, LaBar, & Alison Adcock, 2012; Schmidt et al., 2015, 2017). Finally, participants completed four runs of the Stroop task: two runs with shocks (learning phase) followed by two runs without shock (extinction phase). Each run consisted of 120 trials for a total of 480 trials.

The 120 trials per run consisted of six blocks of 20 randomized trials. Each block consisted of all the possible color-word and color-ink pairings such that there were four congruent trials and 12 incongruent trials, and four neutral trials. Half of each block (10 trials) were with shock-associated ink. The response time (RT) cutoff was adjusted dynamically to be the median of the participants’ RT over the past 40 trials. If the participant responded incorrectly or slower than the cutoff on trials with shock-associated ink, then participants would receive an electric shock, with similar performance-dependent frequency as in Krebs et al. (2010). There were no consequences for responding incorrectly or slowly for non-shock-associated ink colors. However, there was a six-shock limit per block. Before performing the learning phase, participants were informed that they might receive a shock “if [they] respond slowly or incorrectly.” Before performing the extinction phase, participants were explicitly instructed that no electric stimulation would be delivered anymore and shock electrodes were disconnected.

Data Analysis

Analyses were conducted separately for learning phase (i.e., the two runs with shock) and extinction phase (i.e., the two runs without shock). RTs for correct responses beyond three standard deviations of the conditional mean (< 1% of responses for all analyses) were removed from the data for each participant (Grégoire, Perruchet, & Poulin-Charronnat, 2013, 2014, 2015). The remaining data were submitted to repeated measure analysis of variance (ANOVA) with the aversive association (color: potential-shock, no-shock) and the congruity (word: congruent, incongruent, (valence) neutral) as within-subjects variables. The same analysis was conducted for error rates, which included misses and incorrect responses. In order to investigate differential effects of shock-related and shock-unrelated word information, additional ANOVAs were conducted focusing on incongruent trials, with color-ink (color: potential-shock, no-shock) and color-word (shock-related, shock-unrelated) as within-subjects variables. The error rates and RTs used for these analyses were relative to the corresponding neutral condition (i.e., performance from the neutral condition with the same shock-ink association was subtracted; see Krebs et al., 2010) in order to more effectively isolate the influence of the words’ association with shock.

Results

Stroop Effect

The difference in errors and RT in the learning phase amounted to a Stroop effect (incongruent – congruent) of 3.25%, t(33) = 5.18, p < 0.001, d = 0.89 , and 51 ms, t(33) = 8.55, p < 0.001, d = 1.47. The difference in RT in the extinction phase amounted to a Stroop effect of 42 ms, t(33) = 5.53, p < 0.001, d = 0.95, but there was no Stroop effect for errors, p = 0.124.

Learning phase

Errors.

ANOVA revealed a significant main effect of aversive association, F(1,33) = 4.68, p = 0.038, $η_{p}^{2} = 0.124$ , with more errors in potential-shock compared to no-shock trials. We also observed a significant main effect of congruity, F(2,66) = 11.65, p < 0.001, $η_{p}^{2} = 0.261$ , such that errors varied as a function of condition – incongruent, congruent and neutral (Figure 3A & Table 1). Errors were more frequent for incongruent compared to congruent and neutral conditions, ts > 3.25, ps < 0.004, ds > 0.55 The interaction between aversive association and congruity was not significant, F(2,66) = 0.53, p = 0.590.

Figure 3 – — Mean error rates for the learning phase. A. Error rates between potential-shock (red bars) and no-shock ink (blue bars) for congruent, neutral and incongruent trials. Error bars represent standard error of the means. B. Error rates for shock-related and shock-unrelated incongruent trials are depicted as the difference relative to corresponding neutral words.

Table 1.

Effects of relevant and irrelevant shock associations on performance

Color (aversive association)		Word (congruity)
		Congruent	Incongruent		Neutral
		Congruent	Shock-unrelated	Shock-related	Neutral
Learning phase
Potential shock	RT (ms)	651.16 (94.25)	694.07 (85.60)	718.09 (107.35)	684.33 (91.81)
Potential shock	Error rate	5.02 (5.50)	8.29 (4.79)	8.70 (6.52)	7.11 (6.03)
No shock	RT (ms)	642.18 (90.38)	692.00 (80.45)	702 .07 (89.13)	674.26 (80.37)
No shock	Error rate	3.92 (4.46)	5.39 (4.76)	8.70 (4.96)	4.78 (4.60)

Extinction phase
Former potential shock	RT (ms)	633.98 (75.30)	680.54 (66.13)	690.80 (65.75)	668.11 (74.85)
Former potential shock	Error rate	5.99 (4.94)	5.69 (4.10)	7.16 (5.92)	5.21 (4.01)
Former no shock	RT (ms)	647.65 (75.88)	670.68 (66.97)	685.13 (71.28)	649.33 (58.95)
Former no shock	Error rate	4.56 (5.31)	6.58 (5.35)	7.75 (4.51)	5.34 (5.83)

Open in a new tab

Note. Standard deviations are in parentheses.

Focusing on incongruent trials only, we observed a main effect of color-word, F(1,33) = 7.12, p = 0.012, $η_{p}^{2} = 0.177$ , with more errors for shock-related words compared to shock-unrelated words (see Figure 3B). The main effect for the color-ink was not significant, F(1,33) = 0.70, p = 0.409. We also observed a significant interaction between color-word and color-ink, F(1,33) = 4.16, p = 0.049, $η_{p}^{2} = 0.112$ . Errors were comparatively more frequent on incongruent trials with shock-related words and no-shock ink compared to incongruent trials with shock-unrelated words, which was confirmed by pairwise comparisons, ts > 2.59, ps < 0.015, ds > 0.44.

Response times.

ANOVA revealed a significant main effect of congruity, F(2,66) = 35.77, p < 0.001, $η_{p}^{2} = 0.520$ . Incongruent words were slower than congruent and neutral words, and neutral words were slower than congruent words, ts > 3.28, ps < 0.003, ds > 0.56 (Figure 4A). The main effect of aversive association was not significant, nor was the interaction between aversive association and congruity, Fs < 1.

Figure 4 – — Mean RT from the learning phase. A. RTs between potential-shock (red bars) and no-shock ink (blue bars) for congruent, neutral, and incongruent trials. Error bars represent standard error of the means. B. RTs for shock-related and shock-unrelated incongruent trials are depicted as the difference relative to corresponding neutral words.

Focusing on incongruent trials, there was a main effect of color-word, F(1,33) = 7.54, p = 0.010, $η_{p}^{2} = 0.186$ , such that shock-related words slowed responses more than shock-unrelated words (Figure 4B). There was no main effect of color-ink and no interaction effect, Fs < 1.

Extinction phase

Errors.

ANOVA with aversive association (color: potential-shock, no-shock) and congruity (word: congruent, incongruent, neutral) as within-subjects variables revealed no significant main effect or interaction (all ps > 0.10; Table 1). The ANOVA on incongruent trials only yielded no significant main effect or interaction (all ps > 0.10).

Response times.

ANOVA revealed a significant main effect of congruity, F(2,66) = 16.08, p < 0.001, $η_{p}^{2} = 0.328$ . Incongruent words were slower than congruent and neutral words, and neutral words were slower than congruent words, ts > 2.44, ps < 0.021, ds > 0.41 (Table 1). There was no main effect of aversive association, F(1,33) = 0.64, p = 0.430, and a marginally significant interaction effect, F(2,66) = 2.80, p = 0.068, $η_{p}^{2} = 0.078$ .

Focusing on incongruent trials, there was a marginal effect of color-word that mirrored the pattern during the training phase, F(1,33) = 3.84, p = 0.059, $η_{p}^{2} = 0.104$ . The main effect of color-ink and its interaction with color-word were not significant (all ps > 0.10).

Discussion

In Experiment 1, participants performed a modified version of the color-word Stroop task in which a subset of ink colors were associated with punishment (shocks). Participants had an opportunity to avoid shock if they responded accurately and fast enough. In contrast with prior research using reward (Bijleveld et al., 2010; Engelmann & Pessoa, 2007; Krebs, Boehler, & Woldorff, 2010) and loss (Wentura, Müller, & Rothermund, 2014; Müller, Rothermund, & Wentura, 2016) as a motivator, shock-association (in both color-ink and color-words) interfered with, rather than enhanced, task performance.

In the learning phase, there was an overall increase in errors for trials with potential-shock ink compared to no-shock ink, as well as an overall increase in both errors and RT for trials with shock-related words compared to shock-unrelated words. The difference in RT remained marginally significant in extinction, while the difference in errors did not. The observed difference in error rate was particularly striking, as participants made the most errors on trials on which they should have been the most motivated to perform well (to avoid shock), which was distinctly counterproductive in that it resulted in more overall shocks being delivered. That is, not only did aversive motivation fail to improve task performance, it ironically impaired performance. Although the typical Stroop effect was observed across all incongruent trials, such interference was particularly pronounced for shock-related words, consistent with prior results investigating the consequences of color-reward associations (Krebs et al., 2010).

Because our design involved more incongruent trials compared to congruent trials overall (as in Krebs et al., 2010), there was inevitably more shocks received on incongruent trials. Our results could have been influenced by this imbalance in design, which may have affected baseline performance on incongruent trials, but it cannot account for the differential processing of shock-associated stimuli. Such an imbalance is necessary to avoid biases in how frequently each color-word combination is presented, which could compromise the comparisons of interest.

A pronounced difference in our design compared to that of Krebs et al. (2010) is that participants were not explicitly informed of the colors associated with shocks. Awareness may have contributed to more efficient deployment of attention to the relevant colors to maximize gains and minimize loss (Krebs et al., 2010; Carsten et al., 2019). Implicit learning of associations may also have contributed to the persistence of attentional biases in RT. In order to draw direct contrasts between aversive and appetitive motivation in the Stroop task, we conducted a second experiment that exchanged the opportunity to avoid shocks with the prospect to earn rewards while keeping everything else the same. In order to better understand the potential role of awareness of the task contingencies, we also included an awareness assessment at the end of the experiment.