Regulatory Fit Effects on Stimulus Identification

Abstract

This article examines the effects of a fit between a person's global regulatory focus and the local task reward structure on perceptual processing and judgment. On each trial, participants were presented with one of two briefly presented stimuli and were asked to identify it. Participants were placed in a promotion focus (a situationally induced sensitivity to gains) or a prevention focus (a situationally induced sensitivity to losses) and were asked to maximize gains or minimize losses. An asymmetric payoff ratio biased the overall reward toward one identification response over the other. Two experiments tested the role of regulatory fit when internal familiarity and perceptual sensitivity was low or high. When familiarity and sensitivity were low, participants in a regulatory fit (promotion focus with gains or a prevention focus with losses) showed greater perceptual sensitivity, but no response bias differences relative to participants in a regulatory mismatch. When familiarity and sensitivity were high, participants in a regulatory fit showed a response bias toward the high payoff stimulus, but no differences in perceptual sensitivity. Speculation on the neurobiological basis of this effect, as well as implications of this work for clinical disorders, such as depression, is offered.

Motivation influences people's performance on an array of tasks ranging from complex learning to simple perceptual discrimination (Humphreys & Revelle, 1984). It is crucial to understand these influences, particularly on perceptual tasks, because considerable effort is spent trying to motivate individuals like air traffic controllers and radiologists whose job success depends on efficient and effective perceptual processing. Despite the importance of motivation both in laboratory studies and in the workplace, the effect of motivational manipulations like incentives is still poorly understood (Brehm & Self, 1989; Locke & Latham, 2002; Maddox & Markman, 2010).

In this paper, we extend previous work from our lab to focus on the effects of incentives on a perceptual discrimination task. As we discuss, these perceptual discrimination tasks allow us to explore influences of motivation on both judgment and internal perceptual noise. We start with a review of the regulatory fit motivational framework that has driven our research. Then, we introduce the experimental procedure and present two studies that explore the effects of motivational incentives on performance.

The regulatory fit framework

Previous research from our lab suggests that the influence of an incentive on performance is best conceptualized as an interaction between the motivational state induced by the incentive and the rewards or punishments provided by a task (Maddox, Baldwin, & Markman, 2006; Maddox, Markman, & Baldwin, 2006; Markman, Baldwin, & Maddox, 2005; Markman, Maddox, Worthy, & Baldwin, 2007; Worthy, Maddox, & Markman, 2007). This idea is related to Higgins’ concept of a regulatory fit (Avnet & Higgins, 2003; Higgins, 2000).

On this view, a situational incentive induces a promotion focus (when participants try to achieve a positive reward) or a prevention focus (when participants try to avoid a negative punishment). For example, when people are performing a task, they may have the opportunity to earn a ticket into a drawing to win $50 (promotion focus) or they may be given a ticket for the drawing when entering the lab and have to perform well or else they lose the ticket (prevention focus).

Tasks also have local incentives. Points can be obtained on each trial (gains) with the goal of maximizing points or points can be lost on each trial (losses) with the goal of minimizing losses. We believe that both gain maximization and loss minimization are ecologically valid constructs. For gains, consider a seashell collector who finds many good seashells if a good part of the beach is chosen, but only a few seashells if a bad part of the beach is chosen. Regardless of success, the number of seashells collected is always increasing. For losses, consider a delivery driver with a limited amount of gasoline for his many routes. If a good route is chosen then a smaller amount of gas will be consumed, but if a poor route is chosen then a larger amount of gas will be consumed. Regardless of success, the amount of gasoline is always decreasing.

A regulatory fit occurs when there is a global approach incentive and the task involves a local gains incentive, or a global avoidance incentive with a local losses incentive (promotion-gains and prevention-losses). The other states create a regulatory mismatch (promotion-losses and prevention-gains; see Table 1). An example of a promotion-gains regulatory fit would be a seashell collector who is trying to collect enough seashells to gain an entry into a local seashell competition. An example of a prevention-loss regulatory mismatch would be a delivery driver who was trying to avoid being fired by trying not to running out of gas on his route. In our work, we find an interaction between global situational incentives and local task reward incentives on performance.

Table 1.

Regulatory Fit Framework

		Local Task Reward Structure
		Gains	Losses
Global Regulatory Focus	Promotion Focus	Fit	Mismatch
	Prevention Focus	Mismatch	Fit

It is critical to note that this interaction of global and local incentives further interacts with properties of the task at hand. For example, Grimm, Markman, Maddox, & Baldwin (2007) observed a three-way interaction of global incentives, local incentives, and whether flexible strategies facilitate or hinder performance in a classification task. In other words, the interaction of global and local incentives resulted in either a fit interaction or a mismatch interaction, which additionally interacted with task processing requirements. Those in a regulatory fit interaction demonstrated better classification performance when the task required a flexible, rule-based strategy. Conversely, those in a regulatory mismatch interaction demonstrated better performance when the task required a more incremental, implicit strategy. This three-way interaction eludes simple main effect explanations, and signifies that a regulatory fit between global and local incentives does not universally lead to better performance regardless of task properties.

Studying Perceptual Identification

This article applies the three-factor motivation-cognition framework to a perceptual identification task in two experiments. Previous research suggests that a regulatory fit between global and local incentive has consequences for higher-level decision making. Markman, Baldwin and Maddox (2005) demonstrated that regulatory fit leads to differences in the classification of items drawn from continuous and overlapping category distributions. In that study, the decision criterion needed to maximize reward differed from that required to maximize classification accuracy, because rewards associated with one category were three times greater than those associated with the other category (see Green & Swets, 1966, Maddox, 2002for details). People have a natural tendency to maximize accuracy, and so flexibility is required in order to abandon the decision criterion that maximizes accuracy for the one that maximizes reward (Maddox, 2002; Maddox & Bohil, 1998; Maddox & Dodd, 2001). Given this set up, participants in a regulatory fit exhibited flexibility by shifting their decision criterion away from the accuracy maximizing criterion and toward the reward maximizing criterion.

We build on this established higher-level response bias effect and investigate lower-level processing using tools offered by signal detection theory. Detection and identification of briefly presented stimuli is a hallmark of the sensory and perceptual systems, forming the bases for much higher level cognition (Wilson, 2002). The studies outlined above all used classification tasks in which large numbers of physically unique stimuli were each sampled from normally distributed categories, and the participants’ task was to classify each stimulus into its associated category. These paradigms have their roots in early “external noise” distribution studies whose aim was to make the “noise” characteristics observable (i.e., external) so that decision processes could be studied (Healy & Kubovy, 1981; Lee & Zentall, 1967). In the present investigation, we return the more basic signal detection/perceptual identification paradigm in which only two unique stimuli are relevant and the participant is asked to identify which stimulus is presented on each trial. The stimuli are highly perceptually confusable and are presented for short durations significantly increasing the influence of internal noise. To our knowledge this represents the first study to examine the impact of regulatory fit on perceptual and decisional processes in stimulus identification.

To study stimulus identification with internal noise characteristics, we introduce a critical change from our prior work. Markman, Baldwin, and Maddox (2005) examined category learning with simple uni-dimensional stimuli. Each category contained a large number of unique stimuli with each stimulus being sampled from a univariate normal category distribution (i.e., with external noise). Critically, each stimulus was presented in a high contrast, response terminated display. Thus, while perceptual noise was negligible, noise was introduced into the process by utilizing highly overlapping category distributions. In the stimulus identification paradigm used in this article, only two stimuli are relevant, but “noise” is introduced by using highly confusable stimuli and short display durations. Thus, the main source of noise is perceptual and comes from the internal representation of each stimulus. Because there are two stimuli and two responses in stimulus identification, performance in the task can be quantified by calculating the probability of a “hit” and a “miss” for each stimulus. Specifically, for Stimulus₁ there are Hits₁ and Misses₁, and for Stimulus₂ there are Hits₂ and Misses₂. These labels are equivalent to the traditional signal detection theory measures of hits (Hits₁), misses (Misses₁), correct rejections (Hits₂) and false alarms (Misses₂) taken relative to Stimulus₁ as the Signal and Stimulus₂ as Noise.

These quantities are used to calculate two fundamental measures of stimulus identification performance: perceptual sensitivity and response bias (see Figure 1). Perceptual sensitivity is a measure of the ability to discriminate between the two stimuli (i.e., the distance between the perceptual distributions of the stimuli along an evidence x-axis). Response bias is a measure of the tendency to endorse, or select one response choice over the other. Sensitivity is usually influenced by stimulus characteristics such as stimulus discriminability and exposure duration, and response bias is influenced by the rewards and/or costs associated with correct and incorrect responses for each stimulus (McCarthy & Davidson, 1979; Wickens, 2001; but see Balakrishnan, 1999). A person's performance in a stimulus identification task can vary along these two independent measures of performance (e.g., high perceptual sensitivity and moderate response bias, low perceptual sensitivity and moderate response bias, etc).

Figure 1.

Stimulus identification properties of a simple identification task.

Stimulus characteristics can drive both empirical perceptual sensitivity and response bias. Typically, increased discriminability leads to increased empirical sensitivity. Response bias can be manipulated, for example, with the instantiation of an asymmetric local reward structure. In a symmetric payoff situation, equal rewards are given for Hits₁ and Hits₂, and also equal rewards are given for Misses₁ and Misses₂. The optimal response bias (β) is computed as follows:

$$\beta =\frac{(Hit{s}_1-Misse{s}_1)P\left(Stimulu{s}_1\right)}{(Hit{s}_2-Misse{s}_2)P\left(Stimulu{s}_2\right)}$$ (1)

where Hits_i and Misses_i refers to the reward given for either a hit or miss response to Stimulus_i and P(Stimulus_i) is the proportion of times Stimulus_i is shown (i.e., base rate).

Assuming equal base-rates, in a symmetric payoff situation, the reward values in Equation 1 are equal for both stimuli. Under these conditions, β = 1 as shown below:

$${\beta }_{Symmetric}=\frac{(200-100)\left(.5\right)}{(200-100)\left(.5\right)}=\frac{100}{100}=1$$ (2)

The accuracy maximizing criterion, given by the ratio of the base rates, is also 1. Therefore, in a symmetric payoff situation, the accuracy maximizing criterion is the same as the reward maximizing criterion.

Again assuming equal base-rates, in an asymmetric payoff situation, the reward values are unequal. For example, in the present tasks, Hits₁ = 400, Hits₂ = 200 while Misses₁ = Misses₂ =100. Under these conditions, the optimal decision criterion = 3 as described below:

$${\beta }_{Asymmetric}=\frac{(400-100)\left(.5\right)}{(200-100)\left(.5\right)}=\frac{150}{50}=3$$ (3)

The accuracy-maximizing decision criterion = 1. Thus, the reward maximizing criterion differs from the accuracy maximizing criterion requiring participants to abandon their natural tendency toward accuracy-maximization in favor of reward-maximization.

This β value was chosen as a starting point based on our prior research. The local reward structure of the present tasks all have a β = 3. However, there are two between-participant conditions that differ in the specific reward values. In the gains condition, participants gain more points for correct responses than for incorrect responses while approaching a positive bonus point criterion. In the losses condition, participants lose fewer points for a correct response than for incorrect responses while avoiding a negative bonus point criterion (see Figure 2). In both scenarios, more points were given for correct responses, but importantly, more points were given for response to one stimulus relative to the other. Because of the equal base rates and the use of two discrete stimuli, both high accuracy and a shifted response criterion result in high task performance (i.e., more gains or fewer losses). For this reason, it was crucial for overall discriminability to be low such that perfect accuracy was unlikely.

Figure 2.

A. Sample display from the gains condition. B. Sample display from the losses

There are three ways that regulatory fit might affect performance in this internal noise experiment. First, regulatory fit might affect the location of the decision criterion as in prior work with stimuli with external noise. In previous work we found that a regulatory fit increased participants’ willingness to abandon accuracy maximization in the interest of maximizing reward leading to a shift in the decision criterion in a two-category classification task (Markman, et al., 2005). Extending this effect to stimulus identification would suggest that the regulatory fit interaction previously observed with suprathreshold, high discriminable stimuli and response-terminated displays extends to internal noise scenarios with two stimuli. Second, an intriguing possibility is that regulatory fit increases perceptual sensitivity. There is some neuropsychological evidence to suggest that this might be the case. For example, increased dopamine in frontal areas (that might be associated with a regulatory fit; Maddox et al, 2006) could increase the signal-to-noise ratio associated with projections from sensory representation areas into prefrontal cortex (Ashby, Casale, 2003; Ashby, et al., 2005). This is an exciting possibility because it suggests that the interaction of global and local incentives can affect low-level perceptual processing. The neuropsychological basis for this possibility is discussed in more detail in the General Discussion. Lastly, regulatory fit might affect both criterion placement and perceptual sensitivity. Based upon our previous work (Markman, et al., 2005) our a priori prediction is that regulatory fit will affect the placement of the decision criterion. Even so, the perceptual sensitivity hypothesis is compelling.

Overview of the Current Studies

Experiment 1 used a perceptual identification task with two highly confusable stimuli, short exposure duration, and limited training. Global regulatory focus (promotion or prevention focus) and local task reward structure (gains or losses) were manipulated between participants. Based on prior work, we expected global regulatory focus and local reward structure to interact such that those in a regulatory fit (promotion-gains and prevention-losses) would demonstrate a response bias shift away from the accuracy-maximizing decision criterion and towards the rewards-maximizing decision criterion. This prediction was based on a similar finding of the role of regulatory fit in the placement of the decision criterion in category learning (Markman, et al., 2005). To anticipate we were surprised to find that those in a regulatory fit demonstrated increased perceptual sensitivity relative to those in a regulatory mismatch (promotion-losses and prevention-gains).

In Experiment 2, we attempted to bridge the gap between the procedures associated with our stimulus identification task in Experiment 1 with those used in our prior work (Markman et al., 2005). The idea was to distinguish the lower-level consequences of regulatory fit observed with novice participants performing a perceptually difficult stimulus identification task (Experiment 1), from the higher-level consequences common in our prior work (Maddox, Baldwin, & Markman, 2006; Maddox, Markman, & Baldwin, 2006; Markman, Baldwin, & Maddox, 2005; Markman, Maddox, Worthy, & Baldwin, 2007; Worthy, Maddox, & Markman, 2007) by modifying the procedure used in Experiment 1. During pre-training, we provided participants with extensive experience with the task and adjusted the stimulus presentation time so that participants could reliably achieve 70% accuracy. There was no global regulatory focus or local reward structure manipulation during pre-training resulting in a symmetric feedback scenario where optimal β = 1 (see Equation 2). Once performance stabilized (see details in Experiment 2), we introduced the global regulatory focus (promotion vs. prevention) and local reward structure (gains vs. losses) manipulations and modified the payoff matrix so that optimal β = 3. There was also no global regulatory focus or local reward structure manipulation during pre-training. We hypothesized that, under these conditions, a regulatory fit would increase flexibility in such a way that participants would shift their decision criterion away from the accuracy-maximizing criterion toward the reward-maximizing criterion. To anticipate, participants in a regulatory fit demonstrate a shift toward the reward-maximizing decision criterion. In the General Discussion, we speculate on the neurobiological underpinning of these effects, and discuss implications of this work for recent work in clinical populations, such as patients with depression.

Experiment 1

In Experiment 1 participants were given a situational promotion or prevention focus, and were asked to complete several hundred trials in which they identified one of two schematic faces that differed only in the width of the mouth (see Pizzagalli, Jahn, & O'Shea, 2005 for a similar task). On each trial, one of the two (randomly selected) stimuli was presented for 100ms, and the participant decided whether the mouth was “short” (Stimulus₁) or “long” (Stimulus₂). Participants in the gains condition received points for each response, but received more points for correct responses (Hits₁ and Hits₂) than for incorrect responses (Misses₁ and Misses₂; see Figure 2A). Participants in the losses condition lost points for each response, but they lost fewer points for correct responses than for incorrect responses (Figure 2B). Importantly, the payoff ratio was asymmetric with Hits₁ yielding a better payoff than Hits₂. The exact payoffs are displayed in Table 2. In both conditions, a higher score was more desirable than a lower score. Importantly, these payoffs were designed to yield the same optimal decision criterion for gains and losses (β = 3; see Equation 1).

Table 2.

Payoff Matrices and Performance Criterion for the Between-Participants Manipulation of Gains and Losses

	Experiment 1		Experiment 2
	Gain	Loss	Gain	Loss
Hit1(Higher reward)	400	-111	4	-1
Hit2 (Lower reward)	200	-311	2	-3
Miss1	100	-411	1	-4
Miss2	100	-411	1	-4
Performance Criterion	22400	-28600	130	-120

Method

Seventy-six participants from the University of Texas community completed the study with 21, 20, 18, and 17 in the promotion-gains, promotion-losses, prevention-gains, and prevention-losses conditions, respectively. Participants were either paid $6 for their time or were given course credit. In addition, participants were all given the opportunity to receive entries into a drawing for a one-in-ten chance to win $50. The data from 12 participants (3 in promotion-gains, 3 in promotion-losses, 4 in prevention-gains, and 2 in prevention-losses) were eliminated from analysis due to perseverative responding or a majority of trials with response times below 200ms, possibly due to frustration. Testing occurred individually in a well lit room.

The stimuli were two schematic faces that differed only in the width of the mouth, with the short mouth measuring 36 pixels, and the long mouth measuring 38 pixels. Both faces shared an outer oval shape and the position of two dots to represent eyes (see examples in Figure 2). On each trial, one (randomly selected) stimulus was presented for 100ms, followed by 500ms Gabor patch mask. The participant pressed the “short” or “long” mouth key and was informed of the number of points earned or lost. A point meter incremented or decremented following each response. The asymmetric payoff matrix (see Table 2) created a reward dichotomy between Stimulus₁ and Stimulus₂ and was necessary for measuring task-appropriate changes in response bias (see Equation 1).

Participants completed a 40-trial practice phase in which they received corrective feedback with no payoff structure. This was followed by 3 100-trial blocks of stimulus identification. Before the first block, the global regulatory focus manipulation was introduced. Promotion focus participants were told that in each block, if their performance exceeded the criterion, they would receive one raffle ticket. Prevention focus participants were given three raffle tickets at the start of the experiment. They were told that they would lose a raffle ticket in each block of trials that they lost too many points.

After the training phase, the payoff structure was explained to the participant. Gains participants were told that they needed to gain enough points to exceed the criterion to obtain or keep the raffle ticket. Losses participants were told that they needed to minimize their losses to stay above the criterion to obtain or keep the raffle ticket. The performance criterion for two payoff conditions was set at 68% of the difference between the highest and lowest possible point total for a block of trials, and is displayed in Table 2. Points were reset to zero at the beginning of each block.

Results

Sensitivity and response bias are important measures in stimulus identification tasks. Sensitivity is related to accuracy and is a measure of a participant's ability to discriminate between the two stimuli in the task. Response bias is a measure of the tendency of a participant to favor one response over the other. Here, we use empirical measures of sensitivity (log d; Equation 4) and response bias (log b; Equation 5). These measures are well behaved at various levels of accuracy (McCarthy & Davidson, 1979; Tripp & Alsop, 1999). They were chosen in lieu of d-prime and Beta because Beta is most interpretable when a single level of discrimination is imposed on a stimulus space with external noise, or when a small range of empirical sensitivities are recovered from an identification task with internal noise. While both paradigms are parametric in that d-prime and Beta assume a normal distribution and log d and log b assume a logistic distribution, we found log b to be most resilient to a range of empirical sensitivities and less distorted by the level of discriminability. Most applications of either method assume equal variances. We make the same assumption but address its implications in the General Discussion.

For log d, positive values represent a tendency to correctly identify both Stimulus₁ and Stimulus₂. For log b, positive values represent a tendency to identify stimuli as Stimulus₁ (the high payoff stimulus) rather than Stimulus₂ (the low payoff stimulus). Thus, larger values are indicative of larger shifts toward the reward-maximizing criterion.

$$\log d=0.5\ast \log \left(\frac{Hit{s}_1\ast Hit{s}_2}{Misse{s}_1\ast Misse{s}_2}\right)$$ (4)

$$\log b=0.5\ast \log \left(\frac{Hit{s}_1\ast Misse{s}_1}{Hit{s}_2\ast Misse{s}_2}\right)$$ (5)

Perceptual Sensitivity (log d)

The stimulus identification sensitivity (log d) values are displayed in Figure 3A and were subjected to a 2 [focus: promotion, prevention] × 2 [local reward structure: gains, losses] × 3 [block] ANOVA. Global regulatory focus interacted with local reward structure, F(1, 60) = 9.86, p = .003, ε² =.14. Pairwise t-tests suggested higher sensitivity for participants in the promotion-gains condition (log d = .91) relative to participants in the promotion-losses condition (log d = .64), t(33) = 2.20, p = .035, and higher sensitivity for participants in the prevention-losses condition (log d = 1.12) relative to participants in the prevention-gains condition (log d = .82), t(27) = 2.26, p = .016. There was also no significant difference between the promotion-gains and prevention-losses condition, t(31) = 1.69, p = .10, nor the promotion-losses and prevention-gains condition, t(29) = 1.36, p = .19. Proportion correct (i.e., accuracy) is highly correlated with log d, although not a linear transformation. Proportion correct analyses led to the same results.

Figure 3.

Performance measures from Experiment 1 for each block and collapsed across blocks. Error bars represent standard error.

Stimulus Identification Decision Criterion (logb)

The stimulus identification decision criterion (log b) values are displayed in Figure 3B and were subjected to a 2 × 2 × 3 ANOVA. The main effect of local reward structure was nearly significant, F(1, 60) = 3.96, p = .051, ε² =.062), and suggested that participants in the gains condition showed a larger shift toward the reward-maximizing decision criterion relative to the losses condition. Because we expected the regulatory fit interaction to affect response bias, we compared the regulatory fit groups versus the regulatory mismatch groups for both sensitivity and bias. The difference was significant for log d, t(62) = 3.08, p = .003, d = .73, mean difference 95% CI [-0.47, -0.10], but not for log b, t(62) = -.50, p = .62, d = -.17, mean difference 95% CI [-.13, .27].

Effects of Learning

Although block did not interact significantly with global regulatory focus or local task reward structure, an examination of Figure 3 suggests a consistent trend whereby the effects of regulatory fit (i.e., better performance for promotion-gains relative to promotion-losses and prevention-losses relative to prevention-gains) tended to be largest in the first block of trials and diminished over blocks. In fact, performance for regulatory fit participants was fairly stable across the three blocks, whereas there was a general tendency for performance of regulatory mismatch participants to start poor and gradually improve. The interaction of global regulatory focus and local reward structure was significant for log d in blocks 1 and 2, but was non-significant in block 3, F(1, 60) = 14.67, p < .001, ε² =.196 in Block 1, F(1, 60) = 5.45, p = .023, ε² =.083 in Block 2, and F(1, 60) = 2.21, p = .14, ε² =.036 in Block 3. Thus, it does not appear that a regulatory fit leads to higher asymptotic levels of performance, but rather leads to faster initial learning.

In summary, in an stimulus identification task with internal noise and a short exposure duration we find an interaction of global incentives (regulatory focus) and local incentives (local task reward structure), such that a regulatory fit interaction leads to higher sensitivity, but not a decision criterion shift. Although we predicted an effect of regulatory fit on response bias based on previous work, there was some reason to expect a regulatory fit effect on sensitivity. In Experiment 2, we test whether decreased perceptual noise will lead to the regulatory fit effect on response bias which we have shown in previous work.

Experiment 2

The novel results from Experiment 1 suggest that the interaction of global and local incentives has consequences for low-level perceptual processing. Our prior work with regulatory fit and higher-level decision making reveals motivational consequences for higher-level processes such as decision criterion placement in a classification task. To determine the juncture between regulatory fit effects on decision criterion placement in a classification task and perceptual sensitivity in a stimulus identification task, we examined the possibility that increased training and familiarity with the stimulus identification task allows participants in a regulatory fit to develop a response bias towards the high reward stimulus. To this end, we modified Experiment 1 to promote increased overall accuracy and perceptual sensitivity by increasing trial-by-trial exposure duration and offering increased familiarity through training. We predict that increased training and familiarity with the stimulus identification task would lead those in the regulatory fit condition to shift bias towards the higher reward response.

Method

Ninety-one participants from the University of Texas community completed the study with 24, 19, 26, and 22 in the promotion-gains, promotion-losses, prevention-gains, and prevention-losses conditions, respectively. In this analysis, we were only interested in the stimulus identification performance of those who successfully achieved the bonus criterion in at least one of the motivation blocks.² This led to the exclusion of 4 participants from promotion-gains, 4 from promotion-losses, 7 from prevention-gains, and 5 from prevention-losses.

To achieve higher overall accuracy, we introduced a titration phase consisting of 4 blocks of 50 stimulus identification trials with corrective feedback. These blocks occurred before the regulatory focus manipulation was instantiated and they had a symmetric local task reward structure in that no points were given as feedback. After each titration block, the stimulus presentation time was adjusted based on Equation 4.

$$newtime=\frac{oldtime}{accuracy}\ast 0.7$$ (4)

Where oldtime refers to the stimulus presentation time carried over from the previous trial, accuracy refers to cumulative accuracy of the participant on the previous trials, and the target accuracy rate is set to 0.7. The initial stimulus presentation time was set at 200msec. This training procedure was effective yielding a mean accuracy of 77% (SE = 0.01) at the end of the titration phase.

The stimuli in this study were gray 50×50 pixel gray patches at two levels of brightness. We abandoned the face paradigm because the gray patches gave us more control for a titration procedure. We found that even a 1-pixel change in mouth length had a drastic effect on accuracy that was too large for the aims of the titration procedure. On each trial, participants were instructed to classify a presented stimulus by pressing one of two response buttons. After the titration blocks, the motivational manipulation was instantiated as in Experiment 1. Participants engaged in 3 motivation blocks of 50 trials. As in Experiment 1, an asymmetric payoff matrix was constructed with an optimal decision criterion (β) of 3. See Table 2 for the payoff matrix used in Experiment 2. Since the higher accuracy would inevitably lead to more participants reaching the bonus point criterion, the criterion was raised to 80% of the maximum possible points minus the minimum possible points. Stimulus masks and response keys were the same as in Experiment 1. The proportion of those to achieve the bonus was such that there is a main effect of Local Reward Structure, F(1,67) = 8.32, p = .005, but no interaction of global regulatory focus and local reward structure.

Results

Perceptual Sensitivity (logd)

The stimulus identification sensitivity (log d) values are displayed in Figure 4A and were also subjected to a 2 × 2 × 3 ANOVA. The relative ease of the task in comparison to Experiment 1 led participants to greater sensitivity as the task proceeded. The main effect of Block was significant, F(1,67) = 17.3, p < .001, ε² = .21 as well as the main effect of Local Reward Structure, F(1,67) = 9.1, p = .004, ε² = .12. Importantly, and unlike the results from Experiment 1, there was no interaction of global regulatory focus and local reward structure. Analysis of the proportion correct data yielded the same pattern.

Figure 4.

Performance measures from Experiment 2 for each block and collapsed across blocks. Error bars represent standard error.

Stimulus Identification Decision Criterion (logb)

The stimulus identification decision criterion (log b) values are displayed in Figure 4B and were subjected to a 2 × 2 × 3 ANOVA. There was a significant interaction of Global Regulatory Focus and Local Reward Structure, F(1,67) = 4.8, p = .033, ε² = .07. The interaction reflects that participants in a regulatory fit were more strongly biased away from an accuracy-maximizing criterion and towards a higher payoff reward-maximizing criterion (i.e., response shift towards Stimulus₁) than those in a regulatory mismatch. Pairwise t-tests suggested a shifted response bias for participants in the promotion-gains condition (log b = .29) relative to participants in the promotion-losses condition (log b = .02), t(33) = 2.18, p = .037, but only a directional non-significant difference for participants in the prevention-losses condition (log b = .02) relative to participants in the prevention-gains condition (log b = .15), t(34) = 0.95, p = .35. There was also no significant difference between the promotion-gains and prevention-losses condition, t(35) = 0.99, p = .33, nor the promotion-losses and prevention-gains condition, t(32) = 0.02, p = .98. There were no significant main effects of Global Regulatory Focus or Local Reward Structure. Because we expected the regulatory fit interaction to affect sensitivity, we compared the regulatory fit groups versus the regulatory mismatch groups for both sensitivity and bias. The difference was significant for log b, t(69) = 2.26, p = .03, d = .51, mean difference 95% CI [-.39, -.02], but not for log d, t(62) = .29, p = .77, d = .12, mean difference 95% CI [-.40, .24].

In summary, Experiment 2 featured a titration training procedure which led to decreased perceptual noise and a regulatory fit effect on response bias, as found in our previous work. This is in contrast to the regulatory fit effect on sensitivity found in Experiment 1, for which the internal perceptual representation were noisier. We discuss the significance of these results below.

General Discussion

This article addresses the effect of incentives on performance using a simple stimulus identification task. Consistent with previous work, we find not only an interaction between the global situational incentive and the local task reward structure, but also the processing required by the task itself. This study goes beyond previous research, however, by demonstrating that a regulatory fit between the global regulatory focus and the local task reward structure affected performance differentially depending on the level of internal perceptual noise. In the higher internal perceptual noise situation, regulatory fit increased perceptual sensitivity (as measured by log d from signal detection theory and traditional measures of accuracy)³, whereas in the lower internal perceptual noise situation, regulatory fit affected response bias in that those in a regulatory fit were more likely to abandon their natural tendency toward accuracy-maximization in favor of reward-maximization, and thus an increased chance of obtaining or keeping the raffle ticket entry. The increased empirical discriminability in Experiment 2 resulted from the longer exposure duration on each trial and increased familiarity with the stimuli due to the increased training phase. This allowed participants to attend to the asymmetric reward characteristics of the task. With perceptual discriminability no longer a struggle for participants, response bias effects emerged for participants in the two regulatory fit conditions.

Our results do indicate that perceptual sensitivity is influenced by regulatory fit in situations of low perceptual discriminability, although it may be possible that perceptual sensitivity is also influenced in higher discriminability situations albeit to a smaller degree. It is also possible that the regulatory fit effect on sensitivity is larger early in learning followed by the emergence of an effect on response bias later in learning, due to the fact that discriminability oftentimes increases through experience with stimuli

We must emphasize that these studies (as in our previous work) can be explained only by theories that postulate an interaction between the global regulatory focus, local reward structure, and task properties on performance. Theories that posit overall differences between global promotion and prevention focus (e.g., Crowe & Higgins, 1997, Higgins & Scholer, 2009) or local gains and losses only (e.g., Kahneman & Tversky, 1979) are not sufficient to account for the present results. Furthermore, although not manipulated in the present experiment, prior work demonstrates an additional tertiary interaction between regulatory fit and task demands. For example, the increased exploratory response pattern of those in a regulatory fit can hinder performance in tasks where an exploitative response pattern is optimal (Worthy, Maddox, & Markman, 2007; Worthy, Maddox, & Markman, 2008). This ternary interaction (regulatory focus × local reward structure × task demands) attenuates engagement and/or attention hypotheses regarding regulatory fit. In summary, this pattern of results demonstrates that regulatory fit is not a universally beneficial psychological state and further precludes the usefulness of theories that ignore the interactive properties of global and local incentives.

In Experiment 1, we predicted an effect of regulatory fit on response bias. Instead, we found an effect on perceptual sensitivity. Although initially surprising, an examination of the relevant neuroscience literature supports this finding. Neuroscience research suggests that motivational state might influence sensitivity when perceptual noise is large (as with the briefly presented stimuli in Experiment 1). A body of evidence converges in suggesting that a regulatory fit might increase cortical dopamine levels (Maddox, et al., 2006). Briefly, the evidence includes work that suggests a relationship between positive affect and cortical dopamine levels (Ashby, Ell, Valentin, & Casale, 1999; Isen, 1993; Isen, 1999), and a relationship between global regulatory focus and frontal brain activation (Amodio, Shah, Sigelman, Brazy, & Harmon-Jones, 2004; Cunningham, Raye, & Johnson, 2005). Increased cortical dopamine is thought to increase the signal-to-noise ratio on projections from sensory representation areas with prefrontal cortical working memory cells (Ashby, Casale, 2003; Ashby, et al., 2005). An increase in signal-to-noise ratio should increase sensitivity leaving the response bias unaffected. Although based on solid neurobiological evidence, this proposal requires further study.

Previous work suggests that a regulatory fit might affect the placement of the decision criterion in identification. This did not occur in the high perceptual noise situation of Experiment 1, but did occur in the low perceptual noise situation of Experiment 2. Similar to the previous research, those in a regulatory fit shifted response towards the higher payoff options, a reward-maximizing criterion shift that inherently involves a tradeoff from accuracy (Crowe & Higgins, 1997; Markman, et al., 2005).

This work also has important clinical implications. A cardinal attribute of depression is decreased approach-related behavior, which is similar in spirit to a decreased ability to maintain a promotion focus (Davidson, 1998; Sloan, Strauss, & Wisner, 2001; Watson, et al, 1995). In one study, Pizzagalli, et al. (2005) tested the hypothesis that depressed patients are insensitive to manipulations of the payoff structure because of reductions in their approach-related motivational system. Pizzagalli et al. (2005) used a two-stimulus identification task (that inspired our task), but importantly only gains were presented. Pizzagalli et al. (2005) showed that depressed patients had less optimal decision criterion placement than controls. If depression induces a debilitated promotion focus, then a prevention focus may be in effect, meaning depressed patients with a gains reward structure are in a regulatory mismatch and might be expected to show sub-optimal decision criterion placement. Some evidence in support of this hypothesis can be found in our own data. Furthermore, Experiment 2 suggests the intriguing hypothesis that depressed patients might perform better when required to minimize losses, as opposed to maximizing gains, because a losses reward structure would create a regulatory fit for patients with approach related problems, leading to a de facto prevention focus.

The current research also adds to previous support that work in regulatory focus can inform the study of the conscious-unconscious distinction. Markman, Maddox, Worthy, and Baldwin (2007) make the case that explicit versus implicit processing become preferentially activated due to current regulatory state. The current experiments further this distinction by suggesting that task demands dictate which aspects of implicit perceptual processing are affected by regulatory focus (i.e., sensitivity or bias).

In conclusion, a fit between global regulatory focus and local reward structure affects perceptual tasks as well as cognitive tasks. When perceptual noise is higher, regulatory fit leads to greater sensitivity than regulatory mismatch. When perceptual noise is lower, regulatory fit leads to a shift in response bias toward the optimal reward-maximizing decision criterion.

Limitations

Although we are unable to support a strong null effect for response bias in Experiment 1 and empirical discriminability in Experiment 2, our key findings are for the interactions of global situational incentives and local task reward. In Experiment 1 this interaction affected discriminability, while in Experiment 2 it affected response bias. These are both novel findings which demonstrate that motivational influences can interact to moderate low-level stimulus identification in important ways.

In the signal identification paradigm, it is possible that differences in empirical discriminability arise from a difference in variance between the perceptual distributions (as opposed to a difference in the distance between the distribution means). Receiver operator characteristic (ROC) curves would need to be constructed over several criterion levels in order to rule this out. Thus, we cannot conclude with certainty that the regulatory fit effect observed in Experiment 1 is solely due to an effect on sensitivity. Even so, it is clear that a regulatory fit effect does emerge in stimulus identification, and to our knowledge this has not been demonstrated previously.

One potential limitation of the present study is the use of different stimuli between the two experiments. In Experiment 1, line length was relevant and in Experiment 2, the brightness of a gray patch was used. Although it is possible that line length is more likely to lead to regulatory fit effects on sensitivity, and brightness is more likely to lead to regulatory fit effects on response bias, we find this highly unlikely. Over the past 20 years we have run numerous identification and classification studies using line length, brightness and other simple perceptual dimensions as stimulus dimensions and we find no fundamental differences in their effects on performance. In fact, Sagiv and Bentin (2001) found that although humans can recognize schematic faces as a type of face, schematic faces are processed more like nonface stimuli. In our view, the most reasonable interpretation of these data is that the level of internal noise inherent in the two studies (high in Experiment 1 and low in Experiment 2) is the operative factor leading to fit effects on sensitivity in Experiment 1 and fit effects on response bias in Experiment 2.

Our study also lacks a control condition without imposed incentives. While this could be a topic for future research, by imposing local, situational manipulations we hope to alter any initial or default motivational states that the participant may present with. Another topic for future research is the nature of the payoff manipulation and starting conditions of the stimulus identification task.