Rhesus Macaques (Macaca mulatta) Exhibit the Decoy Effect in a Perceptual Discrimination Task

Abstract

The asymmetric dominance effect (or decoy effect) is a form of context-dependent choice bias in which the probability of choosing one of two options is impacted by the introduction of a third option, also known as the decoy. Decoy effects are documented widely within the human consumer choice literature and even extend to preference testing within nonhuman animals. Here, we extended this line of research to a perceptual discrimination task with rhesus monkeys to determine whether decoy stimuli would impact size judgments of rectangular stimuli. In a computerized task, monkeys attempted to choose the larger of two rectangles that varied in their size and orientation (horizontally or vertically oriented). In probe trials, a third stimulus (the decoy) was presented that was smaller than the other two rectangles but matched the orientation of one of them. Half of the probe trials presented a decoy that matched the orientation of the larger stimulus, and the other half presented a decoy that matched the orientation of the smaller stimulus. Monkeys rarely selected the decoy stimulus. However, their performance (selection of the largest rectangle) increased relative to the baseline trials (with only two choices) when the decoy was congruent in its orientation with the largest rectangle, but decreased relative to baseline when the decoy was incongruent with the largest rectangle. Thus, a decoy stimulus impacted monkeys’ perceptual choice behavior even when it was not a viable choice option itself. These results are explained with regard to comparative evaluation mechanisms.

A homebuyer’s realtor has helped her narrow down the many real estate options to two choices. The first home is fully renovated and a half-hour drive from work, whereas the second home is un-renovated but a mere five minute walk from work. The buyer cannot decide between the two options as she values both home quality and short commute-time. Her realtor introduces a third option also known as the decoy – a partially renovated home that boasts a lengthy one hour-commute. As this option is weaker on both dimensions of home quality and commute time than the first home, the buyer chooses the first fully renovated home with the half-hour commute by car. Rational choice theory’s condition of regularity states that the relative preferences for the original alternatives within a set should not change with the introduction of a third choice as each value is made independently of one another (Luce, 1959; Tversky, 1972). However, decision-making sometimes is at odds with rational choice theory and instead, choice behavior may be impacted by context. For the buyer, this so-called decoy option altered her preference for the original two options, such that it increased the preference for the first home that dominated the decoy on both dimensions of home quality and commute time.

The decoy effect, also known as the asymmetric dominance effect, is a form of context-dependent choice in which the probability of choosing one of two options is impacted by the introduction of a third, weaker option to the choice set (Huber, Payne, & Puto, 1982). In a choice set, the original two options vary on multiple dimensions such that the first option dominates the second on one dimension, and the second option dominates the first on a separate dimension. The decoy stimulus increases the probability of choosing the original item that it is most similar to. Further, there is a positive correlation between this choice behavior and the similarity of the two items (Huber et al., 1982).

In our example, the first home was stronger in the dimension of home quality, but the second home was stronger in the dimension of commute time. If the homebuyer values both of these dimensions to an equal extent, the options are rendered equally attractive and a decision-maker is indifferent between the two options. The introduction of a decoy stimulus that is similar, but only dominated by one of the original options (and thus asymmetrically dominated) results in a shift in the preferences for the original options (Huber et al., 1982). In the home example, the buyer’s preference for the first home increased with the introduction of the decoy home that was weaker than the first on both dimensions of quality and commute time. The asymmetric dominance effect is discussed in light of comparative versus absolute evaluations. Within comparative assessments, a decision-maker assigns relative values to an option, which may change as the choice set increases or decreases. This stands in contrast to an absolute mechanism in which an assigned value is made independent of other alternatives, as predicted by the condition of regularity (e.g., Bateson, Healy, & Hurly, 2002; Luce, 1959).

Although decoy effects have been widely documented in the consumer research literature (e.g., Doyle, O’Conner, Reynolds, & Bottomley, 1999; Huber et al., 1982; Parducci, 1965; Pettibone & Wedell, 2000; Wedell, 1991), the phenomenon extends to various forms of human preference testing outside of the consumer domain, including mate-choice behavior (Sedikides, Ariely, & Olsen, 1999), policy decisions (Herne, 1997), and political races (Pan, O’Curry, & Pitts, 1995). Moreover, the asymmetric dominance effect has been investigated and demonstrated to varying degrees within several nonhuman animal and insect species in which the addition of a decoy stimulus resulted in a shift in the preferences for the original options within a choice set. The nonhuman species tested include ants (Edwards & Pratt, 2009), honeybees and gray jays (Shafir, Waite, & Smith, 2002), hummingbirds (Bateson et al., 2002; Hurly & Oseen, 1999), starlings (Bateson, 2002; Schuck-Paim, Pompilio, & Kacelnik, 2004), and cats (Scarpi, 2011). For example, Bateson et al. (2002) investigated hummingbird decision-making behavior in a foraging paradigm that involved choices between food items (artificial flowers) differing on two dimensions (volume of nectar and sucrose concentration), such that one of the original two options was more valuable in terms of sucrose concentration (15 µl; 40% sucrose) and the second option was more valuable in terms of nectar volume (45 µl; 30% sucrose). A third stimulus (the decoy) was introduced that was inferior to the first flower in both dimensions (10 µl; 35% sucrose). The authors documented a shift in the relative preference of the original items with the introduction of the decoy to the choice set, demonstrating that additional alternatives impacted choice behavior in preference testing with nonhuman animals (Bateson et al., 2002). The authors concluded that, like humans, animal choice behavior is subject to violations of the regularity condition as evaluations are made comparatively depending upon context, rather than made absolutely. Thus, the impact of a decoy option on decision-making appears to be widespread across multiple human domains, extending to nonhuman species.

Decoy effects are prevalent not only in traditional decision-making tasks using preference testing, but they also exist within perceptual discriminations (Choplin & Hummel, 2005; Trueblood, Brown, Heathcote, & Busemeyer, 2013; Tversky, 1972). Imagine being offered the choice between a round cake and a square cake that are nearly identical in terms of their total amount. No clear preference for one over the other would exist. But, then imagine a third cake, round in shape and smaller than the first round cake, was offered. Now, if a preference emerged for the larger round cake over the square cake, this would be an example of the decoy effect because the original options within the set were evaluated against a third, weaker option (the decoy) that changed an original indifference to a preference for the member of the original set that dominated the decoy.

Perceptual decoy effects are typically studied using some type of discrimination task in which participants must differentiate stimuli on the basis of a physical attribute(s), such as height and/or width. Performance is then measured after the introduction of an additional alternative, which may enhance the selection of one of the original options. For example, human participants chose between rectangles of variable size, including an asymmetrically dominated decoy that was introduced to enhance preference for one of the two original rectangles (Trueblood et al., 2013). Depending upon the nature of the decoy presented, there was an increase in the probability of selecting the option that the decoy was similar to, the option that it was dissimilar to, and the option that was rendered a compromise between alternatives. The authors discussed their results in light of the preferential-choice literature in which these three context effects have been documented in higher-level decision-making tasks (Huber et al., 1982; Simonson, 1989; Tversky, 1972).

Although there is less research with perceptual decoy effects than value-driven decoy effects, these studies offer the unique opportunity to explore context effects in basic, perceptual-discrimination tasks in which decisions regarding stimuli are generated early and quickly. Moreover, studies within the perceptual domain are excellent candidates for use in comparative research with nonhuman animals (especially primates) that have demonstrated psychophysical performance patterns in size-discrimination studies that rival or even exceed that of human performance (e.g., Menzel, 1960, 1961; Menzel & Davenport, 1962; Schmitt, Kröger, Zinner, Call, & Fischer, 2013). Finally, perceptual tasks of this nature bypass the requirement for language-based instructions or extensive training histories or specific forms of knowledge that may impact decision-making strategies or responses.

Although decoy effects have been documented within several nonhuman animal species using preference testing, we wanted to explore if and how they manifest in a perceptual-discrimination task with monkeys. Our objective was to determine whether a decoy stimulus would impact discrimination performance of rhesus monkeys in a perceptual task to the same extent observed in non-perceptual preference tasks observed with some animal species (e.g., Bateson, 2002; Bateson et al., 2002; Shafir et al., 2002; Scarpi, 2011; Schuck-Paim et al., 2004). Moreover, we were interested in documenting whether the decoy effect would impact performance in the same direction as that observed in humans in a size-discrimination task in which a similar, but inferior decoy enhanced the preference for the focal option.

In the current study, we introduced a perceptual size-discrimination task similar to that used by Trueblood et al. (2013; Exp. 1) with humans. In our study, rhesus monkeys (Macaca mulatta) attempted to choose the larger of two rectangular stimuli that varied in their size and orientation (horizontally or vertically oriented) in each trial. We varied task difficulty to generate a range of difficulty in the ability to choose the larger rectangle, in the hopes that this would also highlight the extent to which a decoy effect could interfere with typical discrimination performance. In baseline trials, only two stimuli were presented – there was no decoy. In probe trials, a third stimulus (the decoy) was presented that was always smaller than the other two rectangles but matched the orientation of one of those rectangles. Half of the probe trials presented a decoy that matched the orientation of the larger stimulus (Congruent condition), and the other half presented a decoy that matched the orientation of the smaller stimulus (Incongruent condition). Essentially, the decoy stimulus should have elicited a comparison between itself and the original rectangle in the same orientation. If the decoy effect emerged in the current study, we anticipated that the Congruent condition would lead to increased performance (selection of the larger rectangle) relative to the Baseline condition as the same-orientation rectangle was truly the larger of the original set and comparatively larger than the decoy. Alternatively, the Incongruent condition should lead to decreased performance relative to the Baseline condition because the same-orientation rectangle was the smaller of the original set but comparatively larger than the decoy.

Experiment 1

Methods

Subjects

Seven adult male rhesus monkeys between the ages of 11 and 31 years were tested. Monkeys were singly housed for six of seven days each week, but with continuous visual and auditory access to other monkeys housed in the same colony room. One day per week each monkey was paired with a socially compatible conspecific in an indoor-outdoor enclosure during which time monkeys did not participate in research projects but instead were free to engage in enrichment activities and interact with each other. When singly housed, monkeys had continuous access to a dedicated computer system for assessing various cognitive capacities (e.g., Agrillo, Parrish, & Beran, 2014; Beran, Evans, Klein, & Einstein, 2012; Beran & Parrish, 2013; Evans & Beran, 2012; Evans, Perdue, Parrish, & Beran, 2014; Smith, Coutinho, Church, & Beran, 2012; Smith, Flemming, Boomer, Beran, & Church, 2013). These test sessions typically lasted between four to six hours, and monkeys worked or rested at their own choosing throughout these sessions. Monkeys always had access to water, and were given a daily meal each afternoon independent of whether and how much they worked on the computer task.

Apparatus

The monkeys were tested using the Language Research Center’s Computerized Test System (Figure 1). This system consisted of a personal computer, digital joystick controller, color monitor, and pellet dispenser (Evans, Beran, Chan, Klein, & Menzel, 2008; Richardson, Washburn, Hopkins, Savage-Rumbaugh, & Rumbaugh, 1990). Monkeys manipulated the joystick with their hands to control a small cursor somewhere on the computer screen. Contacting stimuli with the cursor sometimes resulted in the delivery of 94-mg banana-flavored chow pellets (Bio-Serv, Frenchtown, NJ) via a pellet dispenser that was connected to the computer. The task program was written in Visual Basic 6.0.

Figure 1.

Experimental setup for the individual test stations at which the monkeys worked on the task. Test stations include a personal computer, monitor, joystick controller, and pellet dispenser.

Design and Procedure

We presented monkeys with a size discrimination task on a white background. At the outset of each trial, the monkey had to move the cursor upward into contact with a grey rectangle at the top center of the computer screen. This was the trial initiation response. When the rectangle was contacted, it disappeared, and two or three black rectangles appeared at the top of the screen in the left, center, and/or right positions (positions were randomly assigned across trials by the computer program). The monkey then had to move the cursor, which had reappeared in the center of the screen, into contact with one of those rectangles. If the selected rectangle was the largest rectangle on the screen for that trial, the monkey heard a melodic tone and was rewarded with a single food pellet, the screen was cleared, and after a 1 s inter-trial interval the grey rectangle and cursor reappeared at the start of the next trial. Incorrect responses led to a buzz tone, and the screen was cleared for a 20 s timeout period during which the screen remained blank before presentation of the grey rectangle at the start of the next trial.

Training sessions

During training sessions, monkeys saw only two rectangles on the screen during each trial (Figure 2). One rectangle was presented such that its height was greater than its width (called the “Tall” rectangle) and the other had a width greater than its height (called the “Wide” rectangle). On half of these trials, the program generated a relatively small rectangle with a width that was randomly chosen from the range of 120 to 169 pixels and a height randomly chosen from a range of 60 to 109 pixels. On the other half of the trials, the program generated a relatively small rectangle with a height that was randomly chosen from the range of 120 to 169 pixels and a width randomly chosen from a range of 60 to 109 pixels. A second, larger, rectangle was generated with a height that was determined by multiplying the width of the first rectangle by the value (1 + (Level * 0.06)) and a width that was determined by multiplying the height of the first rectangle by the value (1 + (Level * 0.06)). Level was a randomly chosen value from 1 to 8 on each trial (with level 1 being the most difficult and level 8 being the least difficult). Thus, level acted as an objective value for the degree of difference in size between the two rectangles. These formulas also ensured that on each trial one rectangle was wider than it was tall, and the other was taller than it was wide. The larger rectangle was equally likely to be presented in the Tall orientation or the Wide orientation, and this was true across all 8 difficulty levels.

Figure 2.

The testing conditions. The top panel shows a Baseline trial (no decoy; correct choice at right). The middle panel shows a Congruent trial (decoy in the same orientation as the larger rectangle; correct choice at center), and the bottom panel shows an Incongruent trial (decoy in the same orientation as the smaller rectangle; correct choice at left).

Monkeys completed full sessions in which only these control trials with just two rectangle choices were presented until they completed a test session at greater than 70% accuracy (overall selection of the larger rectangle). Once a monkey reached criterion on the basic discrimination, the test phase began in which Decoy trials could be presented.

Test sessions

In test sessions, three conditions were presented to the monkeys in the same experimental task. In the Baseline condition, trials were identical to those presented in the Training phase, and only two choices were presented to the monkey, one rectangle in a Tall orientation, and one in a Wide orientation. In the other two conditions, however, a third rectangle was presented on the screen in the third available location (Figure 2). In the Congruent condition, this rectangle had the same orientation as the larger of the other two rectangles on the screen. In the Incongruent condition, this rectangle had the same orientation as the smaller of the other two rectangles on the screen. In both cases, this third rectangle (the decoy) was smaller than its same-orientation counterpart, with a width and height that was only 75% that of the focal option (i.e., its counterpart of the same orientation). Thus, if the decoy effect occurred, and this item led monkeys to see the counterpart as larger than it actually was, then performance would be enhanced in the Congruent condition because the already largest rectangle on the screen would benefit from being made to look larger by its decoy. However, in the Incongruent condition, the decoy would make the smaller of the other two rectangles look larger than it really was, and perhaps dampen performance relative to the Baseline condition.

During test sessions, Baseline trials were presented with a probability of .70, and Congruent and Incongruent trials each were presented with a probability of .15. Trial level was randomly selected from the range of 1 to 8 for all three conditions. As in the Training Phase, selection of the largest rectangle onscreen led to food reward and presentation of the next trial whereas selection of a rectangle that was not the largest onscreen led to a 20s timeout before presentation of the next trial. All monkeys were scheduled to complete 3,000 trials in the Test Phase. However, because of a program error, three monkeys, Murph, Hank, and Gale only completed 2,000 trials.

Results

During the Test Phase, selection of the decoy stimulus among the three choices was a very rare outcome: Chewie - 0.5% of trials; Gale – 0.2% of trials; Hank - 1.0% of trials; Lou – 0.4% of trials; Luke - 0.4% of trials; Murph – 0.3% of trials; Obi - 0.07% of trials. Thus, the decoy rectangle was not a viable choice option to the monkeys as they rarely (≤ 1.0%) selected this option.

Prior to analyses, the data from the eight difficulty levels were binned into four levels by combining pairs of levels. Thus, Levels 1–2 became the new Level 1, Levels 3–4 became the new Level 2, Levels 5–6 became the new Level 3, and Levels 7–8 became the new Level 4.

Figure 3 shows the mean percentage of trials correct at each binned stimulus level for each condition and each orientation. A within-subjects repeated-measures analysis of variance (ANOVA) was conducted to examine the effect of binned level (1–4), condition (Baseline, Congruent, Incongruent), and correct stimulus orientation (Tall, Wide) on the selection of the larger rectangle. There was a significant main effect of level: F (3, 18) = 48.23, p < .001, ηp² = .89, condition: F (2, 12) = 7.77, p = .007, ηp² = .56, and orientation: F (1, 6) = 10.99, p = .016, ηp² = .65. There was a significant interaction of level and condition, F (6, 36) = 4.11, p = .003, ηp² = .41, indicating that performance diverged more between conditions at the more difficult levels compared to the easier levels. There also was a significant interaction of orientation and condition F (2, 12) = 5.29, p = .023, ηp² = .47. There was not an interaction of orientation and level, F (3, 18) = 2.10, p = .14, ηp² = .26. There was not a three way interaction, F (6, 36) = 0.59, p = .74, ηp² = .09.

Figure 3.

Mean percentage of trials correct at each binned stimulus level (1–4, with Level 1 being the objectively most difficult discriminations and Level 4 being the objectively least difficult) for each condition (Baseline, Congruent, and Incongruent) and each orientation (Tall and Wide) in Experiment 1. Errors bars represent 95% confidence intervals.

To examine the interaction between level and condition, we compared each condition to each other (Baseline vs. Congruent, Baseline vs. Incongruent, and Congruent vs. Incongruent) at each level (1, 2, 3, 4) using a paired-samples t-test. A Bonferroni adjusted alpha level of 0.017 was used per test (.05/3). Performance did not differ significantly for any conditions at Level 1 (Baseline vs. Congruent: t(6) = −1.55, p = .17; Baseline vs. Incongruent: t(6) = 2.52, p = .04; Congruent vs. Incongruent: t(6) = 2.42, p = .05). For Level 2, performance was significantly higher in the Congruent condition than the Incongruent condition (t(6) = 3.74, p = .01), but did not differ between Baseline and Incongruent (t(6) = 3.22, p = .018) nor Baseline and Congruent (t(6) = −0.76, p = .48). Performance did not differ significantly for any conditions at Level 3 (Baseline vs. Congruent: t(6) = 1.02, p = .35; Baseline vs. Incongruent: t(6) = .99, p = .36; Congruent vs. Incongruent: t(6) = −0.25, p = .81). For Level 4, performance was significantly higher in the Congruent condition than the Baseline condition (t(6) = 3.33, p = .016), but did not differ between Baseline and Incongruent conditions (t(6) = 2.59, p = .04) nor Congruent and Incongruent conditions (t(6) = −1.13, p = .30).

To explore the interaction of orientation and condition, we compared the effect of orientation in each condition collapsing across level using a paired samples t-test. A Bonferroni adjusted alpha level of 0.017 was used per test (.05/3). Performance was significantly higher in the Tall-Baseline condition versus the Wide-Baseline condition (t(6) = 4.05, p = .007), indicating that when the correct stimulus was in the tall orientation monkeys performed better. Performance was significantly higher in the Tall-Incongruent condition versus the Wide-Incongruent condition (t(6) = 3.45, p = .014), also confirming that performance was better when the correct stimulus was in a tall orientation. Performance did not differ in the Tall-Congruent versus Wide-Congruent conditions (t(6) = −1.21, p = .274).

We also explored whether response times to make a choice within a trial varied between the conditions, and we examined this for Correct and Incorrect trials (Figure 4), under the assumption that such an effect of condition was more likely for the correctly completed trials where animals ultimately made the correct response. Thus, we conducted a two-way repeated-measures ANOVA with outcome and condition as variables. First, we removed any response times that exceeded 10 seconds as we classified these as extreme outliers (they comprised less than 1.3% of trials for all of the monkeys), and then calculated the mean response time for each monkey in each condition for Correct and Incorrect trials. There was not a significant effect of condition on response times: F (2, 12) = .52, p = .61, ηp² = .08, nor of outcome on response times: F (1, 6) = 5.6, p = .06, ηp² = .48. However, there was a significant interaction between condition and outcome, F (2, 12) = 6.33, p = .01, ηp² = .51.

Figure 4.

Mean response times for each condition for correctly completed and incorrectly completed trials in Experiment 1. Error bars show standard errors of the mean.

To explore this interaction, we conducted separate repeated-measures ANOVAs for Correct and Incorrect trials comparing response times in each condition. We found a significant effect of condition for Correct trials, F (2, 12) = 60.3, p < .001, ηp² = .91, but not for Incorrect trials, F (2, 12) = 2.2, p = .15, ηp² = .27. We then compared each condition to each another for Correct trials using paired samples t-tests and a Bonferroni corrected alpha level of .016 given the three comparisons. Response times were significantly faster for Congruent-Correct trials than for Baseline-Correct trials (t(6) = 3.58, p = .01) and significantly faster for Congruent-Correct trials than for Incongruent-Correct trials (t(6) = −8.26, p < .001). Response times were significantly faster also for Baseline-Correct trials than for Incongruent-Correct trials (t(6) = −11.64, p < .001). These results indicate that the monkeys responded fastest in the Congruent trials and the slowest in Incongruent trials when they made a correct response, a pattern consistent with the idea that congruent decoy stimuli aided and speeded choice behavior whereas incongruent decoy stimuli hurt and slowed choice behavior.

In consideration of individual differences among the monkeys, we collapsed across orientation and analyzed the performance (percentage correct) of each monkey in each condition (Baseline, Congruent, and Incongruent) across all levels. Further, we analyzed whether each condition differed from the others for each monkey at each level using a Chi Square or Fischer’s Exact Test on frequency data for correct/incorrect choices. These results are presented in Table 1. Although not every condition varied from every other condition at each level, when a significant difference occurred, it supported our hypothesis that Congruent performance would be higher than Baseline performance, that Incongruent performance would be lower than Baseline performance, or that Congruent performance would be higher than Incongruent performance.

Table 1.

Individual results from Experiment 1, including the percentage correct for each animal in each condition (Baseline, Congruent, and Incongruent) and for each level.

		Level 1	Level 2	Level 3	Level 4
	Baseline	83.17	95.97	96.60	97.76
Chewie	Congruent	89.22	96.54	93.94	94.76
	Incongruent	85.96	90.46*	96.07	93.87
	Baseline	77.08	95.77	98.85	98.17
Gale	Congruent	92.05*	96.30	97.06	90.63*
	Incongruent	65.34^	89.74*	94.00*	91.46*
	Baseline	68.22	82.07	87.79	94.61
Hank	Congruent	61.54	76.52	92.03	88.47
	Incongruent	69.27	76.97	85.82	94.82
	Baseline	80.92	93.36	96.97	99.40
Lou	Congruent	86.46	94.00	90.89*	95.21*
	Incongruent	74.39^	91.66	97.56^	96.23*
	Baseline	81.03	94.56	98.15	98.13
Luke	Congruent	82.02	97.88	96.67	99.14
	Incongruent	77.25	90.29^	100.00	96.69
	Baseline	74.57	90.75	95.46	98.22
Murph	Congruent	76.79	95.45	94.60	95.54
	Incongruent	61.21	90.28	94.54	97.14
	Baseline	88.43	97.58	98.84	99.60
Obi	Congruent	92.01	100.00	99.09	97.50
	Incongruent	75.17*^	98.15	99.06	99.19

Note.

The asterisk indicates p < .05 for the comparison of Congruent to Baseline condition or the Incongruent to Baseline condition.

The caret indicates p < .05 for the comparison of Congruent to Incongruent conditions.

Discussion

The decoy stimulus moderately impacted perceptual choice behavior among rhesus monkeys in the current experiment, even when the decoy stimulus itself did not represent a viable choice option. As expected, the monkeys were highly successful in choosing the larger of two rectangular stimuli within the current experiment, with performance increasing as the true difference between the two stimuli increased. Performance was impacted by the presence of a decoy stimulus, either serving to increase (Congruent condition) or decrease (Incongruent condition) the selection of the larger rectangle relative to Baseline levels. This pattern was more pronounced for the most difficult levels and for the Wide orientation. The individual results also indicated that all cases of a significant difference between any two conditions occurred in the direction indicative of a decoy effect.

The decoy stimuli also affected response time. When the decoy was congruent, monkeys made correct choices even faster than when no decoy stimuli were present. However, the reverse was true for incongruent decoys. When those decoys were present, even when monkeys made the correct choice, they took longer to do so. These results also support the idea that decoys affect how monkeys view and respond to choices that include a non-viable decoy option.

In exploring the interaction between condition and orientation, we confirmed that performance levels were higher when the correct choice was in the Tall orientation versus the Wide orientation in the Baseline and Incongruent conditions. The rhesus monkeys chose more accurately when they were supposed to choose the vertical rather than the horizontal stimulus. In the Incongruent condition, the decoy effect was less pronounced for vertical stimuli than horizontal stimuli. Thus, the monkeys were less likely to erroneously choose the horizontal foil stimulus even when a horizontal decoy was present. These results support the Horizontal-Vertical illusion in which a vertically-oriented line appears longer than an identical, but horizontally-oriented line, which has been documented in several nonhuman primate species (Cercopithecinae and Cebus: Dominguez, 1954; Harris, 1968). The current results provide support that rhesus monkeys may also overestimate vertical dimensions relative to horizontal ones, serving to insulate the impact of horizontally-oriented decoy stimuli.

In general, the monkeys performed at very high levels, and this created a potential ceiling effect that might have masked a stronger decoy effect. This was confirmed by the level by condition interaction. Thus, we conducted a second experiment in which we restricted the trial levels to the more difficult half of the continuum used in Experiment 1 to better assess the prevalence of perceptual decoy effects in these monkeys’ discrimination performances.

Experiment 2

Methods

Subjects and Apparatus

These were the same as in Experiment 1.

Design and Procedure

There was no training phase in this experiment. All monkeys began in the Test Phase, which was functionally identical to Experiment 1. The only difference pertained to the difficulty levels of trials that were increased across the full range of eight levels. Now, the second, larger, rectangle was generated with a height that was determined by multiplying the width of the first rectangle by the value (1 + (Level * 0.015)) and a width that was determined by multiplying the height of the first rectangle by the value (1 + (Level * 0.015)). Note that the multiplier of 0.015 was ¼ the multiplier from Experiment 1 (0.06). The same three conditions were presented in the same proportions as in Experiment 1, and each monkey completed 3,000 trials.

Results

Again, selection of the decoy stimulus among the three choices was a very rare outcome (≤ 1.0%): Chewie - 0.4% of trials; Gale – 0.2% of trials; Hank - 1.0% of trials; Lou – 0.07% of trials; Luke - 0.03% of trials; Murph – 0.9% of trials; Obi - 0.2% of trials.

The data again were binned from eight difficulty levels into four levels prior to analysis. Figure 5 shows the mean percentage of trials correct at each binned stimulus level for each condition and each orientation. We conducted a within-subjects ANOVA to examine the effect of level (1–4), condition (Baseline, Congruent, Incongruent), and orientation (Tall, Wide) on the selection of the larger rectangle in Experiment 2. There was a significant main effect of level, F (3, 18) = 108.93, p < .001, ηp² = .95, and condition, F (2, 12) = 14.58, p = .001, ηp² = .71. However, there was no longer an effect of orientation, F (1, 6) = .001, p = .973, ηp² < .01. There were no significant interactions.

Figure 5.

Mean percentage of trials correct at each binned stimulus level for each condition (Baseline, Congruent, and Incongruent) and each orientation (Tall and Wide) in Experiment 2.Errors bars represent 95% confidence intervals.

As in Experiment 1, we analyzed response time data to determine if latency to make a choice varied as a function of condition with a two-way repeated-measures ANOVA comparing average response times in the Baseline, Congruent, and Incongruent conditions for Correct and Incorrect trials. Again, we removed any response times that exceeded 10 seconds as we classified those as extreme outliers. Such outliers were very rare, comprising less than 1.2% of trials for all of the monkeys. There was a significant effect of condition on response times: F (2, 12) = 5.72, p = .02, ηp² = .49, and of outcome on response times: F (1, 6) = 57.72, p < .001 ηp² = .91. Again, there was a significant interaction between condition and outcome, F (2, 12) = 62.07, p < .001, ηp² = .91.

As in Experiment 1, we next conducted separate repeated-measures ANOVAs for Correct and Incorrect trials comparing response times in each condition (Figure 6). We found a significant effect of condition for Correct trials, F (2, 12) = 76.09, p < .001, ηp² = .93, and for Incorrect trials, F (2, 12) = 27.99, p < .001, ηp² = .82. We then compared each condition to each other using paired samples t-tests for Correct and Incorrect trials (using a Bonferroni corrected p value of .017). Response times were significantly faster for Congruent-Correct trials than for Baseline-Correct trials (t(6) = 5.71, p = .001) and significantly faster for Congruent-Correct trials than for Incongruent-Correct trials (t(6) = −10.42, p < .001). Response times were significantly faster for Baseline-Correct trials than for Incongruent-Correct trials (t(6) = −7.52, p < .001). For Incorrect trials, response times were significantly slower for Congruent-Incorrect trials than for Incongruent-Incorrect trials (t(6) = 5.61, p = .001) and significantly slower for Congruent-Incorrect trials than for Baseline-Incorrect trials (t(6) = −9.0, p < .001). As in Experiment 1, these results indicate that the monkeys responded fastest in the Congruent trials and the slowest in Incongruent trials when they made a correct response. However, monkeys responded slowest in the Congruent trials when they made an incorrect response.

Figure 6.

Mean response times for each condition for correctly completed and incorrectly completed trials in Experiment 2. Error bars show standard errors of the mean.

Additionally, we analyzed the individual data for all monkeys in Experiment 2 in the same manner as in Experiment 1, and these results are presented in Table 2. As in Experiment 1, although not every condition varied from every other condition at each level for each monkey, whenever a significant difference occurred, it supported our hypothesis that Congruent performance would be higher than Baseline performance, that Incongruent performance would be lower than Baseline performance, and that Congruent performance would be higher than Incongruent performance.

Table 2.

Individual results from Experiment 2, including the percentage correct for each animal in each condition (Baseline, Congruent, and Incongruent) and for each level.

		Level 1	Level 2	Level 3	Level 4
	Baseline	61.63	74.37	85.70	90.53
Chewie	Congruent	64.00	78.80	85.10	95.99
	Incongruent	63.11	68.16	76.99*	88.11^
	Baseline	58.11	70.39	77.92	85.27
Gale	Congruent	75.04*	75.91	89.25*	90.72
	Incongruent	59.16^	57.84*^	69.75^	80.70^
	Baseline	55.71	59.89	60.55	69.56
Hank	Congruent	53.70	63.35	66.17	78.16
	Incongruent	50.95	50.81	59.85	64.73^
	Baseline	60.44	73.92	81.64	88.52
Lou	Congruent	72.01	80.00	89.09	94.54
	Incongruent	49.59*^	56.30*^	70.21*^	82.45^*
	Baseline	59.05	73.04	80.86	87.43
Luke	Congruent	63.47	69.46	83.18	91.13
	Incongruent	58.57	80.77	81.66	90.43
	Baseline	55.62	64.54	74.42	80.12
Murph	Congruent	56.76	66.32	81.48	79.86
	Incongruent	42.42*^	55.62	68.34	70.75*
	Baseline	64.23	80.00	93.13	93.67
Obi	Congruent	70.99	84.61	89.92	96.32
	Incongruent	55.82^	78.02	83.92*	88.91

Note.

The asterisk indicates p < .05 for the comparison of Congruent to Baseline condition or the Incongruent to Baseline condition.

The caret indicates p < .05 for the comparison of Congruent to Incongruent conditions.

Discussion

As in Experiment 1, rhesus monkeys’ perceptual judgments about rectangle size were impacted by the presence of a decoy stimulus. However, the introduction of more difficult discriminations (via levels that spanned only half of the original continuum) eliminated the interaction between condition and level. This interaction in Experiment 1 was likely due to high levels of performance in the easier levels within Experiment 1, resulting in a ceiling effect that masked a more pronounced decoy effect. In Experiment 2, the decoy stimulus impacted performance more consistently across levels, either serving to increase (Congruent condition) or decrease (Incongruent condition) performance relative to the Baseline condition (see Figure 5). Additionally, there was no longer an effect of orientation, indicating that the monkeys were equally proficient across Tall and Wide trials. This finding was likely due to experience in performing the task and receiving feedback about their responses. Thus, experience diminished the general higher performance when tall stimuli were the correct choice versus wide stimuli, but the effect of level remained (as was expected given that this was the objective factor for trial difficulty), and the decoy effect remained as evidenced by the continued main effect of condition. Thus, across the two experiments, the decoy effect was present, although there were individual differences in the degree to which this effect was evident.

For correctly completed trials, we replicated the response time effect of Experiment 1. However, we also found a different response time pattern for incorrectly completed trials, suggesting that response times when monkeys were incorrect might have reflected some of the response competition they felt when faced with a decoy stimulus. We discuss this in more detail in the next section.

General Discussion

The current work provides evidence for the asymmetric dominance effect (also known as the decoy effect) within the perceptual domain among rhesus monkeys. Monkeys’ perceptual discrimination in the current size-judgment task was impacted by the presence of a third, smaller decoy stimulus. A smaller, decoy rectangle enhanced the selection of the original rectangle that it was most similar to in terms of its orientation, serving to increase performance relative to baseline if the decoy was in the same orientation as the largest rectangle. Alternatively, this effect disrupted performance relative to baseline if the same-orientation rectangle was the smaller of the two original options. These findings complement similar work documenting the asymmetric dominance effect using traditional preference testing in nonhuman species (Bateson, 2002; Bateson et al., 2002; Hurly & Oseen, 1999; Scarpi, 2011; Shafir et al., 2002; Schuck-Paim et al., 2004). Furthermore, these results are consistent with human evidence of the decoy effect within the perceptual domain (Choplin & Hummel, 2005; Trueblood et al., 2013; Tversky, 1972).

Evidence for a perceptual decoy effect in a monkey species supports the notion that context-dependent choice behavior extends beyond higher-order decision-making tasks and emerges during more basic perceptual processing (Trueblood et al., 2013). The decoy stimulus itself was not a viable choice option as it always was the smallest rectangle within the set, and was rarely selected in any experiment. Thus, the decoy impacted choice behavior by creating a contrasting option to which the original options were compared. These comparative evaluations are dependent upon the availability and properties of alternatives within a choice set (Bateson et al., 2002). Comparative evaluations may serve to decrease the complexity of decisions by allowing the decision-maker to assign a relative value based on one dimension, rather than constructing an absolute value for each option within a set (Ariely & Wallsten, 1995). Bateson and colleagues (2002) discussed the adaptive value of comparative mechanisms in terms of potentially lower computational requirements in relation to more costly absolute evaluations (see Tversky, 1969).

This notion is supported by the difference in response time data in the current study. On correctly completed trials, the monkeys responded fastest when a congruent (or helpful) decoy was present, but incurred a time-cost when an incongruent (or hurtful) decoy was present, even though they still made the correct choice. The decoy stimulus created a contrasting option to the stimulus that it was most similar to; this either quickly led to the correct answer (Congruent trials) or instead slowed response time in making the right choice if the decoy matched the incorrect answer (Incongruent trials). In Experiment 2, an additional feature of the response times emerged. When monkeys ultimately made a mistake (and chose the smaller item of the non-decoy pair), they took the longest to make this error when the decoy was congruent with the correct choice. In other words, the monkeys likely noticed the decoy that would have helped performance because it slowed their eventual choice of the wrong answer, the one that was dissimilar to the decoy. Future research that varies the physical proximity or degree of similarity between the decoy and the focal option would be of interest to understanding the limits of this type of contrast effect in monkeys, and the resulting impact on response time.

These findings open future avenues for testing the impacts of decoy stimuli on choice behavior in nonverbal species and among pre-verbal children. Perceptual tasks of this nature require little training or verbal instructions that may impact decision-making strategies, making this paradigm accessible to many species and populations. They easily contrast performance against well-established psychophysical discriminations from baseline conditions, and allow for parametric variations of the stimuli, including the decoys. Parametric variations would be useful if one were interested in assessing the ideal decoy characteristics for enhancing or disrupting discrimination performance. The use of basic perceptual discrimination tasks allow for spontaneous and early-emerging effects, such as the decoy effect, to highlight the mechanisms that underlie perceptual responding and context-dependent choice behavior. That monkeys showed these effects indicates that they are evolutionarily shared and not language-dependent, and perhaps may be quite widespread among animals. If this is true, it would suggest that perhaps some decisional biases likely have roots in foundational perceptual processes with long evolutionary histories.