Perceptual Decision-Making Regarding Phylogenetically Salient Stimuli

Abstract

Phylogenetically salient stimuli such as spiders are commonly reported threats in the general population and the most common object of specific phobias in clinical populations. Several theories have hypothesized that our perceptual systems prioritize such stimuli in an “automatic” or “bottom-up” manner due to their evolutionary salience. However, empirical research on the perceptual processing of these stimuli as well as the influence of “top-down” goal-driven and bottom-up stimulus-driven factors is lacking. Here, we used perceptual psychophysics to determine absolute perceptual thresholds for the detection of spider and crab images. Subsequently, participants used spider and crab cues (that imposed a top-down perceptual set) to detect spiders and crab images presented at their predetermined perceptual threshold in a two-alternative forced-choice perceptual decision-making task. While spiders were detected at lower perceptual thresholds than crabs, they were not immune to top-down influence. Indeed, compared to top-down crab cues, spider cues improved the speed and accuracy of detection of spiders vs crabs. Using a hierarchical drift diffusion model, we found that spider cues biased decision-making not only by shifting the starting point of evidence accumulation towards the spider decision, but also by increasing the efficiency with which sensory evidence accumulated, more so for spider than crab perceptual decisions. Overall, these findings provide evidence for the perceptual prioritization of phylogenetically salient stimuli and highlight the computational mechanisms by which this prioritization is facilitated by bottom-up and top-down factors.

Supplementary Information

The online version contains supplementary material available at 10.1007/s42761-024-00271-z.

Hundreds of stimuli compete for representation in our perceptual systems. Researchers have argued that threat-related stimuli have an edge in these competitive interactions, receiving prioritized processing (Brosch et al., 2010), because their early detection is adaptive for our survival (Öhman et al., 2001; Seligman, 1971). From an evolutionary perspective, facilitated threat processing has been explained by the activation of the brain’s “fear module,” an evolutionarily developed subcortical neural circuit, primarily based in the amygdala, and automatically activated by evolutionarily relevant phylogenetic stimuli (for review see, de Gelder et al., 2011; Öhman & Mineka, 2001).

A particularly common example of such stimuli are spiders which are strongly disliked by more than a third of children (Muris et al., 1997) and adults (Eaton et al., 2018; Matchett & Davey, 1991) and the object of the most reported specific phobia (Fredrikson et al., 1996; Nesse, 1990). Spider stimuli are rapidly detected in a matrix of benign distractors by adults and children (LoBue, 2010; Öhman et al., 2001), and they are detected even when presented subliminally (Öhman & Soares, 1994) or irrelevant to task goals (New & German, 2015). They elicit greater fixations (Devue et al., 2011; LoBue et al., 2014), heart rate changes (Flykt et al., 2012), sensory cortical (Kopp & Altmann, 2005), and amygdala activity (Mobbs et al., 2010) than neutral stimuli, an effect that is pronounced in individuals with greater fear of spiders (Böhnlein et al., 2021; Michalowski et al., 2015; Siminski et al., 2021). Finally, infants with little exposure to spiders look at them preferentially (Rakison & Derringer, 2008) and with increased pupil dilation (Hoehl et al., 2017). Overall, behavioral and neurophysiological evidence across the lifespan has led to the view that stimuli such as spiders and snakes are perceived in a “bottom-up” or automatic manner due to the phylogenetic salience of their physical features (LoBue et al., 2010; Öhman et al., 2001).

However, making perceptual decisions about stimuli around us involves a complex process of integrating bottom-up sensory evidence arising from the stimulus with “top-down” perceptual sets, attention, and expectations derived from the context or our prior knowledge and experiences (Summerfield & de Lange, 2014). This idea is echoed by the predictive processing framework which proposes that perception involves generating predictions regarding incoming sensory information based on prior knowledge and comparing the predictions with the actual sensory input (Rao & Ballard, 1999; Spratling, 2016). In the case of emotional stimuli, evidence for purely automatic bottom-up processing is weak both behaviorally and neurally (Quinlan, 2013), and top-down factors influence the detection of evolutionarily salient stimuli (LoBue, 2014; Vromen et al., 2015, 2016). For example, learning negative information about one of two novel animals speeds its relative detection (Field, 2006a, b), fear-related labels and experimentally induced mood speed the detection of spiders and snakes (LoBue, 2014), and reducing expectations of seeing spiders reduces attentional bias towards them (Abado, Aue et al., et al., 2020, Abado, Sagi et al., 2020). Overall, such evidence has led researchers to propose that rather than looking for evidence of purely bottom-up or top-down processing, research should focus on the relative contributions of these processes to prioritized perception of emotional stimuli as well as the underlying mechanisms (Pessoa, 2009; Pessoa & Adolphs, 2010; Sussman et al., 2016).

While the behavioral and neural literature reviewed above shows facilitated processing of phylogenetically salient threats via bottom-up or top-down pathways, most of the studies do not directly examine perceptual processes. As a result, it remains unclear whether facilitation occurs simply by influencing attention and response bias or whether it involves fundamental processes of perception. Perception can be studied in several ways including estimating absolute perceptual threshold and measuring perceptual decision-making between two alternatives. Absolute perceptual thresholds can be measured using a variety of psychophysical methods which are designed to estimate the stimulus intensity (i.e., “stimulus energy”) at which a specific stimulus type can be identified accurately a certain percentage of the time (Colman, 2009; Leek, 2001; Pirenne, 1943; Spielman et al., 2017). Adaptive staircasing functions are commonly used for estimating absolute perceptual thresholds because they allow for the modulation and estimation of such thresholds based on participants’ trial-by-trial behavioral responses (e.g., correct or incorrect identification of a specific stimulus type). Another way that perceptual processes are measured is via the classification of sensory information or perceptual decision-making. Perceptual decision-making is the basic process by which sensory evidence is collected and combined to form a decision with the goal of identifying a stimulus and subsequently shaping behavior (Gold & Shadlen, 2007; Heekeren et al., 2008). It involves a “sensory system” responsible for gathering sensory evidence, a “decision system” that converts sensory evidence to a decision variable, which upon passing a decision threshold forms a decision, and a “motor system” that initiates an action based on the perceptual decision. The process by which a decision maker samples the noisy sensory evidence arising from stimuli and/or uses top-down information to arrive at a decision can be formalized via models such as signal detection theory (SDT; Green & Swets, 1966), as well as sequential sampling models, such as drift diffusion model (DDM; Gold & Shadlen, 2007; Ratcliff & McKoon, 2008). In the SDT framework, performance can be characterized by sensitivity (d′) to the presence or absence of phylogenetically salient stimulus-related signal and criterion shift (c bias) which can be biased, i.e., liberal or conservative conservative for deciding whether this signal is present or absent.

While SDT provides a snapshot of sensory signals, in real life, the integration of sensory evidence for competing decisions is accumulated over time. Within the DDM framework, the influence of top-down or bottom-up factors on perceptual decision-making can be examined as a (1) shift in the starting point (z) of the evidence accumulation process (Bogacz et al., 2006; Diederich & Busemeyer, 2006; Ratcliff, 2002; Voss et al., 2004; Wagenmakers et al., 2008) and/or (2) increase in the rate of evidence accumulation (drift rate or v) (Ashby, 1983; Diederich & Busemeyer, 2006; Ratcliff, 2002). The starting point of evidence accumulation can index top-down influences of prior knowledge, beliefs, and expectations (Kloosterman et al., 2019). Hence, in a situation where an observer is deciding between two alternatives, top-down factors such as expectation or value can bias decision-making towards the expected or valued alternative by shifting the starting point of evidence accumulation (away from the mid-point between the two decisions) closer to the corresponding decision boundary (Forstmann et al., 2016; Mulder et al., 2012). This allows the observer to reach the decision corresponding to the expected or valued stimulus with less evidence. On the other hand, the drift rate indexes the rate at which noisy sensory evidence accumulates over time until a boundary is reached (Ratcliff et al., 2004). It is influenced not only by the quality of the stimulus (e.g., a clearer or more salient stimulus will have a higher drift rate) but also top-down factors such as attentional or perceptual sets for the stimulus (e.g., more specific perceptual templates will yield larger drift rates) (Dunovan et al., 2014; Ratcliff, 2002).

In the present study, we aimed to examine absolute perceptual thresholds for the detection of phylogenetically salient stimuli and the impact of bottom-up and top-down factors on perceptual decision-making regarding these stimuli. In order to do that, we compared the detection of spider stimuli with neutral crab stimuli, which are perceptually similar with a central body and many legs, but not considered phylogenetically threatening. Firstly, we used perceptual psychophysics to determine each participant’s absolute threshold for detection of spider and crab stimuli. We hypothesized that given their phylogenetic salience, spider stimuli would be detected at lower thresholds (i.e., lower contrasts that make the image harder to identify) than crab stimuli. Next, participants completed a 2-alternative forced choice decision-making task, in which they used spider- or crab-related top-down cues to discriminate between spider and crab images presented at their predetermined perceptual threshold. A unique aspect of our perceptual discrimination task was that it was made challenging by degrading the spider and crab images to subject-specific perceptual thresholds, encouraging participants’ use of cue-related information in perceptual decision-making. This allowed us to assess the top-down cue-related effects on perceptual decision-making. Furthermore, spiders and crabs were equally likely to follow each cue, allowing us to attribute differences in behavior to the cues. Therefore, these cues indicate whether spiders or crabs are relevant, influencing attention. However, they do not provide information regarding the probability of upcoming stimuli, and thus do not influence expectations regarding upcoming spiders and crabs (e.g., Summerfield & de Lange, 2014). Finally, spiders and crabs were matched for luminance and spatial frequency, minimizing the contribution of low-level confounds unrelated to stimulus saliency characteristics.

We hypothesized that spiders would be detected faster and more accurately than crabs due to their salience, but that this effect would be enhanced following spider vs crab cues, demonstrating the interaction of top-down and bottom-up effects on perceptual decision-making. Next, we compared the influence of spider and crab cues on SDT and DDM parameters. In line with research on basic perception (Lu and Dosher, 2008, Smith and Ratcliff, 2009, Wyart et al., 2012; Swets et al., 1961), we hypothesized that spider cues directing top-down attention towards spiders will enhance sensitivity to spider signals, thereby increasing true positives and d′ but not influence decision criterion which tends to be impacted by expectation. Further, in line with research from basic perception (Cravo et al., 2011; Dunovan et al., 2014; Mulder et al., 2012), we hypothesized that top-down attention to spiders will enhance drift rate or the efficacy of extracting spider-related signals from the sensory evidence and shift the starting point of evidence accumulation closer to a spider decision boundary, allowing one to reach that decision with less evidence. Finally, we conducted a second study (Study 2) with the same procedures to replicate the findings from the original study (Study 1).

Method

Participants

Forty-four participants (26 women; 20.86 ± 6.46 years) were recruited from the Stony Brook University Psychology Department subject pool. Within our sample, 45.5% identified as Asian, 29.5% as White, 9% as Black or African American, 9% as Hispanic or Latino, and 7% as more than one race. Each subject provided informed consent prior to participation in the study, which was approved by Stony Brook University’s Institutional Review Board. Outliers were defined as values ± 2.5 standard deviations (SD) from the mean for both accuracy and perceptual sensitivity, with no outliers detected and thus none excluded from the analysis. On the other hand, data from two subjects were excluded from analyses due to technical malfunction, resulting in a final sample of 42 participants (24 women; 20.88 ± 6.61 years). To determine the appropriate sample size for our study, we conducted power analyses using G*Power (Faul et al., 2009). Based on our previous research on the impact of fear cues versus neutral cues on perceptual sensitivity (Sussman et al., 2016; Glasgow et al., 2022; Ozturk et al., 2024), the analyses indicated that a sample of 34 subjects would be necessary to detect similar significant task-related effects. The findings of this sample were replicated in a second study (n = 53), following the exact same procedure. Please see the Supplementary Methods section for detailed information on the participants of Study 2.

Stimuli

Stimuli consisted of 16 spider (spider stimulus) and 16 crab (crab stimulus) images—modified to grayscale (800 × 800 pixels), against a plain, gray background. These images were equalized for luminance and spatial frequency using the SHINE (Spectrum, Histogram, and Intensity Normalization and Equalization) toolbox for Matlab (Willenbockel et al., 2010). This toolbox was designed to minimize the influence of confounds linked to low-level visual properties of the stimuli. Image processing equating low-level image properties across stimulus sets has been used in studies examining the effects of top-down processes on perception (Fiset et al., 2008; Williams et al., 2009). Each stimulus presentation was followed by one of four perceptual masks, created by randomly reorganized 100-pixel squares, consisting of segments of several superimposed spider and crab images that had been processed using SHINE. Therefore, these masks had the same low-level image properties as the target stimuli.

Behavioral Tasks

Threshold Task

Each participant’s absolute threshold for perception (75% correct) was determined using a 2-alternative forced-choice perceptual discrimination task (Summerfield et al., 2006). The task consisted of 16 blocks of 16 trials with a total of 128 spider stimulus trials and 128 crab stimulus trials, which were created and presented via Psychopy software (Peirce, 2007). Each trial began with a fixation cross in the center of the screen (2–3 s), followed by a degraded spider or crab image (100 ms), followed by a mask (300 ms). Participants identified the stimuli by pressing one of two adjacent buttons corresponding to what they perceived the stimuli to be. Contrast was manipulated on a scale ranging from 1 to 0, such that 1 corresponded to no contrast manipulation and 0 corresponded to complete removal of contrast, making the image a gray square. Spider and crab images were initially presented at a reduced level contrast at 1, making images visible, but not easy to see. The level of contrast on subsequent trials was determined by two adaptive staircases (Watson & Pelli, 1983), one for spider stimuli and one for crab stimuli, which allowed determinations of the absolute perceptual threshold for spider and crab stimuli for each participant (Fig. 1A). Thresholds were measured and adjusted using a Weibull psychometric function, resulting in an approximation of the Bayesian estimate of the perceptual threshold of spider stimuli and crab stimuli, such that an incorrect answer led to an easier-to-see image (image presented at a higher contrast level) on the next trial of the same type, while a correct answer led to a harder-to-see image (image presented at a lower contrast level) on the next trial of the same type (Watson & Pelli, 1983), so that participants detected stimuli at 75% accuracy (Summerfield et al., 2006).

Fig. 1.

A Adaptive staircasing to determine absolute threshold: Thresholds were measured via two adaptive staircases for spider and crab stimuli separately. An incorrect answer led to an image presented at a higher contrast level on the next trial of the same type, while a correct answer led to an image presented at a lower contrast level on the next trial of the same type until contrast corresponding to 75% accuracy was determined. B Cued-perceptual discrimination task timeline: Participants viewed a spider cue which was the letter “S” (lower panel) and indicated that they would decide whether the subsequent stimulus was spider or not. S was followed by a fixation cross followed by a perceptually degraded spider or crab stimuli. The stimuli were followed by a perceptual mask after which the participant indicated their decision with a button press. In a similar timeline, participants viewed a crab cue (C) which was the letter “C” (upper panel) which indicated that they will be deciding whether the subsequent stimulus was crab or not

Perceptual Cued Discrimination Task

To determine the impact of cue salience on subsequent perceptual decision-making, participants viewed the spider and crab stimuli from the threshold task in the same manner, but with three main differences (Fig. 1B). First, spider and crab stimuli were perceptually degraded by lowering the contrast and presented at one of 8 contrast levels ranging from 6% less to 8% more than the participant’s previously determined perceptual threshold (Adini et al., 2004). For example, if a participant’s threshold for spiders was determined to be 0.1, they were subsequently shown spider stimuli ranging from 0.108 to .94. Stimuli were presented at several contrast levels with an equal number of images at each level to prevent improved perceptual performance due to practice effects (Adini et al., 2004). Second, prior to each spider or crab stimulus, the letter “S” (spider cue), or “C” (crab cue) appeared for 1 s. Third, following a spider cue, participants were asked to decide whether the subsequently presented images were spiders or not, by pressing the “yes” button when they decided the stimulus was a spider and the “no” button when they did not. Similarly, on trials starting with a crab cue, participants were asked to decide whether the subsequently presented images were crabs or not, by pressing the “yes” button when they perceived it as a crab and the “no” button when they did not. Furthermore, spiders and crabs were equally likely to follow each cue, allowing us to attribute differences in behavior to the cues. Therefore, these cues indicate whether spiders or crabs are relevant, influencing attention. However, they do not provide information regarding the probability of upcoming stimuli, and thus, do not influence expectations regarding upcoming spiders and crabs (e.g., Summerfield & de Lange, 2014). Finally, spiders and crabs were matched for luminance and spatial frequency, minimizing the contribution of low-level confounds unrelated to stimulus saliency characteristics. Trials were presented in 8 blocks of 32 trials each, resulting in 128 spider cue and 128 crab cue trials. The same spider (16) and crab stimuli (16) were presented following spider cues and crab cues (each image was shown 8 times). The timeline of the task and cue image combinations are shown in Fig. 1B. Accuracy and reaction time (RT) were recorded. As an equal number of spider and crab trials were presented after each cue, the cue provided no information regarding the likelihood of seeing an upcoming spider or crab stimulus, and simply indicated the upcoming perceptual decision.

Data Analyses

Signal Detection Theory

We investigated perceptual decision-making by utilizing the two measures based on signal detection theory (SDT) with the equal variance Gaussian assumption: perceptual sensitivity and criterion shift (Green & Swets, 1966). In the present study, perceptual sensitivity (d′) is an index of a participant’s ability to distinguish between the spider and crab stimuli under each cue. Hence, a higher d′ value for a spider or crab cue suggests better discrimination between subsequently presented spider and crab stimuli. The second SDT measure, criterion shift, quantifies the position of the decision criterion, which can indicate a more liberal or conservative decision bias. For example, a liberal bias (a more negative value) of endorsing the “yes” decision would indicate a higher tendency of making a spider decision following a spider cue or making a crab decision following a crab cue, whereas a conservative bias (a more positive value) would be the opposite.

We calculated d′ as the difference between the z-scores (inverse of a normal distribution) of the hit rate and false alarm rate for the spider and crab cues separately (d′ = z (Hit Rate) – z (False Alarm Rate)). Similarly, c was calculated as the average of the negative z-scores of the hit and false alarm rates (c =  − [z(Hit Rate) + z(False Alarm Rate)]/2), also separately for the spider and crab cues (Stanislaw & Todorov, 1999).

Hierarchical Drift Diffusion Modeling (HDDM)

DDM (Ratcliff, 1978; Ratcliff et al., 2016; Wiecki et al., 2013) was fitted to the choice and RT data, in order to investigate the influence of spider and crab cues on decision-making components. DDM represents perceptual evidence that is stochastically sampled over a short period of time by using a single decision variable. This accumulation of evidence starts from a starting point denoted as z, which lies between two decision boundaries, representing different choice alternatives such as spider and crab decisions. The distance between the two decision boundaries is denoted as a. The rate at which evidence accumulates is indexed by the drift rate, v, while the non-decision component, t, represents sensory encoding and motor execution. The parameters (z, a, v, and t) can be determined by analyzing the shape of the RT distributions for the spider and crab decisions.

Previous research has shown that cues can affect decision-making by influencing the z and v (Dunovan & Wheeler, 2018; Dunovan et al., 2014). Therefore, our main objective was to investigate whether spider and crab cues would impact z and v parameters towards the respective decision boundaries. In our task, we assumed that when participants responded “yes” to a spider stimulus after a spider cue or responded “no” to a crab stimulus after a crab cue, their evidence accumulation process ended at the spider decision boundary, and vice versa. We expected that the present study would affect z and v, and thus, these parameters were allowed to vary according to experimental manipulations in each of our models (Table 1), while a and t were estimated at the subject level.

Table 1.

HDDM models and model fit. Three models (Models 1–3) were tested. Drift rate, v, represents the speed or efficiency of evidence accumulation in reaching either of the two decision boundaries. Starting point bias, z, represents the initial amount of bias in favor of each of the two decision choices. Across three models, cue and/or stimulus were allowed to vary by either starting point (z) or drift rate (v). Right two columns display deviance information criteria (DIC) for both tasks as a measure of model fit. DIC deviance information criterion. Lower values indicate better model fit

	Cue	Stimulus	DIC
Model 1	-	v	10,790.11
Model 2	z	v	8448.33
Model 3	z & v	v	8288.36*

The asterisk signifies that the Model 3 (bolded with astericks) is the best fitting model with the lowest DIC

We employed three different model specifications to examine the effect of spider and crab cues on spider and crab decisions (Table 1 provides a list of these models). In Model 1, only v was permitted to vary based on stimuli (spider and crab), whereas z was estimated at the subject level and not allowed to change. In Model 2, z was allowed to vary based on cues (spider and crab), while v was permitted to vary based on stimuli (spider and crab). In Model 3, z was again allowed to vary by cues, while v was allowed to vary based on both the cues and stimuli. By comparing the models, we could determine which decision-making components were influenced by cues in each task. We evaluated the models using the Deviance Information Criterion (DIC), with lower DIC values indicating more favorable models.

We utilized a Python package for conducting DDM modeling, which employs a hierarchical Bayesian approach to estimate both group-level (hyperparameters) and subject-level parameters simultaneously (Wiecki et al., 2013). The package utilizes Markov Chain Monte Carlo (MCMC) to estimate the joint posterior distribution of model parameters. Each chain consisted of 10,000 samples with 5,000 burn-in samples (thin = 5) to ensure chain stability. We assessed model convergence using the Gelman-Rubin diagnostic R̂ and visually examined parameter convergence to ensure the values approached 1 and were smaller than 1.1 (Gelman & Rubin, 1992; Gelman et al., 2013; Wiecki et al., 2013). Finally, we conducted a posterior predictive check (PPC) for the best fitting model by simulating 10,000 samples for each subject (Gelman et al., 2013).

Results

Perceptual Thresholds for Spider Stimuli Are Lowered

We first statistically compared the absolute perceptual thresholds for spider stimuli and crab stimuli with a paired-sample t-test. The comparison of perceptual thresholds at which spider and crab stimuli were correctly identified (75% of the time) showed that spider stimuli were detected at a significantly lower contrast level (M = .097, SD = .063) compared to crab stimuli (M = 0.218, SD = 0.283, t(41) =  − 2.942, p = .005, Cohen’s d = 0.265). Overall, these findings show that spider stimuli were detected at lower absolute thresholds than crabs. We replicated these findings in Study 2. See Supplementary Results for details.

Speed and Accuracy of Perceptual Decision-Making

Next, we utilized 2 × 2 repeated-measures analyses of variance (rmANOVA) to investigate the effects of cue (spider vs crab cues) and stimulus (spider vs crab stimuli) on accuracy and speed (RT) of perceptual decisions. Accuracy results showed a main effect of cue, F(1, 41) = 15.999, p < .001, η_p² = .281, such that spider cues led to more accurate detection of following stimuli than crab cues, while the effect of stimulus type only approached significance, F(1, 41) = 3.765, p = .059, η_p² = .084. No interaction was observed between cue and stimulus (p > .05; see Fig. 2).

Fig. 2.

Accuracy for detection of spider and crab stimuli following spider and crab cues

For RT, results demonstrated a main effect of cue, F(1, 41) = 81.534, p < .001, η_p² = .600; a main effect of stimulus, F(1, 41) = 17.987, p < .001, η_p² = .305; and a significant interaction between cue and stimulus, F(1, 41) = 40.657, p < .001, η_p² = .498 (see Fig. 3). Simple effects tests showed that spider cue led to significantly faster detection of spider compared to crab stimuli (p < .001, mean difference =  − 0.147, 95% CI =  − 0.182 to − 0.113), while no difference in speed was noted following crab cue (p > .05). Accuracy and RT displayed the same pattern of results in Study 2 (see Supplementary Results).

Fig. 3.

Reaction time for detection of spider and crab stimuli following spider and crab cues

Perceptual Sensitivity and Criterion Shift

To determine the impact of cue salience on the sensitivity of subsequent perceptual decision-making, we compared the d′ for both cue types using paired sample t-tests. d′ was higher following spider cues (M = 2.213, SD = 0.979) than crab cues (M = 1.702, SD = 0.893, t(42) = 4.403, p < .001, Cohen’s d = 0.768). This observed effect was driven by a higher hit rate following spider cues (M = 0.847, SD = 0.162) compared to crab cues (M = 0.749, SD = 0.157, t(42) = 3.643, p < .001, Cohen’s d = 0.173), as false alarm rate did not differ significantly between cue types (p > .05). While criterion shift for crab cues (M = 1.784, SD = 2.362) was more conservative than for spider cues (M = 1.072, SD = 1.055), the difference did not reach significance (t(42) =  − 1.678, p = 0.101, Cohen’s d = 2.749). However, there was a significant difference between the two cues in Study 2 (spider cue, M = 0.859, SD = 0.613; crab cue, M = 1.574, SD = 2.030, t(52) =  − 2.387, p < .05, Cohen’s d = 2.208; see Supplementary Results). Hence, in both studies, the criterion shift was larger for crabs than for spiders, meaning that the criterion for crabs is more conservative than spiders, but it reached significance only in Study 2. Overall, our results show that spider cues led to better discrimination of spider and crab stimuli compared to crab cues while the results were more inconsistent for criterion bias.

Spider Cues Bias the Starting Point and Efficiency of Sensory Evidence Accumulation

Examination of DDM parameters with model comparison showed that, based on DIC, the best fitting model was Model 3 (Model 1 = 10790.11, Model 2 = 8448.33, Model 3 = 8282.06) (see Table 1), meaning that cues impacted both z and v. Model 3 also performed better than a control model with the same number of parameters. For Model 3, visual inspection of parameter convergence and Gelman-Rubin convergence statistics both suggested adequate model convergence. The PPC reproduced key aspects of the behavioral findings (see Table 2).

Table 2.

Posterior predictive check (PPC) for the best fitting model (Model 3) generated by simulating 10,000 samples for each subject

		Observed	Mean	SD	SEM	MSE	Credible
Accuracy		0.522272	0.522768	0.351183	2.46E − 07	0.12333	True
Mean_ub		1.118373	1.152696	0.277903	1.18E − 03	0.078408	True
SD_ub		0.42494	0.392496	0.22059	1.05E − 03	0.049712	True
	10q_ub	0.671794	0.793384	0.209557	1.48E − 02	0.058698	True
	30q_ub	0.858213	0.907965	0.22724	2.48E − 03	0.054113	True
	50q_ub	1.021336	1.045487	0.263986	5.83E − 04	0.070272	True
	70q_ub	1.252641	1.245364	0.333631	5.30E − 05	0.111362	True
	90q_ub	1.713776	1.624047	0.504194	8.05E − 03	0.262263	True
Mean_lb		− 1.168634	− 1.190029	0.285319	4.58E − 04	0.081865	True
SD_lb		0.408457	0.395726	0.229148	1.62E − 04	0.052671	True
	10q_lb	0.71621	0.825072	0.22964	1.19E − 02	0.064585	True
	30q_lb	0.922332	0.943778	0.242228	4.60E − 04	0.059134	True
	50q_lb	1.093983	1.084336	0.274189	9.31E − 05	0.075273	True
	70q_lb	1.314168	1.286949	0.339525	7.41E − 04	0.116018	True
	90q_lb	1.72868	1.666411	0.514691	3.88E − 03	0.268784	True

Next, for Model 3, we examined how cues impacted z and v. Inspection of the posterior distribution of the group-level means of the parameter estimates showed that the spider cue shifted z closer to the spider decision boundary (posterior probability of z > 0.50 was q > 0.99). On the other hand, crab cue shifted z slightly towards the crab decision boundary (posterior probability of z > 0.50 was q > 0.99; Fig. 4). However, the spider cue shifted z closer to the spider decision boundary than the crab cue did to the corresponding boundary (q > 0.99). An examination of v showed that, compared to the crab cue, the spider cue led to a higher v for the spider stimulus (posterior probability of the spider cue/spider stimulus v > the crab cue/spider stimulus was q > 0.99) and the crab stimulus (posterior probabilities of the spider cue/crab stimulus v > the crab cue/crab stimulus v were q > 0.95) but the increase for the spider stimulus was greater than for the crab stimulus (posterior probabilities of the spider cue/spider stimulus v > the spider cue/crab stimulus v were q > 0.99; Fig. 4).

Fig. 4.

HDDM results. Group level means of the drift diffusion model parameter estimates from the Model 3. A Starting point (z) of evidence accumulation for spider cue (red line) and crab cue (blue line). The black dashed line indicates the midpoint while higher values on the x-axis are closer to the spider decision boundary while lower values are towards crab decision boundary B Rate of evidence accumulation (v) for subsequently presented spider (solid line) and crab (dashed line) stimuli following spider cues (in red) and crab cues (in blue). More extreme values indicate a faster drift rate, with a positive sign denoting drift towards the spider decision boundary and a negative sign denoting drifting towards the crab decision boundary

In summary, our results showed that the spider cue shifted z more towards the spider decision boundary than the crab cue did to the corresponding boundary. Furthermore, the spider cue increased v for both spider stimulus and crab stimulus, but more so for spider stimulus.

Discussion

Several theories posit that phylogenetically salient stimuli are prioritized by our perceptual systems in a bottom-up automatic manner due to the evolutionary salience of their physical features (Bar-Haim et al., 2007; Brosch et al., 2010; Cisler & Koster, 2010; Larson et al., 2007; Mathews & Mackintosh, 1998; Mogg & Bradley, 2005; Öhman & Mineka, 2003; Öhman et al., 2001). Emerging research is demonstrating that in addition to bottom-up factors, top-down goal-driven factors contribute to the faster detection of these stimuli (LoBue, 2010, 2014; LoBue et al., 2010; Vromen et al., 2015). However, the facilitated detection demonstrated via bottom-up or top-down manipulation in studies so far may be attributable to biased attention or response tendencies and cannot be attributed specifically to perceptual processes without directly measuring these processes. Indeed, most of this research has focused on the speed with which phylogenetic targets are detected in arrays of other stimuli (Abado, Aue et al., 2020; Abado, Sagi et al., 2020; LoBue, 2010; LoBue et al., 2014; Öhman & Mineka, 2001; Vromen et al., 2015, 2016), as opposed to the sensitivity and accuracy of perceptual decisions.

Hence, in the present study, we used perceptual psychophysics to determine whether participants accurately identified spider images at a lower perceptual threshold than crab images. By using adaptive staircasing to manipulate the contrast levels of our image (increasing the contrast sharpens the image making it more identifiable and decreasing contrast softens the image making it blurred and harder to identify), we were able to determine the contrast level that spiders and crabs were identified correctly 75% of the time. Our results showed that spiders can be detected at lower levels of contrast compared to crab stimuli indicating a greater sensitivity to detect phylogenetic stimuli. These findings provide support for threat detection theories, which historically have proposed that threats are prioritized within attentional and perceptual systems (see Öhman et al., 2001; Öhman & Mineka, 2001). Enhanced processing of the global features of the spiders (Givon-Benjio & Okon-Singer, 2022) that are prioritized in the hierarchy of the visual system might be potential mechanisms for the observed lowered thresholds. Moreover, perceptual psychophysics research suggests that fast and accurate detection of a stimulus is associated with having lower perceptual thresholds for that stimulus category (Calvo & Esteves, 2005; Liu et al., 2012); hence, present findings may explain how facilitated perception contributes to faster detection of phylogenetically salient stimuli. While our research was conducted in a non-clinical population, it has implications for phobias, suggesting that perceptual thresholds for stimuli like spiders and snakes may be even more reduced in case of corresponding specific phobias as has been suggested by earlier frameworks (Cisler & Koster, 2010; Mathews & Mackintosh, 1998; Mogg & Bradley, 2005).

Next, we examined finding whether the lowered perceptual threshold for spiders is driven in a bottom-up manner by features or qualities inherent in these stimuli (Lundqvist et al., 2004; Öhman et al., 2001) or a top-down manner by the vigilance towards these stimuli (Vromen et al., 2015). We used a cued discrimination task in which we manipulated the salience of the top-down perceptual set and the bottom-up stimulus evidence, to examine the contributions of two factors to perceptual decision-making. By presenting stimuli at participants’ individual perceptual thresholds, we made it challenging for the participants to detect stimuli, thereby encouraging them to use the top-down cues to make perceptual decisions. Prior spider and crab cues both had an equal likelihood of being followed by spiders and crabs; therefore, differences in prospective decision-making can be attributed to the emotional salience of the cue. While we found that spider cues improved detection accuracy for both spider and crab stimuli, they specifically improved the speed with which spiders were detected compared to crab stimuli (more so than crab cues for corresponding stimuli). The significant interaction in reaction time demonstrates the unique effect of top-down modulation of phylogenetic salience on performance. In line with our hypothesis, spider cues led to improved perceptual sensitivity compared to crab cues; in other words, following spider cues participants were better able to discriminate between spider and crab targets. In the context of the present task, perceptual sensitivity provides a more fine-grained measure of the effect of phylogenetically salient cues compared to accuracy because it allows us to examine their effect on the ability to discriminate between hits and false alarms. However, inconsistent with our hypotheses, spider cues also influenced the decision criterion such that there was a more conservative criterion following crab cues, but a more liberal one for spider stimuli. While attentional cues influence perceptual sensitivity, expectation-related cues influence decision criterion (Wyart et al., 2012). It is possible that despite the spider cues not being intended to affect expectations since they did not provide probabilistic information in our study, their nature as threat-related stimuli may have inadvertently led to a greater influence on expectations compared to crab cues, thereby affecting criterion shift. This needs to be addressed in future research. Overall, our findings show that phylogenetically salient cues improve both the sensitivity and speed with which salient stimuli are discriminated from neutral stimuli.

Next, using HDDM, we examined the computational mechanisms contributing to top-down and bottom-up influences on the processing of phylogenetically salient spider stimuli. Spider cues shifted the starting point of evidence accumulation closer to the spider decision boundary, allowing participants to reach this decision faster, with less evidence. Changes in starting point in response to relevant prior cues are shown to be linked with faster RTs following cues (Mulder et al., 2012) and are associated with neural changes in frontoparietal regions of the brain including left inferior frontal gyrus and intraparietal sulcus (Heekeren et al., 2008; Kayser et al., 2010). Although such a shift in starting point explains the speed of spider decisions, it could compromise accuracy in the absence of other decision-making changes, as it could lead to increased endorsement of spider targets irrespective of stimulus characteristics. Our results demonstrated that spider cues additionally influenced the evidence accumulation process by increasing the drift rate, such that information was gathered more efficiently following the spider cue, for both spider and crab stimuli but more so for spider stimuli. These results align with earlier modeling results showing that attentional cues indicating the relevance of targets, such as in the present study, influence starting point and drift rate (Dunovan et al., 2014).

By utilizing HDDM alongside SDT, we were able to examine the biases in the dynamic, time-based component of perceptual decision-making. The findings from SDT and HDDM models complement each other and align with prior research. While the SDT and HDDM parameters do not necessarily map onto each other in a one-to-one manner, increased drift rate predicts more correct responses (hence higher hit rates) and is closely associated with higher perceptual sensitivity while there are weak correlations between starting point change and SDT’s criterion shift (Kloosterman et al., 2019). In our own data, we see a similar pattern such that drift rates following spider cues correlate uniquely with spider d-prime while drift rates following crab cues correlate uniquely with crab d-prime (see Supplementary results). While the starting point following crab cues correlates with crab criterion shift, the same was not seen for spider cues.

Neurally, these findings align with current conceptualizations of visual stimulus processing. While traditional views posited that affective visual stimuli are processed automatically via a subcortical low-road, newer approaches suggest that emotionally valenced stimuli engage both cortical and subcortical routes (de Gelder et al., 2011; LoBue et al., 2010; Pessoa & Adolphs, 2010). Furthermore, although it is beyond the scope of the current study, future research could explore these findings within the predictive processing framework. This framework posits that, rather than simply expediting the processing of sensory stimulus characteristics, humans generate predictive models of potential sensory input, which are then compared to actual sensory data (Friston, 2010; Rao & Ballard, 1999). According to this framework, to optimize efficiency, facilitated bottom-up sensory information is constantly compared and adjusted to generative models created by expectation, prior knowledge, and memory, which are carried down in neural feedback pathways. Our findings hint at the possibility that perceptual biases in both top-down and bottom-up processing streams might warrant further examination of the dual pathways, involving both subcortical and higher-level cortical areas.

Future research on clinical populations with spider phobia has the potential to uncover additional insights into the mechanisms underlying spider-related biases in perception and their potential for modification. Individuals with spider phobia show expectancy bias, i.e., they overestimate the probability of encountering spiders in their environment (Aue & Hoeppli, 2012; de Jong & Muris, 2002), and attention bias where they engage with phobic stimuli faster and disengage slower (Abado, Aue et al., 2020; Abado, Sagi et al., 2020). Studies on this population may help parse out the differential influences of attention and expectation on heightened perception of spiders. Further, some studies argue that biases for phylogenetic stimuli may not be sensitive to manipulation through prior cues in individuals with or without spider phobia (Aue et al., 2013). However, our results imply the contrary, showing that top-down cues may indeed affect attention-related biases and warrant further examination in individuals with clinical levels of spider phobia. These findings suggest that targeting top-down expectations and vigilance towards threat may be effective in reducing preferential processing of threat (i.e., attention bias modification (Teng, 2022).

It is also important to discuss our findings of increased discrimination between spider and crab stimuli in light of findings of learned overgeneralization from threatening to physically (Ginat-Frolich et al., 2019; Lissek, 2012) or conceptually (Dunsmoor & Murphy, 2015) similar non-threatening stimuli. Paradigmatically, the latter findings arise from implicit learning tasks, whereas our task is an explicit task in which participants voluntarily use cues to discriminate between threatening and neutral stimuli. It is possible that in implicit tasks, there is a tendency to generalize but when explicitly asked to discriminate, individuals are able to discriminate threatening from neutral targets better following top-down threat-related cues. It is also possible that the generalizing tendency is higher in clinically anxious individuals (Ginat-Frolich et al., 2019; Lissek, 2012) than in individuals at lower levels of spider anxiety as in our sample. Finally, it is important to examine whether the overgeneralization effect on perceptual decision parameters can be diminished via top-down cues that encourage discrimination of physically or conceptually similar threatening and neutral stimuli. These are questions that are important to address in future research. Moreover, infants have perceptual templates for spiders that are generalizable to real-world images of spiders as early as 5 months of age (LoBue, 2010; Rakison & Derringer, 2008). Understanding the developmental trajectories of how the top-down perceptual templates are refined and updated over lifetime is an important question to address in future research. A strength of the current study design is that we compared spiders to neutral crab counterparts, which are very similar in terms of stimulus characteristics including physical features, luminance, and spatial frequency, allowing us to better examine the influence of emotional salience and top-down sets. While this allows us to achieve better experimental control, it limits the generalizability of our findings which should be examined for other phylogenetic stimuli like snakes and dogs, as well as other kinds of neutral stimuli. On the other hand, there are several limitations of the current study. The current study sample was constituted of only undergraduate students. Extending the findings to different age groups would both increase the generalizability of the findings and also provide a significant insight into the development and changes of phylogenetically salient biases across lifetime. Overall, our study provides strong empirical support for theories that espouse the role of perception in the biased processing of phylogenetically salient stimuli and elucidates the role of top-down and bottom-up factors in guiding this perception.

Supplementary Information

Below is the link to the electronic supplementary material.

Additional Information

Funding

The study was funded by internal departmental funds from Stony Brook University to Dr. Aprajita Mohanty.

Conflict of Interest

The authors declare no competing interests.

Data Availability

No datasets were generated or analysed during the current study.

Code availability

Not applicable.

Author Contributions

A.M., and T.S. designed the study. T.S. carried out the experiment. S.O. conducted the data analyses. J.J. and A.M. assisted with the data analyses. S.O wrote the main manuscript text with support from A.M, and T.S.. S.O. prepared all figures and tables. A.M. supervised the study. A.M., S.O., T.S., J.J., G.I., and M.S. reviewed the manuscript and provided feedback.

Ethics approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.