The Shared Feature Principle in Person Perception: Shared Features Lead to Assumptions About Other Features

The shared features principle builds on four terms: source objects, target objects, source features, and target features (for more on these terms and the conceptual account from which they derive, see De Houwer et al., 2019). The term ‘object’ is a general one that refers to many different stimuli: a person, a brand product, a nonsense word, or a social group. The term ‘feature’ refers to any state or attribute of an object. Objects have many such features. For instance, if the object is a person, it could have physical features (e.g., being tall), psychological features (e.g., being intelligent), or behavioral features (e.g., acting aggressively).

Target features refer to those aspects of an object about which observers make assumptions. These are typically the dependent variables under investigation. The target object possesses these target features. Source features refer to those aspects of an object that give rise to assumptions about target features. The source object possesses these source features. By speaking abstractly about sources and targets, objects and features, the principle can accommodate a wide range of psychological phenomena across theories, domains, and disciplines.

The central premise of the shared features principle is as follows: when a source and target object share one (‘bridge’) feature, people will make assumptions about other features of the target object. If people assume that, because the source and the target share a bridge feature, they likely share other features as well, then we call this feature transfer: features of the source object are transferred to the target. If feature sharing leads to assumptions about the target that involve more than mere transfer from one object to another, then we call this feature transformation, with the target acquiring or changing various features.

Hughes et al. (2020) examined the idea that feature sharing will transfer features from one object to another. In their studies, participants observed three simultaneously presented stimuli onscreen: a neutral generic word (target), a positive generic word (source), and a negative generic word (source). During the first half of each trial, all three stimuli were presented in white font against a black background. During the trial’s second half, a target came to share a feature (e.g., the same colored font) with one of the two source stimuli. Participants evaluated the targets in line with the source it shared a feature with, both on self-reported (i.e., ratings and behavioral intentions) and automatic evaluative measures (i.e., evaluative priming and IAT scores). These effects also appeared when targets and sources shared other physical features (e.g., size, location) and conceptual features (group membership). Critically, if the observed changes in liking would have followed from the pairing of stimuli as in Evaluative Conditioning, then participants should have evaluated the target objects ambivalently, given that they appeared repeatedly with both positive and negative sources. However, this was never the case.

Feature Transfer. Hughes and colleagues (2020) showed that shared features influence people’s evaluative assumptions about a target object. However, the principle itself is more general. It predicts that many other assumptions may be influenced based on feature sharing. Consider the case of person perception: it may be that when a target individual shares a bridge feature with a source individual, perceivers assume that those individuals also share other features.

Such an idea is consistent with past work in the person perception literature. For instance, physically similar romantic partners are more likely to be perceived as sharing personality traits than physically dissimilar partners (Glassman & Andersen, 1999). From a shared feature principle perspective, this effect follows because a target and source individual share one feature (i.e., romantic partnership), and therefore, people assume that they also share other features (i.e., personality traits). Elsewhere, research on the “guilt-by-association” and “honor-by-association” effects show that perceptions of one person can transfer to another person whenever those individuals share physical features such as co-occurrence (Walther, 2002), facial similarity (Verosky & Todorov, 2010), or even conceptual features such as family membership or blood ties. Indeed, people find it acceptable to judge one person (target) based on what they know about another (source) when they are family members (shared feature). At the same time, mock jurors view criminal defendants (target) as more likely to be guilty when they are similar to other family (shared feature) members (source) who have previously been convicted of a crime (Rerick et al., 2021). Finally, the attribute conditioning literature indicates that pairing one object (e.g., the face of an unknown male) with another (e.g., another male performing sports) leads the former to acquire similar attributes to the latter (e.g., observers perceive the unknown man as more athletic than before). These changes in attribute assessments are evident across many characteristics (e.g., athletic, funny, educated, intelligent) and present even after evaluative changes (i.e., likes and dislikes) are controlled for (see Unkelbach & Högden, 2019; for an overview).

In summary, there are many situations where a source and target individual share one feature (e.g., common group membership, identity, physical resemblance, co-occurrence, or proximity), and as a result, people assume that the source and the target also share other features (e.g., how contaminated, guilty, athletic, or morally good they are). Although these findings certainly seem to be consistent with a shared features perspective, it is currently an open question if person perception can be experimentally influenced by the sharing of features.

With this in mind, we conducted three experiments. We used procedures similar to those of Hughes et al. (2020), along with stimuli and attributes typically employed in the attribute conditioning literature (see Förderer & Unkelbach, 2014). Specifically, we used a learning task wherein three individuals appeared simultaneously onscreen: a neutral target person (i.e., “Chris” or “Bob”), a tall source individual, and a short source individual. Across trials, Chris shared a feature (i.e., color) with tall sources, whereas Bob shared that same feature with short sources. Next, we assessed how tall or short Chris and Bob were perceived to be, using self-reported ratings and an indirect measure (i.e., Semantic Misattribution Procedure; SMP). If feature sharing leads to feature transfer, then Chris should be perceived as tall and Bob as short. Such an outcome is notable because it cannot be explained via mere contiguity (i.e., it is not a mere attribute conditioning effect). If it were, then Chris and Bob should be perceived ambivalently, given that they were always paired with both tall and short individuals.

Feature Transformation. We also predicted that feature sharing would influence not only assumptions about height but also assumptions about other features. Thus, the target persons may acquire additional novel features as well. To illustrate this point, consider the radiating beauty effect in the halo literature. Research shows that the level of attractiveness of a female romantic partner influences the types of assumptions people make about her male romantic partner (e.g., that he is a good leader; Kocoglu & Mithani, 2020). It is not that sharing a feature (i.e., romantic partnership) leads to a transfer of attractiveness from one partner to another. Rather, sharing a feature leads to a change in how the target individual is subsequently perceived by others (i.e., that he is a good leader; see also Rougier et al., 2023).

A similar situation may also unfold in our studies. If participants assume Chris to be taller following the training phase, then he may also be perceived as stronger, more intelligent, and a better leader (i.e., relative to Bob, who shared a feature with a short source). To test this possibility, we asked participants to also rate Chris and Bob on several non-trained features (e.g., their intelligence, leadership, and strength). Observing such an outcome would suggest that feature sharing may trigger a cascade of assumptions about a target individual and not merely a transfer of a single feature from source to target.

We report all data exclusions, all manipulations, and all measures. If we do not discuss a measure in the main text, it is stated and available in the supplements. All experiments’ pre-registration files, materials, data, and analytic scripts are available at https://osf.io/2ps4n/. We conducted this research according to the American Psychological Association’s (2017) Ethical Principles in the Conduct of Research with Human Participants. We have no conflicts of interest to declare.

Experiment 1 was the first time we assessed changes in attribute assessments of people as targets following shared features training. We had no a priori effect size expectations and decided to collect the data of approximately 200 participants based on the availability of resources.

We recruited 209 participants (102 women, 107 men) ranging in age from 18 to 62 (M = 31.2, SD = 10) via the Prolific platform (https://prolific.co/). They participated online for a monetary reward. The following criteria were required to qualify for inclusion in the study: English as a first language, 75% quality rating based on prior study completion on Prolific, and no participation in other studies from our lab. We counterbalanced whether Chris or Bob shared a feature with tall or short sources across participants. The Source Object Feature of being tall or short represented the Independent Variable in our analyses. We also counterbalanced two method factors: order of the attribute measures (i.e., SMP or Self-Reports first) and attribute assessment order (i.e., assumed transfer: height, or assumed transformation: intelligence, leadership, and strength). The attribute assessment measures represented the Dependent Variables in our analyses.

Target Stimuli (People). We conducted a pilot study with a separate set of 51 participants (29 female) prior to Experiment 1. Participants rated a series of faces individually in terms of their perceived height using a scale ranging from 1 (“He looks like he is Short”) to 8 (“He looks like he is Tall”). They also rated the faces in terms of their perceived strength (from 1 [“He looks like he is Strong”] to 8 [“He looks like he is Weak”]), leadership qualities (from 1 [“He looks like a Bad leader”] to 8 [“He looks like a Good leader”]), and intelligence (from 1 [“He looks like he might be Stupid”] to 8 [“He looks like he might be Intelligent”]).

Based on pilot data, we selected two stimuli as targets and labeled them “Chris” and “Bob”. Chris (M = 4.54, SD = 1.15) and Bob (M = 4.65, SD = 1.21) were rated as closest to the mid-point of the height scale and were matched in terms of their perceived height, t(49) = -0.45, p = .65, d = -0.06, 95% CI [-0.34; 0.21], leadership, t(49) = 0.31, p = .76, d = 0.04, 95% CI [-0.23; 0.32], intelligence, t(49) = 1.21, p = .23, d = 0.17, 95% CI [-0.11; 0.45], and strength assessments, t(49) = -1.67, p = .10, d = -0.24, 95% CI [-0.52; 0.05].

Source Stimuli (Height Attributes). We also had the same participants from the pilot study rate a series of tall and short images. We obtained these images online and selected five to serve as tall sources and five to serve as short sources (see Appendix). Results indicate that the “tall” source stimuli (M = 6.96, SD = 0.69) were perceived as significantly taller than the “short” source stimuli (M = 3.64, SD = 0.66), t(50) = 24.13, p \< .001, d = 3.38, 95% CI [2.66; 4.09].

Participants read a general overview of the experiment and provided informed consent. After that, they provided demographic information (i.e., age and gender), proceeded to the shared features training task, followed by attribute assessments, and exploratory questions. The experiment took, on average, 15 minutes to complete.

Before the training phase, participants were told the following: “In the next part of the experiment, you are going to see two individuals. One is called Chris, and the other is called Bob. Chris or Bob will appear on the left, while two other images (one tall and the other short) will appear on the right side of the screen. Chris or Bob and these other images will be surrounded by a colored frame. Please pay close attention to the images and the colored frames. You will be asked some questions about this later on”.

The training phase consisted of three blocks of 16 trials (48 total). Each block had two types of trials that differed in their second half (see Figure 2); either Target Stimulus 1 (TS1) was presented with a frame of the same color as the frame of a tall individual’s image, or Target Stimulus 2 (TS2) was presented with a frame of the same color as the frame of a short individual’ image.

Each trial presented three stimuli simultaneously onscreen: one of the two Target Stimuli (i.e., Chris or Bob) and two Source Stimuli (i.e., an image of a tall and short individual). All three stimuli initially appeared without a colored frame. After 3000ms, the target and one of the source stimuli were surrounded by the same colored frame (e.g., purple), while the second source stimulus was surrounded by a different colored frame (e.g., blue). Stimuli remained onscreen for another 3000ms. After that, they disappeared, and following an inter-trial interval of 1250ms, the next trial began. The color of the frames randomly varied across trials to ensure that no color assumed a specific attribute meaning. The image of the tall target was always larger than that of the short target. We varied the location of the tall and short sources so that tall sources were sometimes presented at the top of the screen and at other times at the bottom and vice versa for the short sources). We employed four colored frames (i.e., blue, green, yellow, and purple).

SMP. The SMP (Imhoff et al., 2011; Sava et al., 2012) is a variant of the Affective Misattribution Procedure (AMP, Payne et al., 2005) to indirectly assess specific attributes (see Förderer & Unkelbach, 2015). We informed participants that they would see two pictures in rapid succession, the first being either Chris or Bob (i.e., one of the Target Stimuli) and the second being a Chinese ideograph. They were also told that the first image would simply serve as an orientation stimulus and should be disregarded. We informed them that their goal was to decide if the Chinese ideograph represents a word with a “short” or “tall” meaning and to do so as quickly as possible. Each trial consisted of a target stimulus presented for 100ms, followed by a blank screen for 125ms, and an ideograph for 100ms. A black-and- white noise picture immediately replaced the ideograph until participants responded (i.e., pressed a left or right key). Each target stimulus served as a prime stimulus thirty times, resulting in 60 trials. “Shorter” was assigned to the left key and “Taller” to the right key.

Please note that due to the task effort, we only used the indirect measure to assess the feature transfer effect (i.e., height), not the feature transformation effects (i.e., strength, leadership, and intelligence).

Self-Reports. As a direct self-report measure, we assessed the transfer attribute “height” by asking participants to: “Please rate Chris [Bob] using the scale below”. Participants saw a picture of Chris or Bob and a response scale ranging from 1 (Short) to 9 (Tall). Participants rated both targets, one at a time, in a counterbalanced manner. We assessed the non-trained features similarly. Participants saw a picture of Chris or Bob and rated the target along the three assumed transformation dimensions (i.e., Intelligence, Strength, and Leadership), using a similar scale as for height.

We also asked participants several questions about perceived demand, reactance, hypothesis awareness, influence awareness, and familiarity with the Chinese ideographs used during the SMP. We do not discuss these exploratory questions here further (see Supplementary Materials).

We excluded participants who: (a) failed to complete the entire experimental session and thus provided incomplete data (n = 10); (b) completed the experiment in less than 10mins or more than 40mins (n = 0), or (c) reported being able to read Chinese, as such an ability would influence responding to the Chinese ideographs during the SMP (n = 2). These exclusions led to a final sample of 197 participants.

We excluded trials with excessively long responses (i.e., those with a latency greater than 2000ms; 6.3%). After that, we computed a ‘tall’ SMP effect by calculating the proportion of ‘taller’ responses emitted on trials with a prime (i.e., the supposed orientation stimulus) that shared a feature with a tall stimulus during training, relative to all trials where this stimulus functioned as the prime. Similarly, we also computed a ‘shorter’ SMP effect.

We tested target height (i.e., feature transfer variable) and target intelligence, strength, as well as leadership (i.e., feature transformation variables) attribute assessments (dependent variables) as a function of tall vs. short source stimuli (independent variable), using a series of t-tests. We tested the SMP scores in a similar set of analyses. As these are within-person comparisons, we report Cohen’s dz for all comparisons. We also report Bayes factors calculated following procedures outlined by Rouder et al. (2009) to estimate the evidence for the hypothesis that there is a difference in attribute assessments for targets that share features with tall vs. short source stimuli (i.e., alternative hypothesis) or no such difference (i.e., null hypothesis).

A feature transfer effect emerged for height on self-reported ratings, t(196) = 11.58, p \< .001, dz = 0.83, 95% CI [0.66; 0.99], BF₁₀ > 10⁴; participants rated the target which shared a feature with a tall source as being taller (M = 7.1, SD = 1.94) than the target that shared a feature with a short source (M = 3.8, SD = 2.2). The ‘taller’ target was significantly taller, t(196) = 14.98, p \< .001, and the ‘shorter’ target was significantly shorter than the mid-point of the scale, t(196) = 7.49, p \< .001. A similar effect also emerged on the SMP, t(196) = 6.89, p \< .001, dz = 0.49, 95% CI [0.34; 0.64], BF₁₀ > 10⁴, such that the target which shared a feature with tall source stimuli led to more tall assessments (M = 0.69, SD = 0.25) than the target which shared a feature with short stimuli (M = 0.45, SD = 0.29).

We found a feature transformation effect for strength on the self-report measure, t(196) = 4.04, p \< .001, dz = 0.29, 95% CI [0.15; 0.43], BF₁₀ = 180.97. Participants rated the target that shared a feature with tall source stimuli as stronger (M = 6.03, SD = 1.52) than the target that shared a feature with short stimuli (M = 5.35, SD = 1.59). However, no such effect emerged for the leadership, t(196) = 1.08, p = .28, dz = 0.08, 95% CI [-0.07; 0.22], BF₁₀ = 0.14, or intelligence dimensions, t(196) = 0.49, p = .63, dz = 0.04, 95% CI [-0.11; 0.17], BF₁₀ = 0.09.

The sharing of features led to a transfer of attributes from one person (source) to another (target). During the training phase, participants encountered a novel individual together with images of tall and short people. Manipulating the type of image the target shared a feature (color) with influenced how that individual was perceived, viewed as tall by some and short by others. This shared features effect was evident as a clear feature transfer for height on self-reported ratings and SMP scores. Of note, the color of the frame is non-diagnostic for inferences about height, and we also varied the color of the frames (i.e., the shared feature). Partial evidence also emerged for feature transformation, such that the target perceived as tall following the learning phase was also viewed as stronger than the “short” target individual. However, we found no change in rated leadership or intelligence.

Hughes et al. (2020) showed that when a source and target object share a bridge feature, people assume that those objects share other features. Critically, they only assessed valence as a feature. Experiment 1 extends this feature transfer from abstract objects to person perception, and from general valence to a particular feature (i.e., height). In addition, Experiment 1 shows that a shared feature influences other assumptions about the target (i.e., Chris was assumed to be taller and stronger, whereas Bob was assumed to be shorter and weaker). If this finding replicates, it would suggest that feature sharing leads targets to changes in perceivers’ assumptions across multiple dimensions. However, if feature transformation effects fail to emerge in Experiment 2, then we can better attribute the absence of such outcomes to the procedure itself rather than the absence of a relationship between trained and non-trained features per se.

We set out to replicate Experiment 1 and extend upon our prior findings for shared feature effects in several ways. First, we sought additional evidence for a feature transfer effect in person perception (i.e., assumptions about a person’s height could be altered via sharing irrelevant features with a tall/short source). Second, we wanted to gather more evidence for a feature transformation effect (i.e., assumptions about a person’s strength, leadership, and intelligence could be altered via sharing features with a tall/short source) effect.

We were primarily interested in comparing attribute assessment differences between targets that shared features with tall vs. short sources using self-report and SMP measures. To detect a medium Cohen’s dz effect size (i.e., 0.35) in a two-tailed paired samples t-test at the conventional alpha level (.05) with 80% power required 52 participants. We decided to sample approximately 280 participants based on the availability of resources.

We recruited 279 participants (151 women, 128 men) ranging in age from 18 to 50 (M = 31.5, SD = 8.7) via Prolific in exchange for a monetary reward. We used a similar design and set of counterbalanced method factors as in Experiment 1.

Target Stimuli (People). We conducted a similar pilot study (n = 51; 34 female) as in Experiment 1. This time participants were asked to rate eleven new target faces along the same dimensions as assessed in Experiment 1. Our aim was to identify two faces that were even more matched on the feature transfer (height) and transformation (strength, intelligence, and leadership) dimensions. Based on this pilot we decided to replace the original “Chris” image with a new “Chris” face. Analyses revealed that this new Chris (M = 4.77, SD = 1.44) and Bob (M = 4.65, SD = 1.21) did not differ in their perceived height, t(100) = 0.45, p = .66, d = 0.09, 95% CI [-0.30; 0.48], leadership, t(100) = 1.08, p = .29, d = .21, 95% CI [-0.18; 0.60], intelligence, t(100) = 0.73, p = .47, d = .15, 95% CI [-0.24; 0.53], or strength qualities, t(100) = 0.68, p = .50, d = .13, 95% CI [-0.26; 0.52].

Source Stimuli (Height Attributes). We used the same set of tall and short images as in Experiment 1.

The procedure was highly similar to Experiment 1 except for the self-report measures for the feature transformation effects.

Self-Reports. We assessed assumptions about target height by asking participants: “Based on what you just learned, would you say that Chris[Bob] is…”. Participants saw a picture of the individual and a response scale ranging from 1 (Short) to 100 (Tall). They were asked to rate both targets, one at a time, in a counterbalanced manner. Participants rated the targets on the feature transformation attributes in a similar manner. Participants were presented with a picture of Chris or Bob and asked to rate him along four dimensions (Leadership [Follower-Leader], Strength [Weak-Strong], Dominance [Submissive-Dominant], and Masculinity [Feminine-Masculine]) using a similar scale as above. ¹

Participants who provided incomplete data (n = 4) or reported being able to read Chinese (n = 5) were removed prior to analysis. These exclusions led to a final sample of 270 participants.

We found a feature transfer effect for height on self-reported ratings, t(269) = 15.46, p \< .001, dz = 0.94, 95% CI [0.79; 1.08], BF₁₀ > 10⁴. Participants rated the target that shared a feature with tall source stimuli as taller (M = 76.86, SD = 24.75) than the target that shared a feature with short stimuli (M = 29.62, SD = 27.79). The ‘taller’ target was significantly taller than the ‘medium’ mid-point of the scale t(269) = 47.71, p \< .001, whereas the ‘shorter’ target was significantly shorter than the mid-point of the scale, t(269) = 14.56, p \< .001.

We also found a shared feature effect on the SMP, t(269) = 8.94, p \< .001, dz = 0.54, 95% CI [0.42; 0.67], BF₁₀ > 10⁴, such that the target which shared a feature with tall source stimuli led to more tall assessments (M = 0.69, SD = 0.24) than the target which shared a feature with short stimuli (M = 0.44, SD = 0.2).

A feature transformation effect emerged for strength, t(269) = 8.07, p \< .001, dz = 0.49, 95% CI [0.36; 0.62], BF₁₀ > 10⁴, dominance, t(269) = 7.29, p \< .001, dz = 0.44, 95% CI [0.32; 0.57], BF₁₀ > 10⁴, leadership, t(269) = 8.99, p \< .001, dz = 0.55, 95% CI [0.42; 0.68], BF₁₀ > 10⁴, and masculinity, t(269) = 7.15, p \< .001, dz = 0.44, 95% CI [0.31; 0.56], BF₁₀ > 10⁴. The target which shared a feature with tall source stimuli was assessed as stronger (M = 65.13, SD = 16.90), more dominant (M = 62.14, SD = 19.21), a better leader (M = 63.67, SD = 19.25), and more masculine (M = 71.53, SD = 17.33) than the target that shared a feature with short stimuli, who was assessed as weaker (M = 51.40, SD = 18.88), more submissive (M = 46.60, SD = 19.82), more of a follower (M = 45.34, SD = 19.79), and more feminine (M = 61.21, SD = 20.92).

We replicated that sharing features again led to a feature transfer of height attributes from one person (source) to another (target). Different from Experiment 1, we also found stronger evidence for feature transformation effects, such that participants rated the “tall” target also as stronger, more of a leader, dominant, and masculine than the “short” target. Our final experiment aimed to replicate and extend these feature transformation effects.

In Experiment 2, the feature transformation effects clearly aligned with the height dimension – either strongly related to one endpoint of that continuum [tall] or the other [short]. In Experiment 3, we selected attributes that were either strongly related (leadership, dominance) or unrelated to the height dimension (truthfulness, caring). We predicted that feature transformation would be limited to the height-relevant attributes and that we would not observe any effects for the height-irrelevant attributes. If this were the case, then it would suggest that feature transformation may be selective (i.e., Rougier et al., 2023). If targets were to acquire both positive and negative features (e.g., leader but less caring), it would also support the idea that a change in specific attributes rather than a change in general evaluations is taking place.

We employed a similar convenience sampling strategy as in Experiment 2 based on the availability of resources.

We recruited 207 participants (131 women, 46 men) ranging in age from 18 to 49 (M = 33.5, SD = 7.8) via Prolific in exchange for a monetary reward. We used a similar design and set of method factors as in Experiment 2.

The same novel target individuals (i.e., Chris and Bob) from Experiment 2 served as target stimuli during the shared features training phase. The same five pictures representing tallness and five pictures representing shortness served as source stimuli as in Experiment 2.

The procedure was similar to Experiment 2 with one change: participants rated the targets on a new set of attributes following the training phase.

Self-Reports. Following training, participants were presented with a picture of Chris or Bob and asked to rate each on the trained (height) dimension from 1 (Short) to 100 (Tall). They also rated those same individuals along two height ‘related’ dimensions (i.e., Dominance [Submissive-Dominant], and Leadership [a Follower- a Leader]), as well as two height ‘unrelated’ dimensions (i.e., Truthfulness [Truthful-Liar] and Caring [Caring-Uncaring]) using the same response scale as noted above. The order of trained and non-trained attribute assessments was counterbalanced across participants. ²

We removed participants with incomplete data (n = 6) or who reported being able to read Chinese (n = 2) prior to analysis. These exclusions led to a final sample of 199 participants.

A feature transfer effect for height emerged on self-reported ratings, t(199) = 14.33, p \< .001, dz = 1.02, 95% CI [0.84; 1.19], BF₁₀ > 10⁴. Participants rated the target that shared a feature with tall source stimuli as taller (M = 77.69, SD = 23.48) than the target that shared a feature with short source stimuli (M = 28.99, SD = 26.08). The ‘taller’ target was significantly taller than the ‘medium’ mid-point of the scale t(199) = 46.69, p \< .001, whereas the ‘shorter’ target was significantly shorter than the mid-point of the scale, t(199) = 15.68, p \< .001. A similar effect emerged on the SMP, t(199) = 9.77, p \< .001, dz = 0.69, 95% CI [0.54; 0.85], BF₁₀ > 10⁴, such that the target which shared a feature with tall source stimuli led to more tall assessments (M = 0.73, SD = 0.22) than the target which shared a feature with short stimuli (M = 0.42, SD = 0.29).

A feature transformation effect also emerged for the attributes related to height: leadership, t(199) = 7.86, p \< .001, dz = 0.56, 95% CI [0.41; 0.71], BF₁₀ > 10⁴, and dominance, t(199) = 8.47, p \< .001, dz = 0.60, 95% CI [0.45; 0.75], BF₁₀ > 10⁴. Participants rated the target that shared a feature with tall source stimuli as more of a leader (M = 64.07, SD = 20.26) and more dominant (M = 64.36, SD = 19.43) than the target that shared a feature with short stimuli, who was assessed as less of a leader (M = 43.65, SD = 20.49), and more submissive (M = 42.35, SD = 19.85).

Unexpectedly, an effect emerged in the opposite direction for the two height-unrelated attributes: truthfulness, t(197) = -3.17, p = .002, dz = -0.26, 95% CI [-0.37; -0.08], BF₁₀ = 9.96, and caring, t(198) = -3.06, p = .003, dz = -0.22, 95% CI [-0.36; -0.08], BF₁₀ = 7.25. Participants rated the target that shared a feature with a short source as more truthful (M = 57.38, SD = 13.91) and caring (M = 58.36, SD = 16.45) than the target that shared a feature with tall stimuli, who was assessed as less truthful (M = 52.65, SD = 15.13), and less caring (M = 52.44, SD = 17.58).

Finally, we wanted to know if the magnitude of attribute assessments was larger for height ‘relevant’ compared to ‘irrelevant’ attributes. For each attribute we measured, we first subtracted the ratings for the short individual from the ratings for the large individual to create a difference score. We then combined the difference scores for leadership and dominance to create a ‘height relevant’ attribute score and combined the difference scores for truthfulness and caring to create a ‘height irrelevant’ attribute score (also see Rougier et al., 2023). A paired-samples t-test revealed a significant difference between height-relevant and height-irrelevant attribute scores, t(197) = 8.22, p \< .001, dz = 0.59, 95% CI [0.43; 0.74], BF₁₀ > 100, such that feature transformation effects were relatively larger for ‘height relevant’ compared to ‘height irrelevant’ features.

We once again replicated the feature transfer effect for height. We also replicate the feature transformation effect observed in Experiment 2 and partially in Experiment 1. This effect emerged for the height-relevant attributes (i.e., leadership and dominance) such that the participants assumed the “tall” target to be more of a leader and dominant than the “short” target. An effect also emerged for the height-irrelevant attributes. Participants assumed the short target to be more truthful and caring than the tall target. Critically, however, height-relevant effects were larger than irrelevant effects. These findings appear to reflect a change in attributes rather than mere evaluations, given that the short target was perceived as more submissive (negative) and caring (positive), whereas the tall target was perceived as more of a leader (positive) and more uncaring (negative).

We introduce the shared features principle into the area of person perception. At its core lies the idea of feature sharing - namely that target individuals often share ‘bridge’ features with source individuals. When this occurs, two different outcomes may emerge: feature transfer, wherein a feature (e.g., being perceived as tall or short) is transferred from source to target, and feature transformation, wherein a target acquires features from its relation to the source that involves more than mere transfer (e.g., being perceived as more or less strong, dominant, or masculine). We tested these predictions in three experiments. Participants first completed a training task wherein three stimuli were simultaneously presented onscreen: a neutral target individual (Chris or Bob), a tall source individual, and a short source individual. During certain trials, one of the targets (Chris) shared a bridge feature (color) with tall individuals, whereas in other trials, another target (Bob) shared a bridge feature with short individuals. After that, self-reported and indirectly assessed perceptions of these individuals to test for feature transfer (i.e., height), and self-reported perceptions tested feature transformation (i.e., strength, dominance, leadership, truthfulness, care).

All our studies showed evidence for feature transfer such that perceptions of Chris or Bob’s height changed in accordance with the source they shared a bridge feature with. Sharing a bridge feature (here: color) with tall sources led Chris and Bob to be perceived as taller than when they shared a bridge feature with short sources. These effects were robust in magnitude and replicability, evident on multiple outcome measures (i.e., ratings, SMP), and emerged even though both individuals were simultaneously paired with tall and short sources on each trial. These experiments are the first evidence supporting the shared features principle in the domain of person perception. Although many semi-experimental effects could be seen as instances of the principle, this is the first set of studies to demonstrate that feature sharing can lead to feature transfer across a range of personal attributes.

We also observed evidence for feature transformation, such that a target individual who was perceived as taller following training was also viewed as being stronger (Experiments 1-2), more masculine (Experiment 2), a better leader, and dominant (Experiments 2-3) than the short individual (who was scored lower on each of these dimensions). Thus, sharing a bridge feature does not merely transfer a single feature from source to target but rather to a cascade of assumptions about multiple attributes target individuals are assumed to possess. One might also label these halo effects from the transferred feature. In what follows, we consider the conceptual implications of these findings, discuss their limitations, and consider open questions and future research directions.

So far, we focused on the shared features principle and tested its core predictions. However, why exactly do these effects emerge? If one looks closely, it becomes apparent that the principle implies an effect-centric functional account: it simply states a functional if-then relationship involving elements in the environment (i.e., source objects, target objects, bridge features) and related changes in behavior (i.e., feature transfer or transformation). IF objects share a bridge feature, THEN this will result in feature transfer or transformation (for more on functional effect-centric approaches, see Hughes et al., 2016). As such, the principle is agnostic to any one theoretical account and is likely compatible with many, both at the functional and mental levels of analysis.

One possibility is that shared feature effects are themselves instances of ‘symbolic’ or relational behavior (i.e., a type of behavior that involves ‘responding to the relationship between stimuli’; see Hughes et al., 2020; Hughes & Barnes-Holmes, 2016). According to this perspective, bridge features constitute relational contextual cues in the environment. These cues signal how one should respond (e.g., much like how the red and green lights on a traffic light signal how one should behave) and state that a relational response should be emitted (i.e., that one stimulus should be related to another). In the case of shared feature effects, bridge features signal that one should assume there is a relation of similarity between one stimulus (target) and another (source). Once this relationship is established, a corresponding transfer or transformation of stimulus properties occurs, influencing subsequent behavior towards those stimuli. This symbolic account would explain why so many types of bridge features can come to trigger a “similarity” relation between objects because people can imbue almost any event with symbolic meaning, even spatial-temporal events such as contiguity (for related arguments, see De Houwer & Hughes, 2016). It would also explain why bridge features established based on observations or instructions (e.g., “Chris is presented in a similar colored background as Tall people”) may be akin to those encountered via personal experience. In short, shared feature effects are consistent with functional accounts of relational behavior, particularly with relational contextual cues.

We have distinguished between two outcomes that follow from the sharing of features (i.e., feature transfer and transformation). It may be that feature sharing only leads to a cascade of assumptions about an individual once a particular initial assumption has been made (e.g., “Chris is presented in a similar colored background as tall people” “Chris is tall” “If he is tall then he is also likely to be strong, dominant, and a good leader”). If so, then several questions remain. Why did feature transformation occur due to transfer, and under what conditions will feature transfer or transformation occur independently? Is it possible to influence or guide the number and type of assumptions following feature sharing to influence whether transfer and/or transformation takes place? Answers to these questions await further investigation. ³

The procedures in Experiments 1-3 bear remarkable similarity to those used in the evaluative and attribute conditioning literatures (De Houwer & Hughes, 2016; Förderer & Unkelbach, 2015). So, too, do the feature transfer effects reported in our studies. This similarity in terms of the procedures and outcomes supports the idea that evaluative, attribute, and perhaps other types of conditioning (e.g., fear and disgust) all constitute shared features effects. Specifically, they may represent effects based on the fact that a specific bridge feature (e.g., contiguity) is shared by a source and target object, leading to a single feature being transferred from one to the other. This novel perspective on conditioning effects has heuristic value insofar as it highlights previously unrecognized similarities and differences between seemingly unrelated effects in psychology. For instance, it helps to understand that all conditioning effects are similar in that they involve a transfer of a single feature from a source to a target via the same type of bridge feature (contiguity). Where conditioning effects differ is in the type of feature that is ultimately transferred from source to target (e.g., for evaluative conditioning, it is general valence; for attribute conditioning, it is a specific attribute; and for fear conditioning, it is fear).

A shared features perspective also highlights that many other phenomena studied in person perception may also result from a sharing of features. For instance, certain stereotype and discrimination effects seem to involve changing how a target person is perceived due to the bridge features they share with others. Participants who initially encounter an unkind Black confederate are less likely to sit next to a different Black confederate later (Henderson-King & Nisbett, 1996). It may be that the feature (here: race) shared by a source (here: unkind confederate) and target (here: novel confederate) influence the assumptions that people make about the latter, which in turn influences their subsequent behavior towards that individual. Similarly, during the COVID-19 pandemic, racial discrimination against Asian Americans soared as they were repeatedly connected with the virus and its origins (Croucher et al., 2020). This phenomenon was so prevalent that social media users on platforms, such as Twitter, Instagram, and Facebook, implemented campaigns to undermine this connection (#IAmNotAVirus; Mcguire, 2020). The fact that a target object (i.e., Asian Americans) and a source object (i.e., virus) share one feature (i.e., common geographical origin) may have led to assumptions that they share other features as well (e.g., that the former may possess the contaminated properties of the latter).

As such, conditioning, radiating beauty, guilt-by-association, and other person perception effects seem to be similar in that they all involve a change in how a target person is perceived as a result of feature sharing. However, these phenomena also differ in important ways from one another. Conditioning effects, for instance, rely on a regularity-based bridge feature (i.e., contiguity), whereas other phenomena rely on different bridge features. Sometimes, these may be physical properties (e.g., a shared shape or color) or conceptual properties (e.g., group affiliation, romantic partnership, family membership). Elsewhere, certain person perception effects (e.g., radiating beauty, stereotyping) involve a cascade of multiple features being acquired by the target rather than the mere transfer of a single feature. This discussion illustrates how a shared features perspective offers a new way of conceptualizing, connecting, and organizing many phenomena long studied in the person perception literature.

We can see many questions concerning feature sharing that still need to be explored. For instance, during the training phase of Experiments 1-3, we exclusively focused on altering perceptions of a target’s height. However, the shared features principle is not limited to height: it predicts that feature sharing can shift people’s perceptions of others in many ways, from how guilty and athletic an individual is to how attractive, dangerous, or contaminated they are assumed to be. Future studies could test this idea and demonstrate that feature sharing holds across the manifold attributes people are characterized by in everyday life.

When doing so, researchers may simultaneously show that the outcomes reported in this paper are evident on other measures as well. Take, for instance, the reverse correlation task, a psychophysical paradigm that uses a person’s responses to randomly varying images to construct a classification image, which is often viewed as a visualization of their mental representation of that individual (e.g., Brinkman et al., 2017). It may be that participants do not simply report that the target is taller but actually visualize the target as possessing that feature. Feature transformation effects due to feature sharing may also be physically evident (e.g., a taller target may be visualized as more attractive or less trustworthy than a shorter one). Elsewhere, individuals who share features with contaminated others may be physically avoided on behavioral approach tasks, while people who share physical or conceptual features with ingroup members may elicit stronger altruistic or pro-group behaviors than those who do not. Put simply, future work could test if feature sharing impacts judgments, decisions, emotions, and behavior.

Experiments 1-3 focused on two specific types of sources and targets (i.e., novel individuals). The shared features principle predicts that feature sharing should also influence the assumptions people make about the attributes of other objects, from brand products and social groups to political messages and corporations. It may explain how a brand becomes desirable, a protein bar healthy, or a person famous. Future work could test this idea by examining if feature sharing not only leads to person-to-person feature transformation (as documented here) but also in person-to-group, group-to-person, group-to-group, and other source/target constellations. Similarly, the principle also predicts that feature sharing should lead to not only the acquisition of new features but also the modification of existing ones. The present experiments and those of Hughes et al. (2020) both focused on feature acquisition. Future work could determine whether feature sharing can modify the attributes that people, groups, brands, and other objects already possess. Similarly, research could examine if eradicating the bridge feature (or modifying it) leads to reductions or even the complete extinction of previously established shared feature effects.

The ideas above center on ‘what’ is transferred or transformed when different objects share features. It is also possible to ask questions about ‘how’ feature transfer and transformation take place. For instance, our experiments showed that one type of bridge feature (i.e., color) triggers feature transfer and transformation. Hughes et al. (2020) found that other physical and conceptual bridge features, such as location, size, and symbolic connection, also result in feature transfer. Future work could examine if this is true for other bridge features, especially those central to known psychological phenomena, such as romantic connections (radiating beauty effect), proximity (mere proximity effect), shape, and other physical characteristics (disgust, brand mimicry effects), skin color (stereotyping), facial similarity (discrimination) and co-occurrence (“by association” effects). Replicating our findings with experimental analogs of these bridge features would provide stronger support for the idea that existing phenomena represent instances of the same underlying principle.

We believe that there may be different ‘categories’ of bridge features that lead to feature transfer and transformation. In some cases, the bridge feature may be a physical property of the objects (e.g., source and target share a similar color, size, or geographical origin). In other cases, it may be a conceptual property (e.g., conceptual connections such as familial, social, or romantic relationships). In still other cases, it may be an environmental regularity (e.g., that source is contiguously paired with a target or leads to similar responses as that object; see De Houwer & Hughes, 2020). Bridge features may also be established based on personal experience, observations, or via instructions. Future research could examine if the extent to which feature transformation takes place depends on the type of bridge feature and how that bridge was established.

So far, research on the shared features principle has focused on situations where objects unambiguously share a single bridge feature with others. However, in everyday life, it is more likely that multiple bridge features connect objects, and as such, feature sharing is not a binary (“all-or-nothing”) issue but rather a graded one (i.e., one that requires that we ask how many, and in what ways, do objects share such features). Research could examine if the number, type, and nature of the bridge features involved in a given situation moderate the extent of feature transformation. Previous work appears consistent with this idea. For instance, Alves et al. (2020) found that attitude generalization effects were stronger for stimuli that are distinctly related to one another (share limited bridge features) than those that are not (share multiple bridge features). Moreover, the diagnosticity of features may constitute an important moderator. In our case, we used colored frames as a bridge feature, but other, already existing, bridges (e.g., profession, race, or gender) may be even more relevant and hence dominate other types of less relevant cues.

Finally, it would be interesting to examine how and when feature transformation occurs and whether people, upon reflection, think this is an acceptable reason for their judgments, decisions, and actions. Ratliff (2021) reports that people find it unacceptable to make assumptions about a target person simply because they share features (e.g., musical interests, employment status, gender, race) with a source person (also see Banaji & Bhaskar, 2000). However, there are many other areas of psychology where feature sharing is viewed as an acceptable basis for making assumptions about the target, particularly when the bridge feature is a familial connection, race, or perceived group entitativity (see Ratliff, 2021). It may be that, upon reflection, mere proximity to, or contiguity to a source (e.g., a famous world explorer) says little about who a target person is and is, therefore, an unacceptable basis to judge them. However, the opposite may be true when the target is the best friend or close colleague of that explorer. Relatedly, it would be good to know if feature transformation occurs quickly once a bridge feature has been established, regardless of how acceptable it is perceived to be or despite deliberate attempts to resist its impact on behavior. In short, future work could explore issues related to the automaticity of shared feature effects and their perceived acceptability.

Our work has several limitations. One concerns the number of bridge features present during the training phase. Targets and sources not only shared a common color but were also presented contiguously with one another. It may be that the effects reported here represent the combined impact of multiple bridge features (i.e., color and contiguity). It is worth noting that Hughes et al. (2020; Experiment 8) controlled for this possibility, manipulated one bridge feature (location) independent of another (contiguity), and still observed shared feature effects. Future work could explore if the same holds here.

The trained attribute of “tallness” in Experiments 1-3 tends to be evaluated positively, whereas “shortness” is evaluated negatively (Jackson & Ervin, 1992). It may be that the two targets became positive and negative following training rather than tall and short and that their valence “radiated” or generalized to other positive and negative traits (e.g., if the target is positive in one way [tall], then they may have been viewed as positive in other ways [strong, dominant, leader]). Something similar could have also occurred for the short stimulus but in the opposite direction (negative in one way [short] leads to negativity in other ways [submissive, follower, weak]). If so, the outcomes in Experiments 1-2 may represent general evaluation changes rather than specific attribute changes. Experiment 3 provides some evidence against this line of reasoning. In Experiment 3, the tall target was perceived as being more dominant (positive) and less caring (negative) than the short target, who was perceived as more truthful (positive) and less of a leader (negative) than the former. It seems that one target was perceived in ways that were more positive than the other target along specific attribute dimensions and also more negative than the other along other attribute dimensions, suggesting that our findings reflected changes in attributes rather than mere evaluations. Future work could offer an even more robust control for this possibility by re-running Experiment 3, have participants rate the targets in terms of their attributes and valence, regress the attribute assessments onto the valence ratings, and analyze the residuals of this regression (i.e., variance for the attribute ratings independent of valence ratings; see Högden & Unkelbach, 2021). Showing that the attribute effects reported here are still present even after statistically controlling for valence would provide strong evidence that those effects cannot be accounted for by valence alone.

Another possibility could be to have a continuous score of relatedness for each target attribute (i.e., how related the attribute is to height) and a continuous score of the attribute’s valence (i.e., how positive vs. negative the attribute is). If the attributes’ relatedness moderates the shared feature effect (but not by the attributes’ valence), this would mean that this effect depends on how closely the source and target attributes relate (for a similar approach, see Rougier et al., 2023). Such a continuous relatedness score would be more fine-grained than the ‘related’ vs. ‘unrelated’ categories.

As Figure 1 suggests, our height manipulation was also confounded with gaze direction (i.e., tall person looking down) and picture size (i.e., tall person was bigger). The size of the image and gaze direction could have led participants to make additional assumptions (e.g., that the tall person is more powerful and more dominant; see Lamer & Weisbuch, 2019). Indeed, a person looking down could be seen as feeling superior to others, while bigger pictures are often more salient and therefore potentially more influential than small images. Hence, tall stimuli could have had a larger impact than the small stimuli. Importantly, however, these confounds cannot explain the shared feature effect itself (i.e., the fact that the target person’s height perception was influenced by the source person it shared the colored frame with). Similarly, the confounding with gaze direction could explain the negative effect on warmth-related traits, but it cannot explain the larger effect on competence-related traits (Experiment 3).

Finally, we did not observe effects for leadership in Experiment 1 but did observe them in Experiments 2-3. Two possibilities come to mind. It could be that leadership effects did not emerge in Experiment 1 due to that study having slightly less power than Experiments 2 and 3. Alternatively, changes in the face stimuli from the former to the latter studies (i.e., selecting a target stimulus closer to the scale’s mid-point) helped us detect the generalization of the shared feature effect. Relatedly, in Experiment 3, we observed effects for the truth and caring attributes, such that, despite being viewed as unrelated to height during pilot testing, participants rated the short target as more truthful and caring than the tall target. Such an unexpected outcome may have been due to the within-participant design employed in our studies and more specifically, to a compensation effect between competence- (leadership, dominance) and warmth-related traits (truthfulness, caring). After rating one target (Chris) on the ‘competent’ dimension, which is stereotypically related to tallness (e.g., “Chris is a leader”), participants may have then compensated by saying that the other target (Bob) might not be competent but rather trustworthy and caring (i.e., ‘warm’). This idea is consistent with past work on the compensation effect. When people are given information about two groups or individuals (e.g., “Group A is competent and Group B incompetent”), the competent group is perceived as less warm than the incompetent group. People seem to compensate for the difference observed on one dimension (e.g., competence) on another non-manipulated one (e.g., warmth; see Judd et al., 2005). According to Judd eand colleagues, the compensation effect should be weaker or even absent without a comparison context (i.e., when people learn information about one group but not the other). If the unexpected findings in Experiment 3 were due to compensation, then they should disappear when height is manipulated between participants. Future work could examine this possibility.

The current paper extended the shared features principle to the domain of person perception and shows that when a source and target individual share a bridge feature, feature transfer and transformation may occur, shifting the assumptions participants make about the target along multiple dimensions. These effects were strong, reliable, and replicable. Although we focused on just one transferred feature (i.e., heights) and set of objects (i.e., two individuals), our conceptual account applies to many more and offers an overarching way of describing and analyzing shared feature effects throughout psychological science. We hope that the perspective outlined here will stimulate research on feature sharing and contribute to a broader and deeper understanding of how people generate and revise assumptions about other people’s traits and attributes.

The authors have no competing interests to disclose.

Contributed to conception and design: SH, JDH, CU, MR Contributed to acquisition of data: SH Contributed to analysis and interpretation of data: SH, MR Drafted and/or revised the article: SH, JDH, CU, MR Approved the submitted version for publication: SH, JDH, CU, MR

The pre-registration files, deviations from pre-registrations, materials, data, and analytic scripts for all experiments can be found at https://osf.io/2ps4n/.