Social Visual Perception: Exam 3

Do variations in preference for sexual dimorphism in faces across cultures seem to be due to random chance?

Somewhat; it doesn't seem that the preference for sexual dimorphism is cross-cultural, but it does seem that the preference has a pattern. That is, these data don't support the idea of sexual dimorphism as an ancestral cue to mate value. However, the preference isn't random across cultures. This may be due to the fact that facial dimorphism is greater in more developed environments, and this may influence face preferences, or people in more urbanized environments encounter more unfamiliar faces, so they may rely more strongly on judgments of appearance. The relationship between masculinity and prestige may be stronger in more developed environments, which would make masculine faces look more attractive. Preferences for facial dimorphism may not have played a role in human society until recently. 1. Scott et al. (2014): 12 groups, women judged male faces and vice versa (maybe more masculinized and feminized male faces for women as well as maybe a neutral face). Evolutionary prediction: preference for sexual dimorphism should be stronger for less developed groups (disease is more prevalent, so health indicated by sexual dimorphism is more important). Looked at whether years lost to disease, homicide rates (as a measure of male intra-sexual competition), and urbanization could predict a preference for sexual dimorphism. ==> greater urbanization predicted a stronger preference for feminized female faces (male preference for female faces). Less of a disease burden was related to increased preference for masculinized faces by women (years lost to infectious disease correlates but negatively with masculinity preference). Cross-cultural agreement in associating masculinity with aggressiveness (female participants, I think, were shown the same composites from before and asked which looks the most aggressive), but degree of urbanization (proportion of the population that was urban) predicted a stronger perceived relationship.

Remember to look at the articles we read for this test/class!

Y

In Ballew and Todorov (2007), what study design did the researchers use to demonstrate that trait judgments based on appearance can predict election outcomes?

-IV: "predict" = a greater proportion (percentage) of winners were chosen as competent than chance (0.50) -Experiment 1: are competence judgments made after 100ms predictive? -89 races (winner and the runner-up) -3x2 = Three time conditions (100 ms, 250 ms, unlimited time) and two positions of the images (counterbalancing: left and right); Three judgments: binary choice, 9-point scale for the more competent face, recognition; For most conditions: shown a fixation cross, grayscale picture of winner & runner-up, gray mask -Results: RT: unlimited time condition > 100-ms, 250-ms conditions; Accuracy of judgments did not differ in the three time conditions; Just 100-ms exposure was sufficient to make "accurate" predictions; = simple, fast judgments: sufficient to predict the outcomes of the election! -Experiment 2: Does deliberating affect the predictive accuracy of competence judgments? -Three instruction conditions x counterbalanced (6 conditions): forced-choice; 250-ms condition, 2 second response deadline condition, Deliberation condition -55 races: gubernatorial candidates are same sex and ethnicity -Results: The winner was more likely to be selected as competent than the runner-up --> Judgments: 250-ms condition ≈ 2 second response deadline, 250-ms condition & 2 second response deadline > deliberation; Proportion of elections in which winner was chosen as more competent than the runner-up differed significantly from chance (0.50): Only the 250-ms and 2 second response conditions, NOT: deliberation condition; Correlation in competency judgments between unreflective judgments (250-ms & 2 second response condition) and deliberation condition-->Partial correlations showed that it's due to rapid, not deliberate judgments; Unlimited time condition (Experiment 1) and deliberation condition (Experiment 2) were slower than the other two conditions. BUT: Predictive accuracy was reduced not by time but by act of deliberating. Therefore, unreflective judgments were made in the unlimited time condition. -Experiment 3: not retrospective, but PROSPECTIVE judgment! Would competence judgments based on facial appearance predict the election outcomes? -2006 gubernatorial and Senate races -Only one condition: Same as Experiment 1's unlimited time condition (told to use gut feeling) -Results: Was combined with Experiment 1 (same procedures) in order to increase the number (35 + 89 gubernatorial races = 124); More likely to choose the winner as more competent; For both elections, competence judgments predicted 68.6% of the gubernatorial races and 72.4% of the Senate races; Incumbency: if the incumbent won again, this may because he was actually competent in office. However, this hypothesis turned out to be false. Competence judgments did not predict the winner only when the incumbent wins.

Why is it important to consider base-rates when examining the accuracy of judgments based on facial appearance?

1. Accuracy with facial cues should be compared to accuracy using other cues that people typically have access to (like base rates). Appearance causes people to neglect base-rates. When both appearance and base rates are available, accuracy appears to drop below what would be achieved by base rates only. The small benefit of visual appearance is offset by the large cost of ignoring base-rates. Only when base rates are close to chance does appearance-based information improve accuracy. 2. Would have done better using base rates/not seeing the photos than appearance, only when base-rates neared 50% did judges perform as well as survey respondents: "What's My Image" website A. Website users posted photos, selected possible characteristics from a list, indicated where they fell (e.g., sexual orientation, use drugs, own gun, college degree) B. Other users presented with random photo and characteristic, got points for accuracy (Does this person use drugs? Yes/No) C. DV: website performance (mean accuracy of judged on website), dominant base-rate (accuracy level judges would achieve by guessing most frequent category on each trial), survey performance (separate group of undergraduate participants reported categories they believe to be most frequent among American adults, accuracy level that could be achieved by relying only on prior beliefs) D. Results: Accuracy > chance, but worse than using base-rates. That is, accuracy of the dominant base rate was higher than accuracy for website performance except for college degree (age-related cues), survey performance ≤ dominant base-rate; website performance ≤ survey performance, except fist fight, long term, & college degree 3. Manipulated the base rates: "Political Guessing Game" website <-- same study as in #2 A. Shown images and asked if Republican or Democrat (House of Representatives) B. Authors varied the proportion of Democrats C. Dominant base-rate should be > probability matching in accuracy D. IV: told the proportion; told the proportion might very E. Results: accuracy was above chance, when base-rates were known, participants were more accurate than those for whom the base-rates weren't known; Especially on the outer ends of the proportion of Democrats: probability matching & dominant base-rate accuracy > base-rates not known & base-rates known

What are two types of evidence that demonstrate the robustness of consensus judgments based on facial appearance?

1. Adults show a high level of agreement in trait judgments based on appearance (in class: .59 for trustworthiness, 1/3 of variance came from shared perceptions) 2. Seems to span over development: Children approach adult-like consensus early in life: IV -- high and low faces on trustworthiness, dominance, and competence. DV -- Which of the faces is nice?. All age groups showed significant consensus in their rating of facial appearance as compared to chance (50%). Younger age groups were slightly less consistent than adults, but not by all that much (3- to 4-year olds (don't know if sig) versus 5- to 6-year-olds versus 7 - to 10-year-olds ≈ adults). Computer-generated and adult-like faces, though, so not as realistic. 3. Even with naturalistic stimuli, children and adults showed consensus for all types of stimuli. Judgments made in childhood are robust enough to generalize to a variety of stimuli (all were above chance, can judge these traits in those faces). A little confused with what is highlighted in blue. 4. Speed of trait judgments/how well judgments made after a brief exposure correlate with judgments made after a longer period of time. IV: neutral faces, exposure times (100, 500, or 1000 ms, unlimited time). DV: attractiveness, liking, competence, trustworthiness, aggressiveness (Is this person competent?). Controlled for attractiveness judgments. After 100 ms, trait judgments were already highly correlated with unlimited time judgments, and not just due to attractiveness (halo effect). All correlations were significant (controlling for attractiveness). No sig difference in increasing time. 5. Subsequent work has shown that people can form impressions even after 34 ms.

If you are investigating the features underlying trait judgments from appearance, what are the drawbacks of manipulating single features?

1. Although you may find an effect by manipulating certain features, you can't conclusively say that this is the effect. That is, you may get larger effects when altering other features. That is, you can't really say that eyebrows are important for perceiving this one trait, because other features may actually affect it more 2. When manipulating one feature, you're also altering other features, and what effect you argue is due to the eyes, for example, may actually be due to the other features that were altered in the process. That is, given the holistic nature of face processing, it's reasonable to assume that features will influence one another. 3. What counts as a feature? Do we even have a good underlying definition/one we can agree upon? For example, in the paper I read, lip heighth, eye heighth, lip width, and eye width were altered. Are the features the lips and eyes, or the height and width? 4. If you manipulate more than one feature, the number of possible combinations grows rapidly. (10 binary features = 1024 possible combinations). A study would most likely not have enough subjects to test these combinations in a meaningful way.

How is studying the consequences of trait impressions different from studying the accuracy of those impressions?

1. Because people show agreement in their trait judgments based on facial appearance, trait judgments have a variety of consequences. However, the fact that trait judgments have consequences don't mean that they are accurate. In fact, there is very little evidence of accuracy. My ideas: consequences have more to do with cause and effect, whereas accuracy has more to do with being right. One could argue that accuracy actually matters less than the consequences. I feel like there's more to say here 2. Maybe there are more consequences for trait impressions than accuracy, especially for traits that take awhile to judge in a person (like trustworthiness); work about accuracy has looked at more objective characteristics (political orientation, sexual orientation, etc.) 3. A subsequent study found that a CEO's appearance didn't relate to the CEO's actual performance (kind of goes against the findings in the other study, but also shows that maybe more successful companies hire certain CEOs or, especially, that maybe they just take "more powerful" photos) 4. Trait judgments predict outcomes in a number of domains, including: A. Mate choice 1. "What is good is beautiful: Face preference reflects desired personality" by Little and colleagues (2006) B. Leadership 1. Business 2. Politics C. Trust, guilt, and criminality 1. Finance 2. Law

How does the informational diagnosticity reverse correlation approach work? What kinds of questions has it been used to address?

1. Reverse correlation: response attributes are fixed, but stimulus attributes are random 2. Participants see images that are good exemplars of the categories of interest (happy face, sad face). See different faces versus the same base face. The images have a randomly generated mask superimposed on the, which only shows parts of the original image. Participants categorize the faces with these randomly generated masks (categorize each face as happy or sad, for example). The parts of the face that are visible that are required to make a correct categorization are counted as having diagnostic value. Add noise from trials where participants performed task correctly. Divide by total noise. Put this over a face. This mask will show which parts of the face were important for correct gender or emotion categorization (if no regions have special status = mask would be uniform; important regions = less noise) 3. Different from eye-tracking method: one of them uses a shorter time course. Not given the whole image with informational diagnosticity form of reverse correlation. May be looking at a different stage of processing than the eye-tracking method. 4. Study: gender & expression: saw male or female faces w/ neutral or happy expressions. Enough noise = performance kept at 75% correct (don't want a ceiling effect). Need mouth for expression and mouth and eyes for gender. Computers need more, like hair to classify gender. 5. Identity: none of the very low spatial information is important. Researchers took diagnostic information and filtered the faces (w/ diagnostic info or opposite diagnostic info). Participants showed better recognition for faces with diagnostic versus non-diagnostic info. Held true for another set of faces. 6. Questions it addresses: What visual information/which visual areas of faces are important for making certain judgments? What visual information do participants rely on to tell visual categories apart -- you need the categories to be present in the task? What parts of the face are important for correctly identifying gender and expression? What about compared to what areas computers need? Something about identity.

How have researchers demonstrated that trait judgments based on facial appearance have real- world consequences?

1. Benefits of real-world studies: can be used to predict important outcomes (maybe we're using the term "predict" very loosely here)/has real-world consequences and may be more ecologically valid. 2. Weaknesses: not manipulating anything (measure appearance, measure outcomes). With this observational data, one cannot determine causality. Like in the finding with the CEOs, correlational data can go either way. 3. In the studies from the laboratory settings notecard, these findings also have real-world consequences. 4. Business setting: Ambady and Rule (2008) IVs: pictures of CEOs from highest 25 and lowest 25 ranked companies of Fortune 1000. Excluded the one female. DVs: rated on competence, dominance, likability, facial maturity, trustworthiness, global leadership ability ("How good would this person be at leading a company?"). Factor analysis --> power (competence, dominance, maturity judgments); warmth (likability, trustworthiness). Remember: these are ratings based only on the appearance of the CEOs. Controlling for age, affect, and attractiveness: company profits were sig correlated with: power composite, leadership; company revenues were sig correlated with company profits <-- appearance predicted success despite uniform appearance of CEOs (male, ≈age, Caucasian). 5. Ballew & Todorov (competency) 6. Different cultural experiments (predicting Bulgarian elections, captain of your boat, Korean vs. US elections) 7. Look at the other notecards 8. Number of studies claiming that people can judge a variety of internal traits based on appearance: political orientation, criminality, sexual orientation

What is binocular rivalry? What is continuous flash suppression and how is it related to binocular rivalry?

1. Binocular rivalry: Rivalrous images = your awareness switches back and forth between two percepts. Image is presented to one or the other eye. Which image is presented to each eye is counterbalanced. See one or the other image (kind of, at least for what I saw with the task), which switches on the order of seconds. That is, one image is dominant, and the other is suppressed. However, what we are interested in is that the properties of the image affect the likelihood that you'll see it. A. Ways to create binocular rivalry: 3-D glasses (one side is blue, the other red), mirror stereoscope (each eye sees a different image). B.Neural mechanisms: competition in low-level visual areas (neurons receiving input from one eye inhibit neurons receiving input from the other, until they become fatigued so that your perception switches to seeing the other image); occurs through competition between higher-level patterns, instead of competition between the eyes. There is evidence for both views, and a hybrid view holds that binocular rivalry occurs at multiple stages along the visual pathway. 2. Continuous flash suppression: an image is presented to one eye, and a rapidly changing pattern to the other. The pattern initially dominates visual awareness, and the other image only slowly emerges into awareness. Amount of time it takes for an image to break suppression is thought to index the degree of unconscious processing. A. Upright faces will emerge into conscious awareness more quickly than inverted faces. 3. Both are used to study unconscious processing. CFS is a variant of binocular rivalry. Certain images/aspects dominate, which shows when, how, maybe even why our awareness occurs and also unconscious processing. To me, it seems that if certain aspects of images make them emerge faster than others, this indicates that something is occurring "behind the scenes" to make this occur. Both can tell us about general cognitive processes (multisensory integration, attention) and face processing (face inversion, emotion, eye gaze)

How have researchers examined the consequences of trait impressions based on appearance using experiments in the lab? What are the benefits of this approach?

1. Economic trust game (amount tripled, different people they thought they were playing with, computer model, invested more pounds (42% more) in trustworthy than untrustworthy faces, after seeing "randomly selected" info from previous rounds, invested more in trustees with good versus bad histories, but invested average of 6% more in trustworthy appearance trustees than untrustworthy appearance trustees) 2. Jury member: severe or less severe crimes + trustworthy or untrustworthy faces, given evidence (5 ambiguous, four increasingly incriminating, 1 exonerating), guilty/not guilty decision after each piece of evidence, indicate confidence; severe crimes: less evidence for untrustworthy (saying guilty), petty crimes: trustworthiness doesn't really matter, latency to guilty verdict was longer for trustworthy than untrustworthy faces in severe crime condition but was longer for both than in petty crime condition, more confident that untrustworthy-looking person committed the severe crime, trustworthy > untrustworthy for petty but may not be significant 3. Voting study 4. Republican appearance study 5. Ballew & Todorov (competency) 6. Rule & Ambady (2008): judge sexual orientation from male faces, 50% had heterosexual orientation, experiment 1: images from personals website, experiment 2: images from Facebook, 3 presentation times: 10,000 ms, 6500 ms, 1500 ms, 11 ms, 50 ms, 33 ms; accuracy was above chance except when faces were shown for 33 ms, corrected accuracy (A') > uncorrected accuracy (% correct) except for presentation time of 33 ms 7. Benefits: can manipulate appearance, measure outcomes, can determine causality. Drawback = less ecological validity. Give more.

What is one piece of evidence supporting the idea that beliefs about specific individuals influence trait judgments based on facial appearance?

1. Faces paired with info about behavior changes trait judgments of those faces (Todorov & Olson, 2008): learning (face w/ description), trait rating task ("How trustworthy is this person?") 2. Deep learning (knowing person for a long time and learning about them over a long time) generalizes to faces = Kraus & Chen (2010) (faces that resemble significant others): survey about sig other (+ and - descriptors), rated a series of faces to how much they resembled sig others, came in 2 weeks later, given a description (good & bad) of person, person resembles sig other or not (yoked), rated person on 15 traits (8 came from description of sig other) ==> rated novel other as more likely to possess traits of sig other, also evaluated this person more positively (than photo resembling yoked) 3. Learning in the lab about specific people also generalizes to the evaluation of novel (similar) faces (despite the face that learning in an experiment pales in comparison to learning in everyday life and despite the fact that participants treated morphed faces no differently than novel faces): level of similarity is substantial but still difficult to detect at 20% and 35% of learned face (morph), learns a face then sees 4 faces (1 is either 20% or 35% morph), DV: is the face the learned face, similar to the learned face, or a new face? ==> percent response was learned > similar for learned faces, and new > similar > learned for 35% morph, 20% morph, and novel faces (maybe so they can continue experiment?); 9 learned faces paired with negative, neutral, or positive information ("How trustworthy is this person?" -- maybe asked just with the face instead of when seeing face and description???) ==> evaluation of learned faces is positive > neutral > negative, similar findings for 20% and 35% morph (looking only at those participants who rating positive faces > negative faces), although not significant for 20% condition and maybe only for neutral > negative for 35% morph but not sure (shows a learning generalization); only participants who rated positive > neutral > negative, similar results as above for 20% and 35% morphs (evaluation of morphs)

How does the internal representation reverse correlation approach work? What kinds of questions has it been used to address?

1. Same face throughout. Superimposed with noise (added or subtracted). Base image, classification image, image and maybe anti-image with the noise superimposed on it, median subject images. You need an ambiguous image (a face that is ambiguously happy/sad, for example, like a morph of a happy and sad face). The noise allows the participant to project an internal mental representation onto that image. Take the noise from the responses (happy and sad responses, for example), and put it over the base face. 2. Study: Mona Lisa. Shown the portrait superimposed with noise and asked to rate how happy each image looked. Noise for the trials chosen as "happy" was averaged, and the same thing was done for those chosen as "sad." Will we be able to see a change in expression across trials? Found that the mouth is important for changing her expression. The mouth seems to make the eyes be perceived as different (same top half with different bottom halves versus same bottom half and different top halves -- no difference). 3. Study: Dutch participants' mental representations of Moroccan faces. Also completed an IAT (examining automatic negative versus positive associations with Moroccans) in addition to the reverse correlation task. Which face looks more Moroccan? Also asked to choose the more Chinese face (used the same base face). Separate group of participants rated the classification images. Mean criminal rating: high prejudice (positive) > moderate prejudice (+) > low prejudice (-); mean trustworthy rating: low prejudice (+) > moderate prejudice (-) > high prejudice (-) 4. Study: Whether mental representations of Mitt Romney differ as a function of political attitudes: tested soon before 2012 presidential election. See 2 faces. "Which looks more like Mitt Romney?" Participants who identified as more Republicans generated more trustworthy representations of Mitt Romney's face (have more trustworthy mental image of Romney). 5. Study: Participants were asked, "Which face is more trait X (trustworthy, untrustworthy, dominant, submissive)?" We see that the faces really do look more trustworthy, anti-trustworthy, etc. The luminance values of the pixels making up the classification images were correlated as a measure of image similarity. Classification images from one side of a trait dimension are different from those at the other side. That is, classification images visualized the intended traits. However, these two dimensions do overlap (correlation ≠ 0). It's also interesting how, on the mean trait rating scale for trustworthiness ratings, trustworthy > untrustworthy and submissive > dominant. For the dominance dimension (mean trait rating scale (dominance ratings)), untrustworthy > trustworthy and dominant > submissive. 6. Questions: What do mental representations of each emotion look like?. Male/female. Tom Cruise/John Travolta (Does this face look more like Tom Cruise or John Travolta?). What makes the Mona Lisa smile? Can be used to study internal representations of gender, emotion, and identity. Can also be used to examine mental representations of outgroup faces (Dutch participants' representations of Moroccan faces, using the IAT). Mental representations of political candidates depending on your own party. Does our mental representation of certain traits actually match up to when those mental images are classified?

In Flowe and Humphries (2011), what role was criminal appearance found to play in lineups? What are the real-world implications of this finding?

1. Study 1 a.Undergraduates b. Conditions i. Description: used a single eyewitness description (randomly selected if there was more than one) w/ an average of 6 physical characteristics (included in the checklist are height, weight, age, race, hairstyle, and eye color) and one other trait (traits from the checklist include perpetrator's temperament, cleanliness, sound of voice, etc.) ii. No description c. 11 lineups from police arrest files w/ 6 people in each i. Different lineups with different races (I think all the same race in the lineup): White, Black, Hispanic d. DV i. Asked to determine for each lineup which person is the police suspect. ii. Then: reported in a free response format why they had selected that particular member iii. In description condition: asked if had used eyewitness description in making choice e. Results i. Identification rates for each face across the two conditions were statistically independent. That is, the rate of choosing a given face varied depending on whether mock witnesses had a description of the culprit. ii. Race of the lineup members didn't influence any of the measures. iii. Suspect identification rates didn't significantly change when mock witnesses were provided with a description of the culprit. iv. In 2/11 of the lineup in the description condition, mock witnesses identified the suspect at a rate above chance expectation. This was the same for the no description condition, although the suspects were different. v. The majority of the lineups were fair based on functional size and Tredoux' E, traditional measures of lineup bias and size. vi. No description condition 1. Criminal appearance was most often reported as the primary reason for mock witness choices 2. Criminal appearance was reported as the basis for the decision significantly more often in this condition than the description condition 3. Significantly more likely to use guilt as the criterion for picking the witness vii. Description condition 1. Primary reason for the choice was: physical appearance of the lineup member matching the description viii. Reporting of guilt and criminality were independent of one another (never used both to make their selection) ix. The rate at which a face was identified based on criminal appearance didn't differ depending on whether the face was a suspect or a foil face. Based on the self-report data then, the suspects did not appear to be more criminal looking than the foils. x. Summary: the sample of lineups seemed to be fair based on traditional measures of lineup fairness and suspect bias. 2. Study 2 a. 69 undergraduates b. Conditions i. Rating conditions: criminality ("the extent to which the face resembled a criminal = someone who would break the law"), typicality (rate "typicality of the face, or the extent to which the face would resemble other faces"), distinctiveness (rate "the distinctiveness of the face, or the extent to which the face would stand apart from other faces.") ii. Rate the physical similarity of the lineup members: two photographs from a given lineup were presented side-by-side. 165 pairwise evaluations across the 11 lineups. Determine the extent to which the face pairs are "physically similar in appearance to one another." Use only physical attributes (facial features, complexion, hairstyle, hair color). Images were in black or white, like in study 1. c. Faces from the lineups in study 1 were used. Individual faces were displayed in random order. d. Used a scale (0-100) below the face: clicked on the portion of the scale that corresponded with their rating. Told to focus on the physical appearance of the person in all of the ratings tasks. Told to ignore clothing, focal size, picture size, etc. e. Results i. Typicality & distinctiveness ratings were significantly associated. Therefore, distinctiveness was entered in the following analyses. ii. Distinctiveness and criminality were significantly associated. iii. No description condition 1. Criminal appearance ratings were significantly associated with mock witness identifications 2. Distinctiveness and similarity were not associated with identification rates 3. I think only for this condition, but maybe also for description condition: criminal appearance was significantly associated with mock witness identifications after controlling for distinctiveness, as well as after controlling for lineup member similarity 4. Rate at which the suspect was identified was significantly related to his criminal appearance ranking. Not with distinctiveness and similarity. 5. Proportion of witnesses indicating they had used criminality as a criterion was significantly related to the independent ratings of criminality and distinctiveness a. Not the case with similarity and its relation to self-reports of having using criminality 6. Proportion of witnesses basing their decision on guilt was negatively associated with distinctiveness ratings, but similarity and criminality weren't related to guilt iv. Description condition 1. Face ratings weren't significantly associated with identification outcomes 2. None of the rankings were associated with suspect choice rates. 3. Proportion of witnesses indicating they had chosen the face because it appeared criminal was unrelated to any of the independent face ratings. 4. Faces that were chosen on the basis of guilt tended to look more similar to the other lineup members, but criminal appearance and distinctiveness were not related to guilt. 3. Real-world implications a. Criminal appearance is a unique dimension of faces that might be separable from guilt b. Eyewitnesses may use criminal appearance as an alternative identification strategy when they have no information about the culprit's appearance and yet are required to pick someone from a lineup. c. In Study 2, faces rated relatively high with respect to criminality were chosen more often in the no description condition; similarity and distinctiveness ratings were unrelated to lineup choices in the absence of a physical description. Ratings of criminality, distinctiveness and similarity were not related to mock witness choices in the description condition. d. Therefore, criminal appearance should be taken into account along with the physical appearance in constructing lineups. In more than half of the lineups sampled, the suspect was rated as the most or second most criminal-looking member. This is especially worrisome considering that the lineups were taken from real lineups. e. When the other lineup members are police officers, public volunteers, or taken from driver license databases, this may be problematic because people may be more likely to display the emotions and physical traits that are associate with criminality then they are under arrest. f. Only criminal appearance was positively associated with mock witness identifications in the no description condition. Thus, measuring solely the distinctiveness and/or the physical similarity of the lineup members may be inadequate as a means of determining whether a lineup is biased for an eyewitness who ahs no memory for the culprit but who is willing to guess. g. Guilt may be a construct separable from criminality. h. Question to myself: why are lineups used?

What evidence suggests that the influence of trait judgments depends on context?

1. Finding about trustworthy versus untrustworthy faces for different types of crimes: Implications for legal decision-making a. Perceived facial trustworthiness influenced the amount of evidence required for a conviction and confidence in that decision (Porter et al., 2015) i. Severity may exacerbate reliance on first impressions b. Judges and jury members rely on more than the evidence in front of them in reaching a verdict (These findings are troubling because judges and jury members aren't solely given descriptions, but they also see the person (I think)) i. For instance, examination of an individual's demeanor is considered important to assess credibility ii. However, this work shows that unchangeable aspects of appearance can also influence decisions; Perhaps education would mitigate this effect 2. Economic decisions A. Participants invested an average of 43.69 virtual pounds in the untrustworthy faces and 61.91 virtual pounds (42% more) in the trustworthy faces B. However, participants invested more in trustees with "good" versus "bad" histories 1. Main effect such that virtual pounds was higher for the "good" history trustees than the "bad" history trustees for both untrustworthy and trustworthy faces 2. Participants also invested an average of 6% more in trustees with trustworthy vs. untrustworthy appearance 3. Especially with the "bad" history trustees, but it also looks like maybe with the "good" history trustees, more virtual pounds were invested in the trustworthy-appearing trustee than the untrustworthy-appearing trustee 3. Voting (Bush & Kerry) A. IV: two faces (used the shape of Bush and Kerry's face??? Maybe added it to a morph?); questions (Which face would you vote for to run your country? Which face would you vote for to run your country in a time of war? Which face would you vote for to run your country in a time of peace?). Also looked at morphs that were more masculinized and feminized (not Bush or Kerry face) <-- data were collected prior to 2004 general election B. DV: Percent vote (& which face they chose) C. Results: Vote: Bush > Kerry, War: Bush > Kerry, Peace: Kerry > Bush; Vote: Masculinized > Feminized (may not be sig.); War: Masculinized > Feminized; Peace: Feminized > Masculinized. Therefore, perceived masculinity may have been one of the underlying reasons for the pattern of results with Bush and Kerry. 4. Republican appearance (looking like a Republican predicted a candidate's success, but only among Republican voters; there was no correlation betweeen looking like a Democrat and trait judgments) IV: pairs of rival political candidates DV: judge which is the Republican ("Which person is the Republican candidate?"); which person they would prefer as their leader Results: (all correlations) Republican participants preferred Republican-looking faces; Democrats didn't show a sig preference for Democrat-looking faces (don't seem to look politicians who look like politicians); as the Democratic candidate vote-share increased, the likelihood of the Democrat being misidentified as Republican decreased 5. Participants showed less of an effect of appearance on judgments when they also had reputation information

How did Oosterhof and Todorov (2008) decide which traits to model? Which traits did they choose?

1. How they decides which traits to model (they wanted to identify the dimensions that underline the judgments): A. Asked participants to generate free, unconstrained descriptions of neutral faces B. Collected judgments of faces on the most frequently mentioned traits (the traits accounted for 68% of all descriptions: attractive, unhappy, sociable, emotionally stable, mean, boring, aggressive, weird, intelligent, confident, caring, egotistic, responsible, trustworthy, + dominance because it's important in social psychological models of person perception) C. Principal components analysis (PCA) on the trait ratings --> PCA looks to find the dimensions (factors) that underline the variability in judgments D. Found "evaluation" and "dominance." This suggests that faces are rated on these two primary dimensions. E. Chose trustworthiness and dominance because they correlated highly with those in D. That is, Oosterhof and Todorov (2008) used trustworthiness and dominance judgments to approximate these dimensions. F. The following may be more information than is necessary to answer this question, but they used a face model to create 300 emotionally neutral faces and asked participants to rate those faces. G. Had different subjects rate the faces along the trustworthiness and dominance dimensions. H. Created vectors in face space whose direction was optimal for changing trait judgments (created vectors that best described change in the trustworthiness (or dominance) dimension). They did this by weighting the existing face space dimensions in a way that best described variability in the traits. For example, if trustworthiness was strongly related to eye size, then the trustworthiness dimensions would correlate. I. Todorov et al. (2011) validated the models. Dominance judgments = good correlation with dominance dimension. Trustworthiness judgments = correlation, but not as linear (curves at the upper end = not as good at discriminating with the upper trustworthiness faces) 2. Evaluation (valence (positive/negative)) -- accounted for 63% of the variance in trait judgments; dominance (18% of the variance; dominance, aggression, and competence correlated most highly with this component). However, the authors ended up choosing trustworthiness and dominance because these two traits correlate highly with the first two principal components in the PCA analysis.

How good are people at evaluating their own ability to draw inferences from faces? How does this finding relate to the study of accuracy of impressions based on appearance?

1. Not very good 2. Occupation study A. IV: pairs of faces, matched on gender and age; different occupations B. DV: Which of the two men above is a (psychologist)?; indicated confidence in their decision (.5 = guessing; 1 = total confidence) C. Results: confidence > accuracy; also true in a 2nd study where pairings were random instead of chosen by experimenters 3. Using these cues effectively requires knowing when not to use them. If one were to know the base-rates, and if he/she knew the base-rates were low, he/she might know not to use his/her own inferences. Here, too, appearance seems to weight out the actual accurate answer (?). If people are more confident than their accuracy really is, this is concerning because their accuracy is also relatively poor. That is, say with a police lineup, people may be very confident about who did the crime, but their accuracy may actually not be all that high.

What was the practice of physiognomy, and how is it different from modern research examining trait judgments based on facial appearance?

1. Physiognomy: the assessment of someone's character based on their appearance, particularly their facial appearance -Aristotle (people, animals) -Giambattista della Porta (Renaissance, temperament & animals) -Johann Kaspar Lavater (18th century Swiss pastor, facial features & traits, special talent) 2. Influence of physiognomy: art (portraits) & literature (19th century authors -- Charlotte Bronte, Charles Dickens, Edgar Allen Poe), Darwin, Sir Francis Galton (composites to pick up on certain character traits) 3. Discredited in 20th century science, but we still are influenced by appearance 4. Differences: not always one feature(s), but holistic face, more scientific/statistical (we have data -- tested empirically), not saying that the features predict the actual behaviors/not assuming accuracy (at least I don't think so), just that certain faces look certain ways, don't need a special talent for the modern work, but rather the opposite, not thinking about the similarity to animals,

What is one piece of evidence supporting the idea that beliefs about social groups influence trait judgments based on facial appearance?

1. Reverse correlation task with Moroccan faces: beliefs about this social group influenced how people created the faces, which influenced others' trait judgments of the faces 2. Trustworthiness evaluation (Stanley et al., 2011): IV: 100 Black, 100 White, 91 other race faces, IAT score (pleasant/unpleasant words, Black/White faces) -- difference in reaction times between the pairings is the IAT effect DV: rates on "How trustworthy is this person?" (1-9); = = > as implicit Pro-White bias increases, trust disparity tends to increase (more trusting of White versus Black faces), similar with implicit Pro-Black bias but with Black versus White faces; this relationship held after controlling for explicit attitudes and participant race 3. Part of same study --> Economic trust game (replication): DV: $ to invest (quadrupled before reaching partner, which partner previously decided to return half of or not, paid based on 3 randomly selected trials), offer disparity; IV: what picture they see, IAT score ==> similar correlation as above, but offer disparity is on y-axis 4. What this shows us: implicit biases predict evaluations of trustworthiness based on appearance, which occurred independently of consciously held beliefs about racial groups = our behavior isn't driven only by what we consciously intend

What properties of faces do we represent in terms of summary statistics? What experimental paradigms have researchers used to demonstrate this?

1. Summary statistics = average of the group (in its very simplest form, although there's probably a better definition) 2. Average emotion: morphs of 4, 8, 12, or 16 faces from happy/sad or neutral/disgust continuum, mean was selected randomly for each trial, set: 2 happier faces (3 and 9 steps from mean), 2 were sadder (3 and 9 steps from mean) (maybe also for neutral and disgust?), mean face never a member==> probability of "yes" responses increased w/ proximity to the mean, bell-shaped curve, doesn't matter what size the set was; also asked if certain face is happier or sadder than the set mean (maybe shown different faces in one condition and identical faces in another -- here, asked if test face is happier than the set or, on a different trial, if more neutral) ==> as separation between set mean and test face increases, performance also appears to increase (people seem to be nearly as good at mean discrimination as regular discrimination); in the same article, participants indicated the location of any one of the faces that changed, set of 16 faces, participants see each set of faces for 1 second and asked which is happier or sadder ==> mean discrimination was above chance (≈72%), change localization ≈50%, mean discrimination with failed change localization ≈ 60% (not detecting the change, but still above chance, each is significant, if participants attended to one face, probability of a correct guess would be 25%) 3. Average identity of unfamiliar faces (exactly why we do that isn't apparent at first): IV: see a set of faces; I think the test face was a matching morph, a non-matching morph, a matching member, and a non-matching member; DV: "Was this face a member of the original set?"; participants were more likely to say the average face was present than the actual members of the set, but proportion "present" responses were higher for matching than non-matching faces for the morph and the member 4. Average identity for familiar (famous) faces: 4 faces + probe afterwards (average or an exemplar (Bill Clinton): same face, different made from the faces, different from other faces) ==> exemplar: similar > different from ame faces > different from other faces; average ==> similar > different from same faces > different from other faces. It appears that we represent the mean identity similarly for familiar and unfamiliar faces. We're forming averages of facial identity. 5. Also: size (circles study), orientation, motion, speed <-- kind of properties of faces

Can variations in images lead to different first impressions? Again, why does this matter for the study of accuracy?

1. Training computer face recognition algorithms -- different images can convey different traits A. 5 headshots of each individual; told to move heads around (so computer can detect identities using different angles), not told to use a specific expression (images vary, but not in any systematic way) B. IV: ratings -- attractive, extraverted, trustworthy; faces C. DV: judged images on different traits (each participant only saw one picture of each identity) D. Wide spread of average ratings for the identities (different images of the person can be seen as expressing different traits) 2. Same study, but pairing images -- context matters when choosing the image A. Individuals with the highest average ratings across all images were paired with those with the lowest ratings B. For each pair, it was possible to find images where the ratings were reversed for extraversion and trustworthiness <-- same person C. The reversal seem to have a lot to do with smiles, but even when excluding open smiles, reversals were still found without such overt smiling behavior 3. Also part of the same experiment: Pick the photo that would best fit that situation (five images of each person) A. Resume, online dating profile, new Facebook picture, campaign poster for running for mayor, audition for the role of a villain (only given one situation) B. Different situations were chosen for different images, even if the image was the same person 4. Different photos of the same person can convey different impressions; participants in accuracy studies are presented with pictures of individuals who vary on a given characteristics. If one cannot rule out selection bias in the images, then there is reason to be skeptical of accuracy claims. That is, the person may differ in the trait just because that picture fits the trait. Moreover, if researchers are picking photos, they may unconsciously be picking pictures that fit the traits. Need more?

Can judgments of competence based on appearance predict election outcomes across cultures?

1. US participants & Bulgarian presidential election (more candidates = more variability in facial appearance) -listed traits considered important in a president --> others rated importance of these traits --> chosen: competence, dominance, likability, honesty (also attractiveness) -US participants -ratings of competence predicted vote share for Bulgarian presidential candidates (increased perceived competence = decreased rank (1 = good?)), only for competence but other traits didn't predict vote share 2. French parliamentary outcomes (found across cultures & age groups) -Swiss children (winner, runner-up) -Who would you choose as the captain of your boat? -Children's judgments predicted French parliamentary outcomes 3. Depends upon whether the culture is independent or interdependent: American and Korean election outcomes (independent cultures: traits motivate behaviors versus interdependent cultures: social contexts motivate behaviors) -Winner & runner-up -Assembly election or senatorial/gubernatorial elections -High levels of consensus in competence ratings (for both in- and out-groups), but perceived competence was better at predicting US vs. Korean elections (for both Korean and American participants, but actually a larger difference maybe in accuracy % for Korean participants between US elections and Korean elections) = judgments of competence based on facial appearance seem to be less important to Korean voters; other evidence shows that Korean voters are influenced by indicators of social relations, like hometown

What are "data-driven methods"? What are the benefits of these methods?

1. Using unbiased methods, data-driven methods try to identify all (not just a few features) of the information in a face that is linked to trait judgments. That is, they don't rely on systematic image manipulated related to a hypothesis, but instead, they take random variation in faces and relate it to judgments. 2. Two main methods: A. Face space models (computer models; randomly generate faces/different identities) B. Reverse correlation methods (superimpose random noise on faces) 3. Benefits: Don't have the issues listed about with manipulating features (I think). They use real faces, although not exactly because the reverse correlation technique may use composites. Face space models, however, have been created by scanning real faces in 3D and superimposing a surface mesh onto them. I'm guessing there are more benefits. Can to principal components analysis. Not driven by our hypotheses, but by how the faces are actually judged. You get a more holistic image than changing certain features. More?

Which facial cues drive face evaluation?

1. Valence evaluation appears to be an overgeneralization of emotional expressions. (emotional expressions drive valence evaluation; we judge trustworthiness by picking up on emotional cues). Also, when one subtracts attractiveness from trustworthy faces, more trustworthy faces are less attractive. A. Faces with exaggerated (exaggerated the features contributing to these judgments) trustworthiness: seen as expressing emotion (the facial cues in trustworthiness drive emotion judgments, maybe?). B. Interesting because the initial faces used to create the model were in the neutral categorization range. C. Submissive to dominance = not much change, rated as mostly neutral D.Untrustworthiness to trustworthiness = change from angry to happy E. Study: evaluated set of neutral faces on trustworthiness --> saw a series of angry, fearful, or happy faces --> rated trustworthiness of the neutral faces ("How trustworthy was the previous face?" 1 = completely untrustworthy, 9 = completely trustworthy) --> adaptation to angry faces led to an increase in perceived trustworthiness; adaptation to happiness had the opposite effect (faces seemed more negative = less trustworthy). No significant change in the fearful condition (don't want fearful faces to change how trustworthy you think the faces are; maybe want the judgments of trustworthiness and fear to be independent???) 2. Dominance evaluations are an overgeneralization of physical strength/weakness. A. Moving from the negative to the positive extreme of dominance changes faces from feminine/baby-faced to masculine/mature-faced

What evidence suggests that different mechanisms are responsible for low- and high-level ensemble perception (multiple mechanisms for forming summary representations)?

1. Visual system forms summary representations of both low- (size, orientation, motion, speed) and high-level (facial emotion, identity) visual properties. These different visual properties are processed in different brain regions. 2. Habermas et al., 2015 (faces & gabor patches): people were told to adjust the image so it matches the average or just one member of the set of, I think, faces or gabor patches ==> very weak correlation across domain (correlation between mean absolute error between individual face & gabor patch; correlation of degree of mean absolute error between a group of faces (average) and group of gabor patches (average)); stronger correlation within domain (correlation of degree of mean absolute error between an individual face and a group of faces; correlation of degree of mean absolute error between a gabor patch and a group of patches); largest correlation was for the faces 3. Same study: authors had participants do a number of different tasks ==> stronger and significant correlation w/in high and low (high- and high-level property tasks, for example), but not between high and low 4. These results suggest that there are multiple ensemble mechanisms.

What is an example of a question you could address using a reverse correlation method? Which of the reverse correlation methods would be better suited to address this question?

Look back at my first test to see where I went wrong on a similar question. How do American participants conceptualize of a "black" face? What about people who have different prejudice levels (IAT scores)? Better to use the internal representation reverse correlation approach because you're not asking which features participants use to make these judgments, but instead, what their mental representation of the face looks like. You're not asking about different visual categories, and you're using just one base face. You don't know what the mental representation is, so you won't have specific categories (like happy/sad for the informational diagnosticity reverse correlation approach). I said "American" because people from different countries will most likely have different representation of what a "black" person looks like because of the history of that place, etc. Would want to use an ambiguous face, so maybe a morph between a Black and White American (or multiple morphs). Then go into detail about how this would be done. Anything else I'm missing?

Is the preference for faces with average appearance present across cultures?

Mostly 1. Developmental studies have been difficult to interpret: infants will look longer at less average (5-face composites) than more average (20-face composites) faces 2. Shown across several cultures, including Japan and the UK 3. Study with the Hadza: Hadza and also Western participants recruited over the Internet, shown White European and Hadza composites (pairs of 20- and 5-face composites --> Which face do you find most attractive?) ==> Westerners: increased averageness attractive overall. Hadza: averageness is attractive in Hadza faces --> cross-cultural agreement in attractiveness of averageness, but Hadza may not have had enough visual experience with Westerners to form a mental representation of an average White European face