Thursday, December 8, 2022

Intelligence was positively and significantly correlated with face detection, face perception, and face memory

The association between intelligence and face processing abilities: A conceptual and meta-analytic review. Dana L. Walker et al. Intelligence, Volume 96, January–February 2023, 101718.

Abstract: Whether there is an association between intelligence and face processing ability (i.e., face detection, face perception and face memory) is contentious, with some suggesting a moderate, positive association and others contending there is no meaningful association. The inconsistent results may be due to sample size differences, as well as variability in the quality of intelligence measures administered. The establishment of a moderate, positive correlation between face processing and intelligence would suggest it may be integrated within the Cattell-Horn-Carroll model of intelligence. Additionally, developmental prosopagnosia, a specific impairment of the recognition of facial identity, may be assessable in a manner similar to a learning disability. Consequently, we employed a psychometric meta-analytic approach to estimate the true score correlation between intelligence and face processing ability. Intelligence was positively and significantly correlated with face detection (r’ = 0.20; k = 2, N = 407), face perception (r’ = 0.42, k = 11, N = 2528), and face memory (r’ = 0.26, k = 23, N = 9062). Additionally, intelligence measurement quality moderated positively and significantly the association between intelligence and face memory (β = 0.08). On the basis of both theoretical and empirical considerations, we interpreted the results to suggest that face processing ability may be plausibly conceptualised within the Cattell-Horn-Carroll model of intelligence, in a manner similar to other relatively narrow dimensions of cognitive ability, i.e., associated positively with intelligence, but also distinct (e.g., reading comprehension). Potential clinical implications for the assessment of developmental prosopagnosia are also discussed.


On theoretical and empirical grounds, some researchers claim that face processing ability is essentially independent of general intelligence1 (Bowles et al., 2009; Shakeshaft & Plomin, 2015; Wilmer, Germine, & Nakayama, 2014), whereas others contend that it is associated positively and meaningfully with other well-known cognitive abilities, including general intelligence (Connolly, Young, & Lewis, 2019; Gignac, Shankaralingam, Walker, & Kilpatrick, 2016; Hildebrandt, Wilhelm, Schmiedek, Herzmann, & Sommer, 2011). Thus, there is currently no consensus on whether individual differences in face processing ability may be considered a conventional cognitive ability or not.

In order to advance the area forward, in this review, we refer to abstract and operational definitions of intelligence, alongside descriptions of some of the key theories and models of cognitive ability, and we note connections with face processing ability and its measurement. We also conduct meta-analyses on the association between intelligence and face processing ability. To foreshadow, we will suggest that several face processing abilities may be plausibly conceptualised within the broadly accepted model of cognitive abilities, the Cattell-Horn-Carroll (CHC) model (McGrew, 2009). We will also contend that there are potential benefits with such an integration, both theoretical and practical.

Several abstract definitions of intelligence have been provided. For example, echoing Pintner (1923), Sternberg (1997, p.1) defined intelligence as “…the mental abilities necessary for adaptation to, as well as shaping and selection of, any environmental context” (see also McIntosh, Dixon, & Pierson, 2012). Gignac (2017, p. 465) defined intelligence somewhat less abstractly as “…an entity's maximal capacity to achieve a novel goal successfully using perceptual-cognitive abilities.” As will be noted in more detail below, face processing ability may be conceptualised as an adaptive capacity relevant to achieving novel goals using perceptual-cognitive abilities, suggesting face processing abilities may be integrated within conventional intelligence conceptualisations. Although abstract definitions of intelligence are useful, especially in the context of theory, they are limited with respect to the generation of psychometric measures. By contrast, operational definitions, which are more concrete than abstract definitions, facilitate psychometric measurement.

In more operational terms, Gignac (2017, p. 465) defined intelligence as “…an entity's maximal capacity to complete a novel, standardised task with veridical scoring using perceptual-cognitive abilities.” Thus, intelligence tests have scoring that is objective and verifiable. For example, the word ‘ambiguous’ has an agreed upon definition and a person can be asked to define the word ambiguous as part of a vocabulary test. Another example is Digit Span Forward (Kaplan, 1991), a short-term memory test where participants are asked to repeat a series of numbers sequentially. As a final example, participants can complete the Trails-B task (Corrigan & Hinkeldey, 1987), a measure of processing speed, by connecting numbers and letters, within a limited time, in an alternating progressive sequence, 1 to A, A to 2, 2 to B, and so on. Thus, intelligence tests can be administered in a way that is objective and performance is verifiable. To foreshadow, published face processing ability tests can also be regarded as objective tasks of performance, much like typical tests of intelligence.

It is important to note that performance on intelligence tests correlate with each other positively, a phenomenon known as the positive manifold, i.e., ubiquitous positive correlations between cognitive abilities (Carroll, 1993; Spearman, 1904). Consider, for example, that the correlation between verbal comprehension and working memory is r = 0.64, based on the Wechsler Adult Intelligence Scale-IV (WAIS-IV) normative sample (Wechsler, 2008). Furthermore, Digit Span, a measure of short-term memory, is correlated at r = 0.50 with Vocabulary, a measure of crystallised intelligence (Wechsler, 2008). Additionally, Matrix Reasoning is correlated with Symbol Search, a measure of processing speed, at r = 0.39 (Wechsler, 2008). In fact, the average inter-subtest correlation across all 10 subtests of the WAIS-IV is 0.43; and none of the inter-correlations are negative or zero. The inter-correlations between cognitive ability measures have facilitated the development of models of intelligence via techniques such as factor analysis.

Over the years, several models of intelligence have been proposed. For example, Spearman's two-factor model (Spearman, 1904) emphasised the prominence of the general factor on the basis of the positive manifold; Cattell/Horn's model that emphasised the distinction between fluid and crystallised intelligence (Cattell, 1941; Horn & Cattell, 1966); and Carroll's (1993) extensive factor analytic work that has culminated into the CHC model of intelligence (McGrew, 2009).

The CHC model of intelligence is an amalgamation of Horn and Cattel's (1966) model and Carroll's model (Carroll, 1993). The first generation of the CHC model aimed to reconcile the differences between the two models (McGrew, 1997). The first CHC model was based substantially upon Carroll's hierarchical three-factor model, although it included a unique broad ability (reading and writing, Grw) and new narrow abilities, such as reading comprehension and reading speed (see Flanagan & Dixon, 2013). The CHC model has been refined based upon current factor analytic research, as well as developmental, neurocognitive, and heritability evidence (Flanagan & Dixon, 2013).

Today, the structure of the CHC model consists of a general factor (known as g), which is referred to as a Stratum III ability within the model. The model also includes 16 broad abilities, called Stratum II abilities, that appear under g (Newton & McGrew, 2010). The Stratum II abilities include: fluid reasoning (Gf),2 comprehension–knowledge (Gc),3 reading and writing (Grw), visual processing (Gv), long-term storage and retrieval (Glr), processing speed (Gs), short-term memory (Gsm), reaction and decision speed (Gt), and quantitative knowledge (Gq; see Table 1). As described by McGrew (2009), the CHC model includes additional possible Stratum II abilities that have not yet been validated fully, including Auditory processing (Ga), General (domain specific) knowledge (Gkn), Tactile abilities (Gh), Kinesthetic abilities (Gk), Olfactory abilities (Go), Psychomotor abilities (Gp), and Psychomotor speed (Gps).

Each Stratum II (broad) ability is divided further into narrower abilities (i.e., Stratum I abilities) that define the depth and breadth of a broad Stratum II ability. For example, memory span (MS) and working memory (WM) are Stratum I abilities and each measures a different aspect of Gsm (a Stratum II ability). In a comprehensive review, Newton and McGrew (2010) listed all nine broad (Stratum II) abilities and nearly 100 Stratum I abilities, with the latter being very narrow in scope. Examples of Stratum I abilities include writing ability (WA), mathematical achievement (A3), simple reaction time (R1), closure speed (CS), and reading comprehension (RC).

Although the CHC model of intelligence is a relatively comprehensive model of individual differences in cognitive abilities, several authors have contended that additional factors may be seriously considered for inclusion into the CHC model, including social and emotional intelligence (Wilhelm & Kyllonen, 2021). Additionally, it has been suggested that face processing abilities may be advantageously considered within the CHC model of intelligence (Meyer, Sommer, & Hildebrandt, 2021). As we detail below, commonly measured dimensions of face processing ability, including face detection, face perception and face memory, may be linked theoretically and empirically to several of the CHC model dimensions (see Table 1 for summary; see also Table S1 in supplementary materials).

Stratum I abilities are correlated positively with g (McGrew, 2009). For example, the correlation between general intelligence and reading comprehension, a Stratum I ability, has been reported to range between ≈ 0.40 and ≈ 0.55 (Jensen, 1998; Joshi & Hulme, 1998; Naglieri & Ronning, 2000; Tiu Jr, Thompson, & Lewis, 2003). Importantly, while reading comprehension is correlated moderately with general intelligence, it is not considered isomorphic with g. In fact, intelligence researchers recognise reading comprehension as a specific ability that can predict various outcomes, above and beyond general intelligence (Gersten, Fuchs, Williams, & Baker, 2001). Such an observation will be important for the theorised role of face processing ability within the context of cognitive abilities more broadly, as described in more detail further below.

Human face processing may be defined simply as the abilities necessary to process facial information, including the ability to detect, match, and recognise faces accurately (Fysh, 2018; Meyer et al., 2021). As mentioned previously, intelligence may be viewed as how well an individual adapts to an environment successfully using cognitive abilities (McIntosh et al., 2012; Pintner, 1923; Sternberg, 1997). Face processing ability, a construct that includes face detection, face perception and face recognition as dimensions (described in more detail below), are all abilities that may be suggested to facilitate successful adaptation. For example, individual differences in face processing ability correlate positively with cooperative interactions (r = 0.25; Corbett, Newsom, Key, Qualls, & Edmiston, 2014) and quality of social networks (r = 0.21; McLaughlin Engfors, Palermo, & Jeffery, 2019). Therefore, face processing ability could be defined as an adaptive ability, as per cognitive intelligence more generally. Furthermore, face processing tasks require perceptual-cognitive skills to solve novel problems. Finally, the tasks are scored objectively – again, as per conventional IQ tests. As one example, face perception tasks (e.g., Cambridge Face Perception Test, CFPT; Duchaine, Germine, & Nakayama, 2007) show a line-up of faces that need to be matched to a target face, based on the degree of visual similarity to the target face. The task is scored based upon the number of accurate matches (quantitative similarity). Therefore, in general terms, face processing ability could be defined operationally as an individual's capacity to use cognitive faculties to complete a novel task involving faces and for which there is a clear procedure to evaluate successful completion of the task (i.e., veridical scoring).

Like cognitive abilities more generally, there is evidence that face processing abilities yield a positive manifold. Verhallen et al. (2017) referred to the face processing general factor as f. McCaffery, Robertson, Young, and Burton (2018) and Verhallen et al. (2017) reported moderate to relatively large correlations (r ≈ 0.20 to 0.50) between measures of face detection, face perception, and face memory (defined below). Although not all of the empirical research is consistent (e.g., Fysh, 2018), the observation of positive correlations between face processing abilities is similar to the observation of positive correlations between cognitive abilities more generally (Carroll, 1993). It should be noted that although detection, matching and recognising faces may be considered positively inter-related processes, they are also considered to be, at least to some degree, distinct. That is, the relatively large correlations (by individual differences research standards; Gignac & Szodorai, 2016) are not large enough to suggest construct redundancy. We discuss each face processing dimension in further detail next.

Face detection is the ability to detect a face generally within a visual scene (Bindemann & Lewis, 2013; Verhallen et al., 2014). Studies show that humans are quicker at detecting a face than any other non-face object (Lewis & Ellis, 2003), implying that faces are an important object to detect for humans. It has been suggested that there may be a dedicated neurophysiological system that mediates the process of face detection, a system distinct from the detection of other objects (Lewis & Ellis, 2003). Individuals with prosopagnosia, the inability to recognise faces, can have impairments in their ability to detect faces (Garrido, Duchaine, & Nakayama, 2008). In fact, de Gelder and Stekelenburg (2005) proposed that some cases of developmental prosopagnosia may originate from deficits in face detection. Furthermore, they proposed that the face detection system is crucial for the normal development of more specialised face processing abilities, such as face memory. Thus, face detection may be considered a relatively more primary face processing ability.

A commonly used test of face detection is the Mooney test (Mooney, 1957), whereby a participant must view degraded images and determine whether an image contains a face or not (e.g., Fig. 1, left-side). Each Mooney face detection image has obstructions of the important local, featural and relational information (e.g., eyes, nose, mouth). Specifically, an individual would have to construct a specific, three-dimensional model of both the face and lighting in order to detect the face (Verhallen & Mollon, 2016). The underlying processes likely draws, to some degree, upon the observer's stored knowledge of faces acquired over their lifetime (Verhallen et al., 2017). Other tasks of face detection involve finding face-like images (see Robertson, Jenkins, & Burton, 2017) and actual face images (see Fysh, 2018) within a visual scene. Both of these tasks require participants to search visual scenes for concealed face images. Comparatively, the forementioned tasks involve visual searching of scenes to detect a real face, in comparison to the Mooney test which involves detection of a face from black and white ambiguous and non-ambiguous images. Some researchers argue that the Mooney test incorporates limited visual searching, a suggested essential component of face detection (Bindemann & Lewis, 2013; Fysh, 2018). Nonetheless, the Mooney test has been shown to be a reliable and valid measure of face detection (Schwiedrzik, Melloni, & Schurger, 2018; Verhallen et al., 2014; Verhallen & Mollon, 2016).

Overall, face detection is a holistic process whereby information is processed in a more general, “big picture” way, compared to local processing. Local processing involves attending to specific details, or processing information in a narrower and more detail orientated way (Navon, 1977). At a superficial level, the Mooney test has seemingly unrelated patches of white and black. An individual completing the task would need to look at the picture as a whole and decide whether the patches of white and black converge together to form the percept of a face. Thus, the Mooney test involves global judgments that are somewhat dependent upon the integration of local elements (Mooney, 1957). This process of organisation is often referred to as closure (or figure closure).

Arguably, the Mooney test may be considered a relatively narrow instantiation of more general figure closure tasks. For example, the Gestalt Figure Completion Task (Eliot & Czarnolewski, 1999; Goodwin, 2012; Street, 1931) is a commonly used measure of general figure closure ability: an ability regarded as a subdimension of intelligence (Closure Speed, CS; McGrew, 2009). Gestalt perception tasks tend to include incomplete figures of familiar objects, animals, or humans. In a manner similar to the Mooney test, an individual must first recognise the ambiguous stimuli and then label it (see Fig. 1, right-side). Arguably, with respect to both the Mooney test and Gestalt Figure Completion Task, an individual would have to create a mental image of the face/object, drawing upon their experience and knowledge of objects observed within their lifetime. Thus, drawing from the CHC model of cognitive abilities, performance on both tasks likely draws upon Gv (visual processing) and to some degree Gc (comprehension-knowledge). Thus, a positive correlation between face detection ability and general figure closure ability would be expected on theoretical grounds. Correspondingly, small-scale (N = 63) empirical research suggests that general figure closure and face detection tasks load onto the same cognitive ability factor (Wasserstein, Barr, Zappulla, & Rock, 2004). Therefore, whether figure closure tasks that include only face stimuli, as per the Mooney test, draw upon unique visual processing ability (i.e., somewhat distinct from general figure closure ability) remains to be determined, convincingly. Theoretically, the observation of some face detection specific (unique) variance would align with current research, suggesting that the ability to detect faces may be a process that is, at least to some degree, distinct from the ability to detect other objects (Lewis & Ellis, 2003).

Despite the fact that the Mooney test was published many years ago, little research has examined the association between intelligence and face detection ability. In one study, Vigen, Goebel, and Embree (1982) estimated the association between IQ (WAIS-R) and face detection ability (Mooney test) at r = 0.25, based on a diverse sample of college, vocation and community member participants (N = 300). By contrast, in another study with a primarily community sample (N = 104), McCaffery et al. (2018) reported a non-significant correlation (r = 0.06) between executive functioning (Card Sorting Task) and face detection ability (Mooney test). McCaffery et al. (2018) suggested that there was little association between face detection ability and other cognitive abilities. Thus, a meta-analysis may be required to help generate a consensus view on this issue.

Theoretically, face perception is an important ability that would be expected to occur after a face has been detected. That is, once a face has been detected, it is possible to discriminate or individualise faces from each other. Face perception ability, at a basic level, involves scanning faces within a group and identifying faces as distinct/similar. Correspondingly, in typical face perception tasks, participants must discriminate, or tell apart, one face from another. Face perception tasks usually require the face stimuli to remain visible, in order to ensure that the task is focused on the visual processing required to perceive faces, with minimal memory requirements. The Warrington Recognition Memory for Faces test (Warrington, 1984) has participants view two photos and make the judgement of whether the identity of the person portrayed is the same or different (see Fig. 2). By contrast, the Benton Face Recognition Test (Benton, Sivan, Hamsher, Varney, & Spreen, 1983) has participants look at a target photo and asks them to choose the target individual from six simultaneously displayed photos (see Fig. 3). In recent years, other face perception tasks have been designed, including the relatively popular CFPT (Duchaine, Yovel, & Nakayama, 2007), the Kent Face Matching Test (Fysh & Bindemann, 2018), the Glasgow Face Matching Test (Burton, White, & McNeill, 2010; White, Guilbert, Varela, Jenkins, & Burton, 2021) and the Faces Card-Sorting Task (Andrews, Jenkins, Cursiter, & Burton, 2015). Arguably, these face perception tasks involve visual processing of faces with minimal memory requirements, thus rendering them relatively pure face perception tasks.

Higher levels of face perception ability have been linked to positive outcomes, whereas lower levels of face perception ability haven been linked to social difficulties. For example, the ability to tell faces apart, or individualise a face, is an important social skill (Fysh, Stacchi, & Ramon, 2020). From a professional perspective, many common professions require at least adequate performance in the ability to perceive and differentiate faces. For example, police officers may have to match a photo of a suspect with video footage of a crime scene (White et al., 2015). Additionally, border control officers and airport security personnel often check identification by matching a passport photo with the face of the person who presents with the identification (White et al., 2015). Similarly, people who work in banks, post offices, and establishments that sell alcohol must often match photo identification to a face.

Not everyone can perceive faces well. For example, individuals with prosopagnosia are often impaired in their face perception ability (Behrmann & Avidan, 2005; Duchaine, Germine, & Nakayama, 2007). Duchaine, Yovel, and Nakayama (2007) found that healthy controls averaged statistically significantly fewer errors than people with developmental prosopagnosia (Cohen's d = −2.13). Correspondingly, people with developmental prosopagnosia often report socialisation difficulties due to their poor face processing abilities and become anxious in public locations (Dalrymple et al., 2014). Therefore, a greater understanding of face perception ability is not only important theoretically, but also practically.

Face perception requires the ability to accurately discern facial configurations and features (Hildebrandt, Schacht, Sommer, & Wilhelm, 2012). More specifically, individuals must detect similarities, or differences, between faces. It could be argued that many visual processing and fluid reasoning tasks require similar detection of image similarities and differences. Consider, for example, the Raven's Progressive Matrices Test (Raven, Raven, & Court, 1998), a measure of fluid reasoning. In this task, participants are presented with a 3 × 3 matrix of geometric figures. Fig. 4 includes an example progressive matrices item from the International Cognitive Ability Resource (ICAR, 2017); it can be seen that the bottom right geometric figure is missing and must be selected from eight multiple choice response options. Interestingly, McGreggor, Kunda, and Goel (2010) found that a computer program designed solely to compare the similarity of images (akin to face perception tasks) was able to accurately complete over half of the Raven's Progressive Matrices Test. Arguably, the computer program exhibited processes related primarily to visual processing (Gv), and perhaps specifically visual matching, in addition to fluid reasoning (Gf). Correspondingly, Raven's Progressive Matrices has been found to measure general intelligence, as well as Gf and Gv (Gignac, 2017). Therefore, it is plausible to suggest that there is a positive association between a person's ability to perceive and differentiate faces and an individual's Gf and Gv ability. Stated alternatively, face perception ability may be considered, in part, a cognitive ability imbued with visual processing and fluid reasoning variance, within the context of the CHC model of cognitive abilities, at least theoretically.

Empirically, the evidence also suggests the possibility of a positive association. For example, Wilhelm et al. (2010) found a significant, positive association (r = 0.56) between their measure of face perception, a custom-made task based upon the part-whole paradigm, and composite intelligence scores defined by multiple cognitive ability subtests (community sample: N = 209). They interpreted their findings as supportive of the hypothesis of an association between intelligence and face perception ability. By contrast, Slone, Brigham, and Meissner (2000) investigated the association between the Benton Face Recognition Task, a measure of face perception ability, and a digit span task, a measure of short-term memory. They reported a small, non-significant correlation (r = 0.09); however, their study was based on a relatively small and restricted sample of university students (N = 129). As per face detection, the inconsistent results in the literature suggest that a meta-analysis may be beneficial.

Face recognition is a term often used in the literature to describe different concepts. Some authors use the term face recognition for a task that involves perceiving faces (Oruc, Balas, & Landy, 2019). Additionally, the term face recognition has been used as a label for tasks and processes that are face perception or memory in nature. For example, the Benton Face Recognition Test (Benton et al., 1983) is a face perception test. Within this review, the term ‘face memory’ will be used, rather than the more ambiguous term ‘face recognition’.

Face memory is the ability to perceive a face, encode that face into memory, and then recall that face, in order to determine if it has been seen previously (Dalrymple & Palermo, 2016). Many face memory tasks have a short interval between viewing the face and recalling the face (e.g., Cambridge Face Memory Test; CFMT; Duchaine & Nakayama, 2006). The CFMT requires participants to recognise six learnt faces across three test stages (see Fig. 5). In the learning stage, the participants learn the faces of six identities in frontal and side-on views. The first test stage requires participants to select which image contains a learnt face amongst two distractors. The images in this stage are identical to the learning stage. The second test stage employs the same three-alternative force choice paradigm, however, the images shown are different to the learning stage, i.e. novel images where the faces have different viewpoint and/or lighting. The third test phase is the same as the second stage, however, participants must recognise a learnt face in novel images covered by heavy visual noise.

There are also face memory tasks that test an individual's ability to recall the identity of a face over a longer period, for example, a time-delayed CFMT (McKone et al., 2011). The standard CFMT and time-delayed CFMT (both 20 min and 24 h) are correlated at 0.84 (McKone et al., 2011). Even though the CFMT is the most popular face memory test used by researchers, there are other valid tasks developed to measure face memory (see Hildebrandt et al., 2011).

Arguably, face memory is an important skill for successful social interaction, as the successful recognition of another person would be expected to determine how we may interact with the person in an appropriate manner. For example, recognising a colleague compared to a family member, will impact the interaction and appropriate socialisation. Correspondingly, individuals with clinical developmental prosopagnosia report that they avoid social situations where face memory is important (Murray, Hills, Bennetts, & Bate, 2018; Yardley, McDermott, Pisarski, Duchaine, & Nakayama, 2008). Furthermore, they also report long-term, negative consequences, as a result, including (but not limited to) dependency on others, restricted social circle, more limited employment opportunities and low self-confidence (Murray et al., 2018; Yardley et al., 2008). The interpersonal struggles shown by people with developmental prosopagnosia, linked to their inability to recognise faces, highlights the importance of face memory for everyday situations.

Theoretically, face memory may be considered to be associated with multiple cognitive abilities. The ability to recognise and remember faces over a short period of time may be linked to an individual's short-term memory (Gsm). Consider that the Digit Span Forward task from the Wechsler scales is similar in structure and task design to the CFMT. In Digit Span Forward, participants must recall a series of numbers previously learnt, whereas the CFMT requires participants to recall the identity of six faces previously learnt. The two tasks arguably tap into a similar process, notably short-term memory (Gsm). By contrast, the ability to recognise and remember faces over a long period of time could be linked to an individual's long-term storage and retrieval (Glr).

Interestingly, the Wechsler Memory Scale –Third Edition (WMS-III) includes two face memory tasks. These tasks, labelled Faces I and Faces II, form part of the Visual Immediate or Visual Delayed indices. In Faces I, participants are shown 24 target faces, and each face is displayed one at a time for 2 s. Then, participants are shown 48 faces (24 targets and 24 distractors) and are asked to identify the target faces by responding either “yes” or “no” to each face. Participants are prompted to keep the target faces in mind. In Faces II, participants are shown 48 faces (24 targets and 24 distractors) after a 30-min delay and are asked again to identify the target faces. Faces I and Faces II correlated with other subtests within the WMS-III. For example, Faces I correlated at 0.14 with Logical Memory I, and also correlated with another Visual Memory Immediate index task (Family Pictures I) at 0.30 (Psychological Corporation, 1997). Ultimately, the correlations between the Faces tasks and the other tasks within the WMS-III were deemed too low (insufficient convergent validity), which led to the removal of these tasks from the WMS-IV (Hawkins & Tulsky, 2004).

On the one hand, the low correlations may be due to methodological considerations. For example, the measure itself differs from the free recall methodology employed by the WMS-III. Moreover, the recognition format of the faces subtest without a recall component may make the test easier than other nonverbal memory tests (Tulsky, Chiaravalloti, Palmer, & Chelune, 2003). On the other hand, it may be acknowledged that facial memory may require a special (unique) type of visual processing. For example, research into face recognition ability has found that recognition of faces activates a cortical region in the brain specialised to the perception of faces, known as the Fusiform Face Area (Kanwisher & Yovel, 2006; Tsao, Freiwald, Tootell, & Livingstone, 2006). Consequently, memory for faces would not necessarily be expected to be meaningfully correlated with intelligence, and some researchers contend that it is not (Bowles et al., 2009; Shakeshaft & Plomin, 2015; Wilmer et al., 2014).

The empirical results on the association between intelligence and face memory are inconsistent. For example, Gignac et al. (2016) reported a positive, significant association between intelligence, as measured by multiple subtests, and the CFMT, a measure of face memory (r = 0.35; N = 211). They interpreted their findings as supportive of an association between intelligence and face memory ability. By contrast, Richler, Wilmer, and Gauthier (2017) failed to find a significant association between intelligence (Matrices from the Wechsler Abbreviated Scale of Intelligence) and face memory (CFMT), based on a community sample (N = 279). It is noted that Richler et al.'s measurement of intelligence would not be considered good or excellent, based on Gignac and Bates (2017) guidelines, whereas several studies that did use good or excellent intelligence measurement (i.e., several subtests; multiple dimensions) did find a significant and positive association between intelligence and face memory ability (Gignac et al., 2016; Herlitz & Yonker, 2002; Zhu et al., 2010). Thus, a meta-analysis may be useful to help synthesise the empirical results and possibly identify intelligence measurement quality as a positive moderator of the effect between intelligence and face processing ability.

In addition to identity, people glean lots of information from faces, such as eye gaze, attractiveness, trustworthiness, speech decoding, first impression and emotion. These face processing abilities are beyond the scope of this review; however, it is important to review briefly the recent research on individual differences in face emotion recognition and intelligence. Face emotion recognition is the ability to accurately and efficiently recognise facial expressions (Palermo, Connor, Davis, Irons, & McKone, 2013). Empirically, individual differences in face emotion recognition have been found to be associated positively with intelligence (Borod et al., 2000; Connolly et al., 2019; Hildebrandt, Sommer, Schacht, & Wilhelm, 2015). Furthermore, a meta-analysis estimated the association between face emotion recognition and cognitive abilities at r ≈ 0.19 (Schlegel et al., 2019); however, the correlations were not corrected for measurement error and range restriction, nor was intelligence measurement quality taken into consideration. Thus, the reported 0.19 correlation is likely a substantial underestimate.

It is plausible to postulate that face processing abilities facilitate successful adaptation and involves goal/problem solving using cogntive-perceptual abilities. Stated alternatively, we define face processing ability as an adaptive cognitive-perceptual ability to detect, match or recognise facial identity and facial expressions. Such a definition aligns with abstract definitions of intelligence that focus upon successful environmental adaptation (McIntosh et al., 2012; Pintner, 1923; Sternberg, 1997).

Beyond theoretical similarities, face processing tests have characteristics that align with operational measures of cognitive abilities. That is, in more operational terms, face processing abilities can be defined as an individual's ability to complete a novel, standardised visual task involving faces and for which there is veridical scoring. For example, the CFMT is a standardised visual face task that includes novel problems/stimuli and is scored objectively; i.e., in line with conventional operational definitions of intelligence (Gignac, 2017).

Both higher and lower than average exposure to male sex hormones increases same sex attraction in male mice and in men

Carving the Biodevelopment of Same-Sex Sexual Orientation at Its Joints. Doug P. VanderLaan, Malvina N. Skorska, Diana E. Peragine & Lindsay A. Coome. Archives of Sexual Behavior, Aug 12 2022.

Abstract: Sexual orientation is a core aspect of human experience and understanding its development is fundamental to psychology as a scientific discipline. Biological perspectives have played an important role in uncovering the processes that contribute to sexual orientation development. Research in this field has relied on a variety of populations, including community, clinical, and cross-cultural samples, and has commonly focused on female gynephilia (i.e., female sexual attraction to adult females) and male androphilia (i.e., male sexual attraction to adult males). Genetic, hormonal, and immunological processes all appear to influence sexual orientation. Consistent with biological perspectives, there are sexual orientation differences in brain development and evidence indicates that similar biological influences apply across cultures. An outstanding question in the field is whether the hypothesized biological influences are all part of the same process or represent different developmental pathways leading to same-sex sexual orientation. Some studies indicate that same-sex sexually oriented people can be divided into subgroups who likely experienced different biological influences. Consideration of gender expression in addition to sexual orientation might help delineate such subgroups. Thus, future research on the possible existence of such subgroups could prove to be valuable for uncovering the biological development of sexual orientation. Recommendations for such future research are discussed.


We recognize that “gynephilia” and “androphilia” are terms that denote sexual attraction toward adults, and as such these terms do not apply in the case of those who are sexually oriented toward minors (i.e., children and/or adolescents). In studies of those sexually oriented toward minors, the terms “homosexual” and “heterosexual” have been used to denote attraction to the same- and opposite-sex, respectively (e.g., Blanchard et al., 2020). We, however, did not choose to use these terms because here we review to a greater extent the considerable cross-cultural literature on transgender, nonbinary, or third gender populations for which gynephilia and androphilia have been the terms typically employed.

By using same-sex sexual behavior as a proxy for sexual orientation, Ganna et al. (2019) were able to maximize inclusion of genetic data from 477,522 individuals. Importantly, the authors also reported supplemental analyses on genetic associations when using sexual attraction, fantasy, and identity measures available for subsets of participants, and the results generally aligned with those found when using the proxy behavioral measure in the full sample.

Digit ratio has been a particularly contentious biomarker given debate regarding the mechanisms that influence this trait, idiosyncratic findings across left- and right-hand 2D:4D, and the possibility that sex differences in 2D:4D are simply a consequence of allometry (i.e., reflect sex differences in physical size). For further discussion of these issues and data analyses indicating left and right 2D:4D are most appropriately analyzed separately from one another as well as from measures of physical size (e.g., hand length, height), we refer readers to Skorska et al. (2021a).

Gender expression has also often been regarded as a psychological marker of pre-/perinatal hormone exposure (Hines, 2020), but it has also been linked to the genetic and immunological mechanisms discussed in this section (e.g., Alanko et al., 2010; Blanchard, 2018; Coome et al., 2018). We refer readers to the section entitled “One Biodevelopmental Pathway or Many?” for in-depth discussion regarding the importance of gender expression to uncovering the bases of sexual orientation biodevelopment.

A large Dutch national probability sample reported a fraternal birth order effect among women belonging to female-female civil unions, suggesting birth order may be related to female sexual orientation as well (Ablaza et al., 2022); however, caution is warranted in interpreting this finding given prior inconsistencies, and primarily null findings, in birth order studies of female sexual orientation (Blanchard, 2022; Bogaert & Skorska, 2011; Semenyna et al., 2022).

Neuroscience research on transgender populations has often not included information regarding participants’ sexual orientations or reported that participants’ sexual orientations were heterogeneous (for a recent review, see Frigerio et al., 2021). Here, in addition to comprehensive review articles, we only cite examples of individual neuroscience studies that either reported on transgender samples of individuals described as being sexually oriented toward the same sex assigned at birth or that explicitly examined transgender participants’ brain features in relation to varying sexual orientation.

In the study by Rahman et al. (2020), prevalence rates varied depending on how sexual orientation was defined. For example, if sexual orientation was defined by heterosexual, bisexual, and homosexual identity, then prevalence rates were estimated at 90.7, 7.2, and 2.1% for women and of 90.0, 5.1, and 4.9% for men. In contrast, if defined by sexual attractions that were predominantly not toward the same sex, moderately toward the same sex, or predominantly toward the same sex, then prevalence rates were estimated at 66.2, 27.3, and 6.5% for women and 82.6, 10.2, and 7.2% for men.

To further verify the meaningfulness of the subgroups derived from their latent profile analysis, Swift-Gallant et al. (2019a) compared the subgroups on several psychological variables, including gender expression. Details regarding subgroup differences in gender expression are described in the subsection of the present article entitled “Research on Gender Expression and Sexual Orientation Biodevelopment.”.

Industrial policy in China: Little evidence that the Chinese government consistently “picks winners.” Firms’ ex-ante productivity is negatively correlated with subsidies, & subsidies appear to have a negative impact on productivity growth afterwards

Picking Winners? Government Subsidies and Firm Productivity in China. Lee G. Branstetter, Guangwei Li & Mengjia Ren. NBER Working Paper 30699, December 2022. DOI 10.3386/w30699

Abstract: Are Chinese industrial policies making the targeted Chinese firms more productive? Alternatively, are efforts to promote productivity undercut by efforts to maintain or expand employment in less productive enterprises? In this paper, we attempt to shed light on these questions through the analysis of previously underutilized microdata on direct government subsidies provided to China’s publicly traded firms. We categorize subsidies into different types. We then estimate total-factor productivity (TFP) for Chinese listed firms and investigate the relationship between these estimates of TFP and the allocation of government subsidies. We find little evidence that the Chinese government consistently “picks winners”. Firms’ ex-ante productivity is negatively correlated with subsidies received by firms, and subsidies appear to have a negative impact on firms’ ex-post productivity growth throughout our data window, 2007 to 2018. Neither subsidies given out under the name of R&D and innovation promotion nor industrial and equipment upgrading positively affect firms’ productivity growth. On the other hand, we find a positive impact of subsidy on current year employment, both for the aggregated and employment-related subsidies. These findings suggest that China’s increasingly prescriptive industrial policies may have generated limited effects in promoting productivity.

Estimating data corruption by publication selection bias: The presence of an effect in economics decreased from 99.9% to 29.7% after adjusting for the bias: in psychology, 98.9% → 55.7%; in medicine, 38.0% → 27.5%

Footprint of publication selection bias on meta-analyses in medicine, economics, and psychology. František Bartoš, Maximilian Maier, Eric-Jan Wagenmakers, Franziska Nippold, Hristos Doucouliagos, John P. A. Ioannidis, Willem M. Otte, Martina Sladekova, Daniele Fanelli, T.D. Stanley.

Abstract: Publication selection bias undermines the systematic accumulation of evidence. To assess the extent of this problem, we survey over 26,000 meta-analyses containing more than 800,000 effect size estimates from medicine, economics, and psychology. Our results indicate that meta-analyses in economics are the most severely contaminated by publication selection bias, closely followed by meta-analyses in psychology, whereas meta-analyses in medicine are contaminated the least. The median probability of the presence of an effect in economics decreased from 99.9% to 29.7% after adjusting for publication selection bias. This reduction was slightly lower in psychology (98.9% → 55.7%) and considerably lower in medicine (38.0% → 27.5%). The high prevalence of publication selection bias underscores the importance of adopting better research practices such as preregistration and registered reports.