Elsevier

Intelligence

Volume 57, July–August 2016, Pages 96-104
Intelligence

Short-term memory for faces relates to general intelligence moderately

https://doi.org/10.1016/j.intell.2016.05.001Get rights and content

Highlights

  • Short-term memory for faces correlated positively with several stratum II factors.

  • Short-term memory for faces was associated with general intelligence at .34.

  • Short-term memory for faces should not be considered “special” (i.e., independent of g).

  • Prosopagnosia may be best characterised as a learning disability.

Abstract

The results associated with a small number of investigations suggest that individual differences in memory for faces, as measured by the Cambridge Face Memory Test (CFMT), are independent of intelligence. Consequently, memory for faces has been suggested to be a special construct, unlike other cognitive abilities. However, previous investigations have measured intelligence with only one or two subtests. Additionally, the sample sizes upon which previous investigations were based were relatively small (N = 45 to 80). Consequently, in this investigation, a battery of eight cognitive ability tests and the CFMT were administered to a relatively large number of participants (N = 211). Based on a correlated-factor model, memory for faces was found to be correlated positively with fluid intelligence (.29), short-term memory (.23) and lexical knowledge ability (.19). Additionally, based on a higher-order model, memory for faces was found to be associated with g at .34. The results are interpreted to suggest that memory for faces, as measured by the CFMT, may be characterised as a relatively typical narrow cognitive ability within the Cattell–Horn–Carroll (CHC) model of intelligence, rather than a special ability (i.e., independent of other abilities). Future research with a greater diversity in the measurement of face recognition ability is encouraged (e.g., long-term memory), as the CFMT is a measure of short-term face memory ability.

Introduction

The capacity for face identity recognition has been a significant source of research over the years (e.g. Carey et al., 1980, Galper and Hochberg, 1971, Tanaka and Farah, 1993), perhaps in part because of the sensational phenomenon of prosopagnosia: the incapacity of otherwise cognitively able individuals to recognise familiar faces (Duchaine, 2011). In recent years, a number of investigators have begun to investigate face identity recognition ability as an individual difference construct (e.g. Dennett et al., 2012, Rhodes et al., 2014, Sekiguchi, 2011). The empirical evidence suggests that face recognition ability lies along a continuum, with some individuals in possession of relatively poor levels of face recognition ability to those who may be considered “super recognizers” (Russell, Duchaine, & Nakayama, 2009).

At least superficially, individual differences in the capacity to memorise and recall faces may be suggested to be a cognitive ability, given that it is similar in nature to other types of well-established cognitive abilities such as short-term memory (Gsm) and visual–spatial ability (Gv): two lower-order constructs known to be associated with general intelligence (g; Carroll, 1993). To-date, the empirical research relevant to the association between face recognition ability and intelligence is very mixed. Some research suggests that there is a substantial association between face recognition ability and other cognitive abilities (e.g., Hildebrandt, Wilhelm, Schmiedek, Herzmann, & Sommer, 2011). By contrast, others have contended that face identity recognition ability is a construct completely distinct from other cognitive abilities, including g (Wilmer, Germine, & Nakayama, 2014).

Arguably, previous investigations may be suggested to be limited, as they have not administered a comprehensive battery of cognitive ability tests, or they have not administered the most commonly administered measure of face recognition ability, the Cambridge Face Memory Test (CFMT; Duchaine & Nakayama, 2006). Consequently, the purpose of this investigation was to estimate the latent variable association between face recognition ability and other cognitive abilities, including g, through administration of a battery of cognitive ability tests and the CFMT.

Although it has been stated that all adult humans are experts at face recognition (Haxby, Hoffman, & Gobbini, 2000), the empirical research suggests that there are, nonetheless, a non-negligible amount of individual differences in the capacity to recognise faces. For example, individual differences in the capacity to recognise faces are apparent in the distribution of scores associated with the CFMT (Duchaine & Nakayama, 2006). The CFMT is the most commonly used test of face recognition ability (Cho et al., 2015). The items within the CFMT (72 in total) consist of photos of faces displayed on a computer monitor. The photos are ellipsoid in shape such that they exclude characteristics such as the model's hair and neck/clothes. Additionally, the models are not wearing make-up or jewelry. Consequently, the participant viewing the images cannot rely upon non-intrinsic characteristics of the face for the purposes of memorisation.1 For each trial, the participant must first memorise three faces on a computer screen over a period of 20 s, after which the faces disappear from the screen. Then, another series of three faces appear on the screen and the participant must identify which one of the three faces was presented during the memorisation phase. Because the test phase within the CFMT occurs essentially immediately after the memorisation phase, a score on the CFMT is probably best considered as an indicator of short-term face memory, rather than long-term face memory. Also, note that for each CFMT item, there is a correct response alternative, and the participant must select one of the three faces. Thus, the CFMT is arguably not susceptible to response biases (e.g., tendency to respond “haven't seen”).

Short-term face recognition ability has been found to be a dimension associated with a moderate amount of variability. For example, based on a university sample (N = 50), the CFMT has been reported to be associated with a mean of 57.92 and a standard deviation of 7.91 (Duchaine & Nakayama, 2006), which corresponds to a coefficient of variation of .14 (7.91 / 57.92 = .14). Based on a larger sample recruited from the general community (N = 107), Bowles et al. (2009) reported a mean of 54.6 and a standard deviation of 9.4 on the CFMT, which corresponds to a coefficient of variation of .17. For the purposes of comparison, Gignac (2015) reported a coefficient of variation of .19 for digit span forward, across several normative samples. Consequently, with respect to variability, short-term face recognition ability, as measured by the CFMT, is very comparable to serial recall of digits — a cognitive capacity well-known to be associated positively with g. Thus, short-term face recognition ability may be considered as a possible correlate of g, as it shares approximately the same amount of variability in the normal population as other indicators of short-term memory.

It has been well established that short-term memory capacity is related positively to g (Bachelder and Denny, 1977, Gignac and Watkins, 2015, Miller and Vernon, 1992). Within the Cattell–Horn–Carroll (CHC) three stratum model (McGrew, 2009), short-term memory (Gsm) is known as one of the nine broad (stratum II) factors, alongside fluid intelligence (Gf), crystallised intelligence (Gc) and processing speed (Gs), for example (Carroll, 2003). Based on the Wechsler Adult Intelligence Scale — IV (Wechsler, 2008) normative sample (N = 2200), Gignac (2014) found that a Gsm lower-order factor was associated with g at .84; a result replicated closely with Wechsler Intelligence Scale for Children — V (Wechsler, 2014) normative sample (Gignac & Watkins, 2015). In another investigation based on a combination of the Differential Ability Scales (Elliott, 1990) and the Woodcock–Johnson Tests of Cognitive Abilities-III (Woodcock, McGrew, & Mather, 2001), Sanders, McIntosh, Dunham, Rothlisberg, and Finch (2007) reported an association of .61 between Gsm and g. Additionally, a lower-order visual processing factor (Gv) was reported to be associated with g at .76. Similarly, Reynolds, Keith, Fine, Fisher, and Low (2007) reported Gsm and Gv associations with g of .70 and .83, respectively based on the KABC — II (Kaufman & Kaufman, 2004). Thus, as it may be suggested that the completion of short-term face recognition ability tests involves short-term memory and visual processing processes, it is plausible to suggest that short-term face recognition ability may be related to g.

Typically, individual tests of short-term memory capacity are observed to relate to g moderately (.30 to .50), rather than very appreciably in magnitude (.60 to .70). For example, digit span forward has been shown to relate to g at approximately .40, based on a bifactor model of the Wechsler Adult Intelligence Scale (Wechsler, 2008) normative sample (Gignac and Weiss, 2015). Thus, serial recall for digits (verbal memory) shared approximately 15% of its variance with g. Somewhat more appreciably, backward digit span, which is considered to involve some working memory capacity processing, was found to be related to g at .48.

Measures of visual memory have also been found to relate to g moderately. Consider, for example, the Rey–Osterrieth Complex Figure test (Osterrieth, 1944, Rey, 1941), which requires participants to copy a visually displayed complex figure on paper with a pencil. After a particular period of time, during which the participant completes other tasks, the participant is requested to re-draw the complex figure without forewarning (delayed recall). Higher scores are achieved contingent upon the accuracy with which a participant recreated the complex figure. Based on a higher-order model of cognitive abilities, Irwing, Booth, Nyborg, and Rushton (2012) found that the Rey-Osterrieth Complex Figure test was associated with g at .32 and .19 (Schmid–Leiman decomposed) as a cross-loading indicator of GvGf and Gsm/Glr, respectively. Thus, visual memory, as measured by the Rey–Osterrieth Complex Figure test, shared approximately 14% of its variance with g (.322 + .192 = .14). In another investigation, Reynolds, Keith, Flanagan, and Alfonso (2013) reported the results associated with a cross-battery higher-order model of intelligence, which included the Picture Recognition subtest from the Woodcock–Johnson III (Woodcock et al., 2001). Picture Recognition was found to load onto an associative memory lower-order factor at .58. The associative memory lower-order factor loaded onto g at .82. Thus, based on a Schmid–Leiman decomposition of the higher-order effects, Picture Recognition was found to be associated with g at .48.

Given that face recognition ability may be, at least qualitatively, classified as a construct relevant to memory and visual cognitive processes, it may be suggested that individual differences in performance on face recognition ability tests (e.g., the CFMT) would be a representative of cognitive ability, at least to some degree. Theoretically, in order for face recognition ability to be classified as a cognitive ability, it would arguably have to be demonstrated to correlate positively with other well-known cognitive abilities (e.g., Gf, Gc, Gsm). Additionally, and relatedly, it would be expected that face recognition ability would share variance with g. To-date, only a relatively small amount of empirical investigations have examined the association between face recognition ability and other cognitive abilities.

Davis et al. (2011) examined the association between face recognition ability and intelligence through administration of the CFMT and the Culture Fair Intelligence Test (CFIT; Cattell, 1963). Based on a sample of university students (N = 63), Davis et al. reported a correlation of −.08 between the CFMT and the CFIT. Thus, individual differences in face recognition ability were interpreted to be unrelated to non-verbal fluid intelligence. It will be noted, however, that the sample was found to be associated with a CFIT mean of 122 (the CFIT normative sample mean is 100 with an SD of 15). Consequently, performance on only a small number of items would have discriminated between many of the participants, as the CFIT subtests consist of only 10 to 14 items. Additionally, an estimate of intelligence based on, essentially, a single test may be considered rather limited. Jensen (1998) recommended the administration of nine tests for the purposes of estimating general intelligence. Finally, a sample size of only 63 is arguably insufficient to help support null hypothesis conclusions, convincingly.

In a very similar investigation, Palermo, O'Connor, Davis, Irons, and McKone (2013) reported a correlation of −.01 between the CFMT and the CFIT, based on a sample of mostly undergraduate university students (N = 80). Thus, again, the results were interpreted to suggest that there was an absence of evidence for an association between face recognition ability and non-verbal fluid intelligence. Consistent with Davis et al., however, the sample was found to be associated with a CFIT mean of 123, which suggests that the CFIT may not have been an appropriate discriminator of intelligence for the sample. Both the Davis et al. (2011) and the Palermo et al. (2013) samples were highly selected, as they were based on third-year university students at a highly rated university (R. Palermo, personal communication, December 7, 2015). Additionally, only one test of intelligence was administered, which precluded the possibility of estimating g.

In contrast to Davis et al. (2011) and Palermo et al. (2013); Peterson and Miller (2012) administered the CFMT and two subtests from the Wechsler Abbreviated Scale of Intelligence (WASI; Wechsler, 1999): Vocabulary and Matrix Reasoning. Based on a sample of 45 university students, Peterson et al. reported a correlation of .21 between the CFMT and the Matrix Reasoning subtest. However, the correlation was not significant statistically (p = .166), which may not be surprising, considering the lack of statistical power associated with an analysis based on a sample size of only 45 participants. The correlation between the CFMT and Vocabulary was reported at .01. Thus, again, there was no compelling evidence to suggest that face recognition ability was related to intelligence.

In a recent review of the above empirical research, Wilmer et al. (2014) reported a mean correlation of .01 between the CFMT and “general abilities” (pp. 2), despite the fact that none of the investigations measured general intelligence. Wilmer et al. (2014) concluded that face recognition ability “…dissociates almost completely from standardized IQ tests.” Prior to the publication of the above empirical research, Bowles et al. (2009, pp. 452) expressed a very similar view, based principally upon indirect evidence: “…intelligence probably does not affect face memory (CFMT)…” Thus, the common view in the literature is that face recognition ability is completely distinct from other cognitive abilities, and, for this reason, may be considered “special” (Shakeshaft & Plomin, 2015, pp. 12890). Arguably, however, all of the investigations reviewed above may be suggested to have been seriously limited.

First and foremost, all of the investigations estimated intelligence with either one test or a maximum of two. Consequently, it would be unjustified to suggest that the previous investigations yielded an adequate estimate of g, as a minimum of nine relatively diverse cognitive ability tests has been suggested for such purposes (Jensen, 1998). An additional benefit associated with the administration of several cognitive ability tests is that it affords the opportunity to estimate associations between latent variables. Latent variables are not contaminated by measurement error (Fan, 2003). Consequently, the observed effects between latent variables (e.g., correlations) are not attenuated due to measurement error (Nunnally & Bernstein, 1994). Finally, none of the samples in the above investigations were large, nor were they representative of the population (university students). In light of these limitations, it may be contended that the conclusion that face recognition ability is completely dissociated from intelligence is premature.

In addition to the above intelligence and CFMT research, there are also a small number of studies that have examined the association between intelligence and measures of face recognition ability less well-known than the CFMT. For example, Hildebrandt et al. (2011) investigated the age moderating effects on the association between face recognition ability and g, as measured by their own developed face recognition tasks (N = 448). Unfortunately, Hildebrandt et al. (2011) did not report the standardised effect between their general intelligence latent variable and the face recognition ability latent variable. However, they stated in the discussion that general intelligence “…accounted for only about half of the variance of face perception and face memory factors” (p. 711). Such a statement would imply that the standardised effect (standardised beta weight) was approximately β = .70. Similarly, based on a bifactor model of intelligence and a latent variable defined by three face memory tests, Wilhelm et al. (2010) found that 48% (β = .69) of the variance associated with a memory for faces latent variable was accounted for by a combination of g, object cognition, and immediate and delayed memory. These results are in stark contrast to the N-weighted mean correlation of .01 between cognitive abilities and the CFMT reported by Wilmer et al.’s (2014) review, which did not include the Hildebrandt et al. (2011) or the Wilhelm et al. (2010) studies.

The Hildebrandt et al. (2011) results suggest that face recognition ability may be a very appreciably associated with g. It is worth noting, however, that the g factor modeled by Hildebrandt et al. (2011) was composed of nine subtests, five of which were tests of memory, and three of which were tests of processing speed. The remaining test was a short-form (16 items) of Raven's (Raven, Court, & Raven, 1979). Consequently, given the heavy weighting toward memory tests in the battery, an effect of .70 between the general intelligence factor and face recognition ability, which is itself a test of memory, may be considered substantially upwardly biased. Nonetheless, the results of Hildebrandt et al. (2011) do suggest that face recognition ability may possibly be related to g positively.

In light of the above, the firm conclusion drawn by Wilmer et al. (2014) in relation to face recognition ability and intelligence may be suggested to be premature. By contrast, the effects reported by Hildebrandt et al. (2011) may be considered excessively large. Consequently, the purpose of this investigation was to estimate the association between face recognition ability, as measured by the CFMT, and a relatively diverse battery of cognitive ability tests via latent variable modeling. Based on previous research relevant to relatively narrow memory tests and intelligence, it was hypothesized that face recognition ability would associate with g positively and moderately (≈.30 to .50).

Section snippets

Sample

The total sample consisted of 211 participants (68% female) who spoke English as a first language. The age range of the participants was 17 to 35 (M = 19.8, SD = 2.9). The original sample contained an additional 12 participants over the age of 35 years. However, as age-based norms were not available for most of the tests used in this investigation, they were omitted from the final sample. The participants were recruited principally from a first-year undergraduate unit within a psychology program at

Results

As can be seen in Table 1, most of the correlations between the CFMT subtests and the other cognitive ability subtests were positive in direction and statistically significant (p < .05). In particular, it was noted that the correlations between the CFMT2 subtest (novel faces) and the two Culture Fair Intelligence subtests were r = .14 (p = .040) and r = .19 (p = .006), respectively. Additionally, the largest numerical correlation with the CFMT2 subtest (novel faces) was with the visual spatial span

Discussion

The results of this investigation suggest that face recognition ability, as measured by the CFMT, was related positively to other cognitive abilities, including Gf, Gsm, and VL. Additionally, based on a higher-order model, face recognition ability was related to g moderately at .34. Thus, the hypothesis that face recognition ability would relate to g in manner similar to other memory span abilities was supported.

In contrast to previous studies (Davis et al., 2011, Palermo et al., 2013, Peterson

Conclusion

Individual differences in face recognition ability is an interesting construct, in part, because the task of recalling previously perceived faces is performed on a very regular basis in a typical person's day-to-day life. Furthermore, a high level of face recognition ability may be speculated to be associated with both social and career-related benefits, perhaps even independently of g. Researchers are encouraged to investigate such hypotheses (and more) with face recognition ability

References (78)

  • J.V. Haxby et al.

    The distributed human neural system for face perception

    Trends in Cognitive Sciences

    (2000)
  • P. Irwing et al.

    Are g and the general factor of personality (GFP) correlated?

    Intelligence

    (2012)
  • W. Johnson et al.

    Just one g: Consistent results from three test batteries

    Intelligence

    (2004)
  • E. Luders et al.

    Neuroanatomical correlates of intelligence

    Intelligence

    (2009)
  • D.T. Lykken

    Mental energy

    Intelligence

    (2005)
  • N.J. Mackintosh et al.

    The fractionation of working memory maps onto different components of intelligence

    Intelligence

    (2003)
  • K.S. McGrew

    CHC theory and the human cognitive abilities project: Standing on the shoulders of the giants of psychometric intelligence research

    Intelligence

    (2009)
  • L.T. Miller et al.

    The general factor in short-term memory, intelligence, and reaction time

    Intelligence

    (1992)
  • M. Peters et al.

    A redrawn Vandenberg and Kuse mental rotations test-different versions and factors that affect performance

    Brain and Cognition

    (1995)
  • M.R. Reynolds et al.

    A cross-battery, reference variable, confirmatory factor analytic investigation of the CHC taxonomy

    Journal of School Psychology

    (2013)
  • T. Sekiguchi

    Individual differences in face memory and eye fixation patterns during face learning

    Acta Psychologica

    (2011)
  • N. Unsworth

    Consistency of attentional control as an important cognitive trait: A latent variable analysis

    Intelligence

    (2015)
  • J.L. Arbuckle

    Amos (version 21.0)

    (2012)
  • D.C. Bowles et al.

    Diagnosing prosopagnosia: Effects of ageing, sex, and participant–stimulus ethnic match on the Cambridge Face Memory Test and Cambridge Face Perception Test

    Cognitive Neuropsychology

    (2009)
  • S. Carey et al.

    Development of face recognition: A maturational component?

    Developmental Psychology

    (1980)
  • J.B. Carroll

    Human cognitive abilities: A survey of factor-analytic studies

    (1993)
  • R.B. Cattell

    The IPAT culture fair intelligence scales 1, 2 and 3

    (1963)
  • S.J. Cho et al.

    Item response theory analyses of the Cambridge Face Memory Test (CFMT)

    Psychological Assessment

    (2015)
  • K.A. Dalrymple et al.

    Guidelines for studying developmental prosopagnosia in adults and children

    Wiley Interdisciplinary Reviews: Cognitive Science

    (2016)
  • J.M. Davis et al.

    Individual differences in the ability to recognise facial identity are associated with social anxiety

    PloS One

    (2011)
  • H.W. Dennett et al.

    Face aftereffects predict individual differences in face recognition ability

    Psychological Science

    (2012)
  • B. Duchaine

    Developmental prosopagnosia: Cognitive, neural and developmental investigations

  • C.D. Elliott

    Differential ability scales

    (1990)
  • X. Fan

    Two approaches for correcting correlation attenuation caused by measurement error: implications for research practice

    Educational and Psychological Measurement

    (2003)
  • N. Furl et al.

    Fusiform gyrus faces electivity relates to individual differences in facial recognition ability

    Journal of Cognitive Neuroscience

    (2011)
  • F.B. Furlow et al.

    Fluctuating asymmetry and psychometric intelligence

    Proceedings of the Royal Society of London B: Biological Sciences

    (1997)
  • R.E. Galper et al.

    Recognition memory for photographs of faces

    The American Journal of Psychology

    (1971)
  • G.E. Gignac et al.

    Digit span is (mostly) related linearly to general intelligence: Every extra bit of span counts

    Psychological Assessment

    (2015)
  • R. Gonzalez et al.

    Testing parameters in structural equation modeling: Every "one" matters

    Psychological Methods

    (2001)
  • Cited by (34)

    • The association between intelligence and face processing abilities: A conceptual and meta-analytic review

      2023, Intelligence
      Citation Excerpt :

      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.

    • On the nonlinear association between intelligence and openness: Not much of an effect beyond an average IQ

      2020, Personality and Individual Differences
      Citation Excerpt :

      Processing speed was measured with two tests (Connections) similar to Trails-A and Trails B (Reitan, 1958) adapted for online administration (a composite processing speed variable was created by averaging the two scores together). Finally, crystallised intelligence was measured with a 20-item multiple-choice test known as the Advanced Vocabulary Test (AVT; Gignac, Shankaralingam, Walker, & Kilpatrick, 2016). In this sample, the descriptive statistics internal consistency reliabilities (coefficient alpha) were as follows: APM α = 0.74 (M = 4.74; SD = 2.69); Letter-Number Sequencing α = 0.90 (M = 10.54; SD = 4.44) Connections α = 0.83 (M = 27.65; SD = 11.82); and AVT α = 0.84 (M = 13.52; SD = 4.26).

    View all citing articles on Scopus

    Thanks to Romina Palermo, Linda Jeffery, and Laura McLaughlin Engfors for comments during the preparation of this manuscript.

    View full text