Jeffery P. Braden. (1994). Deafness, deprivation, and IQ. Springer.
The book is a compilation of studies on deaf people, which concludes that cultural deprivation due to deafness lowers verbal IQ but not nonverbal IQ. Braden sought to prove Arthur Jensen wrong about his conclusions on the genetic component in racial differences in IQ. At the end, his research culminated in a trauma well known to scientific history, namely, his perfectly good theory was ruined by his data. Being born deaf does not affect g. And genetic theories are the most powerful arguments to account for the pattern of the data.
Not too many ways have been devised to evaluate the impact of environmental deprivation. Typically, it is done by multiple regression, holding constant environmental variables. But genetic effects are also partialled out, and such variables do not capture all of the existing environmental factors. Adoption studies of environmentally deprived children coming into healthy families (Capron & Duyme, 1989, 1996; Duyme et al., 1999) are helpful but difficult to obtain.
The study on deaf people is another way to evaluate the strength of the environmental hypotheses. The deaf belong to the deaf community, subculture. Deaf children are raised in families who belong to another cultural group (p. 47). It is analogous to an adoption study, except that the within-family environment, rather than within-family genetics, is the primary characteristic that is manipulated. This would be an effective way to test the additive genetic model assuming there is no link between inherited deafness and intelligence (by disaggregating genetically and nongenetically deaf groups) and the trauma leading to deafness does not include neurological sequelae that affect intelligence (p. 194). Deaf people afflicted by deafness due to medical trauma are not expected to have lower genotypes for intelligence. The deaf-hearing IQ gap is expected to be purely environmental. Consequently, data on deaf children offers a quasi-experimental study of the environmental effects on IQ. With the adequate sibling and twin data, a thorough investigation of the additive genetic model is possible (pp. 171-172).
The state of research is that deaf and normal-hearing people perform similarly when the language demands, but not the intellectual demands, of the tasks are reduced. Verbal language is therefore not a mediating factor (p. 9). The language barrier is so serious that Binet & Simon (1914), the inventors of modern intelligence tests, have declared long ago that verbal IQ was absolutely impracticable for deaf people (pp. 35, 77).
More than 90% of deaf children in the Western hemisphere are born into families in which both parents have normal hearing (p. 26). The parents may be psychologically unable or unwilling to take steps to alter the way in which they communicate with their deaf child. Even the enthusiastic parents require some training in the use of gestural language. This results in delay and deficit in language exposure for deaf children (p. 28). Parental practice is such that the parent-child interaction in the deaf child’s family is even suspected to be less effective for cognitive development (p. 45). Mother-child interaction seems to be punitive, nonsupportive and oriented toward compliance rather than understanding (p. 181). The deaf children receive little support from their normal-hearing peers, and are often subject to distinctly negative proximal social interactions (p. 48). As for peers, the only language to which they are exposed is language delivered by the deliberate, intentional efforts of others (p. 29). Thus, individuals who are not fluent in gestural systems are less likely to initiate and engage in conversation (p. 31). But deaf still store and retrieve language in a similar manner to that of hearing people (pp. 40, 80).
Hearing loss leads to language deprivation and consequenty to social isolation. Such isolation extends into adulthood (p. 48). Braden (p. 43) has even encountered one particular case of deaf adolescent (aged 15) who regularly communicates with her imaginary friend. Some psychologists can conceive that there is nothing abnormal for an isolated deaf child to create an imaginary friend.
About 64.7% of families in USA do not use (american) sign language (ASL or SL) with their deaf children. At the same time, 66% of educational programs in USA incorporate SL in their instructional curricula (p. 35). SL is used by 50% of the 6-8 year-olds while 85% of the 18-20 year-olds use SL as their primary mode of communication.
About 60% of the school-aged deaf children has a known or reported cause of deafness, of which 49% is reportedly due to some form of medical trauma, which includes maternal rubella (12%), meningitis (7%), premature birth (4%), pregnancy complications (3%), otitis media (3%), and other at-birth or adventitious causes. A large proportion of children (40%) report an unknown cause of deafness. To make things even worse (p. 42), 30.4% of deaf children whose etiology is known have additional disabilities (AD). The percentage with an additional handicap among hereditarily deaf children (17.8%) is less than the proportion for meningitic deafness (25%), matemal rubella (38.6%), and otitis media (42%). Physical handicaps (e.g., visual impairment, brain damage, orthopedic impairment) are found in 15.2% of the hearing-impaired population, whereas 21.3% have a cognitive-behavioral disability in addition to deafness. The most prevalent cognitive-behavioral disabilities are mental retardation (8.5%), specific learning disabilities (8.1%), visual problems (6.1%), and emotional/behavioral problems (5.6%). Braden argues it is likely that the true prevalence of additional handicaps is higher than what is usually reported, because it is difficult to sort out the effects of cognitive-behavioral disability (e.g., learning disability) from the effects of deafness. Some research on deaf children reporting no additional handicaps are in fact likely to include deaf children with such handicaps.
About 12% of the hearing-impaired children in North America and England are diagnosed as having an hereditary cause of deafness (p. 43). However, population genetics estimates suggest that 52% of all deaf children are deaf due to genetic causes. Current researchers have generally accepted this estimates. The wide discrepancy is attributed to inadequate methods for medical diagnosis of genetic deafness. For example, parents may pass on autosomal (i.e., not sex-linked) recessive genes (i.e., having inherited two copies of an abnormal gene in a pair) that produce deafness in their child, without having any previous cases of deafness in either family.
After this description, Braden examines closely the magnitude of the group difference in IQ scores. He has collected a large amount of studies and graphed all the data points. The samples range from 4 to 21,307 with a median of 60.5.
Given the normality of distribution in scores depicted in the above graph, we have no doubt that the performance IQ of deaf children tends to be near 100. This is the central value on which the data points tend to converge, and was derived from a total of 193 studies. The 324 distinct samples reveal a grand mean IQ of 97.14 (SDm=10.79, min.=56, max.=122). There were 199 distinct samples having standard deviations, and the grand average of SD is Msd=15.33 (SDsd=3.38). So, the mean and SD appear quite similar to the normal-hearing population, usually 100 and 15. Braden (p. 68) correlated the year of publication with study’s IQ and found a positive correlation (r=0.25) suggesting a Flynn effect in this group (see also, Bakhiet et al., 2014).
A closer inspection of the 195 reports (i.e., distinct samples) shows the IQ of deaf by type of test, as depicted in the above graph. The mean verbal IQ of 85.54 (SD=9.54) from 32 distinct samples is close to the IQ estimates usually given to blacks (Rushton & Jensen, 2010). The standard deviation of VIQ (MSD=12.84, SDSD=3.26) is also lower than among the normal population. The mean of performance IQ is 99.95 (SD=10.52). The average standard deviation derived from 119 studies of PIQ is also strikingly similar to that of the normal population (MSD=15.36, SDSD=3.11). With regard to motor-free nonverbal IQ, the mean (77 studies, M=94.57, SD=7.97) is lower but not the standard deviation (51 distinct samples, MSD=15.95, SDSD=3.64).
We first examine the different nonverbal IQ tests. Braden (pp. 81-82) suggests 4 explanations for the lower motor-free nonverbal IQ. First, the processes required for motor-free tests of intelligence are not easily demonstrated. Second, manual manipulation and dexterity skills boost the performance on this kind of test. Third, the solution of motor-free items is facilitated with verbal mediation. This is suggest by a study of Conrad (1979) in which the performance of deaf with internal language skills on this test is higher than the deaf without internal language skills. Fourth, performance IQ tests yield greater feedback than motor-free tasks, and consequently provide impulsive individuals with cues to let them know when their problem-solving strategies are unsuccessful. This in turn gives them the opportunity to correct their responses.
Braden’s (1987) earlier study showed that deaf children of deaf parents are faster (in movement times) than deaf children of hearing parents, who in turn are faster (in movement times) than normal hearing children on RT tasks. This implies that speeded, psychomotor intelligence tests (i.e., performance tests of intelligence) may produce higher IQs for deaf people than nonspeeded intelligence tests (i.e., motor-free, untimed IQ tests). Hearing (HC) and deaf children of hearing parents (HP) have equal RTs but deaf children of deaf parents (DP) deaf have faster RT. Their raw scores on Raven SPM are, respectively, 50.31, 45.71, 47.16. This implies that hearing children have an advantage on nonverbal IQ but not on RT. Braden (1987) suspected the reason behind the lower RSPM score is that the test was untimed; I have calculated the d gap between hearing and deaf children of deaf parents, which was about 5 IQ points. As Braden (1987, p. 265) noted, it was empirically found that deaf people have a deficit in information processing tasks requiring strategy and metastrategy (coordination of strategies) skills, which are learned from others, and, as a result, make poor choices about which problem-solving strategies to use, when the tests do not reward speed (of completion). Finally, Braden (1987) also found that exposure to sign language (ASL) is correlated with faster MT (r=0.180). The reason could be that the genetically deaf (e.g., DP) acquire above-average motor dexterity because dexterity and speed are practiced via the repeated and consistent use of sign language (pp. 132-133), which is ultimately transferred to performance IQ tests because they typically reward rapid and dexterous manipulation of materials. That is consistent with the finding that the genetically deaf have above-average PIQs but average or slightly below-average IQs on motor-free nonverbal tests. This view is based on the suggestion that deaf people compensate for their hearing loss by developing other skills, e.g., psychomotor. In that case, IQ tests that reward speed might provide inaccurate estimates of intelligence, although there is no good evidence that deaf people could develop such skills. In fact, orthopedic, visual, and gross motor disabilities are more prevalent among deaf people than they are among normal-hearing people (pp. 132-133). The greater prevalence of dysfunction in vision and motor skills is attributed to organic, rather than environmental, causes.
But Braden apparently changed his views in Braden et al. (1994). The authors found that deaf children (n=21) have lower scores than hearing children (n=21) on a motor-reduced nonverbal IQ test (MAT-SF) but also completed the test more quickly, even though they had similar WISC-III PIQ scores. Both groups had similar movement time (in median MT and SD of MT) in two nonverbal SIP paradigms (Odd-Man-Out and Hick). Deaf children made more errors on the Odd-Man-Out but probably not on the Hick. They argue that the best explanation is that deaf children are more impulsive, i.e., they initiate responses more quickly because they engage in less anticipation and planning. But this perspective is credible only if MT is equal in both group; this is the case in Braden et al. (1994) but not Braden (1987). So, the issue is not resolved. On the other hand, if impulsivity artificially reduces the IQ of deaf in motor-reduced nonverbal tests, their true IQ level is closer to 100 than to 95.
At this point, we need to further examine the apparent finding that the hearing status of the parents moderate IQ levels in the child. The data on parents and siblings are hopefully available (p. 99). Braden expected (pp. 52-60) that deaf children with hearing parents and hearing siblings (HP) will be depressed compared to deaf children with hearing parents and deaf siblings (HP/DS) or deaf children with deaf parents (DP). This is because in the latter group, the deaf siblings will adopt similar communication methods, thus helping each other. And the presence of more than one member of family having hearing loss certainly gives more incentive to the parents in learning alternative methods of communication.
The IQ of deaf children with unidentified parental status (mean=94.89) is lower than the IQ of deaf children of HP (mean=99.21), of HP/DS (mean=103.63), and of DP (mean=108.00). Furthermore, Braden (p. 102) reports a “significant” interaction between IQ of deaf according to parental hearing status (HP, HP/DS, DP, unknown) and the type of test (performance, motor-reduced nonverbal, verbal, unknown) with IQs of HP/DS and DP above that of normal-hearing peers on performance and unknown types of IQ. He apparently uses ANOVA on his entire collection of studies. One problem with significance test is that it does not tell us the effect size, and the difference can be “significant” only because of the large sample size.
The fact that the genetically deaf (HP/DS and DP) have IQs higher than deaf HP, despite having a more severe degree of hearing loss than HP (p. 56), seems to validate Braden’s social theory. But other theories have been proposed too. It has been hypothesized that the genetically deaf (HP/DS and DP) have less prevalence of AD because the deaf HP are usually deaf due to medical trauma. So, the higher IQ of genetically deaf would be due to lower additional disability than among the nongenetically deaf (p. 54). Several studies (p. 98) report that deaf children with additional disabilities (AD) have much lower IQs than those who don’t have AD. The IQ for deaf people is 89.80 with AD, 102.16 without AD, and 95.69 for unknown status. But Conrad and Weiskrantz (1981) do not believe that the rate of ADs among genetically deaf could be lower than in the normal-hearing population, and the data shows indeed that genetically deaf have higher ADs than hearing people (p. 54). Consequently, they propose that the higher IQ of HP/DS and DP is due to selection bias. But Braden (p. 101) rejects this hypothesis in the face of the data. The debate continued with Kusche et al. (1983) who discovered that the genetically deaf (HP/DS and DP) have higher IQ than deaf HP even after matching for hearing loss, AD, race, gender, SES, and other variables. They propose that there might be a link between genes associated with genetic deafness and genes associated with high IQ. Since then, Paquin (1992) found that both DP and their parents have above-average nonverbal IQ, which is consistent with a genetic hypothesis, in that assortative mating may lead to higher IQ among DP. Braden noted that DP have higher academic achievement. He advances two reasons (p. 102). One is that DP have an internal language base, which facilitates acquisition and storage of academic knowledge. The other is simply that DP are just more intelligent than their deaf peers, perhaps due to assortative mating effects.
Given what have been said, it seems likely that DP deaf IQs provide the more accurate estimates of deaf IQs, because they are fluent in ASL and have less additional disabilities than HP deaf.
It would be interesting to see if the mean IQ differs by tests. Braden (1994) did not report those data in the book, but these are available in his 1992 meta-analysis. The table is shown below, and is from Mayberry (2002).
It seems that deaf people obtain higher IQs when estimations come from norms based on deaf people (pp. 86-87). The mean IQ from 161 studies using norms from normal-hearing people is 95.12 (SD=9.57) while the mean obtained from 19 studies using norms from hearing-impaired people is 99.43 (SD=9.11). Braden advances three explanations. The first is that the slightly lower distribution of IQ among deaf people is sufficiently different from the IQ distribution of normal-hearing people that norms on the slightly lower group yield slightly higher IQs. The second is the age of the samples used to calculate deaf norms. For example, the norms for the Hiskey-Nebraska Test of Learning Aptitude (HNTLA), which are based on deaf children in residential schools, appear to be at least 25 years old and may have been collected in the 1920s. Due to the occurrence of the Flynn effect, the inflated IQs of deaf obtained on the HNTLA could be due to the fact that deaf people are being compared to cohorts of four generations ago. A third is that sampling methods employed in the development of deaf norms systematically skew the normative samples toward lower IQs, which in turn results in higher IQs when the normative samples are used. For clarification, the following analogy may help. If one’s height were compared to a normative sample, one’s normative height score increases as the average of the normative sample decreases. An individual normative height appears high if compared to dwarfs but small if compared to basketball players. Here, the normative samples of deaf children are neither randomly selected nor stratified, and most samples are taken exclusively from residential schools for the deaf. And the IQ of deaf children in those schools are artificially low, although these children reduce their deficit as they age. That is, the potential bias is more problematic for children samples than for adult samples.
Now, we examine verbal deficit. Two hypotheses (p. 79) can be expected. One is that verbal deficit is a function of test bias. The other is that verbal deficit reflects delayed and impoverished verbal reasoning ability. Contra Braden however, the two hypotheses are not in competition against each other. Throughout his books, test bias is referred to how the children organize the cognitive tasks they are asked to resolve (factor loading bias). There is another level of bias, which is referred to intercept bias, or group differences in difficulty (the b-parameter in IRT models). This occurs when the groups differ in the knowledge required to resolve the test. For this reason, it is also referred to cultural bias.
The studies are quite revealing. Braden cites (p. 79) a study by Conrad (1979) which shows that deaf children who used some form of subvocal, or subgestural, language for mediating behavior had better performance on tests of achievement and intelligence (verbal and nonverbal) than deaf children with no internal language for mediating behavior. Braden (p. 79) also cites a study by Miller (1984) on the Verbal scale of WISC-R. An item is presented in signed english, and if it was failed, the item is repeated up to 3 additional trials, with each successive rendition approaching the american sign language (ASL). The children obtain a score of 96.43, and Miller concludes that when children are assessed in their native language, they have a somewhat normal verbal aptitude. Braden (p. 80) cites another study by Moore (1970). It was found that deaf children’s performance on verbal reasoning tasks was largely below that of normal-hearing peers, even when they were matched for academic achievement levels. In other words, there is more than schools.
The method of administration matters (p. 83). Even for nonverbal tests. These methods include oral speech or written (or combined), total communication (speech, signs, and other methods needed to convey meaning presented concurrently or in sequence), gestured, and unknown/unreported communication methods.
We have an illustration with Goetzinger and Rousey (1957) who give the Performance scale of the Wechsler to deaf and normal-hearing children using oral and gestural administrations. When the test is given in a gestural format, the deaf performed less well than normal-hearing peers who were given the test following standard administration in speech. But when the test is given in gestural administration for both groups, there were no differences. Sullivan’s (1982) research likewise suggests that gestural or oral administrations yield lower PIQs (Wechsler) than signed or combined administration methods. A group of hearing-impaired children is assessed numerous times, each time with a different method of test administration. The method of total communication presented directly by the examiner or through an interpreter, produces higher PIQs than gestural or oral administration methods. A general overview is depicted in Table 3.2.
Measurement bias emerges when an external factor that has an impact on the performance of members in a particular group has less weight (i.e., relative importance) on the performance of members in another group (Lubke et al., 2003, pp. 551-553). Then, the lower IQs for oral/speech and written forms for deaf people is the best proof of measurement bias.
Another way to examine test bias is to estimate reliability coefficient. If the IQ test is less stable in one population, it is suspected to be biased. The mean reliability coefficient from studies reported by Braden (p. 114) is 0.83 for nonverbal IQ (SD=0.10, range 0.64 to 0.98). This is close to the estimates for normal-hearing population (Jensen, 1980, p. 272). Unfortunately, there is no study reporting reliability on verbal IQ.
Braden (pp. 91-92) says that factor analyses reveal that deaf and normal-hearing people have similar factor structure on nonverbal IQ tests. However, he says that such conclusion differs depending on the heterogeneity of the age groups (e.g., aged 2-4 versus aged 6-16). For example, deaf persons aged 14-18 have different factor structure than normal-hearing persons aged 6-9. But the structures are found to be similar in homogeneous age groups. It seems that deaf children lag behind normal-hearing peers in the differentiation of intellective abilities over the age span. Longitudinal differences in nonverbal intellectual structures exist between deaf and normal-hearing children at early ages, but these differences disappear over time. Starting at age 11, the structures of nonverbal tests become similar. Another indication that deafness is unrelated with nonverbal IQ comes from factor analysis (p. 95) showing that verbal IQ and academic achievement together with hearing loss load on a common verbal factor. But hearing loss does not load on nonverbal factor. This result corroborates with other correlational studies (pp. 121, 125). There is no correlation between nonverbal IQ and degree and onset of hearing loss, as the mean correlation from 12 studies is 0.05. The mean correlation from 4 studies using verbal IQ is -0.49.
One last question not answered yet concerns intercept bias, i.e., the presence of difference in means of subtest scores when the groups are equated in latent factor scores (and eventually in factor loadings). The few studies using modern techniques produced no evidence of such bias (Maller, 1997, 2000).
Several studies (p. 94) indicate that children who acquire deafness before age 5 have much lower verbal IQ and scholastic achievement than those who acquire deafness after age 5. This is not surprising, as the latter group surely had more early learning opportunities.
Braden investigates whether the type of school matters for cognitive development among deaf children (pp. 96-97). The institutions considered are residential programs (which are treatment programs aimed to help child’s behavioral problems under the clinical supervision), day schools, mixed (combined residential and day programs), universities, clinics, public schools (with no special programs), unknown settings.
Longitudinal studies are rare (only 5) and all sampled children from residential programs. And 3 of them show IQ gains over time. One study by Braden et al. (1993) shows that selective placement seems to have taken place because the (performance) IQ of deaf is lower in residential programs than on day programs. However, the children placed on residential programs show large PIQ gains but not the children placed on day programs. Regression to the mean (the tendency of lower-scoring group to score higher at re-test because of measurement error) seems to be an insufficient explanation for these trajectories.
There is no convincing evidence that these estimates are due to sampling bias (pp. 134-135). It is true that 50% of the studies sample children in residential schools, versus 11% for children in public school (commuter) programs, and 10% for children in non-educational settings. The over-representation of deaf children in residential schools tends to lower their scores because the children in residential schools have additional disabilities (AD). However, given that residential schools enroll more deaf children of deaf parents than public schools, this must raise their IQ. Braden conducted meta-analytic comparisons across studies that suggest no “significant” relationship between sources of samples of deaf children and their IQ. If this “non-significance” can be trusted, it contradicts the evidence that children in day and residential programs have different IQs. There is no definitive answer to this difficult question.
However, intelligence tests in deaf children may suffer from insufficient construct validity. Braden (p. 120) found 50 reports of correlations between IQ tests, and the mean of these correlations is modestly lower (r=0.68, SD=0.18, range between 0.16 and 0.91) than the mean correlation (mean r=0.67, mean of the median r=0.77) in normal population (Jensen, 1980, pp. 314-315). The range of the correlation seems much larger among deaf children, but one can easily expect the sample size is small. This can introduce random measurement errors. On the other hand, Braden speculates (p. 120) that a perusal of studies suggests that smaller correlations are associated with studies using pantomime administration or restricted variability in sampling (e.g., correlations between tests for mentally retarded deaf people). In fact, studies using more appropriate administration methods and samples yield moderate to high correlations between tests.
Studies on IQ correlates with academic achievement are important. A marked difference between hearing and deaf populations can only mean there is predictive bias, and hence test bias. Braden found only a few studies (p. 123) reporting predictive validity coefficients for nonverbal and performance IQ tests, but they are well within the range of values expected for normal-hearing people (i.e., rs>0.50). More studies is needed.
Braden also (p. 124) found 153 reports of concurrent validity, in which IQ was correlated with achievement, teacher ratings, grades, or other indices that could be related to intelligence. The mean correlation is quite acceptable (r=0.42) but the range of correlation is large (-0.06 to 0.89). Some criteria have stronger correlations with IQ than other criteria. For example, the average correlation between spelling achievement and a composite achievement index (mean r=0.29) is less than the average correlation between IQ and a composite achievement index (mean r=0.53). Correlations between IQ and achievement also vary as a function of the type of IQ test. The average correlation between verbal IQ and all concurrent criteria (mean r=0.52, N=26) is higher than the average nonverbal IQ-criteria correlation (mean r=0.42, N=132). The verbal IQ-achievement correlations are higher than nonverbal IQ-achievement correlations in all three academic achievement areas (i.e., reading, mathematics, and language) where verbal IQ and nonverbal IQ data are available. This conclusion is also supported by studies contrasting verbal IQ to nonverbal IQ as a predictor of concurrent achievement within the same sample.
There are at least two viable explanations for the finding of higher verbal IQ-achievement correlations in deaf people (which is also true in normal population; Jensen, 1998, p. 279; Postlethwaite, 2011). The first is that verbal IQ and achievement tests have a greater overlap in content, have shared effects of verbal reasoning, because verbal reasoning underlies both verbal IQ and achievement scores. This would result in a stronger correlation between the two tests. The second is that verbal IQ may covary with achievement more than nonverbal IQ simply because both verbal IQ and achievement are related to a third variable (hearing loss).
Other factors also affect concurrent validity coefficients, and in particular, IQ-achievement correlations. One such factor is restriction of range in the criterion variable. It may occur when grade equivalents or other nonstandard scores are used to estimate deaf children’s achievement, because grade equivalents are not normally distributed in samples of deaf people. There is a substantial restriction of range in grade equivalents at the lower grade levels, and a serious attenuation of the IQ-achievement correlation when samples span large age ranges. For example, a bright first grader might earn a grade equivalent of 3.0 on a reading test, whereas a below average teen might earn the same 3.0 grade equivalent. The inclusion of young, talented children with older, less successful children attenuates the IQ-achievement relationship if grade equivalents are used. This is probably true. When grade equivalent scores are replaced with normative-based achievement scores, correlations between IQ and achievement improve to a large degree (p. 125). Also, correlations between IQs and raw scores on achievement tests, when controlled for age, are considerably higher than correlations between IQ and grade equivalents in samples of deaf children. Therefore, low correlations between IQ and achievement reported in some studies of deaf children may have more to do with the metric used to represent achievement. Substituting an appropriate, age-based metric for achievement eliminates the measurement problems associated with grade equivalents. Conversely, some of the unusually high correlations are based on samples with excessive variation in IQ and achievement. Such samples are typically composed of children varying widely in age and ability, but for which intelligence is typically represented as a mental age. This concerns studies that select small samples of children ranging from preschool to high school ages.
Given the similarity in IQ-achievement correlation, one would think there won’t be any external bias (pp. 126-127). And yet, there is. Because deaf people score 1 SD below normal-hearing people on verbal IQ and achievement, there won’t be any predictive bias if based on a common regression slope (and the regression slope is determined by the correlation). But simply because nonverbal IQ is similar between the two groups, while scholastic achievement is not, the prediction of achievement using nonverbal IQ will make it a biased predictor. The reason is due to the difference in intercepts; the two groups have similar means in one variable but different means in the other variable. The nonverbal IQ would overpredict achievement. Of course, this is just a purely statistical phenomenon. To achieve unbiased prediction, either nonverbal test should be at 85 or achievement test should be at 100. But contra Braden, if we have every reasons to believe achievement is biased, there is no logical reason to affirm that nonverbal test is a biased predictor.
Having reviewed the research on deaf children, Braden evaluates the strength of the hypotheses proposed to explain group differences in IQ. Genetic models seem to have a better fit to the data.
Perhaps what is the most popular theory assumes that early disadvantages (initially caused by hearing loss) will magnify group differences over time. Deaf children should fall further behind normal-hearing peers as age increases, due to the cumulative effects of deprivation. Cross-sectional studies (p. 158) show that deaf children fall further behind normal-hearing peers at each advance in chronological age in terms of academic achievement. Braden says that on absolute terms, the cumulative effect (i.e., deprivation) on deaf is substantially larger than on black children’s IQ. However, as one should know, cross-sectional analysis is a very inefficient way to study growth curves. Braden (p. 159) cites several longitudinal studies that agree with the cumulative deficit hypothesis, but the effects are quite varied. But he has noted a smaller deaf-hearing achievement differences in more recent studies than in older studies. One possible explanation is that the education of deaf children has improved relative to the normal-hearing children. Even though most deaf children are still functionally illiterate, the deaf adolescents complete 12 or more years of schooling.
There are indeed plenty of cross-sectional studies for nonverbal IQ but the results are far less consistent. One possibility is the kind of test used (p. 159). Performance IQ shows no cumulative deficit. Motor-free tests of nonverbal intelligence shows the cumulative deficit, but the deaf adolescents aged 17-20 are closer to normal-hearing peers than are deaf children aged 9-16 of years. Either there is a catching up effect or such reduction is due to attrition (dull deaf students leave the school, and as a consequence raise the nonverbal IQ of the other students). There is no way to know for sure.
Another proposed one is that the parents of black children do not ask abstract questions or require explanations for a response (pp. 151-155). In other words, they use some sort of restricted code that does not encourage abstract thinking. For example, US blacks frequently use “be” for all forms of the verb “to be” so that “he is walking” is appropriately expressed in the Standard Black Dialect as “he be walking”. This is said to account for the black-white IQ gap, and is a theory suggested by Nisbett (2009, p. 116). Such model is said to apply to deaf-hearing comparison as well. If it were true that speaking with children enhances cognitive development and g, that would have caused a general impairment in all cognitive domains.
Polemic theories of test bias argue that IQ test is an unbiased predictor of academic achievement because they both measure knowledge of the dominant culture to the same degree (pp. 143-145). For this reason, IQ will show no statistical bias. Deafness restricts knowledge of the dominant culture, such that the absence of cultural deprivation on nonverbal IQ refutes this theory (which holds if we consider culturally loaded achievement and verbal tests only). By the same token, the theory cannot explain the black-white IQ gap. One possible counter would be the statement that deaf children, due to their condition, spend more time playing with puzzles, blocks, i.e., the kind of activities suspected to improve nonverbal intelligence. In other words, deaf people will have greater access to the type of knowledge measured on nonverbal tests. First, there is no evidence that such activities are related to nonverbal ability (Sigal & McKelvie, 2012). Second, the parents tend to insist on the child being fluent in their language (p. 145). Deaf people are effectively deprived of any kind of stimulating environments.
Braden next considered the additive and non-additive genetic models. The fact that nonverbal IQ is little affected by large environmental shift implicated in deafness suggests that most variation in nonverbal IQ comes from additive genetic effects (p. 173). Braden (pp. 175-176) cited a study by Paquin (1992) which has directly tested the genetic hypothesis. The correlation between deaf parents and deaf children in nonverbal IQ was close to the predicted relationship of 0.50. Thus, with regard to DP, the additive genetic hypothesis is tenable. Unfortunately, there seems to be no other studies of this kind with or without other deaf groups. Paquin (1992) has also tested the patterns of regression to the mean. It was expected that deaf children will regress to the mean (population) to the same extent as the normal-hearing children (i.e., half the distance from midparent IQ to the population mean). The result is not robust. The IQs of the deaf children were only slightly lower than their deaf parents.
There is still a question unanswered in the additive genetic model, such as why genetically deaf have a nonverbal IQ above nongenetically deaf and normal-hearing persons. One possibility is assortative mating. People with high(er) intelligence mate with each other. Some research support this assumption. Paquin (1992) found that both DP and their parents have above-average nonverbal IQ while Kusche et al. (1983) found that the genetically deaf (HP/DS and DP) have higher IQ than deaf HP even after matching for hearing loss, AD, race, gender, SES, and other variables. But assortative mating cannot explain the above-average nonverbal IQ of HP/DS deaf because their parents would be unlikely to select each other on the basis of recessive genes for deafness. Another possibility is that alleles for genetic deafness are physically linked to alleles for higher intelligence, so that children who receive an allele for deafness are also likely to receive an allele for high intelligence (pleiotropism). Pleiotropic effects have been found to link myopia and IQ, and could function within populations of deaf people. Braden considers this single pleiotropic model untenable because DP are thought to inherit at least one allele for dominant genetic deafness while HP/DS must inherit two recessive alleles to cause their deafness (p. 173). Different alleles must be implicated in the inheritance of deafness. But see page 177.
Since Braden, other environmental theories have been proposed to account for group differences. One is the genotype-environment model of social multiplier effects proposed by Dickens & Flynn (2001) in which a superior genotype for a particular trait will become matched with superior environments for that trait, the two variables mutually affecting (upwardly or downwardly) each other. The model ultimately assumes a reciprocal causation of phenotypic IQ and environment. The fact that a poor environment impedes the development of verbal but not nonverbal abilities speaks against such story. The model assumes that the most heritable tests exhibit the strongest environmentality, and ultimately, result in more malleability. Another G-E theory is the mutualism model of van der Maas et al. (2006) which predicts that g is caused by the mutualistic causation of narrow abilities due to environments magnifying the effect of genes. Given their model, a deficit in one cognitive domain should be accompanied by a deficit in all other domains. This pattern has not been observed in the data (Willerman & Bailey, 1987). A related model has been proposed by Kan (2011, ch. 3, 4, 6) to explain the positive correlation between (sub)tests g-loadings and cultural loadings. But the magnitude of cultural loadings attributed to each (sub)test is surely an arbitrary one. A more objective cultural loading variable has been (unintentionally) provided by te Nijenhuis et al. (2014) with regard to early preschool intervention programs, Jongeneel-Grimen (2007, unpublished dissertation) with regard to adoption gain, and Franssen (2010) with regard to deprivation due to blindness and/or deafness. These cultural indices always show negative correlation with g-loadings. The main problem with all of these models is to rely on the assumption that the vector of means of each cognitive domain (denoted alpha in CFA-SEM modeling) should be treated as a continuous variable. For example, Gf and Gc must exhibit different magnitude of deprivation, which would be a positive function of their respective heritabilities. If Gc is more heritable, as Kan (2011) has concluded, then Gc would be more affected by environmental effect than is Gf. If, on the other hand, Gc is affected but Gf is not, this is sufficient to reject all these models.
Most importantly, the study of deaf people considerably reduces the array of possible causes of racial differences. There are only a few data (p. 93) on racial differences but they are quite revealing. The black deaf children perform about 0.5 to 2 SD below comparison groups in IQ. Sometimes, the comparison group is white deaf children, and sometimes it’s the normal-hearing normative data that is used as a mean of comparison. There are smaller race differences in achievement tests than on intelligence tests. The black-white deaf children differ in read and math achievement (p. 186) by -0.60 to -0.88 SD gap, which magnitude is similar to that found about normal-hearing population (Jensen, 1973, p. 249; Yeung & Conley, 2008, p. 310) but apparently much smaller than the deaf-hearing achievement gap (p. 167). Regarding IQ, Braden thinks that 1 SD difference between black and white deaf children is a safe approximation. This claim is odd because he himself reports that the mean nonverbal IQ of black deaf children from 7 studies is 75.98, ranged from 60 to 90 (p. 186). In other words, the deaf-hearing gap among blacks seems to be 10 points, but zero among whites. On the other hand, these studies sampled blacks in the southern states and Jensen (1980, p. 99) reported that the mean IQ of blacks in these regions was 80. Another reason to suspect deaf black IQs to be under-estimated is that hereditary deafness is more prevalent among whites than among blacks (pp. 59, 187). But if these factors (and some others ?) cannot help to restore black (nonverbal) IQs to the expected mean of 85, either deafness truly (but modestly) deprives black IQs or more samples need to be collected. In the first case, the hypotheses predicting that blacks will react differently to the same environments affecting whites should be taken seriously.
The (plausible) similarity with the black-white IQ difference in the normal population suggests that the developmental processes behind the cultural deprivation due to hearing loss may be similar across racial groups. That is, the impact of culture on IQ could be similar for blacks and whites. But such finding also implies that environmental factors other than those associated with deafness must bring about between-group differences in IQ (p. 187). Environmental differences between whites and blacks that cause different distributions of IQ must therefore cross the hearing barrier relatively unchanged, or interact in unexpected ways to produce the same net effect. Another implication is that such finding can be predicted by additive genetic model (p. 187) if there is no reason to suspect a three-way interaction between genotypes for race, hearing loss, and genetics.
Given the above discussion, we can derive several conclusions. If the black-white environmental difference is a subset of deaf-hearing environmental difference, and that both group comparison yields a d gap of 1 SD, it could be expected that the IQ gap between blacks and whites must have a genetic component, because blacks would be less deprived in terms of environments than are deaf people. If being deaf deprives black IQ by 15 points and that the black-white environmental difference is only a subset of deaf-hearing environmental difference, then the black-white gap must be entirely genetic. If the IQ gap in black-white deaf children is lower than 1 SD, this is consistent either with the idea that the environment separating blacks and whites is a subset of the environment separating deaf and normal-hearing groups or that the kind of environments depressing black IQs is different than that affecting the IQs of deaf people. Furthermore, if we accept the idea that the black-white IQ gap is a function of test’s g-loadedness (Jensen, 1998, ch. 11) and is “generalized” over all cognitive domains whereas environmental deprivation due to deafness is inversely related with test’s g-loadedness (Braden, 1984, 1989; Franssen, 2010) and is domain-specific, it may be expected that the environments depressing black IQs must be different rather than being a subset of those environments involving in depressing IQs of deaf people.
This has implications for group difference in heritability (pp. 183-184). If the verbal d gap between deaf and normal-hearing children is 0.96, assuming IQ within-group heritability (environmentality) is 0.60 (0.40), the between group difference in environment with respect to verbal IQ is 0.96/SQRT(0.4)=1.52 SD. Braden uses 0.96/0.4=2.4 SD, but 0.4 is expressed in terms of r-squared, which is not an effect size (Hunter & Schmidt, 2004, pp. 289-291). On the other hand, if the nonverbal d gap between deaf and normal-hearing children is 0.19, the between group difference in environment on nonverbal IQ is 0.16/SQRT(0.4)=0.25 SD. One would wonder if environmentality can solely account for such a large shift in verbal IQ. For example, an environmental SD gap of 1.52 (or 0.25) would place deaf people at the 6.4th (or 40.1th) percentile of the distribution of environments among hearing people. The key point of such argument is that whenever the actual environmental difference is insufficient to account for group difference in IQ, either the heritability in at least one group is lower or a genetic component must be involved in the IQ gap. Regarding deaf-hearing IQ gap, Braden did not expect deaf and hearing people to have different genotypes for verbal IQ, because the high correlation between verbal IQ and the degree of hearing impairment would attenuate the correlation between genotype and verbal IQ, hence lowering the heritability of verbal IQ in deaf people. Thus, he believed that the heritability of 0.60 is too high. On the other hand, the heritability of nonverbal IQ is not suspected to be depressed.
However, these estimates are misleading. If we were to estimate the expected impact of environmental effects on the black-white gap or the deaf-haring gap, we must examine the between-family (shared environmental) effects, not the entire environmental effects. Jensen (1976) estimated the within- and between-family variation to be 44% and 37%, the remaining belonged to group differences and measurement errors. Such estimations are consistent with other reports (Jensen, 1970; Willerman, 1979). But Jensen (1973) preferred to assume that half of the environmental variation has its source between the families. In this case, 1/(SQRT(0.4)/2)=3.16. That means the blacks will be at the 0.08th percentile of white environments. Given the available data, Jensen (1973, p. 169) suspected that such large environmental black-white gap is unlikely and concluded that the black-white IQ gap must have a genetic component. More recent estimates from the CNLSY79 on incomes show that blacks are located at the 16th percentile of white income distribution (explaining only 2 points in the black-white PPVT-R gap) but that an environmental composite variable based on every measures available would place them at the 9th percentile of white environments (Jencks & Phillips, 1998, pp. 115, 132). This is far too much to approach the 0.08th percentile. This prediction can be applied to deaf-hearing comparisons, but there is no data on environmental distribution. But we can imagine that the deaf-hearing gap in between-family environmental effect regarding nonverbal IQ gap, 0.16/(SQRT(0.4)/2)=0.50 SD, does not exceed the black-white environmental difference, which is suspected to be lower than deaf-hearing difference.
Cultural theories seem to have reached a dead end. In definitive, what Braden seems to believe, and correctly, I would say, is that if environments account for some portion of the black-white IQ gap, the biological-environmental explanation appears to be far more promising road.
Leave a Reply