500 words minimum, masters level, apa format, discussion question, REVIEW ATTACHMENT

Avoiding Type II Error in Assessing Lead Toxicity Plainti�s

Frank J. Dyer, Ph.D.

This article discusses statistical parallels between exces-

sive conservatism and insu�cient conservatism in

rendering forensic opinions. The elements of a tort are

reviewed and their relation to psychological and psy-

chiatric opinions is also discussed, as are psychometric

and clinical approaches to assessment of impairment

and causation in individual lead poisoned children. It is

argued that assessments in lead poisoning cases consist-

ing solely of cranial nerve examinations result in con-

siderable Type II Error. Sources of Type II Error in

research using analysis of covariance techniques to study

the toxic e�ects of lead include variance stealing, use of

excessive numbers of covariates, lack of attention to

interactions, and use of covariates that are actually

substitute measures of lead ingestion. When experts cite

nonsigni®cant ®ndings of studies of low-level lead

exposure, it inappropriately negates lead e�ects in more

severely lead poisoned plainti�s. In true experimental

studies where there is no ambiguity regarding causation,

the destructive e�ects of lead are quite clear. # 1998 John Wiley & Sons, Ltd.

Lead poisoning litigation has increased in recent years due to greater awareness of the e�ects of lead toxicity on child development and subsequent mental functioning (Dyer, 1993). Consulting in these types of cases presents special challenges to the forensic psychologist that are not signi®cant issues in purely clinical consultations. Whereas a successful clinical consultation may render an accurate diagnosis and treatment recommendations, forensic cases, especially those involving lead toxicity issues, frequently involve abstract questions of scienti®c validity of research ®ndings as applied to an individual case. In order to render e�ective consultation in

CCC 0735±3936/98/010131±15$17.50 # 1998 John Wiley & Sons, Ltd.

Behavioral Sciences and the Law

Behav. Sci. Law, 16, 131±145 (1998)

This article is based on the author's presentation as part of a symposium entitled ``Integrating Research and Practice in Forensic Psychology'' at the 104th convention of the American Psychological Associ- ation, Toronto, Ontario, Canada. Correspondence concerning this article should be addressed to: Frank J. Dyer, Ph.D., 70 Park Street, Montclair, NJ, 07042. USA.

Contract grant sponsor: the 104th convention of the American Psychological Association, Toronto, Ontario, Canada.

lead poisoning cases, forensic psychologists must buttress their expert opinions with empirical research results, as outlined in the Boulder Model of psychologists as scientist-practitioners.

The scientist-practitioner model stresses the empirical roots of psychology, as opposed to its origins as a branch of speculative philosophy. This distinction is re¯ected in William James's division of the psychological world into the tough minded and the tender minded in the beginning of this century. This categor- ization highlighted the schism between those psychologists whose priorities were more in line with the scienti®c model's conservatism in interpreting data and those whose dazzling command of rational argument generated grand theoretical systems that were as replete with persuasive conceptual brilliance as they were free of actual data.

In any research study intended to generalize results obtained on a sample to the broader population, there are two types of possible errors. The ®rst, called Type I Error, occurs when the researcher inappropriately concludes that there is a non-zero e�ect in the population when, in fact, no such e�ect exists. In other words, Type I Error occurs when statistical signi®cance is found although the null hypothesis is actually true (Hays, 1973). This type of error corresponds to what would be termed a false positive in the jargon of psychological measurement. Another commonly used term for Type I Error is alpha error. The second type of error, or Type II Error, occurs when null hypothesis is actually false but the researcher concludes that the data fail to show an e�ect. This corresponds to a false negative in psychometric terms. It is also called beta error. Contemporary psychological statistics and experimental design courses emphasize the avoidance of Type I Error, viewed as a fatal attribute of the tender minded who naively see relationships where none exist, based upon inappropriate generalization from limited data.

In this climate of scienti®c rigor, Type II Error, or failure to detect a non-zero population e�ect, receives far less attention in statistics courses than does Type I Error. Of course it is not any more scienti®c to err on the side of missing a genuine population e�ect because of excessive conservatism than to err on the side of ``®nding'' an illusory e�ect because of insu�cient alpha level concerns. In both cases the researcher has made an error and the interpretation of the results does not re¯ect the true state of Nature. This point of view is elaborated in the work of Jacob Cohen (1969; 1975), who has developed power analysis techniques that a�ord Type II protection as a complement to alpha level concerns.

PARALLELS OF TYPE I AND TYPE II ERRORS IN FORENSIC PSYCHOLOGY

Much has been written about what may be described metaphorically as avoiding Type I Error in forensic psychology. Type I Error in this context consists of overinterpretation of data to form what are in truth merely speculative conclusions that are presented authoritatively on the witness stand. Simply put, this is junk science that courts blanket criticism of forensic mental health experts (Dyer, 1996).

The forensic analog of Type II Error consists of failing to ®nd the condition or disorder or causal nexus of interest due to inappropriate reliance on theories or beliefs that obscure what is actually present in the person being examined. One

132 F. J. Dyer

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

situation typically encountered is that in which an attorney advances a theory of the litigant's condition, or in tort cases its causal relationship to the defendant's behavior, that is upheld by nothing more than distortion and by the citation of outlandish ``possibilities''. Because of attorneys' a�rmative duty to provide aggressive representation of their clients, they are more or less ethically bound to raise these marginal considerations on the outside chance that they will succeed in creating a degree of doubt that prevents the other side from meeting the applicable evidentiary standard, even though everything that the opposing experts say may be entirely accurate.

Whereas it is the attorney's job to create Type II Error through e�ective representation of the client's interests, it is also frequently the case that mental health experts engage in Type II Errors both of omission and commission. The use of assessment instruments that are simply not sensitive to the condition purportedly being assessed constitutes a common error of the former type. An egregious example of the latter is a psychiatric expert's assertion that a lead poisoned child was exposed to lead toxicity from mouthing the ``lead'' on a pencil (which is actually graphite) rather than anything stemming from the child's living conditions, which were documented by various health inspectors as having lead based paint peeling o� the walls, creating lead paint chips and lead bearing dust.

LEAD POISONING TORT CASES

The special challenges to forensic evaluators presented by lead poisoning cases proceed from the requirements of proof in personal injury, or tort, cases. The elements of a tort include duty, violation of duty, compensable damages, and proximate cause (Barton, 1985). The ®rst of these elements refers to the fact that under the law, the defendant had a duty to behave in a particular manner toward the plainti�. Examples of such duties include the duty to operate a motor vehicle in a safe and responsible manner, the duty of a landlord to keep the premises in habitable condition, and the duty of psychotherapists to refrain from acting out sexually with their patients, no matter how thoroughly they may rationalize such abuses. The second element of a tort is an alleged violation of this duty by the defendant. Examples include operating a motor vehicle recklessly, failing to main- tain premises in a habitable condition, and making sexual advances toward one's therapy patients. The third element refers to injuries or losses that the plainti� has su�ered and for which the plainti� seeks to be made whole. Compensable damages su�ered by the plainti� may include physical or psychic injury, as well as economic loss. Finally, the element of proximate cause is de®ned as the ``near cause'' or ``the straw that broke the camel's back'', linking the defendant's violation of duty to the injuries su�ered by the plainti�, even if the plainti� had preexisting conditions that were aggravated by the defendant's tortious behavior. It is the concept that ``but for'' the acts of the defendant, the plainti� would be free of injury or other loss (Barton, 1985). While psychologists are seldom called upon in this type of litigation to provide testimony regarding the defendant's duty toward the plainti� or the defendant's alleged violation of the duty, lead poisoning litigants typically employ psychologists and psychiatrists to assess damages (injuries) and, especially, proximate cause.

Avoiding Type II Error 133

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

ASSESSMENT OF COMPENSABLE DAMAGES

When examined by defense psychiatrists, infant plainti�s in lead toxicity cases frequently get an absolutely clean bill of health. This is because they are admin- istered the standard neurological examination to test for cranial nerve functioning via re¯exes and other gross responses. The cranial nerve examination assesses the functioning of the 12 pairs of cranial nerves that connect central nervous system processors with the rest of the body. These examination procedures include exposing patients to odiferous substances and asking them to identify them, examining for ptosis and appropriate dilation in response to light, application of pinpricks, heat, cold, and cotton swabs to the face, having the patient imitate movements such as wrinkling the forehead, frowning and raising the eyebrows, assessment of swallowing, assessing various tongue movements, and other tasks (Berg, Franzen, & Wedding, 1987). In the typical case cited above, the child passes all of these screening procedures and the psychiatric report comes back with a diagnosis of ``cranial nerves grossly intact''. This diagnosis has been rendered by defense psychiatric experts even in cases where the child is clearly hyperactive and distractible and has substantial impairments in verbally mediated learning. It is an egregious example of Type II Error, even though the diagnosis itself is essentially correct. The child's cranial nerve functioning is grossly intact, although there are obvious learning and behavior problems present. However, as Bellinger (1995) observes, ``Goldstein . . . proposed that lead may disrupt this [synaptic] `tuning' or `matching' process, producing a nervous system that appears grossly normal but in which the connections are `poorly chosen', perhaps producing functional impair- ment'' (p. 207). Bellinger proceeds to enumerate various destructive e�ects of lead at the cellular level that disrupt neuronal synapse connections supporting learning and behavior.

While practitioners who are accustomed to clinical or forensic cases in which a high standard of practice is adhered to may regard the above example as a rare aberration, in actual fact such distortions are very common in this type of litigation. Further, while many practitioners might regard this type of blatant distortion as something that is easily remedied by judicial scrutiny and vigorous cross examination, the unfortunate reality is that such ``medical'' testimony is quite impressive to lay jurors and resistant to cross examination in cases where the expert delivers these conclusions with an air of conviction and authority or where the expert's credentials are perceived as superior. Sukel, Bennett, and Cooper (1995), for example, found that while mock jurors were not swayed by the credentials of the expert in cases where the issues were easily understood, they were swayed by an expert's superior credentials where complex issues were the subject of testimony.

Dyer (1993) notes that there has been extremely little written about the clinical assessment of lead poisoned children. This is especially true for neuropsycho- logical assessment and as of this writing there is no published literature demon- strating that neuropsychological testing o�ers anything in the way of incremental validity over the standard clinical or school psychological assessment battery. An adequate clinical assessment of a lead poisoned child must include measures of cognitive, perceptual-motor, language, and abstract reasoning abilities. There should also be an assessment of academic achievement, or in the case of younger children, readiness skills. It is also highly desirable to include a measure of

134 F. J. Dyer

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

adaptive behavior to assess the nonintellective capacities of the child in meeting the challenges of daily living appropriate to his or her developmental level. Apart from the psychometric components of this type of assessment, evaluators should be alert to behavioral soft signs of neurological damage including hyperactivity, distractibility, motor incoordination, drooling, peculiar pencil grasp, articulation problems, and excessive dys¯uencies of speech.

ASSESSMENT OF CAUSATION

In addition to Type II Errors relating to assessment of compensable damages, a greater source of error is in the assessment of proximate cause. Contrary to some defense attorneys' assertions that the e�ects of lead on a young child's mental functioning can only be established through ``medical'' testing, experts' conclusions regarding causal relationship in these cases should be founded not only upon careful clinical observation and analysis of psychometric test ®ndings, behavioral observation, and case history data, but also on the large body of empirical literature that has developed in this ®eld in the last three decades. There is no ``medical'' test capable of assessing causal relationship of lead ingestion and a particular child' s learning problems. Certainly, medical testing can determine the child's blood lead level and level of free erythrocyte protoporphyrins (FEP), which some investigators regard as an index of tissue damage due to lead (Rabinowitz, Bellinger, & Leviton, 1989). However, the causal relationship of lead to school performance and psychometric test results is a complex matter that is best assessed through an integration of clinical and empirical ®ndings.

At the clinical level it is necessary to examine all of the causal factors impinging upon the individual child's functioning. The examiner must mentally perform what is in e�ect a one-subject analysis of covariance in order to tease out the probable causal factors that are behind whatever impairments may be disclosed by the observation and testing procedures discussed above. This analysis requires the collection of family history and developmental data to a degree that goes somewhat beyond normal clinical practice. At the conclusion of the process, all pathogenic background factors elicited by the interview and review of records are ordered in terms of rationally determined causal priority, taking into account the nature and severity of the child's observed impairments, if any.

The preferred informant for the background interview is the biological mother, if the child has been in her care. Otherwise, a family member, caretaker, or social service worker familiar with the history should be interviewed. Being that the child's genetic background is routinely introduced as an issue in this type of litigation, it is necessary to inquire as to the presence of mental retardation, attention de®cit disorder, hyperactivity, learning disabilities, organic impairment, and emotional problems in the families of both birth parents. The birth mother's prenatal care should be explored, as well as her use of tobacco, alcohol, prescription medications, and illicit drugs during the pregnancy. Any illnesses or complications during pregnancy should be discussed. A careful birth history should be obtained, including such details as whether the baby was born full term, birth weight, length of labor, complications in the delivery process, use of forceps, special medical

Avoiding Type II Error 135

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

intervention required, fetal anoxia, jaundice, breech birth, whether the umbilical cord was wrapped around the child's neck, and similar data.

It is essential to elicit a detailed developmental history to identify conditions that might provide an alternate explanation of the child's learning and behavior prob- lems. It should be determined whether the child su�ered from any unusual illnesses during infancy and early childhood and whether there were any falls, head injuries, or extremely high fevers. The examiner should record the ages at which developmental milestones, including walking, individual words, speaking in short sentences, and toilet training were achieved. A common pattern among lead poisoned children is one in which there is a stalling or decline in developmental progress, especially in the area of language skills, around the time that the child acquires mobility and is thus vulnerable to ingestion of lead paint chips or lead bearing dust.

It is not unusual to encounter a pattern in which the lead poisoned child registers normal scores on measures of adaptive behavior, apart from areas that are signi®cantly impacted by hyperactivity and distractibility, and to ®nd severe impairments in abilities measured by psychometric testing of school related mental abilities. As noted by Dyer (1993), pattern analysis of cognitive test results is something to be undertaken with great caution. With this caveat in mind, however, there does appear to be a pattern of verbal vs. nonverbal ability test results that is fairly speci®c to lead. Dyer reports that the con®guration of Wechsler Performance IQ greater than 1 standard deviation above Wechsler Verbal IQ or Goodenough- Harris Drawing Test standard score greater than 1 standard deviation above Wechsler Verbal IQ was signi®cantly more common among clinically lead poisoned subjects (as de®ned by a diagnosis of lead poisoning listed in the social history) that in either a comparison sample similar in socioeconomic status or in the WISC-R standardization sample. While the encountering of this pattern in an individual child is not by itself proof of a lead etiology, it should be regarded as consistent with such an etiology, providing support for the hypothesis of lead as the major determinant of the child's impairments. Absence of this pattern does not indicate that lead is an insigni®cant factor, as this is not a pattern that invariably occurs in lead exposed children, but one that is more frequently found in that group.

STATISTICAL ISSUES IN CAUSATION

Complementing the clinical assessment of causal factors in lead poisoning plainti�s is an extensive body of empirical research studies that statistically examine the performance of large numbers of children with varying levels of blood lead. While such studies typically show a substantial zero order correlation between blood lead level and a large number of cognitive, perceptual, language, and behavioral variables, the caveat stressed in elementary statistics courses that correlation is not causation is particularly applicable here. When other predictor variables are also introduced in these studies, the correlation between lead and the dependent variables becomes severely attenuated, sometimes to the point of falling short of statistical signi®cance. This is due to the fact that lead ingestion is also correlated with socioeconomic variables, creating interpretive problems of daunting complex- ity. It is also a phenomenon that is very frequently stressed by defense counsel to

136 F. J. Dyer

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

bolster the assertion that the infant plainti� does not have any impairments that were not present before the lead exposure and that the real causal culprit is economic and educational disadvantage.

Studies of the e�ects of lead exposure are essentially drug studies that may be compared with tests of the e�cacy of new medications, except that in the case of lead the object of the research is to determine whether there is a toxic e�ect rather than a therapeutic e�ect. Given this circumstance, researchers are ethically pre- cluded from following standard drug experimentation protocols with human subjects that permit precise assessment of the therapeutic e�cacy of medications. These procedures include double blind administration of the agent, precise dosages, and randomization in assignment to experimental and control groups. Of these various factors, it is primarily the absence of randomization that complicates research studies of lead toxicity. In a completely randomized study of the e�ects of lead there would be absolutely no question of whether any changes in the dependent variable (such as IQ, academic achievement, or perceptual-motor measures) were caused by changes in the independent variable. In the real world, however, researchers must deal with what are termed confounders or covariates, these being variables such as maternal IQ, socioeconomic level, and condition of the home environment, that are correlated with the presence or absence of lead.

The typical design of lead toxicity studies of children involves a statistical procedure known as analysis of covariance (ACV). In this design, potential confounding factors such as maternal IQ, socioeconomic level, and home environ- ment measures are entered into a statistical prediction equation using multiple regression, with blood lead entered into the equation after these covariates. This is intended to have the e�ect of setting all subjects statistically equal on the covariates, removing the e�ect of those variables from the blood lead variable. The resulting semipartial correlation of lead with the dependent variable is then assessed for statistical signi®cance.

Typically this procedure results in a much smaller semipartial, or adjusted, correlation between lead and the dependent variable than the zero order correla- tion, leading some researchers to speculate that if all of the covariates could be measured and entered into the prediction equation, the lead e�ect would disappear entirely. Needleman and Bellinger (1989) state: ``Because cognitive function is determined by multiple factors, careful investigators of lead's e�ects try to identify and evaluate those non-lead covariates which could confound. This partitioning of the variance usually, but not always, has the e�ect of reducing the size of the lead e�ect'' (p. 295). Typifying the school of thought that asserts that all lead e�ects are statistical artifacts due to confounding with socioeconomic factors, Smith et al. (1983), quoted by Needleman and Bellinger, state ``The ®ndings in this study show that if outcome measures are controlled, di�erences between lead groups on all tests become nonsigni®cant and the null hypothesis that the di�erences are not statistically di�erent from zero must be accepted'' (p. 295).

The Problem of Variance Stealing

Actually, ACV is a very highly conservative statistical procedure with a substantial Type II bias. This is inherent in the fact that where there is overlapping variance

Avoiding Type II Error 137

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

between the covariate and the main e�ect in predicting the dependent variable, 100% of the overlap is attributed to the covariate, regardless of the true state of nature. Smith (1989) states that ``the cautious convention in multivariate statistics is to attribute any shared variance to the confounder, so if no association is found after control for covariates, there is assumed to be no e�ect of lead. If a confounder is to some extent a cause of the child's lead burden, as well as being related to the outcome, then variance attributable to lead will be removed along with the `social' variance, relevant to the outcome'' (p. 41). In lead ACV studies, where as many as 20 covariates are entered into the equation before the lead e�ect, the fact that there is often a substantial correlation between each of the covariates and lead, with some of this variance shared by both variables with the dependent variable, causes the lead e�ect to shrink enormously.

Cohen (1975) employs the term ``variance stealing'' to describe situations in which variance that properly belongs to the main e�ect is erroneously attributed to the covariates. In the same vein, Needleman and Bellinger (1989) state: ``To control for such outcome measures as school placement . . . , hyperactive behavior . . . , or developmental delay . . . may be to subtract out variance which properly belongs to the main e�ect, lead'' (p. 296). In regard to socioeconomic variables, the variance stealing is particularly egregious due to the fact that lead exposure is more common in lower SES groups. Smith (1989) states that ``The clinic studies indicated that lead was not randomly distributed in normal populations, as children identi®ed by screening clinics as having high lead levels were are likely to be disadvantaged. Studies of normal populations in several di�erent countries have con®rmed this relationship of lead measures with broad social indicators of disadvantage'' (p. 17). Smith further states that ``It has been said that controlling for a large number of variables, and particularly interrelated variables, can result in overcontrol, that is removing variance due to lead, and masking a real e�ect'' (p. 40).

Cohen's (1975) counter to the ``cautious convention'' of entering covariates ®rst and then automatically attributing shared variance to the covariate is that common sense should guide the development of a hypothesized causal model to determine the hierarchy of variables to be entered into the regression equation. Cohen notes that ``We summarize this principle, then, as `least is last'Ðwhen research factors can be ordered as to their centrality, those of least relevance are appraised last in the hierarchy, and their results taken as indicative rather than conclusive'' (p. 162).

Use of Questionable Covariates in ACV Studies of Lead

In the case of ACV studies of lead e�ects, common sense dictates that because results from animal studies unequivocally document the neurotoxic properties of lead, it is the child's body burden of lead, and not social covariates, that should receive priority in the causal hierarchy. Bornschein (1985), for example, demon- strates that in at least one study scores on the Home Observation for Measurement of the Environment (HOME; a measure of conditions of the home environment commonly employed as a covariate in ACV studies) are actually a substitute for body burden of lead. Bornschein notes that prior studies showed a relationship between scores on the HOME and blood lead levels. However, certain HOME

138 F. J. Dyer

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

subscale scores predict both the amount of lead dust in the house and children's hand-to-mouth behavior, which is a signi®cant pathway of lead ingestion. Thus, the HOME-lead blood level correlation is mediated by the hand-to-mouth variable and by HOME's sensitivity to level of lead dust in the environment. In most ACV studies, however, the HOME measure is entered into the regression equation ®rst, as though it were entirely independent of lead, whereas it is actually a substitute measure both for the amount of lead dust to which the child is exposed and for the amount of lead ingestion due to hand-to-mouth behaviors. The fact that all of the overlapping variance between HOME and blood lead level is attributed to the HOME measure is ``variance stealing'', to use Cohen's (1975) term, at its worst.

Use of Excessive Numbers of Covariates

Another way in which current ACV methodology creates a Type II Error bias has to do with number of covariates employed. The concept of statistical power as elaborated by Cohen (1969; 1975) relates the number of subjects, alpha level, and assumed population e�ect size to the probability that the ®ndings will reach statistical signi®cance. The general rule is that the larger the number of subjects, the greater the likelihood that the ®ndings will reach statistical signi®cance. Likewise, the larger the number of variables studied in a single regression analysis, the less likely it is that statistical signi®cance will be achieved, if the number of subjects is held constant (Cohen, 1975, p. 160). In lead ACV studies it is the lead variable that is entered last, frequently after an inordinate number of covariates, making it unlikely that the lead e�ect will reach signi®cance simply on the basis of reduced degrees of freedom alone.

For example, the Boston lead study conducted by Stiles and Bellinger (1993) employed 10 covariates; the Port Pirie study by Baghurst et al. (1995) employed 12 covariates; the Needleman (1996) study of lead and delinquency in an inner city Pittsburgh sample employed 9 covariates. It is also of interest that a study by Pocock et al. (1989) employed 17 covariates; however, using an optimality criterion the number was reduced to 11. While Pocock and colleagues employed tests for interaction between the covariates and the lead variable, there was no mention of statistical power considerations. Hatzakis et al. (1989), in a study of Greek children residing in a lead smelting area, employed 17 covariates and still found that lead accounted for 8.1% of the total variance when entered into the regression last. Possible factors associated with this ®nding include large sample size (n�509) and wide range of lead levels. Moreover, the cultural homogeneity of the sample prevented race or other variables to steal variance from lead due to unequal exposure among ethnic groups such as that found in United States studies. Further, the fact that the sample was from a lead smelting area meant that since the presence of lead was pervasive, there was no variance stealing due to unequal exposure based on whether children lived in tenements versus more elaborate dwellings. Again, ACV procedures such as these, which found signi®cant relationships between lead and many of the dependent variables, are in fact loaded against ®nding signi®cance for the lead e�ect because of the reduction of statistical power due to the excessively large numbers of covariates.

Avoiding Type II Error 139

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

Interactions Invalidate ACV

Astoundingly, virtually all of the major ACV studies of lead toxicity were performed without any attention paid to the subject of interaction between the covariates and the lead main e�ect variable. This is of crucial signi®cance because of the logic of ACV that ``setting everyone equal statistically'' on the covariates permits assessment of the true impact of the main e�ect. However, this requires the assumption that the relationship between the main e�ect and the dependent variable remains constant across levels of the covariates on which the subjects are presumably being equalized. If the relationship di�ers across levels of the covariates, then such ``equalization'' of subjects is impossible because a funda- mental assumption of the procedure has been violated. Cohen (1975) states categorically that a signi®cant interaction between the covariates and the main e�ect invalidates ACV, which assumes homogeneity of regression.

The failure to recognize the presence of interaction e�ects is not merely a speculative criticism of current research; it is grounded in empirical ®ndings. Bellinger (1995) states that several studies have reported a signi®cant statistical interaction between social class and lead. In a study of higher SES, higher IQ children than are typically examined, Stiles and Bellinger note, ``. . . because the e�ects of early CNS insult may be expressed to a greater extent in populations at higher socioeconomic risk . . . associations between lead and neuropsychological functioning may be underestimated in this sample'' (p. 34). This observation stresses the SES-lead interaction in regard to neuropsychological measures. In spite of Bellinger's acknowledgment of a social class-lead interaction, this is not included in the sources of Type II error listed in Needleman and Bellinger's (1989) article on ACV methodology.

BLOOD LEAD LEVELS IN RESEARCH SUBJECTS AND IN TYPICAL LEAD PLAINTIFFS

The Centers for Disease Control (CDC) (United States Department of Health and Human Services, 1991) lists interpretations for 5 ranges of blood lead levels. Below 10 mg/dl blood lead is considered not to be indicative of lead poisoning. A blood lead of 10 to 14 mg/dl is described as representing a ``border zone''. E�ects of lead at this level are considered subtle and not likely to be recognizable or measurable in the individual child. The CDC recommends follow-up blood testing at this level. Blood lead of 15±19 mg/dl requires more careful followup. Children at this level are at risk for decreases in IQ up to several points, according to the CDC. Environ- mental investigation and remediation are recommended. Children with blood lead levels of 20 to 69 should have a full medical evaluation, with greater urgency indicated for levels at or above 45 mg/dl. Practitioners working with children at these blood lead levels are advised to investigate reading or other learning disability and language development. Children should be referred to a lead poisoning clinic for management and there should be an environmental investigation for the purpose of abating the lead. The CDC considers blood lead levels at or above 70 mg/dl to constitute a medical emergency, with the child critically ill with lead poisoning.

140 F. J. Dyer

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

Experience has shown that the majority of lead toxicity plainti�s present with a history of having had a body burden of lead of 50 mg/dl or greater at some time between ages 1 and 3 years. The majority of studies of lead toxicity in the general population are concerned with low-level lead exposure (Dyer, 1993). This is the case for those studies summarized in Grant and Davis (1989) where lead level is reported in a mg/dl metric. McBride et al. (1982) lists a range of 2±29 mg/dl; Yule et al. (1981) lists a range of 7±32 mg/dl; Landsdown et al. (1986) lists a range of 7±24 mg/dl; Harvey et al. (1984) lists a range of 6±30 mg/dl; Silva et al. (1986) lists a range of 4±50 mg/dl with a mean of 11.1 mg/dl. However, some general population studies such as those of Schroeder et al. (1985) and Schroeder and Hawk (1987), examined samples of children with lead levels near those of typical lead plainti�s and, not surprisingly, found signi®cant relationships between lead exposure and subjects' performance on psychometric measures. Clinic studies summarized in Grant and Davis (1989) where lead toxicity levels for high lead groups were approximately the same as for typical lead plainti�s also found signi®cant e�ects for lead on intelligence, perceptual-motor, and behavioral measures (de la Burde & Choate, 1972; Rummo, 1974; Rummo et al., 1979; Kotok, 1972).

Defense attorneys in lead litigation and their experts frequently cite certain ACV general population studies of low level lead exposure as though their nonsigni®cant ®ndings had relevance for the situation of the typical lead poisoned plainti� whose lead toxicity is 2 or 3 times more severe. Actually, the fact that there have been several low-lead level ACV studies such as those of Needleman et al. (1979) that have found signi®cant e�ects underscores the severity of the e�ects of lead on children at 50 mg/dl and above.

RESULTS OF LABORATORY STUDIES OF LEAD TOXICITY

There are ample empirical reasons for rejecting the ``cautious convention'' of attributing all overlapping variance in ACV studies of lead to the covariates. Clear lead e�ects appear in animal studies, where randomization is possible, and in studies of human subjects where the dependent variable is a physiological measure that is not a�ected by sociocultural factors. Further, in vitro studies provide graphic evidence of lead's destructive e�ects on brain tissue.

Where standard balanced cell ANOVA research is possible, permitting direct determination of causality as is the case with animal subjects, studies demonstrate very clear lead e�ects on maze learning (Barrett & Livesey, 1985), altered maternal behavior (Barrett & Livesey, 1983), activity level (Silbergeld & Goldberg, 1973), visual discrimination (Bushnell et al., 1977), retarded early development (Donald et al., 1987), and spatial learning and short-term memory (Rice & Karpinski, 1988). Rice (1989), in a series of lead studies employing monkeys, states ``These results collectively provide strong evidence for developmental exposure to lead causing behavioural impairment in the monkey, even at PbB levels near the current average for children in the United States and below presently accepted US criteria for undue risk of lead toxicity'' (p. 428). Rice concludes that his research demon- strates ``clear, dose-related de®cits as a result of lead exposure on tests of activity, attention and memory, distractibility and adaptability. In fact, de®cits analogous to

Avoiding Type II Error 141

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

those observed in children were observed in monkeys at PbB concentrations below the criterion value (425 mg/dl) currently established for children by both the Environmental Protection Agency and the Centers for Disease Control as being associated with risk for lead toxicity'' (pp. 437±438). Of course it is not possible to state with precision the degree to which the observed dose e�ects of lead in monkeys and other animal subjects apply to humans, owing to interspecies physiological di�erences that a�ect the results in unknown ways. However, the above animal studies constitute powerful evidence for at least a moderate degree of neurological damage in humans at dosage levels characteristic of infant plainti�s in lead poisoning litigation.

Employing both in vivo and in vitro methodology, Regan et al. (1989) demon- strated that chronic low level lead exposure impairs the early structuring of the central nervous system. They relate blood lead levels to impairments of cell acquisition, ®ber outgrowth, and synapse formation. They also state that these observed physiological changes would explain the neurobehavioral de®cits observed in general population ACV studies. Pelling et al. (1989), in reviewing previous studies of lead e�ects on brain microvasculature in the developing rat, found that this is a primary target for lead in the central nervous system. They cite prior research demonstrating that high blood levels of lead are associated with extensive capillary breakdown, hemorrhage, and edema. Using in vitro techniques, Pelling et al. demonstrated reduced glucose uptake for various central nervous system regions in the rat, associated with exposure to lead, particularly acute exposure. Otto (1989) found that evoked brain potentials provide a sensitive index of lead-induced change in the central nervous system functioning of children. Using an active conditioning paradigm, he found a linear relationship of slow wave voltage and current blood lead levels at three scalp recording sites. He con- cludes: ``Audiometric and electrophysiological assessments both suggest that auditory function in children is altered by lead absorption'' (p. 290). Relating these ®ndings to behavioral phenomena, Otto states that the shorter latencies and faster peripheral nerve conduction velocities and shorter visual evoked potential latencies in lead exposed children observed in his and other research ``. . . could be mani- festations of a common underlying problem resulting from lead absorption. Learning disorders, poor performance, developmental delays, and hyper- excitability of the peripheral and central nervous system are all compatible with attention de®cit disorder. Increased nervous system excitability and cognitive impairment viewed as common symptoms of attention de®cit disorder could thus be observed in the same children as a consequence of lead exposure'' (pp. 293±284).

CONCLUSIONS

Psychological consultation in lead poisoning litigation demands a thorough integration of research knowledge with clinical practice. Being that causal relation- ships are of paramount importance in these cases, experts must go beyond standard observation and testing approaches to address this central question through reference to large scale statistical studies of lead exposed children.

Practitioners must be on guard against Type II errors at two stages of the tort process: assessing compensable damages and assessing proximate cause. In the area

142 F. J. Dyer

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

of assessment of compensable damages, ``medical testing'' approaches, usually in the form of a standard cranial nerve examination, often yield false negatives, as lead exposed children who have signi®cant neurological damage at the synaptic level typically present as grossly normal. These children perform poorly on psycho- metric measures including tests of general mental ability, expressive and receptive language skills, perceptual-motor skills, and attentional skills. In clinical assess- ment of causation in the individual child it is essential to elicit data on family background, pregnancy complications, birth history, developmental history, and accidents or illnesses not related to lead.

It is in the area of assessing proximate cause that practitioners encounter com- plex issues of proof: ACV conservatism due to attribution of overlapping variance to the ®rst variable entered into the regression equation; use of excessively large numbers of covariates so as to reduce the statistical power for variables entered last; using covariates that are actually substitute variables for children's exposure to lead dust and ingestion of lead through hand-to-mouth behavior; failure of most research to explore interactions between covariates and lead that would invalidate the ACV; and inappropriate citation of nonsigni®cant ACV studies of children with much lower levels of lead than typical plainti�s to suggest that higher levels of lead have no deleterious e�ects. When true experimental methods are employed, such as in animal and in vitro studies, a pronounced e�ect for lead emerges. Human physiological studies employing measures that are free of sociocultural in¯uences also demonstrate unequivocal lead e�ects.

REFERENCES

Baghurst, P. A., McMichael, A. J., Tong, S., Wigg, N. R., Vimpani, G. V., & Robertson, E. F. (1995). Exposure to environmental lead and visual-motor integration at age 7 years: The Port Pirie cohort study. Epidemiology, 6, 104±109.

Barrett, J., & Livesey, P. J. (1983). Lead induced alterations in maternal behavior and o�spring development in the rat. Neurobehavioral Toxicology and Teratology, 5, 557±563.

Barrett, J., & Livesey, P. J. (1985). Low-level lead e�ects on the learning and extinction of passive avoidance in the rat. Australian Journal of Psychology, 37, 1±13.

Barton, W. A. (1985). Recovering for psychological injuries. Washington, DC: The Association of Trial Lawyers of America.

Bellinger, D. C. (1995). Neuropsychologic function in children exposed to environmental lead. Epidemiology, 6, 101±102.

Bellinger, D. C. (1995). Interpreting the literature on lead and child development: The neglected role of the ``experimental system''. Neurotoxicology and Teratology, 17, 201±212.

Berg, R., Franzen, M., & Wedding, D. (1987). Screening for brain impairment: A manual for mental health practice. New York: Springer.

Bornschein, R. L. (1985). In¯uence of social factors on lead exposure and child development. Environmental Health Perspectives, 62, 343±351.

Bushnell, P. J., Bowman, R. E., Allen, J. R., & Marlar, R. J. (1977). Scotopic vision de®cits in young monkeys exposed to lead. Science, 196, 333±335.

Cohen, J. (1969). Statistical power analysis for the behavioral sciences. New York: Academic Press. Cohen, J., & Cohen, P. (1975). Applied multiple regression/correlation analysis for the behavioral

sciences. Hillsdale, NJ: Lawrence Erlbaum Associates. de la Burde, B., & Choate, M. S. (1972). Does asymptomatic lead exposure in children have latent

sequelae? Journal of Pediatrics (St. Louis), 81, 1088±1091. Donald, J. M., Cutler, M. G., & Moore, M. R. (1987). E�ects of lead in the laboratory mouse:

Development and social behavior after lifelong exposure to 12 mM lead in drinking ¯uid. Neuro- pharmacology, 26, 391±339.

Dyer, F. J. (1993). Clinical presentation of the lead poisoned child on mental ability tests. Journal of Clinical Psychology, 49, 94±101.

Avoiding Type II Error 143

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

Dyer, F. J. (1996). Scienti®c validity of forensic assessments: The end of the expert's black box. New Jersey Psychologist, 46(3), 20±22.

Grant, L. D., & Davis, J. M. (1989). E�ects of low-level lead exposure on paediatric neurobehavioral development: Current ®ndings and future directions. In M. A. Smith, L. D. Grant, & A. I. Sors (Eds.), Lead exposure and child development: An international assessment (pp. 49±115). Dordrecht, Netherlands: Kluwer.

Harvey, P. G., Hamlin, M. W., & Kumar, R. (1984). Blood lead, behavior, and intelligence test performance in preschool children. Science and Total Environment, 40, 45±60.

Hatzakis, A., Kokkevi, A., Maravelias, C., Katsouyanni, K., Salaminios, F., Kalandidi, A., Koutselinis, A., Stefanis, C., & Trichopoulos, D. (1989). Psychometric intelligence de®cits in lead- exposed children. In M. A. Smith, L. D. Grant, & A. I. Sors (Eds.), Lead exposure and child develop- ment: An international assessment. Dordrecht, Netherlands: Kluwer.

Kotok, D. (1972). Development of children with elevated blood lead levels: A controlled study. Journal of Pediatrics (St. Louis), 80, 56±71.

Landsdown, R., Yule, W., Urbanowicz, M. A., & Hunter, J. (1986). The relationship between bloood-lead concentrations, intelligence, atttainment, and behavior in a school population: The second London study. International Archives of Occupational and Environmental Health, 57, 225±235.

McBride, W. G., Black, B. P., & English, B. J. (1982). Blood lead levels and behavior for 400 preschool children. Medical Journal of Australia, 2, 26±29.

Needleman, H. L., Gunnoe, C., Leviton, A., Reed, R., Peresie, H., Maher, C., & Barrett, P. (1979). De®cits in psychologic and classroom performance of children with elevated dentine lead levels. New England Journal of Medicine, 300, 689±695.

Needleman, H. L., & Bellinger, D. C. (1989). Type II fallacies in the study of childhood exposure to lead at low dose: A critical and quantitative review. In M. A. Smith, L. D. Grant, & A. I. Sors (Eds.), Lead exposure and child development: An international assessment (pp. 293±304). Dordrecht, Netherlands: Kluwer.

Needleman, H. L., Riess, J. A., Tobin, M. J., Biesecker, G. E., & Greenhouse, J. B. (1996). Bone lead levels and delinquent behavior. Journal of the American Medical Association, 275, 363±369.

Pocock, S. J., Ashby, D., & Smith, M. A. (1989). Lead exposure and children's intellectual performance: The Institute of Child Health/Southampton study. In M. A. Smith, L. D. Grant, & A. I. Sors (Eds.), Lead exposure and child development: An international assessment (pp. 149±165). Dordrecht, Netherlands: Kluwer Academic Publishers.

Rabinowitz, M., Bellinger, D. C., & Leviton, A. (1989). Which measures of lead burden best predict a child's 2-year mental development? In M. A. Smith, L. D. Grant, & A. I. Sors (Eds.), Lead exposure and child development: An international assessment (pp. 475±476). Dordrecht Netherlands: Kluwer.

Regan, C. M., Cookman, G. R., Keane, G. J., King, W., & Hemmens, S. E. (1989). The e�ects of chronic low-level lead exposure on the early structuring of the central nervous system. In M. A. Smith, L. D Grant, & A. I. Sors (Eds.), Lead exposure and child development: An international assessment (pp. 440±452). Dordrecht, Netherlands: Kluwer.

Rice, D. C. (1989). Behavioural e�ects of low-level develpomental exposure to lead in the monkey. In M. A. Smith, L. D. Grant, & A. I. Sors (Eds.), Lead exposure and child development: An international assessment (pp. 427±439). Dordrecht, Netherlands: Kluwer.

Rice, D. C., & Karpinski, K. F. (1988). Lifetime low-level exposure produces de®cits in delayed alternation in adult monkeys. Neurobehavioral Toxicology and Teratology 10, 207±214.

Rummo, J. H. (1974). Intellectual and behavioral e�ects of lead poisoning in children. Unpublished doctoral dissertation, University of North Carolina, Chapel Hill. (University Micro®lms No. 74-26, 930).

Rummo, J. H., Routh, D. K., Rummo, N. J., & Brown, J. F. (1979). Behavioral and neurological e�ects of symptomatic and asymptomatic lead exposure in children. Archives of Environmental Helath, 34, 120±124.

Schroeder, S. R., & Hawk, B. (1987). Psycho-social factiors, lead exposure, and IQ. In W. R. Schroeder (Ed.), Toxic substances and mental retardation. Neurobehavioral toxicology and teratology (pp. 97±136). Washington, DC: American Association on Mental De®ciency.

Schroeder, S. R., Hawk, B., Otto, D. A., Mushak, P., & Hicks, R. E. (1985). Separating the e�ects of lead and social factors on IQ. Enviorionmental Research, 38, 144±154.

Silbergeld, E. K., & Goldberg, A. M. (1973). A lead-induced behavioral disorder. Life Sciences, 13, 1275±1283.

Silva, P. A., Hughes, P., & Faed, J. M. (1986). Blood lead levels in 576 Dunedin eleven year old children. New Zealand Medical Journal, 99, 43±52.

Smith, M. A. (1989). The e�ects of low-level lead exposure on children. In M. A. Smith, L. D. Grant, & A. I. Sors (Eds.), Lead exposure and child development: An international assessment (pp. 3±47). Dordrecht, Netherlands: Kluwer.

Smith, M. Delves, T., Landsdown, R., Clayton, B., & Graham, P. (1983). The e�ects of lead exposure on urban children: The Institute of Child Health/Southampton study. Developmental Medicine and Child Neurology, 25, 47.

144 F. J. Dyer

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

Smith, M. A., Grant, L. D., & Sors, A. I. (Eds.). (1989). Lead exposure and child development: An international assessment. Dordrecht, Netherlands: Kluwer.

Stiles, K. M., & Bellinger, D. C. (1993). Neuropsychological correlates of low-level lead exposure in school age children: A prospective study. Neurotoxicology and Teratology, 15, 27±35.

Sukel, H., Bennett, E., & Cooper, J. (1995, August). Complex scienti®c testimony: How do jurors make decisions? Paper presented at the 103rd convention of the American Psychological Association, New York, NY.

United States Department of Health and Human Services/Centers for Disease Control. (1991). Preventing lead poisoning in young children. Atlanta, GA: U.S. PubIic Health Service.

Yule, W., Landsdown, R., Millar, I. B., & Urbanowicz, M. A. (1981). The relationship between blood lead concentrations, intelligence, and attainment in a school population: A pilot study. Developmental Medicine and Child Neurology, 23, 567±576.

Avoiding Type II Error 145

# 1998 John Wiley & Sons, Ltd. Behav. Sci. Law, Vol. 16, 131±145 (1998)

Copyright of Behavioral Sciences & the Law is the property of John Wiley & Sons, Inc. and its content may not

be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written

permission. However, users may print, download, or email articles for individual use.