Attention Deficit Hyperactivity Disorder

INTRODUCTION

Despite extensive research, Attention Deficit/Hyperactivity Disorder (ADHD), its diagnosis, and its treatment remains the object of controversy. The differences in the number of children diagnosed and treated across countries, along with major differences in the use of stimulant medications, support the argument that ADHD may be a value-laden social label, rather than a legitimate medical condition. In particular, methylphenidate is prescribed at a considerably higher rate in the United States (US) than in other developed nations (International Narcotics Control Board, 1994), and even within the United States patterns of use are highly variable (LeFever, Dawson, & Morrow, 1999; Rappley, Gardiner, Jetton, & Houang, 1995). In some countries, such as Italy, methylphenidate is not even available.

The diagnostic criteria have undergone significant changes historically. In 1980, in the DSM-III, the focus of the disorder shifted to inattention because of research by Dr Virginia Douglas and her colleagues (Douglas, 1974; Douglas & Peters, 1979). The diagnostic criteria in the DSM-III (APA, 1980) included three dimensions (inattention, impulsivity, and hyperactivity) and two subtypes (Attention Deficit Disorder with and without Hyperactivity). This was followed by DSM-IIIR (APA, 1987), which retained the focus on inattention, impulsivity, and hyperactivity, but eliminated all dimensions and subtypes. Most recently, in the DSM-IV (APA, 1994), subtypes were reinstituted defining two dimensions (inattention and hyperactivity-impulsivity) and three subtypes (the predominantly inattentive, the predominantly hyperactive-impulsive, and the combined type; APA, 1994).

ADHD has remained a disorder primarily defined by specific behavior symptoms, and there is no simple and definitive method by which it can be diagnosed. It does not even lend itself to objective psychometric testing for diagnosis as has been employed with Mental Retardation or Learning Disabilities. The changes from Attention Deficit Disorder (ADD) with or without Hyperactivity to ADHD and now to ADHD with three subtypes reflect the conceptual disagreements and evolving scientific evidence about this condition.

Although the dimensions and subtypes were eliminated in the criteria for DSM-IIIR (APA, 1987), the review process for DSM-IV(APA, 1994), based on data from the field trials, led to revised criteria, permitting the current categorization of two dimensions and three subtypes. Since the impulsive and hyperactive behaviors appear different and distinct on face value, as well as clinically, they have been grouped separately within the dimension. Several subsequent studies have supported the two current dimensions of hyperactivity/impulsivity and inattention (Bums, Boe, Walsh, Sommers-Flanagan, & Teegarden, 2001; DuPaul et al., 1997, 1998; Hartman et al., 2001; Hudziak et al., 1998; Molina, Smith, & Pelham, 2001).

DSM-IV also added the requirement of dysfunction in more than one setting, emphasizing the need for information from multiple sources. This requirement has furthered the prominence of teacher information in making the diagnosis, which is emphasized by the DSM-IV ADHD criteria requirement that "Some impairment from the symptoms be present in two or more settings (e.g. at school or work) and at home."

Rating scales have been valuable tools for examining school or community-based samples (Baumgaertel, Wolraich, & Dietrich, 1995; Gaub & Carlson, 1997; Wolraich, Hannah, Pinnock, Baumgaertel, & Brown, 1996). The correspondence of rating scales and direct observation of the child's behavior provides evidence of rating scale validity (Kazdin, Esveldt-Dawson, & Loar, 1983; Schachar, Sandberg, & Rutter, 1986). Popular teacher behavior rating scales for ADHD (Conners, Sitarenios, Parker, & Epstein, 1998; Swanson, 1992; Wolraich, Feurer, Hannah, Pinnock, & Baumgaertel, 1998) use DSM-IV criteria. One such rating method is the Vanderbilt ADHD Diagnostic Teacher Rating Scale (VADTRS; Wolraich, Feurer, et al., 1998). Previous research (Wolraich, Feurer, et al., 1998) reported an exploratory factor analysis concluding that the VADTRS ratings fit the DSM-IV conceptual model, which posits two distinct-but-correlated sets of symptom criteria (inattention and hyperactivity/ impulsivity) for ADHD. Two limitations of the previ ous research include (a) exploratory factor analyses without significance tests or comparative fit indices of competing prestated hypotheses, and (b) use of a single sample of children in suburban schools in the US. The current study extends the earlier one by employing confirmatory factor analysis to compare three models and utilizing additional samples, including two from outside the US. The three models to be compared include (1) A general single-factor ADHD model including all 18 symptoms of ADHD; (2) A two-factor model with 9 symptoms of inattention and 9 symptoms of hyperactivity/impulsivity (like DSM-IV); (3) A three-factor model with 9 symptoms of inattention, 6 symptoms of hyperactivity, and 3 symptoms of impulsivity.

The present study uses the VADTRS to compare four samples: the suburban US samples previously reported on (Wolraich, Hannah, Baumgaertel, Pinnock, & Feurer, 1998); an urban US sample; a Spanish sample; and, a German sample. The study examines four questions:

1) Is it necessary to distinguish two aspects of ADHD (inattentive vs. hyperactive/impulsive) as done by DSM-IV?

2) How well does a two-factor model of the 18 ADHD symptoms fit a wide range of samples?

3) Do the 9 hyperactivity/impulsivity symptoms represent 1 factor or 2 factors (hyperactivity and impulsivity)?

4) How do ADHD symptoms relate to performance in school?

METHOD

Participant

This section describes how the four samples were collected; later in the results, we will review statistical tests of mean differences among the samples.

US Suburban Samples

Participants were elementary school children (kindergarten through fifth grade) in a suburban county of middle Tennessee. The county included two school districts with an overall population of 81,021 based on the 1990 census. During the 1993-94 and 1994-95 school years two samples were collected. The earlier sample was a whole county sample including all 16 schools described previously (Wolraich et al., 1996); in it 100% of the teachers in the participating schools completed rating scales on 100% of the students in their classes. The later suburban sample, which also has been previously described in greater detail (Wolraich, Hannah, et al., 1998), contained 10 of the 16 schools, 214 teachers, and 4,323 students.

US Urban Sample

The sample consisted of 6,171 out of 19,145 (32%) children in kindergarten through the fourth grade in an urban school system in Tennessee. Out of 67 schools, 58 contributed ratings. Data were collected from 321 teachers out of a total of 1,080, which represents a 30% response rate for teachers. The sample was culturally diverse with 46.1% Caucasian, 41.4% African American, and 12.5% other.

Spanish Sample

The sample comprised 1,332 students distributed evenly across Grades 1 through 4. They were from 10 schools and were rated by 59 teachers from Barcelona, a city with 1.5 million inhabitants of which 11.9% are children. The population of Barcelona is 98% Spanish. Barcelona schools were chosen randomly within six strata in a 2 by 3 design: (a) public or private; and (b) socioeconomic status (high, middle, low). This classification was obtained from the data of the department of education. The questionnaire was distributed in the last trimester of the academic course, a time when teachers knew their students' behavior in class. The questionnaires were completed by all the teachers on all of the students in their classes, a response rate of 100%.

German Sample

The sample, previously reported by Baumgaertel et al. (1995), included 1,077 children, 10 schools, and 55 teachers. The students, in Grades 1 to 4, were from rural areas (5 0.4%) as well as urban areas (49.6%) from the same district in Bavaria. The urban sample was from five schools in a city of 120,000. The schools were chosen by the school board as typical for the area and provide a fairly even mix of children with varied socioeconomic status. At least one teacher from each Grade 1-4 from each school volunteered to complete the questionnaires on all children in their classes. Seventeen children were in a bilingual Turkish class in one of the urban schools; otherwise, all of the children were in regular German-speaking classes.

Assessment Instrument

The Vanderbilt ADHD Teacher Rating Scale (VADTRS; Wolraich, Feurer et al., 1998) enables teachers to report on ADHD symptoms and some common comorbid complications. Teachers rate 35 symptoms and 8 performance items. The 35 symptoms consist of four groups, two measuring ADHD and two measuring common comorbid complications: (a) all 9 DSM-IV behaviors for inattention; b) all 9 DSM-I V symptoms for hyperactivity/impulsivity; c) an abbreviated 10-item scale for oppositional defiant and conduct disorders; and d) an abbreviated 7-item scale for anxiety and depression symptoms from the Pediatric Behavior Scale (Lindgren & Koeppl, 1987). Teachers rate each symptom on frequency (0 = never, 1 = occasionally, 2 = often, and 3 = very often). When symptom counts are needed, symptoms that occur often or very often are considered present, and those that occur never or occasionally are considered absent. In the present study, raw scores (0123) were used to maximize precision.

The school performance section evaluates functioning in the classroom with 8 items having 5-point Likert scales. Three items evaluate academic performance (reading, mathematics, and written expression), and five items evaluate classroom performance (peer relations, following directions, disrupting class, assignment completion, and organizational skills). In more recent samples (Suburban 2, Urban and Spain), all eight individual problems were rated, but in older samples (Suburban 1 and Germany) teachers used 2 items, 1 rating academic problems and the other rating behavior problems. Because of this difference among samples, performance problems (ADHD symptoms and school performance) were dichotomized (0 = absent, 1 = present) for analysis. Low performance ratings (1 or 2 on a 5-point scale) were considered evidence of impaired performance.

Assessment of reliability and validity was reported previously (Wolraich, Feurer, et al., 1998). In an exploratory and confirmatory analysis, a four-component solution was indicated (inattention, hyperactivity/impulsivity, opposition/aggression, and anxiety/depression) for the behavioral scales, internal consistency for items was .7 or greater, and correlations between symptoms and problems ranged from .25 to .66.

Language Translations

In the case of the Spanish translation, a front translation was accomplished by two qualified translators independently from the original to the target language. Once completed inconsistencies between translations were resolved. There were similar back translations and a comparison front to back was done to correct inconsistencies until the translators judged the items to be identical in content and meaning. The German translation was completed by a developmental-behavioral pediatrician (AB), a native German speaker who was fluent in both German and English, and reviewed by German special educators also fluent in both languages. An earlier version of the VADTRS was used with the German sample (Baumgaertel et al., 1995). This version contained more comorbid items, and only two performance items (academic and behavior).

Rating Procedure

Research personnel at the four sites held meetings with teachers to explain the study and to review the instrument and its use. Teachers were told to report averages for the whole school year, and to consider what is appropriate for the child's age. Each elementary school teacher received a packet of rating scales and a mailer for returning completed scales to the research staff. Teachers were asked to return rating forms without identifying information in order to preserve the anonymity of every child. Teachers completed behavioral rating scales on students in their classes. Follow-up contacts prompted teachers who did not respond. The Institutional Review Board or its equivalent and the Board of Education in each school district approved these procedures along with Vanderbilt University Institutional Review Board.

Statistical Methods

Confirmatory Factor Analysis (CFA)

CFA tests the fit of a priori models describing the covariance among items. The three models, which were described in the introduction and appear in Fig. 1, were tested. The double-headed arrows in Fig. i mean that factors were free to correlate with each other as determined by the data.

CFA was conducted with EQS 5.7b (Bentler & Wu, 1993; Byrne, 1994). In CFA, the prestated model is expressed as a set of equations in order to determine how well the model fits the data. Likelihood ratio tests determine whether one nested model is a significant improvement over another, but with thousands of participants, very small differences may be statistically significant. Bentler's comparative fit index (CFI; Bentler, 1988) was used to compare the fit of models. According to Bender (1992, p. 93), the CFI reflects model fit at all sample sizes. To ensure that departures from normality did not distort results, Satorra-Bentler scaled chi-square tests, robust standard errors (Hu, Bentler, & Kano, 1992), and robust CFIs were used when possible.

CFA offers tests for factor structure invariance across samples through multigroup structural equation modeling, a procedure introduced by Joreskog (1971) and extended and streamlined by Bentler (1992). Following the example of Byrne (1994, p. 177), the equivalence of the ADHD measurement model in all four samples was tested to determine whether item loadings and between-factor correlations differed significantly among the four samples.

Between-sample equivalence was tested by comparing the goodness of fit of a pooled four-sample factor model in which the item-factor loadings and the between-factor correlations were constrained to equality with the fit of a model in which loadings and correlations were free to vary among samples. If the samples differ, the constrained model would show a worse fit than the free-to-vary model. Because there were thousands of cases, likelihood ratio tests had the power to detect small and perhaps trivial differences; therefore, we examined the fit index (CFI) of constrained and unconstrained models as well as statistical significance.

RESULTS

Sample Characteristics

Table I shows sample characteristics related to ADHD as reported by each child's teacher. Gender was 50 or 51% male in all four samples (p = .91), but all other characteristics in Table I showed differences among the samples. Some differences were large; for example, the percentage of children having academic problems ranged from 13 to 28%. The presence of so many differences among samples suggests that the samples are distinct on many characteristics, including age, average severity of ADHD symptoms, and comorbid psychopathology. In the rest of this paper, we examine how well a single model of ADHD symptoms based on DSM-IV applies to samples with so many geographic and symptomatic differences.

CFA Model Fit

The CFA proceeded in two stages. First, the fit of the three models (1 factor, 2 factors, 3 factors) was tested on the second suburban sample. After that, the between-sample fit of the models was compared using the four main samples, in which cross-national differences could be observed. The single factor solution shows poor fit (CFI = 0.76). The two and three-factor models show adequate fit (CFI > .90). There are reductions in [chi square] "misfit" when the second and third factors are added. These reductions are statistically significant in this very large sample. The improvement in fit index resulting from adding the third factor (impulsivity) is only about 1% (CFI from .92 to .93 when parameters may vary between samples, or .91 to .92 when the sample parameters are constrained to equality). This small improvement suggests that while the third factor is statistically significant, it may not be important in this analysis of 18 symptoms from the DSM-IV. Impulsivity might be a stronger independent factor if t he list of symptoms were expanded to include more than three items.

Table II compares the fit of the three nested models with likelihood ratio tests and fit indices. Model 1, the single factor model in which all ADHD symptoms resulted from a single latent dimension, showed unacceptable fit (CFI = .73), which was far below an acceptable fit (CFI > .90).

In row 2 of Table II, ADHD symptoms were divided into two subtypes (inattentive and hyperactive/impulsive) according to the DSM-IV criteria. Model 2 showed a significantly better fit, one that was within the acceptable range (CFI > .90). Model 3, which distinguishes between hyperactivity and impulsivity, showed a significant improvement in fit over Model 2, [DELTA][chi square](2, N = 4,273) = 2,247, p < .000 1, and better CFI than Model 2 ([CFI.sub.2] = .9 18, [CFI.sub.3] = .945). In each case the robust fit indices are slightly lower than the normal fit indices, indicating that departures from normality are a slight problem, rather than a serious distortion.

Results so far suggest that teacher reports distinguish three separate aspects of ADHD in the suburban US sample. In the next step, we use four fresh samples to test the generalizability of this result.

The first step in the cross-national analysis was exploratory, determining how many factors should be extracted by examining eigenvalue plots for all five samples, as shown in Fig. 2. This analysis used principal components without any rotations or other elaborations. Overall, the five plots show great similarity. The first two factors explained from 68 to 77% of the variance, and the third factor's eigenvalue was always less than one. Having an eigenvalue <1.0 suggests that a third component explains less variance than one symptom, which is why 1.0 is a traditional lower limit below which additional factors have no value. Later likelihood ratio tests and fit indices will evaluate the statistical and practical significance of the third factor in a confirmatory factor analysis.

We now compare the three hypothesized models to determine whether there is a single "best model" for all four samples.

Table III compares the three models (1, 2, or 3 factors) using two methods (constrained--unconstrained) for a total of six CFAs. The two methods, "constrained and unconstrained" refer to the equality of the model's coefficients across the four samples. If we constrain all the samples to share a single solution, differences among the samples would lower the fit indices compared with unconstrained models in which each of the four samples has its own parameters.

The a-b pairs of rows in Table III (parameters "Vary" or are "Equal") ask whether the same factor structure of the ADHD symptoms applies in all four samples. Likelihood ratio tests in Table III column 5 compare two solutions: (a) Constrained, in which the four samples are forced to have identical factor loadings and cross-factor correlations; and (b) unconstrained, in which the four samples may each have their own unique parameters. In every case the LRT was statistically significant, suggesting that there are reliable small differences among the samples. However, comparing fit indices suggests that ignoring cross-sample differences reduces the model fit index by only about 1% (Table III, columns 8 & 9). Evidently, the differences among samples, while not due to chance, are small.

Another way to see whether the same model fits these four disparate samples is to inspect the loadings of the site-specific models, as shown in 3-factor models in Table IV.

Table IV shows the factor loadings in the four samples when parameters were free to vary among samples. If factor structure were different in the samples, the loadings would differ. Since it would be difficult to evaluate so many parameters by inspection of 18 x 4 = 72, an overall index of difference was calculated in the last column (Table IV, column 7). This index is the average deviation in loadings across the four samples in each row (average absolute deviation = [SIGMA]/D - M\/4, where D is the deviation and M is the mean). The grand mean deviation is very small, 0.030 for the whole table, suggesting that the variation across samples is slight. Evidently, models for the four samples are not identical--there are statistically significant differences--but the differences are small.

Internal Consistency of Inattention, Hyperactivity, and Impulsivity Scales

The next step was to evaluate the internal consistency reliability of the Inattention, Hyperactivity, and Impulsivity scales. Could Factor III, impulsivity, with only 3 items, be long enough to be reliable? Results appear in Table V.

The lowest Cronbach's alpha in Table III is 0.87, which is adequate, and 9 of 12 alphas are over 0.90. Even the smaller factors of hyperactivity (6 items) and impulsivity (3 items) appear consistent. The fact that the third factor can be measured, however, does not prove that it is important. Impulsivity has an extremely high correlation with hyperactivity (r = .975), and ignoring it reduces model fit by only 1%. These results suggest that the independent effect of impulsivity is too slight to make a practical difference when working within the 18-symptom list from DSM-IV.

Tables III, IV, and V favor the two-factor model for ADHD. The third factor is less parsimonious and improves the fit indices only slightly. In addition, while there are statistically significant differences between large samples in the ADHD measurement model, these differences are slight, and ignoring them reduces the model fit by only about 1%. These results suggest that the same two-factor model fits all four samples fairly well.

ADHD Symptom Scores

The last section of results concerns the relationship between the ADHD behaviors and performance in school. Teacher-reported information was available on the presence or absence of academic and behavior problems in all four samples. To see how ADHD factor scores relate to performance problems, we calculated the mean rating for each item to produce 5 subscale scores: ADHD total (18 items), inattention (9 items), hyperactivity/impulsivity (9 items), hyperactivity only (6 items), and impulsivity only (3 items). Subscale means and standard errors appear as profiles in Fig. 3. Standard errors were plotted to appear as error bars, which are small, often invisible, due to the large number of cases.

The four profiles show four groups: (a) children with both behavior and academic problems (9%); (b) children with behavior problems only (6%); (c) children with academic problems only (14%); and (d) children with neither problem (72%). In this large sample, only two means are not significantly different (noted as NS on the figure).

As expected, children with no performance problems (white triangles) are low on all five measures of ADHD. Children with academic problems only (black triangles) are characterized mainly by inattentiveness, with slightbut-significant elevations on the three hyperactivity and/or impulsivity scores. Children with behavior problems and no academic problems (white circles) have approximately equally elevated scores on all aspects of ADHD, including inattention, hyperactivity, and impulsivity. Children with both academic and behavior problems (black circles) have an additional elevation on inattention. Overall, Fig. 3 suggests that inattention is a key ingredient of poor academic performance, and that all aspects of ADHD (inattention, hyperactivity, and impulsivity) are higher in school children with behavior problems.

Other Moderators of Model Fit

Having seen that the two-factor model works in four different geographical samples, we next examine whether model fit is moderated by gender, age, school grade, clustering, and level of ADHD. The last two items need a brief introduction.

Factor analytic models assume that observations are independent, for example, different children taking a test. In a school screen for ADHD, children are nested in classrooms rated by a single teacher. A brief follow-up analysis will check to see if this dependence affects the model.

Another technical moderator may be level of ADHD. Consider the ADHD symptom sum, which can range from o to 54 (18 symptoms each rated 0-3). In the largest and most complete sample, the Suburban 1 population sample, the modal ADHD item sum is 0, the median is 1, and the range is 0-51. It is conceivable that the same model wouldn't apply both to low scoring children without ADHD and high scoring children with ADHD.

For simplicity, the same statistical approach is used for each moderator. A single sample will be used in which all variables are available (some lack age, grade, or teacher ID). This follow-up will show whether moderators have any effect, but it is not an in-depth study, for example of gender and ADHD.

In the most transparent analysis, separate two-factor models were estimated for boys and girls, then constrained to a single set of parameters. If misfit increases significantly, and the CFI fit index goes down, we would conclude that the same model doesn't fit both boys and girls. Row 1 in Table VI shows the result. Separate models had a CFI of 0.923, and forced-equal models had a fit of 0.919. According to an LRT, this difference in fit was nonsignificant (p = .185). This result suggests that the model fits both boys and girls equally. Of course, fitting the same model in no way contradicts the common finding that boys have higher mean problem scores and greater prevalence of ADHD.

The second moderator, age, was handled in the same way. First age was split into two groups (5-7, N = 2965, and 8-11, N = 3928). Cases with missing ages or ages <5 or >11, or missing symptom items, were dropped. There was a statistically significant loss of fit when younger and older children were forced into the same model (p < .001). However, the change in CFI was very small (.934 down to .932). The practical conclusion is that the two factor models apply almost equally to young children and older ones.

The result for school grade was very similar. Grades were split into lower (K-1-2 and higher 3-4-5) with a median split. Outliers were dropped. The difference in fit was statistically significant but small (CFI declining from .935 to .931).

Clustering and model fit was studied by creating a subsample of N = 359 in which each teacher had only one case randomly selected from their class. Empirically the intraclass correlation (ICC) due to teacher was significantly greater than zero (z = 10.31, p < .0001), with an ICC = .19 for the ADHD symptom sum in the Suburban 1 sample. This ICC indicates a moderate lack of independence among children in a given classroom. When the two 18 x 18 covariance matrices were compared, there were significant differences, [chi square](171, N = 8,252) = 345, p < .0001. There was a significant loss of fit when the same parameters were forced on both samples (Row 4, p < .001). However, the CFI went down by only 0.002, a small difference. This small change suggests that clustering, while present, has little effect on the model.

To test the moderating effect of ADHD level, the sample was split at an ADHD sum score below which the diagnosis of ADHD could not be made (scores of 2 on 6 items, total = 12). Splitting at 12 created two sub-samples (Low scores N = 6,928, high N = 1,328). These two samples had significantly different 18 x 18 covariance matrices, [chi square](171, N = 6,818) = 23577, p < .0001. Overall fit went down to 0.868, and the increase in misfit (0.02) was the largest in Table VI. However, it is not clear that this creates a practical problem. Artificially splitting any sample into low and high scores would necessarily change the variances of items (changing the covariance matrix). Judging by full-sample results, teacher screening with the 18 ADHD symptoms from DSM-IV and two factors seems to work with a full range population sample.

DISCUSSION

In a confirmatory factor analysis of ADHD symptoms, teachers rated 19,542 grade school children in four samples: Suburban US, Urban US, Spanish, and German. This study generally supports the DSM-IV model of ADHD, at least in terms of symptom dimensions. Several other studies have yielded similar results that support a two-dimension model of ADHD. Bums and colleagues (Burns et al., 2001) examined the factor analysis of ADHD and ODD symptoms and found the best model fit was one that used a two-dimension model of ADHD. Comparable research conducted by Molina et al. (2001) had similar findings. Studies done with adolescents by DuPaul and associates (DuPaul et al., 1997, 1998) also found the two-dimension model of ADHD to be the best fit. The same two-dimension model was found to fit best for female adolescents (Hudziak et al., 1998). Additionally, evidence for the two-dimensional model of ADHD was provided by research on the internal construct validity of DSM-IV based models of ADHD, CD, ODD, generalized anxiety, and depression (Hartman et al., 2001).

In any large study with thousands of cases, statistical power makes it possible to detect small differences that are statistically significant but too small to be clinically meaningful. In the present study, there were statistically significant between-sample differences in ADHD factor structure in this large study, but these differences were small and, for most purposes, unimportant. Two factors, inattention and hyperactivity/impulsivity, were found in all samples. A hypothesized third factor, impulsivity, could be detected, but it had negligible importance in analysis based on the 18 items from the DSM-IV. Impulsivity contains only 3 items, and when recognized as a separate factor, impulsivity made only slight improvements in fit index (about 1%) and produced a factor so highly correlated with hyperactivity (r = .975) that there is no reason to distinguish it in practice. However, the output of factor analysis is dependent on the items put into the analysis. It remains possible that a measure of impulsivity could be constructed if more items were used to measure it.

Although there were statistically significant differences in factor loadings among the four samples, these differences were small, and the same two-factor ADHD measurement model fit all four samples adequately. Finding approximately the same factors and loadings in suburban and urban samples in the US, and in Germany and Spain, suggests that the model reflects the characteristics of children, as opposed to the value-laden culturally distinct perceptions of teachers. Small differences were found between males and females and between younger (5 to 7) and older (8 to 11) age groups so that the two factor solution appears to be consistent for gender and age at least for elementary school age children.

There also appears to be a relationship between scores on the behavior items and problems in performance. Academic problems were connected most closely to inattention. Hyperactivity/impulsivity was related to problems in both academic and behavioral domains; the combined effect further increases the number of academic problems. This finding further confirms the relationship established in two of the samples (Baumgaertel et al., 1995; Wolraich et al., 1996).

Milich, Balentine, and Lynam (2001) have argued for two distinct disorders of inattention only and combined hyperactivity/impulsivity and inattention based on the different types of impairment and outcomes that occur between these two groups. The results from this study at least support the two distinct dimensions that have differing relationships to function. However, it is not sufficient evidence that the two subtypes are distinct disorders.

The findings do not reflect true diagnostic rates because the information was only obtained from teachers so that there is no information about how the children behave or perform in other settings. In addition, there is no information about the initial onset or duration of the behaviors. The factor structure reflects how teachers perceive the behaviors. It has yet to be determined that parental perception, the more common clinical source of information on children's behavior is similar.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

ACKNOWLEDGMENTS

Portions of this study were supported by a grant from the National Institute of Mental Health (HS/MH 0905), the United Way of Williamson County, the Catalonian Society of Pediatrics, and the John F. Kennedy Center for Research on Human Development.

Received April 23, 2002; revision received August 19, 2002; accepted September 28, 2002

REFERENCES

American Psychiatric Association. (1980). Diagnostic and Statistical Manual for Mental Disorders (3rd ed.). Washington. DC: Author.

American Psychiatric Association. (1987). Diagnostic and Statistical Manual of Mental Disorders (3rd rev. ed.). Washington, DC: Author.

American Psychiatric Association. (1994). Diagnostic and Statistical Manual of Mental Disorders (4th ed.). Washington, DC: Author.

Baumgaertel, A., Wolraich, M. L., & Dietrich, M. (1995). Comparison of diagnostic criteria for attention deficit disorders in a German elementary school sample. Journal of the American Academy of Child and Adolescent Psychiatry, 34, 629-638.

Bentler, P.M. (1988). Causal modeling via structural equation systems. In J. R. Nesselroade & R. B. Cattell (Eds.), Handbook of multivariate experimental psychology (pp. 317-335). New York: Plenum Press.

Bentler, P. M. (1992). EQS Structural Equations Program Manual. Los Angeles, CA: BMDP Statistical Software.

Bentler, P. M., & Wu, E. J. C. (1993). EQS/Windows User's Guide. Los Angeles, CA: BMDP Statistical Software.

Burns, G. L., Boe, B., Walsh, J. A., Sommers-Flanagan, R., & Teegarden, L.A. (2001). A confirmatory factor analysis on the DSM-IV ADHD and ODD symptoms: What is the best model for the organization of these symptoms. Journal of Abnormal Child Psychology, 29, 339-349.

Byrne, B. M. (1994). Structural Equation Modeling with EQS and EQS/Windows. Thousand Oaks, CA: Sage.

Conners, C., Sitarenios, G., Parker, J. D., & Epstein, J. N. (1998). Revision and restandardization of the Conners Teacher Rating Scale (CTRS-R): Factor structure, reliability, and criterion validity. Journal of Abnormal Child Psychology, 26, 279-291.

Douglas, V. I. (1974). Differences between normal and hyperkinetic children. In C. Conners (Ed.), Clinical use of stimulant drugs in children (pp. 12-23). Amsterdam: Excerpta Medica.

Douglas, V. I., & Peters, K. G. (1979). Toward a clearer definition of the attention deficit of hyperactive children. In G. Hale & M. Lewis (Eds.), Attention and the development of cognitive skills. New York: Plenum Press.

DuPaul, G. J., Anastopoulos, A. D., McGoey, K. E., Power, T. J., Reid, R., & Ikeda, M. J. (1997). Teacher ratings of attention deficit hyperactivity disorder symptoms: Factor structure and normative data. Psychological Assessment, 9, 436-444.

DuPaul, G. J. Anastopoulos, A. D., Power, T. J., Reid, R., Ikeda, M. J., & McGoey, K. E. (1998). Parent ratings of attention-deficit/hyperactivity disorder symptoms: Factor structure and normative data. Journal of Psychopathology and Behavioral Assessment, 20, 83-102.

Gaub, M., & Carlson, C. L. (1997). Behavioral characterisitics of DSM IV ADHD subtypes in a school-based population. Journal of Abnormal Child Psychology, 2S, 103-1ll.

Hartman, C. A., Hox, J., Mellenbergh, G. J., Boyle, M. H., Offord, D. R., Racine, Y., et al. (2001). DSM-IV Internal Construct Validity: When a taxonomy meets data. Journal of Child Psychology and Psychiatry, 42, 8 17-836.

Hu, L. T., Bentler, P. M., & Kano, Y. (1992). Can test statistics in covariance structure analysis be trusted? Psychological Bulletin, 112, 351-362.

Hudziak, J. J., Heath, A. C., Madden, P. F., Reich, W., Bucholz, K. K., Slutske, W., et al. (1998). Latent class and factor analysis of DSM-IV ADHD: A twin study of female adolescents. Journal of the American Academy of Child and Adolescent Psychiatry, 37, 848-885.

International Narcotics Control Board. (1994). Report of the International Narcotics Control Board for 1994. Vienna: United Nations.

Joreskog, K. G. (1971). Simultaneous factor analysis in several populations. psychometrika, 36, 409-426.

Kazdin, A., Esveldt-Dawson, K., & Loar, L. L. (1983). Correspondence of teacher ratings and direct observations of classroom behavior of psychiatric inpatient children. Journal of Abnormal Child Psychology, 11, 549-564.

LeFever, G., Dawson, K. V., & Morrow, A. L. (1999). The extent of drug therapy for attention deficit-hyperactivity disorder among children in public schools. American Journal of Public Health, 89, 1359- 1364.

Lindgren, S., & Koeppl, G. G. (1987). Assessing child behavior problems in a medical setting: Development of the Pediatric Behavior Scale. In R. J. Prinz (Ed.), Advances in behavioral assessment of children and families (Vol. 3, pp. 57-90). Greenwich, CT: JAI Press.

Milich, R., Balentine, A. C., & Lynam, D. R. (2001). ADHD combined type and ADHD predominantly inattentive type are distinct and unrelated disorders. Clinical Psychology-Science and Practice, 8, 463-488.

Molina, S. G., Smith, B. H., & Pelham, W. E. (2001). Factor structure and criterion validity of secondary school teacher ratings of ADHD and ODD. Journal of Abnormal Child Psychology, 29, 71-82.

Rappley, M. D., Gardiner, J. C., Jetton, J. R., & Houang, R. T. (1995). The use of methylphenidate in Michigan. Archives of Pediatric and Adolescent Medicine, 149. 675-679.

Schachar, R., Sandberg, S., & Rutter, M. (1986). Agreement between teachers' ratings and observations of hyperactivity, inattentiveness and defiance. Journal of Abnormal Child Psychology, 14, 331-345.

Swanson, J. (1992). School-based assessments and interventions for ADD students. Irvine, CA: K. C. Publishing.

Wolraich, M. L., Feurer, I., Hannah, J. N., Pinnock, T. Y., & Baumgaertel, A. (1998). Obtaining systematic teacher report of disruptive behavior disorders utilizing DSM-IV. Journal of Abnormal Child Psychology, 26, 141-152.

Wolraich, M. L., Hannah, J. N., Pinnock, T. Y., Baumgaertel, A., & Brown, J. (1996). Comparison of diagnostic criteria for attention deficit hyperactivity disorder in a county-wide sample. Journal of American Academy of Child and Adolescent Psychiatry, 3S, 319-323.

Wolraich, M., Hannah J. N., Baumgaertel, A., Pinnock T. Y., & Feurer, I. (1998). Examination of DSM-IV critieria for ADHD in a countywide sample. Journal of Developmental and Behavioral Pediatrics, 19, 162-168.

Mark L. Wolraich, (1, 5) E. Warren Lambert, (2) Anna Baumgaertel, (3) Santiago Garcia-Tornel, (4) Irene D. Feurer, (2) Leonard Bickman, (2) Melissa A. Doffing (1)

(1.) University of Oklahoma Health Sciences Center, Oklahoma City, Oklahoma.

(2.) Vanderbilt University, Nashville, Tennessee.

(3.) University of Pennsylvania, Philadelphia, Pennsylvania.

(4.) Hospital Saint Joan de Deu, Barcelona, Spain.

(5.) Address all correspondence to Mark L. Wolraich, University of Oklahoma Health Sciences Center, 1100 N.E 13th Street, Oklahoma City, Oklahoma 73117; e-mail: mark-wolraich@ouhsc.edu.

COPYRIGHT 2003 Plenum Publishing Corporation
COPYRIGHT 2003 Gale Group