• Clinical Group Differences
  • ADHD and General Population Comparisons
  • ADHD, Depression/Anxiety, and General Population Comparisons
• Classification Accuracy

Clinical Group Differences

To provide evidence of the criterion-related validity of the scores from the CAARS 2, differences among test scores from various relevant clinical groups and the general population were explored. Mean score differences were compared within the following groups:

1. Individuals diagnosed with ADHD Inattentive Presentation (ADHD Inattentive group), individuals diagnosed with ADHD Combined Presentation (ADHD Combined group), and individuals from the general population. Note that consistent with very low population estimates for ADHD Predominantly Hyperactive/Impulsive Presentation (APA, 2013), it was difficult to obtain a sufficient number of participants with this presentation; the three individuals who reported this presentation were excluded from the study as the sample size was too small to provide stable estimates.
   
   
2. Individuals diagnosed with Depression and/or Anxiety (Depression/Anxiety group; specific diagnoses included Major Depressive Disorder, Persistent Depressive Disorder, Generalized Anxiety Disorder, Panic Disorder, Separation Anxiety, and Social Anxiety Disorder, all with no co-occurring diagnosis of ADHD), individuals diagnosed with ADHD (with no co-occurring diagnosis of Depression or Anxiety), and individuals from the general population.
   

All scale-level analyses were conducted via a series of analysis of variance tests (ANOVA; conducted in R via the stats package, version 3.6.1; R Core Team, 2013). Given the large number of comparisons conducted, a conservative significance level (p < .01) was adopted to determine statistical significance. Effect sizes, as measured by eta-squared (η²; Cohen, 1973) and by Cohen’s d, are provided for all analyses.

Item-level analyses were based on the determination of an Associated Clinical Concern Item as “Endorsed” or an Impairment & Functional Outcome Item as “Elevated” (for more information about item elevations, see chapter 4, Interpretation). For the Associated Clinical Concern Items regarding suicidality¹ and self-injury, “Endorsed” refers to a response greater than 0, “Not at all true; Never/Rarely“; for Depression and Anxiety screener items, “Endorsed” refers to a response greater than 1, “Just a little true; Occasionally” [with the exception of individuals aged 70 years or older for the Depression screener item, in which “Endorsed” refers to a response greater than 0, “Not at all true; Never/Rarely“]).

ADHD and General Population Comparisons

To provide additional evidence of the criterion-related validity of the scores from the CAARS 2, mean score differences between individuals from the general population and those with ADHD were compared across the CAARS 2 (including Content Scales, Symptoms Scales, Associated Clinical Concerns items, and Impairment & Functional Outcome items). ADHD samples included in this section included ADHD Inattentive (Self-Report N = 96; Observer N = 63) and ADHD Combined (Self-Report N = 101; Observer N = 76). To reflect the heterogeneous nature of the ADHD population, participants were not excluded on the basis of comorbid diagnoses such as anxiety or depression. To facilitate comparisons between groups, a subsample of the general population sample was selected to match the demographics of the ADHD sample (i.e., unifying ADHD Inattentive and ADHD Combined), in terms of age, gender, race/ethnicity, and education level (EL) (Self-Report N = 197; Observer N = 139). The demographic characteristics of the General Population and ADHD groups and their raters are presented in appendix J.

Content and DSM Symptom Scales. Individuals in the ADHD Inattentive and ADHD Combined groups were expected, on average, to have higher scores on all CAARS 2 scales than individuals in the General Population group as the CAARS 2 is designed to capture symptoms related to adult ADHD. Individuals in the ADHD Combined group were expected to score higher on Hyperactivity, Impulsivity, and Emotional Dysregulation than individuals in the ADHD Inattentive group, as these scales capture features of ADHD that are not as prominent for the Predominantly Inattentive presentation.

Comparisons between the General Population and ADHD groups were analyzed via a series of analysis of variance tests (ANOVA; conducted in R via the stats package, version 3.6.1; R Core Team, 2013), and significant omnibus F-tests were followed up with Tukey’s honestly significant difference (HSD) post-hoc tests for pairwise comparisons. Results are displayed in Tables 9.9 to 9.10 and depicted graphically in Figure 9.1 to 9.2. As expected, significant differences were observed for all scale-level comparisons between the ADHD groups (including both presentations) and the General Population groups, for both the CAARS 2 Self-Report and Observer. The size of these differences varied and are reported in terms of eta-squared (η²; Cohen, 1973) and Cohen’s d effect size values. Values ranged from η² = .25 to .66 for the CAARS 2 Self-Report, representing large effect sizes. For the CAARS 2 Observer, values ranged from η² = .13 to .48, indicating a large effect size for most scales; a moderate effect size was noted for Impulsivity and Emotional Dysregulation. In looking at specific pairwise comparisons, results of the ADHD and General Population comparisons were first explored, followed by an exploration of the differences between the ADHD groups.

When comparing the individual ADHD groups to the General Population group, for CAARS 2 Self-Report and CAARS 2 Observer, both ADHD groups scored significantly higher than the General Population on all scales, with moderate to very large effect sizes (median Cohen’s d = 1.35 for Self-Report and 1.07 for Observer). The effects were even larger when comparing the ADHD Combined and General Population groups (median Cohen’s d = 2.64 for Self-Report and 1.34 for Observer). These results provide strong evidence to support the validity of the CAARS 2, as they demonstrate that the CAARS 2 scale scores successfully distinguish the profiles of individuals with and without ADHD.

It is worth noting that the effect sizes for Self-Report exceed those for Observer (that is, there are larger differences between individuals with and without ADHD when examining self-reported ratings, compared to observer-reported ratings). This trend is consistent with previously reported patterns of results for different rater types (e.g., Van Voorhees et al., 2011).

Significant differences between presentations of ADHD (i.e., Inattentive vs. Combined) were also observed. For Self-Report, the ADHD Combined group scored significantly higher on Hyperactivity, Impulsivity, Emotional Dysregulation, ADHD Hyperactive/Impulsive, and Total ADHD Symptoms scales than the ADHD Inattentive group (see Figure 9.1). For Observer, the ADHD Combined group scored significantly higher on the ADHD Hyperactive/Impulsive scale (see Figure 9.2). These findings were as expected, given that a diagnosis of ADHD Combined by definition has more hyperactive and impulsive symptoms than a diagnosis of Predominantly ADHD Inattentive presentation. These clear, statistically significant, and expected group differences provide strong evidence for the validity of the CAARS 2 scores included in these analyses.

Click to expand

Figure 9.1. Profiles for General Population and ADHD Groups: CAARS 2 Self-Report

Note. INA/EDF = Inattention & Executive Dysfunction; HYP = Hyperactivity; IMP = Impulsivity; EM DYS = Emotional Dysregulation; NEG SC = Negative Self-Concept; ADHD-I = ADHD Inattentive Symptoms; ADHD-HI = ADHD Hyperactive/Impulsive Symptoms; ADHD-TOT = Total ADHD Symptoms.

Click to expand

Figure 9.2. Profiles for General Population and ADHD Groups: CAARS 2 Observer

Note. INA/EDF = Inattention & Executive Dysfunction; HYP = Hyperactivity; IMP = Impulsivity; EM DYS = Emotional Dysregulation; NEG SC = Negative Self-Concept; ADHD-I = ADHD Inattentive Symptoms; ADHD-HI = ADHD Hyperactive/Impulsive Symptoms; ADHD-TOT = Total ADHD Symptoms.

Associated Clinical Concern Items. The Associated Clinical Concern Items were expected to reveal differences between the General Population and ADHD groups; in particular, the suicidality and self-injury items were expected to be endorsed more frequently by individuals with ADHD than individuals in the General Population group, given the documented higher prevalence rates of both suicide and self-injurious behaviors among individuals with ADHD (James et al., 2004; Meszaros et al., 2017; Septier et al., 2019; see also chapter 2, Background). Additionally, due to the high level of co-occurring internalizing disorders in individuals with ADHD (mirrored in our ADHD samples; refer to appendix J), it was expected that individuals with ADHD would, on average, rate the depression and anxiety screener items more highly than the General Population group.

For these item-level analyses, the endorsement rates and differences in endorsement (via Cliff’s d effect sizes) are presented in Table 9.11 and Figure 9.3. Trends were consistent with hypotheses that suicidality is more commonly endorsed by those with an ADHD diagnosis than by those in the general population. For Self-Report, individuals with ADHD were up to 2.8 times more likely to endorse the suicidality and self-harm items than those in the General Population. A relatively high number of individuals in the General Population group (29.0%) indicated that they had thought about killing themselves. Alarmingly, this number was considerably higher for the ADHD groups, with 58.3% of the ADHD Inattentive and 81.2% of the ADHD Combined group endorsing the suicidality item. With regard to past self-injurious behavior, individuals in the ADHD Inattentive (28.1%) group were slightly more likely to endorse this item than individuals in the General Population (18.0%), and those in the ADHD Combined group (41.6%) were 2.3 times more likely to endorse this item than individuals in the General Population. Overall, the trend of ADHD groups endorsing these items more frequently than individuals without ADHD is consistent with previous research that reports individuals with ADHD are at higher risk for self-injurious behavior, particularly those with ADHD Combined Presentation (see Hinshaw et al., 2012, see also chapter 2, Background).

A similar pattern of results was evident in the Observer data; the ADHD groups were up to 3.2 times more likely than the General Population group to endorse the suicidality item and were 2.4 times more likely to endorse the self-injury item. In contrast to the Self-Report, however, Observer data did not show the same degree of difference between ADHD Inattentive and ADHD Combined endorsement rates. Overall, Observer ratings were more moderate compared to Self-Report (as can be seen by the relatively smaller effect sizes; Self-Report median Cliff’s d = .28; Observer median Cliff’s d = .16). The lower rate of Observer endorsement in general could serve to highlight the often covert nature of these items; even close observers may not be aware of self-injury or suicidal thoughts. This difference serves as another reminder that gathering multi-informant perspectives, including self-report, is critical in an adult ADHD evaluation.

Additionally, results revealed that both the ADHD Inattentive and ADHD Combined groups were more likely to endorse the depression and anxiety screener items than the General Population group, with small to large effect sizes (Self-Report median Cliff’s d = 0.45; Observer median Cliff’s d = 0.59). Specifically, on the Self-Report, individuals with ADHD were up to 2.8 times more likely to endorse these items than those in the General Population group, and Observers rating individuals with ADHD were up to 6.4 times more likely to endorse these items than ratings of individuals from the General Population group. Given that many people with ADHD have a comorbid internalizing disorder (such as anxiety and/or depression), it is not surprising that 40.5% for Self-Report and 49.8% for Observer in the ADHD samples presented with co-occurring diagnoses (such as Major Depressive Disorder and/or Generalized Anxiety Disorder); thus, the higher rate of endorsement for these items is in line with expectations.

Click to expand

Figure 9.3. CAARS 2 Associated Clinical Concern Items: ADHD and General Population Comparisons

Impairment & Functional Outcome Items. Individuals with either presentation of ADHD were expected to endorse the Impairment & Functional Outcome Items to a higher degree than individuals in the General Population group, as these items were designed to capture challenges commonly faced by individuals with ADHD and associated impairment, as required by diagnostic criteria. Although both ADHD groups were expected to endorse the Impairment & Functional Outcome Items, it was expected that individuals in the ADHD Combined group would have slightly higher ratings and/or demonstrate a greater number of impairments than individuals in the ADHD Inattentive group. This hypothesis was based on (a) the likelihood that some impairments are uniquely associated with hyperactive-impulsive symptoms, and (b) the high correlations that have been reported between number of symptoms and degree of impairment, given the fact that individuals with ADHD Combined typically have more total ADHD symptoms than individuals with ADHD Inattentive (Mannuzza et al., 2011).

Results of these analyses are summarized in Table 9.12 and 9.13. For Self-Report, the differences between each ADHD group and the General Population were statistically significant for all Impairment & Functional Outcome Items, with larger effects observed for items referring to general impairment (e.g., “Things are harder for me” yielded Cliff’s d = .68 for ADHD Inattentive and .72 for ADHD Combined when compared to individuals from the General Population). A similar pattern emerged in the Observer results; ratings for individuals from the ADHD Inattentive and ADHD Combined groups on the item “This person has a harder time with things than other people do” were markedly higher than for ratings of individuals in the General Population (Cliff’s d = .69 for ADHD Inattentive and .66 for ADHD Combined). Upon reviewing the percent of the sample with elevated items, relatively few individuals in the General Population had item elevations for these global impairment items (less than 12% for Self-Report, less than 16% for Observer); in contrast, the majority of individuals in the ADHD groups had elevations for these items (up to 82.2% for Self-Report, up to 88.9% for Observer).

More modest effect sizes were found for items that inquired about specific types of impairment, such as risky driving behaviors, when comparing the ADHD groups to the General Population (median Cliff’s d = .39 for Self-Report and .17 for Observer). The pattern of differences matched expectations. When comparing the ADHD Inattentive and the General Population groups, effect sizes range from small to medium (median Cliff’s d = .43 for both Self-Report and Observer). When comparing ADHD Combined and the General Population, the median effect size was large (Cliff’s d = .59) for Self-Report and medium (Cliff’s d = .37) for Observer.

Consistent with the preliminary hypothesis, these results demonstrate that the ADHD groups are much more likely to be elevated on the Impairment & Functional Outcome Items than individuals in the General Population. Similar to results found for the CAARS 2 Content Scales, Symptoms Scales, and Associated Clinical Concern Items, differences are larger for the Self-Report than the Observer report. These results highlight the importance of asking the individual themselves about these concerns, and not relying solely on observer ratings.

The differences between ADHD Inattentive and ADHD Combined groups on the Impairment & Functional Outcome Items are smaller than those noted above for comparisons between the ADHD groups and the General Population. The Self-Report results matched initial hypotheses that those in the ADHD Combined group would have slightly higher ratings and/or demonstrate a greater number of impairments than individuals in the ADHD Inattentive group, albeit with small effect sizes (median Cliff’s d = .11 for Self-Report across all items). These results, however, were not found in the Observer report (median Cliff’s d = |.10| for Observer; interestingly, the ADHD Inattentive group was slightly more likely to have elevated ratings for some items than the ADHD Combined group).

ADHD, Depression/Anxiety, and General Population Comparisons

Mean score differences between the General Population, individuals with Depression and/or Anxiety (i.e., Major Depressive Episode, Major Depressive Disorder, Persistent Depressive Disorder, Generalized Anxiety Disorder, Separation Anxiety Disorder, Social Anxiety Disorder, or Panic Disorder) without a co-occurring ADHD diagnoses, and individuals with ADHD (all presentations) were compared across the CAARS 2 (including Content Scales, Symptoms Scales, Associated Clinical Concerns items, and Impairment & Functional Outcome items). For the General Population sample, a subsample of the full General Population sample was selected that corresponded to the combined ADHD and Depression/Anxiety samples in terms of gender, age group, race/ethnicity, and education level (EL). The demographic characteristics of the ADHD, Depression/Anxiety, and General Population groups are presented in appendix J.

Content and DSM Symptom Scales. On average, individuals with Depression/Anxiety were expected to score higher than the General Population group but lower than those in the ADHD group on most CAARS 2 scales. Specifically, individuals with Depression/Anxiety were anticipated to have elevated scale scores where there is symptom overlap between these internalizing disorders and ADHD (i.e., Inattention/Executive Dysfunction, Negative Self-Concept, and Emotional Dysregulation; Newark et al., 2012; Tahmassian & Jalali-Moghadam, 2011). However, individuals with ADHD were expected to score significantly higher than individuals with Depression and Anxiety on scales that reflect features more unique to ADHD (i.e., Hyperactivity and Impulsivity). Differences in average scores for Emotional Dysregulation in particular were expected to be more modest, given that emotion regulation is a transdiagnostic feature that is common to ADHD and internalizing disorders, such as depression and anxiety (Newark et al., 2012; Tahmassian & Jalali-Moghadam, 2011).

Results from the ANOVAs met expectations and are found in Tables 9.14 to 9.15, as well as depicted graphically in Figures 9.4 to 9.5. As expected, significant differences were observed between the groups for all scales (p < .001). Results confirmed expectations in terms of distinct profiles between groups. Specifically, for both rater-types, the ADHD group and the Depression/Anxiety group had significantly higher scores than the General Population (with large effect sizes), indicating a higher overall level of impairment in the ADHD group and Depression/Anxiety group than in the General Population group for nearly all contrasts (median Cohen’s d = 1.62 for Self-Report, 1.27 for Observer). The Depression/Anxiety group also scored significantly higher than the General Population sample for the expected scales (median Cohen’s d = 0.69 for Self-Report, 0.58 for Observer). In addition, all scale-level comparisons were statistically significant, with the exception of the Hyperactivity, Impulsivity, and DSM ADHD Hyperactive/Impulsive Symptoms scales of the Observer form between Depression/Anxiety and the General Population group, consistent with the a priori hypothesis.

Moreover, differences between the ADHD group and Depression/Anxiety group revealed moderate to large effect sizes on nearly all scales (median Cohen’s d = 1.06 for Self-Report, 0.90 for Observer). The largest differences were observed for scales that target features unique to ADHD (i.e., hyperactivity and impulsivity symptoms). As expected, there were no significant differences between the groups for Emotional Dysregulation (small effect sizes; Cohen’s d = 0.24 for Self-Report and 0.16 for Observer). As predicted, the differences between the ADHD group and Depression/Anxiety group were negligible to small and not statistically significant for Negative Self-Concept. Notably, the ADHD group scores were significantly higher than the Anxiety/Depression group, with moderate to large effect sizes, for scales containing features and symptoms of ADHD that do notsuicidality typically overlap with those of internalizing disorders (e.g., for Hyperactivity, Cohen’s d = 0.94 for Self-Report and 0.92 for Observer). Overall, this pattern of observations reveals that, as expected, the CAARS 2 is able to distinguish between ADHD and General Population samples, but also can effectively distinguish ADHD from key clinical samples such as internalizing disorders (e.g., anxiety, depression).

Click to expand

Figure 9.4. Score Profiles of ADHD, Depression/Anxiety, and General Population Samples: CAARS 2 Self-Report

Note. INA/EDF = Inattention/Executive Dysfunction. HYP = Hyperactivity. IMP = Impulsivity. EM DYS = Emotional Dysregulation. NEG SC = Negative Self-Concept. ADHD-I = ADHD Inattentive Symptoms. ADHD-HI = ADHD Hyperactive/Impulsive Symptoms. ADHD-TOT = Total ADHD Symptoms.

Click to expand

Figure 9.5. Score Profiles of ADHD, Depression/Anxiety, and General Population Samples: CAARS 2 Observer

Note. INA/EDF = Inattention/Executive Dysfunction. HYP = Hyperactivity. IMP = Impulsivity. EM DYS = Emotional Dysregulation. NEG SC = Negative Self-Concept. ADHD-I = ADHD Inattentive Symptoms. ADHD-HI = ADHD Hyperactive/Impulsive Symptoms. ADHD-TOT = Total ADHD Symptoms.

Associated Clinical Concern Items. The Depression/Anxiety group was expected to show greater endorsement of the suicidality and self-injury items compared to individuals with ADHD or those from the General Population, and to show the highest endorsement for the Depression and Anxiety screener items relative to the other groups.

For these item-level analyses, differences are presented as Cliff’s d effect sizes, and results are presented in Table 9.16 and Figure 9.6. Both the ADHD group (Cliff’s d = .36 for Self-Report and .21 for Observer) and the Depression/Anxiety group (Cliff’s d = .40 for Self-Report and .39 for Observer) were far more likely to endorse the suicidality item than the General Population group. On the Self-Report, approximately 65% of both the ADHD and Depression/Anxiety groups endorsed the suicidality item (being 2.2 to 2.3 times more likely to endorse this item than the General Population group). These findings were somewhat more moderate on the Observer report, where 32.1% of the ADHD group and 48.9% of the Depression/Anxiety group endorsed this item (raters were 2.8 to 4.3 times more likely to endorse this item than the General Population group). A similar trend was found for the self-injury item; however, effect sizes were considerably smaller, with a lower rate of endorsement by all groups (Self-Report and Observer; ADHD, Depression/Anxiety, and General Population samples). On the Self-Report form, individuals in the ADHD and Depression/Anxiety groups were 1.4 to 1.9 times more likely to endorse the self-injury item relative to those in the General Population group. Similarly, on the Observer report, raters of individuals in the ADHD and Depression/Anxiety groups were 1.2 to 2.7 times more likely to endorse the self-injury item relative to those in the General Population group.

As expected, the highest endorsement of the depression and anxiety items occurred with individuals in the Depression/Anxiety group (up to 57.7% for Self-Report, up to 71.4% for Observer), followed by individuals with ADHD (up to 38.5% for Self-Report, up to 51.9% for Observer), and then by individuals in the General Population (up to 27.3% for Self-Report, up to 9.0% for Observer). There were large effect sizes between the Depression/Anxiety and General Population groups (Self-Report: Cliff’s d = .51 to .68, with individuals in the Depression/Anxiety group being 2.1 to 2.7 times more likely to endorse these items relative to the General Population; Observer: Cliff’s d = .62 to .74, with raters in the Depression/Anxiety group being 5.7 to 7.9 times more likely to endorse these items relative to the General Population). There were moderate to large effect sizes between the ADHD and General Population groups (Self-Report: Cliff’s d = .34 to .39, with individuals in the ADHD group being 1.4 to 1.8 times more likely to endorse these items relative to the General Population; Observer: Cliff’s d = .38 to .55, with raters in the ADHD group being 3.5 to 5.8 times more likely to endorse these items relative to the General Population).

Click to expand

Figure 9.6. CAARS 2 Associated Clinical Concern Items: ADHD, Depression/Anxiety, and General Population Comparisons

Impairment & Functional Outcome Items. Because the majority of the Impairment & Functional Outcome items were written specifically to address common impairments in adults with ADHD, individuals in the ADHD group were expected to have higher levels of endorsement on these items than individuals in the Depression/Anxiety group or individuals in the General Population group; however, individuals in the Depression/Anxiety group were expected to exhibit higher overall levels of impairment than the General Population group.

For these item-level analyses, differences are presented as Cliff’s d effect sizes, and results are presented in Tables 9.17 and 9.18. As hypothesized, the differences between the Depression/Anxiety group and the General Population were small to moderate in size (median Cliff’s d = .32 for Self-Report and .31 for Observer), whereas the differences between the ADHD group and the General Population group were more pronounced, with moderate to large effect sizes (median Cliff’s d = .51 for Self-Report and .38 for Observer). Additionally, small effects were observed for comparisons between the ADHD group and the Depression/Anxiety group (median Cliff’s d = .26 for Self-Report and .14 for Observer). Positive values indicate higher ratings of impairment, and results for most of the items followed the predicted pattern (viz., ADHD group > Anxiety/Depression group > General Population). It is not surprising that the one negative effect found within Self-Report data (viz., higher ratings for the Depression/Anxiety group) involved the sleep problems item, a diagnostic symptom of Major Depressive Disorder (APA, 2013).

Overall, hypotheses were confirmed in that individuals with ADHD endorsed considerably more impairment than individuals in the general population and somewhat more impairment than individuals with Depression/Anxiety.

Overall, these results lend support to the validity of constructs measured by the CAARS 2, as the items (and scales, discussed earlier in this section) effectively distinguish between individuals with distinct clinical disorders that share some, but not all, features and yield meaningfully different profiles of scores and endorsement.

Classification Accuracy

Classification accuracy is an index of validity akin to predictive validity evidence in that it considers the portion of respondents that are correctly classified into their respective groups (e.g., General Population vs. ADHD) based on their performance on a criterion measure (Jenkins et al., 2007). Binary classification modelling was used to predict the accuracy with which CAARS 2 scales correctly classified individuals from the General Population sample and individuals with ADHD based on (a) the Content Scales in concert with the Impairment & Functional Outcome Items, and (b) the DSM Symptom Scales in concert with the Impairment & Functional Outcome Items. Classification accuracy was modelled for these results because each set of scales and items represents the combination of symptomatic and impairment criteria required for the identification and diagnosis of ADHD (APA, 2013).

For the CAARS 2 Self-Report and Observer, binary classification modelling consisted of two sets of analyses for each form: (a) binomial logistic regression, followed by (b) the creation of confusion matrices and the derivation of classification accuracy statistics. Logistic regression is a statistical approach used to predict a dichotomous dependent variable from one or more predictor variables. For the first set of analyses, General Population vs. ADHD group membership represented the dichotomous dependent variable of interest, whereas the Content Scale T-scores, DSM Symptom Scale T-scores, and Impairment & Functional Outcome Item raw scores represented the predictor variables. The second set of analyses involved using the classification(s) predicted by the logistic regression models and actual group membership to construct confusion matrices. Using the approach outlined by Kessel and Zimmerman (1993), the confusion matrices were then used to derive classification accuracy statistics (see Response Style Analysis: Item Selection and Score Creation in chapter 6, Development, for a detailed explanation of these statistics).

A common approach to deriving classification accuracy statistics is to utilize matched samples with an equal number of individuals in each class (general population vs. clinical diagnostic group) to control for the influence of confounding variables. The approach outlined below employed samples matched on key demographic characteristics (including gender, age, race/ethnicity, and education level [EL]). Utilizing matched samples cultivates a base rate of 50%, or an equal likelihood that the individual being rated could belong to one class or the other (see Janes & Pepe, 2008, for a review). However, the real-world prevalence of a given clinical disorder can be higher or lower than 50% and thus the subsequent positive predictive value (PPV; i.e., the probability that an individual with a positive assessment result actually has a clinical disorder) and negative predictive value (NPV; i.e., the probability that an individual with a negative assessment result actually does not have a clinical disorder) will vary as a function of real world prevalence (see Robinson et al., 2016 and Lavigne et al., 2016 for reviews). The prevalence (or base rate of a given disorder in the relevant referral population) can vary widely depending on the purpose of the evaluation and the setting, and predictive ability will vary based on prevalence. For example, in a screening setting you might expect the prevalence of ADHD to be around 10% or less; whereas, in a clinically referred sample, a prevalence of approximately 50% may be more likely, rising to 60%–80% in an ADHD-specific clinical practice. Accordingly, the classification accuracy statistics of the CAARS 2 scales (assuming a 50% base rate) are presented in the following sections, and the PPV and NPV based on varying base rates are provided in subsequent tables for reference.

The matched samples described in appendix J were submitted to a series of logistic regressions to explore how well the scores predicted group membership. Specifically, logistic regressions were used to predict how well the

1. CAARS 2 Content Scale T-scores and Impairment & Functional Outcome Items distinguished between individuals from the General Population and those diagnosed with ADHD (all presentations),
   
   
2. DSM Total ADHD Symptoms T-scores and Impairment & Functional Outcome Items distinguished between individuals from the General Population and those diagnosed with ADHD (all presentations),
   
   
3. DSM ADHD Inattentive T-scores and Impairment & Functional Outcome Items distinguished between individuals from the General Population and those diagnosed with ADHD Predominantly Inattentive or Combined Presentations, and
   
   
4. DSM Hyperactive/Impulsive T-scores and Impairment & Functional Outcome Items distinguished between individuals from the General Population and individuals diagnosed with ADHD Predominantly Hyperactive/Impulsive or Combined Presentations.
   

The classification accuracy statistics are summarized in Tables 9.19 to 9.22. High levels of classification accuracy were demonstrated for both the combination of the CAARS 2 Content Scales and Impairment & Functional Outcome Items (overall correct classification rate = 92.5% for Self-Report, 91.7% for Observer) and the combination of the CAARS 2 DSM Symptom Scales and Impairment & Functional Outcome Items (overall correct classification rate = 89.7% to 92.3% for Self-Report, 84.7% to 88.5% for Observer). These findings demonstrate that the CAARS 2 exhibits a high level of accuracy with respect to correctly classifying individuals from the General Population and those with ADHD into their respective groups.