The Conners 4–ADHD Index was developed using gradient-boosting machine learning (GBM; Friedman, 2001) algorithms to select the most effective items at distinguishing between ratings of youth with and without ADHD. Through an iterative process, 12 items were selected (separately for Parent, Teacher, and Self-Report) from the full-length Conners 4 for the Index. Raw scores were calculated based on relative importance of the items on the Index, and the ratio of the distribution of scores for youth with and without ADHD was used to derive a probability score. The probability score communicates the likelihood of a score profile resembling youth with ADHD.

Evidence of reliability was demonstrated through

• high internal consistency (median omega coefficients: .93 for Parent, .90 for Teacher, and .87 for Self-Report);

• high precision of measurement, both in terms of low values for standard error of measurement and as high values in target ranges of test information;

• moderate to high test-retest reliability, in which raw scores were found to be stable over a 2- to 4-week period, showing strong to very strong and statistically significant relationships (r = .91 for Parent, .89 for Teacher, and .66 for Self-Report); and

• inter-rater reliability, which showed strong relationships between raters of the same type (2 parent raters: r = .83, 2 teacher raters: r = .62), and weak to moderate relationships between different types of raters (Teacher/Self-Report r = .11; Parent/Self-Report r = .21; Parent/Teacher r = .48), highlighting the unique value of different perspectives.

Evidence of validity was supported through an investigation of the Index’s ability to correctly classify youth with and without ADHD. The overall correct classification rate was 85.7% for Parent, 73.6% for Teacher, and 72.8% for Self-Report.

Evidence of fairness, in terms of an absence of measurement bias and empirical differences, was investigated regarding gender, race/ethnicity, country of residence, and parental education level (PEL):

• Gender. There was no evidence of measurement bias for Parents, Teachers, and Self-Report (maximum ETSSD = 0.03), and negligible to small differences in probability scores were found (Cliff’s d = .03 to .25) when comparing ratings of males and females.

• Race/ethnicity. There was no evidence of measurement bias for Parents, Teachers, and Self-Report (maximum ETSSD = |.02|) when comparing ratings of White youth to ratings of Black and Hispanic youth. For Parent and Self-Report, there were negligible differences in probability scores for all group comparisons (White vs. Hispanic, White vs. Black, White vs. Asian; Cliff’s d = -.05 to .11). Teachers provided similar scores for White and Hispanic youth; however, teachers rated Black youth slightly lower (Cliff’s d = -.18) and Asian youth slightly higher (Cliff’s d = .18) than White youth, with small effect sizes.

• Country of Residence. Ratings of youth from the U.S. and Canada displayed equivalent measurement for Parents, Teachers, and Self-Report (maximum ETSSD =|0.05|) and negligible differences in probability scores were found (Cliff’s d = |.02| to |.06|) when comparing scores of youth from the U.S. and Canada.

• Parental Education Level (PEL). There was no evidence of measurement bias for Parents and Self-Report (maximum ETSSD = |0.02|) when comparing ratings of youth with high and low PEL, and negligible differences in probability scores were found (Cliff’s d = -.06 to .09) between different levels of parental education.