UniODA vs. Chi-Square: Deciphering R x C Contingency Tables

Paul R. Yarnold

Optimal Data Analysis, LLC

Whereas omnibus effects identified using chi-square to assess association in R x C contingency tables are often difficult or impossible to disentangle when R and C are greater than two, identifying the structure underlying R x C tables using UniODA is straightforward. An investigation of the association between furniture factory production shift and type of defect is presented as an example.

View journal article

UniODA vs. Eyeball Analysis: Comparing Repeated Ordinal Scores

Paul R. Yarnold

Optimal Data Analysis, LLC

The Adverse Drug Reaction Probability Scale (APS) is an algorithm that is used to rate the probability that an adverse drug event is drug-induced. Prior research compared the reliability of APS ratings generated by 15 pairs of independent experts between baseline and training, versus between training and three-month follow-up. Findings of eyeball comparison of ranges of the intraclass correlation coefficient of reliability, versus findings obtained using UniODA to contrast the two testing periods, are compared.

View journal article

UniODA vs. Sign Test: Comparing Repeated Ordinal Scores

Paul R. Yarnold

Optimal Data Analysis, LLC

The Adverse Drug Reaction Probability Scale (APS) is used to rate the probability that an adverse drug event is drug-induced. Prior research compared the inter-rater reliability of APS ratings generated by 15 pairs of independent experts at two testings: (1) immediately after training, and (2) three months after training. Findings obtained using the sign test versus UniODA to contrast testings are compared.

View journal article

UniODA vs. Spearman Rank ρ: Between-Raters Reliability of Scores on the Adverse Drug Reaction Probability Scale

Paul R. Yarnold

Optimal Data Analysis, LLC

The Adverse Drug Reaction Probability Scale (APS) algorithm is widely-used for rating the probability that an adverse drug event is drug-induced. Prior research (Figure 1, p. 695) presented APS ratings generated by two independent experts for N = 129 challenging cases. Between-rater reliability of these ratings is computed by and compared between Spearman’s rank-order correlation and UniODA.

View journal article

UniODA vs. t-Test: Low Brain Uptake of L-[11C]5-hydroxytryptophan in Major Depression

Paul R. Yarnold

Optimal Data Analysis, LLC

Prior research employed t-test to compare mean uptake of [11C]5-HTP across the blood-brain barrier for eight healthy volunteers versus six patients with unipolar depression, however no statistically significant mean difference was identified for basil ganglia, caudate nucleus, or lentiform nucleus. When analyzed via UniODA, significantly lower mean uptake was identified for basil ganglia and caudate nucleus.

View journal article

UniODA vs. Logistic Regression and Fisher’s Linear Discriminant Analysis: Modeling 10-Year Population Change

Paul R. Yarnold

Optimal Data Analysis, LLC

UniODA-based classification models involving one attribute yielded greater overall, maximum, and minimum classification accuracy in modeling 10-year population change than legacy models involving five attributes and derived using either logistic regression analysis or Fisher’s linear discriminant analysis.

View journal article

Gender and Psychology Concentration for Graduate Students

Paul R. Yarnold

Optimal Data Analysis, LLC

Gender and psychology concentration are cross-classified for N = 100 graduate students, resulting in a 2 x 5 contingency table. Estimating the nature and magnitude of the association between these variables by legacy statistical methods is inappropriate owing to computational and interpretative difficulties. However, obtaining and interpreting the maximum-accuracy model for these data is straightforward.

View journal article

Is Going First an Advantage in Cribbage?

Paul R. Yarnold

Optimal Data Analysis, LLC

To assess if going first is an advantage in the card-game cribbage, an experienced human cribbage player competed against a computer for 100 games. When findings were analyzed using non-directional chi-square analysis there was no statistically significant effect. However, analysis using confirmatory UniODA revealed that the first move is a marginally significant (p<0.08) predictor of winning a cribbage game, and is a statistically significant (p<0.04) predictor if weights that are used by rule to indicate the magnitude of win (win = 1 point; skunk = 2 points; double-skunk = 3 points) are applied in analysis.

View journal article

UniODA vs. Kruskal-Wallace Test: Gender and Dominance of Free-Ranging Domestic Dogs in the Outskirts of Rome

Paul R. Yarnold

Optimal Data Analysis, LLC

The Kruskal-Wallace test and UniODA are both used to compare investigator-ranked dominance ratings of male versus female dogs. Exposition highlights the insufficiency of statistical reliability as a singular criterion for evaluating model performance, as well as the importance of testing confirmatory hypotheses in the context of increasing statistical power when assessing statistical reliability.

View journal article

Optimal Statistical Analysis Involving Multiple Confounding Variables

Paul R. Yarnold

Optimal Data Analysis, LLC

This paper demonstrates a maximum-accuracy statistical approach that assesses the effect of two or more confounding variables on the estimated association of a class variable and attribute(s). An example involves modeling patient self-ratings of the likelihood that they will recommend an Emergency Department to others (class variable), based on five patient-rated dimensions of physician care behavior (attributes), and patient satisfaction ratings for amount of time waiting to see the physician, and for amount of time waiting in the registration room before going to the treatment area (confounders).

View journal article

GO-CTA vs. Marginal Structural Model: Observed Data from a Point-Treatment Study, Stratified by Known Confounder

Paul R. Yarnold

Optimal Data Analysis, LLC

Statistical evaluation of a treatment effect in the context of a known confounder is contrasted between globally-optimal classification tree analysis (GO-CTA) and marginal structural model (MSM) methods. Unlike MSM, the validity of GO-CTA doesn’t require data to satisfy numerous assumptions, and maximum-accuracy models explicitly identify underlying structure and strength of the confounding effect.

View journal article

Optimal Statistical Analysis Involving a Confounding Variable

Paul R. Yarnold

Optimal Data Analysis, LLC

This paper compares maximum-accuracy statistical approaches that address the effect of a confounding variable on estimated association of class variable and attribute. An example involves modeling patient ratings of likelihood they will recommend an Emergency Department to others (class variable) on the basis of five aspects of physician care behavior (attributes), as well as patient satisfaction with the amount of time spent waiting to see the physician (confounder).

View journal article