2018 Stern AssessingAccuracyCloseRelationships
The article reviews methods for measuring how accurately individuals perceive their romantic partners, distinguishing two primary forms of accuracy—mean‑level bias (systematic over‑ or under‑estimation) and correlational accuracy (tracking the truth). It outlines multilevel modeling techniques for assessing each form separately at both the sample (nomothetic) and individual (idiographic) levels, highlighting the importance of accounting for dyadic non‑independence. The authors then introduce the Truth and Bias (T&B) model (West & Kenny, 2011), which simultaneously estimates mean‑level bias and correlational accuracy while incorporating bias variables, truth variables, and moderators. Detailed step‑by‑step examples using a fictitious longitudinal dyadic dataset (distinguishable couples by political party) illustrate model specification, interpretation of parameters (directional bias, truth force, bias force), and extensions such as moderator effects, indirect accuracy, and calculation of individual accuracy scores. Limitations of traditional difference‑score and correlational approaches are discussed, and the T&B model is presented as a flexible solution that can address response‑bias, stereotype effects, and allow researchers to test whether bias can enhance or diminish accuracy. The article concludes with recommendations for researchers to adopt the T&B framework to advance understanding of accuracy and bias in close‑relationship research.
Topics
Mean‑Level Bias vs. Correlational Accuracy: Conceptual Foundations
Defines the two core accuracy constructs—mean‑level bias (systematic deviation from the truth) and correlational accuracy (extent to which judgments track the truth).
Multilevel Modeling for Dyadic Accuracy: Handling Non‑Independence
Explains why dyadic data require multilevel (crossed) models, distinguishes distinguishable vs. indistinguishable dyads, and outlines random‑intercept specifications.
Sample‑Level (Nomothetic) and Individual‑Level (Idiographic) Bias Analyses
Shows how to compute and interpret mean‑level bias and correlational accuracy at the group level and for each partner, including fixed‑effect predictors (e.g., extraversion).
Limitations of Difference‑Score and Simple Correlational Approaches
Highlights methodological caveats such as response‑bias confounding, inability to separate truth from shared bias, and issues with interpreting small discrepancy scores.
The Truth and Bias (T&B) Model: Integrated Framework
Introduces the T&B model that simultaneously estimates directional bias, truth force, and bias force, allowing bias variables to contribute positively or negatively to accuracy.
Operationalizing Truth, Bias, and Moderator Variables in T&B Analyses
Guides researchers on selecting appropriate truth variables (direct measures), bias variables (conceptually distinct cues), and moderators (e.g., relationship closeness) for model inclusion.
Interpreting T&B Model Parameters: Direct, Indirect, and Total Accuracy
Describes how to compute and interpret the truth force (direct accuracy), bias force, indirect accuracy (bias × truth), and overall accuracy scores.
Practical Example: Longitudinal Satisfaction Data from Distinguishable Couples
Walks through a fictitious dataset (10 couples, monthly satisfaction ratings) to illustrate model syntax, parameter estimation, and interpretation for both bias types.