Psychometrics Matter in Health Behavior: A Long-term Reliability Generalization Study.

OBJECTIVES: Despite numerous calls for increased understanding and reporting of reliability estimates, social science research, including the field of health behavior, has been slow to respond and adopt such practices. Therefore, we offer a brief overview of reliability and common reporting errors; we then perform analyses to examine and demonstrate the variability of reliability estimates by sample and over time.

METHODS: Using meta-analytic reliability generalization, we examined the variability of coefficient alpha scores for a well-designed, consistent, nationwide health study, covering a span of nearly 40 years.

RESULTS: For each year and sample, reliability varied. Furthermore, reliability was predicted by a sample characteristic that differed among age groups within each administration.

CONCLUSIONS: We demonstrated that reliability is influenced by the methods and individuals from which a given sample is drawn. Our work echoes previous calls that psychometric properties, particularly reliability of scores, are important and must be considered and reported before drawing statistical conclusions.