Estimating causal effects from multiple-baseline studies: implications for design and analysis.

Area of Science:

• Behavioral research methodology
• Psychometrics
• Quantitative psychology

Background:

• Traditional average causal effect estimation in multiple-baseline designs relies on within-series comparisons.
• This method may be susceptible to bias from event effects like history and maturation.
• Concerns about bias motivate the development of alternative estimation techniques.

Purpose of the Study:

• To propose and evaluate a novel between-series estimator for multiple-baseline data.
• To compare the accuracy and bias of the between-series estimator against the traditional within-series estimator.
• To assess the utility of the between-series estimator in detecting inaccuracies in modeling assumptions.

Main Methods:

• Simulation studies were conducted with participants randomly assigned to intervention start points.
• The accuracy and bias of within-series and between-series estimators were assessed.
• The Type I error rate and power to detect event effects were evaluated.

Main Results:

• The within-series estimator demonstrated higher power but was biased by event effects, leading to inaccurate causal inferences.
• The between-series estimator remained unbiased and controlled the Type I error rate, irrespective of event effects.
• The difference between the two estimators can identify inaccuracies in modeling assumptions.

Conclusions:

• The between-series estimator provides unbiased causal effect estimates in multiple-baseline studies, even with event effects.
• The proposed method enhances the reliability of causal inferences in behavioral research.
• Researchers can use the comparison between within-series and between-series estimates to validate their analytical models.