Conducting sensitivity analyses to identify and buffer power vulnerabilities in studies examining substance use over...

Area of Science:

• Addiction Sciences
• Quantitative Psychology
• Longitudinal Research Methods

Background:

• A priori power analysis is vital for efficient research design in addiction sciences.
• Challenges exist in estimating parameters for power analyses due to uncertainty, limiting their utility.
• Longitudinal studies in addiction research often involve complex designs and numerous parameters.

Purpose of the Study:

• To demonstrate a sensitivity analysis approach for evaluating the impact of model parameters on statistical power.
• To illustrate how sensitivity analyses can inform strategies for managing power vulnerabilities in addiction research.
• To systematically investigate the influence of base rates, individual differences, and correlated predictors on power in longitudinal substance use studies.

Main Methods:

• Employed a sensitivity analysis framework to assess parameter effects on statistical power.
• Utilized three key design aspects relevant to substance use researchers: base rates, individual differences (random slopes), and correlated predictors (co-use).
• Examined how varying these parameters influences the power to detect associations in longitudinal data.

Main Results:

• Low or high substance use base rates can significantly reduce power, even with large sample sizes or numerous assessments.
• Substantial individual differences (random slopes) can negatively impact power, irrespective of increased assessments.
• Correlated predictors (co-use) generally have minimal detrimental effects on power unless the correlation is very high.

Conclusions:

• Increasing the number of participants is generally more effective for enhancing power than increasing the number of assessments.
• Optimizing assessment frequency relative to the occurrence of substance use can improve study power.
• Sensitivity analyses are instrumental in identifying parameter spaces vulnerable to estimation errors, enabling the design of more robust and resource-efficient addiction research studies.