Causal analyses of existing databases: no power calculations required

Area of Science:

• Epidemiology
• Biostatistics
• Health Research Methods

Background:

• Observational databases are crucial for causal inference.
• Researchers often face requirements for high statistical power before accessing data or funding.
• This focus on high power can hinder valuable observational research.

Purpose of the Study:

• To challenge the restrictive attitude towards observational analyses based on statistical power.
• To advocate for the estimation of causal effects rather than their detection.
• To promote the conduct and meta-analysis of multiple observational studies.

Main Methods:

• Critique of the "detect or undetected" misconception of causal effects.
• Argument for embracing imprecise estimates from multiple observational studies.
• Proposal for meta-analysis of findings from numerous studies.

Main Results:

• Causal effects are numerical quantities to be estimated, not binary signals.
• Imprecise estimates from observational analyses are not a reason to withhold research.
• Multiple studies with imprecise estimates are more valuable than no study.

Conclusions:

• The justification for withholding observational analyses based on imprecise estimates is flawed.
• Ethical considerations for power calculations in randomized trials do not apply to observational studies.
• Encouraging multiple observational analyses and subsequent meta-analysis is key to answering important causal questions.