A flexible method for aggregation of prior statistical findings

Area of Science:

• Quantitative synthesis
• Meta-analysis
• Scientific modeling

Background:

• Scientific output is rapidly growing, necessitating advanced methods for synthesizing prior research.
• Current meta-analysis techniques are limited to studies with similar designs, restricting broader data aggregation.
• A need exists for flexible methods to combine diverse prior models into a unified meta-model.

Purpose of the Study:

• Introduce and validate Generalized Model Aggregation (GMA) as a novel quantitative synthesis method.
• Demonstrate GMA's ability to aggregate models with varied structures into a meta-model.
• Showcase GMA's utility in advancing scientific understanding and assessing prior research reliability.

Main Methods:

• Developed Generalized Model Aggregation (GMA) for combining prior estimated models.
• Imposed minimal restrictions on the structure of prior models and the resulting meta-model.
• Validated GMA using published equations for Basal Metabolic Rate and through numerical examples.

Main Results:

• GMA successfully aggregated 27 published equations from 16 studies into a predictive meta-model for Basal Metabolic Rate.
• The GMA-derived Basal Metabolic Rate equation outperformed existing models.
• GMA identified novel nonlinearities and estimated biases in measurement methods, and provided unbiased estimates from mis-specified prior studies.

Conclusions:

• Generalized Model Aggregation (GMA) significantly extends the capabilities of meta-analysis.
• GMA can be applied across various scientific domains to compare theories, develop new models, and evaluate research reliability.
• This method leverages previous findings more effectively, advancing quantitative synthesis in science.