Alternative Model-Based and Design-Based Frameworks for Inference From Samples to Populations: From Polarization to...

Area of Science:

• Psychological research methodology
• Statistical inference
• Quantitative psychology

Background:

• Traditional statistical inference relies on either model-based (Fisher) or design-based (Neyman) frameworks.
• These frameworks have limitations regarding sample types (nonrandom/random vs. only random) and inference types (descriptive vs. analytic) for finite/infinite populations.
• Current implementation in observational psychology research shows varied adoption of these frameworks.

Purpose of the Study:

• To compare model-based and design-based inference frameworks in psychology.
• To identify limitations of existing frameworks.
• To introduce and illustrate a novel hybrid model/design-based framework for enhanced statistical inference.

Main Methods:

• Comparative analysis of Fisher's model-based and Neyman's design-based frameworks.
• Examination of framework implementation in observational psychology.
• Development and application of a hybrid model/design-based framework using the High School and Beyond dataset.

Main Results:

• Model-based frameworks accommodate nonrandom/random samples for descriptive/analytic inference to finite/infinite populations.
• Design-based frameworks are restricted to random samples for analytic inference to finite populations.
• The hybrid framework overcomes limitations, allowing both descriptive and analytic inference for both finite and infinite populations with random samples.

Conclusions:

• A hybrid model/design-based framework offers a more flexible and comprehensive approach to statistical inference in psychology.
• This unified framework enhances the ability to draw robust conclusions from diverse psychological research data.
• The High School and Beyond dataset serves as a practical example of the hybrid framework's utility.