Related Experiment Video
Updated: Jul 14, 2025

The Collective Trust Game: An Online Group Adaptation of the Trust Game Based on the HoneyComb Paradigm
Published on: October 20, 2022
In models we trust: preregistration, large samples, and replication may not suffice
Martin Spiess1, Pascal Jordan1
1Institute of Psychology, Department of Psychology and Human Movement Science, University of Hamburg, Hamburg, Germany.
Statistical models in psychological research must account for data generation processes, especially concerning sample selection and assumption validity. Ignoring these factors, even with large samples or preregistration, can lead to biased conclusions and flawed inferences.
Area of Science:
- Psychological research methodology
- Statistical modeling in social sciences
Background:
- Replicability in psychological research is debated, yet selection mechanisms and statistical model assumptions are often overlooked.
- These overlooked issues question whether statistical tools accurately reflect the data generation process.
Purpose of the Study:
- To investigate the impact of selection mechanisms and violated model assumptions on the validity of inferences in psychological research.
- To highlight the limitations of common strategies like preregistration and large samples in addressing these issues.
Main Methods:
- Analysis of statistical inference validity under selective sampling conditions.
- Discussion of consequences of violated assumptions in linear models and generalized linear mixed models (e.g., in item response theory).
Main Results:
- Inferences from selective samples are only valid under very specific conditions, contrary to common practice.
- Violated model assumptions in psychological research are frequently ignored, leading to potential biases.
- Standard strategies (preregistration, large samples, replication) do not resolve these fundamental problems.
Conclusions:
- Addressing selection mechanisms and model assumptions is crucial for valid psychological research.
- Strategies like model diagnostics, methods for selectivity compensation, and non-parametric estimation can help mitigate bias.
- A fundamental solution involves iterative theory development with statistically justified assumptions and empirical research on self-selection.
Related Concept Videos
Mechanistic Models: Compartment Models in Individual and Population Analysis
Longitudinal Research
Group Design
Surveys
Contaminants and Errors
Another key consideration is determining the appropriate number of samples required to...
Naturalistic Observations

