Exploring Replication in Statistical Modeling: A Guide to Avoiding Misleading Conclusions

Music City Center 

Constructing effective statistical models can be challenging for researchers of all experience levels. This talk highlights the critical role of replication and the risks of analyzing dependent observations as if they were independent. Such misrepresentation can inflate sample sizes, leading to incorrect p-value calculations and unreliable confidence intervals. Without properly assessing replication, researchers risk drawing invalid conclusions. Using illustrative examples, we clarify key concepts, offer practical techniques for distinguishing between types of replication, and highlight the consequences of common missteps. Finally, we introduce What Would Fisher Do (WWFD), a model-building tool that helps researchers determine appropriate degrees of freedom and build statistically sound models by accurately addressing sources of variation.

replication

model-building tool 

Section on Statistics and Data Science Education