Origin and Nature of 0.05 Threshold

• 0.05 threshold relatively arbitrary, not biologically/naturally optimal
• Used widely in science publications for long time
• Generally effective: 5% wrong, 95% right - cost-effective for biological sciences
• Recent issues with p-hacking in high-profile articles

Publishing Data Threshold Guidelines

• 0.05 almost always appropriate for publication
• Editors and reviewers expect this standard
• Going below 0.05 is beneficial when possible
• Exceptions exist for specific circumstances

Exploratory Data Analysis Threshold

• Downloaded datasets from internet for hypothesis testing
• P-values slightly above 0.05 acceptable when:
  • Not publishing immediately
  • Using limited/free datasets
  • Testing basic hypotheses for later validation
  • Generating new hypotheses for future testing
• Don't need massively tiny p-values for exploration

Effect Size Considerations

• Small p-value insufficient alone
• R-squared example: 10% variation explained = poor despite tiny p-value
• Need good effect size AND small p-value
• Good correlation/explanation of data essential
• Publications with p-value 0.000000 but horrible R-squared are meaningless

Extraordinary Claims Standard

• Extraordinary claims require extraordinary data
• Example: extraterrestrials flying around NYC in UFOs
• P-value of 0.047 (< 0.05) completely insufficient
• Need "crazy small" p-value that's "unbelievable"
• Threshold must be so extreme it forces acceptance of extraordinary conclusion

Summary Guidelines

• Publishing: 0.05 threshold fine with good R-squared/correlation
• Exploration: flexible threshold for validation-intended research
• Extraordinary claims: extremely small p-values required