Quick reference
Common significance thresholds
p < 0.001
Highly significant
p < 0.01
Very significant
p < 0.05
Significant (default)
p < 0.10
Marginal
p ≥ 0.10
Not significant
Rules of thumb
Pick one-tailed only when the effect direction is decided in advance.
Report effect size and confidence intervals alongside p — never on their own.
Don't p-hack: fix your test, sample size, and α before peeking.
Combine it with practical significance, confidence intervals, and domain knowledge before you make a call.
State the hypothesis first
Define your null and alternative before collecting data. Deciding after the fact inflates false positives.
Mind the effect size
With a huge sample, tiny irrelevant differences become 'significant'. Always report the magnitude, not just p.
Correct for multiple tests
Running 20 comparisons at α = 0.05 gets you a false positive by chance. Use Bonferroni or FDR corrections.
