AIC / BIC Model Comparison Calculator
Compare two regression models from their residual sum of squares, sample size and parameter counts — AIC, corrected AICc, BIC, ΔAIC and Akaike weights.
How it works
AIC = n·ln(RSS/n) + 2k; AICc = AIC + 2k(k+1)/(n−k−1); BIC = n·ln(RSS/n) + k·ln(n); weight ∝ exp(−ΔAIC/2)
Adding parameters always lowers the residual sum of squares, so raw fit cannot choose between models. AIC penalises each parameter by 2 points; BIC penalises by ln(n), which is harsher for n above 7 and tends to pick simpler models. Only differences matter: a ΔAIC under 2 is a toss-up, over 10 is decisive. AICc should be used whenever n/k is below about 40. Both models must be fit to exactly the same observations.
Worked example
n = 50: model 1 (RSS 120, 3 parameters) AIC = 49.8; model 2 (RSS 105, 6 parameters) AIC = 49.1. ΔAIC = 0.7 — equivalent, and BIC prefers the simpler model 1.
Frequently asked questions
Can I compare AIC across different datasets?
No — the likelihood depends on the data, so AIC is only comparable for models fit to the same observations.
Why does BIC prefer simpler models?
Its penalty grows with sample size, so with lots of data a parameter must improve the fit substantially to be kept.
Related calculators
This calculator is for educational purposes. Double-check important results before acting on them.