akaikeInfoCriterion

Computes Akaike's information criterion (AIC) from the number of parameters estimated for the model numParameters and the maximized value lnMax of the log-likelihood function of the model. That is, 2p - 2L, for p parameters and maximized log-likelihood L. Smaller is better. The penalty 2p grows by 2 for each additional parameter, which is what makes the criterion trade fit against complexity.

Unlike the other information criteria, the plain criterion does not depend on the sample size. For a small-sample corrected form that does, see akaikeInfoCriterionCorrected.

Statistic.akaikeInfoCriterion(numParameters, lnMax)

Computes Akaike's information criterion (AIC) from the number of parameters estimated for the model numParameters and the maximized value lnMax of the log-likelihood function of the model. The sampleSize is validated but otherwise unused: AIC does not depend on it.

Retained only so that code written against R1.5 and earlier still compiles. Note that the value returned has changed: through R1.5 the penalty was the ratio (n - 2p + 2)/(n - p + 1), which lies in (0, 1] and shrinks as parameters are added, so the criterion fell as the model grew rather than rising. Callers that ranked models by it were ranking by log-likelihood alone, with a small bonus for complexity.