hannanQuinnInfoCriterion

Computes the Hannan-Quinn criterion based on the sample size sampleSize, 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*ln(ln(n)) - 2L, for n observations, p parameters, and maximized log-likelihood L. Smaller is better.

The penalty per parameter, 2*ln(ln(n)), is negative for samples smaller than e (about 2.72) and remains below AIC's penalty of 2 until the sample size reaches e raised to the power e (about 15.15), beyond which it exceeds it and keeps growing. This is inherent to the criterion, not an artifact of the implementation, but it means that the criterion is intended for large samples.

Implementation based on Vose