Fit Statistics

The Adjusted R-squared value attempts to account for the addition of more effect variables. Values can range from 0 to 1. Values closer to 1 are preferred.

Akaike’s Information Criterion. Smaller values indicate better models, and AIC values can become negative. AIC is based on the Kullback-Leibler information measure of discrepancy between the true distribution of the response variable and the distribution specified by the model.

Corrected Akaike’s Information Criterion. This version of AIC adjusts the value to account for sample size. The result is that extra effects penalize AICC more than AIC. As the sample size increases, AICC and AIC converge.

The average squared error (ASE) is the sum of squared errors (SSE) divided by the number of observations. Smaller values are preferred.

The value of the F test in a one-way ANOVA after the variances are normalized by the degrees of freedom. Larger values are better, but can indicate overfitting.

The mean squared error (MSE) is the SSE divided by the degrees of freedom for error. The degrees of freedom for error is the number of cases minus the number of weights in the model. This process yields an unbiased estimate of the population noise variance under the usual assumptions. Smaller values are preferred.

The number of observations used in the model.

The p-value associated with the corresponding F statistic. Smaller values are preferred.

The R-squared value is an indicator of how well the model fits the data. R-squared values can range from 0 to 1. Values closer to 1 are preferred.

Square root of the MSE.

The Schwarz’s Bayesian Criterion (SBC), also known as the Bayesian Information Criterion (BIC), is an increasing function of the model’s residual sum of squares and the number of effects. Unexplained variations in the response variable and the number of effects increase the value of the SBC. As a result, a lower SBC implies either fewer explanatory variables, better fit, or both. SBC penalizes free parameters more strongly than AIC.