Dear all,

Hope you all are doing well.

I have graphical results of K-fold cross-validation checks and I am wondering how to interpret those results precisely. In fact, the K-fold cross-validation checks were used to compare models performance of counts data, including Negative Binomial Regression I (NB1), Hurdle-NB1, Hurdle-NB2, and Zero-inflated NB2. Followings are author's interpretations "Figure 8.7 shows a comparison of the NB models for the number of office-based visits. Although the NB2 model is the worse performer for each replicate, its hurdle counterpart performs quite well - either the best or close to the best-performing model" By Deb, Norton, and Manning.

Array

Note: NB1 is the first bar, Hurdle-NB2 is the second, Hurdle-NB1 is the third, and Zi-NB2 is the fourth. The figure is taken from Health Econometrics Using Stata Book written by Deb, Norton, and Manning.

Results of log likelihood, AIC, and BIC are presented below.
Code:
                        K      LogLik         AIC         BIC
     Poisson          37  -10682.042   21438.085    21729.36
         NB2          38  -9995.2177   20066.435   20365.583
         NB1          38  -10020.414   20116.828   20415.976
Hurdle_Poi~n          74  -10118.265   20384.531   20967.082
  Hurdle_NB2          75  -9947.3831   20044.766   20635.189
  Hurdle_NB1          75  -10057.127   20264.254   20854.677
         ZIP          74  -10113.589   20375.177   20957.728
       ZINB2          75  -9937.9712   20025.942   20616.365
Could anyone help me understand the figure more clearly?

Thank you and have a nice wk!

DL