New normal-based classes of probability distributions
classes of probability distributions; normal distribution; maximum likelihood; Monte Carlo simulation; real data modelling
In this work we present three classes of probability distributions build upon an innovative method that considers the Lebesgue integration of a function denoted by H(t). For the three classes we set H(t) to be the standard normal cumulative distribution function, differing only in the limits of integration. The upper limit of integration of the second class is unprecedented in the statistical literature and the third class admits two different baselines, being a competing alternative to twocomponent mixture models. The classes have the advantage of demanding no extra parameters, besides those of the baseline(s), providing thereby parsimonious models. Under certain conditions, the distributions generated by the new classes are identifiable. We present some mathematical properties of the classes, like the linear representation of the probability density function, the moments and the moment generating function. Monte Carlo simulation studies are performed to investigate the behavior of the maximum likelihood estimates of the parameters for some submodels of the classes. The same submodels examined in the simulations are also used in applications to real datasets. Information criteria and formal goodness-of-fit statistics (Cramér-von Mises and Anderson-Darling) are used as criteria for model selection in comparisons considering the studied distributions emerged from the classes and some well-known distributions commonly employed in modelling similar data. The results suggest that the distributions from the proposed classes outperform the competing models