If the model fits, use it: Methods and benchmarks for evaluating NBD-Dirichlet goodness-of-fit

•There are no formal tests of fit for the NBD-Dirichlet model.•This paper develops a suite of tests of fit to address this issue, including the novel ‘relative AAE’.•Benchmarks are empirically derived from 54 FMCG markets and the literature.•Finally a single test (purchase frequency AMAD) is identified that classifies 93% of categories the same as the full suite. The Dirichlet model is an empirical generalization describing and predicting repeated choice amongst a set of competitive alternatives. With the advent of big data, there are many new potential applications for this model. Its developers emphasized one goodness-of-fit statistic, and subsequent researchers have used this along with others. There is, however, no consensus in the literature regarding which measures to use or, more importantly, benchmarks. This paper proposes a suite of six goodness-of-fit statistics developed from the literature to assess the fit of the model and develops two new measures that account for category specific factors enabling the development of benchmarks. It also provides appropriate benchmarks for all statistics derived from 54 FMCG categories in the UK. Dirichlet模型是描述和预测在一组竞争性替代品中重复多次选择的 经验泛化模型。随着大数据的到来，这种模型有了很多新的应用潜力。它的开发者强调了一个拟合度统计量，随后的研究者也随之使用该方法。然而，对于选择什么量度以及更重要的选择什么基准，各文献中并没有共识。本文提出了一套从文献中开发的六种拟合度统计量，以评估模型的拟合度，并提出两个新的量度，考虑类别特定因素，便于研发基准。本文还为从英国54种消费品类别衍生的所有统计量提供了适当的基准。