A Comparative Study On Modeling And Testing Methods For Zero-inflated Continuous Data

The zero-inflated model is widely used in many fields such as health care,dietary intake,and clinical trials.At present,the domestic scholars’ research on the zero-inflated model mainly focuses on the discrete distribution field,while the research on the continuous distribution is somewhat failing behind.Therefore,this paper intends to make innovation contributions in the hypothesis testing of the zero-inflated continuous model.First of all,according to the characteristics of the data,this paper proposes three models: ZILN,ZIE and ZIG.Then the hypothesis is established based on the consideration of mean equal and same distribution,and the hypothesis test is carried out under all three models.Finally,by random simulations comparing the first type of errors and efficacy levels of T,Wilcoxon,Wald,LR,and Score under the above conditions.The simulation results show that under the null hypothesis of the mean,the three models can control the first type of error near 0.05 in other four tests except Wilcoxon test.In terms of efficacy,Wald has the best test results for the ZIG and ZIE models,followed by LR,Score,and T;in the ZILN model,the test results are different where p =0 or p ?0.Under the assumption of same distribution,the first type of error can be controlled at around 0.05 and tends to be stable as the sample increases.In terms of test effect,the LR test is significantly better than other tests.When the sample is large enough,except the Wilcoxon test,all the other tests are close to 1.After obtaining the above results,the methods were applied to an example of a longitudinal dietary intervention.Excessive zero values for certain dietary indicators of adolescent patients with type 1 diabetes(e.g.total fruit,whole fruit,dark green orange vegetables and beans and whole grains)increase the difficulty of the analysis.This paper introduces a zero-inflated model.According to the AIC information criterion,the AIC values of ZILN,ZIE and ZIG models under four indexes are compared.Finally,ZIG model is selected to model and analyze the data.In this paper,five commonly used hypothesis testing methods are compared for zeroinflated continuous data under three models and two null hypotheses.The research results in this paper are for the workers in the fields of clinical experiment,medical and health,and provide theoretical guidance for the test of zero-inflated continuous data.According to the model fitted by the data,the statistician can directly use the optimal test method to improve the efficiency of the statistical test.