Longitudinal zero-inflated count data are gathered from a series of measurements of experimental individuals at multiple time points and possess the very characteristics of longitudinal data, count data and zero-inflated data simultaneously.The number of zeros contained in count data far exceeds that randomly generated by classical discrete distributions such as Poisson and negative binomial distributions.In the regression analysis of longitudinal zero-inflated count data, the data are generally assumed to consist of zero-inflated components and random sampling components, and the mean, zero-inflation rate and dispersion are three main characteristic parameters of the data.Nowadays, most regression models for zero-inflated data only consider the influence of covariates on the mean and zero-inflation rate and the dispersion is unfortunately igored or set to a fixed value.As a result, these models cannot be applied to the situations where dispersion at different observation time points dynamically changes with time or other covariates.For the count data following the zero-inflated negative binomial distribution, a generalized regression model is develpoed to describe the relationship between all three data characteristics and the covariates.The maximum likelihood estimate of the regression coefficients is obtained by using the EM (Expectation-Maximization) algorithm to simultaneously the three regression coefficients.Theoretical analysis shows that, in comparison with the two-parameter regression model estimation, the proposed likelihood estimator is consistent and asymptotically normal, enabling accurate simultaneous estimation of the three regression coefficients.Finally, simulation results demonstrate that the proposed model is more accurate and effective than the regression models that do not consider time-varying dispersion.
MullahyJ.Specification and testing of some modified count data models [J].J Econom, 1986, 33(3): 341-365.
[2]
LambertD.Zero-inflated Poisson regression, with an application to defects in manufacturing [J].Technometrics, 1992, 34(1): 1-14.
[3]
YauK K W, WangK, LeeA H.Zero-inflated negative binomial mixed regression modeling of over-dispersed count data with extra zeros [J].Biom J, 2003, 45(4): 437-452.
[4]
FontaniveV N, BrochB, SeoaneM, et al.Oral impact of dental prosthesis use in different patterns of tooth loss [J].J Prosthet Dent, 2025, 134(5): 1781-1786.
[5]
ChoH, LiuC, PreisserJ S, et al.A bivariate zero-inflated negative binomial model and its applications to biomedical settings [J].Stat Methods Med Res, 2023, 32(7): 1300-1317.
[6]
OzorJ, RondeP, TungS, et al.The “Middleground” as a catalyst for the dynamics of innovation in an ecosystem? The case of Eurasanté in Hauts-de-France [J].Technol Forecasting Social Change, 2024, 209: 123731.
[7]
CanigliaM.Economic insecurity in young adulthood among children maltreated during early adolescence [J].Child Abuse Neglect, 2024, 157: 107047.
[8]
KangK I, KangK, KimC.Risk factors influencing cyberbullying perpetration among middle school students in Korea: Analysis using the zero-inflated negative binomial regression model [J].Int J Environ Res Public Health, 2021, 18(5): 2224.
[9]
WengJ, YangD, QianT, et al.Combining zero-inflated negative binomial regression with MLRT techniques: An approach to evaluating ship accident casualties [J].Ocean Eng, 2018, 166: 135-144.
[10]
ChoH, LiuC, PreisserJ S, et al.A bivariate zero-inflated negative binomial model and its applications to biomedical settings [J].Stat Methods Med Res, 2023, 32(7): 1300-1317.
[11]
HuY N, TianM Z.Joint modeling and variable selection from zero-inflated count data [J].Statistical Research, 2019, 36(1): 104-114.
AlfòM, MaruottiA.Two-part regression models for longitudinal zero-inflated count data [J].Can J Stat, 2010, 38(2): 197-216.
[16]
AtkinsD C, BaldwinS A, ZhengC, et al.A tutorial on count regression and zero-altered count models for longitudinal substance use data [J].Psychol Addict Behav, 2013, 27(1): 166-177.
[17]
SamaniE B, AmirianY, GanjaliM.Likelihood estimation for longitudinal zero-inflated power series regression models [J].J Appl Stat, 2012, 39(9): 1965-1974.
[18]
ZhuH, LuoS, DeSantisS M.Zero-inflated count models for longitudinal measurements with heterogeneous random effects [J].Stat Methods Med Res, 2017, 26(4): 1774-1786.
[19]
ZhuH, DeSantisS M, LuoS.Joint modeling of longitudinal zero-inflated count and time-to-event data: A Bayesian perspective [J].Stat Methods Med Res, 2018, 27(4): 1258-1270.
[20]
WangJ, Reyes-GibbyC, SheteS.An approach to analyze longitudinal zero-inflated microbiome count data using two-stage mixed effects models [J].Stat Biosci, 2021, 13: 267-290.
[21]
ZhangW, QianJ W F, ChenY.A joint mean-correlation modeling approach for longitudinal zero-inflated count data [J].Braz J Probab Stat, 2020, 34(1): 35-50.
[22]
McLachlanG J, KrishnanT.The EM algorithm and extensions [M].Hoboken: John Wiley Sons, 2008.
[23]
ZeileisA, KleiberC, JackmanS.Regression models for count data in R [J].J Stat Softw, 2008, 27(1): 1-25.
[24]
CameronA C, PravinK T.Regression analysis of count data [M].Cambridge: Cambridge University Press, 2013.
[25]
HilbeJ M.Negative binomial regression [M].Cambridge: Cambridge University Press, 2011.