Modelling Excess Zeros in Count Data: A New Perspective on Modelling Approaches

Abstract

Summary We consider the analysis of count data in which the observed frequency of zero counts is unusually large, typically with respect to the Poisson distribution. We focus on two alternative modelling approaches: over‐dispersion (OD) models and zero‐inflation (ZI) models, both of which can be seen as generalisations of the Poisson distribution; we refer to these as implicit and explicit ZI models, respectively. Although sometimes seen as competing approaches, they can be complementary; OD is a consequence of ZI modelling, and ZI is a by‐product of OD modelling. The central objective in such analyses is often concerned with inference on the effect of covariates on the mean, in light of the apparent excess of zeros in the counts. Typically, the modelling of the excess zeros per se is a secondary objective, and there are choices to be made between, and within, the OD and ZI approaches. The contribution of this paper is primarily conceptual. We contrast, descriptively, the impact on zeros of the two approaches. We further offer a novel descriptive characterisation of alternative ZI models, including the classic hurdle and mixture models, by providing a unifying theoretical framework for their comparison. This in turn leads to a novel and technically simpler ZI model. We develop the underlying theory for univariate counts and touch on its implication for multivariate count data.

Enabling targeted mass drug administration for schistosomiasis in north-western Tanzania: Exploring the use of geostatistical modeling to inform planning at sub-district level

“…ZIP models were used to explore associations between the different socio-ecological covariates and schistosomiasis prevalence. These models are specified by simultaneously estimating the probability of finding zero cases (logistic distribution) as well as the expected count of cases [46]. To assess whether different endemic schistosoma species (S. haematobium and S. mansoni) occurred in different environmental conditions, separate models were fitted for urogenital and intestinal schistosomiasis prevalence data.…”

Moments of the Negative Multinomial Distribution

“…) (y j 1 + y j 2 + y j 3 ) +(r + 4) (5) (y j 1 y j 2 + y j 1 y j 3 + y j 2 y j 3 ) + (r + 5) (6) y j 1 y j 2 y j 3 , E[η 2 j 1 η 2 j 2 η j 3 η j 4 ] = y j 1 y j 2 y j 3 y j 4 (r + 3) (4) + (r + 4) (5) (y j 1 + y j 2 ) + (r + 5) (6) y j 1 y j 2 , E[η 2 j 1 η j 2 η j 3 η j 4 η j 5 ] = y j 1 y j 2 y j 3 y j 4 y j 5 (r + 4) (5) + (r + 5) (6) y j 1 , E[η j 1 η j 2 η j 3 η j 4 η j 5 η j 6 ] = y j 1 y j 2 y j 3 y j 4 y j 5 y j 6 (r + 5) (6) .…”

“…(r + 3) (4) + (r + 4) (5) (3y j 1 + y j 2 ) +(r + 5) (6) (y 2 j 1 + 3y j 1 y j 2 ) + (r + 6) (7) y 2 j 1 y j 2 , E[η 3 j 1 η j 2 η j 3 η j 4 η j 5 ] = y j 1 y j 2 y j 3 y j 4 y j 5 (r + 4) (5) + 3(r + 5) (6) y j 1 + (r + 6) (7)…”

“…E[η 2 j 1 η 2 j 2 η j 3 η j 4 η j 5 ] = y j 1 y j 2 y j 3 y j 4 y j 5 (r + 4) (5) + (r + 5) (6) (y j 1 + y j 2 ) +(r + 6) (7) y j 1 y j 2 , E[η 2 j 1 η j 2 η j 3 η j 4 η j 5 η j 6 ] = y j 1 y j 2 y j 3 y j 4 y j 5 y j 6 (r + 5) (6) + (r + 6) (7) y j 1 , E[η j 1 η j 2 η j 3 η j 4 η j 5 η j 6 η j 7 ] = y j 1 y j 2 y j 3 y j 4 y j 5 y j 6 y j 7 (r + 6) (7) .…”

“…Another motivation for using the negative multinomial distribution is its ability to handle excess zeros in count data, see, e.g., Haslett et al [5]. Count data often exhibit zero inflation, where there are more zeros than would be expected under a Poisson distribution.…”