eda.zi {ZIGP} | R Documentation |
'eda.zi' performs an exploratory data analysis on the influence of a covariate on the zero-inflation design (where the logit-link is assumed). Thereby, a discretization using scoring classes will be applied and empirical logits be calculated for each scoring class (see Czado et. al (2007)). Here, a shift of 1/2 is used to obtain well defined empirical logits even for 0. The dashed line is the empirical logit of 1/(number of scoring classes). Empirical logits further away from this line indicate high influence on zero-inflation.
eda.zi(x, y, numberclasses=5)
x |
Covariate considered |
y |
Response considered |
numberclasses |
Number of classes for discretization. Defaults to 5. |
As covariate x, discrete or continuous variables may be considered. Categorical covariates with two levels are allowed as well.
Notwithstanding the description in Czado et. al (2007), the empirical logits are now adjusted for individual class sizes.
Czado, C., Erhardt, V., Min, A., Wagner, S. (2007) Zero-inflated generalized Poisson models with regression effects on the mean, dispersion and zero-inflation level applied to patent outsourcing rates. Statistical Modelling 7 (2), 125-153.
data(Seatbelts) DriversKilled <- as.vector(Seatbelts[,1]) # will be response kms <- as.vector(Seatbelts[,5]/mean(Seatbelts[,5])) # will be exposure PetrolPrice <- as.vector(Seatbelts[,6]) # will be covariate 1 law <- as.vector(Seatbelts[,8]) # will be covariate 2 # artificially create some zeros DriversKilled[PetrolPrice<0.09] <- 0 eda.zi(x=PetrolPrice, y=DriversKilled) eda.zi(x=PetrolPrice, y=DriversKilled, numberclasses=200) eda.zi(x=law, y=DriversKilled)