# [R] predictions from hurdle model

erin erinzylstra at gmail.com
Thu Aug 2 20:08:47 CEST 2012

```I ran a negative binomial logit hurdle model and am now trying to plot the
effects of a continuous predictor variable (the only variable in my model)
on the count and zero component and the overall mean response.  I'm confused
because for some values, the predicted overall mean is higher than the mean
of the non-zero counts (range of predicted overall means=2.2-11.0; range of
non-zero count means=2.8-4.6).  Is this possible or did I do something
wrong?  I did not expect these results because I thought the probability of
getting a zero response should reduce the overall mean below the non-zero
count mean.

I attached the relevant code and output  below.  Any thoughts or suggestions
would be much appreciated!  I'm new to hurdle models and their
interpretation.  Thank you,
Erin

> library(pscl)
> nbhurdle.prev<-hurdle(LLF.a~y.prev,data=full.cts,dist="negbin")
> summary(nbhurdle.prev)

Call:
hurdle(formula = LLF.a ~ y.prev, data = full.cts, dist = "negbin")

Pearson residuals:
Min      1Q  Median      3Q     Max
-0.6398 -0.3333 -0.3333 -0.1363  7.3145

Count model coefficients (truncated negbin with log link):
Estimate Std. Error z value Pr(>|z|)
(Intercept)  1.031508   0.470804   2.191  0.02846 *
y.prev       0.005042   0.001835   2.747  0.00601 **
Log(theta)  -1.856429   0.561709  -3.305  0.00095 ***
Zero hurdle model coefficients (binomial with logit link):
Estimate Std. Error z value Pr(>|z|)
(Intercept) -0.915944   0.103853  -8.820  < 2e-16 ***
y.prev       0.048832   0.009079   5.379 7.51e-08 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Theta: count = 0.1562
Number of iterations in BFGS optimization: 20
Log-likelihood: -931.7 on 5 Df

> y.prev<-seq(0,100,by=1)
> newdata<-data.frame(y.prev)
> nonzero<-1-predict(nbhurdle.prev,newdata=newdata,type="prob")
> countmean<-predict(nbhurdle.prev,newdata=newdata,type="count")
> allmean<-predict(nbhurdle.prev,newdata=newdata,type="response")
> plot(y.prev,nonzero[,1],type="b")
> plot(y.prev,countmean,type="b")
> plot(y.prev, allmean, type = "b")

--
View this message in context: http://r.789695.n4.nabble.com/predictions-from-hurdle-model-tp4638924.html
Sent from the R help mailing list archive at Nabble.com.

```