I am new to R's pscl package, and struggling to interpret output from a hurdle model. I have a data frame
df which contains a count variable named
count (containing a lot of zero's) and two predictor variables:
VAR1 contains two levels(
VAR2 contains three (
I fit the model with
fit = hurdle(count ~ var1 + var2, data = df, link = 'logit', dist = 'negbin'
and summary(fit) returns something like this:
Zero hurdle model coefficients (binomial with logit link): Estimate Std. Error z value Pr(>|z|) (Intercept) -2.799e+00 8.638e-01 -3.241 0.00119 ** VAR1lev2 8.924e-01 3.257e-01 2.740 0.00614 ** VAR2levB 1.236e+00 4.972e-01 2.487 0.01288 * VAR2levC 5.217e-01 3.379e-01 1.544 0.12259
I have two major concerns.
1) I understand that the intercept represents the 'baseline' likelihood of a positive value for
count. How does one interpret a negative intercept? Does it simply mean that
count is more likely to be 0, as opposed to >0? More importantly, does a negative intercept affect interpretation of estimated values for my predictor variables? (and if so, how?)
2) I understand that each level of
VAR2 is being compared to a reference level. So, there is a significant difference in estimated count between
VAR1lev2. Is there an sensible way to infer differences between two non-reference levels - in this case, between