

Zeroinflated poisson coefficients 

Message/Author 

Tracy Witte posted on Thursday, August 26, 2010  8:50 pm



I am very confused about how to interpret the coefficients for zeroinflated poisson regressions. For each regression, you get coefficients predicting the preponderance of zeroes and for the count predictors. To interpret them, do you take the antilog of the coefficients and interpret them as an odds ratio for increasing the variable of interest by 1 unit? I've done many web searches and I seem to find conflicting information on this. Any guidance would be greatly appreciated. I've pasted some of my output below. As you can see, some of the estimates are negative. I don't understand what this means...(and do you interpret the coefficients for the preponderance of zeroes differently than the coefficients for the count variable?) Estimate S.E. Est./S.E. PValue Y ON X1 1.692 0.455 3.715 0.000 X2 0.221 0.266 0.831 0.406 Y#1 ON X1 2.807 1.685 1.666 0.096 X2 0.059 0.746 0.079 0.937 


For the Y# ON equation you are using a standard logistic regression with a binary dependent variable (being in the zero class or not). So even though this binary DV is latent, the usual rules apply in terms of odds ratios. A negative slope simply means that the odds ratio is lower for being in the zero class when x2 increases. For the Y ON equation, things are a little different because the DV relates to counts. Look at Scott Long's book that we refer to in the UG. His pages 223226 are relevant. The factor change on p. 225 uses exp (i.e. antilog), but the wording is different from that of logistic regression. 

Back to top 

