请教关于zero inflated negative binomial 和 hurdle model 的区别

5056

收藏 2015-02-13

不知道有没有热心人可以指点一二，多谢赐教：）

I'm attempting to use zero-inflated or hurdle model to predict 1) the prob. of hours on care-giving >0 and 2) the amount of hours of care, if hrs>0. Here comes the problem.

1. When using command zinb with inflate(), it doesn't converge. I guess this could be due to not enough zeros in my data. So I changed to hurdle model using hnblogit. But this command wouldn't allow me to have factor variables. Is there anyone know about any alternative command/model? What's the actual difference between these two model? I'm also confused on the distinction between a general hurdle model and a double-hurdle model.

2. When adding more zero case in my data, my zero-inflated model converge. Using the margins command after zinb with inflate(), I obtained means at (variable =(0 1)). But I'm not quite sure whether this margin is derived from zinb part of the model, the inflate() part or both?

Your comments are highly appreciated!

扫码加我拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

全部回复

sweet91

2015-2-13 21:53:21

They differs in the sense that the Hurdle model(Mullahy,1986) allows that the zero and positive observations be generated by different mechanisms. However, both models are for obs contaminated by individuals having 0. As you said, you don't have enough zeros, then maybe they are not good models to choose from. Hope this helps.