03-01-2015 12:16 AM
Greeting to SAS communities,
My question is how to correct for sample selection bias in discrete choice modeling done by proc mdc? My research has two steps: Step 1) Predict whether a household chooses to purchase an internet connection in their home or not (yes vs no: binary logit), Regressors for step 1 includes household characteristic such as age, education, income, etc. Step 2) If a household chooses to purchase an internet connection, predict a choice of an internet package available in a household's area. (say five choices: conditional logit model). Regressors for step 2 includes household characteristics and choice characteristics such as monthly price, download speed and upload speed.
I am currently using doing only step 2 by proc mdc. However, my professor comments that my estimation may be biased. He suggests adding step 1 into estimation. Although proc qlim has a selection bias correction option, it cannot fit proc mdc-like dataset (dataset that have number of rows equal to number of choices for an individual). Could you pls suggest me an idea? Thank so much for every help.