From the list:
Dear Nlogit users,
is there a prevailing opinion how to combine RP and SP data in joint estimation: do I consider each individuals' RP choice only once or do I repeatedly estimate it with all SP choices?
I have seen both.
Your opinions are very welcome.
My response:
My opinion is that one should first employ each RP observation only once. Using them more than once seems like you are generating some artificial efficiency gains. That said, I know there is the issue where a large number of SP observations can swamp the influence of the lone RP observation and you lose some of the calibration gains. So it may depend on the ratio of SP to RP observations. What is your ratio?
Also, other than the increased cost to the researcher, there is little reason not to do it both ways and compare your results.
I'd be interested in what others might think as I don't have a lot of experience with the choice experiment type of SP data.
What do ya'll think?
