Page 1 of 1
IIA
Posted: 04 Feb 2021, 09:23
by svenne
Stephane & David,
I would appreciate some sort of IIA test as an apollo_ function.
Best
Sven
Re: IIA
Posted: 04 Feb 2021, 14:01
by stephanehess
Hi Svenn
David and I have discussed this but have decided against implementing something. It would be very difficult to do something that would be generic enough to work across case studies. I suggest you instead have a look at the discussion on IIA tests in chapter 3 in Kenneth Train's book, and follow some of the suggestions there
Best wishes
Stephane
Re: IIA
Posted: 05 Feb 2021, 16:07
by svenne
Stephane,
I see. I have done a Hausman-MacFadden test
Code: Select all
hmft <- ( (sub.m.coef- full.m.coef) %*%
solve(sub.m.vc - full.m.vc)
) %*% (sub.m.coef- full.m.coef)
abs(hmft)
where sub.m is the model with the reduced choice set and full.m is the model with all alternatives. coef and vc are extracted from the apollo model object. hmft then is checked against qchisq().
Best,
Sven
Re: IIA
Posted: 05 Feb 2021, 17:46
by stephanehess
Sven
nice, and that works. But of course generalising it to a function that creates a reduced choice set is the hard part, which led to our decision
Stephane
Re: IIA
Posted: 06 Jul 2021, 12:26
by asilvestri
Sven,
Thank you for suggesting how to compute a hausman test in R. I was able to do as you explain and derive the hmft for my model, however I wonder how do you check it against the qchisq(). I am not particularly familiar with r, I would appreciate if you could explain how do you compuite the hypotesis testing once calculated the hmft.
Best,
Alessandro
Re: IIA
Posted: 07 Jul 2021, 12:40
by svenne
If I remember right, hmft is compared to the critical value from the X^2 with the respective degree of freedom (i.e., 1): qchisq(hmft,1), I think it is. If hmft is larger than this value, the two models are considered to be different, i.e., IIA does not hold.
Does this help?
Best,
Sven
Re: IIA
Posted: 08 Jul 2021, 10:50
by asilvestri
Yes, thank you very much!
Best,
AS
Re: IIA
Posted: 03 Mar 2022, 18:41
by malemu588@gmail.com
Hi Sven, I tried to estimate full and reduced models, and then directly applied the code you provided. It didn't work for me. By the way, are sub.m and full.m model names in your case?
Thanks in advance.
Best,
Mohammed
Re: IIA
Posted: 07 Mar 2022, 09:29
by svenne
Hi Mohammed,
yes, sub.m and full.m are the model object names. What makes the code not work?
Best
Sven