I conducted a choice experiment with the best-worst answering format. Each choice set had three alternative, so when I create the exploded dataset the dependent variable is given by the first best choice among three alternatives and then the second best between the remaining two, for each choice card. In Nlogit you can analyze this by adding an additional variable in the Lhs command (;Lhs=y, Nij)
clogit for MNL and mixlogit for MXL. For more info on the syntax: type in help clogit or help mixlogit. (that's the beautiful thing about stata)
in case you haven't installed the mixlogit command. first type in: ssc install mixlogit and stata will do it for you.
I agree with Sebastien on the choice of models. I endorse mixlogit over nlogit in this instance.
Ariel
You may also question the IIA assumption in your context.
You may use the hausman-McFadden test for that. The syntax for this test depends on your version of STATA. Type help hausman, as suggested by Sebastien.
Multinomial logit assumes IIA, but not mixed logit.
And I agree with Sebastien on using Stata's help files and built-in pdf manuals!
Additionally you can use the runmlwin command that allows you to fit models in MLwiN from Stata. You need to install that software first: This is the command with runmlwin: gen cons=1
runmlwin dv cons x1 x2 x3 , level2(schools: cons) level1(students) discrete(distribution(multinomial) link(mlogit) denom(cons) basecategory(4)) nopause
Where dv is Y
Sorry to bother you but I'm confused by this problem a lot. May I ask you a question about your words “In Nlogit you can analyze this by adding an additional variable in the Lhs command (;Lhs=y, Nij)”?
I have been written in my Nlogit program the code, but it can't run well.
the attachment is the picture of the code.
code1 is the original code. code2 is the code what I changed the form by data.What's wrong in it?
Really appreciate your help.
Dear Zhou,
my question was related to rplogit with variable number of choices, which usually occur when you adopt the best-worst format of answers. Say, for example, that you implement a choice experiment with 8 choice tasks, in which there are 3 alternatives (alternative A, alternative B and the status quo). You ask respondents to state their preferred (best alternative) alternative and the least preferred alternative (worst alternative). The respondent makes the best chiice among alternative A, alternative B and status quo, so he has 3 alternatives. When he chooses, say alternative A, the worst choice is between alternative B and the status quo, so between the remaining 2 alternatives. With this method, you collect more information for each respondent (16 observations instead of 8). In nlogit, you have to let the program understand that the number of choices is variable, in the sense that the first time there are 3 alternatives and the second time only 2, by creating a new variable that i called nij, to be added in the lhs command. When you include such variable, you don't have to include the ;choices command anymore.
i think your problem is slightly different. First of all, looking at your code, i see you did 't specify the lenght of the panel, i.e. The number of choice tasks. You can do this with the command ;pds (in the previous example would have been ;pds=8). Secondly, it is not clear to me how you organized your dataset. Is injury_1 your dependent variable? If so, what are the other variables injury_?
Dear Ginaluca,
Thanks for your help. I don't know how to express the gratitude because my poor English.
My data from the traffic accident reports. So I think my data look strange because there are only the characteristic of the accident happened. So It doesn't like your data have 3 choices, There just one choice of each accident, So the data not like the panel format. I'm very confusing about problem about data format.
Anyway, Thanks for your help! Have a nice day.
Best
Zhou
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation