An question about hierarchical（multilevel）logistic regression，thank you for your generous help！

[Cassie Liu]

Hello everyone：
I recently read some references about hierarchical linear model and hierarchical logistic regression model，therefore，I want to use hierarchical logistic regeression to do my research about fertility disire.As for my dependent variable,it is classified into 4 aspects such as:boy preference,girl preference,none preference and both preference.I dont know if i can use hierarchical logistic regression or the hierarchical logistic regressin only suit for bivariate dependent variable?My second question is the commands of this model,"meqrlogit or xtmelogit or gllamm?"I was wondering which command is suitable for my research question.The last question is about the group variable ,my independent variables are classified into two levels,region (city)and individual level,so I was wondering which level can be considered as the group variable?I am not good at english,thank you for your generous help.
--Sincere,Cassie

[Cassie Liu]

Sorry，“a” question，not “an” question.

[Joseph Coveney]

You might want something more like hierarchical (multilevel) multinomial logistic regression. I've not done this, but I understand that you can use either the user-written gllamm, or the official Stata command gsem to fit such a model.

To answer your questions, you would probably not want to pursue using either meqrlogit or xtmelogit, as these are for binary responses, and you seem to have four preference choices.

There are a couple of questions and considerations that you might need to address during setting up of the regression model.

1. I assume that your dataset is from a simple sample of couples at a convenience sample of fertility clinics, that is, the dataset is not from a complex, designed population survey. Is that right?

2. If your dataset is from fertility clinics and your dataset has a lot of regions / cities with only a single fertility clinic, then it might be better to use the clinic, itself, and not the region /city as the second level.

3. You have four choices for an "individual level" preference for the child's sex, in particular, "both". There are less-likely alternative possibilities, but does that mean that the "individual level" is actually a couple, and that "both" represents a split decision (one member of the couple wants a girl and the other wants a boy)? If that's the case, then you might want to consider a three-level hierarchy: regions / cities (or clinics), couple and then individual. That way, you could simplify the response (dependent variable) to three categories: girl, boy and no-preference. Introducing another hierarchical level might complicate model fitting, though.

[Cassie Liu]

Dear Jseph，

Many thanks for your time and generous help.Thank you for letting me know when I use “qrmelogit“ ，its ’ DV must be bivariate.There must be something puzzled you because of my poor illustration of English.Here,I'd like to express something about the data I used to make you understand well about it.

1.My dataset came from a designed survey.In this survey,our respondents are migrants in china from 11 provinces,they migrated form small village to big city because of the development of urbanization,many of them make living by work in city as factory worker.We collected their information from comprehensive aspects containing their personal foundamental information(age、eduction...)、their family's situation,their "marriage and family"、"their public service"and so on.

2.Besides the 5219 respondents' individual-level information,we also collect the province and city macro data such as perGDP ...and have already matched the individual-level data with the macro(province&city)-level data.I was wodering if my individual nested in the city-level and provice level?

3.Our questionnaire is for individual,not for couple,everyone has its' own choice about children preference.
Thank you again for your serious and perfect answer,while you know more about my reserch,I am looking forward to gaining your precious advices.I will take it seriously to cosider.Reply me any time if you are convinient.
Sincerely,Cassie

[Cassie Liu]

Sorry for the wrong spelling of your name,I always do things carelessly.

[Joseph Coveney]

1.My dataset came from a designed survey.In this survey,our respondents are migrants in china from 11 provinces,they migrated form small village to big city because of the development of urbanization,many of them make living by work in city as factory worker.We collected their information from comprehensive aspects containing their personal foundamental information(age、eduction...)、their family's situation,their "marriage and family"、"their public service"and so on.

2.Besides the 5219 respondents' individual-level information,we also collect the province and city macro data such as perGDP ...and have already matched the individual-level data with the macro(province&city)-level data.I was wodering if my individual nested in the city-level and provice level?

It's not clear from your description, but if the data come from a complex survey design, then you'll need to look into Stata's estimation commands for survey data.

help survey

for more information.

[Cassie Liu]

"Joseph Coveney”，many thanks for your answer which have already helped a lot.