questions about gologit2 command*and related models

[Donna Kim]

Hi,

I want to ask some questions about gologit2 command and partial proportional odds model.
Here's my questions below.

1. making output table in gamma parameterization

I've tried to export my results in gamma using outreg2 or estout in several ways, but all failed.
I read help gologit2 and learnt gamma(name) option, but don't know how to use the option correctly.
This is one of my code. Please let me know the problem and better codes you have.

gologit2 allow_poor i.gender conage sbstatus fmincome edu i.relig, autofit or gamma(m1)
est store m1, title(Model 1)
est out m1 using "[my filename]", cells(b(star fmt(3)) se(par fmt(2))) stats(aic bic N) replace

2. Negative probability

I got 1~10 cases of negative probability from more than 7000 observations. Is this acceptable number to use the gologit model?


3. Relationships between ordered regression models

I read few books and articles about ordered regression models and tried to understand relationships between the concepts.
I need help to figure out if there's something wrong with my rough understanding below.

1) parallel cumulative logit model = proportional odds model = ordered logit model(in narrow sense)
2) partial(unconstrained, constrained) cumulative logit model= partial proportional odds model
3) non-parallel cumulative logit model = generalized ordered logit model(in narrow sense)
4) 1)~3) ⊂ generalized ordered model(in broad sense) or ordered logit model(in broad sense)

Thank you for reading.
Please help me out!

[Richard Williams]

Donna, welcome to Statalist.

I don't have time to answer all your Qs now but will try to get to them later if nobody else does. Negative predicted probabilities, and what you might do about them, are discussed in the troubleshooting FAQ:

https://www3.nd.edu/~rwilliam/gologit2/tsfaq.html

The article introducing gologit2 is available for free at

https://www.stata-journal.com/articl...article=st0097

[Richard Williams]

1. The gamma estimates were stored in m1 with your gologit2 command. When you followed with the estimates store command, you overwrote the gamma results with the regular gologit2 results. If you have esttab installed, do something like

webuse nhanes2f, clear
gologit2 health weight height i.black, auto gamma(m1) or
esttab m1, cells(b(star fmt(3)) se(par fmt(2))) stats(aic bic N) replace eform

2. Negative predicted values. See the troubleshooting FAQ:

https://www3.nd.edu/~rwilliam/gologit2/tsfaq.html

3. The paper I linked to earlier shows how various models are special cases of the generalized ordered logit model. If you only had gologit2, you could estimate the same models as

logit
probit
cloglog
ologit
oprobit

For each model, you would specify the link option (default is logit) and the pl option (to constrain all vars to meet parallel lines).