Average Marginal Effects for Gologit2 Model

[Valerie Aulia]

Hallo,

I'm Valerie and currently on my last year of bachelor degree, thus I'm in the process for writing my bachelor theses. I wanna estimate using Gologit as my dependent var is in categorical (Keluhan Kesehatan/Health Complaints (Y) 1 (lowest) and 4 (highest or complaining about health, illness, interference with work, to outpatient care)). I choose Gologit as my model doesnt fit the parallel trend assumption for the usual ordered logit. I have an independent var of Car Ownership (binary, 1 refer to owned car, and 0 otherwise).

After calculating the gologit, how can I calculate the marginal effect of change in my dependent var Health Complaints? I've used this command:

**AMEs using margins
*outcome 1
margins r.CarOwnership, predict(outcome (#1))
*outcome 2
margins r.CarOwnership, predict(outcome (#2))
*outcome 3
margins r.CarOwnership, predict(outcome (#3))
*outcome 4
margins r.CarOwnership, predict(outcome (#4))

But my supervisor told me that I have to pay attention to the other control variables too, especially for those who are in a binary form(?)

My next appointment with my supervisor still next week but I'm really curious about how to calculate the AMEs, hehe. Someone explain to me, please?

Thank you,
Valerie

[Felix Bittmann]

Your syntax is fine, you could also use

margins, dydx(CarOwnership) predict(outcome(#1))

which is equivalent.

You did not show your main model. Just make sure to include all relevant controls there (you do not need to include them in margins, only if you want to show their effects as well). Including binary variables is simple, you can also use the i.prefix (not required for these). I am not sure why the binary variables should pose any special problem for this kind of analysis. The effect is just the change from 0 to 1. Say, the average marginal effect for women is 0.50, you can state the probability increases by 50 percentage points (compared to men).

[Valerie Aulia]

here's my main model

with X refers to the control variables factors.

Anyway, my dependen var (health complaints) categories are below:

• Category Y = 1 if the respondent has not experienced any health complaints during the past month.
• Category Y = 2 if the respondent has experienced health complaints during the past month, but they have not interfered with work, school, or daily activities.
• Category Y = 3 if the respondent has experienced health complaints during the past month that have interfered with work, school, or daily activities, but did not seek outpatient treatment.
• Category Y = 4 if the respondent has experienced health complaints that have interfered with work, school, or daily activities during the past month and sought outpatient treatment.

With the syntax above, the result came out like this: (I'm using Richard Williams notes)

Does my interpretation correct: "On average, Car's owners are 0.8 percentage point more than non-Car's owner to report that they had not experienced any health complaints during the past month"

I just remembered that my supervisor said I should use ME instead of AME, so which one actually better to use, ME or AME? If i use the syntax above, it will produce the AME, is it right?

But, if the result that I want is the usual ME, is this syntax correct?

margins r.CarOwnership, predict(outcome (#1)) atmeans

Using ME syntaxt, the result is:

Thus, could I interpret it with: "Marginal effect at the mean shows that, when all other variables are held at their average values, owning a car increases the probability of not reporting any health complaints (outcome #1) about 0.08 percentage points compared to non-Car's owner"

Regards,
Valerie

[Felix Bittmann]

This seems fine, regarding the first output:

On average, Car's owners are 0.8 percentage point more likely than non-Car's owner to report that they had not experienced any health complaints during the past month.

Regarding the second output, you shifted the decimal point one digit.

Marginal effect at the mean shows that, [when all other variables are held at their average values,] owning a car increases the probability of not reporting any health complaints (outcome #1) by about 0.8 percentage points compared to non-Car's owner"

I think you can omit the part in brackets since this is implied by MEMs.

[Valerie Aulia]

Alright, I'll try to continue my estimation first. Thank you very much for the explanation, it helps a lot!

Regards,
Valerie

[Richard Williams]

Unless you are using an ancient version of Stata you are making this too hard. You don’t need to specify each outcome separately. See

https://academicweb.nd.edu/~rwilliam.../Margins05.pdf

Several other handouts on margins are available at

https://academicweb.nd.edu/~rwilliam...rginal_effects

i am away on vacation now but I’ll try to check back in a day or two to see if you are getting along ok.