Calculating predicted likelihood of a baseline profile using mlogit result!!!

[Arren Zhou]

Hello guys,

I have 8 independent variables: friendly_CHN (from 1 to 4)、friendly_USA (from 1 to 4)、pol_interest、age、gender、 education、urban、income

and one dependent variable: CH_power_type(from 1 to 5)

I have conducted the regression analysis using

mlogit CH_power_type friendly_CHN friendly_USA pol_interest age gender education urban income

I assume a baseline profile: 40-year-old male (1), lives in an urban area (1), has 10 years of education, has the highest political interest (2), earns 15,000 per month, and views the United States as the least friendly (1). How to calculate predicted likelihood as the "friendly_CHN" (friendly attitude toward China) changes from the lowest (1, unfriendly) to the highest (4, very friendly)?

I have spent hour on this and I have no idea how to do it with STATA codes.

I really do not have to know the math behind it since it is a random task that I was asked to complete from my relentless boss, who knows that I have no knowledge about multinomial logistic regression...

Please give me the codes if possible. Any advice will be greatly appreciated!!!!

Thank you!

[Joseph Coveney]

I assume a baseline profile: 40-year-old male (1), lives in an urban area (1), has 10 years of education, has the highest political interest (2), earns 15,000 per month, and views the United States as the least friendly (1). How to calculate predicted likelihood as the "friendly_CHN" (friendly attitude toward China) changes from the lowest (1, unfriendly) to the highest (4, very friendly)?

I assume that by "predicted likelihood" you mean changes in the predicted proportion of each category of the outcome variable as one of the predictors varies and the others are held constant.

If that's correct, then maybe something similar to the following line (executed immediately after you've fitted the multinomial logistic regression model).

margins , at(urban = 1 education = 10 pol_interest = 2 income = 15000 freindly_USA = 1) ///
    over(frindly_CHN)

I've attached the do-file and log-file of a worked example of the technique (simplified to three predictors with an artificial dataset) if you're interested.

[Arren Zhou]

Thank you! Mr. Coveney.

You saved my life.

[Joseph Coveney]

Thank you!

You're welcome. And thanks for the closure.