using nlcom to calculate combinations of parameter estimates from different models

[Julie Watson]

Dear Stata users,

I am estimating two mixed logit models:

(1)
mixlogit choice price1, group(idgroup) rand(var1 var2) id(idn)

(2)
mixlogit choice price2, group(idgroup) rand(var1 var2) id(idn)

The only difference between the two models is the non-random variable where price2=price1*happy

I should note that:

• the two models are estimated on the same dataset
• there are no missing values
• price2 is not a re-scaling of price1 (happy varies over idn)

Post-estimation, I want to use both _b[price1] and _b[price2] in nlcom calculations.
Can you please help me figure out how I go about that?

Many thanks in advance!

Best wishes,
Julie

[Hong Il Yoo]

Hi Julie, is -suest- the command you are looking for?

[Julie Watson]

Hi Hong, thanks for your reply. I have tried but suest does not accommodate mixlogit unfortunately.

[Hong Il Yoo]

I see, could you elaborate on the parameter that you are planning to derive using the two price coefficients?

[Julie Watson]

Sure. I’m looking to calculate a set of ratios so in this example I would like to be able to calculate e.g. _b[price1]/_b[price2]

[Hong Il Yoo]

Fitting the two model specifications that you have listed to the same estimation sample and then using the results to derive the ratio seems rather odd to me. I wonder whether it may more make sense to estimate -mixlogit choice price price3, group(idgroup) rand(var1 var2) id(idn)- where price3 = price*(happy - mean of happy across idn) instead. The price coefficient is the average agent's price coefficient and the price3 coefficient captures whether the price coefficient varies with happy.

[Julie Watson]

Thanks for the suggestion Hong (much appreciated) but I'm afraid it is the ratio (with confidence intervals) that I need