Mixlogitwtp:lognormal specification of price coefficient

[Julia Schreiner]

Dear Statalisters,

I am working with the mixlogitwtp command by A.R. Hole to estimate a model in WTP space. We designed an experiment with two price ranges for a product to elicit the effect of high and low prices. According to the examples from the help file the price coefficient is assumed to have a log-normally distributed coefficient. I created a new variable "mprice" by multiplying prices from both ranges by minus one. When I execute the command, I receive an error message for the lower price range: The price variable should be multiplied by -1 so that it has a positive coefficient in the conditional logit model. r(498).

mixlogitwtp Wahl if RangeA ==1, group(Fall_neu) id(ID) price(mprice) rand(SiegelWeide SiegelCO2 SiegelRegio) nrep(500) corr
estimates table, star(.1 .05 .001)

nlcom (Mean_price: -exp([Mean]mprice + 0.5*[SD]mprice^2))
nlcom (SD_price: exp([Mean]mprice + 0.5*[SD]mprice^2) * sqrt(exp([SD]mprice^2)-1))

In my opinion it might be that the price coefficient of the model with the lower price range is not clearly negative (or positive after the change of sign). Would it be possible to assume a log normal distribution in one model and a normal distribution in the model with the lower price range? I am afraid that it won't be possible to compare the WTPs for the attributes between them.
Any ideas how to solve the problem?
Thank you!
Julia

[Jens Petersen]

Are you sure that you used the given commands?

By using the "corr" option there is no sd stored in the estimation results.

[Jimmy Floyd]

Once you run the mixlogitwtp command and get the estimates, are you supposed to also transform back the coefficients of the other variables apart from price which were not restricted to a lognormal distribution?
e.g using:
nlcom (Mean_price: -exp([Mean]mprice + 0.5*[SD]mprice^2)) nlcom (SD_price: exp([Mean]mprice + 0.5*[SD]mprice^2) * sqrt(exp([SD]mprice^2)-1))

[ABD HILO]

Julia Schreiner did you figure it out? I have similar issue.