ApolloChoiceModelling forum

Hi all,

I am trying to conduct a choice analysis on a set of 5 alternatives. The survey data that I am using only has price data available for the choice that the participant made, not the prices for all options (this seems like a common problem with survey data). I have a couple of questions please:

I assume that I should obtain data on each of the other options through imputation methods?
I assume that I include price or cost data for each choice into the model?

Thankyou for your help, and assistance or reference would be greatly appreciated.

Hi

this is quite a common issue in some areas.

You would indeed need to find the price data for the other alternatives, for example through imputation.

However, it is VERY important that you then impute the price for all alternatives, including the chosen one. It would be unwise to use the observed price for the chosen one alongside the imputed one for the unchosen ones

Stephane

Thanks for this.

So I am trying to estimate a model for choice of childcare. My 4 categories are "formal", "informal" (friend/family member), "Other" and "No Childcare" (self). I am hoping to use this model to forecast the demand for childcare.

As I mentioned, I only have price information on the choice that each person made, and as you said I will impute ALL prices for each person and each category based on the subset of price information that I have front he survey data.

However, I think that it is a reasonable assumption to make that "No Childcare" has $0 price, and that the price of "informal" care is also very small. Given this, I may well just be imputing the price for the "Other" and "Formal" categories.

I was also wondering on whether the fact that two of my categories has very low prices would effect the statistical significance of price within the model? - as there is not much variation in prices between categories. Does this thinking make any sense?

Thanks again for your help.

Hi

yes, you could make those assumptions.

In terms of your ability to estimate the effects, this really depends on the richness of the trade-offs that you will have in the data

Stephane

Thanks, one more question if I may.

I plan on using multiple imputation (MICE) (https://stefvanbuuren.name/RECAPworksho ... al_II.html) as my method, do you have any experience with this?

I assume that I should not use the same explanatory variables to impute the price variable as what would be used in my discrete choice model (income, age etc), as this would create multicollinearity between the explanatory variables in the discrete choice model?

Thanks again.

Hi

whether this causes problems is going to depend on other factors relating to your model specification. The key issue with imputation is whether you believe that for people with missing information, the link between the explanators and that variable is the same as for those without missing information. Have a look at https://doi.org/10.1016/j.jocm.2014.10.001

Stephane