Dear all,
I am considering a study to estimate the partial utility of each attribute by asking respondents to directly answer the utility for health states that combine multiple attributes with multiple levels.
In such a study, can Ngene be used to create an efficient set of health states to be evaluated by each respondent?
Also, would it be possible to estimate the sample size given the expected value of the partial utility of each attribute?
Any help would be greatly appreciated.
Many thanks in advance.
Experiments that directly evaluate utilities
Moderators: Andrew Collins, Michiel Bliemer, johnr
-
Michiel Bliemer
- Posts: 2055
- Joined: Tue Mar 31, 2009 4:13 pm
Re: Experiments that directly evaluate utilities
I am unclear how a respondent could "directly answer the utility for health states". Utility is a latent construct, it is not observable. You can ask respondents to make choices, but you cannot ask respondents to provide utility.
You could ask a respondent to rate a health state, e.g. on a scale from 1 to 10, but then you would not be estimating a choice model, instead you would estimate a linear regression model. Rating alternatives is generally not considered acceptable in the scientific community as it is not a natural decision process, so publishing research based on ratings will be difficult.
If you want to do ratings, you could generate an orthogonal design in Ngene, which can be used for both choice models as well as linear regression models, although I do not recommend using ratings to estimate a linear regression model.
Ngene assumes that a choice model is estimated, not a linear regression model, so any sample size estimate that Ngene produces is only valid for choice models (assuming sufficiently reliable priors).
Michiel
You could ask a respondent to rate a health state, e.g. on a scale from 1 to 10, but then you would not be estimating a choice model, instead you would estimate a linear regression model. Rating alternatives is generally not considered acceptable in the scientific community as it is not a natural decision process, so publishing research based on ratings will be difficult.
If you want to do ratings, you could generate an orthogonal design in Ngene, which can be used for both choice models as well as linear regression models, although I do not recommend using ratings to estimate a linear regression model.
Ngene assumes that a choice model is estimated, not a linear regression model, so any sample size estimate that Ngene produces is only valid for choice models (assuming sufficiently reliable priors).
Michiel
Re: Experiments that directly evaluate utilities
Thank you so much for your insightful and informative response.
I will consider research methods based on your advice.
I will consider research methods based on your advice.