post-estimation scenario analysis
Posted: 20 Sep 2023, 11:36
Dear Stephane & David,
I have some questions on producing hypothetical scenarios post-estimation as an extension of predictions.
I have looked through the manual, examples and forum - the only thing I can find was a previous similar question on the forum which was never answered (here: viewtopic.php?p=2150&hilit=scenario#p2150). Apologies if I have missed anything simple in advance.
Background: I have an unlabelled experiment with 2 choices: Drug A and Drug B. Each have attributes (including WAIT TIME, numerical variable) and there are also important scenario attributes (e.g. QoL (categorical) and Life Expectancy (numerical)) which are added as interactions (because they're constant across alternatives). I have used Apollo Predictions to calculate impact on choice share (A/B) on increasing Wait time with no issue.
I would now like to create some hypothetical (policy-relevant) scenarios and look at the impact on choice share of changing certain attributes - I guess similar to a sub-group analysis (but instead of using actual data, use the model to predict pre- and post- when different attributes are changed. As an example, I might like to have in Scenario A: individuals have "X" QoL and "X" Life Expectancy and the Drug A attributes also have certain levels, I would then like to show what would be the impact of increasing wait time versus when the drug attributes have different level etc.
The purpose of doing this would be to create "policy-relevant" scenarios that would be easier for people to interpret. It might also serve as a validity check by looking at the model base prediction to see if it is as we would expect when certain attributes are fixed.
Is there a way to use specify multiple attributes in the predictions_base and then change multiple attribute in predictions_new? For example, when changing the database to show an increase in DrugA_WAIT of 1 year, could I also specify that QoL = 1 and Life = 2:
E.g. what I currently have:
database$DrugA_WAIT=1+database$DrugA_WAIT
I would then add something like database$QOL==1 to select only the cases where QOL =1 etc.?
Thanks in advance for your advice,
Rob
I have some questions on producing hypothetical scenarios post-estimation as an extension of predictions.
I have looked through the manual, examples and forum - the only thing I can find was a previous similar question on the forum which was never answered (here: viewtopic.php?p=2150&hilit=scenario#p2150). Apologies if I have missed anything simple in advance.
Background: I have an unlabelled experiment with 2 choices: Drug A and Drug B. Each have attributes (including WAIT TIME, numerical variable) and there are also important scenario attributes (e.g. QoL (categorical) and Life Expectancy (numerical)) which are added as interactions (because they're constant across alternatives). I have used Apollo Predictions to calculate impact on choice share (A/B) on increasing Wait time with no issue.
I would now like to create some hypothetical (policy-relevant) scenarios and look at the impact on choice share of changing certain attributes - I guess similar to a sub-group analysis (but instead of using actual data, use the model to predict pre- and post- when different attributes are changed. As an example, I might like to have in Scenario A: individuals have "X" QoL and "X" Life Expectancy and the Drug A attributes also have certain levels, I would then like to show what would be the impact of increasing wait time versus when the drug attributes have different level etc.
The purpose of doing this would be to create "policy-relevant" scenarios that would be easier for people to interpret. It might also serve as a validity check by looking at the model base prediction to see if it is as we would expect when certain attributes are fixed.
Is there a way to use specify multiple attributes in the predictions_base and then change multiple attribute in predictions_new? For example, when changing the database to show an increase in DrugA_WAIT of 1 year, could I also specify that QoL = 1 and Life = 2:
E.g. what I currently have:
database$DrugA_WAIT=1+database$DrugA_WAIT
I would then add something like database$QOL==1 to select only the cases where QOL =1 etc.?
Thanks in advance for your advice,
Rob