choice-metrics.com

Hi,

in my latest post (see topic "effects coding") I already mentioned that I would like to generate an experimental design that includes 5 different contexts that I would like to treat as an attribute which is kept constant across alternatives within each choice set. I plan to generate 30 choice sets and block the design into 6 blocks with 5 choice sets for each respondent which means that each respondent sees a choice in each context only once or at most twice. I will show three alternatives plus a no-choice alternative per choice set, all attributes are generic.

I was wondering how I should interpret the S estimate in case of a scenario-design. If I get an S estimate of 250 or 300, does this mean this is the minimum required sample size for one single context/scenario or for all contexts/scenarios in the design?
I consulted the Jaeger/Rose (2008)-paper which makes me assume the S estimate refers to all contexts. Am I right? I just want to make sure.

Regards,
Julia

The S-estimate refers to one whole replication of the design, so if you have a design consisting of 30 choice tasks (6 blocks of 5 choice tasks), then an S-estimate of 300 means that in order to get a t-ratio of at least 1.96 one needsat least 300 respondents completing the entire design of 30 choice tasks, or in case of 6 blocks, requiring 6x300 = 1800 respondents completing 5 choice tasks. In other words, it will not be easy to estimate such a parameter with statistical significance.