choice-metrics.com

Hi There,

I am a bit confused as to how to interpret the S-estimate and could not find clear directions in the manual. Basically I ran my code and got an S-estimate (for the Bayesian Mean) of 7.10 (it is 4.2 fixed). I have 5 blocks so I assumed to get the sample size I multiply the S-estimate by the blocks, but this would be a maximum of 36, which does not make sense if this was the total sample size as it sounds much to small? Am I missing something?

Regards,
Melanie

The S-estimate assumes that all choice tasks are given to a single respondent. If you have blocks, you need to multiply with the number of blocks. An S-estimate of 36 is certainly possible, especially if the attributes are important (so their impact on utility of large) and if you have no dummy/effects coded attributes (which often require a larger sample size).

Note that the S-estimate is the sample size that is sufficient for estimating a parameter at a statistically significant level with 95% confidence. This does not mean that the parameter is very reliable, one would generally like p values much smaller than 0.05 or t-values much larger than 1.96. So you should treat it as a minimum bound. Further, this sample size estimate is based on the priors, so if your priors are not reliable, then your sample size estimate is also not reliable. Finally, if you want to include sociodemographics in model estimation you will often need larger sample sizes; with 36 respondents you only get 36 observations of such covariates.

Michiel