Timeline for Continuous sample size determination based on the control group
Current License: CC BY-SA 4.0
3 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 23, 2021 at 11:21 | comment | added | ReneBt | If the application is such low risk then why do you need a rigorous statistical design? Are there no rules of thumb used in the application area? The point is if you feed empirical results from your ongoing experiment back into a continuous sample size estimator you will end up propagating the errors collected in the experiment into your stopping decision, compounding the risks. This feedback loop is pointless and more likely to cause serious errors. If it is impossible to set parameters without empirical estimates, then do a small initial study then exclude that data from the next analysis. | |
| Aug 23, 2021 at 10:53 | comment | added | Ivan | Thank you! I am still trying to understand if I phrased the question correctly. Take, for instance, the formula you gave and assume a minimal effect size that is still of interest to detect. What remains for the calculation is the variance. One might just have a good estimate off the top of one’s head or have to do a calculation. In the latter case, representative historical data might be used. What I am asking in the question, would it be sound to use the control group to estimate this variance, as it comes from the right context and at the right time, i.e., very much representative? | |
| Aug 23, 2021 at 9:53 | history | answered | ReneBt | CC BY-SA 4.0 |