0
$\begingroup$

I am having a report written on very complex EHR data where the data needed comes from different formats, from different locations. I need to choose a sample from about 20,000 accounts to manually review to show how accurate the report is.

What is the minimum sample size I would need? Do I calculate this statistically, or do I use a general business rule of n=30? And how can I say how accurate the report is based off of my sample size and the results?

$\endgroup$
0
$\begingroup$

It depends on what model you want to use for a simple $t$-test $n=30$ would be enough although you still get higher accuracy by increasing the sample size.

If you are using standard $t$-test you can start by defining the maximum margin of error you are willing to accept at given confidence interval and then calculate necessary number of observations to achieve that using $$n=(t^*SE/m)^2$$ where $t^*$ is t critical value given by the confidence interval, SE is the standard error and m the margin of error.

For more complex parametric models you usually need 30 observations per independent regressors, but some models may require more. Without knowing what model you use and more specifics it’s not possible to give more precise answer.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.