StATS: Do I have enough data after 24 months of time? (created 2005-04-05)

Someone asked me about a correlation coefficient that he computed on a data set that represented 24 months of data collection. A particular correlation of interest (a correlation between staff turnover and resident falls) was not significantly different from zero, but this person wanted to know how much more data to collect before safely concluding that no relation has been or likely will be established.

First compute a confidence interval for the correlation coefficient. If that interval is so narrow that you can rule out the possibility of a clinically important shift, then your sample size is large enough. How large a correlation is clinically significant? That's very hard to say. The correlation is a unitless quantity, and usually you need some measure in physical units (meters, kilograms, etc.) before you can talk about clinical importance.

You might want to look instead at the regression coefficient which does have units of measure in it. I assume that turnover is your independent variable and falls is your dependent variable. Think, then, about how much of an increase in falls per unit change in turnover is important from a clinical perspective. If that value is (I'm just making up a number) 0.5, then your sample size is adequate as long as the confidence interval for the slope is entirely inside plus/minus 0.5.

Please realize that an outsider like me can't tell you what's clinically important, because that requires clinical judgment, something I lack. A good general overview about clinical importance is on my web pages at

--> Stats: Confidence intervals

If this is an ongoing project, perhaps you might also find some value to using a control chart. A control chart allows for continuous monitoring of important processes. Who knows, maybe something that is not apparent now will become apparent because of some of the recent changes in health care? I have a brief outline of control charts at

--> Stats: Guidelines for quality control models

Another issue is that it is dangerous to look at 12 months worth of data, then 13, then 14, etc. because you are testing multiple times on a single hypothesis. It's sort of like being dealt three poker hands and choosing which one you like best. It would be better to select a sample size (time interval) prior to data collection and then test only once. If you do test multiple times, you need to adjust your alpha level. See

--> Stats: Interim analysis and

--> Stats: Early stopping in an animal study (July 1, 2004)

This page was written by Steve Simon while working at Children's Mercy Hospital. Although I do not hold the copyright for this material, I am reproducing it here as a service, as it is no longer available on the Children's Mercy Hospital website. Need more information? I have a page with general help resources. You can also browse for pages similar to this one at Category: Clinical importance or Category: Early stopping.