Correlation Question

Attached are two graphs comparing death from sepsis (blood stream infection) and a treatment (bundle)

First, if you look at the aggregate data for all Facilities (17), you see a very nice negative correlation between bundle and mortality (correlation coefficient = -0.31).

If you plot the same data relationship broken down by individual facility, there is actually a REVERSE correlation for (CORREL = + 0.29)






How is that possible?

michael menen said:

Attached are two graphs comparing death from sepsis (blood stream infection) and a treatment (bundle)

First, if you look at the aggregate data for all Facilities (17), you see a very nice negative correlation between bundle and mortality (correlation coefficient = -0.31).

If you plot the same data relationship broken down by individual facility, there is actually a REVERSE correlation for (CORREL = + 0.29)

How is that possible?

Click to expand...

What were the original data? Thank you!

michael menen said:

Attached are two graphs comparing death from sepsis (blood stream infection) and a treatment (bundle)

First, if you look at the aggregate data for all Facilities (17), you see a very nice negative correlation between bundle and mortality (correlation coefficient = -0.31).

If you plot the same data relationship broken down by individual facility, there is actually a REVERSE correlation for (CORREL = + 0.29)

How is that possible?

Click to expand...

Regression lines and correlation coefficients don't necessarily have anything to do with 'reality'. As an example look at the four data sets (graphs) at
https://en.wikipedia.org/wiki/Linear_regression#Interpretation
which all have the same regression line.