Concordance Correlation Coefficient


3/1/16

This is a slightly modified web version of a presentation I gave at work on the concordance correlation coefficient and a macro to calculate and make a graph. I include all slides below, some notes for some slides, as well as a link to the .pdf file of the slides here.


I would have some confidence in the electronic cuff, and I would recommend to start using the electronic cuff because it is measuring the same thing as the gold standard (and is cheaper). Note that this data here is simulated.

Assessing can be done at micro (record) or macro (tabulation) level. Note that intercept=0 and slope=1 means the 45 degree Y=X line that goes through the origin.

These are two errors: if small spread around the 45 degree line, but “just off”, reject H0 (but should fail to reject here), and if large scatter/spread around the 45 degree line, fail to reject H0 (but should reject here).



This expected squared difference is just some number, so we scale it.

This might need to be modified for surveys by incorporating weights.

The variables X and Y in this example have good Pearson correlation, but you wouldn’t want to replace X with Y. The points don’t lie on the 45 degree “Y=X” line. The macro also outputs a confidence interval for the concordance correlation coefficient.

I'm not saying don’t use the other methods, but use them in combination. Certainly don’t use Pearson correlation by itself. Also make sure to use “low tech” approaches like graphing. In survey work, this could be used to see if a variable X could be replaced via substitution using an administrative data version of X. For example, if a respondent did not report employment, substitute in their administrative employment. Also, there is an "Overall Concordance Correlation Coefficient" to compare agreement between more than 2 raters.




I hope you found the presentation informative.


If you enjoyed any of my content, please consider supporting it in a variety of ways: