Archived Comments for:
The effect of carbon dioxide on near-death experiences in out-of-hospital cardiac arrest survivors: a prospective observational study
Fascinating article. On a minor point, I notice that the article abstract states that "The logistic regression model explained 46% of the variation" in NDE's. However, unlike ordinary least squares regression, logistic regression does not produce a true explained variance R-squared figure (given that the goal of the maximum likelihood estimation process is not to maximise explained variance). A number of pseudo R-squared measures are available for logistic regression, but these measures are not usually correctly interpreted as indicating explained variance (although DeMaris 2002 found that the McKelvey-Zavoina pseudo R-squared does a pretty good job at estimating explained variance).
Other examples of pseudo R-squares are Nagelkerke and Cox & Snell's , both available in SPSS. The measure used to obtain the estimate of 46% explained variation also doesn't seem to be stated in the study, which would be useful, given that the R-squared estimate tends to vary quite a bit between the different pseudo R-squared measures. Could you clarify how the estimate of explained variance was calculated?
Critical Care
Logistic regression and "explained variance"
27 April 2010
Fascinating article. On a minor point, I notice that the article abstract states that "The logistic regression model explained 46% of the variation" in NDE's. However, unlike ordinary least squares regression, logistic regression does not produce a true explained variance R-squared figure (given that the goal of the maximum likelihood estimation process is not to maximise explained variance). A number of pseudo R-squared measures are available for logistic regression, but these measures are not usually correctly interpreted as indicating explained variance (although DeMaris 2002 found that the McKelvey-Zavoina pseudo R-squared does a pretty good job at estimating explained variance).
Other examples of pseudo R-squares are Nagelkerke and Cox & Snell's , both available in SPSS. The measure used to obtain the estimate of 46% explained variation also doesn't seem to be stated in the study, which would be useful, given that the R-squared estimate tends to vary quite a bit between the different pseudo R-squared measures. Could you clarify how the estimate of explained variance was calculated?
Competing interests
None.