I compared the heatmap correlation coefficient with one that I got from a correlation after filtering the same values in R. Seems to be pretty much the same. However it is not possible to tell exactly, as the user does not know which correlation method is used. (Pearson, I assume?). Equally important is the information of how the algorithm handles missing values:
If in one row we have one parameter with a missing value, we cannot correlate with this value. One solution is to only correlate all other values in this row, and leave out the missing one. Another approach is to leave out the entire row, if one parameter has a missing value. Both methods have their justification, but we need an information box maybe, to show the user a bit of information about the algorithm.
I compared the heatmap correlation coefficient with one that I got from a correlation after filtering the same values in R. Seems to be pretty much the same. However it is not possible to tell exactly, as the user does not know which correlation method is used. (Pearson, I assume?). Equally important is the information of how the algorithm handles missing values:
If in one row we have one parameter with a missing value, we cannot correlate with this value. One solution is to only correlate all other values in this row, and leave out the missing one. Another approach is to leave out the entire row, if one parameter has a missing value. Both methods have their justification, but we need an information box maybe, to show the user a bit of information about the algorithm.