1

I just started learning about Bland-Altman plots. So my data consists of angles from a particular device and another set of angles from another device. The angles on both devices vary between -5 degrees to 20 degrees. So I did my bland-Altman plot and the diagram looks fine.

However, I am wondering whether I should be considering the absolute value of the angles instead of both negative and positive values.

So should I be using the absolute value for the calculations?

Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357
Ali Parahoo
  • 13
  • 2
  • Related: [Creating and interpreting Bland-Altman plot](http://stats.stackexchange.com/q/99835/1352), in particular [@Glen_b's observation](http://stats.stackexchange.com/questions/99835/creating-and-interpreting-bland-altman-plot#comment193487_99835) that this plot is also known as a *Tukey mean-difference plot*. – Stephan Kolassa Oct 23 '15 at 06:37

1 Answers1

1

On the one hand, it all depends on your application. If your devices' measurements are taken in absolute value after measurement for further use, anyway, then of course doing the Bland-Altman plots on absolute values makes sense. (Similarly, if you take logarithms of raw measurements and then work on logs, then you should do the plots on logged data.)

That said, I find it hard to envisage a situation where you can measure an angle in either a positive or a negative direction and don't care about the sign. When would a measurement of -5 degrees lead to the same subsequent decisions as a measurement of +5 degrees? So my first impulse, without knowing your specific problem, would be to do the plot on the raw data.

Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357
  • @Kolasso Thank you so much for your response. So I looked at my raw data and taking the absolute value does not make sense. That said, if I want to calculate my CV(coefficient of variability) and the reproducibility coefficient of the error (RPC value) as a percentage of the mean, the percentage values I get are sort of big (above 100%) I am guessing this is because the mean value is close to zero since i have positive and negaitive values. So should I ignore those percentages? – Ali Parahoo Oct 23 '15 at 15:12
  • Yes. Any percentage metric gets unstable and useless if zero is a "realistic" value that can occur as the denominator. (The same applies to the Mean Absolute Percentage Error.) Only use percentages if all values are "far enough" away from zero. Maybe you can simply use the numerators and compare these. – Stephan Kolassa Oct 23 '15 at 18:37