I'm working on a Bland-Altman plot. I would like to ask if the difference between variables should be normally distributed or not? Is this an assumption of Bland Altman plot?
Whether it will follow normal distribution or not depends on the mean obtained by the method used in measuring some parameter. To follow normal distribution with mean you will need variance too. The concept of Bland-Altman is just plotting the data points between two different assays. The data may be or may not be normally distributed depending upon your system under study.
But if the difference is not normally distributed standard deviation of distribution will lose its meaning. Standard deviation is one of the basis of the plot with mean of difference. It does not make sense. Mean of the difference also will not be a good measure when the difference is not normally distributed.
OK, I think the debate is over. Douglas and Altman said normal distribution of difference is essential in their article:
http://www-users.york.ac.uk/~mb55/meas//ba.pdf
But somewhere I have seen different ideas. Maybe I can wait for alternative responses.
The importance of differences in a Bland Altman plot is that they indicate if a systematic error happens and their closure to line 0 shows strength of the difference between two methods. In case of a systematic error, one finds an identifiable shape of differences.
There is no need to check if the differences distribute normally for the obtained general impression from the plot. You can check normality in order to make sure about the precision of estimated limits of agreement as mentioned by George and apparent in the article.
In Bland Altman plotting the variables or difference between variables need not be normally distributed.
- Badges
- Science topic
- Similar topics
- Correction