st: RE: comparison of agreement plot for non-Normal data

["Nick Cox" <n.j.cox@durham.ac.uk>]

The Bland-Altman plot I take to be a plot of difference versus mean, which long predates either Martin Bland or Douglas Altman, not that they invented the term. 

I know no sense in which this assumes a normal distribution, or even that its efficient or proper use requires a normal distribution. I've often used such plots very happily with skewed distributions. 

-concord- from SJ supports various related operations. 

Nick 
n.j.cox@durham.ac.uk 

Pinto, Daniel

I am analyzing the results of a method comparison study assessing agreement between two methods of capturing health service use and costs, N=50. Due to the small sample size and the analysis of cost data my distribution is non-Normal. I believe that Bland-Altman plot is the best statistic to use except that it assumes a Normal distribution. To address non-Normality Bland and Altman recommend using a double-regression half-Normal distribution method in the following paper: Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res. 1999 Jun;8(2):135-60.

I have attempted the performance of this method in STATA 10, plotting the residuals of the difference in GP count against the mean of the GP count.

regress gpcntdif meangpcnt

predict gpcntdifresid, resid

regress gpcntdifresid meangpcnt

I have tried to produce the plot including 95% CI using: twoway lfitci gpcntdifresid meangpcnt, stdf || scatter gpcntdifresid meangpcnt

This does not produce the same 95% CI as the equation from the double regression (recommended in the paper mentioned above): 
b0 + b1A ± 1.96 * residual SD from the regression. 
 
Although I can calculate the 95% CI using the equation I am unable to apply the 95% CI lines to the B-A plot. Is there anyway how to do this?


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/