|
Size: 1476
Comment:
|
Size: 1476
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 5: | Line 5: |
| Bland and Altman (1995) suggested checking agreement between a pair of measures by plotting their difference on the y-axis against their sum on the x-axis. They also suggest working out the mean inter-test difference and the confidence interval for this difference (equal to mean difference +/- 1.96 SD of the differences) and adding these to the scatterplot. These statistics [:FAQ/balSPSS: can be worked out in SPSS] although SPSS will not add the lines representing the limits of the confidence interval or the mean for the inter-test difference to the scatter plot. Further details with illustrations are given [http://www.medcalc.org/manual/blandaltman.php here] abd [attachment:balt.pdf here.] If the scatterplot is random, the mean difference is around zero and the inter-test differences within +/-1.96 of the mean then the tests may be used interchangeably. | Bland and Altman (1995) suggested checking agreement between a pair of measures by plotting their difference on the y-axis against their sum on the x-axis. They also suggest working out the mean inter-test difference and the confidence interval for this difference (equal to mean difference +/- 1.96 SD of the differences) and adding these to the scatterplot. These statistics [:FAQ/balSPSS: can be worked out in SPSS] although SPSS will not add the lines representing the limits of the confidence interval or the mean for the inter-test difference to the scatter plot. Further details with illustrations are given [http://www.medcalc.org/manual/blandaltman.php here] and [attachment:balt.pdf here.] If the scatterplot is random, the mean difference is around zero and the inter-test differences within +/-1.96 of the mean then the tests may be used interchangeably. |
Plotting agreement between two measures using the Bland-Altman plot
Two tests which purport to measure the same underlying variable may be highly correlated (form a straight line when plotted against each other) but not agree. Agreement would be indicated by both tests taking the same value with the line x=y representing the best fitting straight line in a scatterplot of the tests graphed against one another.
Bland and Altman (1995) suggested checking agreement between a pair of measures by plotting their difference on the y-axis against their sum on the x-axis. They also suggest working out the mean inter-test difference and the confidence interval for this difference (equal to mean difference +/- 1.96 SD of the differences) and adding these to the scatterplot. These statistics [:FAQ/balSPSS: can be worked out in SPSS] although SPSS will not add the lines representing the limits of the confidence interval or the mean for the inter-test difference to the scatter plot. Further details with illustrations are given [http://www.medcalc.org/manual/blandaltman.php here] and [attachment:balt.pdf here.] If the scatterplot is random, the mean difference is around zero and the inter-test differences within +/-1.96 of the mean then the tests may be used interchangeably.
Reference
Bland JM, Altman DG (1995). Comparing methods of measurement: why plotting difference against standard method is misleading. The Lancet 346 1085-1087.