Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Quick question of using -concord- to investigate proportional errors

From   Nick Cox <>
Subject   Re: st: Quick question of using -concord- to investigate proportional errors
Date   Thu, 19 Jan 2012 10:34:52 +0000

You have been posting to the list since October. See my reply to you
last Saturday spelling out the longstanding and often repeated request
to explain where user-written programs come from.

In this case -concord- is a program last updated in SJ 10(4).
-concord- is a small toolkit for thinking quantitatively about the
structure of agreement or disagreement between two continuous
measures. Homing in on significance test results misses most of what
-concord- can tell you.

You only cite part of your results but they show only weak agreement
between your measures: the concordance correlation will be about .262.
The low P-value of 0.002 just says that the concordance correlation is
definitely not zero; that is a conventional test for those who need
it, but is unlikely to be an answer to the underlying scientific

Your results also show a correlation between difference and mean that
is a bit high.

So there is a lot of scatter in your data and the error structure is
only roughly multiplicative; there's some extra structure on top of

Most of what is written on -concord- is freely available in the STB
and SJ archives, so I won't repeat explanations already given.


On Thu, Jan 19, 2012 at 12:29 AM, cecilia sam <> wrote:

> I am new to statistics and Stata, and need help for using -concord- in
> Stata 11.0 for Windows.
> I am comparing two datasets derived from 2 different methods. I
> natural log transformed both datasets on the original scale, and typed
> -concord y x, loa(regline)-. The output provides the slope of the
> regression line, and a p-value listed besides the Pearson r (I
> attached a part of the Stata output and attached below). Is this the
> p-value that indicate the signficance of the slope of the regression
> line?  In this case, does the p-value <0.05 mean that there are
> statitically signficant proportional errors ?
> Pearson's r =  0.271  Pr(r = 0) = 0.002  C_b = rho_c/r =  0.966
> Reduced major axis:   Slope =     1.301   Intercept =    -2.727
> Difference = y - x
>         Difference                 95% Limits Of Agreement
>    Average     Std Dev.             (Bland & Altman, 1986)
> ---------------------------------------------------------------
>      0.010       0.333                 -0.643      0.663
> Correlation between difference and mean = 0.266

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index