Stata: Data Analysis and Statistical Software

From: Emily Watson [mailto:emilywatson632@gmail.com] 
Sent: 21 November 2012 21:58
To: Seed, Paul
Cc: statalist@hsphsun2.harvard.edu
Subject: Re: Interpretation of the Bland-Altman method on Stata

Dear Paul,

Thanks for your reply - I have now joined the Stata list

I am using Stata/IC 12.1, these are the codes I used and the Stata output:

. gen energy_log_r = log( energypd_r)

. gen energy_log_q = log( energypd_q)

. baplot energy_log_r energy_log_q

Bland-Altman comparison of energy_log_r and energy_log_q
Limits of agreement (Reference Range for difference): -0.939 to  0.099
Mean difference: -0.420 (CI -0.462 to -0.379) 
Range :  7.108 to  8.911
Pitman's Test of difference in variance: r = -0.212, n = 153, p = 0.012

. baplot energy_log_q energy_log_r

Bland-Altman comparison of energy_log_q and energy_log_r
Limits of agreement (Reference Range for difference): -0.099 to  0.939
Mean difference:  0.420 (CI  0.379 to  0.462) 
Range :  7.108 to  8.911
Pitman's Test of difference in variance: r =  0.212, n = 153, p = 0.009

My main question is, in which order should I put the variables? The record is the reference method (energypd_r) and I know from looking at the data that the energypd_r is lower than the energypd_q, so I would like to say that the questionnaire overestimates energy by X amount compared to the record but I'm not sure which output to use.
When you say it is not designed to use logged data, does that mean it will not work?

Kind regards
Emily

On Thu, Nov 22, 2012 at 5:24 AM, Seed, Paul <paul.seed@kcl.ac.uk> wrote:
Dear Emily,

(I received this privately, but am posting my answer to Statalist
to be shared.)

You might post such questions on Statalist
See http://www.stata.com/support/faqs/resources/statalist-faq/
for information on how to joins and how to write posts.

In any case, please the read the FAQ in detail, in particular
the advice about how to ask a question.

A difference of 34 on the logged scale is massive. Astronomical.
Larger than the US national debt.
It corresponds to a ratio of 10^34 (if you are working
with logs base 10) or exp(34) ), which is roughly 5.8*10^14
(if you are using natural logs).
Either number is so large that I am sure you have an error
somewhere else.

My command -baplot- is not designed for handling logged values.
And I can't say what the answer means without seeing the question.

The general advice to show exactly what you typed and what happened
applies.

Best wishes,

Paul Seed


From: Emily Watson [mailto:emilywatson632@gmail.com]
Sent: 21 November 2012 03:16
To: Seed, Paul
Subject: Interpretation of the Bland-Altman method on Stata

Hi Paul,

I came across your email address when searching for help on Stata - I hope you don't mind me emailing you.

I am using baplot to compare two methods of measuring dietary intake. I have entered the code -baplot record questionnaire- (record and questionnaire being my two methods(or variables)), both of which are the log versions.

I am of the understanding when log versions are used, once the output is back transformed the output will be a ratio which when multiplied by 100 is then converted to a percentage.

If the mean difference is then 34 (after the above adjustments) does this mean the questionnaire is 34% higher than the record? or the record is 34% higher than the questionnaire? or the questionnaire is 34% of the record?

Thank-you for your time and help

Kind regards
Emily



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