[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: per standard deviation change

From   "Martin Weiss" <[email protected]>
To   <[email protected]>
Subject   RE: st: per standard deviation change
Date   Thu, 11 Sep 2008 15:28:25 +0200

That is a very good point by Ronan! I overlooked in my initial response that
you had a dummy as explanatory variable where the -beta- option to -regress-
does not make sense. Note that the calculation you envisage toward the end
of your post is easily accomplished via -egen, std-. 


-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Ronan Conroy
Sent: Thursday, September 11, 2008 2:08 PM
To: [email protected]
Subject: Re: st: per standard deviation change 

On 11 Sep 2008, at 12:35, Mohammed El Faramawi wrote:

> I want to ask about modeling a continuous variable as per standard  
> deviation change. For example if I have a continuous variable such  
> as arterial blood pressure(BP) measured in mmhg and I want to  
> express the relationship between per standard deviation change of  
> blood pressure and the physical activity (coded yes, no). How can I  
> do that? Should I calculate the mean of BP abd then subtract it from  
> the individual observation then divide the product by standard  
> deviation.

You are making life too complicated! Here's an example with mean  
arterial pressure, measured in women, late in pregnancy:

. regress map smoking if visit==5

       Source |       SS       df       MS              Number of obs  
=     256
-------------+------------------------------           F(  1,   254)  
=    8.35
        Model |  788.039627     1  788.039627           Prob > F       
=  0.0042
     Residual |  23975.3197   254  94.3910226           R-squared      
=  0.0318
-------------+------------------------------           Adj R-squared  
=  0.0280
        Total |  24763.3594   255  97.1112132           Root MSE       
=  9.7155

          map |      Coef.   Std. Err.      t    P>|t|     [95% Conf.  
      smoking |  -3.562013   1.232783    -2.89   0.004    -5.989791    
        _cons |   93.37333   .7932676   117.71   0.000     91.81111     

You can see that smoking reduces the mean arterial pressure by 3.6 mm/ 
Hg (the coefficient for smoking) with a confidence interval of 1.1 to  
6.0 (5.989791 rounded). This is an important reason why smokers'  
babies are born lighter than non-smokers' (they lose the equivalent to  
one week of gestational development).

Reporting the effect thus has the advantage that a clinician can  
understand it, because it is expressed in the units in which blood  
pressure is measured. And from your point of view, the advantage is  
that you can extract this information very easily from -regress-.

You can do likewise with your exercisers versus non-exercisers  

No-one understands standard deviations in clinical medicine.

> Or should I subtract the standard deviation from the individual  
> observation?
> Also, Should I take the absolute variable when the new variable is  
> created or should I keep the signs when the regression is conducted?

You will notice how I reported the effect size and its confidence  
interval with the minus signs removed. As a general rule, no-one  
understands negative numbers and no-one understands risk ratios or  
odds ratios that are less than one.

Ronan Conroy

[email protected]
Royal College of Surgeons in Ireland
Epidemiology Department,
Beaux Lane House, Dublin 2, Ireland
+353 (0)1 402 2431
+353 (0)87 799 97 95
+353 (0)1 402 2764 (Fax - remember them?)

P    Before printing, think about the environment

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index