It's a property of the deviation contrast that the constant will be
equal to the unweighted mean of the category means (in a oneway
model). Other programs such as SPSS will give the same values.
Best,
John Hendrickx
--- Kim Lyngby Mikkelsen <[email protected]> wrote:
> Dear Statalisters--
>
>
>
> The constant term reported by �desmat- using the defcon(dev#))
> option
> seems to be wrong.
>
>
>
> In the example given below, the constant term reported by -desmat-
> is an
> un-weighted mean of the within category means, where I would have
> expected the overall mean to be reported!
>
>
>
> Is this a bug or the intension?
>
>
>
>
>
> I am using Stata 8 for Windows updated 15 Dec 2003, and my -desmat-
> is
> up to date:
>
> . which desmat
>
> c:\ado\stbplus\d\desmat.ado
>
> *! version 3.1, 29Oct2002, HYPERLINK
> "mailto:[email protected]"[email protected]
>
>
>
>
>
> Example:
>
> In the following, somewhat silly, example I would like to see if
> one
> category of �education years� (soev1: 3 categories (<8, <11, or
> more))
> have a mean height which differs from the overall mean height:
>
>
>
> The mean height of the sample population is:
>
> . su height if soev1!=.
>
>
>
> Variable | Obs Mean Std. Dev.
>
> ---------+-------------------------------
>
> height | 15089 167.933 9.044364
>
>
>
>
>
> And the height by category of educational years is:
>
> . table soev1, c(mean height freq) f(%9.3f) row
>
>
>
> ----------------------------------------
>
> education |
>
> , years | mean(height) Freq.
>
> ----------+-----------------------------
>
> 1 | 165.955 6,687
>
> 2 | 168.886 6,772
>
> 3 | 172.089 1,630
>
> |
>
> Total | 167.933 15,089
>
> ----------------------------------------
>
> (Higher people have more education years (or vice versa)).
>
>
>
>
>
>
>
> From this table the �wrong� mean (of the within category means) can
> be
> calculated as:
>
> . di (165.955 + 168.886 + 172.089)/3
>
> 168.97667
>
>
>
> This �wrong� mean of the within category mean, is the mean which is
> calculated by �desmat-:
>
>
>
>
>
>
>
>
>
>
>
> . desmat: reg hojharm soev1, defcon(dev(3))
>
> ------------------------------------------------
>
> regress
>
> ------------------------------------------------
>
> Dependent variable height
>
> Number of observations: 15089
>
> F statistic: 388.653
>
> Model degrees of freedom: 2
>
> Residual degrees of freedom: 15086
>
> R-squared: 0.049
>
> Adjusted R-squared: 0.049
>
> Root MSE 8.821
>
> Prob: 0.000
>
> ------------------------------------------------
>
> nr Effect Coeff s.e.
>
> ------------------------------------------------
>
> soev1
>
> 1 1 -3.022** 0.108
>
> 2 2 -0.090 0.108
>
> 3 _cons 168.977** 0.089
>
> ------------------------------------------------
>
> * p < .05
>
> ** p < .01
>
>
>
>
>
> As can be seen, the constant term reported by �desmat- (_cons =
> 168.977)
> is the �wrong� un-weighted mean calculated above, and not the
> overall
> mean (= 167.933), which I would have expected.
>
>
>
>
>
>
>
> Regards
>
> Kim
>
> MD, Ph.D, Epidemiologist
>
> Institute of Occupational Health, Denmark
>
>
> ---
> Outgoing mail is certified Virus Free.
> Checked by AVG anti-virus system (http://www.grisoft.com).
> Version: 6.0.547 / Virus Database: 340 - Release Date: 02-12-2003
>
>
__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
