I was asked off-list to provide an example demonstrating how to
contrast a single group mean against the grand mean. You'll see below
that I provide various types of contrasts using the LMATRIX subcommand
via the GLM procedure.
It's probably worth noting that the first LMATRIX statement defines
the grand mean as the sum of the group means divided by the number of
groups.
The GLM procedure is certainly not the only procedure that is capable
of testing various types of contrasts. I typically use the MIXED
procedure for such contrasts for many reasons.
Ryan
..
data list list / ID Group Y.
begin data
1 1 24
2 1 22
3 1 21
4 1 23
5 2 67
6 2 66
7 2 67
8 2 65
9 3 43
10 3 44
11 3 45
12 3 42
13 4 12
14 4 11
15 4 10
16 4 13
end data.
GLM Y BY Group
/PRINT = PARAMETER
/DESIGN = Group
/LMATRIX = 'Grand vs Group 1' Group 3/4 -1/4 -1/4 -1/4
/LMATRIX = 'Group 1 vs Group 2,3,4' Group 1 -1/3 -1/3 -1/3
/LMATRIX = 'Group 1 vs Group 2' Group 1 -1 0 0
/LMATRIX = 'Group 1,2 vs Group 3,4' Group 1/2 1/2 -1/2 -1/2.
On Sat, Dec 4, 2010 at 8:19 AM, Art Kendall <Art@drkendall.org> wrote:
> YMMV but in my experience, when people say they want to "test one group
> against the total" they mean several things.
>
> <soapbox> It is important to tease out the precise question. <\soapbox>
>
> Sometimes they mean they want something like a standard score type idea,
> i.e., how extreme is this subpop in the whole pop of which it is a part.
> [mean vs grand mean] This can be problematical when a large portion of the
> cases are on both sides of the comparison. [IQ this class vs norms.]
> Sometimes they mean that they want to compare this subpop to all of the
> other members of the pop without regard to which subpop they in (1 residual
> subpop). [mean a vs mean of all residual cases (cases not a)] This can
> compare apples to all other fruit.
> Sometimes they mean that they want to compare this subpop to each of the
> other subpops. [mean a vs mean b, mean a vs mean c, etc, but not b vs c
> etc.]
> Sometimes they mean that they want to compare each pair of subpops.[All
> pairs of means]
> Sometimes when the subpops are ordered they want to compare ordered subpops
> to the lowest value for any subpop. (e.g., doses vs zero dose).
>
> aside: Does anyone know of a site that succinctly lays out in one place 1)
> statements like those above 2) specifies the names for kinds of contrasts
> 3) gives examples and 4) gives some visualizations?
>
> Art Kendall
> Social Research Consultants
>
> On 12/3/2010 10:21 PM, R B wrote:
>
> I didn't respond to this post initially because the question pertained
> to excel. But, since someone responded, I would like to point out that
> it is possible to compare the mean of one group against the grand
> mean. I recently showed how to do this in SPSS via the MIXED procedure
> employing the TEST subcommand. Really, there are several procedures in
> SPSS that could be used to carry out this type of contrast.
> Ryan
> On Fri, Dec 3, 2010 at 4:26 PM, Richard Ristow <wrristow@mindspring.com>
> wrote:
>
> At 04:45 AM 11/24/2010, Alexandra Chirilov wrote:
>
> Could you help me with an excel which perform test on means for
> subgroups vs. total?
>
> None of the standard tests compare subgroups with the total; they
> compare non-overlapping subgroups with each other.
> The general term for comparing subgroup means is 'analysis of
> variance' ('ANOVA'). There are many variations on ANOVA, but the
> place to start is with a simple one-way ANOVA, unless there's reason
> in your study to do otherwise. (A still better place to start is to
> get advice from someone who knows the statistical methods and the
> details of your study, who can tell what methods would be appropriate.)
> If you want to do this in Excel, the main advice you'll get here is,
> don't. Lots of people do statistical analysis in Excel, but the
> general professional opinion is that Excel isn't very good for that,
> and there are many risks in using it.
> If you have SPSS, and can load your data into SPSS, from the menus try
> Analyze > Compare Means > One-Way ANOVA
> You *must* have someone statistically knowledgeable help you to
> understand the results. You can also post SPSS and statistical
> questions here, though there's no substitute for someone you can talk
> with face to face.
> -Best of luck,
> Richard Ristow
> =====================
> To manage your subscription to SPSSX-L, send a message to
> LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
> ===================== To manage your subscription to SPSSX-L, send a message
> to LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command SIGNOFF SPSSX-L For a list of
> commands to manage subscriptions, send the command INFO REFCARD
=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD