GLM Univariate

[Mike]

As Paul Swank has already pointed out, the Scheffe test can be

applied to pairwise comparisons and combinations of means.

Kirk's Experimental Design text covers this as does Howell's

"Stat Methods for Psychology" (7th ed -  see pages 394-395)

as well as other sources on experimental design and multiple

comparisons.  If you use "a priori" or planned comparisons,

you can also make combinations of means (in Howell 7th ed,

see p371-377 and note his presentation on orthogonal versus

nonorthogonal contrasts).

 

In general, SPSS provides the option to use the Scheffe test

in its ANOVA procedures (oneway, UNIANOVA, GLM) but

because it has the lowest level of statistical power, it is usually

used in situations where one has to make a large number of

comparisons and Type I errors are much more costly than

Type II errors. 

 

-Mike Palij

New York University

[hidden email]

 

----- Original Message -----

From: [hidden email]

To: [hidden email]

Sent: Friday, March 05, 2010 9:23 AM

Subject: Re: GLM Univariate

Marta wrote:

>>>Post hoc tests focus on pairwise differences between means, while the
overall F tests whether there is any difference among any pair of
combination of means. For instance, the F could be significant because
mean1+mean2 combined differ significantly from mean3+mean4 combined. No
post hoc test would reveal that.

 

Is the posthoc you are referring to is true also to post hoc in one-way anova?

--- On Tue, 3/2/10, Marta García-Granero <[hidden email]> wrote:

From: Marta García-Granero <[hidden email]>
Subject: Re: GLM Univariate
To: [hidden email]
Date: Tuesday, 2 March, 2010, 5:21 PM

Eins Bernardo wrote:

> Hi Marta, et al.
>
> The problem was already solved.  Thank you, Marta, for the solution.
>
> I would like to raise the issue based on what I have discovered a few
> days back.
> What is the statistical explaination of the instances wherein
> the ANOVA is highly significant but none of the posthoc test is
> significant?
>

Post hoc tests focus on pairwise differences between means, while the
overall F tests whether there is any difference among any pair of
combination of means. For instance, the F could be significant because
mean1+mean2 combined differ significantly from mean3+mean4 combined. No
post hoc test would reveal that.

HTH,
Marta GG

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

---

Try the new FASTER Yahoo! Mail.. Experience it today!

[E. Bernardo]

You answered already my question. Thank you, Marta.

--- On Fri, 3/5/10, Marta García-Granero <[hidden email]> wrote:

From: Marta García-Granero <[hidden email]>
Subject: Re: GLM Univariate
To: [hidden email]
Date: Friday, 5 March, 2010, 2:43 PM

Eins Bernardo wrote:

> Marta wrote:
> >>>Post hoc tests focus on pairwise differences between means, while the
> overall F tests whether there is any difference among any pair of
> combination of means. For instance, the F could be significant because
> mean1+mean2 combined differ significantly from mean3+mean4 combined. No
> post hoc test would reveal that.
>
> Is the posthoc you are referring to is true also to post hoc in
> one-way anova?
>

Beg your pardon? I don't understand the question. Anyway, what I said
about "classical post-hoc tests" (Tukey ...) is true for any ANOVA
design (ONEWAY, nested, repeated measures...).

Marta

=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@... (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

---

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!