**Date:** Mon, 5 Sep 2005 22:21:45 -0400
**Reply-To:** "Frank J. Gallo" <fjgallo@verizon.net>
**Sender:** "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
**From:** "Frank J. Gallo" <fjgallo@verizon.net>
**Subject:** Re: ANOVA and Post-hoc Tests
**In-Reply-To:** <015e01c5b27d$38aa7b00$63076cc8@datatrend>
**Content-Type:** text/plain; charset="US-ASCII"
Hi Isacc,

Fisher's LSD test does not correct for multiple comparisons as the Tukey HSD
post test does. Tukey HSD sets the 5% significance level for the entire
family of comparisons. Generally, researchers do not recommend Fisher's LSD
test but Tukey's HSD. This is my understanding.

Thanks,

Frank

-----Original Message-----
From: Isaac Dialsingh [mailto:consult@tstt.net.tt]
Sent: Monday, September 05, 2005 8:51 PM
To: Frank J. Gallo
Subject: Re: ANOVA and Post-hoc Tests

Why didn't you use the LSD test?

Isaac

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

From: "Frank J. Gallo" <fjgallo@verizon.net>

Newsgroups: bit.listserv.spssx-l

To: <SPSSX-L@LISTSERV.UGA.EDU>

Sent: Monday, September 05, 2005 3:19 PM

Subject: ANOVA and Post-hoc Tests

> Hi All,

>

>

>

> Looking for some help on interpreting contradictions found in an ANOVA run

> with TukeyHSD follow-up. The F-test indicates significant differences

among

> groups.

>

>

>

> ANOVA

>

>

>

>

> Sum of Squares

>

> df

>

> Mean Square

>

> F

>

> Sig.

>

>

>

> Between Groups

> 2.849

>

> 3

>

> .950

>

> 2.830

>

> .041

>

>

>

> Within Groups

> 46.649

>

> 139

>

> .336

>

>

>

>

>

>

>

> Total

> 49.498

>

> 142

>

>

>

>

>

>

>

>

>

>

>

>

>

> Levene = 5.168, p=.002, the null hypothesis that the error variance of the

> dependent variable is equal across groups suggests no.

>

> But, post-hoc comparisons show no significant differences among groups.

>

> And, the homogenous subsets table identifies one set (4 groups, p=.143)

> indicating that the means of the subgroups are not significantly different

> from each other.

>

>

>

> I do have unequal sample sizes: 23, 42, 30, 48. The run used the harmonic

> mean (32.934), so I am aware that the Type 1 error level is not

guaranteed.

> Is this what I am seeing? That is, the p-value for the F-test is

> underestimated and it should be larger. This is my first experience with

> this kind of situation.

>

>

>

> Yours thoughts are greatly appreciated.

>

> Frank