Re: ANCOVA
- From: "Old Mac User" <chendrixstats@xxxxxxxx>
- Date: 30 May 2006 08:38:36 -0700
Bill:
Without knowing more, my answer could be off-target.
I'm assuming your have two sets of data somewhat like this:
Case: X2 = 40 Case X2 = 60
X1 = 3.1 Y = 9.6 X1 = 2.3 Y = 8.4
X1 = 6.4 Y = 11.3 X1 = 4.1 Y = 9.0
X1 = 5.6 Y = 10.1 X1 = 10.3 Y = 14.9
X1 = 7.9 Y = 13.0 X1 = 7.2 Y = 10.2
etc. etc.
when when plotted as Y vs. X1 would yield two lines...
one associated with Case: X2 = 40 and the other
one associated with Case: X2 = 60.
If this is so, then the most direct way through this
is to multiple regression. Use a model that is linear
in X1 and linear in X2 and also an "interaction"
term. The interaction measures the degree to which
the lines are not parallel. If they are parallel, then
the interaction is "zero" (or near-zero) and the
signal for this will be that the t-ratio associated
with the interaction will be small. If the lines are
truly not parallel, then the t-ratio for the interaction
term will be "large" (significant).
While "ANOVA" deals with what is and what is not
"significant", it is not focused on the slopes of lines
and/or modeling the data.
Am I on the right track with this?
If the data set is small (not hundreds or thousands
of rows of data) and if you are free to share the data
(and provide enough background so I can be sure
I'm doing the right thing) I'd be willing to help you
with this... no charge or obligation.
.
- References:
- ANCOVA
- From: bill
- ANCOVA
- Prev by Date: Re: Minimum number of cases for a correlation?
- Next by Date: Re: Multiple Linear Regression with Nominal, Ordinal and Interval Data
- Previous by thread: ANCOVA
- Next by thread: Re: ANCOVA
- Index(es):
Relevant Pages
|
