On Apr 28, 10:36 pm, "Phil Holman" <piholmanc@yourservice> wrote:
I have 4 high school math classes. One pair is relatively high and one
pair relatively low achievers. One out of each pairing was randomly
selected for a treatment and the other 2 were controls. I taught a
chapter of math and conducted the usual end of chapter test.
I have mean scores for all 4 classes and also the scores for their end
of chapter tests for the whole year. Per the subject line, which would
be the best test to run to see if the treatment had an effect on their
test scores. This test was a little harder than normal so the mean
scores were lower.
Phil H
You are describing a 2x2 ANOVA (or ANCOVA) type of model. Of course,
it could be run equivalently as a regression model with two binary
explanatory variables (low v high achievement, treatment v control)
and their product. If you run it as a regression, you might want to
centre the outcome variable on its mean or some other convenient in-
range value. With either approach, the other variables you mention
could also be included as covariates.
Note that this approach treats class as fixed. If you want to treat
it as random, you have a couple options: a mixed (fixed and random)
ANOVA, or a multi-level model. Neither will work very well, IMO. The
mixed ANOVA will have the problem of very low df for the error term,
so not much power. Re the multilevel model approach, Snijders &
Bosker (1999, p. 44) say this:
"In order to choose between regarding the group-dependent intercepts
U[0j] as fixed statistical parameters and regarding them as random
variables, a rule of thumb that often works in educational and social
research is the following. This rule depends mainly on N, the number
of groups in the data. If N is small, say, N < 10, then use the
analysis of covariance approach: the problem with viewing the groups
as a sample from a population is in this case, that the data will
contain only scanty information about this population. If N is not
small, say N >= 10, while n[j] is small or intermediate, say n[j]
100, then use the random coefficient approach: 10 or more groups is
usually too large a number to be regarded as unique entities. If the
group sizes n[j] are large, say n[j] >= 100, then it does not matter
much which view we take. However, this rule of thumb should be taken
with a large grain of salt and serves only to give a first hunch, not
to determine the choice between fixed and random effects."
--
Bruce Weaver
bwea...@lakeheadu.cawww.angelfire.com/wv/bwhomedir
"When all else fails, RTFM."
