Interpreting repeated measures ANOVA results in Intervention Study

cait

New Member
#1
I'm having one of those late night working problems, where I'm suddenly second guessing all my work and I seem to have lost all my statistics know-how in the last 4 hours.

Here's my problemo: I did an intervention study. I have pre and post test measures, and participants were split into 2 conditions - called "inclusive" (experimental) and "non-inclusive" (control). So I'm comparing participants scores, based on their condition, in their pre and post test measures. I ran a 2 (condition) x 2 (age group: older or younger) x 2 (time 1, time 2), ANOVA with time as a repeated measure, and the DV was scores pre and post test.

So now I'm trying to interpret my results. I have a significant effect of time, which is cool. Regardless of condition, my participants got better (I know I shouldn't be excited about that, but with the nature of this project, it's pretty a-okay). I have no interaction of time and condition. So it's not that the inclusive got better than the non-inclusive. However, I did get a significant effect of condition. And this is where I'm getting all twisted in my brain. What does this actually mean? I have been thinking that it means that the inclusive group got better than the non-inclusive, but wouldn't that be what the interaction was telling me? Does it just mean that my scores (pre and post) are different in the different conditions? Why didn't I pick up on that when I tested all the pretest measure for difference? Before I took any steps, I made sure to compare the 2 groups on this measure, and there were no differences.

So basically, I'm trying to figure out what the different between that time*condition interaction and just the condition effect is.

Help?
 

Karabiner

TS Contributor
#2
Well, perhaps some questions for clarification first.

What kind of intervention, and which population?
What did you actually measure as dependent variable?
How large is your sample size?

And, besides,
2 (age group: older or younger)
Was this necessary? If you have the partcipants' age,
then do not split your sample, use the original age
variable instead.

What does this actually mean?
Was this a randomized trial?
I have been thinking that it means that the inclusive group got better than the non-inclusive, but wouldn't that be what the interaction was telling me?
Disn't you already make a graphic display of your data?
E.g. boxplots of your groups pre and post. Or a bar
chart for the group means at t1 and t2, respectively
(including SDs).

Generally speaking, a condition effect without interaction
should indicate that the means in one condition are higher
than in the other, across time points. Therefore, the question
how subjects were allocated to groups.

With kind regards

K.
 

cait

New Member
#3
It was a classroom lesson plan intervention, so half of the classroom got assigned to the control and half to the experimental. The DV was their score on a measure of heterosexism. It was based on their creation of same-sex families out of deck of cards of people. They completed this measure before and after the lesson plan. I have 99 kids in the study.

In my field, because they're kids (5-10 years old), and huge developmental changes can happen based on age, it's always standard to create some sort of age grouping (i used a median split). The results don't change when age is taken out, so I've been advised to keep it out because I know it will be something my reviewers tell me to add.

We matched children somewhat to different groups. After the pretest, kids who knew what gay people are were split even into the control and experimental lesson plans, and then the rest of the children were randomly assigned.

In regards to the issue being if my groups were already different at time 1 - I believe I already checked for that, in the beginning when I ran analysis on all my pretest measures to make sure that there weren't differences among groups. In this case, their pretest score on this measure served as the DV in a 2 (gender, because this is also something we worried about causing differences in groups) x 2 (condition) anova, and this showed no difference between the control and experimental groups on this measure before the intervention. Even when I take gender out and just do condition x scores, the result is non-significant
 
#4
When it is related to kids, grouping should not be done on the basis of pretest as such, age grouping is may be possible. In preschools such as Huntington Preschool grouping is done so that it becomes easy for teachers and staff to manage kids and can look after them.
 
Last edited by a moderator: