My moderation model have a indepent variable: "before-after treatment", dependent variable: -health questionarie score, and moderation variable is personality score.
What problems there are with to use paired samples in moderation analysis ?
The research question : Is the personality of a moderating effect of whitening teeth in the oral health perceived score? but the problem is that the data has no control group. The data have before and after oral health perceived score.-
-------------------------------
My solution is paired-sample t-Tests, then calculate a new variable ( before oral health perceived score - after oral health perceived score) and correlate this new variable with sub-scales of NEO-PI.
The problem is that paired-sample t-Tests don't allow you to do multi-variate analysis, but moderation requires multivariate testing -- usually with regression.
In this case, it would be something like:
Y = a + B1X1 + B2X2
Y = a + B1X! + B2X2 + B3 X1*X2
Where Y is the post-test, X1 is the pre-test, X2 is the personality test, and X1*X2 is the product of the two independent variables. Note that the interaction term here is the product of two continuous variables, which is not the easiest thing to interpret, but the first step is to see if that term is significant (i.e., does it increase the R-sq relative to the first equation, which does not include the interaction effect).
Note also that using regression to test the relationship between the pre- and post- test allows for B1 to be something other than 1.0, while the paired-comparison simply tests whether the mean of subtracting the pre- from the post-test is different from zero. In other words, the regression approach is a more sophisticated model of the relationship between the two variables.
Moderation analyses usually involve the equivalent of interaction effects, so that the impact of the treatment would depend on the personality score. Before specifying the framework for the analysis, it would help to clarify the design. In particular, I presume that there is no control group, so that when you say "paired samples" you are thinking about something like a t-test to prepare the pre- and post- tests.
Well,
It is important to distinguish here between moderation and paired sample t test.
When you are describing your moderation model it seems that personality is a mediator between treatment and health. I am not sure if it is possible to set your independent variable at two time points (pre and post treatment).
Usually moderator is based on two categories (e.g. age - young, old, anxious - high, low etc). It is also possible that moderator may be continuous.
What I would suggest for you is to establish UNEQUIVOCALLY what your research question is.
Then you may have an idea if you need t test to check for the intervention efficacy (if there is a significant statistically mean difference pre and post intervention) or moderation (usually one of the categories of moderator operates differently that the other).
Without clarifying this issue you are on the way to statistical abyss. You could also talk to your peers if they are dealing with the same assignment or your teacher.
Finally when you type in moderation or t tests in the tube you will find easy to understand materials there.
Regards, Witold
The problem is that paired-sample t-Tests don't allow you to do multi-variate analysis, but moderation requires multivariate testing -- usually with regression.
In this case, it would be something like:
Y = a + B1X1 + B2X2
Y = a + B1X! + B2X2 + B3 X1*X2
Where Y is the post-test, X1 is the pre-test, X2 is the personality test, and X1*X2 is the product of the two independent variables. Note that the interaction term here is the product of two continuous variables, which is not the easiest thing to interpret, but the first step is to see if that term is significant (i.e., does it increase the R-sq relative to the first equation, which does not include the interaction effect).
Note also that using regression to test the relationship between the pre- and post- test allows for B1 to be something other than 1.0, while the paired-comparison simply tests whether the mean of subtracting the pre- from the post-test is different from zero. In other words, the regression approach is a more sophisticated model of the relationship between the two variables.
@David L Morgan Thanks for your reply. Indeed, i have to do moderation analysis with the effects about treatment (teeth whitening) but without control group, I think this is not adequate because we dont know the effects about treatment , then we can´t calculate how this effects are moderate for personality (NEO-PI). My solution is paired-sample t-Tests, then calculate a new variable ( before oral health perceived score - after oral health perceived score) and correlate this new variable with sub-scales of NEO-PI. However, i think your solution (Y = a + B1X! + B2X2 + B3 X1*X2) is a good idea, i will try it-
@Witold Orlik Thanks for your reply. The research question : Is the personality of a moderating effect of whitening teeth in the oral health perceived score? but the problem is that the data has no control group. The data have before and after oral health perceived score.
Regards
Hi,
I am attaching articles that used moderation testing. no control group was there.
hope this help
If you want to simply things, you can create a difference score between the two versions of the dependent variable, and correlate that with the personality measure(s).
That mirrors what a paired comparison t-Test does, but it misses out on some of the things that the regression can detect.
Depending on the size of your data set, it would probably be wise to examine the scatterplot of the difference score and the personality measure(s), since the relationships could be non-linear.
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics