1. Regression vs. t-test

Hello all,

I have this small study, and I am wondering how to analyze it.

I have a measure X taken before some event, and after for every participant (subject).
In addition, I have two groups, "treatment" and "control".

I want to show, that for the treatment group there is a very small change between the measure after and the measure before, in compare to the control group, in which I expect a much bigger difference. I could also show that for the treatment group the values after are much lower than for the control group (will also show the effect I am after).

I am considering two options:

1) Calculate the difference for each subject. Then, compare the means of the differences in the two groups (using a t-test).

2) Use regression: the dependent variable will be the measure after the event, and the independent variables will be the baseline value (the value before) and the group.

Which way would you choose ? Do you think there are special advantages and disadvantages for each method ?

thank you !

2. Re: Regression vs. t-test

I have this small study, and I am wondering how to analyze it.
Do you mean a study with small sample sizes?
How large were your sample sizes, and BTW what is this all about,
what are the topic and the research question of your study?
I have a measure X taken before some event, and after for every participant (subject).
What did you actually measure, on which scale?
In addition, I have two groups, "treatment" and "control".
Were subjects allocated randomly to groups?
I want to show, that for the treatment group there is a very small change between the measure after and the measure before, in compare to the control group, in which I expect a much bigger difference.
How do you define large/small change? - But apart from that,
in a small study, you will be happy if you are able to show at
all whether there is a change or not, regardless of how big or small.
I could also show that for the treatment group the values after are much lower than for the control group (will also show the effect I am after).
I don't know what you mean by "much lower".
But if you restrict yourself to the more easy question
whether the change is just different between goups,
this will be much more easy to handle.

If you have an interval scaled measure, and sample sizes
are large enough, then you could consider using repeated
measures (mixed) ANOVA with time as within-subject factor
and group as between-subjects factor.

With kind regards

K.

3. Re: Regression vs. t-test

I am talking about an operation, that usually causes a damage to the flexibility of some muscle. There is a new treatment that should prevent or reduce this damage to the muscle.

The flexibility is measured in angles, but not periodic, it doesn't go to 360 degrees, the scale is 0 to 90 degrees.

For each subject, the flexibility is measured before the surgery, and after. There are two groups, one with the treatment and one without. 8 subjects in each group, randomized !

I want to show that for the treatment group the flexibility before was kept, or almost kept, and for the control, not (big difference between before and after).

The sample is small, however, without specifying numbers (don't have it in front of me), the differences are very large between the groups, so from power point of view, the sample should be sufficient.

My question is however more general, which way to go:

1) calculate (After-before) for each subject and use a t-test for comparison
2) Use the flexibility after as a dependent variable vs. the flexibility before and the group in a regression model.

I tried avoiding a mixed model or RM ANOVA because there are only 2 observations for each subject, so I am trying to simplify it.

Let's also ignore for now the assumptions of normality and all that, I will deal with it later, now I am asking in general which approach is better.

4. Re: Regression vs. t-test

I think the advantage of the regression approach is that it will adjust for the baseline measurement of flexibility. In other words, some patients may already be highly flexible prior to surgery so any impact that surgery has on them will not be as noticeable. In this sense there is a ceiling effect. So adjusting for baseline will, in my opinion, be the best way to determine if treatment does differ from control.

Another possibility is developing a regression model of the change (post less pre) that includes the pre baseline measure as an independent variable.

Hope this helps. Good luck with your study!

5. Re: Regression vs. t-test

I want to show that for the treatment group the flexibility before was kept, or almost kept, and for the control, not (big difference between before and after).
You can demonstrate this by descriptve statistics of your sample
data. But you probably won't be able to use a statistical
test to make such statements for the populations.
1) calculate (After-before) for each subject and use a t-test for comparison
With n=16, I'd consider U-test.
2) Use the flexibility after as a dependent variable vs. the flexibility before and the group in a regression model.
If you think you can do a linear regression, you can
also do repeated measure ANOVA, which IMHO is
easier to interpret.
I tried avoiding a mixed model or RM ANOVA because there are only 2 observations for each subject, so I am trying to simplify it.
I have never heard of such a justification before.

With kind regards

K.

6. Re: Regression vs. t-test

mostater, thank you, what you said is more or less what I was thinking.

Karabiner, by U test do you mean Mann-Whitney ?

I wouldn't imagine that interpreting a RM ANOVA would be easier than linear regression. How would you set the model ? Y=flexibility after, X1=flexibility before X2=time ?

7. Re: Regression vs. t-test

Ok, some might find "the baseline-adjusted post-treatment scores were different between groups" mor easy to interpret than "the difference between baseline and posttest was higher for group1 than group 2".

Regarding the model, you would have 1 within-subjects (repeated measures) factor and 1 between-subjects (grouping) factor. But if you are not accustomed to ANOVA models, you might indeed preferably use regression.

With kind regards

K.

 Tweet

Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts