BookmarkSubscribeRSS Feed
blbaker28
Calcite | Level 5

Hello All,

This is my first post but I've been using SAS for quite a long time.  Here is the background and my question.  I basically have a repeated measures pre and post test but only group.  Each participant filled out the survey twice. The first time was in response to the first scenario they were presented and the second time was in response to the second (treated) scenario.  The regression equation is the same, the respondents are also the same, the only thing that is different is that the first regression equation is run using the data the respondents gave in response to the untreated scenario and the second time I ran the equation, it was done using the data the same respondents gave after responding to the second, treated condition.  I"m wondering if there is any was to test if the regression coefficients that are produced after the first and second running are significantly different?

Any and all help is very much appreciated.

thank you

Brent

4 REPLIES 4
Doc_Duke
Rhodochrosite | Level 12

Brent,

There are a couple of ways to attack this problem, though the approach you described is not among them (it ignores the dependence within an individual).

If you don't have any covariates, then this reduces to a paired t-test.  Each observation has a pre and post value.

If you have covariates, you could look at this as controlling for the baseline assessment.  Use the post value as an outcome variable and add the pre-value to the predictor list.  The coefficients for the other covariates would then represent the marginal impact of covariate on the treatment effect.

You could also use the difference score as the outcome and the other variables a covariates.

There is a large literature on the appropriateness of each approach, just Google

difference scores vs baseline covariate

for more than you might want to know.

blbaker28
Calcite | Level 5

Doc,

Thank you very much for the reply.  I think I'm on the brink of 'getting it' but I just want to be sure.  I did run a paired t-test on the means of each construct (as an aside, it worked very, very well, as far as supporting my story).  However, I was thinking an interesting post-hoc analysis might be to take the examination a little deeper by also looking at how the relationships between factors (5 predictors 1 outcome) change after the treatment.  For example, one of the predictor variables is trust and the outcome variable is purchase intention.  The paired t-test revealed that after treatment, the means for both variables went up significantly (in the statistical sense).  I'm wondering now, if the relationship between these two variables also changed.  Would the approaches you describe above, regarding the use of difference scores, work for this analysis and provide added insight the t-test doesn't, if I don't include covariates?

Thanks again for the help

SteveDenham
Jade | Level 19

I think you have covariates--you are fitting some sort of regression equation.  So (pulls out soapbox) use the covariates and don't do difference scores.  This assumes a perfect linear response across all covariates over time, which is what I think you want to explore.  So something like:

proc mixed data=yourdata; /* This assumes that the response is continuous and has normally distributed errors.  If not, we need to shift to GLIMMIX. */

class pred1 pred2 pred3 pred4 pred5; /* I'm assuming that these are class variables */

model post = pred1 pred2 pred3 pred4 pred5 pre*pred1 pre*pred2 pre*pred3 pre*pred4 pre*pred4/solution noint;

run;

In the output, the pre*pred1, etc. values will give the change in the predictor effects comparing the pre-test to the post-test values.

Hopefully this will get you started.

Steve Denham

blbaker28
Calcite | Level 5

Steve,

Thanks for the helpful answer.  Just to be certain I understand, when you say I have covariates are you saying that I should use the pre-treatment measures as covariates and the post treatment measures as predictors of my post treatment dependent variable?  If that's all I had to do I feel a little silly since that's a pretty simple solution that was eluding me for some reason.  Maybe I was trying to make it harder than it is, assuming I'm interpreting your response correctly.

thanks for the help

Brent

sas-innovate-2024.png

Join us for SAS Innovate April 16-19 at the Aria in Las Vegas. Bring the team and save big with our group pricing for a limited time only.

Pre-conference courses and tutorials are filling up fast and are always a sellout. Register today to reserve your seat.

 

Register now!

What is ANOVA?

ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 4 replies
  • 1408 views
  • 7 likes
  • 3 in conversation