Solved
Contributor
Posts: 52

# Not a chi-sq, but...similar?

Okay SAS friends I know one of you has an answer for me!

I know how to run a chi-sq, love me a good chi-sq when appropriate.  Now I need to do something similar, but different...I think...LOL.

I have a set of pre-test scores (values of 0-8), I also have post-test scores (values of 0-8), finally, I have also computed the change in pre-test to post-test scores and the change has a normal curve so life is good...kind of.

I'm trying to figure out who learned more in the class. I ran a regression with the available demographic variables, I even broke it down into t-tests to check in baby steps, but there's nothing that's significant in predicting the change in score.

I *think* the best way to do this is by checking pre-test scores against post-test scores. So I want to know for those who scored a 0 on the pre-test, for those who scored 0 on the pre-test what were the resulting post-test scores, for those who scored 1 on the pre-test what were the resulting post-test scores, for those who scored 2 on the pre-test what were the resulting post-test scores...all the way up to 8 on the pre-test and the corresponding score(s) on the post-test.

How do I do that?

K8

Accepted Solutions
Solution
‎10-07-2016 03:09 PM
Super User
Posts: 10,770

## Re: Not a chi-sq, but...similar?

Yeah. It is matched/paired data . If you only care about pre-score and post-score,

Check PROC TTEST+PAIRE, or dif=post-score - pre-score and use PROC MEANS check dif=0 or not.

If you want include other variables to do covariance analysis,use PROC GLM.

``````proc glm data=have;
class sex;
model post_score=pre_score age sex/solution;
quit;``````

All Replies
Posts: 5,526

## Re: Not a chi-sq, but...similar?

Before demographics, I would think that the best predictor of difTest = postTest - preTest would be preTest. Did you check that, using simple anova?

PG
Contributor
Posts: 52

## Re: Not a chi-sq, but...similar?

In terms of prediction, yes, pre-test predicts the difference, but I was being asked to come up with something different. Thank you, PG for getting back to me!!

Solution
‎10-07-2016 03:09 PM
Super User
Posts: 10,770

## Re: Not a chi-sq, but...similar?

Yeah. It is matched/paired data . If you only care about pre-score and post-score,

Check PROC TTEST+PAIRE, or dif=post-score - pre-score and use PROC MEANS check dif=0 or not.

If you want include other variables to do covariance analysis,use PROC GLM.

``````proc glm data=have;
class sex;
model post_score=pre_score age sex/solution;
quit;``````
Contributor
Posts: 52

## Re: Not a chi-sq, but...similar?

I haven't used proc glm much, but will try it Ksharp! Thanks for the help!

Kate

Community Manager
Posts: 681

## Re: Not a chi-sq, but...similar?

"...love me a good chi-sq when appropriate..." It was so much fun to read your post! Have a great weekend.

Contributor
Posts: 52

## Re: Not a chi-sq, but...similar?

@BeverlyBrown LOL...just trying to keep it real! If we can't laugh while doing stats we may very well cry...so I'll choose laughter every time.

I hope you have a great weekend too!!

K8

Community Manager
Posts: 681