For example, if I have a pre-post design experiment with 2 groups A and B. I can get change scores(Post minus Pre). And I can count the number of positive changes and negative changes. Hence I create a 2*2 table(GroupA/B*Positive/Negative Changes)

```

A B

Positive Change Score n1 n2

Negative Change Score n3 n4

```

Then can I use Chi-Square Test?

I have learned that ANCOVA is a better choice, but the change scores violate normality test assumptions so maybe ANCOVA is not a good one.