Attrition Test on Practice Questions

[madisonfrazee]

On the practice final under Attrition Tests, we are unsure what the regression for the random attrition test is.

This is the equation we came up with,

Y= X +b1(dummy_D) 1 = attrition 0 = stayed B1=0 means random attrition

Does that seem right?

[danafuller]

I am also confused by the progression of the bullet points under Attrition Tests. Are they just informative points leading up to having all of the information to construct the regression? If so, would it be y=x + b1time + b2 D (attrition)? I only slipped in the b1*Time because it says it is reflexive, but I obviously need some clarification

[lecy]

From the discussion yesterday to make sure this is clear:

Most studies will report group equivalence before and after attrition occurs (tests 0 and 1 in the diagram). But you could test for random attrition in two other ways as well.

I am using the term "random attrition" a little imprecisely. You could have non-random attrition, but as long as it's non-random in the same way in both groups it would still result in balanced groups (all high performers leave from each group, for example).

Test 1 looks at group balance after attrition. It would be agnostic to the type of attrition, but as long as it did not impact balance it should not matter.

Test 2 is the only real test for random attrition. Test 3 would ensure that both groups experience the same type of attrition, but it does not necessarily have to be random.

For any of these tests you would subset the data to isolate the two groups, create a dummy variable for one group, then run the regression:

Y = b0 + b1(dummy) + e

If b1 is significant it means the groups are not balanced (1), attrition is non-random (2), or attrition is non-equivalent (3).

Y can be the pre-treatment outcome, or it can represent a series of study participant traits, similar to what you did in Lab 2.

---

From CH21 - tests 0 and 1 reported in the table.