DID

Question 1

how is DID implemented?

Answer 1

basically post=1 (change in both treated and control in post period), treated=1 (intercept difference…difference between treatment and control NOT due to effect of treatment), posttreatment (the subset…change in treated in post period). If there are different effects over time then include a year dummy (for control) and yeartreated for treated, latter is the treatment effect.

Question 2

can we use DID if the control and treated don’t move in parallel prior to treatment?

Answer 2

depends. If control remains constant in the post-period (via ITS?), then we may be able to use a DIDID where we compare the difference between the treated in the pre- and post. Time is slope for all over time, timetreated is the differential slope for treated over time not due to treatment administration, timetreated*post gives the effect of treatment relative to the trajectory of the treated pre-treatment

Question 3

why might we use DID? How do we know it’s needed?

Answer 3

if there was a change (treatment) between two groups and we believe that change was the only important difference between the groups. The alternative here is ITS, which we might not trust due testing bias, history, unstable instrumentation, maturation, etc.

Question 4

what are the limitations of DID?

Answer 4

if anything changed between treated and control after treatment archetypal DID only picks up average changes for treated in post-treatment period. differential interactions with treatment over time could be modeled using time and time*treatment.

Multiple treatment variables, because all the combinations among treated groups and controls and nesting becomes too complicated
If treatment starts at different points in time for different groups
A treatmetn with dosages, for example, PPE, cannot be used because we can’t use dummies.

Question 5

For DID to provide a valid counterfactual, what must we believe?

Answer 5

both are exposed to same external influences as treatment (if there’s a maturation or time shock then it needs to apply to both groups)

Response to the external influences have the same impact on control and treatment