I use Difference-in-difference (DID) method to evaluate impact of a policy. Stata 17.0 has the command “didregress” for the DID estimation. It also allows us to test parallel trend assumption by using “estat trendplots” and “estat ptrends” after a “didregress” estimation. I have some questions as below:
1) Assume that using "estat ptrends", I have the following results:
. estat ptrends Parallel-trends test (pretreatment time period)
H0: Linear trends are parallel
F(1, 45) = 3.16
Prob > F = 0.0515 When (Prob>F) is greater 0.05, we cannot reject the null hypothesis of parallel trends, right? In this case, the test suggest a parallel trend? How about the value of F(1,45) here?
2) Using “estat trendplots”, I have the below graphs: How can I interpret the graph results? Should I look at the “Observed means” graph or the “Linear-trends model” graph to consider if the parallel trends assumption is satisfied? In my case, can I state that prior to the policy implementation, treated and control groups followed a parallel path?
3) The guide of STATA 17.0 writes that “We can also think of nonparallel as an indication of an anticipatory treatment effect. We saw that the trends were not parallel before the treatment took place, which could indicate a treatment effect even before the treatment is implemented. Thus, another way to state our parallel-trends assumption is that there should be no treatment effect in anticipation of the treatment. To test this assumption, we could fit a Granger-type causality model.”
In case that the “estat trendplots” cannot provide a clear evidence of parallel trends, should I use the Granger-type test (“estat granger”) instead? If “estat ptrends” suggest parallel trends but “estat granger” rejects the assumption, can I report only the results of “estat ptrends” and ignore thoses of the “estat granger”?
1) Assume that using "estat ptrends", I have the following results:
. estat ptrends Parallel-trends test (pretreatment time period)
H0: Linear trends are parallel
F(1, 45) = 3.16
Prob > F = 0.0515 When (Prob>F) is greater 0.05, we cannot reject the null hypothesis of parallel trends, right? In this case, the test suggest a parallel trend? How about the value of F(1,45) here?
2) Using “estat trendplots”, I have the below graphs: How can I interpret the graph results? Should I look at the “Observed means” graph or the “Linear-trends model” graph to consider if the parallel trends assumption is satisfied? In my case, can I state that prior to the policy implementation, treated and control groups followed a parallel path?
3) The guide of STATA 17.0 writes that “We can also think of nonparallel as an indication of an anticipatory treatment effect. We saw that the trends were not parallel before the treatment took place, which could indicate a treatment effect even before the treatment is implemented. Thus, another way to state our parallel-trends assumption is that there should be no treatment effect in anticipation of the treatment. To test this assumption, we could fit a Granger-type causality model.”
In case that the “estat trendplots” cannot provide a clear evidence of parallel trends, should I use the Granger-type test (“estat granger”) instead? If “estat ptrends” suggest parallel trends but “estat granger” rejects the assumption, can I report only the results of “estat ptrends” and ignore thoses of the “estat granger”?
Attached Files
