Hi,
I ran a DID regression and now want to test the parallel trends assumption MANUALLY. I write the manual joint significance test in the following form:
test 1.treatment_dummy#2.time_dummy 1.treatment_dummy#3.time_dummy 1.treatment_dummy#4.time_dummy 1.treatment_dummy#5.time_dummy 1.treatment_dummy#6.time_dummy 1.treatment_dummy#7.time_dummy 1.treatment_dummy#8.time_dummy 1.treatment_dummy#9.time_dummy 1.treatment_dummy#10.time_dummy 1.treatment_dummy#11.time_dummy 1.treatment_dummy#12.time_dummy 1.treatment_dummy#13.time_dummy 1.treatment_dummy#14.time_dummy 1.treatment_dummy#15.time_dummy 1.treatment_dummy#16.time_dummy 1.treatment_dummy#17.time_dummy 1.treatment_dummy#18.time_dummy 1.treatment_dummy#19.time_dummy 1.treatment_dummy#20.time_dummy 1.treatment_dummy#21.time_dummy
These are all the interactionvariables pre-treatment. I get F( 4, 5) = 3.42 & Prob > F = 0.1050.
Hence the parallel trends assumption should hold.
But when I plot the coefficient and their confidence intervals in an Event Study plot it really does not look like parallel trends holds.
Please see the attached PDF.
What do you think about this? Should I treat this as parallel trends? Or argue that even though we get a statistically insignificant value on the joint test, we see a problem in treating it as such?
Kind regards,
Oliver
I ran a DID regression and now want to test the parallel trends assumption MANUALLY. I write the manual joint significance test in the following form:
test 1.treatment_dummy#2.time_dummy 1.treatment_dummy#3.time_dummy 1.treatment_dummy#4.time_dummy 1.treatment_dummy#5.time_dummy 1.treatment_dummy#6.time_dummy 1.treatment_dummy#7.time_dummy 1.treatment_dummy#8.time_dummy 1.treatment_dummy#9.time_dummy 1.treatment_dummy#10.time_dummy 1.treatment_dummy#11.time_dummy 1.treatment_dummy#12.time_dummy 1.treatment_dummy#13.time_dummy 1.treatment_dummy#14.time_dummy 1.treatment_dummy#15.time_dummy 1.treatment_dummy#16.time_dummy 1.treatment_dummy#17.time_dummy 1.treatment_dummy#18.time_dummy 1.treatment_dummy#19.time_dummy 1.treatment_dummy#20.time_dummy 1.treatment_dummy#21.time_dummy
These are all the interactionvariables pre-treatment. I get F( 4, 5) = 3.42 & Prob > F = 0.1050.
Hence the parallel trends assumption should hold.
But when I plot the coefficient and their confidence intervals in an Event Study plot it really does not look like parallel trends holds.
Please see the attached PDF.
What do you think about this? Should I treat this as parallel trends? Or argue that even though we get a statistically insignificant value on the joint test, we see a problem in treating it as such?
Kind regards,
Oliver
