Bacon Decomposition bacondecomp Chart Interpretation

[Camila Alvayay]

Hi there, I am having problems interpreting the chart from the

bacondecom

p output. I would really much appreciate any help in this regard.

I am evaluating a staggered intervention. Units are census blocks and years are from 2006 to 2019. The first round happened in 2009, and the second round in 2011. I want to provide some motivation to use the C&S estimator, and I wanted to see the results from the Bacon Decomposition first. And even though I read the paper, see the slides, and see other sources, I still have trouble understanding what is really showing us.

This is the code:

xtreg        sale_price age  lotsizesquarefeet noofstories fullbath GarageAreaSqFt i.year trend1-trend1067 post, fe vce(cluster bg00)
bacondecomp  sale_price post age lotsizesquarefeet noofstories fullbath GarageAreaSqFt trend1-trend1067 , stub(Bacon_)

And this is the figure I obtained:

My problem with this picture is the labels. In the theoretical explanation, I always see "early vs later" groups, and in this picture, I see "Timing groups", or "Always treated vs timing", which I do not understand. If anybody has a clue on this I would really appreciate it! Thank you.

[Venessa Yu]

Hello Camila,

Timing_groups: means the weights of (1) Earlier Treated vs. Later Control, and (2) Later Treated vs. Earlier Control. And the latter comparison, which is Later Treated vs. Earlier Control usually is a bad comparison, which cause bias in TWFE regression, because TWFE put the earlier treated units as the control group of the later treated units, and this is an evidence of bias.

Can you also show the result of bacondecomp, which includes the beta and total weights of Timing groups and Never v. timing?