Timeline for How to plot the graph or perform a formal test of parallel trends for generalized DiD?
Current License: CC BY-SA 4.0
22 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 18, 2021 at 1:17 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Further explication provided.
|
| Jul 17, 2021 at 18:13 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Further explication provided.
|
| Jul 7, 2021 at 3:22 | comment | added | Thomas Bilach | The authors tested the effect of liberalizing bar hours on traffic accidents. Their results suggest they found a decrease in accidents once bar hours were extended. Notice how the effects grow larger in subsequent periods. It suggests a growing policy influence over time. | |
| Jul 3, 2021 at 0:52 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Minor textual edit.
|
| Jul 1, 2021 at 1:39 | comment | added | Phil Nguyen | Hi @Thomas Bilach, can I ask what does the decrease in coefficients over time in the last graph means (from Green, 2015)'s paper? | |
| Jun 17, 2021 at 16:37 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Minor textual edits.
|
| Jun 9, 2021 at 23:13 | comment | added | Thomas Bilach | Correct. It's one way to demonstrate the stability of the trends in the pre-period. It doesn't prove it, it's just evidence to support it. | |
| Jun 9, 2021 at 21:01 | comment | added | Phil Nguyen | Thank you for you amazing explanation, Thomas, everything is clear now, just one last one curiosity in this topic, so, regarding the question (5) above, I remember I read one of your answer previously. And it seems that coefficients plotting is also a way to do pre-trend testing, is not it? and we can circumvent the hardness regarding plotting the control trend in Dagupsta Figure 1 case? | |
| Jun 8, 2021 at 18:11 | comment | added | Thomas Bilach | (7) Surveys were conducted biennially since 1991, hence why the time intervals are expressed this way. They dropped the "2-year period" before the ban to act as a reference. | |
| Jun 8, 2021 at 18:04 | comment | added | Thomas Bilach | (5) As indicated in my answer, it is difficult to define the relative periods when the enactment years vary so widely across countries. However, when it comes to plotting the coefficients, then you do not need to drop Jordan. Simply put the lead and lag indicators into your model. The values for Jordan will be consistently 0. (6) I simply meant that it's permissible to include more leads. The authors decided to plot leads for 1 and 2 years before adoption. In other studies, the authors may plot estimates 3 or even 4 years before adoption. It's up to you to decide how many to include. | |
| Jun 8, 2021 at 17:51 | comment | added | Thomas Bilach | (1) By "schism" I was drawing attention to how the lines seem to break apart around $t = 2$. (2) By "entities" I mean firms. And the time relative to treatment is when we center all treated units around $t = 0$. (3) Yes. You create a new variable to show the relative periods. In other words, the group of treatment/control firms are all some number of time units approaching the enactment year (i.e., $-3, -2, -1$) and some number of time units away from the enactment year (i.e., $0, +1, +2, +3$). (4) The variable $d^{+\bar{3}}_{kt}$ is a "binned" indicator. It 'turns on' and stays on. | |
| Jun 7, 2021 at 22:04 | vote | accept | Phil Nguyen | ||
| Jun 7, 2021 at 22:00 | history | bounty ended | Phil Nguyen | ||
| Jun 7, 2021 at 10:02 | comment | added | Phil Nguyen | Regarding the question (5) above, I am wondering if only it without trend plotting is okay enough for pre-trend testing then ? I sent an email to the author but there is no response so far so I am finding an alternative way to do so. | |
| Jun 7, 2021 at 9:55 | comment | added | Phil Nguyen | (7) I am quite confused about the event time (-2/-1) in your amazing reproduction work from Venkataramani,2019 . I do not understand what does this number means and why we need to perform the number like that. | |
| Jun 7, 2021 at 9:44 | comment | added | Phil Nguyen | (6) Regarding the sentence "It's also worth highlighting that the authors acquired employment data from 1979–1995, so they didn't have to limit themselves to a finite number of adoption leads." I understand what you mean, but I do not know why you put it here as a note "so they didn't have to limit themselves to a finite number of adoption leads", maybe I miss something important then ? | |
| Jun 7, 2021 at 9:40 | comment | added | Phil Nguyen | (5) Regarding your full specification, it seems that we only can run with Belgium and** Iceland**, isn't it? Because we cannot define the relative indicator (lead/lag) of Jordan, I guess. And the full specification if it is explained like mine , it seems that it is also an approach to test pre-trend without using the control group, I hope? | |
| Jun 7, 2021 at 9:09 | comment | added | Phil Nguyen | (4) I do not fully understand the meaning of the word "binned" in this sentence "The endpoint is "binned" to index all periods 3 years or more after the law change". Could you please kindly help me to clarify it? | |
| Jun 7, 2021 at 9:02 | comment | added | Phil Nguyen | (3) "it's a plot of the relative period indicators", do you mean converting the year to relative year around the treatment year (-2;-1;0;1;2;3) ? | |
| Jun 7, 2021 at 8:08 | comment | added | Phil Nguyen | (2) regarding this sentence "But do you even observe outcomes beyond 2012? Remember, the authors are plotting aggregate trends. Each entity is some amount of time units relative to the immediate adoption period.". So, do you mean "But do you even observe outcomes beyond 2012?" that whether I have the data of dependent variables from 2012 to 2017? And can you please help me to clarify this sentence more "Each entity is some amount of time units relative to the immediate adoption period.". I guess that "entity" here is firm but I cannot fully understand the sentence. Much appreciated. | |
| Jun 7, 2021 at 8:04 | comment | added | Phil Nguyen | Thank you for your dedicated answer, Thomas Bilach, can I ask a couple of questions: (1) I do not fully understand the meaning of this sentence "Note the schism after t=2", or in another word, what does "schism" mean in this sentence? Sorry I am from a country that English is the second language so sometimes I cannot catch the idea reasonably | |
| Jun 6, 2021 at 0:43 | history | answered | Thomas Bilach | CC BY-SA 4.0 |