Timeline for Is a DID model appropriate?
Current License: CC BY-SA 4.0
23 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 26 at 23:20 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
added 16 characters in body
|
| Mar 30, 2021 at 4:48 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Minor textual edit. Replaced the dash with an en dash.
|
| Mar 27, 2021 at 21:24 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edited the equation (dropped the intercept). Other textual edits.
|
| Oct 3, 2020 at 15:16 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edited the text.
|
| Aug 18, 2020 at 15:59 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edited the equation.
|
| Aug 18, 2020 at 15:51 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Further explication to understand why this is important.
|
| Aug 17, 2020 at 22:46 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Further explication to understand why this is important.
|
| Aug 17, 2020 at 22:27 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edits to answer.
|
| Aug 17, 2020 at 22:14 | comment | added | Thomas Bilach | @Bryan You can think of $\text{Prohibition}_{st}$ as your interaction term. It is equal to 1 if it is a treated state and it is in the post-treatment period. I updated my answer and simulated some toy datasets to help with your intuition. Let me know if anything is unclear! | |
| Aug 17, 2020 at 21:57 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Further explication to understand why this is important.
|
| Aug 17, 2020 at 16:34 | comment | added | Bryan | So, no interactions in the model? Since you say 'dummies', I should use dummy coding and not effect coding. | |
| Aug 17, 2020 at 15:37 | comment | added | Thomas Bilach |
@Bryan I’m not sure what you mean when you say we are disposing of the ordered nature of years. The DiD model requires a full set of $T-1$ year effects. The as.factor(year) notation in R will create these year dummies for you. This accounts for the time shocks common to all states. Also, clustering your standard errors at the state level should help with the dependency among observations within a state across years. If you opt to remove the “always treated” states, you should look into other finite sample adjustments.
|
|
| Aug 17, 2020 at 15:27 | comment | added | Thomas Bilach | @Bryan To be clear, you do have a panel and not “repeated cross-sections” correct? The difference is not trivial in this setting. | |
| Aug 17, 2020 at 13:47 | comment | added | Bryan | Wouldn't I want to use a model that also also partially takes care of the autocorrelation that would be expected in a time course, such as random intercepts for state? After all, it is repeated measures data. | |
| Aug 17, 2020 at 13:42 | comment | added | Bryan | Why "as.factor(year)"? Why do I need to dispose of the ordered nature of years? Likewise, that would make the year-based variable into a factor with over 50 levels. | |
| Aug 15, 2020 at 23:11 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edited the text.
|
| Aug 15, 2020 at 17:05 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Further explication to understand why this is important.
|
| Aug 14, 2020 at 20:47 | comment | added | Bryan | There have been no repeals. I could attempt to find a surrogate outcome variable to be able to go back the decades when all states lacked the prohibition. | |
| Aug 14, 2020 at 20:09 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edited the text.
|
| Aug 14, 2020 at 19:55 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edited the text.
|
| Aug 14, 2020 at 19:44 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Edited the equation.
|
| Aug 14, 2020 at 19:38 | history | edited | Thomas Bilach | CC BY-SA 4.0 |
Further explication to understand why this is important.
|
| Aug 14, 2020 at 19:30 | history | answered | Thomas Bilach | CC BY-SA 4.0 |