Vlad Surdea-Hernea
@vladsurdea.bsky.social
Ph.D. in Political Economy at the Central European University. Master of Public Policy at the Hertie School of Governance.
Causal inference with observational data. I really like maths, history, and politics.
Causal inference with observational data. I really like maths, history, and politics.
The article is (now) published in Comparative Political Studies:
journals.sagepub.com/eprint/CAN8D...
End of 🧵
journals.sagepub.com/eprint/CAN8D...
End of 🧵
Sage Journals: Discover world-class research
Subscription and open access journals from Sage, the world's leading independent academic publisher.
journals.sagepub.com
September 27, 2025 at 10:37 AM
The article is (now) published in Comparative Political Studies:
journals.sagepub.com/eprint/CAN8D...
End of 🧵
journals.sagepub.com/eprint/CAN8D...
End of 🧵
Process tracing, followed by a survey of Revolution participants provides further proof. A vast majority stated they were motivated by a "duty to participate" and it being the "right thing to do", while most did not believe regime change was possible at the outset.
11/🧵
11/🧵
September 27, 2025 at 10:37 AM
Process tracing, followed by a survey of Revolution participants provides further proof. A vast majority stated they were motivated by a "duty to participate" and it being the "right thing to do", while most did not believe regime change was possible at the outset.
11/🧵
11/🧵
They were also more likely to vote against neo-communist presidential candidates in the 1992 and 2004 elections.
10/🧵
10/🧵
September 27, 2025 at 10:37 AM
They were also more likely to vote against neo-communist presidential candidates in the 1992 and 2004 elections.
10/🧵
10/🧵
These anti-communist norms persisted long after 1989. In the early 1990s, "Gulag localities" saw more anti-government protests against the communist successors.
9/🧵
9/🧵
September 27, 2025 at 10:37 AM
These anti-communist norms persisted long after 1989. In the early 1990s, "Gulag localities" saw more anti-government protests against the communist successors.
9/🧵
9/🧵
We also provide preliminary evidence for our norm-based mechanism. First, exposure to the Gulag is linked to lower membership in the Romanian Communist Party. People in these areas were less likely to join the party, even when it was beneficial to do so.
8/🧵
8/🧵
September 27, 2025 at 10:37 AM
We also provide preliminary evidence for our norm-based mechanism. First, exposure to the Gulag is linked to lower membership in the Romanian Communist Party. People in these areas were less likely to join the party, even when it was beneficial to do so.
8/🧵
8/🧵
The result is robust to different model specifications, sub-sample choices, causal identification strategies, and estimators that deal with both the distribution of the main variables and the spatial nature of the data. It also holds for alternative outcomes.
7/🧵
7/🧵
September 27, 2025 at 10:37 AM
The result is robust to different model specifications, sub-sample choices, causal identification strategies, and estimators that deal with both the distribution of the main variables and the spatial nature of the data. It also holds for alternative outcomes.
7/🧵
7/🧵
Our key finding?
On average, localities with Gulag facilities had nearly 5 times more people seriously injured in confrontations with government forces during the Revolution (on average 3 injured, compared to a mean of 0.605).
6/🧵
On average, localities with Gulag facilities had nearly 5 times more people seriously injured in confrontations with government forces during the Revolution (on average 3 injured, compared to a mean of 0.605).
6/🧵
September 27, 2025 at 10:37 AM
Our key finding?
On average, localities with Gulag facilities had nearly 5 times more people seriously injured in confrontations with government forces during the Revolution (on average 3 injured, compared to a mean of 0.605).
6/🧵
On average, localities with Gulag facilities had nearly 5 times more people seriously injured in confrontations with government forces during the Revolution (on average 3 injured, compared to a mean of 0.605).
6/🧵
To test this, we examine one of the 20th century's most brutal episodes: the Romanian Gulag before 1965. We collect data on the distribution of labor camps, prisons, and extermination sites, matching it to the locality-level death and injury counts during the Revolution.
5/🧵
5/🧵
September 27, 2025 at 10:37 AM
To test this, we examine one of the 20th century's most brutal episodes: the Romanian Gulag before 1965. We collect data on the distribution of labor camps, prisons, and extermination sites, matching it to the locality-level death and injury counts during the Revolution.
5/🧵
5/🧵
This shifts the protesters' decision-making from a "logic of consequences" (risk-reward) to a "logic of appropriateness". Dissent becomes a moral duty, an expression of who they are and what their community stands for, regardless of the cost.
4/🧵
4/🧵
September 27, 2025 at 10:37 AM
This shifts the protesters' decision-making from a "logic of consequences" (risk-reward) to a "logic of appropriateness". Dissent becomes a moral duty, an expression of who they are and what their community stands for, regardless of the cost.
4/🧵
4/🧵
To address this, we came up with a theory of repression-induced norm formation. We argue that exposure to extreme repression, like the Gulag, forges powerful, lasting anti-regime identities and social norms within communities.
3/🧵
3/🧵
September 27, 2025 at 10:37 AM
To address this, we came up with a theory of repression-induced norm formation. We argue that exposure to extreme repression, like the Gulag, forges powerful, lasting anti-regime identities and social norms within communities.
3/🧵
3/🧵
Romania's was the only violent anti-communist revolution in 1989, with thousands either killed or injured. Yet protests grew monotonically with violence: the more the army shot, the more people joined. This is hard to explain using a purely consequentialist logic.
2/🧵
2/🧵
September 27, 2025 at 10:37 AM
Romania's was the only violent anti-communist revolution in 1989, with thousands either killed or injured. Yet protests grew monotonically with violence: the more the army shot, the more people joined. This is hard to explain using a purely consequentialist logic.
2/🧵
2/🧵
And more about collider bias than you hear in most such courses, if my memory of the class from the early Covid era is correct.
October 29, 2023 at 6:12 PM
And more about collider bias than you hear in most such courses, if my memory of the class from the early Covid era is correct.
Isn't clustering on a discrete running variable deemed generally problematic? I think Kolesár and Rothe (2018) show this, and there are some alternative estimators for CIs they propose in the RDHonest package.
October 25, 2023 at 12:07 PM
Isn't clustering on a discrete running variable deemed generally problematic? I think Kolesár and Rothe (2018) show this, and there are some alternative estimators for CIs they propose in the RDHonest package.
A firm that makes a particular type of cosmetics?
October 14, 2023 at 5:51 AM
A firm that makes a particular type of cosmetics?
I also now use it to upload nicely formatted figures and ask it to use my raw R results to produce something similar.
October 13, 2023 at 12:02 PM
I also now use it to upload nicely formatted figures and ask it to use my raw R results to produce something similar.
Is EB Garamond the boring and mainstream answer?
October 7, 2023 at 12:14 PM
Is EB Garamond the boring and mainstream answer?
To be fair, that's partly because the analysis rarely discusses the margin of error in any depth. This gives the impression that the uncertainty that people naturally associate with sampling is not properly addressed.
September 30, 2023 at 5:31 PM
To be fair, that's partly because the analysis rarely discusses the margin of error in any depth. This gives the impression that the uncertainty that people naturally associate with sampling is not properly addressed.
That's one the cleanest applications of mediation analysis I have seen in applied work, great paper!
September 30, 2023 at 11:46 AM
That's one the cleanest applications of mediation analysis I have seen in applied work, great paper!
Sounds to me like an ideal scenario for a vignette/conjoint to also measure the relative importance of different features of the estimators in the researcher's choice.
September 30, 2023 at 11:17 AM
Sounds to me like an ideal scenario for a vignette/conjoint to also measure the relative importance of different features of the estimators in the researcher's choice.
Taking this a step further, I would be curious to know how political scientists would choose their estimator given this information. It seems to me that we are in a race to keep up with the literature (e.g. DiD) without being critical of what the real benefits are of moving away from the basics.
September 30, 2023 at 11:08 AM
Taking this a step further, I would be curious to know how political scientists would choose their estimator given this information. It seems to me that we are in a race to keep up with the literature (e.g. DiD) without being critical of what the real benefits are of moving away from the basics.