Large-scale online deanonymization with LLMs shows how easily pseudonymous users can now be re-identified from unstructured text alone.
When discussing our EJMR paper at NBER SI 2023 Catherine Tucker argued that rising compute would soon erode online anonymity. The future may already be here.
The figure shows a descriptive overview of age, education, gender and party vote distribution in the full sample (grey bars) alongside willingness (green) and completion rates (blue) by group. With regard to age, younger respondents appear more likely to being willing to participate as well as to complete the donation task. For education, there is a tendency for individuals with higher educational attainment to be more likely to complete the donation. The very small proportion of individuals with lower levels of education does not allow for conclusions on their likelihood of being willing to participate or completing the data donation study. Overall, the sample contains more self-identified men than women, and men show a slightly higher likelihood of being willing to participate and completing the data donation. Regarding partisan affiliation, supporters of the Green Party appear substantially more likely to finish the data donation, followed - though to a lesser extent - by voters of the FDP and Die Linke.
We study differences in participation by party vote and socio-demographics based on election survey data @gles.bsky.social linked with a data donation study during the 2025 German federal election 👇Preprint link: doi.org/10.31235/osf...
To account for the two-step process of participation, first being willing, then actually donating, we use a bivariate probit model with partial observability. Also, we experimented with naming Google in the invitation or not. Check it out in our preprint: doi.org/10.31235/osf...
The figure shows the predicted probabilities by party vote, derived from the partially observed bivariate probit model, separately for willingness and completion, and the joint probability. Voters of the CDU/CSU, the AfD, the SPD, the LINKE, and BSW exhibit predicted probabilities of willingness to participate close to the sample average of 50%. Green Party voters show a higher predicted probability of willingness to participate, while the elevated probabilities among FDP voters are not statistically different from those of CDU/CSU voters. In contrast, voters of other parties and non-voters, including those who did not disclose their vote choice, display lower predicted probabilities of willingness. When it comes to donation completion, both Green Party and FDP voters stand out with higher predicted probabilities. As a result, the joint predicted probabilities indicate that voters of the Greens and FDP are the most likely to both opt-in and follow through with the data donation, even when controlling for age, gender, and education.
Who opts into a data donation study asking for Google Search histories – and more importantly, who actually donates their data? Sina Chen and I found that Green Party and FDP voters were more likely to follow through, hinting at political skew in donated data 👇
#DataDonation #DBD #GLES #ElectionResearch
Who actually donates their #GoogleSearch data during an election #BTW25?
Barbara Binder and Sina Chen explore the potential of enriching election surveys with donated platform data in their new paper "Opting in but not donating?
🚨 New GLES data release for #BTW25:
GLES Tracking March 2025 – repeatedly questioned respondents T60W (ZA10132)
💻📱 CAWI survey
👥 n = 977 interviews
📅 Fieldwork: 05–21 March 2025
🗳️ Post-election follow-up survey of a subset of the T60 (ZA10105) respondents.
📥 Access the data: doi.org/10.4232/5.ZA...
Wow!
The “official” poverty guidelines were quite problematic. But hard to see how eligibility for Medicaid & other programs will be determined given they were based on these guidelines.
The field of gender and sexual minority (LGBTQ) health is being erased. Hundreds of ongoing NIH grants being immediately terminated. Deb Umberson and my decade long longitudinal study on marriage and health joined the termination list last night.
The current administration is straight up censoring the scientific record by requiring authors to alter an already-accepted research paper because it uses banned words such as "gender" and "equitable" and presents data broken down by sexual orientation.
🚨 New Preprint! 🚨
How are generational shifts shaping hormonal contraceptive use in Germany? We analyze societal, economic, and personal predictors across three cohorts using machine learning.
💡 Key findings below 👇
Read here: doi.org/10.31235/osf...
Thanks to co-authors Theresa Nutz & Bryan Chan
Looking for linkable social media data of candidates? 🔎 Check out the candidates’ social media accounts (ZA7959) and the candidates’ social media activities (ZA8917) and contact the GLES team for possibilities to link the data!📧
🚨 New data release! 🚨 GLES Candidate Study European Election 2024 (ZA7960)!🎉
Questions include background & nomination, election campaign, attitudes, democracy & representation. Special focus on use of social media in election campaigns.
➡️ dx.doi.org/10.4232/1.14...
🚨Check out our new paper with @schechtlm.bsky.social on the role of public redistribution in wealth-income correlations across countries! 🚨💰💶
OA in Socius doi.org/10.1177/23780231241261599
Data collection of GLES Candidate Study European Election 2024 has started!🎉
Questions on candidates' backgrounds, recruitment & selection, attitudes, policy positions, campaigns & special focus on social media campaigning
Pre-release questionnaire: www.gesis.org/en/gles/over...
Der Grundfreibetrag soll ein festgelegtes steuerfreies #Existenzminimum garantieren. Die absolute Entlastungswirkung steigt jedoch kontinuierlich mit dem Einkommen, wie unsere neue Studie zeigt. #ifo_Institut 🧵1/2 ➡️ zentrum-neue-sozialpolitik.org/projekte/ifo...
🚨 New data release! 🚨 GLES Candidate Study European Elections 2024, Social Media Accounts (ZA7959)! 🎉
The data covers accounts from Facebook, Instagram, X, and personal websites of German candidates for the upcoming European Elections.
➡️ doi.org/10.4232/1.14...
A graph with the headline "The lowest-wage workers had the strongest wage growth during the pandemic" and a sub-headline that reads "Real wage growth across the wage distribution, 2019-2023"
🚨 Policy works! 🚨
Low-wage workers saw dramatically fast wage growth from 2019-2023—even after accounting for inflation—largely because of federal and state policy measures like COVID relief and minimum wage increases. This is in stark contrast to prior decades. 🧵 1/4
Eine Chance für die Quote?
In Hessen einigen sich Land und Gewerkschaften auf eine bundesweit einzigartige Regelung zum Ausbau von Dauerstellen an Hochschulen. Was das für die #WissZeitVG-Debatte bedeuten könnte. #IchbinHanna
Im Blog: www.jmwiarda.de/2024/03/18/e...
That very viral "Holocaust denial among young people" finding doesn't replicate.
t.co/9wT82XfBzW
We want to thank everyone who commented on earlier versions presented at conferences, and we are happy to hear your thoughts!
Further research should improve knowledge on country-specific dynamics, particularly on the interaction between employment and social protection and how it affects poverty outcomes. Another challenge is to improve the anchored poverty measure to account for enduring trends in median income.
Our results can enrich the ongoing debate on the European Union’s social target indicator and provide further insights for the developments of alternative poverty thresholds.
When using the anchored at-risk-of-poverty rate among the active-age population as the dependent variable, the association between employment and poverty becomes substantially stronger – up to four times higher than when using the floating poverty line.
However, the most important factor in this regard proves to be the change in the median incomes! Thus, a large part of the story is how the distribution of incomes, and not only the situation among the poor population changed.
Furthermore, the expansion of part-time employment, measured both on individual and household levels as a form of non-standard work added to a blurred relationship and to disappointing poverty trends.
The poverty trends would have been more encouraging during economic upturns if new jobs were distributed more favorably across households and if poverty risks among the jobless wouldn’t have increased.
So why don’t EU (and many country)-level descriptive statistics reflect this relationship after all? We suggested four mechanisms that blur the relationship, and our analysis provides empirical evidence in support of each of these mechanisms.
Using decomposition analysis and macro-level regressions on #EUSILC data for years 2004 to 2017, we find a consistent, negative relationship between employment (individual and household levels) and relative income poverty. This challenges previous findings and concerns of social policy experts.
This figure shows trends in employment (left axis), low work intensity and poverty (right axis) for the EU-27 average (UK included), from 2004 to 2017 (in %). @fdzgml.bsky.social @eu-eurostat-mirror.bsky.social
Consistently high EU-average poverty rates, despite rising employment, continue to generate debates about their relationship and the drivers of poverty rates.
