iiinova — where your digital self keeps evolving

Rich Frank @richwfrank.bsky.social · 10/10/2026

Methods & disclosure: Claude Opus 5.5 scraped the narratives from state.gov, ran the text comparisons, and made the figures. I designed the comparison and checked the numbers against the reports. Shorter texts lower “share mentioning” rates, so I only flag big shifts. Any errors are mine. 11/11

000

Rich Frank @richwfrank.bsky.social · 10/10/2026

Finally, the 2026 #TIPreport includes the highest reported # of prosecutions (24,254) and victims identified (156,912) of any year since this table begins in 2019. Nevertheless, the country narratives that explain those numbers are much thinner than a year ago. 10/n

Three small bar charts of TIP report figures for the 2020 to 2026 reports, with 2026 highlighted in blue. Each report's law enforcement figures cover the previous calendar year. Left: global prosecutions, from 11,841 in the 2020 report to a high of 24,254 in 2026 (previous high 18,774 in 2024). Middle: victims identified, from 111,932 in 2020 to a high of 156,912 in 2026 (previous high 133,939 in 2024). Right: average words per country narrative, which rises from 2,900 in 2020 to 3,962 in 2024, then falls to 3,528 in 2025 and 2,175 in 2026, the lowest of the seven reports. Data processed and chart made by Claude Opus 5.5; checked by Richard W. Frank.

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

The US narrative shrank most of all: 13,396 words in 2024, 3,250 in 2026. It reports that federal funding for NGO victim services fell from $82.3m to $18.9m and that the national action plan lapsed. The US remains Tier 1. New passages praise Texas and Florida. 9/n

Two bar charts side by side. Left: words in the US country narrative of the US State Department's Trafficking in Persons report, 2010 to 2026. The US narrative runs between about 5,500 and 7,000 words from 2010 to 2018, rises to 13,499 in 2021, and is 13,396 words in 2024. It falls to 9,283 in 2025 and to 3,250 in 2026, highlighted in blue: 76% shorter than in 2024. The US has been Tier 1 in every report since 2010. Right: federal funding for NGO victim services, as reported in the 2026 US narrative, fell from $82.3 million in FY 2024 to $18.9 million in FY 2025. Data processed and chart made by Claude Opus 5.5; checked by Richard W. Frank

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

Ghana, Israel, & Norway rose to Tier 1; Seychelles fell to Tier 2. Syria and PNG left Tier 3. Syria also came off the child soldier (CSPA) list and the state-sponsored trafficking list. The CSPA list shrank from 17 to 14 governments; Nigeria was added. 8/n

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

On TIP rankings, 35 countries are Tier 1, the most since 2018. The Tier 2 Watch List shrank to 18, its smallest in at least 20 years. Of the 25 WL countries in 2025, 10 moved up to Tier 2 in 2026, 13 stayed, and 2 (Hong Kong, Rep. of Congo) fell to Tier 3. 7/n

Line chart of the number of countries on the TIP report's Tier 2 Watch List, 2006 to 2026 (2008 not coded). The list starts at 32 countries in 2006, peaks at 58 in 2010, and holds between 38 and 45 from 2011 to 2021. It then declines to 34 in 2022, 26 in 2023, 32 in 2024, and 25 in 2025, reaching 18 in 2026, the lowest in the series. Data processed and chart made by Claude Opus 5.5; checked by Richard W. Frank.

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

Victim-centred language is shrinking too: "survivor" now appears in 23% of narratives (43% in 2024) and "trauma-informed" in 22% (45% in 2025). New emphases: "scam" appears in 43% (22% in 2024) and "supply chain" in 27% (13%). #ForcedLabor 6/n

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

The vocabulary shift began in 2025 and continued in 2026. "LGBTQI+" was in 43% of 2024 narratives and in none since. "Irregular migration" fell from 22% to 1% of narratives, while "illegal" migration or entry rose from 7% to 23%. 5/n

Dot chart of the share of TIP report country narratives mentioning selected terms at least once in 2024, 2025, and 2026. LGBTQI+: 43% in 2024, 0% in 2025 and 2026. "PRC" (replaced by "China"): 44%, 0%, 0%. "Irregular migration": 22%, 2%, 1%. "Trauma-informed": 38%, 45%, 22%. "Survivor": 43%, 40%, 23%. "Refugee": 57%, 59%, 44%. "Illegal" migration or entry: 7%, 25%, 23%. "Supply chain": 13%, 13%, 27%. "Scam": 22%, 29%, 43%. Data processed and chart made by Claude Opus 5.5; checked by Richard W. Frank.

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

The narratives are also more formulaic. In 2026, 6% of the narrative text is in sentences repeated word for word in 10 or more countries. That is double 2025 (3%) and roughly 3x the 2015-24 average (2%). 4/n

Line chart of the share of TIP report narrative text in sentences repeated word for word in 10 or more country narratives in the same year, 2004 to 2026. The share stays between about 1% and 2.6% from 2004 to 2024, then rises to 3.0% in 2025 and jumps to 6.0% in 2026, the highest in the series. 2008 is omitted. Data processed and chart made by Claude Opus 5.5; checked by Richard W. Frank.

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

What was cut? In 2025, every country narrative (outside four special cases) included prioritised recommendations to the government. In 2026, none do. Protection sections also lost (on average) 41% of their length, prevention 40%, and trafficking profiles 32%. 3/n

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

The average narrative is 2,175 words, down 38% from 2025 and 45% from 2024. This is the shortest since 2017. 2/n

Bar chart of the average number of words per country narrative in the US State Department's Trafficking in Persons report, 2001 to 2026. Length grows steadily from 263 words in 2001 to about 1,700 to 1,900 words in the early 2010s, then rises sharply to a peak of 3,962 words in 2024. It falls to 3,528 words in 2025 and to 2,175 words in 2026, highlighted in blue: 45% shorter than in 2024 and the shortest since 2017. Data processed and chart made by Claude Opus 5.5; checked by Richard W. Frank.

100

Rich Frank @richwfrank.bsky.social · 10/10/2026

The 2026 #HumanTrafficking #TIPreport is (finally) out. Given the delay and (waves hands wildly), I was keen to compare 2026 with previous years. A quick thread of initial findings, with data work and figures by Claude Opus 5.5 and checked by me (details at end). 🧵 1/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

The article is open access here: doi.org/10.1080/1036... The data and replication files are available on my Dataverse: dataverse.harvard.edu/dataverse/ri.... If you are part of rethinking your own department's curriculum, the program-level coding might be a useful benchmark. 14/14

doi.org

Undergraduate political science curriculum design in five democracies

Most published research on political science curriculum design focuses on the United States, leaving unclear whether findings generalise internationally. This paper analyses 218 programs at 140 uni...

021

Rich Frank @richwfrank.bsky.social · 22/09/2026

My takeaway, especially for #AusPol colleagues: Australia's relatively unstructured approach probably reflects path dependence and institutional autonomy more than any pedagogical consensus. This matters as debates about quality assurance (and AI use) continue. 13/n

120

Rich Frank @richwfrank.bsky.social · 22/09/2026

Student body size and international enrolments matter less than I expected, although more international students are associated with fewer capstone requirements. 12/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

PhD-granting universities are also much more likely to require methods training. Holding other factors at their means, it's 80% vs 46%. Larger faculties also predict methods requirements, and higher-ranked universities are somewhat more likely to require a capstone. 11/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

So what might help explain the cross-country differences? I test six hypotheses about institutional characteristics. The strongest and most consistent predictor is how much of the degree is required coursework. Programs that prescribe more courses are more likely to adopt the full sequence. 10/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

30% of degrees require no explicit methods class at all. In Australia, only 7 of 31 programs require one. Where methods are required, they are usually general intro/research design courses, not stats. 9/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

Finding #3 is one I found interesting, given that this project started while designing an ANU undergrad methods course. Put politely, methods requirements are uneven. 8/n

110

Rich Frank @richwfrank.bsky.social · 22/09/2026

Finding #2. Intro to IR is the most frequently required first-year class (144 of 218 programs), followed by Intro to Politics (102). Each country has its own trends, though. Six UK degrees, for example, require a first-year US politics course. 7/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

My (untested) guess is that the UK's national benchmark for Politics and IR degrees, which emphasizes building up to independent research, does a lot of the work here. Unsurprisingly, national frameworks seem to matter. 6/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

Finding #1. The sequential model is far more common *outside* the US. 37% of programs in my sample (81 of 218) require all three elements. But that is mostly a UK story. 76 of those 81 programs are British. None are in Canada or New Zealand. 5/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

My benchmark is Wahlke's (1991) sequential model: introductory courses, research methods training, and a capstone. This model has been recommended to US departments for over 30 years, yet only 18% of US programs have adopted all three elements. 4/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

To find out, I coded every political science, politics, and IR undergraduate degree in Australia, Canada, Ireland, New Zealand, and the UK. 218 programs at 140 universities in all. 3/n

100

Rich Frank @richwfrank.bsky.social · 22/09/2026

Most of what we know about political science curricula comes from the US. My systematic literature review finds that over 2/3rds of recent studies look only at American programs. So…what happens everywhere else? 2/n

110

Rich Frank @richwfrank.bsky.social · 22/09/2026

I have a new open-access article in @ausjpolsci.bsky.social on how we design undergraduate political science degrees, and why they differ so much across countries. 🧵

120

Rich Frank @richwfrank.bsky.social · 25/08/2026

My takeaways for future research: - Focus more on election-specific triggers, not just slow-moving structural factors, - Better id perpetrators & targets, - stop generalizing across levels of analysis, - fix measurement. Findings from 157 measures from 40 sources are unlikely to generalize. 11/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

Sub-nationally the question shifts from where to whom. Specifically, younger citizens and female candidates face higher risk. But most studies do not test the whom. 81% of subnational models never say who the perpetrator was, and 97% of national models never say who the victim was. 10/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

Finding #3. Level of analysis matters more than we usually admit. Of the 13 robust predictors, only two (population size and domestic conflict) work at both levels. A competitive election predicts violence nationally but not sub-nationally. 9/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

What does predict violence are election-specific factors. Specifically, fraud, competitiveness, an executive contest on the ballot, and the election month. These are shorter-term and strategic factors. Unlike GDP or ethnic geography, they are things you could in principle intervene on. 8/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

Finding #2 is one I found quite surprising. The usual suspects fail. For instance, democracy does not predict election violence, whether measured as a level, a dummy, or years democratic. Neither does GDP per capita, tested in 177 models across 25 articles. 7/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

Finding #1. Of 440 predictors, only 13 are both widely used and consistently significant. Nine national, four subnational. The strongest by far is previous election violence. Violence surrounding the last election is the best single predictor of violence at the next one. 6/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

So, I ran the first meta-analysis of election violence. For every predictor in 581 models, I code how often it is statistically significant in the direction the author expected using an approach used in electoral turnout and electoral integrity meta-analyses. 3 quick findings… 5/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

Thus, the problem is not a shortage of research. It's the inconsistency. 65 articles, 157 different ways of measuring "election violence", drawn from 40 different data sources. Reading different slices of this literature can lead us to different answers as to what explains election violence. 4/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

Basically, I read every quantitative study focused on the causes of election violence published between 2010 and 2022, and I analyzed their 440 predictors. Surprisingly, only 13 of them are used more than twice and are consistently statistically significant. 3/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

Now that I have sent my conference paper out, I thought I might post a bit about the meta-analysis, because I (in a completely unbiased way, of course) think it is pretty cool and potentially useful. 2/n

100

Rich Frank @richwfrank.bsky.social · 25/08/2026

As I madly tried to finish a paper for APSA in the last few weeks (and a book project the paper springs from), I found myself repeatedly revisiting my recent election violence meta-analysis in JPR. 🧵

110

Rich Frank @richwfrank.bsky.social · 02/10/2025

I received some great feedback on my paper “Climate Change and FDI” from discussants and attendees and was able to catch up with my former SPIR colleague Woo Chang Kang!

030

Rich Frank @richwfrank.bsky.social · 02/10/2025

Thanks to Woo Chang Kang, Sung Eun Kim, the Department of Political Science & IR, the Peace & Democracy Institute, and Korea University for their hospitality and for hosting a terrific conference!

130

Rich Frank @richwfrank.bsky.social · 02/10/2025

Back from the inaugural “Conference on Peace and Democracy in Asia-Pacific and the World”, held at Korea University. A great way to celebrate 80 years since the founding of the Department of Political Science and International Relations.

130

Rich Frank @richwfrank.bsky.social · 08/05/2025

Enjoyed tonight's #SPIR roundtable at #ANU unpacking last week's election with Nick Biddle, Ian McAllister, @marijataflaga.bsky.social, and @palesl.bsky.social. Ballots still being tallied, but the expert analysis is already in full swing! #AusPol

170