Two series of short videos about amorem. Every number on screen is computed by running the package; the narration is in English.
Night Sky Atlas
One real dataset per card, seen as a night sky: actors are stars, each event is a streak of light between two of them, and time runs along the horizon. Each card ends on one finding computed with amorem. The cards stand alone and can be watched in any order.
No. 1
Over Washington
Every leader-to-leader visit of the United States, 1945–2019. A country whose leader has just visited Washington is about twice as likely to be the president's next stop (2.08 times the rate, 95% interval 1.60 to 2.69).
Try it in R: amorem::us_leader_visits
No. 2
Over a Classroom
One 44-minute US high-school class period, 16 October 1996: 20 people and 511 turns of talk. A turn back to someone who spoke to you in the last minute comes at 9.69 times the rate of an otherwise similar pair where that did not happen (95% interval 7.95 to 11.83).
Try it in R: amorem::classroom_events
No. 3
Over the football world
Men's international football, 1872 to 2026: 49,945 matches between 337 teams. A pair that met a year ago plays again at 2.3 times the rate of a pair that last met ten years ago (95% interval 2.2 to 2.5).
Try it in R: amorem::football_results
No. 4
Over Europe
18,703 flights the OpenSky Network recorded between 60 European airports in April 2020. A route flown the other way in the last 12 hours carries the next flight at 3.83 times the rate of one flown the other way only earlier (95% interval 3.64 to 4.03).
Data: OpenSky Network, CC BY 4.0
No. 5
Over the Office
The first 60 days of e-mail in a manufacturing company in Poland, 2010: 14,962 e-mails among 167 employees. Copies flow to someone who has already written to the sender at 7.04 times the rate of a pair where they have not (95% interval 5.61 to 8.82).
Try it in R: amorem::radoslaw_email
No. 6
Over Rome
Italy's diplomatic visits, 2000–2023. Next to come to Rome: a country Italy had visited two times since 2000 comes at 2.02 times the rate of one it had not visited (95% interval 1.78 to 2.28).
Data: Corda & Casiraghi, Italy's Diplomatic Visits, CC BY 4.0
No. 7
Over Copenhagen
28 days of calls and texts among 608 students of the Technical University of Denmark. 20.9% of all calls (754 of 3,600) go back to someone who called the caller in the hour before.
Try it in R: amorem::copenhagen_calls
No. 8
Over the Earth
11,730 first records of alien species in 259 regions of the world, 1880 to 2005. A species already in 4 regions reaches its next region at 1.97 times the rate of a species in 1 (95% interval 1.76 to 2.20).
Try it in R: amorem::alien_invasions
No. 9
Over Capitol Hill
Every bill of the 118th US Congress, 2023 to 2025: 16,213 bills and 162,424 cosponsor signatures. 21.7% of the signatures cross party lines (35,300 of 162,424): 20.4% in the House, 28.2% in the Senate.
Try it in R: amorem::congress_bills_118
No. 10
Over London
19,120 bike journeys between London's 60 busiest docking stations, February 2026. Riding back along a journey made in the last 12 hours comes at 2.42 times the rate of riding back along one made only earlier (95% interval 2.31 to 2.54).
Data: TfL cycle hire journeys, February 2026. Powered by TfL Open Data.
No. 11
Over Lichess
7,025 rated games between the 60 most active Lichess players, January 2013. A pair that met in the last hour is more than ten times as likely to play next as one that met only earlier (seed 1: 17, 95% interval 12 to 24; five random draws: 14 to 21).
Data: Lichess open database, CC0
No. 12
Over Africa
2,959 foreign-minister visits between 170 countries, reported by 30 African ministries, 2010 to 2022. A minister whose counterpart came in the past year visits back next at 1.38 times the rate (95% interval 1.19 to 1.59).
Data: Burman Roy et al., CC BY 4.0
Time, Rate, History
Eleven short lectures that build relational event models from the ground up: events, rates, history as a covariate, the partial likelihood, case-control sampling, smooth effects, goodness of fit and global covariates. They are meant to be watched in order. They also play on the Luche channel on YouTube.
Episode 1
The data are events, not a network
A relational event model starts from a log: who acted towards whom, and when.
Episode 2
Every pair has a clock
Give every ordered pair its own exponential clock.
Episode 3
Coefficients multiply rates
Each pair's rate is a baseline times exp(beta times x).
Episode 4
History becomes a covariate
Endogenous statistics turn a log's own past into covariates: reciprocity, transitivity, recency and their variants.
Episode 5
Which dyad fired?
Given that an event happened, which pair was it? The partial likelihood asks only that question.
Episode 6
Compare with a few who did not
The partial likelihood sums over every pair at risk.
Episode 7
How many controls?
How many non-events should be sampled per event? On a fixed history, more controls narrow the spread of the estimate without moving its centre: from one control to ten, the spread falls about three times.
Episode 8
Ghosts read the past but never write it
A sampled non-event did not happen, so it must read the history to get its covariates but must never write into it.
Episode 9
With one control, a regression on differences
With one control per event, the conditional logit becomes a logistic regression without intercept on case-minus-control differences.
Episode 10
Is the model wrong?
Rank the candidate models by AIC on one shared case-control sample, then check the winner.
Episode 11
Why the weekday cancels
A covariate shared by every pair at the same moment, such as the weekday or El Nino, takes the same value for an event and its controls, so it cancels like the baseline.