~/projects/general-conference/events
Responding to the world
Wars, pandemics, assassinations, a moon landing, a court ruling: when the world moves, how soon does the pulpit answer, how loudly, and for how long? Thirty-one events, each with a date and a vocabulary, measured against the share of talks that use that vocabulary in the conferences before and after.
What stands out
- COVID-19 drew the loudest response. The share of talks mentioning it rose by 44.19 points over baseline to 46% in 2020-10; Great Depression (34.38 points) and Pearl Harbor (18.76 points) follow.
- The pulpit usually answers at the next conference. Median lag was 4 months for events before 1970 and 2.5 since, across 14 and 12 external events with a response. Influenza pandemic was answered within 0 months.
- First World War stayed in the preaching longest, above baseline for 6 consecutive conferences.
- Some events pass without a word. Civil Rights Act, Watergate, Obergefell, War in Ukraine lifted the share of mentioning talks by under one point.
- When the world shocks, the vocabulary turns to war, global, nations, forces, united states. A classifier told the responding talks from the talks before with AUC 0.83 after the event words themselves were removed.
Loading the study…
Model card
- Events
- 31 events chosen by hand, each with a date and a regular expression for its vocabulary (for example terrorist, twin towers and September 11 for 2001). Twenty-eight are external, three are the Church's own announcements for comparison.
- Measure
- For each event, the share of talks (200 words or more, April and October conferences) matching the lexicon in each of the four conferences before and the six after. Baseline is the mean share before; lag is months from the event to the first conference after it with a matching talk (a conference is dated to its first Saturday); lift is the peak share after minus the baseline, in percentage points; a conference counts as above baseline while the share exceeds one and a half times the baseline plus one point.
- Classifier
- Positives are the lexicon-matching talks in the two conferences after each external event; negatives are the talks in the four conferences before. Masked TF-IDF, logistic regression with balanced weights, five folds grouped by speaker. Its strongest terms are reported after removing anything an event lexicon could match.
- Known failure modes
- A lexicon can miss an oblique response (a talk on fear after a war) and can catch an unrelated use (recession as a word, iraq in a scripture). No lag means no match within six conferences, not necessarily silence. The 1942–1970 and 1880–1941 texts are transcriptions and OCR, and a few events fall where coverage is thin.
Methods and data for the corpus as a whole.