~/projects/general-conference/compared-with-byu/lead-lag
Which pulpit moves first?
Does the campus anticipate the Church, or follow it? For 129 tracked phrases and 1,996 common words with an adoption year in both corpora since 1970, and for the thirty joint topics by decade, this study asks which pulpit took up each term first and by how long.
What stands out
- Conference usually says it first. Of 129 phrases, BYU adopted 40% first and conference 50%; the median lead is -1 years, and the cross-correlation of the two series peaks at a median lag of -1 years.
- For common words the picture is the same: BYU first on 24%, conference first on 61%, median lead -1 years across 1,996 words.
- “doctrine of christ” reached BYU 31 years before conference (1981 against 2012); “peacemakers” reached conference 20 years before BYU (1982 against 2002).
- At the grain of topics, BYU peaked first on 10 of thirty, conference on 8, the same decade on 12. Decades are coarse; the phrase series are the finer instrument.
- Two caveats shape every number. The BYU corpus is a tenth the size, so a rare phrase is detected later there by chance alone, which biases leads toward conference; and the two pulpits share speakers, so an apostle can carry a phrase from one to the other within a year.
Loading the study…
Model card
- Data
- 4,203 conference talks and 1,950 BYU speeches of 300 words or more from 1970 to 2026. Terms: the union of the two phrase studies' tracked phrases and replacement pairs, plus the 2,000 most common content words across both corpora.
- Adoption and lead
- For each term and corpus, the share of talks per year that use it, smoothed over five years. The adoption year is the first year the smoothed share reaches half its eventual peak after at least two talks have used the term. Lead is conference's adoption year minus BYU's. Terms adopted in both corpora within two years of 1970 are set aside as already in use. Terms need ten or more talks in each corpus.
- Cross-correlation
- The Pearson correlation of the two smoothed series at lags from −10 to +10 years; the best lag is reported for terms whose best correlation reaches 0.3. Topics use the joint thirty-topic model's decade mixes and compare peak decades.
- Known failure modes
- Sample size differs by an order of magnitude, which delays detection in the smaller corpus. Shared speakers blur the question. A term that appears in a quotation of the other pulpit counts as use. Half-of-peak adoption is sensitive to a late peak.
Methods and data for the corpus as a whole.