# Proposed protocol for an archive of prayers for rain

Version 1.0, 4 September 2026. This is a draft for registration before analysis. It has not been registered in a study registry and reports no completed study. It was prepared with AI assistance; no cited researcher or author has endorsed it.

## Question and primary outcome

How often do documented public prayers for rain precede at least one day with precipitation above 5 mm in the following seven days? Compare that frequency with days under similar weather conditions without a documented prayer.

The primary estimate is the average difference between those frequencies, in percentage points. The analysis describes an association between documented events and weather. Ritual timing, forecasts and selective reporting can explain an association. The estimate does not identify a causal effect of prayer or measure participants' beliefs.

## Freeze the archive before inspecting weather outcomes

1. Fix the period as 1 January 2000–31 December 2025 and locations within Romania's current borders. Register the dated protocol version in a public repository before matching events to precipitation.
2. Freeze the archive list: Basilica, Doxologia and official websites of Orthodox dioceses based in Romania. Save the domains and accessible years. This denominational boundary defines the archive, not all practices in Romania.
3. Use the same search expressions with and without diacritics: “rugăciuni pentru ploaie”, “rugăciune pentru ploaie”, “procesiune pentru ploaie”, “încetarea secetei”, “vreme de secetă”. Inspect archive categories too. Record the query, domain, period, search date and pages examined.
4. Close collection after every domain–year–query combination and the selected categories have been examined. Do not stop at a convenient event count. Retain a log of inaccessible pages.
5. Two coders independently assess each report without seeing measured precipitation after the event. Resolve disagreements before joining the datasets. Preserve initial coding and the reason for each resolution.

## Registry inclusion

Include a public prayer or procession for rain when its report identifies a locality and a calendar date. Keep announcements separate from reports that the event occurred. Announcements without confirmation remain in the registry but do not enter the main meteorological sample.

Exclude general prayers without a dated local event, prayers to stop rainfall and reports outside the study period. Retain each exclusion reason. Caloian, Paparuda and other practices require a separate registry and search protocol.

Assign one event identifier to an event covered by several publications. For the same locality, repetitions no more than 14 days apart form one campaign. The main analysis uses its first confirmed ceremony; retain every repetition. This rule can merge distinct events. Repeat the analysis with 7-day and 28-day limits as sensitivity analyses.

Store event and campaign identifiers, locality, coordinates and their source, event date and its precision, time when known, publication date, URL, access date, source type, announced/confirmed status, practice, stated reason, repetitions, weather claims and inclusion decision. Code reports published before and after the event separately. An absent article does not establish an absent ritual.

## Match meteorological observations

Use GHCN-Daily PRCP and preserve its version, files, flags and SHA-256 hashes. Apply the calculator's quality rules. Select the nearest eligible station within 30 km and no more than 200 m difference in elevation. These limits are protocol choices. Keep events without a suitable station in the registry without a meteorological outcome.

Let D denote the ceremony date. Calculate prior dry duration using observations through D−1. The primary outcome is at least one strict exceedance of 5 mm on D+1 through D+7. Require 28 complete following days to retain one cohort for secondary windows of 1, 14 and 28 days. Exclude the ceremony day from the outcome. Record the daily observation closing time; flag possible date misalignment when it is unknown.

Secondary thresholds are strictly >1 and >10 mm. No threshold automatically marks the end of agricultural drought. Repeat the analysis for April–September only as a check on season and solid precipitation.

## Comparison dates

For each campaign, use all eligible dates at the same station, in the same calendar month, from the other years of 2000–2025. Match prior dry duration in categories 0–6, 7–13, 14–29, or at least 30 days. Exclude dates within 28 days of any known ceremony in that locality. A date without a documented ceremony may contain an unrecorded event. This comparison retains the month but not the particular weather of the ceremony year.

Include a campaign in the comparison only if it has at least five comparison dates. Calculate the mean outcome on those dates and subtract it from the campaign outcome. Give every campaign equal weight. Report unmatched campaigns and their characteristics separately.

A secondary analysis includes only events and comparison dates with archived forecasts issued before the assessed date. Also match forecast probability in categories 0–20%, above 20–50%, above 50–80%, and above 80–100%. Never replace a missing forecast with one reconstructed from subsequent weather. If a forecast source cannot be fixed before analysis, register an amendment before inspecting this subset's outcomes.

## Reporting and verification

Report the complete registry, inclusion flow, distribution by year and locality, source coverage and primary difference, including zero or negative results. Do not estimate population prevalence of rituals from this news archive.

Use 10,000 resamples of calendar years with replacement. Apply shared year weights to campaigns and meteorological observations, then recalculate comparisons in each draw. Fix the random seed at 20260904. Report the 2.5th and 97.5th percentiles as an approximate interval, alongside year and campaign counts. With fewer than 30 matched campaigns or 10 years, publish counts and treat the comparison as exploratory. Dependence across years and selective reporting remain limitations.

Manually verify every event date, every station match and a random 20% sample of meteorological outcomes. Recalculate the primary difference in a separate implementation. Publish amendments, code and the data dictionary. Interviews about beliefs or intentions require a separate study with participant consent; this protocol does not authorize contact.

Starting sources: [QJE study and appendices](https://academic.oup.com/qje/article/141/3/2363/8676046), [NOAA specification](https://www.ncei.noaa.gov/pub/data/ghcn/daily/readme.txt), [Suceava example, 2015](https://basilica.ro/procesiune-pentru-ploaie-la-suceava/). These examples are already known; disclose them in the registration and do not treat them as cases discovered without prior knowledge.