Scotland accused 3,840 people of witchcraft between 1563 and 1736, most of them in a handful of sudden panics. I mapped where they lived, counted who they were, and tested whether the panics followed cold, hard years.
In the winter of 1590, James VI sailed home from Denmark with his new queen through storms, and a group of men and women from East Lothian were soon accused of raising them. That trial, at North Berwick, opened the first of Scotland's great witch panics. Over the next century and a half the country accused 3,840 people of witchcraft, most of them in a few terrible bursts.
Historians of continental Europe, notably Wolfgang Behringer and Emily Oster, have argued that witch hunts tracked the Little Ice Age: cold years ruined harvests, and frightened communities looked for someone to blame. It is a powerful idea, and Scotland is a good place to test it, because the Survey of Scottish Witchcraft records nearly every known accusation with a date and often a place. So I put the Survey's 3,413 cases next to five centuries of reconstructed European temperature, and asked three questions. Where were the accused? Who were they? And did the panics follow cold, hard years?
The short answer to the last question: not in any way I can detect. The panics were sudden and local, the people caught up in them were overwhelmingly women, and the cold years were not the witch years.
The Survey lists accusations from 1563, when the Scottish Parliament made witchcraft a capital crime, until 1736, when the law was repealed. In a typical year there were only a handful: the median year had 5 people accused, and 33 years had none at all. Then, every so often, the number exploded. The worst year, 1649, saw 565 people accused. The five panics marked below cover 11 years, yet they hold 2,181 of the 3,821 dated accusations, or 57%.
The five panics are the ones historians usually name: 1590 to 1591, 1597, 1628 to 1631, 1649 to 1650 and 1661 to 1662. The data also show smaller surges outside them: 134 people in 1643 and 83 in 1644, and 170 in 1658 to 1659. After the Restoration panic the hunt faded. Only 13% of accusations came after 1662.
Each circle is a place where named accused lived, sized by how many; the dark centre shows how many of them are recorded as executed. Drag the slider to move through the years, press play to watch the panics flare, or jump straight to one. The density layer shows where accusations were thickest. The gold diamond in Lancashire marks Pendle, for comparison (see section 05).
The pattern is unmistakable. Accusations piled up in the Lowlands around the Firth of Forth: 54% of mapped accused lived within 50 km of Edinburgh. Three counties, Haddington (East Lothian), Fife and Edinburgh (Midlothian), account for 42% of all accused with a recorded county. The single place with most accused is Haddington, with 218. The Highlands and Islands are almost empty in the record: only 21% of mapped accused lived north of the Tay.
The panics were not national waves. Each one had its own heartland.
The 1597 hunt was centred on Aberdeenshire and Fife, and only 31% of its mapped accused lived within 50 km of Edinburgh. The 1649 to 1650 and 1661 to 1662 panics, by far the largest, were concentrated in the Lothians. The cards also show how warm or cold each panic's first year was. Only two of the five onset years were colder than the average year; the 1661 to 1662 panic began in the 172nd coldest of 174 years, which is to say one of the warmest.
If cold drove the hunts, the panic years, or the years just before them, should be colder than usual. I tested this several ways, each designed to respect two awkward facts: there are only five panics, and both temperature and accusations run in streaks, which makes chance coincidences look stronger than they are.
The chart below stacks up the five panics and averages the temperature from five years before each onset to five years after. The grey band shows where 95% of averages fall when I pick five random years in the window instead, repeated 20,000 times. If the cold came first, the line should dip below the band just before the onset.
For annual temperature, the onset years averaged +0.04 °C against the window mean, well inside the range chance produces (−0.34 to +0.27 °C). Winters at onset were, if anything, mild (+0.47 °C). There is one hint: the summer before a panic averaged −0.28 °C, and 4 of the 5 were below average. On its own that would pass a conventional test (p = 0.033). But I looked at five seasons, two measures and three windows, 30 comparisons in all, and five random onsets turn up a result at least this striking somewhere among them 25% of the time. A cool summer before the panic is worth a closer look with local Scottish evidence, not a finding.
Across all 174 years, I correlated the number accused with temperature in the same year and one and two years before, for each season, both as raw anomalies and as year-to-year swings (the temperature with its slow 31-year trend removed). To judge each correlation fairly I slid the temperature series against the accusations by every offset of at least ten years, which keeps the streakiness of both series, and asked how often a shifted series correlates as strongly.
None of the 30 correlations reaches p < 0.05; the smallest p is 0.13. The strongest headline candidate, annual temperature in the same year, gives a correlation of −0.08 (p = 0.27). The panic years were in fact slightly warmer than the rest: +0.15 °C against −0.01 °C on average, relative to the window mean.
The 17 coldest years of the window (the coldest tenth) saw 215 accusations, 5.6% of the total, less than the 9.8% they would get if accusations were spread evenly. Only one panic, 1628 to 1631, began in or just after one of those cold years.
The Little Ice Age argument is often made at the scale of decades rather than single years, so I also compared the ten-year totals with the ten-year mean temperature. The colder decades did see somewhat more accusations (correlation −0.21), but with only 16 decades the chance of a link this strong from random temperature series with the same rhythms is 0.15. Leave out the eighteenth century, when accusations had already died away, and the sign flips (+0.15).
Not a small one. With five panics, the onset years would have needed to average about −0.28 °C (annual) to stand out at the 5% level, against a typical year-to-year spread of 0.36 °C. So this is not proof that weather never mattered. It is evidence that Scotland's great panics were not set off, in any consistent way, by the kind of Europe-wide cold the reconstruction records.
Of the 3,212 named accused, 2,697 were women and 468 were men (47 unknown): 85% women. That share barely moved between panic years (86%) and quieter years (85%). Where marital status is recorded, 575 were married and 140 widowed. Where an age is known or estimated (166 people), the median is 45. Most whose social standing is recorded were of middling or lower status.
The records of what happened to the accused are badly incomplete. For 2,687 of the 3,212 named accused no outcome survives. Of the 525 with a known outcome, 248 (47%) were executed, most often strangled and then burned (124 cases where the method is recorded). The share executed among known outcomes was higher during panics (57%) than between them (41%). The Survey records torture for 52 people, most often sleep deprivation, and pricking (searching for the "witch's mark" with a needle) for 105. These are floors, not totals.
Lancashire's Pendle witches, tried in 1612, are England's most famous witch trial: 12 people accused, 10 hanged. In Scotland that same year the Survey records 10 people accused, a quiet year. English law and courts produced isolated, notorious trials. Scotland, where local church courts gathered accusations and the Privy Council issued commissions to try them, produced panics that could sweep up hundreds in a season.
Accusations. I read 38 tables from the Survey of Scottish Witchcraft's database export. The text files have no header row, so the column names are inferred: I matched each column I use to the Survey's fields by checking its contents. Each case is one row. Named cases count as one person; a few cases describe unnamed groups ("3 unnamed witches"), which I count by the number given (628 people in 201 cases, 109 of them recorded as three). I date each case by its case date, falling back to its start date; using start dates instead changes almost nothing (correlation 0.995 between the two annual series). 14 cases have no date. Outcome is the most serious result in any linked trial record: execution if any trial records one.
Places. The Survey gives settlements, parishes and counties but no coordinates for most accused. Wikidata holds an item for nearly every named accused, with the Survey's reference and a place of residence that has coordinates, which I joined on the reference. These are place centres, and I have not checked every Wikidata residence against the Survey, so any errors there carry through.
Temperature. The temperature series is a reconstruction for European land as a whole (25°W to 40°E, 35°N to 70°N), built from early instrumental records, documentary sources and natural proxies. It is not Scottish weather. Scotland sits at its north-western edge, and a cold year in central Europe could be ordinary in Fife. Winter is December to February, labelled by the year of January. A Scottish series this long and detailed does not exist, as far as I know, which is a real limit on this test.
Tests. All p-values account for autocorrelation: circular shifts of at least ten years for the annual correlations (155 shifts), random sets of five onset years from 1568 to 1731 for the event study (20,000 draws), and random-phase surrogate series with the same spectrum for the decadal test. I report the family-wise check for the event study because I tried 30 combinations. The five panics were defined by historians who already knew the counts; they are not tuned to temperature.
Respect. These are real people, most of them women, many of them killed. I have shown their names as the Survey records them, with links to their case pages, because a name is more than a dot.
Other history data studies: Is Domesday destiny? · Within Sight · The Triskell Boom