Women raised runestones more often in Svealand than in Götaland – 31 against 17 percent

The women of the runestones have been a field of their own since the 1990s. Judith Jesch (1991) gathered what the inscriptions say about women, and Birgit Sawyer (2000) counted the sponsors province by province and showed that the share of women varies between provinces. Sawyer read the stones as documents of inheritance: whoever raised the stone presented herself or himself as the heir, and where women could inherit they appeared more often. Anne-Sofie Gräslund (1995) has discussed the same women in relation to Christianisation. The counts in that literature are raw shares. They take no account of the fact that the stones of Svealand are on average younger than those of Götaland, or that a few carvers account for many stones in Uppland.

The aim is to test whether the difference between Svealand and Götaland holds when the age of the stones and the carvers are held constant, with a model fixed before it was run, and to let it pass the same sensitivity analyses as every other association in the project. The study says nothing new about why women raised stones. It tests whether the regional difference is a finding to build on.

The hypothesis, the model and the rules of interpretation were written down in an analysis plan on 28 October 2026, before any model had been run. The plan concerned a larger study of who raised the runestones and for whom, with six tests; this is the first. The plan is time-stamped in the project's version control and has the SHA-256 checksum ea26a388379e29eb…, so that any later change would show (see Data and code).

The test reads in the plan: the outcome is at least one woman among the sponsors, among stones with at least one identified sponsor; the model is outcome ~ region + dating period + carver as a random effect; the measure is the odds ratio Götaland against Svealand. The six tests are corrected together by Holm's method. The rules of interpretation were fixed in advance: a test confirms when Holm p is below 0.05 and the direction is the same with p below 0.05 in each of the material's two halves; a pattern to test has Holm p below 0.05 but does not hold in both halves; an indication has only uncorrected p below 0.05.

Before the plan was locked I had seen one figure: 391 legible Swedish stones have a woman among the sponsors. No comparison between regions had been made. The plan says so.

The analyses in sections 5.1–5.3 are those the plan prescribes. The follow-up in 5.4 comes from a second analysis plan, written and locked on 29 October 2026 (SHA-256 6a03f6fd1a5f50f5…) after the main result was known. It is thus pre-registered but not part of the first test, and it is reported as a follow-up. No exploratory analyses are included.

• Source. The Scandinavian Runic-text Database (Rundata), Department of Scandinavian Languages, Uppsala University: transliteration, normalisation, translation, province, parish, carver and style group for each inscription. The material was frozen for the analysis on 29 October 2026.
• The stones. All Viking-Age runestones in the Swedish provinces: 2,320 stones. Runestones only; other objects with runes are not included. Skåne and Blekinge, Danish in the Viking Age, are counted as Denmark in Rundata and are not included.
• Selection. Stones whose text is preserved well enough for the sponsors to be read, and where at least one sponsor can be identified: 1,425 stones. Of these, 1,013 lie in Svealand and 370 in Götaland. 42 lie on Gotland or in Norrland; they enter the model as their own levels but are not compared.
• The regions by today's provinces: Svealand is Uppland, Södermanland, Västmanland, Närke, Gästrikland, Dalarna and Värmland; Götaland is Östergötland, Västergötland, Småland, Öland, Bohuslän, Halland and Dalsland.
• The outcome is that at least one woman stands among the sponsors. The sponsors' sex is read from Rundata's English translation by rules in four steps: first the relation (mother, daughter, wife; son, father, brother), then a lone sponsor's possessive in the formula (her son, his son), then a name lexicon built from the first two steps, and last the name's ending. The rules were checked against hand coding on a new sample of 40 stones the rules had never seen: 39 of 40 right. In the hand-coded samples together (80 stones that could be compared) the rule's yes was right 13 times of 13 and its no 67 times of 67. The errors sit where the text has gaps, and they miss women more often than they add them.
• Dating period by the stone's ornamental style (Gräslund 2006): before 1010, 1010–1040, 1040–1070, after 1070, and undated as a group of its own.
• The carver. The carver Rundata gives. Stones without a known carver are grouped by parish.

4.1 The model. Logistic regression with the R package mgcv (R 4.6.1; Wood 2017): the outcome against region and dating period, with the carver as a random effect. The random effect gives each carver a level of their own, so that a prolific carver's habits are not counted as a region's custom. The shares per region are the model's marginal effects, that is, the average over the stones as they are.

4.2 Correction and replication. Holm correction (Holm 1979) over the study's six tests. Internal replication: the material was divided into two halves parish by parish, with whole parishes dealt alternately within each province. The division was made once, before the study and the same for all the project's analyses, and the test must hold in each half on its own. In addition, five robustness checks prescribed by the plan: only stones with at least ten legible words, only stones where every sponsor's sex is determined, the model without the carver effect, Uppland only, and sons against fathers (the last applies only to the study's tests about the dead).

4.3 Sensitivity analyses. The project applies the same fixed series of sensitivity analyses to every association it reports, whatever study the association comes from: the raw model; the model adjusted for dating period; the carver as a random effect; the parish as a random effect, to see whether the difference is the peculiarity of a few parishes; each half on its own; and the model refitted without one province at a time (provinces with at least 30 stones). An analysis with the stone's location as a spatial surface is not made here, since the comparison itself is geographic. An analysis holds when p is below 0.05 and the direction is the same as in the adjusted model. When a province is removed, significance is required only where at least two thirds of the stones remain; otherwise the direction suffices.

Since the project tests many associations, the p-values are also adjusted together for all 48 associations reported (Benjamini & Hochberg 1995). The adjusted value is called q below. An association counts as confirmed when it holds in both halves and q is below 0.05. The model's assumptions are checked with simulated residuals (DHARMa; Hartig 2022). The errors in the sex determination are tested by refitting the model 200 times, with each stone's ”true” outcome drawn from the hit rates in the hand-coded sample.

The sensitivity analyses are computed on the 1,383 stones in Svealand and Götaland. The main model has 1,425, since Gotland and Norrland are included there as levels of their own. The sensitivity analyses use the project's common dating periods, so their figures can differ slightly from the main test's.

4.4 Deviations from the plan. None for the main plan. In the follow-up plan, one of the tests (a lone woman raises the stone) was first run with a variable that was missing from the data file, which gave zero stones. The variable was then computed exactly as the plan defines it and the test rerun. Nothing else was changed.

5.1 The main test. Raw, a woman stands among the sponsors on 324 of 1,013 stones in Svealand (32 percent) against 61 of 370 in Götaland (16 percent). With time and carver constant the shares are 31 against 17 percent. The odds ratio Götaland against Svealand is 0.46 (95 % interval 0.32–0.66), p < 0.001, Holm-corrected p < 0.001. In plain words: draw a stone at random in each region, and it is nearly twice as common that the Svealand stone has a woman among the sponsors; a swing that large from chance alone is too rare to reckon with. Half A: 730 stones, odds ratio 0.41 (0.24–0.69), p < 0.001. Half B: 695 stones, odds ratio 0.51 (0.31–0.85), p = 0.009. Same direction and significant in both halves: the test confirms by the plan's rule.

5.2 The plan's robustness checks.

• Only stones with at least ten legible words: 951 stones, odds ratio 0.51, p = 0.005.
• Only stones where every sponsor's sex is determined: 1,390 stones, odds ratio 0.46, p < 0.001.
• Without the carver effect: odds ratio 0.46, p < 0.001.
• Uppland only: cannot be tested, since the comparison concerns regions and Uppland lies wholly in Svealand.

5.3 The sensitivity analyses (1,383 stones, 385 with a woman among the sponsors). All 7 hold:

• Raw model: odds ratio 0.42 (0.31–0.57), p < 0.001.
• Adjusted for dating period: 0.46 (0.32–0.65), p < 0.001.
• Carver as a random effect (483 carvers and parish groups): 0.45 (0.31–0.64), p < 0.001.
• Parish as a random effect (469 parishes): 0.46 (0.32–0.66), p < 0.001. The difference is not the peculiarity of a few parishes.
• Half A (701 stones): 0.43, p < 0.001. Half B (682 stones): 0.49, p = 0.004.
• Without one province at a time (6 provinces tested): the direction holds without each of them, and no province turns it. Weakest without Uppland, when 643 stones remain: odds ratio 0.66 (0.43–1.03), p = 0.07. Without Uppland more than half of Svealand's stones are gone, and there only the direction is required.
• The difference in percentage points in the adjusted model: 31 against 17 percent, 14 points (probably between 9 and 20).
• The model's assumptions: residual distribution p = 0.59, overdispersion p = 0.71, remaining spatial pattern p = 0.67. Nothing amiss.
• Errors in the sex determination: the association holds in 100 percent of the 200 refits; median odds ratio 0.47, range 0.44–0.55.
• Adjusted for all 48 associations the project reports: q < 0.001. The association is one of 15 that hold in every sensitivity analysis.

5.4 Follow-up (second pre-registered plan). Five pre-set explanations of what the difference consists of, tested with the same model and the same rules of interpretation:

• Widows after the journeys – not supported. A woman raises the stone for her husband: 96 of 1,013 in Svealand against 23 of 370 in Götaland, odds ratio 0.82, no difference. Without the stones that mention journeys the difference remains and grows slightly (odds ratio 0.40, holds in both halves).
• The custom of listing the family – explains a fifth. With the number of sponsors constant: odds ratio 0.53. 20 percent of the difference on the log-odds scale disappears, below the plan's limit of half.
• A position of her own – not supported. A lone woman raises the stone: 137 of 1,013 against 44 of 370, odds ratio 0.77, no difference.
• The roles – this is where the difference lies. On stones for sons the mother appears on 107 of 195 in Svealand (55 percent) against 13 of 70 in Götaland (19 percent), odds ratio 0.23, Holm-corrected p = 0.015. In the halves p = 0.058 and 0.031, so by the plan's rule it is an indication, not a confirmed finding. On stones for fathers a daughter or wife appears on 66 of 415 against 9 of 117, odds ratio 0.37, also an indication.
• Christianisation – no association. A Christian feature on the stone (cross or prayer) and a woman among the sponsors: odds ratio 1.27, p = 0.16, and no difference between the regions.

What the finding says. The difference between Svealand and Götaland is not due to Svealand's stones being younger or to a few carvers there carving many stones. It remains when both are held constant, in each half and in every sensitivity analysis. It is a difference in custom, not in population: a woman is more often named among those who raised the stone in Svealand. It does not say that women there had more power or property.

Against earlier research. Sawyer's (2000) observation that the share of women varies between provinces thus stands also when age and carver are held constant. The follow-up points the same way as her interpretation. The women of Svealand were not more often widows and not more often lone sponsors; the mother appeared on her son's stone together with the father and the siblings. That fits a rule of inheritance under which a mother could inherit from her child better than it fits the journeys or Christianisation (cf. Gräslund 1995). But the role finding is an indication by the plan's rule, and the interpretation is an interpretation. The report rests on the main finding.

Weaknesses, in order of weight.

1. Uppland carries much. Without Uppland the odds ratio is 0.66 and the interval reaches above 1. The direction holds in every province, but the certain difference is to a large degree a difference between Uppland and Götaland.
2. Sex is determined by rules. The rules err on about one stone in twenty, mostly where the text has gaps, and they miss women more often than they add them. If gaps were more common in Götaland that would strengthen the difference on false grounds. With only stones of at least ten legible words the odds ratio is 0.51, so that does not explain the finding.
3. The dating rests on ornamental style. Undated stones are a group of their own. Another dating could move the figures slightly but hardly the direction, since the raw and the adjusted models nearly coincide.
4. Region is a coarse division. That the difference is largest between Uppland and Götaland is already visible. What happens in the border provinces is not tested here.

What would overturn the finding. A hand coding of sex on a large sample showing more errors in one region than in the other, or a recomputation with another dating in which the adjusted model no longer holds.

A woman stands among the sponsors on 31 percent of Svealand's runestones with a known sponsor against 17 percent of Götaland's, with age and carver constant, and the difference withstands every sensitivity analysis. It is due to how the family remembered its dead – the mother on her son's stone – and is therefore a finding about commemorative custom, not a measure of women's standing. The next question is why the custom differed, and it needs other material than the inscriptions' own words.