January Isn't Divorce Month: What 11 Years of Wikipedia Data Actually Shows
The famous January divorce spike mostly disappears once you control for Wikipedia's own seasonal traffic. What's left is a real, large, unexplained collapse every July — and we're upfront about not knowing why.
StoriesFly Team
This is the first piece in a research series where every claim comes with its source and every dataset is published alongside the article, so you can check the work rather than take it on faith.
Before anything else, the limitation: everything below comes from Wikipedia pageviews — how many people read articles like "Divorce" or "Breakup" in a given month, in a given language. That measures interest, curiosity, or someone Googling their own situation at 1 AM. It does not measure how many people actually divorced or broke up that month. Keep that distinction in mind for every number that follows.
With that said, here's the headline: raw pageviews for English Wikipedia's "Divorce" article do rise 4.6% every January. But English Wikipedia as a whole runs 8.2% above its own yearly average every January — everything gets read more in January, divorce included. Once you divide out that site-wide surge, interest in "Divorce" doesn't spike in January. It falls to 3.3% *below* its yearly average.
The popular belief and the controlled data point in opposite directions.
The belief, and where it actually came from !July deviation in interest in the Divorce article across eight Wikipedia language editions, sorted. Turkish, Indonesian, Spanish, English, Portuguese and German all fall in July; Russian and Japanese rise.
"January is divorce month" is a real claim with a real paper trail — it's just not the one people think. Sociologists Julie Brines and Brian Serafini, then at the University of Washington, pulled divorce filings from Washington state courts covering 2001 to 2015 and presented the results at the American Sociological Association's 2016 annual meeting: filings peaked twice a year, in March and in August, not in January. Their explanation was that January is when couples decide, but spring and late summer — after the holidays clear and after summer vacation ends — are when the paperwork gets filed.
That's actual legal filings in one U.S. state. What we're measuring is Wikipedia reading behavior, worldwide, in eight languages. They're different instruments pointed at a related question, and it's worth saying plainly that they don't agree with each other on *when* the peak lands, either — our English-language data, once controlled, peaks in November and October, not March or August. Two independent datasets can each debunk "January" without agreeing on a replacement. That's normal, and it's a reason to hold the replacement loosely too.
What "controlling for the baseline" means, concretely
Wikipedia isn't read at a flat rate all year. Every article, on every topic, gets a lift every January — more people online after the holidays, New Year's resolutions driving searches of every kind. If you don't account for that, every single topic on Wikipedia looks like it "spikes in January," because the whole site does.
So for each article, we divided its monthly views by that Wikipedia edition's *total* monthly views, then expressed the result as a percentage deviation from the article's own yearly average. The same calculation was run without that adjustment, so the two numbers can sit side by side.
| Edition — topic | January, raw | January, controlled | |---|---|---| | English — Divorce | +4.6% | -3.3% | | English — Breakup | +6.3% | -1.8% | | German — Divorce | +18.3% | +2.3% | | Spanish — Divorce | -23.5% | -20.7% |
Two things stand out. First, both English topics flip sign once controlled — divorce and breakup are separate articles with independent readerships, and both tell the same story. Second, Spanish Wikipedia never showed a January spike in the first place, controlled or not: January is that edition's *lowest* month for "Divorcio," at -20.7%. Nowhere in this dataset does January end up as a real, controlled peak for an editorially independent topic. For more details, see our guide on view stories anonymously. For more details, see our guide on how long Instagram stories last.
Controlled, English "Divorce" actually peaks in November (+11.8%) and October (+10.2%), with the low point in June (-11.3%). "Breakup" peaks in March (+10.9%) and troughs in July (-13.0%). Different months, same conclusion: not January.
The part that's real: a July collapse
Related tools
Here's what doesn't go away when you control for it. Every edition we pulled shows a sharp drop in "divorce"-topic reading somewhere around July, and in most of them, July is the single lowest month of the year:
| Edition — topic | July, controlled | Is July this edition's yearly low? | |---|---|---| | English — Divorce | -10.1% | close (June is slightly lower) | | English — Breakup | -13.0% | yes | | Spanish — Divorce | -15.4% | no (January is lower) | | Portuguese (Brazil) — Divorce | -8.4% | no (December is lower, -15.6%) | | Indonesian — Divorce | -24.2% | yes — deepest drop in the whole dataset | | Turkish — Divorce | -30.1% | yes | | German — Divorce | -5.0% | yes | | Japanese — Divorce | +14.4% | no — July is this edition's peak |
The obvious explanation is "summer, people are on vacation and happier, fewer breakups." That explanation cannot survive contact with two rows in this table. Indonesian Wikipedia's readership is equatorial — Indonesia doesn't have a summer to be happy in, and July is not meteorologically distinct from March there. It still shows the single largest drop in the entire dataset: -24.2%. And Portuguese Wikipedia's readership is dominated by Brazil, where July is the middle of winter, the opposite season from the "everyone's relaxed and tanned" story — and it still dips in July, just not as far as its own December low.
Japanese Wikipedia is the other wrinkle: it's the one edition here where July is the peak, not the trough, for "divorce" reading. Whatever mechanism produces a July collapse in seven of these eight series, it isn't universal, and Japan is the live counter-case, not a rounding error.
One more honesty check: Russian Wikipedia's "Divorce" article only accumulated 15,923 total views across the full 11-year window we pulled — tiny next to English's 3.75 million. Its month-to-month swings (a reported +126% in June, for instance) are almost certainly just small-sample noise, not signal. We're not using it to argue anything either way.
What we don't know
We do not have a confirmed mechanism for the July trough, and we're not going to manufacture one. Here are the candidate explanations, labeled honestly as untested against this specific data:
- Academic and holiday calendars. Many countries cluster school breaks and family holidays around July, which could plausibly suppress searching for heavy topics generally. Untested here — we haven't checked whether *other* serious topics (illness, unemployment, grief) show the same July dip, which would support a general "people search less about hard things in midsummer" pattern rather than anything divorce-specific.
- General summer information-seeking patterns. People may simply read less reference material of any kind in July, and the "controlled" figure — which nets out the *whole wiki's* seasonal traffic — might not fully capture a topic-category effect (serious/informational topics vs. entertainment) if that category dips harder than the site average. We haven't tested this.
- A measurement quirk in the pageview pipeline. Mobile-vs-desktop mix, regional server routing, or some other artifact of how the Wikimedia API aggregates July traffic specifically. We consider this the least likely of the three, but we haven't ruled it out, and we'd rather say so than pretend the other two are proven.
Resisting the pull to pick one of these and present it as the answer is the point of this series. We found a real, large, cross-linguistic pattern. We did not find why it happens.
Method
- Source: Wikimedia REST Pageviews API (https://wikimedia.org/api/rest_v1/), all-access, user-agent traffic only — bots are excluded by the API itself, not by us.
- Coverage: July 2015 through July 2026 (11 full years) across eight Wikipedia language editions: English, Spanish, Portuguese, Indonesian, Turkish, Japanese, Russian, and German.
- Control calculation: for each article, monthly views were divided by that Wikipedia edition's total monthly views for the same month, then expressed as a percentage deviation from the article's own yearly average. This is the number we call "controlled." The "uncontrolled" figure is the same deviation calculation run on raw article views alone, with no adjustment for the wiki's own seasonal traffic.
- License: the underlying pageview data is CC0-1.0 (Wikimedia).
- Reproduce it yourself: pull monthly views for any article and for the `all-projects`/edition-wide total from the same REST endpoint, over the same date range, and divide one by the other per month. No special access or tooling is required beyond the public API.
Raw data, including every month and every edition referenced above:
We also collected a companion dataset on breakup and divorce-related search vocabulary, which we'll cover in a later piece in this series: relationship-vocabulary.json.
If you use these numbers, use the controlled figures, not the raw ones — and if you find a July mechanism we didn't, we'd genuinely like to hear it.