Updated August 7, 2026. Quick answer: the arithmetic people use to get there compares two different countries. CDC publishes a 2023 marriage rate of 6.1 per 1,000 and a divorce rate of 2.4 — but the divorce figure excludes five states, and CDC divides it by a population of 279.8 million against 334.9 million for marriages. The rates do not share a denominator. Census cohort data puts the share of ever-married adults who have ever divorced at 34% of women and 33% of men.
The claim: “Half of all marriages end in divorce”
Verdict: MUTATED — Right figure, wrong meaning.
Evidence tier: primary. Verdict class chosen from the taxonomy fixed before the evidence.
Where the number comes from
The claim is almost always produced by putting the annual divorce rate next to the annual marriage rate. Here is what CDC actually publishes, from the National Vital Statistics System provisional tables for 2000–2023:
“Provisional number of marriages and marriage rate: United States, 2000-2023 … 2023 2,041,926 334,914,895 6.1”
“Provisional number of divorces and annulments and rate: United States, 2000-2023 … 2023 672,502 279,800,079 2.4”
Read from cdc.gov/nchs — National marriage and divorce rate trends, 2000–2023 on 7 August 2026.
🔴 Look at the two population columns. CDC divides marriages by 334,914,895 and divorces by 279,800,079. They are not the same denominator, and the reason is in the table’s own footnote 1, which is attached to the 2023 divorce row:
“Excludes data for California, Hawaii, Indiana, Minnesota, and New Mexico.”
The divorce denominator is 83.5% of the marriage denominator — our division, from their two columns. So the published “national” divorce rate is not national: it omits California, the largest state in the country, and has omitted it every year in the series. The exclusion list also changes from year to year, which the footnotes show plainly — Georgia, Louisiana and Oklahoma appear and disappear across the earlier rows — so the trend line does not hold its own population base constant either.
What that source actually measured
Even with matched denominators the arithmetic would not mean what people think. A crude rate counts events in one calendar year against the whole population. The roughly 2,042,000 marriages of 2023 and the roughly 672,500 divorces of 2023 are almost entirely different couples: the divorces come out of marriages formed across the previous four or five decades, while the marriages formed in 2023 will not have known outcomes for decades yet. Dividing one by the other compares a flow of new marriages to a flow of exits from an older, larger stock. It cannot produce the share of any real group of marriages that ends in divorce.
That explanation is ours. We looked for a current CDC page cautioning against the calculation in those terms and did not find one, so we are not attributing the reasoning to CDC.
The measure the claim wants is a cohort measure, and the Census Bureau publishes it. From Number, Timing, and Duration of Marriages and Divorces: 2016 (Current Population Reports P70-167, April 2021):
“Among ever-married adults 20 years and over, 34 percent of women and 33 percent of men had ever been divorced, while the percentage ever divorced was highest for adults aged 55 to 64 (about 43 percent for both sexes).”
Read from census.gov — Number, Timing, and Duration of Marriages and Divorces: 2016 (P70-167) on 7 August 2026.
And on whether marriages last, the same report follows actual marriage cohorts:
“Table 10 shows that about 58 percent of men who married between 1970 and 1974 stayed married for at least 20 years, and about 54 percent of men who married between 1985 and 1989 stayed married as long.”
More than half of those marriage cohorts reached their twentieth anniversary, which is difficult to reconcile with half of all marriages ending in divorce.
The honest current figure
There is one federal figure that does land near 50%, and it is the strongest case for the claim, so here it is. CDC’s National Survey of Family Growth publishes the probability of first-marriage disruption: by 20 years, 48% for the 2006–2010 period and 53% for 2011–2015.
Read from cdc.gov/nchs — National Survey of Family Growth, key statistics on 7 August 2026.
Two things stop it from supporting the claim as stated, and CDC says the first one itself:
“Marital disruption is defined as separation, divorce or death (rare for this age group)”
Disruption is not divorce. It counts separations that never become divorces, and it counts widowhood. And the universe is narrow: the table covers women aged 15 to 44, so it describes marriages formed young by women still under 45 — a group with historically higher disruption than marriages formed later. It is not “all marriages.”
So the honest summary is that a near-50% federal number exists and measures something other than what the claim says it measures. That is why the verdict here is MUTATED rather than false: the figure is real, and its meaning has drifted.
We did not establish this claim’s lineage. We are not asserting that the popular saying was derived from the NSFG table; we have not traced it, and we are not going to invent a chain. The verdict rests on the mismatch between the claim and the federal figures nearest to it, not on a proven derivation.
What to cite instead
“Census Bureau data puts the share of ever-married adults who have ever divorced at 34% of women and 33% of men, and more than half of the 1970s and 1980s marriage cohorts reached their 20th anniversary. The widely repeated 50% figure comes from comparing annual divorce and marriage rates, which CDC computes on different populations – the 2023 divorce rate excludes five states, including California.”
Source: Clear Money Guide, The Provenance File, 2026. Free to reuse with attribution under CC BY 4.0.
What this audit does not establish
No single national figure for the eventual share of first marriages ending in divorce is published in the Census report, and we are not constructing one. P70-167 gives race-specific eventual shares and a five-year checkpoint; it does not give an all-groups total. UNVERIFIED, and left that way.
The Census figures are 2016 data published in 2021. They are the most recent edition of that report we located, not necessarily the most recent data in existence.
The NSFG table is dated. CDC’s page carries a 2017 review date and the underlying cycle is 2011–2015. Whether a newer cycle has been published in this form is unverified; we did not find one.
This audit does not measure whether divorce is rising or falling by age. That is a separate question with its own owner on this site, and the figures there are not restated here.
Related
Gray divorce statistics owns the age question — where divorce has actually moved, with its own sourced figures. Divorce and your money covers the financial mechanics.
This is an audit in the Provenance File. Our method, the six verdict classes and the desk laws are published at how we check money claims. If you think this is wrong, tell us — with a source. No advertising appears on this page and we earn nothing from it.