Educational only. This page describes public statistics about populations. None of it bears on what you personally must withdraw. For your situation, consult your CPA.
Most writing about how Americans draw down retirement accounts rests on a survey, a fund company's own book of business, or a number nobody checked.
The IRS publishes better material — free, annual, drawn from filed returns. It is dry, late, and largely unread. This page is where we keep our work with it, and where we write down what those files cannot show, which is the part almost nobody states.
In short: The IRS publishes two families of retirement data. One has ages but no geography. The other has geography but no ages. Neither marks a distribution as required, so in the public record an RMD is indistinguishable from a withdrawal somebody simply wanted. Both arrive two to three years late and both get revised. Your own required minimum is the opposite kind of number: a division you can compute to the dollar.
On this page
- What We've Published
- The Two Datasets Don't Cross
- Nothing Marks a Distribution as Required
- What the Data Can't Tell You
- The Year a Field Vanished
- Your Own Number Is Not an Estimate
- Where These Files Come From
- When the Next Numbers Land
- How We Handle Numbers
- Frequently Asked Questions
What We've Published
Tax Year 2023: The Same IRA Dollars, Spread Across More People — the new county and state files, verified against our own pipeline run: total distributions flat at $432 billion, 242,170 more returns reporting one, and the average per recipient down — with a plain statement of what this data cannot explain.
Mapping Retirement Income: What the IRS Publishes, and the Year One Field Vanished — retirement income by state, county, and ZIP, drawn from the SOI geographic files, and why a multi-year map has to be labeled "retirement income" rather than "IRA distributions."
For the rules themselves rather than the analysis, start at What is an RMD? or the inherited IRA hub.
The Two Datasets Don't Cross
| Product | Grain | Age? | Geography? | Basis |
|---|---|---|---|---|
| Accumulation and Distribution of IRAs | National totals | Yes — five-year bands | None, in any table, in any year | Sample estimates |
| Individual income tax statistics by geographic area | State, county, metro, district, ZIP | None | Yes | Essentially all filed returns |
The age data has no map. The map has no ages. No published IRS product joins them, so a claim like "retirees over 75 in this county withdrew X" cannot be sourced to the IRS at all — only modeled.
This is the most common thing people assume is available. It isn't, and the absence is structural rather than an oversight: the two products are built from different samples, on different schedules, for different administrative purposes.
The age bands carry a second limit. They run five years wide and top out at "80 and over," so the population most affected by required distributions is flattened into one open-ended band. Because the current RMD age is 73, the "70 under 75" row straddles the threshold — some people in it owed a distribution that year and some didn't, and the table can't separate them.
Nothing Marks a Distribution as Required
Form 5498 does carry required-distribution boxes: a checkbox for the coming year, an RMD date, an RMD amount. The IRS receives the figures. Three things keep them out of the statistics.
- They're prospective. Box 12b is what your custodian calculates you will owe next year, not what you took. The withdrawal is reported separately on Form 1099-R, with no link back.
- They're incomplete by instruction. The form's own instructions state that an RMD may be required even when the box is not checked, and the amount box may be blank when the custodian sends the figure separately.
- SOI never publishes them. No tabulation of that box exists in any IRS statistical product.
Meanwhile Form 1099-R — the record of money that actually left — has no code for required. Its codes describe circumstances, not obligation. An eighty-year-old's required distribution and a sixty-two-year-old's discretionary withdrawal both carry code 7, "Normal distribution." A beneficiary's required distribution and a beneficiary's full liquidation both carry code 4, "Death."
So in the public data, a distribution is a distribution. The taxable IRA distributions reported for any county in any year contain, mixed together:
- Required minimums taken by owners who have reached the age at which they're due
- Voluntary withdrawals by anyone past 59½ who wanted the money
- Distributions from inherited IRAs, including beneficiaries working through the 10-year rule
- The taxable portion of Roth conversions
- Early and hardship withdrawals
A county with unusually high taxable IRA distributions might be full of dutiful eighty-year-olds taking their minimums. It might be sixty-two-year-olds running Roth conversions ahead of retirement. It might be beneficiaries deliberately taking more than required to avoid a balloon in year ten. The files can't distinguish those worlds.
What the Data Can't Tell You
The Line Between Reporting and Guessing
Questions public IRS data answers, and questions it only appears to.
The data can answer
- How much taxable retirement income was reported in a state, county, or ZIP code
- How many returns reported any of it
- Aggregate IRA balances and withdrawals by five-year age band, nationally
- How those totals have moved across years
The data cannot answer
- How much of any withdrawal was required rather than voluntary
- Which withdrawals came from inherited accounts
- Anything about age within a state or county
- What a typical household did — there are no medians
Based on the published table structures and record layouts of the IRS SOI IRA publication and geographic data files.
Two limits deserve naming beyond that card.
Every dollar figure is a total. Divide dollars by returns and you get a mean, which retirement data drags hard toward a small number of very large accounts. SOI publishes no medians for these items, and the gap between mean and typical is widest where people most want the number.
The published numbers get revised. Data that is already two years old is not therefore final. In June 2026 SOI reissued the IRA publication under a revised methodology, noting that tax year 2023 and later will use only the new approach. Tax year 2022 now exists in two versions, and a chart built from the earlier release won't match one built from the current file. Nothing is hidden; it's documented briefly, in places nobody checks. Any figure cited from here carries a tax year and a retrieval date.
The Year a Field Vanished
The geographic files count lines on a tax form. When the form changes, the statistic changes with it.
For tax year 2018 and that year alone, the field for taxable IRA distributions does not exist. The redesigned "postcard" Form 1040 collapsed the separate IRA and pension lines into a single 4a/4b, and SOI's documentation records the consequence flatly:
IRA distributions (N01400 and A01400) and Pensions and annuities (N01700 and A01700), which were previously separated, have been combined to create a new field: IRAs, pensions, and annuities (N01750 and A01750).
The 2019 form split them apart again and the separate IRA field returned. But the hole in the middle of the series is permanent, and any multi-year comparison has to work around it rather than splice across it.
Nothing about American retirement behavior changed in 2018. A form was simplified, and a statistical series lost a field for twelve months. Public data records what a form happened to ask in the year it asked it — which means changing the form changes what the past looks like. Full write-up here.
Your Own Number Is Not an Estimate
Set against all that uncertainty, your required minimum distribution is a division: prior year-end balance, divided by a life expectancy factor the IRS publishes in Appendix B of Publication 590-B.
Two tables do most of the work. Table III, Uniform Lifetime, covers unmarried owners and most married owners. Table I, Single Life Expectancy, is titled "(For Use by Beneficiaries)." The factors are printed to one decimal place, fixed for the year, and identical for everyone the table covers — 26.5 at age 73, 24.6 at age 75 in the current edition.
| Population behavior | Your required minimum | |
|---|---|---|
| Source | Sampled and aggregated returns | Published IRS factor tables |
| Precision | Estimates with sampling error | To the dollar |
| Timeliness | Two to three years behind | Current year |
| Revisions | Happen, quietly | The table for the year is the table |
| Applies to | Populations | You |
Two people with the same age and the same December 31 balance owe the same number. Research on this page can tell you something about the country and nothing about your account — for that, see how the calculation works, which table applies, or the inherited-IRA version.
Where These Files Come From
Both products come from the Statistics of Income division, and both are built from the same two information returns.
Form 5498 reports contributions, rollovers, conversions, and the account's fair market value at year end — the balance side, and the same input a required minimum is computed from. Form 1099-R reports each distribution: gross amount, taxable amount, withholding, and a code describing the kind of distribution. It's also the form your own RMD shows up on at tax time.
The IRA publication is a sample. Every table carries the line "all figures are estimates based on samples," and SOI publishes companion confidence-interval files so the uncertainty can be quantified.
The geographic files are near-census, built from the returns themselves rather than a panel — but they discard a great deal on purpose. ZIP codes with fewer than 100 returns are rolled into a catch-all 99999. Items with fewer than 20 returns in an income class are collapsed or dropped. Return counts are rounded to the nearest 10. Returns with negative AGI are excluded. Where one return would dominate a cell, the item is suppressed, and the IRS states that the threshold it uses cannot be released.
That's disclosure protection, not sloppiness, and it's why the files can be published at all. But it isn't neutral: the rules bite hardest on sparse and rural places, which are systematically thinner in this data than dense ones. Reading that as a finding about where retirees live is a mistake.
One trap that produces wrong numbers silently: all money amounts are in thousands of dollars.
When the Next Numbers Land
Tax year 2023 county and state data posted in late August 2026; ZIP-level data had not yet posted as of August 27, 2026. The scheduled dates were:
| Scheduled date | Release | Tax year |
|---|---|---|
| August 13, 2026 — posted late August 2026 (our analysis) | State, County, Metropolitan and Micropolitan Data | 2023 |
| August 27, 2026 | Congressional District and ZIP Code Data | 2023 |
When the ZIP-level files land, we will publish a what-changed analysis within 48 hours.
The IRS calls its own dates tentative, and the tax year 2023 dates above have already moved once. The lag has been drifting longer rather than shorter — tax year 2022 landed about two years after the tax year closed; 2023 is now scheduled closer to two and a half.
The IRA publication runs on a separate clock and is currently ahead of the maps: tax years 2022 and 2023 were both posted in June 2026. For now the national age data covers a more recent year than the geographic files do, which is an easy way to compare two different years without noticing.
If you're reading this after those dates, check the source rather than this paragraph.
How We Handle Numbers
- Primary sources only — every figure comes from an IRS file we pulled ourselves, not an article citing an article.
- A tax year on every number, because the underlying form changes.
- No illustrative statistics. If a point can't be supported by a figure we verified, we make it qualitatively or cut it.
- Limits next to findings — sampling error and suppression rules belong in the same paragraph as the result, not a footnote.
- Aggregates stay aggregates. Nothing about a population is presented as though it applied to your account.
If something here violates one of those, it's a defect and we'd like to know.
Frequently Asked Questions
Does the IRS publish data on required minimum distributions specifically?
No. Form 5498 has RMD boxes, so the IRS receives figures — but they're a prospective estimate for owners, an RMD may be required even when the box isn't checked, and SOI publishes no tabulation of them. Form 1099-R, which records what actually left the account, has no code marking a distribution as required. Every published statistic mixes required and voluntary withdrawals together.
Can I find IRA distributions by age for my state?
Not from published IRS data. The IRA publication has ages but no geographic breakout in any of its ten tables, in any year. The geographic files have geography but no age classifier. No published IRS product joins them.
Why is tax year 2018 missing from IRA distribution series?
The redesigned 2018 Form 1040 merged IRA distributions with pensions and annuities into one line, so two statistical fields became one. The 2019 form split them apart again, but tax year 2018 has no standalone IRA figure and never will. Full explanation.
Is this data a sample or does it cover everyone?
Depends which product. The IRA publication is a sample and says so on every table. The geographic files are built from essentially all returns filed in the relevant window, though they exclude negative-AGI returns, returns without a usable ZIP, and foreign addresses, and they suppress small cells.
Does any of this affect what I have to withdraw?
No. Your required minimum is your prior year-end balance divided by a factor from the IRS tables. It has nothing to do with what anyone else in your state or age band did. See how to calculate an RMD.
Where can I get these files myself?
They're free, and they're ordinary spreadsheets — linked in the sources below. Read the documentation guide for the specific tax year before touching the data; the "Nature of Changes" section is where the IRS records what moved. And remember that all money amounts are in thousands.
Research is context. Your number is a calculation.
- How to calculate an RMD — the arithmetic, with the table that applies to you.
- Retirement income by state, county, and ZIP →
- Terms defined in the glossary → · Source documents →
This article is for informational purposes only and does not constitute tax, legal, or financial advice. Aggregate statistics describe populations, not individuals, and have no bearing on any particular taxpayer's required minimum distribution. IRS rules and tax laws are subject to change. Consult a qualified tax professional or financial advisor for guidance specific to your situation. SimpleRMD is a calculation and tracking tool — not a financial advisory service.
Sources: IRS Statistics of Income — Accumulation and Distribution of Individual Retirement Arrangements (tax years 2022 and 2023, posted June 2026, including Table 4 by age of taxpayer); Data by geographic area; Individual Income Tax Statistics — ZIP Code Data; Upcoming data releases; SOI Tax Stats: What's New. Field labels, disclosure rules, and the "Nature of Changes" quotations are taken from the SOI ZIP Code Data documentation guides for tax years 2018, 2019, and 2022, and confirmed against the published data files. Form line references verified against the 2017, 2018, and 2019 Forms 1040 in the IRS prior-year forms archive. Distribution factors: IRS Publication 590-B (2025), Distributions from Individual Retirement Arrangements (IRAs), Appendix B, Tables I and III. Form box labels and instructions read from the current Form 5498, IRA Contribution Information and Form 1099-R. All sources retrieved August 2026.