The Age Structure of the Electorate,
1966–2024
Who is old enough to vote, who says they voted, and
sixty years of the gap between the two
Democracy’s Data
Last updated September 15, 2026 at 4:42p EDT
Older Americans run American elections. You have heard the sentence,
or one of its relatives: the electorate is grey, the young do not vote,
politicians answer to the people who show up. It is repeated as though
it were a permanent fact about the country. Is it true — and if it is,
how much more do older Americans actually vote, and has the gap always
been there?
Start with a distinction the word “electorate” hides. It sometimes
means everybody entitled to vote, and it sometimes means everybody who
did. Those are different populations. In 2024, 260.4m
Americans were old enough to vote and 154.4m told the
Census Bureau they had — 59.3%. The clearest way to see how different
the two groups are is to draw them as two pyramids on one axis, and to
watch the pyramids for sixty years.
The test
The data is the Census Bureau’s Current
Population Survey, Voting and Registration Supplement — historical
Table A-1, whose age blocks cover 30 federal elections from 1966 to
2024. The Current Population Survey is the Bureau’s monthly survey of
households, best known as the source of the unemployment rate. After
every federal election since 1964, it has asked the roughly 60,000
households in that month’s sample whether the adults in them voted, and
the Bureau publishes what they said, by age.
Be clear about what kind of source that is. The same Bureau runs the
decennial census, which is an enumeration: an attempt to count everyone,
one by one. The CPS is the other instrument: a survey, a sample of
households chosen to stand in for the country. Its answers are weighted,
which means counting some responses more than others so the sample
matches the population.
Nobody’s voting answer is checked against a record; every “voted” in
this brief is a self-report — a person’s own account, taken at its word.
This brief sits in the survey section because that is what the source
is, whoever publishes it. The surveys
chapter lays out what a sample can establish that a count cannot,
and this series is a sixty-year example.
Why this source, of all sources? Because nothing else reaches. No
administrative record pairs a national age profile of the adult
population with an age profile of its voters, election after election,
back to the 1960s. The CPS supplement does exactly that, in one
continuously published table, from the agency that also measures the
population itself.
The data
The workbook is not one table. It is ten tables stacked in a single
spreadsheet column: five age groups (the total, then four bands), each
appearing twice, once for voting and once for registration. Pulled apart
and reassembled, those ten blocks become two flat files.
bands.csv holds one row per year per age band:
| year |
the federal election year, 1966 to 2024 |
discrete |
| band |
one of the Bureau’s four age bands |
ordered categorical |
| vap |
voting-age population, in thousands |
count |
| pct_voted |
the share who told the survey they voted |
continuous |
| voters |
vap multiplied by that share — a count this chapter
derives |
count |
The Bureau publishes the population and the percentage; the count of
voters is the two multiplied, which is arithmetic on its own figures
rather than an estimate of ours. Here is 2024:
| 18–24 |
30.0m |
13.2m |
43.9% |
53.5% |
| 25–44 |
88.8m |
46.5m |
52.4% |
61.5% |
| 45–64 |
81.2m |
51.3m |
63.2% |
69.8% |
| 65+ |
60.4m |
43.3m |
71.7% |
77.1% |
shares.csv is derived from it: each band’s
pop_share of everybody old enough to vote, its
vote_share of everybody who said they voted, and
gap, the second minus the first. The four bands are
published separately from the total, so they can be checked against
it:
| age blocks found |
10 |
| years covered |
30 |
| largest gap, bands against total (population) |
0.001% |
| largest gap, bands against total (voters) |
0.207% |
Before you read on, commit to a view. People aged 65
and over are 23.2% of American adults. What share of the votes
do you think they cast? Write it down, and write down what you
would have guessed for 1968.
What it says about democracy
Figure 1. The adult population and the voters, 2024.
Two pyramids on one axis, because the question is whether the second
population is a scale model of the first, and a pyramid shows the answer
band by band. Grey to the left is each band’s share of everybody old
enough to vote; teal to the right is its share of everybody who said
they voted. Drag the year; the button switches from shares to counts of
people.
The two sides are not the same shape. People aged 65 and over
are 23.2% of the adult population and 28.1% of the voters.
People aged 18 to 24 are 11.5% of adults and 8.5% of voters. Compare
that with what you wrote down.
The mechanism is not mysterious — it is turnout. 71.7% of the oldest
band said they voted against 43.9% of the youngest, a ratio of 1.6 to
one. Some of that gap opens before the ballot: 77.1% of the oldest band
were registered against 53.5% of the youngest, so roughly half the
difference is a registration difference rather than a turnout one.
So the claim survives its first test. The people who show up are
older than the country that could have shown up, by a margin large
enough to see from across the room. The second question is whether it
was always so, and the answer is the finding of this brief.
Figure 2. Over-representation, presidential elections 1968 to
2024. One line per band: its share of the voters minus its
share of the adults, so zero means a band cast exactly its share of the
votes. A line, rather than more pyramids, because the question here is
change over time.
The oldest band starts at -0.5 in 1968, meaning it
voted almost exactly in proportion to its size. It ends at
+4.9. The over-representation of older voters is not an
old feature of American elections. It is roughly as old as the people
currently complaining about it.
Two things drove it and they are easy to confuse. The band grew:
65-and-over went from 15.8% of adults in 1968 to 23.2% now, which is
demography and would have raised their share of the votes even if
nothing about their behaviour had changed. And on top of that their
turnout advantage widened. The first is a fact about the
country. The second is a fact about its politics. A count of
votes cast cannot separate them, which is why Figure 2 is drawn as a
difference between two shares.
The youngest band, meanwhile, is roughly where it was. It has been
between -8.9 and -2.5 points under-represented in every presidential
election in the series. Whatever has changed about young Americans and
voting, their standing relative to their own numbers is not it.
One property of Figure 2 is worth knowing before you lean on it.
Suppose turnout rose by ten points in every age band at once: nothing in
the figure would move, because a uniform rise cancels out of a share.
That is what makes the lines readable across high-turnout and
low-turnout years, and it is also what makes them blind to which years
those were. Figure 1’s counts button is there for that.
What this brief cannot tell you
These are people who said they voted. The survey
asks after the election, and nobody checks. The validated-turnout
chapter establishes that the answer is consistently too high — in
2024 the survey implies more voters than there were ballots. If
over-reporting differs by age, and there is good reason to think it
does,
then the right-hand pyramid is distorted in a way none of these figures
can correct.
Four bands is a coarse pyramid. A conventional
population pyramid has five-year bands and splits by sex; this one has
four unequal bands and no sex, because that is what the historical table
publishes. The 25-to-44 band spans twenty years and the 65-and-over band
has no upper bound, so both hide structure — the oldest band in
particular mixes people at very different turnout rates. And the
youngest band changes its meaning once. Before 1972 the voting age was
21 in most states, so the youngest block for the earlier elections
counts mostly 21-to-24-year-olds; the Twenty-sixth Amendment, ratified
in 1971, lowered the age to 18 everywhere, and the band’s
population jumps from 11.6m in 1968 to 24.6m in 1972 for that reason.
Figure 2’s first point for that band describes a narrower group than its
label.
Nothing here is about a state. These are national
figures. The turnout-denominator
chapter shows how much states differ on the denominator alone, the
bottom number a turnout rate is divided by. An age pyramid for Florida
would not look like one for Utah.
Being over-represented is not the same as being
decisive. A band that casts 28.1% of the votes decides nothing
on its own, and this brief has said nothing about how any band voted.
Who shows up is prior to, and separate from, what they do when they get
there.
What you should have learned
- In 2024, people 65 and over were 23.2% of American adults and 28.1%
of the people who said they voted; people 18 to 24 were 11.5% of adults
and 8.5% of voters.
- The mechanism is turnout — 71.7% against 43.9%, a ratio of 1.6 to
one — and roughly half of that gap is a registration gap.
- The oldest band’s over-representation is new: -0.5 points in 1968,
+4.9 points in 2024. Part of that is the band growing, part is its
turnout advantage widening, and the two are different kinds of
fact.
- The CPS supplement is a survey: a weighted sample of self-reports,
published by the Census Bureau but not a count, and every “voter” in it
is a respondent’s claim.
Extensions
shares.csv carries a gap for every year
and age band. Which band’s gap has moved the most across the series, and
which has barely moved?bands.csv reports pct_registered and
pct_voted separately. Work out what share of the registered
actually voted, band by band. Do the youngest fail to register, or fail
to turn out?- Pick a presidential year in
bands.csv and the midterm
beside it. Which age band falls away the most between the two?
- Everything here is people saying they voted. Add up
voters across the bands in bands.csv and set
the total against a turnout figure from anywhere else. Which way does
the error run?
- Stretch: the Bureau publishes the same table for single
states in recent years. Build Florida’s two pyramids and Utah’s, and see
how far a national figure can mislead about either.
- Stretch: the CPS microdata records age in years, not bands.
Sketch what Figure 2 would look like with a 65-to-74 line and a
75-and-over line, and say which result above you would expect to
change.
Sources
Data
The data itself
Every figure above is drawn from these tables. They are plain CSV,
they open in a spreadsheet, and they are yours to take further.
bands.csv, shares.csv
Two more files here are not data about the subject at all – they are
how this chapter checked its own work: facts.csv, the individual
numbers the text above quotes, and checks.csv, the
arithmetic verified before anything was drawn.
Rebuild this data yourself, with an AI assistant
I want to rebuild a public dataset from its original source, without any
starting files. Please do the following and show your work.
THE SOURCE. U.S. Census Bureau, Current Population Survey, Voting and
Registration Supplement, historical Table A-1. One spreadsheet at
https://www2.census.gov/programs-surveys/cps/tables/time-series/voting-historical-time-series/hst_vote01.xlsx
No account or key is needed. The sheet is not one table: it is ten
tables stacked in a single column. Five age groups -- the total, then
18 to 24, 25 to 44, 45 to 64, and 65 and over -- each appear twice,
once under a heading ending "Percent Voted" and once under "Percent
Registered". Find those headings and read years downward until the
years stop. Populations are in thousands. And keep in mind what this
is: a survey that asks people after the election whether they voted,
and nobody checks. People say yes more often than they voted.
WHAT TO BUILD.
1. One table with a row per election year per age group: the
voting-age population, the percent who said they voted, and the
count of voters that implies.
2. Each group's share of all adults beside its share of everyone who
said they voted, year by year -- two age pyramids, one of the
population and one of the reported electorate.
CHECK YOUR WORK. The copy this book used was downloaded in August 2026.
If your rebuild matches it, you should find:
- ten stacked tables, together covering 30 federal elections from
1966 through 2024
- in 2024, an adult population of 260,363 thousand, of whom 154,395
thousand said they voted
- in 2024, 43.9% of 18-to-24-year-olds said they voted, against
71.7% of people 65 and over
- the four age groups reconstruct the published total to within
0.207%
The Bureau extends this workbook after each federal election; a later
final year means the series grew, not that you made an error.
