The US Census Bureau's first LGBT data is buried in a pandemic survey

Canada started its 2026 census this month.

(Yes, the Canadian census happens on years ending in 1 and 6. No, I don’t know why.)

I clicked over to StatsCan to look at the data tables from the last census, and thought: this should be a map.

So I built one: censusmaps.ca.

It’s a zoomable map of Canada, broken down by census division (cities-ish), then by census tract (neighborhood-ish), then by dissemination area (block-ish). You see population, age, language, ethnic origin, all the rest.

Once you’ve wired up one country’s census, the next one is easy.

uscensusmaps.org

So I did the U.S. version too: uscensusmaps.org.

Same idea, American Community Survey five-year data, drilling all the way down to block group.

While I was knee-deep in Census Bureau FTP folders, I noticed a dataset I’d never heard of: Household Pulse Survey

The Pulse

The Household Pulse Survey is what the Census Bureau spun up during COVID to publish data faster than the ACS could. Two-week turnaround instead of two years.

Most of it is pandemic stuff: who’s working, who’s hungry, who’s depressed, who’s vaccinated.

But it also asks questions the US census has never asked:

Which of the following best represents how you think of yourself?

1. Gay or lesbian
2. Straight, that is, not gay or lesbian
3. Bisexual
4. Something else
5. I don’t know

Plus a question on current gender, plus a question on sex assigned at birth.

This was the first time the U.S. Census Bureau has ever collected sexual orientation and gender identity data from a national sample.

And almost nobody has plotted it.

So I pooled seven non-overlapping HPS cycles spanning September 2021 to June 2024, weighted everyone by PWEIGHT (Census’s per-respondent weight, i.e. how many U.S. adults each respondent stands in for), and made some charts.

I thought these graphs should be out there, so here they are:

By birth year

By income

By state

By metro area

Per metro area breakdowns

Show all 15 metro area breakdowns

Caveats

Before reading too much into any of this:

• HPS response rates are 6 to 7%, with documented nonresponse bias toward higher-income and higher-education respondents. The absolute levels are likely overstated. The shapes (across cohorts, across geographies) are more defensible than the magnitudes.
• HPS is online-only, and online surveys tend to report higher LGBT shares than phone or in-person modes (more privacy, less social desirability bias). Julian, Manning, and Westrick-Payne (Demography, 2024) put HPS at 8.5% LGBT in 2022 vs Gallup at 7.2%, NHIS at 7.2%, and BRFSS at 6.7%. The Census Bureau itself warns that HPS “should not be interpreted as the single indicator of LGBT prevalence.”
• These charts almost certainly undercount the youngest cohorts. HPS offers “something else” for sexual identity and “none of these” for gender; Julian et al. show those buckets are increasingly picked by younger respondents using identities outside the L/G/B/T labels (pansexual, queer, nonbinary, gender-fluid, etc.). 4.87% of the 1995-2004 cohort picked “something else” vs 0.71% of the 1945-1954 cohort, and 3.44% picked “none of these” vs 0.67%. My charts treat both as not-LGBT, so the real cohort gradient may be steeper than what’s plotted.
• Older cohorts also skip the SOGI questions at higher rates (2.75% missing in 1945-1954 vs 0.75% in 1995-2004), so part of the cohort gradient is who answers, not just who identifies.
• The transgender count here includes respondents whose sex assigned at birth differs from their current gender but who did not pick “transgender” themselves. Census imputed the sex value for roughly half of these, which Julian et al. flag as warranting caution.
• Geography in the public file stops at state, region, and the 15 largest MSAs. There’s no county, ZIP, or PUMA.
• INCOME is household income for the prior calendar year, not personal.
• Birth year 2006 is dropped from the cohort chart. Phase 4.0 ran a split-panel test that asked SOGI of only half the sample, and 2006 only shows up in two cycles, so the effective N is too small to plot.
• Phase 4.2 (late 2024 onward) transgender data is excluded. Census changed the available answers for the SOGI questions in that phase, so the numbers don’t line up with the earlier cycles.

Raw data

U.S. Census Bureau Household Pulse Survey Public Use Files