How many people could you actually date?

Geography. Dating markets are metro areas, not city limits, so every city maps to its Census metropolitan or micropolitan statistical area (a county-based region around an urban core of 10,000+). Search for Decatur, GA and you get the Atlanta metro. Towns outside any metro or micro area use their county, so Crested Butte, CO is Gunnison County. County samples are small, so treat rural numbers as rough.

Counts. Population, sex, age and marital status come straight from the ACS 2020-2024 5-year, table B12002 (sex by marital status by age), the most recent release that covers every county (the 2025 release has been delayed indefinitely). Those counts are centered around 2022, so each sex and age group is scaled to the Census Vintage 2025 population estimates by metro and county, sex and age while keeping the ACS marital-status mix. ACS reports ages in five-year bins; custom ranges are split using single-year-of-age shares from the microdata. “Not married” means never married, divorced or widowed. (ACS 2020-2024 5-year, scaled to Vintage 2025 population estimates (July 1, 2025).)

Cohabitation, education, kids, living situation, income. Census tables don't cross marital status with these, so they come from the ACS 2020-2024 5-year Public Use Microdata Sample: for each sex, age bin and marital status we measure the share living with a partner (or in an institution), and the joint mix of education, own children under 18 at home, living in a parent's home and personal income, for the microdata areas that overlap each metro. Small metros borrow strength from their Census division. Measuring these jointly, instead of multiplying separate percentages, keeps real correlations: unmarried 35-year-olds are more likely to have a degree and less likely to have kids than 35-year-olds overall. The microdata carry income in four brackets (under $25k, $25k–60k, $60k–120k, $120k+); an income floor inside a bracket keeps the national share of that bracket above it, by sex, age and education.

The survey steps come from national surveys and are applied as shares of whoever is left, so treat them as educated guesses (they're marked est. and can be switched off). Each is worked out separately for every combination of education, kids at home and race/ethnicity among who's left, so a step can depend on those where the survey supports it. The table below says exactly what each one depends on and what it assumes away. When you stack several Census filters in a small metro, fewer than about 30 survey records may stand behind the result; those steps get a thin data tag.

• Orientation: Gallup, LGBTQ+ identification (2025 data, published Feb 2026), with the age gradient from CDC NSFG 2011-13, sexual attraction by age (NHSR 88). Bisexual people count toward both sexes.
• Already seeing someone: weighted estimates, by sex and age, of unpartnered people with a boyfriend, girlfriend or partner they don't live with, from General Social Survey 1996-2024 (microdata: ROMANCE, POSSLQ), Stanford How Couples Meet and Stay Together 2017 (microdata) and CDC National Survey of Family Growth 2022-2023 (microdata). Question wording matters most: strict “steady partner” items give about 15%, broad “romantic partner” items 32–38%; the sieve uses the middle (about 42% of unpartnered women in their 20s, 24% of men, falling after 55).
• Looking (off by default): Pew Research Center, 5 facts about single Americans (2023). Under 50, about 57% of singles say they're looking for a relationship or dates. It's one stated-intent survey question, and plenty of people who aren't “looking” would date the right person, so it's an optional extra rather than part of the default sieve.

Race and ethnicity. The optional filter uses each metro's race and ethnicity mix of available singles by sex, age and marital status from the ACS 2020-2024 5-year Public Use Microdata Sample (Hispanic of any race; the other groups non-Hispanic). That mix is then adjusted for each education and income group by how race varies with education and income nationally among singles of the same sex, age and marital status (for example, far more Asian singles hold graduate degrees), so combining a race filter with a degree or income filter no longer treats them as unrelated. The local mix is the metro's own; only the education and income pattern is national. In the competition model, race pairing is a revealed preference like education: how often recent couples paired across groups relative to who was single and available. Those patterns reflect segregation, social networks and history as much as anyone's preferences.

Religion and politics. The Census asks about neither. Both come from one model of the Cooperative Election Study 2024 (60,000 adults): religion for unmarried adults by sex, age, education and race/ethnicity, and party (leaners counted as partisans) given all of those plus religion, with the religion–party link allowed to differ by race (white and Black Protestants vote very differently). So a degree, race or religion filter changes who's left politically, and vice versa. The national religion level is calibrated to Pew's 2023–24 Religious Landscape Study, since the CES panel runs less religious. Each place is then shifted by how its religious mix (Pew's metro figures where they exist, else its state's, plus the CES respondents who live there, the unmarried ones most) and its 2024 vote (2024 presidential results by county (compiled from official returns)) differ from the US beyond what its make-up already explains. Where one religion dominates and its members marry young (Latter-day Saints in Utah), estimates for young singles are rough. In the competition model, religious pairing comes from married couples under 45 in the General Social Survey 1996-2024 (microdata: ROMANCE, POSSLQ) and political pairing from both partners' party in Stanford How Couples Meet and Stay Together 2017 (partners' party), each relative to random pairing.

Kids someday. Unpartnered adults answering “Do you, yourself, want to have a(nother) baby at some time in the future?” in the CDC National Survey of Family Growth 2022-2023: yes, no or don't know, by sex, age, education, whether they already have kids and race/ethnicity (the public survey lumps Asian singles into “other”). It follows that mix among who's left instead of assuming independence; it doesn't depend on religion or income. The survey covers ages 18–49 here; older ages reuse the 45–49 rates, which likely overstates how many older women want kids. The unsure (about 4%) are never ruled out. It's national, and intentions change.

Height and health. The height range uses measured heights from CDC NHANES 2017-2023 (measured height and BMI, smoking) by sex, age and race/ethnicity. Smoking and BMI (measured, not self-reported) come from a model of NHANES adults by sex, age, degree (bachelor's or not), race/ethnicity and whether they live with a partner, used for the unpartnered; so a college-grad filter now finds fewer smokers, and BMI differs by race the way it really does. Each place is then shifted by how its 2024 CDC BRFSS 2024 metro (SMART) and state prevalence rate (or its state's, where the CDC doesn't survey the metro) differs from the US beyond what its race, education and age make-up already explains, so a place isn't counted twice for its demographics. None of these depend on income, and education only as degree or not. The 2024 vote in the context card sums county results from 2024 presidential results by county (compiled from official returns).

Competition is a revealed-preference choice model fitted to couples formed in the last one to three years (age, education, income and marriage history), plus published estimates for kids and height. It scores every available competitor by how well they match what your targets go for. Your competition-adjusted pool is your pool divided by the number of serious competitors per person in it (counting competitors by appeal, so long shots count for little), times your appeal relative to a typical serious competitor. With one serious competitor per person and you exactly typical, it equals the raw pool. Your filters are your preferences, not your competitors', so how crowded the market is gets measured on the pool before your filters, while how you stack up uses what your filtered pool goes for. Details are in the competition section.

Known gaps. Census records sex, not gender identity. Men's children are inferred from household relationships, so kids who live elsewhere are invisible. Nothing here knows whether anyone would like you. Inspired by Backus (2010), Why I don't have a girlfriend.