How the immigrant share of the population shifts across the country. Press play to animate. Source: US Census ACS 5-year.
Spatial access by the enhanced two-step floating catchment area method (E2SFCA). Darker green = better access; orange = a service desert (people who need it, but no organization within reach); gray = no foreign-born population. Source: org locations + US Census ACS 2024.
How inclusive each state’s policies are toward immigrants, across seven domains: health coverage, language access, higher education, identification, integration & licensure, legal protection, and economic inclusion. Darker green = more inclusive; orange/red = more restrictive or limited. Click any state for its full profile.
Standardized gaps between foreign-born and US-born residents (uninsured, crowded housing, rent burden, education, home internet, vehicle access, homeownership) from US Census ACS. Green = smaller gaps; red = larger gaps. This is an outcome measure, separate from the Policy map: it describes how immigrant and US-born residents differ, and is shaped by who has immigrated where (recency and composition), not by state policy alone. Click any state for its full profile.
Project Implicit is a self-selected volunteer sample, not a representative survey; within each area, respondents are reweighted to the Census population by age, sex, race and ethnicity, and citizenship. Higher = less welcoming. Areas with too few respondents are left blank. See the note below for the exact survey item. Source: Project Implicit 2023-2025.
Project Implicit is a self-selected volunteer sample, not a representative survey; within each area, respondents are reweighted to the Census population by age, sex, race and ethnicity, and citizenship. Higher = less welcoming. Areas with too few respondents are left blank. See the note below for the exact survey item. Source: Project Implicit 2023-2025.
Pick a group to see its own profile: how large it is and how it lives, with the all-foreign-born and US-born averages shown for context. National estimates from US Census ACS 1-year microdata (PUMS).
Single countries of birth, single languages, and ancestries that include the US-born, for the nation, each state, and each metro area. Source: IPUMS USA, ACS 2020-2024 5-year microdata.
Adults 18 and older in the United States, by nativity, years in the US, citizenship, and the language of the interview. National Health Interview Survey 2019-2025, pooled. Each bar is an estimate; the line through it is the 95% confidence interval.
How this dataset was built and what every field means. This page doubles as the data dictionary for the accompanying paper.
A national, geocoded directory of 15,264 US nonprofit organizations that serve immigrants and refugees. It is built on IRS administrative records, adds a service taxonomy, an organizational profile and a website language-access reading taken by large language models from each organization’s own website, and is linked to American Community Survey demand and a spatial-accessibility analysis. The screening and attribution behind it are described below, step by step, with what each step was measured against.
Every registered US nonprofit was passed through five detectors, and their union is a candidate frame of 61,240 organizations: an ethnonym or community name in the registered name, an ethnonym in the Form 990 program text, an immigrant or refugee word in either field, a National Taxonomy of Exempt Entities code for ethnic and immigrant centers (P84) or international migration and refugee issues (Q71), and membership in one of the reference directories listed above. Each detector is permissive by design, and most organizations were found by one detector alone.
A candidate is then judged on the evidence it has, never on the detector that found it. The evidence is the text of a website proven to belong to the organization, together with its own Form 990 program narrative. Three open-weight language models read that evidence independently, and each returns include, maybe, or exclude; a model voting to exclude must name a ground from a fixed list, in the way a systematic review charts its exclusions. All three must agree for include or for exclude, and a split panel returns maybe, because a wrong exclusion costs an organization its place in every later use while a wrong maybe costs one more page read. Evidence enough to screen on existed for 49,880 of the 61,240 candidates, and the panel was unanimous on 73.2 percent of them.
The caution that makes a split return maybe leaves a large maybe population, so the panel’s own instrument was put to a fourth and larger model of a different family: same prompt, same criteria, same evidence, and only the model answering differs. It was asked about organizations the panel had left at maybe and about nothing else, so no verdict the panel settled was reopened. It answered 8,968 of 8,969, raising 3,259 to include and excluding 912.
Three deterministic rules then apply to the whole frame on the same terms. Dormancy removes an organization that filed a terminating 990-N postcard or has no filing in five years. Authoritative federal rosters admit State Department Reception and Placement affiliates and Department of Justice recognized organizations. Geography restricts the dataset to the fifty states and the District of Columbia. 19,079 organizations remain in the census file. A curation ledger of hand-judged removals, additions from the national immigration legal services directory, and the requirement of a geocodable address bring the published directory to 15,264.
How much of the population is missing is not known, and this page does not estimate it. An organization that files no informative program text, appears in no directory and has no findable website is outside the frame altogether, and that description fits the smallest and least formal organizations first. What is audited instead is treatment received: for every processing step the organizations owed the step are enumerated against the organizations that received it, any shortfall is listed as identifiers rather than summarized as a percentage, and the release does not build while a shortfall remains.
A website becomes a source for an organization only when a fact the organization itself filed with the IRS appears on the page. Ten channels implement that test: the Employer Identification Number, the filed street address and ZIP code, the filed street number and street name, the filed telephone number, a pair of filed officer names, a text shingle from the Form 990 narrative, the in-care-of name from the Business Master File, an officer named on a 990-N postcard, the website the organization filed on its own Form 990, and a name rare enough to identify the organization by itself.
Each channel was measured against matched controls before it was adopted, where a control pairs an organization with a different organization’s page. Any channel that fired on a control was discarded, so the ten in use have a measured false-positive rate of zero on that test. The test bounds one error, a page assigned to an organization it does not belong to, and it says nothing about the pages the channels miss.
Nothing read at an unproven page is published. When a page matched neither a filed fact nor the organization’s name, the address, the language-access reading, the social accounts and the services taken from it are withdrawn together, because every one of them would otherwise describe a different organization. Each published record states its own evidence state: confirmed by a filed fact, attributed through a shared host with a confirmed parent organization, read but matched nothing, found but unusable, host answered but unreadable, no website found, or no website.
Web addresses were located in two stages of rising cost. Grounded search poses a plain question with web search enabled and then chooses among the links the search returned, in a second call with search disabled, so that an address cannot be invented. An agentic loop runs on the organizations grounded search failed on, reading pages, following aggregator listings because those print an address, and abstaining when it cannot tell a local chapter from the national body. On 100 randomly drawn organizations the agentic loop was right for 90 against 74.5 percent for grounded search, and 83 percent of the addresses it produced across the whole frame were later confirmed by the attribution test.
Each proven website was fetched and read, and its language access classified into four states:
The distinction between an authored page and a widget is the one Rodriguez and colleagues (2019) drew by hand for hospital websites, applied here at national scale. Of the 13,137 published organizations with a proven page, 8,624 are English only. Among the 4,513 whose sites offer any non-English option, 1,171 do so through a widget alone.
The classification is made from the page text a visitor actually reads, not from what the page declares about itself. Each captured page is split into paragraph-sized chunks, every chunk is language-detected, and the shares are weighted by characters. Metadata alone is not enough: a site written entirely in Korean or Spanish often still declares English in its template, in the html lang attribute, its hreflang tags, or a language switcher.
The text signal may only raise a classification, never lower it. A capture is a handful of pages and usually lands on the English version of a site, so English-looking text is not evidence that a site’s other-language pages are absent. An organization with no proven page receives no label at all, rather than an English-only label.
Services were read from each organization's website and IRS Form 990 narrative by a large language model (Claude Opus), against a fixed category set grounded in Cordero-Guzmán (2005) on immigrant community-based organizations and cross-walked to the AIRS or 2-1-1 taxonomy and Open Eligibility. An organization can offer several. A missing tag means the evidence did not state the service, not that it is unavailable. There is no catch-all: a service that fits no category is left untagged rather than filed under Other, so a tag always means something specific.
The colours group the categories into the three functions of immigrant community-based organizations that are in scope for a US service directory. Immigration and legal status covers the immigration process itself; daily needs and services covers education, health, housing, food, work, money, and care; community and voice covers advocacy, civic participation, economic development, and culture. Cordero-Guzmán's fourth function, transnational activity toward the country of origin, is out of scope here.
| Legal / immigration | Immigration legal representation and consultation: asylum, visas, removal defense, DACA, VAWA, U and T visas. |
| Citizenship | Citizenship and naturalization preparation, N-400 help, and citizenship classes. |
| ESL / English | English as a Second Language instruction. |
| Education / youth | Academic education for children and youth: K-12 support, tutoring, after-school, mentoring, scholarships, college access, early childhood education. |
| Interpretation / translation | Spoken-language interpretation or document translation for the people served. |
| Case management | Case management, navigation, referrals, and wraparound support. |
| Refugee resettlement (R&P) | Formal refugee reception and placement in the first months after arrival, including sponsor and co-sponsor programs. General ongoing services are case management instead. |
| Housing | Housing assistance, shelter, and rental help. |
| Food assistance | Food pantry, meals, and groceries. |
| Emergency / basic needs | Emergency cash, clothing, utilities, and other basic needs other than food. |
| Benefits enrollment | Enrolling people in public benefits or insurance: SNAP, Medicaid, marketplace plans, WIC, cash assistance. |
| Health / medical | Direct clinical care: a clinic, medical, dental or vision services, screenings by clinicians, medication or treatment. |
| Health education | Health education and promotion as a named activity: classes and workshops, community health workers or promotores, help navigating the health-care system. |
| Mental health | Counseling, psychological or behavioral health care, and trauma services. |
| Employment / job training | Job placement, workforce and job training, career services. |
| Economic development | Small-business and entrepreneurship support, microloans, community economic development. |
| Financial literacy | Financial education, banking access, tax help, VITA. |
| Digital literacy | Computer, technology, and digital-skills training. |
| Civic engagement | Voter registration and mobilization, census outreach, civic and leadership development, know-your-rights education. |
| Advocacy / policy | Policy advocacy, legislative campaigns, immigrant-rights litigation, and organizing aimed at systems change. |
| Domestic violence services | Services for domestic or intimate-partner violence. |
| Family reunification | Family-based petitions, sponsorship, and reuniting separated families, including unaccompanied minors. |
| Childcare | Custodial childcare and daycare, including care during programs. Preschool academics are education instead. |
| Transportation | Transportation assistance. |
| Cultural programming | Cultural festivals, celebrations, heritage events, and arts programming run by the organization. |
| Heritage language | A heritage or community-language school teaching the community its own language, such as a Korean weekend school or a Chinese school. Distinct from ESL. |
Beyond services, the model coded the following organizational attributes from each website.
| Population focus | children, youth, unaccompanied_minors, older_adults, women, men, lgbtq, people_with_disabilities, survivors, undocumented, veterans, farmworkers, day_laborers, domestic_workers, workers | A sub-population the organization names as a priority BEYOND serving immigrants generally, through a dedicated program or in its own mission. Empty means it serves the general immigrant population. |
| Community served | e.g. Vietnamese, Afghan, Latino, East African | The ethnic or national-origin community the organization primarily serves. Occupation, income, and religion are excluded. |
| Languages served | list | Non-English languages the organization states it provides services in. A translate widget on the website does not count. |
| Website language access | true_multilingual, machine_translate, machine_translate_error, english_only | Measured from the captured page text, not from what the page declares: paragraphs are language-detected and weighted by characters. A machine-translation widget is recorded as such and is not counted as in-language content. An organization with no proven page receives no value. |
| Grassroots | true when found | The organization describes itself as grassroots, community-led, or mutual aid on its own site. |
| Interpretation offered | yes when found | The site states the organization provides interpretation or translation to the people it serves. Interpreter job postings and training courses do not count. |
| Faith-based | yes when found | The organization presents itself as faith-based, in its name or in a sentence about itself. A verse on the page or a partner congregation does not count. |
| Federal recognition and funding | eoir_recognized, orr_resettlement, orr_unaccompanied_children, uscis_citizenship, dol_farmworker, ovc_trafficking, ovw_culturally_specific | Taken from federal records, not from the website: the DOJ EOIR roster of recognized organizations, and USASpending award recipients by program. |
| Federal provider directories | samhsa_mh, samhsa_sud, hrsa_hc | Whether a federal facility inventory lists the organization, matched on the name plus a second agreement (ZIP, city, street address or telephone) and never on the name alone: the SAMHSA mental health and substance use treatment locators (2025) and the HRSA Health Center Service Delivery Sites. The health and mental health service tags are read from the organization’s own website or Form 990, so this is the one place a second party confirms them. Neither inventory is a licence and neither is complete, and an organization absent from both has not been contradicted. |
| Founding year / latest filing | year | IRS ruling year; for a subordinate covered by its parent organization's group exemption, the year of formation reported on Form 990, or blank. Latest filing: the most recent filing across the 990-N e-Postcard and the full-990 e-file corpus. |
The directory lists nonprofits registered in the United States that provide a service or an active community function to immigrants, refugees, or a specific ethnic-immigrant community here. Several categories are deliberately excluded, and each exclusion was applied to every organization in the frame rather than case by case.
What each filter shows when you turn it on.
| Has website | Shows only organizations that have a website. |
| Service tags | Shows organizations that offer the service you pick. Choosing several shows organizations offering any of them. |
| Language served | Shows organizations that provide services in the language you pick. |
| Community served | Shows organizations that serve the ethnic or national-origin community you pick. |
| Web language access | Shows organizations whose website reaches non-English speakers the way you pick: truly multilingual, an auto-translate widget, or English only. |
| Population focus | Shows organizations that focus on the group you pick, such as youth, women, or survivors of violence. |
| Grassroots or community-led | Shows organizations that describe themselves as grassroots or community-led. |
| DOJ-recognized legal provider | Shows organizations recognized by the Department of Justice to give immigration legal help. |
| Federal funding received | Shows organizations that receive federal funding for immigrant or refugee services. |
Foreign-born population and share: ACS B05002. Countries of origin: ACS B05006. Foreign-born by race and ethnicity: ACS B05003. Limited-English speakers by detailed language: ACS B16001 and grouped ACS C16001. Poverty by nativity: ACS B17025. Health insurance by nativity: ACS B27020. Education by nativity: ACS B06009. Median income by place of birth: ACS B06011. Race and ethnicity: ACS B03002. Vehicle availability: ACS B08201. Additional well-being measures compared by nativity (Medicaid, disability, home internet and computer, SNAP, homeownership, vehicle access, housing cost burden, crowding, and educational attainment) are computed from ACS Public Use Microdata.
Access to organizations is measured with the enhanced two-step floating catchment area method (Luo and Qi 2009). Demand units are census tracts (foreign-born population, or limited-English speakers of a language, or those in no-vehicle households); supply is the 15,014 geocoded organizations, each given equal capacity because capacity is not observed. Distances are driving distances on the road network within each state, from OpenStreetMap roads with Dijkstra shortest paths, for every layer at every level, including the ZIP-code view. Because the road network is built one state at a time, an organization is matched only to demand in its own state. A place with no road path to any organization is shown as unreachable, that is, as a service desert. Both use a 30-mile catchment with step weights of 1.0, 0.68, and 0.22 across 10, 20, and 30 miles; no-car layers use a roughly 10-mile walk or transit catchment. Tract scores are aggregated to county, metro, and state as a demand-weighted mean, and a ZIP-code (ZCTA) view is computed directly from ZIP-code Census demand. Access is computed for:
Results are shown at the ZIP-code, county, metro, and state level (ZIP uses ZCTA-level Census demand). On the Access map a service desert is an area with people who need a service but no organization offering it within reach (shown in orange); areas with no foreign-born population are shown in gray. Public-transit travel time is a planned refinement, and the no-car layers approximate transit dependence with a shorter catchment.
The Telehealth filter of the mental-health Access map counts the clinicians a resident could see by video, where distance does not matter and licensure does. A clinician treats a client where the client is, so each clinician’s catchment is the set of states where they may practice: their own state and, with the compacts, every other member state of a compact that covers their license when their own state is a member. Private supply consists of the 2025 Psychology Today listings that offer online sessions, each in the state of its listing. Public supply consists of the SAMHSA facilities of the in-person map whose 2025 National Directory entry lists telemedicine or telehealth therapy (service code TELE), each in the state of its location. That flag was linked to 11,962 of the 11,964 facilities on the in-person map, 11,947 of them through identical geocoded coordinates and identical language and Medicaid codes and 15 through the nearest record with the same codes; the 2 not linked count as not offering telehealth. Facilities have no clinician license and serve their own state only.
Three scenarios are shown. In-state only allows no practice across state lines. Current compacts uses the compacts in operation on 28 September 2026, read from their official sites that day: PSYPACT for psychologists in 43 of the 50 states and DC, and the Counseling Compact among the 9 states that issue privileges (Arizona, Arkansas, Georgia, Indiana, Louisiana, Minnesota, Ohio, Tennessee and Wyoming). The Social Work Licensure Compact had been enacted in 35 states and had issued no multistate license, and no compact existed for marriage and family therapists. Proposed compacts (all) takes the February 2026 membership coded by Yoo and Weaver (2026, American Journal of Public Health) together with the compacts in operation on 28 September 2026, so Alaska, whose PSYPACT membership took effect on 20 September 2026, is included, and treats every compact as in operation: PSYPACT (43), the Counseling Compact (39), the Social Work Licensure Compact (31) and, for marriage and family therapists, the Access MFTs portability framework (26). California has joined no compact but is in that last channel, so under this scenario its licensed marriage and family therapists are shared with the residents of the other member states, and its value falls while theirs rise.
The license decides the compact. Where a listing’s title states a license type, only the compact that covers that type applies: PSYPACT for a psychologist, the Counseling Compact for a professional counselor (LPC, LPCC, LMHC and equivalents), the Social Work Licensure Compact for a social worker (LCSW, LMSW, LICSW and equivalents) and the marriage and family therapist channel for an LMFT. An associate, intern or limited license, an addiction counselor license or a medical license opens no channel across state lines. Where the title states no license type, the profession flags of the listing file apply, as in Yoo and Weaver (2026). Of the 297,957 listings that offer online sessions, 251,592 state a license a compact covers, 39,938 state one no compact covers and 6,427 state none, 1,460 of which have a profession flag.
Access is computed with the two-step floating catchment method on these state catchments, counting only the residents who can see a clinician by video from home. That takes a computer (the Census counts a smartphone or tablet) and a broadband subscription (fixed, or a cellular data plan), so each group’s demand in a census tract is its population times the share with both, from the American Community Survey 2024 five-year estimates: table B28008 for all residents, and B28009B, C, D, E and I for Black, American Indian and Alaska Native, Asian, Native Hawaiian and Pacific Islander, and Hispanic residents. Each clinician’s capacity is divided among the residents with broadband in every state where the clinician may practice, and a resident with broadband scores the sum of those shares per 100,000; a resident without it scores 0. A tract’s value is therefore its broadband share times its state’s score, and a county, metro area or state takes the population-weighted mean of its tracts. ZIP codes take the same formula on the ZIP code units of the in-person map, with their own broadband shares from the same tables. Demand and the language, race-focus and Medicaid subsets are those of the road-network access measures, and there are no no-car variants.
The survey reports no broadband share by ability to speak English, so the limited-English language groups take the share of all residents of their tract or ZIP code. The race tables count each race alone of any ethnicity while the demand counts the non-Hispanic group, so those shares include Hispanic members of the race. The survey asks about households; people in group quarters, such as dormitories, prisons and nursing homes, take their state’s household share. The compacts and broadband move capacity between places without adding any, so the national average is the same in every scenario, a state with more clinicians per resident than its compact partners scores lower with them, and under in-state only a state’s value is the same with or without broadband. An area with fewer than 100 residents in households to measure its broadband share on (of the group itself, for a race group) takes its state’s share, and a metro area that crosses a state line applies this to each state part. A value above 0 is withheld where fewer than 300 people of the group live in the states that share clinicians with it. The scenarios assume that every eligible clinician holds the privilege and that the state of a listing is the clinician’s home state.
What has been measured, and what has not. Three parts of the construction were measured against a standard outside themselves. The ten attribution channels fire on 0.0 percent of matched controls. The agentic search stage was right for 90 of 100 randomly drawn organizations, against 74.5 percent for grounded search. On 54 organizations known from hand review to have been excluded in error, the most permissive panel seat recovered 66.7 percent and the least permissive 29.6 percent, so on that set permissiveness and accuracy moved together. None of the three is a substitute for human coding of the service and language labels. A stratified sample of 400 organizations, a coding manual and a blind coding sheet are prepared; completed sheets from two coders, an adjudication of their disagreements and the resulting agreement and accuracy tables are not. Per-category precision and recall against trained human coders are therefore not available, and every service and attribute tag should be read as machine-assisted rather than hand-checked.
Coverage. Coverage of the frame is bounded and the bound is not known. An organization that files no informative program text, appears in no directory and has no findable website never entered the frame, and that population overlaps with the smallest and least formal organizations, which is exactly where this dataset can measure least. A capture-recapture design against a second independent enumeration of a defined geography is the next step and has not been done.
Blank is not zero. A service language is recorded for 6,649 of the 15,264 published organizations and a population focus for 7,385, and the blanks are not random: they concentrate among organizations with no proven page, which are the organizations that file least. A blank language field records an absence of evidence and is not a statement that the organization serves in English. Nothing in the dataset is imputed for that reason. Any analysis of the language or population fields should be restricted to organizations with a proven page or should model the missingness, and the language layers of the accessibility analysis inherit the same undercount.
What a service label means. A label describes what an organization advertises or reported to the IRS, which is not the same as what it delivers today. The source of every label is recorded per organization for that reason, so an analysis can be restricted to labels read from a proven website.
Other bounds. 15,014 of the 15,264 published organizations have a map location; the rest are searchable in the directory and absent from the map. Locations come from IRS addresses, and where only a ZIP code was filed the record is placed at the ZIP centroid and flagged as approximate. The accessibility analysis for organizations gives every organization equal capacity, because capacity is not observed. Distance is driving distance on the road network for every layer at every level. Per-state road graphs also mean an organization is matched only to demand in its own state, so an organization just across a state line does not count toward a border tract's score. Medicaid acceptance reflects what each provider lists. Always confirm details directly with an organization before relying on a listing.
Nari Yoo, University of Michigan School of Social Work. Questions and corrections: Suggestions and corrections page. Organization data derive from public IRS and federal records and the US Census ACS.
Created and maintained by Nari Yoo, PhD, University of Michigan School of Social Work.
Collaboration and data-access inquiries are welcome: nariyoo@umich.edu.
Immigrants and refugees in the United States depend on a dispersed patchwork of nonprofit organizations for legal aid, resettlement, English classes, health and mental health care, and basic needs. There has been no national, openly available map that identifies these organizations, describes what each one actually does, and records whether they can serve people in languages other than English. Administrative records cannot answer these questions on their own: the IRS classification system undercounts immigrant- and refugee-serving organizations and is silent on the services they deliver and the languages they use (Gleeson & Bloemraad, 2013). The inventories that do exist tend to cover a single domain, such as immigration legal services, or rely on hand coding that does not scale to the whole country.
This project addresses those gaps. It assembles a national, geocoded census of immigrant- and refugee-serving nonprofits from IRS records, Form 990 program text and federal and sector directories, and screens every candidate with a panel of language models reading the evidence that candidate actually has. A language-model layer then reads each organization’s website to tag the specific services it offers and to audit its language access, separating genuinely multilingual sites with professionally translated content from those that only add an automatic-translation widget, and recording the languages each organization states it serves. Producing these organization-level service and language-access measures at national scale is the dataset’s distinctive contribution; they are not derivable from administrative data and have not previously been compiled into a single open resource.
The supply side is then placed next to demand. Each organization is linked to Census American Community Survey measures of the foreign-born population, its growth over time, and limited-English proficiency by language, at the state, county, city, and metropolitan level. Seeing provision and capacity alongside where immigrants live and which languages they need makes it possible, for any place in the country, to identify both who provides help and where the gap between need and available, language-accessible help is widest.
It is built for three audiences. Newcomers and the people who assist them can see what resources exist in a new place, identify the organizations that provide a specific service, and use the directory to make referrals in the language a client needs. Practitioners and funders can see where coverage is thin relative to need. Researchers and policymakers get a reproducible, openly documented dataset that links the immigrant nonprofit sector to local linguistic demand.
The dashboard now extends beyond the directory. The Access map measures spatial access to organizations, and to public (SAMHSA) and private (a private practice directory) mental-health care, with a two-step floating catchment method, by language, for households without a car, and for providers that accept Medicaid. The Policy map carries a state Immigrant Inclusion Policy Index (immigrant-specific inclusive minus restrictive laws across health, language, education, identification, and legal protection), and the Disparity map shows a companion Immigrant-Native Disparity Index of foreign-born versus US-born outcome gaps. A Language access guide explains the right to free interpreting and translation in health and mental-health care, and a grounded assistant answers questions from the underlying data.
Loading coverage figures…
A national directory and dashboard of US nonprofit organizations serving immigrants, refugees, asylees, and other foreign-born newcomers, paired with Census foreign-born, language-need, and immigrant social/economic context at state, county, and city levels.
Every demographic figure is foreign-born specific and available at state, county, city, and metro levels, with US, Census-region, and state averages shown as benchmarks. Time-series (2009–2024 / 2011–2024) cover foreign-born share, language need by language, immigrant and racial/ethnic diversity, and residential segregation.
Loading coverage statistics…
Yoo, N. (2026). Immigrant Support Atlas [Dashboard and data]. University of Michigan School of Social Work. https://www.immigrantmap.us
The data are released under CC BY 4.0 and may be reused with this citation. When quoting a figure, also name the page it came from and the data year printed beside it, such as the ACS 2020-2024 5-year estimates. For the organization records as a file, write to nariyoo@umich.edu.
General educational information only, not legal, medical, immigration, or mental-health advice; using it creates no attorney-client or provider-patient relationship. Compiled from public sources and may be incomplete or out of date, so verify with the official source before relying on any listing. Outside links are for convenience and do not imply endorsement, and individual providers are not vetted. Provided “as is,” without warranty. © 2026 Nari Yoo. Data CC BY 4.0.