240,000 people
About one in seven Philadelphians was born outside the United States. Across the metropolitan area, 468,039 residents were born abroad in 2022–2024. The largest groups were born in India (52,106), Mexico (35,992), China (35,265), the Dominican Republic (25,322) and Vietnam (17,363).
Asia and Latin America each account for roughly a third of the foreign-born. Europe and Africa make up most of the rest. Singer's 2004 typology filed Philadelphia under former gateways, cities that drew immigrants before 1930 and few afterward. Its immigrant population has grown again since.
The city's foreign-born population rose from 187,000 in 2013 to 240,000 in 2024, the highest of the twelve years. This is a story about what those people earn, and about whether living among other immigrants changes it.
Newer arrivals started closer
Every arrival cohort starts below comparable natives and closes the gap at about 1.3 percent a year. Immigrants who came before 2000 started 21 to 22 percent behind.
Those who arrived after 2000 started 17 to 18 percent behind. The difference of 4 to 5 points is about two standard errors (p = 0.02). It runs opposite to the declining cohort quality Borjas found for national arrivals of the 1950s to 1970s. Curves stop at each cohort's longest observed duration.
Where immigrants live
The area unit is the PUMA, about 100,000 people, larger than most cities. Immigrant density peaks in Northeast Philadelphia, where the Central and Lower Northeast, Far Northeast and North Delaware PUMAs are 28, 26 and 23 percent foreign-born. Next come two suburban PUMAs in Chester and Montgomery counties at 19 percent.
The same map can show where movers go. The circles show which suburban PUMAs received the foreign-born residents who left the city in the past year. The globe shows every country of birth. Chapter 10 asks who those movers are.
Immigrant geography by county
Each rose is a county and each petal one measure, all percentages on the same scale, 0 at the center and 100 at the outer ring. Philadelphia County holds the most foreign-born residents and the highest foreign-born percentage, 15.5 percent, and 31 percent of them speak limited English. Its foreign-born median wage is 21.48 dollars against 27.62 for the native-born, 78 percent, and a third of foreign-born adults hold a degree.
Salem County in New Jersey is the other end. Nine percent foreign-born, but 41 percent of them limited-English and 15 percent with a degree, and a foreign-born median wage of 17.43 dollars against 26.86 for the native-born, 65 percent, the lowest ratio of the eleven. Camden County sits between the two, with the same limited-English percentage as Philadelphia and the same wage ratio of 78 percent.
Chester County is where immigrant and native wages are both highest, 35.32 and 39.05 dollars, a ratio of 90 percent, and 60 percent of foreign-born adults hold a degree. In Bucks and Burlington counties the two medians are within a dollar of each other, ratios of 100 and 98 percent.
The wage gradient by English
Among foreign-born workers in 2022–2024, median hourly pay rises from 15 dollars for those who speak no English to 18 for "not well," 23 for "well," and 36 for "very well," and the 90 percent intervals do not overlap between adjacent categories. The step from "well" to "very well" is larger than the two steps below it.
Part of that is composition. Workers who speak English very well are also more educated and have been in the country longer. Holding those things, arrival cohort, occupation and area constant, each step up the four-point scale is still associated with about 6 percent higher pay, 0.059 log points in the main specification.
Between natives and immigrants the raw hourly gap is about 11 percent. Age, education and family structure explain a quarter of it. Add English and they explain 72 percent. What is left is three percentage points.
Now the same wages by area. Foreign-born median wages are highest in western Delaware County, 59 dollars an hour, and northern Chester County, 54 dollars, and no other PUMA exceeds 44. The wage map is close to the inverse of the density map. Each dot is one of the 48 PUMAs, sized by limited English among its foreign-born, and the correlation between foreign-born percentage and foreign-born median wage is −0.37. Hover a dot or drag a box across the chart.
A two-level model nests 7,355 foreign-born workers in the 48 PUMAs, with a random intercept and a random limited-English slope for each PUMA. For immigrants who speak English well, hourly pay is about 8 percent lower per standard deviation of area immigrant density, a step of 5.4 percentage points, with a standard error of 1.8. The band is the interval of that slope. Move the mouse across the chart to read the fitted wage at any density.
Limited English lowers the log wage by 0.234 at mean density, a differential of 21 percent. For immigrants with limited English the density slope is about 3 percent and not distinguishable from zero. The difference between the two slopes is the interaction the protective and isolation views argue about. It is 0.049 with a standard error of 0.029 (p = 0.09), so the gap narrows from 25 percent one standard deviation below the mean to 13 percent two above it, but the interval does not exclude zero. The 2017–2021 window, dashed, gives 0.050 with a standard error of 0.028. The interval rules out a large isolating effect and does not rule out a narrowing of up to about 10 percent per standard deviation.
Is this just cheap housing? Cheaper areas hold both more immigrants and lower-wage workers of every origin. The same model on the 47,893 native-born workers in the same 48 PUMAs gives a gradient of 2.0 percent per standard deviation, with a standard error of 1.1. Adding the PUMA's native median wage as a control leaves the immigrant gradient at 6.4 percent and moves the interaction to 0.059 (p = 0.05). About a quarter of the gradient is shared with natives or absorbed by the area's wage level. The rest is specific to the foreign-born.
Two lines, one occupation
The vertical distance between the lines is the limited-English differential at mean density, 21 percent. Both lines fall with density, the proficient line faster.
Compare workers in the same occupation and the differential falls to 9 percent. Holding occupation constant removes about 60 percent of the differential, 40 percent of the density gradient, and all of the interaction. A limited-English worker earns 9 percent less than a proficient worker in the same occupation and the same area, and 21 percent less overall, because limited English is associated with sorting into lower-wage occupations. Drag the slider to read both gaps at any density.
The same two panels for the New York metropolitan area, from the same sample rules and specifications. The limited-English differential there is 20 percent, and 10 percent within an occupation, the same split as in Philadelphia. The occupation check is descriptive. Occupation is itself an outcome of English, so the change in the coefficients measures how much of the differential is associated with occupational sorting, not a causal share.
Which occupations limited English sorts into
The fixed effects say most of the differential is occupational. This is what that looks like. Thirty-seven percent of limited-English workers are in service occupations and 14 percent each in production and in transportation, against 12, 3 and 6 percent of the native-born. Proficient immigrants cluster where the native-born do, in management and in computing, engineering and science, and in the last of those they hold a larger proportion than the native-born.
Inside an occupation the gap is smaller but does not vanish. The right column divides the limited-English median by the proficient median in the same occupation group. In service work the two groups earn within 17 percent of each other. In sales and in construction the limited-English median is under three quarters of the proficient one. These are descriptive medians for the same sample the models use, not model estimates.
Own-language density
Immigrant density as a whole may be the wrong measure. An exploratory check uses the density of the worker's own home language instead, for the 5,732 workers who speak a language other than English at home. Its five highest values are all Spanish. In North Philadelphia 41 percent of residents aged 5 and over speak Spanish at home, in the city of Camden across the river 38 percent, in the Central and Lower Northeast 22 percent, in the Salem and Cumberland PUMA around Bridgeton 18 percent, and in the North Delaware section of Philadelphia 17 percent.
On this measure the differential does narrow. The interaction is 0.046 with a standard error of 0.022 (p = 0.04), and 0.021 with occupation fixed effects. But look at which line moves. Limited-English wages do not vary with own-language density, minus 0.011 per standard deviation with a standard error of 0.016. Proficient wages fall with it, by 0.057 with a standard error of 0.017. The differential narrows because proficient workers earn less where their language is common, not because limited-English workers earn more. Generic density is held constant in the same model and keeps a slope of minus 0.061 for both groups.
The same model in the New York metropolitan area, for 54,089 workers with a non-English home language in 141 PUMAs, gives an interaction of 0.045 with a standard error of 0.009, and 0.028 within occupations, against 0.046 and 0.021 in Philadelphia. The interactions agree. The proficient slopes do not, at minus 0.032 in New York against minus 0.057 in Philadelphia per standard deviation, and further apart per percentage point, since a New York standard deviation is 17.8 points against 7.7. In both areas the narrowing comes from the proficient side.
One account is selection. Higher earners leave the city, and if proficient immigrants with higher earning capacity are the ones who leave areas where their language is concentrated, the proficient immigrants who remain are negatively selected and the pattern follows without any effect of the area on pay. A direct check uses the 4,370 proficient workers who lived in the metropolitan area a year earlier, of whom 98 had left their prior migration area. The leavers earn 6 percent less than stayers, with a standard error of 18 percent, and 13 percent less within occupation, with a standard error of 12. Nothing here indicates that leavers earn more, and the intervals leave a difference of about 35 percent in either direction inside them, so the check is too imprecise to test the account. A further caveat: in North Philadelphia and Camden, 77 and 64 percent of Spanish speakers are native-born, so the measure there describes a largely native-born linguistic community.
Philadelphia against New York
Would a traditional gateway show the same pattern? The same sample rules, variables and specifications are applied to the New York-Newark-Jersey City metropolitan area, with the five boroughs as the city. It is a comparison, not part of the design, and the Philadelphia conclusions do not depend on it. New York holds 68,058 foreign-born workers in 141 PUMAs, and immigrant density there averages 30 percent of PUMA residents with a maximum of 64, against 12 and 28 in Philadelphia. The limited-English differential is 20 percent in New York and 10 percent within an occupation, the same split as Philadelphia's 21 and 9. Wages fall with density for proficient workers in both areas.
Does the differential change with density? On generic density the two areas give interaction terms that are marginal or null in both and differ by about one standard error. On region-based co-ethnic density they differ by about two standard errors, with opposite signs, minus 0.033 with a standard error of 0.024 in Philadelphia against 0.018 with a standard error of 0.009 in New York. Neither area supports the protective view on the generic or the co-ethnic measure.
On own-language density the interactions agree, 0.046 in Philadelphia against 0.045 in New York, and 0.021 against 0.028 within occupations. The New York estimates are five and four standard errors from zero and are the only ones among the seventeen interaction terms the paper reports that would survive a correction for that number of tests. Philadelphia's own-language interaction, at p = 0.04, does not. In both areas the narrowing comes from the proficient side. A test that fixes own-language density as the main measure in advance and runs it across several metropolitan areas is the design these results call for.
Who leaves the city
The one-year migration question in the ACS records where each adult lived a year earlier. Across the three windows the pattern of moves among foreign-born adults aged 25 to 64 barely changes. Moves within the suburbs stay near 6 percent of foreign-born adults and moves within the city between 2 and 3 percent. Arrivals from abroad were 3.3 percent in 2012–2016, 2.0 percent in 2017–2021 and 2.7 percent in 2022–2024. Moves from the city to a suburb ran at 0.5, 1.0 and 0.7 percent, and moves from a suburb into the city at 0.6, 0.5 and 0.6 percent.
In 2022–2024 the flows are small but not random. The 2,400 a year who moved from the city to a suburb earned a median of 33 dollars an hour. The 1,900 who moved the other way earned 20. New arrivals from abroad landed in the city and the suburbs at 3,400 to 5,500 a year, earning 16 and 22 dollars. Hover a ribbon for the count and the movers' median wage.
The model below pools the three windows and colors the ribbons. A worker whose hourly wage is one log unit higher is 1.30 times as likely to move from the city to a suburb, with an interval from 0.97 to 1.74, and 0.71 times as likely to move within the city, with an interval from 0.61 to 0.83. Among prior city residents, doubling hourly pay raises the odds of leaving the city by about 40 percent. Immigrants who moved from the city to a suburb earn 5 percent more than observationally similar stayers, with a standard error of 6 percent, an interval too wide to detect selection on unobserved earning capacity. Switch the covariate to English, a graduate degree or marriage to recolor the ribbons.
Put the stayers back and the flows shrink to their true size. In 2022–2024, 85.7 percent of foreign-born adults lived in the same house as a year earlier. Of the remainder, 5.9 percent moved within the suburbs, 2.7 percent arrived from abroad, 2.3 percent moved within the city, 2.1 percent arrived from another state, 0.7 percent moved from the city to a suburb and 0.6 percent from a suburb into the city.
Native-born adults move between the city and the suburbs at nearly the same rates, 0.6 and 0.7 percent. What differs is arrival from abroad, 2.7 percent of foreign-born adults against 0.2 percent of the native-born, and it lands in the suburbs more often than in the city, 5,500 to 3,400 a year.
Cohorts over twelve years
Following the same arrival cohorts across three survey windows, real median wages rose in eleven of the twelve cells observed in all three windows, by about 10 dollars for Asian-born arrivals of the 1990s and 2000s and by 1 to 8 dollars elsewhere, and were flat for African-born arrivals of the 1990s. The suburban percentage moved much less. For cohorts that arrived before 2010 it sits between 59 and 86 percent in every window. The cell fixed-effects regression gives 9.7 percentage points of suburbanization per log point of wage growth with a standard error of 10.7, so the data cannot detect a group-level link between wage growth and suburbanization.
The 2010s arrivals look different, but part of that is duration. Compare them with the 2000s cohort at the same number of years after arrival and one difference holds. Latin American arrivals of the 2010s are 7 points less suburban, a difference of about two standard errors. European arrivals show the same size with too few respondents to establish it. Asian arrivals are 5 points more suburban.
The same windows by country of birth
The paper's cells group arrivals by region of birth and arrival cohort. Pooling the cohorts instead gives 53 countries of birth with at least 50 respondents in a window, from the same sample of foreign-born workers aged 25 to 64. Each row is a country, the three columns are the survey windows, and color is the real median hourly wage in 2024 dollars. Among Asian-born workers the median ranges from 19 to 26 dollars for Bangladesh to 44 to 53 dollars for India and Taiwan. Among Latin American-born workers it runs from 15 to 20 dollars for Mexico, Guatemala and El Salvador to 23 to 30 dollars for Jamaica and Trinidad and Tobago. Countries are ordered by weighted population within each region. Hover a cell, or click a country for its three windows.
Color the same grid by the percentage with limited English and the low-wage rows light up. Between 59 and 62 percent of Guatemalan-born workers report limited English in every window, 42 to 67 percent of Honduran-born, 40 to 53 percent of Salvadoran-born, 38 to 51 percent of Mexican-born and 44 to 47 percent of Dominican-born. Among workers born in Canada, England, Ireland, Jamaica, Trinidad and Tobago, Guyana, Ghana and Nigeria, fewer than 2 percent do in any window.
Color it by the percentage living outside Philadelphia County and the ordering changes again. Workers born in Mexico and India are the most suburban, 86 to 91 and 85 to 87 percent outside the city in every window. Those born in Albania, Cambodia, Brazil, Haiti and the Dominican Republic are the most concentrated in the city, with between 24 and 64 percent outside it. Wage and residence do not line up: the Mexican-born and the Indian-born sit at opposite ends of the wage grid and at the same end of this one.