Learning a Language Pays Off — Just Not the Way You Think
Second-language learning delivers large economic gains for immigrants mastering a host language but shows little broad cognitive advantage and may reduce US earnings, while programs carry high dropout rates and demand hundreds to thousands of hours.
What documented cognitive, economic, and social effects does second language acquisition have on individuals across different age groups and contexts?
- 1Dutch proficiency raises immigrant employment probability by roughly 30 percentage points, and each level of Spanish fluency in Spain adds about 3.2 percentage points.
- 2In the US labor market, bilinguals earn 0.9 percentage points less than monolinguals after controlling for hours worked and industry, according to a peer-reviewed analysis of American Community Survey data.
- 3A meta-analysis of 99 studies found only about one in seven comparisons showed a bilingual cognitive advantage, while roughly four in five were null.
- 4Bilinguals show dementia symptoms about 4 to 5 years later than monolinguals, the most consistent cognitive finding, though it rests on a single source.
- 5Nearly a quarter of French-immersion pupils drop out before fifth grade, and foreign-language teachers quit at around 44 percent over five years.
Second-language acquisition delivers a small set of well-documented effects that depend heavily on who is learning, which language, and in what context. The economic returns are large and consistent for immigrants mastering a host country's dominant language — Dutch proficiency raises the chance of employment by roughly 30 percentage points, and each level of Spanish fluency adds about 3.2 percentage points — but in the US labor market bilinguals generally earn the same or slightly less than monolinguals after controlling for skills and industry. The broad claim of a cognitive advantage is contradicted by most evidence: a meta-analysis of 99 studies found only one in seven comparisons showed any advantage, while roughly four in five returned null. The clearest cognitive benefit is narrow and late-life: bilinguals show dementia symptoms about 4 to 5 years later, based on one source. Language programs carry high costs rarely disclosed in the sales pitch: nearly a quarter of French-immersion pupils drop out before fifth grade, foreign-language teachers quit at around 44 percent over five years, and reaching proficiency takes 600 to 2,200 hours. The popular narrative treats these as a single story of universal uplift; the evidence says they are separate outcomes for different people, and conflating them leads to mistaken expectations.
The Full Investigation
8 sections · 10 min read
Confirmed facts and attributed reporting read normally; only contested, unverified, or speculative sentences are highlighted. Hover any sentence for its grade and sources.
A benefit everyone agrees on, until you read the fine print
Language education is sold as one of the safest investments a person can make - sharper thinking, higher pay, easier belonging in a new country. The evidence libraries that supply that pitch are real and deep. But those same libraries hold the receipts: studies showing bilinguals earning less than monolinguals, cognitive tests coming back flat, and immersion classrooms losing a quarter of their pupils before fifth grade. This report tests the sales pitch against its own sources.
The research spans three very different worlds that are easy to blur together. One is cognitive science, asking whether juggling two languages reshapes the brain. Another is labor economics, asking whether a second language shows up in a paycheck. A third is immigrant integration, asking whether newcomers who master the local tongue find work and belonging. The findings in one world rarely transfer cleanly to another, and much of the confusion in public debate comes from treating them as a single story.
One early warning colors everything that follows. A 2015 study by Angela de Bruin and colleagues found that conference abstracts fully supporting the bilingual advantage were more than twice as likely to be published as those that challenged it. Whether that early tilt still distorts the literature is itself a live question this report returns to. The point for now is simple: in this field, the published record and the full record are not always the same thing.
The broad cognitive advantage is mostly missing when researchers go looking
If bilingualism reliably sharpened the mind, the effect should show up when researchers pool hundreds of comparisons. Mostly it does not. A meta-analysis by Kenneth Paap and colleagues, published in a Cambridge University Press journal, examined 99 studies covering 177 separate comparisons: only about one in seven - 14.7 percent - showed a bilingual advantage, while 81.4 percent were null and a small remainder actually favored monolinguals. A companion study of 320 University of Alberta undergraduates went further, finding monolinguals were faster than bilinguals on inhibition and shifting tasks.
The individual findings tell a mixed, task-specific story rather than a clean win. A Frontiers in Psychology brain-imaging study found that under emotional interference, proficient bilinguals showed a significantly smaller early-stage neural response - an N1 amplitude of -2.39 microvolts against -4.02 in less proficient bilinguals - which the authors read as more efficient early processing. But the same study found no significant bilingual advantage at the later, decision-making stage of the task. An early flicker of efficiency did not carry through to performance.
The one meta-analysis in this evidence set that reports an overall advantage is both weak and modest. It pooled 32 studies from 2018 to 2024 with 3,875 participants and found a small positive effect on executive function, a Hedges' g of 0.18 - a figure so small that even if real it is close to trivial. That study comes from a journal of unclear credibility, so it carries far less weight than the peer-reviewed work pointing the other way. Taken together, these origins diverge sharply, and the strongest of them lean toward no broad advantage.
2015-01-01
- de Bruin et al. (2015) published finding that conference abstracts fully supporting bilingual advantage hypothesis were more than twice as likely to be published compared to those challenging it
- Georgia's workforce report found foreign language teachers had highest attrition rate (44% after five years) among high school subjects
2020-01-10
- Alzheimer's Research & Therapy published study finding bilinguals showed no significant brain atrophy or cognitive decline over ~7 months while monolinguals did
2023-01-01
- Re-analysis of de Bruin et al. publication bias study found 82.35% of null-result and 82.5% of negative-result abstracts were published by 2023 once grey literature included
2023-03-31
- Frontiers in Psychology published ERP study finding proficient bilinguals had significantly smaller N1 amplitude than non-proficient bilinguals under emotional interference
2023-06-29
- PLOS ONE published analysis of ACS data finding bilinguals earned 0.9 percentage points less than monolinguals after controlling for hours worked and industry
2023-10-01
- Behavioral Sciences (MDPI) published re-analysis challenging publication bias claims in bilingual advantage research
2024-01-01
- Meta-analysis of 32 studies (2018-2024) with 3,875 participants found small positive effect of bilingualism on executive function (Hedges' g = 0.18)
2024-04-25
- Bilingualism: Language and Cognition published study of 320 University of Alberta undergraduates finding monolinguals faster than bilinguals on inhibition and shifting tasks
Open: Whether pre-registered replication studies using standardized executive-function batteries across diverse ages, language pairs, and socioeconomic backgrounds would confirm or erase the small effects that survive.; Whether the divergence between studies reflects genuine task-specific effects or differences in sample age and proficiency that the evidence set does not resolve.
The one strong cognitive benefit is narrow, geriatric, and rests on a single source
Where the executive-function story falls apart, one health finding holds firm. Research published in Alzheimer's Research & Therapy reports that the most consistent result across studies is a delay in the clinical onset of dementia among bilinguals of about 4 to 5 years compared with monolinguals. In a follow-up averaging roughly seven months, monolinguals showed significant brain atrophy and cognitive decline while bilinguals did not.
This is the affirmative case's most durable claim - and also its most isolated. Within this evidence set the 4-to-5-year figure appears in only one source, with no independent corroboration. It is a specific outcome in an aging population, not a general boost to thinking. It does not rescue the broad executive-function claims contradicted by the meta-analytic evidence; a benefit that shows up decades later in a subset of elderly people is a different thing from the everyday mental edge the popular pitch implies.
Open: Whether the 4-to-5-year dementia delay would replicate in independent samples, since the estimate has no corroborating source in this evidence set.; Whether the delay reflects bilingualism itself or unmeasured differences in education, migration, or lifestyle among the people studied.
In the US, bilinguals earn slightly less - not more - once you control for skills
The paycheck is where the sales pitch meets the hardest resistance, and the resistance comes from the strongest sources. A PLOS ONE analysis of American Community Survey data found that after controlling for hours worked and industry, bilinguals earned 0.9 percentage points less than monolinguals, and late learners of a second language 3.1 points less. Adding controls for non-language skills barely changed the picture: bilinguals still trailed by 0.8 points and late learners by 2.4. A Clemson University thesis pointed the same way, associating non-Spanish bilinguals with 9 percent lower earnings while finding the Spanish-bilingual effect statistically insignificant. A related thesis figure putting the overall bilingual shortfall at 7.7 percent is contested: the cited source confirms the number itself but does not, in the passage available, confirm the survey dataset it is attributed to, so the figure should be treated as unverified.
An independent research institute reinforces the null. The IZA World of Labor concludes that US workers who know Spanish plus English are not necessarily paid more than English-only workers. Three separate analyses of the same US labor market - a top-tier journal, an academic thesis, and a research institute - thus converge on the same unglamorous verdict: in the American labor market, a second language adds little or nothing to pay by itself.
Against this stands a single loud dissent. Preply, a commercial language-learning platform, reports that bilingual workers earn on average $9,353 more per year, a 14.3 percent premium across 18 countries, with the US figure at $14,050, or 18.8 percent. As a commercial language-learning platform, Preply has an incentive to report results showing its product pays off, and its figures do not appear to apply the skills-and-industry controls that flip the academic results. Its cross-country numbers hold together internally - the US premium sits about half again above the global average, as one would expect of an above-average market - but internal consistency is not the same as verification against neutral data.
Open: Whether a causal design using exogenous variation in bilingualism - such as childhood language-policy changes or random assignment to immersion - would separate the effect of the second language from the immigrant or minority status it often signals.; What controls and sample Preply applied, since the gap between its premium and the academic penalty cannot be reconciled from the information disclosed.
For immigrants learning the local language, the gains are real and large
The economic story changes completely when the learner is an immigrant mastering the host country's dominant language - and this is where the evidence is strongest and most consistent. An IZA discussion paper found that Dutch proficiency raises an immigrant's probability of being employed by about 30 percentage points. In Spain, a Springer migration journal found each additional level of Spanish fluency lifted the chance of employment by 3.2 percentage points, rising to 4.3 once parental education was controlled, and raised the odds of holding a highly skilled job by about 2.1 points per level. The IZA World of Labor puts the general earnings advantage for dominant-language fluency in the range of 10 to 20 percent.
Three independent origins from three countries agree in direction even as the exact size varies with context. That convergence is what makes this the report's most solid economic finding. The flip side is equally documented: migrants who keep using their native language predominantly tend to earn much lower wages than the mainstream economy, and immigrants who arrive after early adolescence carry literacy-test gaps of up to a full standard deviation behind native speakers, roughly double the gap for those who arrive young.
The crucial caveat is what these numbers do not measure. They capture the payoff of learning the local language as a second language - not the effect of being bilingual as such. The IZA finding that US Spanish-English bilinguals earn no premium and the immigrant integration gains can both be true at once, because they are answering different questions. Blurring them - treating an immigrant's employment jump into general proof that bilingualism pays - is the single most common distortion in how this literature is popularized.
Open: Whether the integration gains would survive a design comparing immigrants who speak only the host language against those who also retain a heritage language, which would isolate any effect of bilingualism beyond host-language proficiency.; Whether the labor-force participation gap of 57 percent versus 76 percent attributed to the Migration Policy Institute [C-027] holds, since it reaches this report only through a single commercial source citing the institute without a direct link.
The costs the pitch leaves out: dropout, teacher flight, and thousands of hours
Every benefit above assumes the learner actually reaches proficiency. Most do not, and the programs meant to get them there leak badly. The Center for Advanced Research on Language Acquisition at the University of Minnesota found that across six cohorts at a French-immersion school, nearly a quarter of pupils - 24.7 percent - dropped out before the end of fifth grade. Canadian French-immersion programs fared worse, with reported elementary attrition of 40 to 50 percent. The teachers leave too: Georgia's 2015 workforce report found foreign-language and mathematics teachers had the highest attrition of any high-school subject, around 44 percent over five years.
These figures measure different populations over different windows - pupils across six years, teachers across five - so they cannot simply be stacked. What they share is a direction: language programs shed both learners and instructors at high rates.
Then there is the raw arithmetic of time. The Foreign Service Institute estimates it takes about 600 hours to reach professional proficiency in languages close to English, and up to 2,200 for the hardest ones. Set against that, typical classroom provision falls far short: one practitioner estimate, presented as an optimistic reading, holds that most UK students get around 2.5 hours a week over five years - roughly 450 contact hours - and rarely climb past a basic A2 level, with only a small subset reaching partial intermediate proficiency by the end of their GCSE exams. That characterization is one informed observer's interpretation rather than a hard measurement, but the underlying hours match: 2.5 hours across roughly 36 weeks a year for five years lands near 450, below even the easiest FSI threshold.
It is not all attrition. Intensive exposure can move fast: a thesis literature review reports that after a five-week study-abroad program, learners improved at spotting grammatical errors with a large effect size of 0.77, and one cited study found speech rates jumping from 65 to 111 syllables a minute against no gain for stay-at-home learners. The lesson is about dose, not futility - the returns come with immersion and hours, precisely what most school programs cannot supply.
Open: Whether dose-response trials relating cumulative contact hours to standardized proficiency outcomes would confirm a threshold below which programs predictably fail.; Whether the far-off-track catch-up figures - fewer than 10 percent of struggling fourth graders and 6 percent of eighth graders reaching benchmark [C-031] - hold, since they reach this report through a single blog citing ACT data rather than the original report.
Weighing the competing explanations
The evidence supports several competing readings, and they do not all stand equally. The first is that bilingualism provides a genuine, measurable cognitive advantage. Its support is real but thin: the narrow N1 processing efficiency, the dementia delay, and a small meta-analytic effect from a weak source. Against it sits the bulk of the executive-function evidence - the 81.4 percent null rate across 177 comparisons, the monolingual speed advantage on inhibition tasks, and the absence of any late-stage behavioral gain. This explanation stands weak. What would settle it is pre-registered replication with standardized batteries and a commitment to publish null results.
A second reading holds that in the English-dominant US context, bilingualism carries a mild labor-market penalty, likely because it signals immigrant status or non-native English rather than because the second language itself hurts. This is the better-supported economic story: the PLOS ONE penalty after controls, the thesis finding on non-Spanish bilinguals, and the IZA conclusion of no premium all point the same way, with only a commercially interested source dissenting. What it lacks is a causal design - an audit study or a natural experiment - that could separate the language from the status it flags.
A third reading is the cleanest: for immigrants, proficiency in the host country's dominant language is the primary engine of economic integration, dwarfing any separate benefit of bilingualism. Nothing in the evidence set contradicts it. The Dutch employment jump, the Spanish fluency gains, and the general 10-to-20 percent premium converge across countries. The open frontier is whether any bilingual bonus survives once host-language proficiency is held constant.
A fourth reading treats the high failure rates as a problem of insufficient instructional time rather than a verdict against language learning itself: classroom programs deliver a few hundred hours against the 600-to-2,200 the FSI says proficiency requires, while intensive immersion produces rapid gains. This is plausible and internally coherent, though it rests partly on interpretive rather than measured claims, and no dose-response trial in the evidence set nails down the threshold.
What the evidence forces us to conclude
The evidence does not support the popular story of second-language learning as a broad, near-automatic upgrade to mind, income, and belonging. It forces a more segmented conclusion. On cognition, the weight of testing runs against a general advantage: most comparisons come back null, and where advantages appear they are narrow, task-specific, or small. The clearest exception is the delayed onset of dementia, and even that stands on a single source addressing a single late-life outcome.
On earnings, the evidence compels a split verdict. For US-born or established bilinguals, a second language does not reliably raise pay and may slightly lower it once skills and industry are controlled. For immigrants learning the host country's dominant language, the return is large and consistent. Any account that merges these two into a single bilingual premium misreads the data.
On costs, the record is unambiguous where it exists: programs lose learners and teachers at high rates, and proficiency demands an hour count few classrooms approach. It is worth stating plainly, as a matter of grade rather than fact, that some of the most eye-catching numbers - the commercial wage premiums, the participation gap, the school catch-up rates - reach this report only through single, interested, or secondhand sources and should be treated as claims awaiting independent confirmation, not established findings. As a labeled speculative extension: if the publication tilt documented in 2015 shaped which studies entered the canonical literature before the 2023 correction, then the true average cognitive effect may be even smaller than the already-modest published figures suggest - a possibility the Bayesian re-analysis leaning weakly toward no difference is consistent with but cannot confirm.
Why it matters
Governments, schools, and families pour money and years into second-language learning on the promise of sharper minds, fatter paychecks, and smoother integration. This evidence says the promise is real for some people and largely illusory for others - strong for immigrants learning a host language, weak or absent as a general cognitive or wage boost - while the costs in dropout and time are heavy and rarely disclosed. Knowing which benefit applies to whom is the difference between an investment that pays and one that mostly does not.
- Whether the 2015 publication tilt still shapes the corrected literature in ways the 2023 re-analysis could not detect.
- How the effects documented here translate across the report's stated geographies - the US, UK, Canada, and Australia - given that nearly every economic and integration finding is tied to one country's institutions and labor market.