College Graduate Supply vs. Job Readiness Pathways Before Graduation and Entry-Level Job Market After · 1970–2030 · Job Ventures, Inc. | April 2026 · v13
Enrollment, per-student annual cost, and total system investment · Hover any year for full investment and aid breakdown
Graduate supply, qualifying internship creation, degree-level job creation, underemployment rates, and persistence · 1970–2030 · Toggle series · Hover to compare · Click ⓘ for methodology
This visualization documents a structural labor market mismatch between degree-production volume and degree-level opportunity creation over a 55-year period. The following boundaries apply to all claims in this document:
"Structural" is used throughout this document in the labor economics sense: a persistent supply-demand mismatch that does not self-correct through normal market mechanisms (price adjustment, wage signals, or cyclical recovery). This is distinct from cyclical mismatches driven by business cycle conditions, which reverse as the economy expands. The 55-year graduate supply vs. opportunity creation gap documented here persisted through multiple full business cycles — including the 2000s expansion, the post-2010 recovery, and the 2019 labor market tightening — confirming its structural character. The rising underemployment floor (from ~37% in 1985 to ~40–42% at recent employment peaks) is the primary structural signal.
Source: NCES IPEDS Table 318.10, Bachelor's Degrees Conferred by Postsecondary Institutions. Tier A Confirmed for 1970–2023. Federal data; peer-reviewed compilation. 2024–2030 values are projected per NCES Projections of Education Statistics to 2030, which uses a +20% trajectory from the 2020 baseline. Projection years are directional. Graduate counts are in thousands (K) and represent all U.S. bachelor's degree conferrals in a given academic year.
1970 equilibrium baseline — externally sourced threshold: 1970 is the last year in which degree production and degree-level job openings were within Freeman's (1976) ±10% approximate parity band. The equilibrium criterion is sourced from Freeman, Richard B. (1976), The Overeducated American (Academic Press), which defines equilibrium as "approximate parity" between degree supply and degree-level demand and documents the divergence beginning in the early 1970s. The ±10% threshold is not self-defined by Job Ventures. Formal test: 1970 ratio = −0.3% (degree production ÷ job openings − 1); within the Freeman band. 1975 ratio = +28.2%; clearly outside the band. The "55 Years" framing counts forward from 1970 and is accurate under this externally sourced criterion. See JVI-TBD-001 v1.2 §4.6 for full year-by-year divergence table.
OLS projection diagnostics (JVI-TBD-001 v1.2 §5): The 2024–2030 projection for this series is based on OLS trend regression on the 1970–2022 confirmed IPEDS data (n=16). Full diagnostics: R²=0.951, Adjusted R²=0.947, F-statistic=269.05 (p<0.0001). Durbin-Watson = 0.34 — positive autocorrelation detected (DW <1.5 threshold). This means naive OLS confidence intervals are understated; HAC (heteroskedasticity and autocorrelation consistent) corrected standard errors have been applied for the projection band. Ljung-Box p=0.000 (residual autocorrelation significant). ADF test p=0.078 (residuals borderline non-stationary). The shaded band above the projected series represents the 90% HAC-corrected prediction interval: wider than a naive OLS band would produce. The trend direction is not in question; only the width of the projection band is affected by the autocorrelation correction. 2025 interval: [1,877K – 2,288K]. 2030 interval: [2,003K – 2,422K]. Tier B projected with documented uncertainty bounds. Note on selective CI band display: The prediction interval band is shown only for the Annual Graduates series — the series with the most data points (n=16) and the most severe autocorrelation (DW=0.34), where the HAC correction has the greatest impact on interval width. Other projected series carry equal or greater uncertainty: WBL supply (n=10, DW=1.44, sparser data), underemployment projections (DW=1.18, extrapolated beyond confirmed data), and net new jobs (flat line from single anchor). The selective display reflects data density, not relative certainty. All projected series beyond 2024 should be treated as directional.
Source: Georgetown CEW, Falling Behind: How Skills Shortages Threaten Future Jobs,
September 2025. Tier A Confirmed.
Derivation: The report states the U.S. economy will add approximately 685,000 new jobs
requiring postsecondary education and training from 2024 through 2032 (an 8-year window), yielding
~86K per year. Important scope note: the 685K figure covers all postsecondary levels
(certificates, associate's degrees, and bachelor's degrees). The BA-or-higher subset is approximately
73–75% of this total, consistent with the report's finding that 4.5 of 5.25 million additional
workers needed will require a bachelor's degree or higher. The chart label "Net new degree-level jobs"
is a reasonable approximation but should be cited as Georgetown CEW Falling Behind (Sept. 2025) with
the scope clarification above if cited directly.
Interpretive note — the paradox: Falling Behind reports a projected worker shortage (18.4M experienced postsecondary workers retiring through 2032, vs. 13.8M new entrants with equivalent credentials). This is not contradicted by high graduate underemployment. Both conditions coexist because: (a) new graduates tend to be distributed across majors inconsistently with where shortages exist (management, healthcare, teaching, engineering); (b) AI has specifically eliminated entry-level roles in tech, data, and consulting — the sectors that historically absorbed new graduates; (c) the retirement wave creates demand for experienced workers, not entry-level positions. The underemployment and shortage statistics are measuring different phenomena in the same labor market.
Construction: EREPL = TREPL × EFRAC, where TREPL is total BA+ replacement demand and EFRAC is the entry-accessible fraction of that demand.
TREPL — documented discontinuity: Pre-2020 values were modeled from older Georgetown CEW "Recovery" reports (2013, 2016) that used a "new jobs + replacement jobs" methodology. Post-2020 values (2020: 3,900K; 2023: 4,214K) are drawn from Georgetown CEW After Everything: Projections through 2031 (Nov. 2023), which introduced a new methodology based on annualized total job openings — combining both growth demand and all replacement demand. Georgetown CEW explicitly acknowledged this shift, noting: "In previous reports…the Georgetown University Center on Education and the Workforce broke down projections into two categories: new and replacement jobs. However, we became increasingly convinced that those projections misrepresented 'churn' in jobs…The Bureau of Labor Statistics changed its methods for forecasting jobs beginning in 2016." The result is a 4.5× step increase in the series between 2015 and 2020 that is entirely an artifact of methodology change, not a real labor-market event. A visual break is marked on the chart. Pre-2020 values should be treated as Tier C directional estimates only; the confirmed post-2020 anchor is 4,214K (After Everything, 2023). Tier C pre-2020; Tier A for 2023 anchor.
EFRAC — entry-accessible fraction: Derived from BGI/Strada No Country for Young Grads (July 2025), which documents the decline in entry-level hiring rates in software (43%→28%), data (35%→22%), and consulting (41%→26%) between 2018 and 2024. EFRAC pre-2018 is a modeled backward extrapolation from confirmed 2018–2024 trend data. Tier C pre-2018; Tier B 2018–2024. Do not cite pre-2018 EFRAC values as primary-source data.
Compounding uncertainty — pre-2018 EREPL values: For years before 2018, EREPL is the product of two independent Tier C estimates: TREPL (pre-2020, Tier C) × EFRAC (pre-2018, Tier C). When two uncertain estimates are multiplied, the compounded uncertainty is greater than either input individually. Pre-2018 EREPL values should be treated as illustrative directional estimates only — they cannot be cited as derived data. They are retained in the visualization to show the historical shape of entry-level accessibility, not to assert specific headcount values. Pre-2018 EREPL: Tier C compounded — directional illustration only.
Source: Federal Reserve Bank of New York, Abel & Deitz Are Recent College Graduates Finding Good Jobs? quarterly tracker. Tier A Confirmed for the confirmed period. Q4 2025 = 42.5% confirmed. Pre-1990 values are estimated from Freeman (1976) and BLS historical data — Tier C directional. 2026–2030 values are extrapolated consistent with BGI's 7–11M graduate surplus projection — Tier B derived.
Structural floor vs. cyclical oscillation: This series contains both a structural and a cyclical component. The rate oscillates with the business cycle — rising in recessions (44% in 2010), falling in expansions (38–39% in the late 1990s). The structural signal is the floor: the minimum rate observed at peak employment conditions. That floor has risen from approximately 37–38% (1985–1990 peak) to approximately 40–42% (2015–2019 peak). The rising floor — not the cyclical oscillation — is what constitutes the structural deficit. Forward projections (2026–2030) reflect the structural trajectory; they should not be interpreted as forecasts of cyclical conditions.
Classification method — NY Fed vs. BGI (not conflicting data): The NY Fed series (41–43%) and the BGI scatter points (43%, 52%) measure the same phenomenon through different classification systems and should not be treated as corroborating or conflicting. NY Fed uses BLS Standard Occupational Classification (SOC) — a normative system that asks whether an occupation typically requires a bachelor's degree. BGI uses posting-level classification — asking whether each specific job posting explicitly requires or prefers a degree. BGI's result is higher because postings in "degree-preferred" occupations where most employers do not post degree requirements are classified as non-degree. The 9–11 percentage-point gap between the two series is fully explained by this methodological difference, not by data error. NY Fed SOC is the longer historical series (available from ~1990); BGI posting-level is more current and operationally precise. Both are presented because they answer different questions. See JVI-TBD-001 v1.2 §4.4 for full reconciliation table.
Sources: BGI/Cleveland Fed anchors: 5.2% (2018–2019), 6.2% (2023–2025); NY Fed: 5.7% (Q4 2025). Tier A for named anchors. The 2020 data point is nulled — no named primary source was identified for a COVID-year recent-graduate unemployment rate that meets citation standards. Pre-1990 values are omitted. 2026–2030 values are extrapolated — Tier B.
What is plotted: For each chart year Y, this series shows the predicted headcount of graduates who will be underemployed at their 5-year or 10-year mark — not an observed current stock of underemployed workers. It is a per-cohort forward prediction, not a measurement of current employment status of past graduates.
Construction: Graduates from cohort year (Y−5 or Y−10, interpolated) × year-1 underemployment rate for that cohort × BGI persistence factor.
Tier B derived throughout.
The persistence rates themselves are confirmed Tier A; the calculated headcount series is Tier B because it combines confirmed rates with estimated or derived cohort-level underemployment inputs.
Selection bias caveat: The 75% and 73% persistence rates were measured on graduates who were already underemployed at year 1 — not on all graduates. In labor economics terminology this is selection on unobservables: graduates underemployed at year 1 are likely systematically different from those who are not (by major, institution selectivity, geography, socioeconomic background, and quality of internship experience prior to graduation). The persistence rates therefore reflect the experience of the underemployed population specifically and should not be applied to the full graduating class as a universal probability. The D5 and D10 headcount series correctly apply the rates only to the estimated underemployed cohort — not to all graduates — which is the methodologically appropriate use.
Anchor: 2023 = 3,574K — BHEF Expanding Internship Programs to Improve Student Outcomes (Sept. 2024) + NSCI 2023 + IPEDS enrollment. Tier A Confirmed for 2023. 2024 derived at –3.1% per NACE Internship & Co-op Report (2025). Tier B for 2024. 1995–2022 values are directional trend estimates from Shandra (2022), Socius Vol. 8 — Tier C directional; not citable as confirmed annual headcount series pre-2023. Note: Shandra (2022) is a study of disability and labor market outcomes; the specific finding used here is the incidental observation that internship participation "more than doubled since the mid-1990s." No other data from the paper is used.
Anchor: 2023 = 2,502K — BHEF Expanding Internships (Sept. 2024): 70% of employers with internship programs offer medium-quality or better (employer-side rating). Tier A Confirmed for 2023. 2024–2030 values derived by applying the 70% quality rate to projected total internship supply — Tier B.
Construction: WBL Access Rate = (INTERN_TOTAL × 0.30 ÷ 17,000K enrolled students) × 100. The 30% quality rate is the learner-side standard from Strada Scaling Work-Based Learning (2025). Tier B Derived.
Denominator note — 17M vs. 18M: This series uses 17,000K (NCES IPEDS total undergraduate enrollment, confirmed) as the denominator. The figure "18M candidates" used elsewhere in this document refers to BHEF's broader population definition — enrolled students plus recent graduates still seeking WBL opportunities — and is sourced from BHEF Expanding Internships (Sept. 2024). These are two distinct populations used for distinct calculations: 17M for the per-enrolled-student access rate; 18M for the total WBL candidate demand estimate. They are not interchangeable and should not be used to cross-compute.
"Quality" vs. "Qualifying" — an important distinction: The chart title uses the term "Qualifying Internship" — meaning an internship that satisfies ATS domain-matching and experience requirements for a specific target role. This is a stricter standard than Strada's "quality" definition. A quality internship (clarity, supervision, skill development) may still be non-qualifying if it is in the wrong domain, company type, or skill configuration for the role the student is targeting. No confirmed dataset currently measures qualifying internship supply in the ATS-matching sense. This series uses quality as the best available proxy. The gap between quality and qualifying is itself an unquantified finding and an active research gap in workforce development literature.
Qualifying position ceiling (JVI-TBD-001 v1.2 §4.5): No primary source directly measures the count of WBL positions meeting the ATS-qualifying threshold. The hard upper bound is derived from confirmed outcome data: approximately 966,000 graduates achieved degree-level employment in 2023 (IPEDS 2,013K graduates × (1 − BGI 52% underemployment rate) = 966K; Tier B Derived). Qualifying positions cannot exceed this ceiling — it is mathematically impossible for more positions to qualify students than students who achieved the outcome. The actual count of qualifying positions is some fraction of 966K, constrained further by WBL participation rates (BHEF: 44% of graduates) and NACE's 62% FT offer rate for 2024 paid intern completers. The ceiling alone — 966K vs. 18M candidates — demonstrates structural insufficiency without requiring a point estimate. A direct measurement study quantifying the ATS-qualifying subset does not exist in current public literature. This is a named research gap, not a data error.
Two bundled deficits: The 7% figure is the product of two separate gaps: (1) Access gap — only ~21% of enrolled students can access any internship; (2) Quality gap — only 30% of those who get one rate it as quality. Gap 1 (supply creation) and Gap 2 (quality design) are analytically distinct problems that require different structural interventions.
The 2023 computed anchor = 6.3% (3,574K × 0.30 ÷ 17,000K × 100). The 7.0% reference line (MMI v2.0) uses ~4M internship participants ÷ 17M enrolled × 30%.
Intern-to-full-time offer rate (corrected): 62% for the 2024 cohort, down from 67% for the 2023 cohort — lowest in five years. Source: NACE Internship & Co-op Report (2025).
The 43% (Permanent Detour, 2018) and 52% (Talent Disrupted, 2024) scatter points come from two separate BGI/Strada studies with potentially different sampling frames, resume databases, and occupational classification methodology vintages. Plotting them on the same axis implies methodological continuity that has not been independently confirmed. The directional finding (underemployment increased from the ~2015 cohort to the Class of 2023) is consistent with NY Fed aggregate data over the same period and is therefore plausible. However, the specific 9-percentage-point magnitude should not be cited as a precisely measured change — only as the directional finding from two separate confirmed studies conducted by the same institution using broadly comparable methods.
All series in this visualization represent aggregate postsecondary totals across all fields of study. The structural deficit is substantially more severe in some fields than others. NY Fed Abel & Deitz data shows underemployment rates by major ranging from approximately 20% (nursing, computer science, engineering) to 55–65% (arts, humanities, general social sciences). Georgetown CEW's Falling Behind (Sept. 2025) documents that the projected worker shortage is concentrated in management, healthcare, teaching, and engineering — fields where the qualifying internship and certification deficit is simultaneously most acute. Aggregate figures are appropriate for market-sizing purposes and for documenting the structural condition at the system level, but they understate mismatch severity in the most affected disciplines. Field-level disaggregation is available from NY Fed and NCES sources and is recommended for any field-specific investment thesis.
The following changes align this visualization with JVI-TBD-001 v1.2 Technical Basis Document (Job Ventures, Inc., April 2026). All changes are additive — no data series, confirmed values, or v12 QA fixes were modified.
Seventeen issues identified in comprehensive technical review. All corrections are documented below. No confirmed data values were changed; all changes are methodological clarifications, label corrections, scope additions, or framing improvements.
| # | Citation | Series Used In | Tier | URL |
|---|---|---|---|---|
| 1 | NCES IPEDS Table 318.10 — Bachelor's Degrees Conferred by Postsecondary Institutions, 1970–2023. National Center for Education Statistics. | Annual Graduates | A | nces.ed.gov |
| 2 | NCES Projections of Education Statistics to 2030. National Center for Education Statistics, 2022. | Graduate projections 2024–2030 | B | nces.ed.gov/programs/projections |
| 3 | Georgetown CEW, Falling Behind: How Skills Shortages Threaten Future Jobs, Nicole Smith et al., September 16, 2025. | Net new jobs (~86K/yr); 4.5M BA+ worker shortage | A | cew.georgetown.edu/cew-reports/skills-shortages |
| 4 | Georgetown CEW, After Everything: Projections of Jobs, Education, and Training Requirements through 2031, Carnevale et al., November 2023. | TREPL 2020–2023 anchor (4,214K); methodology discontinuity source | A | cew.georgetown.edu/cew-reports/projections2031 |
| 5 | Burning Glass Institute / Strada, No Country for Young Grads, July 2025. Entry-level hiring rates in tech, data, consulting, 2018–2024. | EFRAC (2018–2024) | A | burningglassinstitute.org |
| 6 | Federal Reserve Bank of New York — Abel & Deitz, Are Recent College Graduates Finding Good Jobs? Quarterly tracker, Q4 2025 = 42.5%. | Underemployment % (NY Fed) | A | newyorkfed.org/research/college-labor-market |
| 7 | BGI/Strada, Permanent Detour: Underemployment's Long-Term Effects on the Careers of College Graduates, May 2018. First-job underemployment = 43% for ~2015 entry cohort; 5-year persistence rate = 75%. | BGI scatter 43%; D5 persistence | A | burningglassinstitute.org/research/permanent-detour |
| 8 | BGI/Strada, Talent Disrupted: Underemployment of College Graduates Worsened and Shows No Sign of Improving, February 2024. Class of 2023 = 52% yr-1 underemployed; 10-year persistence = 73%. | BGI scatter 52%; D10 persistence | A | burningglassinstitute.org/research/talent-disrupted |
| 9 | BGI, Graduate Surplus Projection 7–11M, December 2025. Directional trajectory consistent with ~55% yr-1 underemployment by 2028. | BGI scatter ~55% (projected); UNEMP 2026–2030 extrapolation | B | burningglassinstitute.org |
| 10 | BHEF, Expanding Internship Programs to Improve Student Outcomes: Employer Perspectives on Internship Quality, September 2024. 2023 anchor: 3,574K total internships; 2,502K at medium-quality or better (70% employer standard). | Internships available; Quality internships BHEF | A | bhef.com/publications |
| 11 | NACE, Internship & Co-op Report, 2025. Intern-to-full-time offer rate 62% (2024 cohort); –3.1% internship supply decline 2023→2024. | Intern-to-hire rate; INTERN_TOTAL 2024 | A | naceweb.org |
| 12 | Shandra, Carrie L. (2022). "Disability, Internship Participation, and Labor Market Outcomes among Young Adults." Socius: Sociological Research for a Dynamic World, Vol. 8. Specific finding used: The incidental observation that U.S. internship participation rates "more than doubled since the mid-1990s" — the only published academic source for this directional trend. The paper's primary subject (disability and labor market outcomes) is not used here. Used for directional internship trend 1995–2022 only. Tier C — directional, not confirmed annual series. | Internships available (pre-2023, directional) | C | journals.sagepub.com |
| 13 | Strada Education, Scaling Work-Based Learning, 2025. "Just 30% of learners had WBL experiences marked by clarity, supervision, and skill development." | WBL Access Rate % (Strada 30% learner standard) | A | stradaeducation.org |
| 14 | Georgetown CEW, Recovery: Job Growth and Education Requirements Through 2020, 2013. Source for pre-2020 directional replacement demand estimates (old methodology — see TREPL discontinuity note). | TREPL pre-2020 directional (Tier C) | C | cew.georgetown.edu |
| 15 | Freeman, Richard B. (1976). The Overeducated American. Academic Press. Two distinct uses: (1) Pre-1990 underemployment directional estimates (Tier C — shape only); (2) Source of the ±10% equilibrium parity criterion used to validate 1970 as the structural break point (Series 1 methodology, §4.6 of JVI-TBD-001 v1.2). The equilibrium use is methodological, not a data citation. | Underemployment % pre-1990 (Tier C); 1970 equilibrium threshold (methodological) | C | Academic Press, 1976 (no public URL) |
The three-layer deficit: Row 1 documents compounding barriers — not a single aggregate shortfall. The economy has never created enough net-new BA-level positions for graduating volume (29:1 graduates per net-new BA+ job, using 2023 confirmed data). The replacement pathway that historically absorbed new graduates is being systematically closed via experience inflation (43%→26%). The experience requirement itself is mathematically unsatisfiable through the standard educational pathway — 3 summers of internships produce 7.5 months of experience; employers require 12–36 months. The result: 52% first-job underemployment and a 73–75% persistence rate at 5 and 10 years for graduates who are already underemployed at year 1 — meaning those who miss the degree-level window within 12 months of graduation face a high probability of sustained underemployment. (Note: the 73–75% rates apply to the already-underemployed population specifically — selection on unobservables applies. See Series 6 & 7 methodology.)
Why existing institutional responses have not closed the deficit: Every current intervention — career services, stipend programs, Launchpad initiatives, internship fairs, Handshake, credit incentives — is a demand-side response. When 7,600+ institutions simultaneously fund students to compete for qualifying internships, they intensify competition for a fixed supply that does not expand because demand-side investment does not expand the supply of qualifying positions. The structural deficit persists because no mechanism currently operates at a scale sufficient to close the gap between 18M candidates and ~966K qualifying positions. Existing supply-creation programs — DOL registered apprenticeships, NSF Research Experiences for Undergraduates, and select employer co-op partnerships — serve important roles but collectively reach fewer than 800K students annually, with significant overlap with the populations already counted in the internship supply figures.
Students need not one qualifying experience but a designed sequence — foundational, intermediate, advanced, and capstone — compounding across 4–5 years of undergraduate enrollment and 1–2 years of graduate study, each stage in the right domain, credential stack, and employer context. The effective demand is not 17M students competing for 3,574K positions once. It is 17M students needing 4–5 sequential qualifying experiences — a system-level impossibility no stipend program addresses because stipends fund competition, not creation.
Note on the Georgetown CEW apparent contradiction: The same report that confirms 86K net-new jobs/yr also projects a 5.25M worker shortage. Both are true — field mismatch, experience requirements, and ATS screening ensure that graduate oversupply and employer shortages coexist in the same labor market. Both conditions are symptoms of the same broken pipeline — and both require supply-side intervention at the qualifying experience layer, not demand-side competition for fixed supply.