How we project a 10-year college payoff
Honest doesn't mean precise. This tool gives you ranges, not points, because nobody — including us — can project a decade of cost, loans, and earnings down to a single dollar. Below we lay out exactly what we measure, what we don't, and how we check whether we're being honest. Jump to calibration evidence.
Inputs the model uses
Your college list (1–3)
We project a 10-year trajectory for each. The list order doesn't change the math; we render the trajectories side-by-side in input order.
Intended major
We fuzzy-match to each college's bachelor's program data and pull the program's published median Y1 earnings as the earnings anchor.
Household income range
Used by the SAI calculator to estimate Pell + institutional grant aid in year 1 and to size the family's expected cash contribution.
State of residence
Triggers in-state tuition pricing at public colleges in your state. You can flip this per-college on the result page.
GPA range
Reserved for a future merit-aid model. v1 doesn't condition on it — disclosed on the form.
Dependents (optional)
Sizes the income-protection allowance in the SAI calculator.
How the model works
The simulator walks 10 years from start of college: four in-school years (cost accumulates, loans originate) and six post-grad years (earnings flow, loans amortize). Every output renders as a range that combines two components: how confident we can be given the data we have (statistical uncertainty), and how much the things we can't model might move the answer (out-of-model uncertainty).
More technical
Cost & aid. Year-1 cost = tuition (in- or out-of-state per your residency) + room & board. Year-1 grant aid comes from the SAI calculator (Pell + institutional + work-study) seeded with the bracket midpoint of your household income. Both inflate 3.5%/yr nominal — a single national prior; per-college multi-year history is a v2 follow-up.
Loan layering. Stafford fills the gap after family contribution first (subsidized then unsub up to AY 2024-25 caps: freshman $5.5k / sophomore $6.5k / junior+ $7.5k). PLUS covers any remaining gap. Sub doesn't accrue during enrollment; unsub + PLUS accrue simple interest and capitalize at graduation.
Repayment. 10-year standard plan with a weighted-average rate across sub / unsub / PLUS. Closed-form monthly amortization, constant payment across the 6 post-grad years in the simulation window.
Earnings. Year-1 anchor = the matched program's ``median_earnings`` (Scorecard 4-years- out path; ≈ first year post-grad for a 4-year degree). When no program matches, we fall back to the college-wide 10yr earnings figure scaled to a Y1 estimate (0.7×) — disclosed with a per-prediction note. Wage growth: 3.0%/yr nominal national prior. BLS Employment Projections per-major differentiation is filed as a v2 follow-up issue.
Earnings band. Statistical component from cohort sample size — SE(median) ≈ 1.253 × σ/√n with σ = 30% × earnings (CV prior on individual earnings dispersion), scaled to a 95% CI via z=1.96. Out-of-model: ±20% at post-grad Y1 widening linearly to ±35% at post-grad Y6.
Cost band. Statistical component near zero (Scorecard publishes a point estimate, not a distribution). Out-of-model: ±15% per year for year-over-year price changes and family circumstances.
Loan balance + annual payment. Deterministic given the cost band (no additional envelope — would double-count cost uncertainty already in net price).
Break-even year. First post-grad year where cumulative post-grad earnings overtake the gross college bill (sum of net price during the four in-school years). ``None`` when break-even doesn't occur within the 10-year window.
Earnings clamps. ``lo`` band ≥ federal poverty line ($15,060/yr for a 1-person household, 2024). ``loan_balance.lo`` ≥ 0. Negative-debt and Y10-earner-at-$0 aren't true claims about any degree-holder.
What we don't model
- your career choices after graduation
- wage differentiation by major (we use a single national prior)
- individual employer outcomes (Scorecard reports federally-aided cohort medians only)
- life events (recession, illness, relocation, family circumstances)
- occupational disruption from automation or industry shifts
- graduate school (we model bachelor's earnings only)
- geographic cost-of-living variance
Repetition is intentional — these are the same factors the result page calls out. They're the part of your payoff trajectory you actually control.
Synthetic profile on per-school pages
The per-row trajectory bands on /colleges/[id]/career-outcomes are computed with a synthetic profile so the page is cacheable and not visitor-specific. The profile uses:
- Household income range: $60k–$100k (covers the U.S. median ~$75k).
- Family size: 4 (dependents = 2 + the student, per the SAI calculator's
dependents + 2convention). - State of residence: the school's own state, which surfaces in-state public-tuition treatment when the school is a public flagship in that state, and is neutral for everywhere else.
- GPA range: 3.0–3.5 (central bracket; v1 doesn't consume GPA, but the input schema requires it).
Your personal projection — run via “See your own 10-year projection” from the per-school page — uses your actual income, family size, residency, and intended major.
Calibration evidence
Last successful calibration: October 19, 2026 at 6:30 AM
Mean absolute error (USD): $0
Smaller is better. This number measures how close the model's Y1 earnings midpoint lands to each program's published median for a synthetic median student. Because the model anchors on Scorecard's published median by construction, the healthy v1 score is ~$0; a non-zero MAE signals a math regression (unit error, off-by-one in inflation, etc.) — not a real test of accuracy against student outcomes.
Predicted vs. published median earnings, by earnings bin. Points near the dashed y=x line mean the model recovers the program's published median; points off the line indicate a back-fit miss.
26554 (college, bachelor's program) pairs scored this run.
v1 calibration is a back-fit consistency check: does the model recover each program's published median year-1 earnings from College Scorecard when scored on a synthetic median student? It is NOT a holdout test against real outcomes — we don't have that data yet. The MAE here (USD) measures fit to a published prior, not real-world accuracy. Same posture as the admissions-chance v1 calibration. Score is mean absolute error in USD between the model's year-1 earnings midpoint and the program's published median.
Version history
| Model version | Deployed | Notes |
|---|---|---|
| life_path_v1.1 | 2026-05-16 | Restored the bachelor's-only filter on the calibration sample after #387 populated Program.degree_level from Scorecard credential.level. Model math is unchanged from v1; only the eligibility gate shape changed. The 2026-05-16 hotfix sentence ("v1 also scores all program types — Scorecard's degree-level data is not yet populated...") is removed from the persisted disclaimer; calibration MAE is once again scored only against bachelor's programs, matching the v1 4-year-window assumption. |
| life_path_v1 | 2026-05-15 | Initial v1 model — 10-year window (4 in-school + 6 post-grad); SAI-driven year-1 aid + Stafford / PLUS layering at AY 2024-25 rates with in-school capitalization of unsubsidized + PLUS interest; 10-year standard repayment with weighted-average rate; 3.0% nominal wage growth prior; 3.5% nominal COA inflation prior; binomial-flavored earnings statistical band (CV=0.30, cohort-size SE); ±20% Y1 → ±35% Y6 out-of-model envelope on earnings; ±15%/yr out-of-model envelope on cost. |
Privacy & data
This tool is anonymous-first. We don't require signup, we don't collect your name, and we don't ask for an email unless you opt into lead capture. Your inputs and the resulting trajectory are stored in a 30-day Redis session keyed by an opaque ID; nothing leaves that scope. We do not condition the model on race, ethnicity, or any protected class. See our privacy policy for the full picture.