MD Finance Calculator

Physician Specialty Career Earnings and Financial Independence

You are choosing a specialty. This shows you the money side of that choice for 64 fields of medicine - what you earn, when you start earning it, what is left after tax and loans, and the age at which work could become optional. No finance background needed. Every field has a ? beside it if you want more.

● In process Version 2.1.0 · Built 16 Aug 2026 · 64 specialties · Compensation anchor: Doximity 2025 + Medscape 2026, trended to mid-2026 dollars

Checking…

Six things worth knowing before you start

These are the findings that surprise people most. Each one is explained in full under the tabs above.

Saving beats choosing

20 yrs

The difference between saving 10% and 35% of your pay is about twenty years of working life. That is a bigger gap than any two specialties.

Longer training can pay less

6 fields

Six children’s specialties pay less than general paediatrics, after three extra years of training. Paediatric endocrinology is about $45,000 a year lower.

Owning beats earning

1.35×

Becoming part-owner of a practice is worth more than twenty-five years of experience. Almost nobody factors this in when choosing.

Pay per hour tells a different story

2.8×

Yearly pay varies 3.5 times over between fields. Per hour worked it is 2.8 times, and the order changes a lot. Dermatology beats anaesthesia per hour but not per year.

Loans got harder in 2026

$215k

Typical debt now exceeds the new $200,000 government borrowing cap. Students starting after July 2026 will need private loans, which are worse in every way.

A high salary is not enough

1 in 4

One in four doctors in their sixties is not a millionaire, despite a career of top-percentile income. Spending, not earning, is what decides.

The short version. Which specialty you pick sets a pay range, and that range is wide. But three other things usually matter more over a whole career. How long you train turns a bigger salary into a later start, and every extra year is a year at about $75,000 instead. What kind of place you work in moves pay more than experience and location put together. And how much of your pay you save decides more than all of it: at 10% saved, no specialty gets you free before your late fifties; at 35%, almost all of them do before fifty.

The point of this app is to let you see all three at once, because looking at pay alone leads people to the wrong answer.

When could you stop working?

Your salary is only the starting point. What actually decides your life is the gap between what you earn and what you spend, invested over time. Once your savings are big enough to cover your spending, work becomes optional. This works out when that happens for you — including the things that really eat a doctor’s income: loans, children, school fees, and university.

Which specialty

How much of your pay you save 25%

How big a nest egg you want 25×

Number of children 2

Who pays for college

Pay for private school before university

Also pay for graduate school

Use tax-friendly retirement accounts?

Latest age you would want to still be working 65

Median path 10th–90th percentile of 1,500 return paths FI target Coast-FI crossover

COULD STOP AT

—

IF MARKETS GO BADLY

—

COULD STOP SAVING AT

—

NEST EGG NEEDED

—

Where the money actually goes

One mid-career year, broken down. The last line — what is left to live on — is the number that decides your standard of living, and it is far smaller than the salary figure people compare.

The most important chart here

How much you save against the age you could stop working. Your specialty shifts the whole curve up or down. How much you save changes its shape — and the horizontal distance between saving 10% and saving 30% is bigger than the distance between family medicine and orthopaedic surgery.

How big should the nest egg be?

Multiple	Implied withdrawal rate	Evidence over a long horizon	Grade
20×	5.0%	Only defensible with genuinely flexible spending. Trinity-style backtests to 2025 put 6% below 75% success at 30 years; 5% sits in between and fails materially over 40 to 50 years. A physician retiring at 50 should not use this.	B
25×	4.0%	The familiar Bengen/Trinity figure, and it is a 30-year number. At 50 years it holds at roughly 90% success at best, and under 90% at 40 years unless the portfolio is near 100% equity. Adequate for retirement at 60, marginal for retirement at 50.	A
29×	3.5%	Kitces: extending the horizon from 30 to 45 years cuts the safe rate from 4.1% to 3.5%, and it does not fall materially further beyond about 45 years. This is the defensible target for early retirement, and it recommends 60–65% equity rather than the 50–60% appropriate to a 30-year horizon.	B
33×	3.0%	The robust perpetual multiple. Trinity update through 2025 puts 3.5% above 98% success at 50 years with a high equity allocation; 3.0% is more conservative still. The cost of this safety is roughly four extra working years for a typical physician profile.	B

Two opposing revisions, both recent, and the model uses neither uncritically. Morningstar's 2026 edition raised its starting safe withdrawal rate to 3.9% from 3.7%, on a 30-year horizon at 90% success with a 30–50% equity allocation. Bengen, separately, revised his own SAFEMAX from 4.15% up to 4.7% in August 2025 after adding small-cap and international exposure, and has suggested 5.25–5.5% is supportable in current conditions. These bracket a wide range, and all of them assume 30 years. The horizon correction matters more than the dispute: a physician who reaches financial independence at 50 needs a 45-year number, and the 45-year number is 3.5% regardless of which 30-year figure you start from. The model therefore defaults to 25× because that is what most readers will expect, and flags in the output when your projected FI age implies a horizon the 4% rule was never tested against.

The detail behind the numbers

Where every figure came from, how much to trust it, and what nobody publishes. You do not need any of this to use the app — but if you are going to make a decision on these numbers, it is worth ten minutes.

Model specification

Fully transparent and reproducible. Attending compensation is modeled as log-normal: a specialty reference median scaled by multiplicative modifiers, with a residual dispersion term. The financial layers on top of it are deterministic accounting plus one stochastic return process.

Compensation: ln(C) ~ Normal(ln(M₀ₓ · Π mᵢ), σₛ) where M₀ₓ is specialty s's reference median and σₛ depends on practice structure, not specialty.

Reference profile (all modifiers = 1.00): full-time attending, 5–9 years post-training, national practice-setting mix, suburban area, state at the national average price level, single household earner.

Derivation of M₀ₓ. Two sources cover physicians at specialty granularity, and they disagree systematically. Doximity's 2025 report (2024 data, n>37,000 self-reported surveys, regression-adjusted for specialty, metro, gender, tenure, and hours) publishes 52 specialties including real subspecialty detail. Medscape's 2026 report (2025 data, n=5,916 self-reported) publishes 29. Across the 28 specialties in both, after trending each to mid-2026 at 4.0%/yr, Doximity runs 9.8% above Medscape at the median, ranging from 0.995× (pathology) to 1.202× (rheumatology). That is a source fixed effect, not noise. The model takes the geometric mean of the two trended figures where both exist, and applies a ±√1.098 level correction to put a single-source figure on the same scale. The resulting average is then converted to a median by dividing by exp(σ²/2) with σ=0.30, because both sources publish means and a log-normal mean sits above its median.

Why BLS is not the anchor, despite being the only grade-A source. BLS OEWS May 2025 publishes mean annual wages for 17 physician codes, and they are far lower than the survey figures — anesthesiology $360,570, radiology $381,530, family medicine $255,820. Four structural reasons, all pushing the same direction, disqualify OEWS as a level anchor while making it a useful floor check. (1) The self-employed are excluded by design, confirmed verbatim: "The survey does not include the self-employed, owners and partners in unincorporated firms." That removes the highest earners in exactly the office-based and procedural specialties where partnership income concentrates. (2) Residents and fellows are pooled into the same codes as attendings, which is why the published 10th percentile for surgeons is $78,000 — not a real attending figure. (3) Annual wages are a 2,080-hour arithmetic construct, not observed earnings, so call pay, productivity bonuses, and partnership distributions are missing. (4) The May 2025 estimate is a three-year rolling average of panels back to November 2022, so it is centered near late 2023. Separately, 45% of all physician employment in OEWS sits in the single residual code "Physicians, All Other" (342,720 of 756,160), so specialty-level structure is largely unavailable there regardless.

Modifier register

Parameter	Multiplier	Grade	Evidence and derivation

Dispersion terms

Practice structure	Residual σ (log)	P90 ÷ P10	Basis
Academic / university faculty	0.24	1.85×	Narrowest. Salary is set by rank and department scale within an institutional band; AAMC reports rank medians, not spreads. Derived. D
Hospital or health-system employed	0.26	1.94×	Reference. Contract base plus a wRVU incentive; SullivanCotter reports 75% of organizations use individual productivity measures, which widens the band above a pure salary. Derived. D
Private practice, employed (non-partner)	0.28	2.03×	Derived. D
Private practice, partner or owner	0.38	2.55×	Practice profit, ancillary capture, and payer mix are all uncontrolled. Widest legitimate spread. Derived. D
Private-equity-owned practice	0.30	2.13×	Set between employed and partner. The literature does not measure physician pay in PE-acquired practices at all. Derived. D
Locum tenens / 1099, full-time equivalent	0.40	2.66×	Hourly rate is observed but annual income depends entirely on days worked, which is discretionary and unobserved. Derived. D

This table is graded D throughout, and that is the honest state of the evidence. No source publishes a usable physician earnings percentile distribution. BLS suppresses P75 and P90 for 100% of physician specialty codes at every vintage. Doximity, Medscape, AMGA, and SullivanCotter publish averages or medians without percentiles. MGMA publishes percentiles but only behind a licensed subscription. The σ values above are set from the qualitative structure of each arrangement — how much of pay is contractual versus discretionary — and are not measurements. Every percentile this model displays inherits that. Treat the median as roughly right and the width as illustrative.

The financial engine

Layer	Treatment	Key parameters, all verified for 2026
Timeline	Age 22 at college graduation, four years of medical school with no income, MD at 26, then T years of residency and fellowship at the AAMC stipend for that PGY level, then attending.	AAMC 2025 stipend survey, 350 institutions, 114,361 residents: PGY-1 $68,166 rising to PGY-8 $94,215. Nominal growth 2.2%, real growth −0.48%.
Debt	Balance accrues at the grad/professional unsubsidized rate from disbursement. Repayment follows the selected path. Under RAP, unpaid accrued interest is waived, so the balance never negatively amortizes, and a $50/month principal match applies when the payment reduces principal by less than $50.	Grad/professional unsubsidized rate 8.07% for loans first disbursed 1 Jul 2026–30 Jun 2027. Origination fee 1.057%. Default debt $215,000 = Class of 2025 median for indebted graduates.
Federal tax	2026 brackets and standard deduction, applied to gross less pre-tax retirement contributions. FICA on wage income, with the Social Security cap and the additional Medicare tax.	Rev. Proc. 2025-32. MFJ standard deduction $32,200; brackets 10/12/22/24/32/35/37% with 37% starting at $768,700 MFJ. Social Security wage base $184,500. Additional Medicare 0.9% above $250,000 MFJ.
State tax	An effective-rate approximation, not a full bracket schedule: the top marginal rate scaled by where its threshold sits relative to physician income. Eight states are zero.	Tax Foundation, rates in effect 1 Jan 2026. CA 13.30% but above $1M; OR 9.90% above $125,000 and MN 9.85% above $203,150 — those two bind on ordinary physician income, the coastal headline rates largely do not.
Tax-advantaged capacity	Contributions fill sheltered accounts before taxable. Capacity depends on setting, which is a real and underappreciated academic advantage.	402(g) deferral $24,500; 415(c) total annual additions $72,000; governmental 457(b) a separate $24,500 that stacks on a 403(b); HSA family $8,750; backdoor Roth $7,500 each. Notice 2025-67 and Rev. Proc. 2025-19.
Returns	Real returns, log-normal, drawn independently by year. 1,500 Monte Carlo paths for the band; the median path is deterministic.	US equities 6.6% real annualized 1900–2025 (UBS Global Investment Returns Yearbook 2026, 126 years); bonds 1.6% real, bills 0.5%. Equity SD 18.5%. Expense drag 0.05%, plus 0.40% tax drag on the taxable portion.
Family costs	Per child: general cost of raising to 18, optional private K–12 for 13 years, optional college for 4 years, optional graduate school for 4 years. Removed from spending at retirement, which lowers the FI target.	Raising a child to 18 ≈$299,000 (2025 estimate; the USDA discontinued its official series after 2015). Private K–12 average $14,923/yr. College total published price: in-state public $25,850, out-of-state public $45,780, private nonprofit $60,920 (College Board, AY 2025–26).
FI condition	Portfolio ≥ multiple × retirement spending, where retirement spending is current spending less child-related costs and less debt service. Coast-FI is the earlier age at which the existing portfolio alone compounds to the target by the selected retirement age with no further contributions.	Default 25×. The output flags when the implied retirement horizon exceeds 35 years, since 25× is a 30-year figure.

Calibration check

The model's reference medians must reproduce the published sources they were built from, and must sit in a defensible relationship to the one grade-A source that was deliberately not used as an anchor.

Specialty	Doximity 2025 (2024 → mid-2026, → median)	Medscape 2026 (2025 → mid-2026, → median)	Model median	BLS OEWS mean May 2025	Verdict

Result: PASS on the two survey sources by construction, and PASS on the BLS relationship as a test. The model median sits between the two survey averages at every specialty, which it must, since it is their geometric mean converted to a median. The non-trivial test is BLS: the model median exceeds the OEWS mean for every specialty, by 30–60%. That is the expected direction and roughly the expected magnitude given that OEWS excludes partners and owners, pools residents with attendings, and is centered on late-2023 data. Had the model fallen below OEWS anywhere, something would be wrong.

Where the three sources disagree most, and what it means for you. Dermatology is the clearest case. Doximity puts it at $550,000 in mid-2026 terms and Medscape at $461,000, a 19% gap — and BLS shows the derm mean declining across three consecutive vintages ($342,860 → $347,810 → $323,530) while nearly every other specialty rose. Three sources, three directions. The model splits the survey difference at $505,000 and grades it B, but a student choosing dermatology on compensation grounds should know that the level is genuinely contested. The same caution applies at smaller magnitude to rheumatology, infectious disease, and plastic surgery, all with Doximity/Medscape ratios above 1.17.

Training length and opportunity cost

Every additional training year is a year at roughly $75,000 instead of an attending salary, with interest compounding at 8.07% on the balance behind it. For a $500,000 attending job, one fellowship year costs about $425,000 in forgone gross income; for a pediatric subspecialty it can cost more than the fellowship ever returns.

Training years against reference median compensation

Each point is one specialty. The dashed line is the fitted relationship. Points well below it are specialties where the training investment does not price into pay.

The pediatric subspecialty penalty is the cleanest negative-return finding in the dataset. General pediatrics is a 3-year residency at a reference median near $272,000. Six of its own subspecialties — each requiring three additional fellowship years — pay less than that: endocrinology −$45k, rheumatology −$44k, infectious disease −$27k, heme/onc −$20k, nephrology −$13k, and pulmonology roughly at parity. The adult analogues are paid 21–40% more for the same fellowship length: adult endocrinology versus pediatric endocrinology, adult cardiology versus pediatric cardiology at a 40% gap. Doximity's own report states adult heme/onc earns 93% more than the pediatric equivalent. This is not a measurement artifact; three independent sources agree. B

The correlation between training length and earnings is weaker than students assume. The only quantitative analysis located put it at 0.46 — and that study is from 2009, assumed $160,000 of loans, and predates current debt levels and the 2026 loan regime entirely. It found infectious disease and endocrinology showing negative financial return versus not subspecializing at all, a conclusion this model independently reproduces with 2026 inputs. The model therefore computes net present value internally rather than importing any published estimate. C for the historical study, A for the training durations themselves.

Post-MD training length by specialty

Years from MD to first attending contract, including intern year and fellowship. Verified against ACGME program requirements where retrievable; the interventional cardiology requirement was confirmed verbatim at 12 months following a three-year cardiovascular disease fellowship. Integrated pathways (IR/DR, vascular, cardiothoracic, plastics) carry a ±1 year band because program structures vary.

Specialty	Group	Post-MD years	Attending at age	Stipend paid over training	Reference median	Forgone gross vs a 3-year path

"Forgone gross" is the additional attending-level compensation not earned during the extra training years, relative to a 3-year residency in the same specialty's pay band. It ignores the compounding of that money, which the comparison chart does not.

Practice setting and employment model

The lever students think about least and that moves earnings most. Note first what has happened to the landscape: physicians in private practice fell from 60.1% in 2012 to 42.2% in 2024, hospital-owned rose from 23.4% to 34.5%, and private equity now owns the practice of 6.5% of physicians.

In private practice, 2024

42.2%

Down from 60.1% in 2012. AMA Physician Practice Benchmark Survey, n=5,000. A

Employees rather than owners

57.5%

Owners 35.4%, independent contractors 7.1%. A

Practices owned by private equity

6.5%

And the literature does not measure what happens to physician pay after acquisition. A

Turnover after PE acquisition

+265%

Year-over-year resignation rate versus matched unacquired controls, ophthalmology, 200 practices / 1,980 clinicians. A

Predicted median by setting

Selected specialty held constant at the reference profile; only setting varies. Bar spans P25 to P75; the tick marks the median.

Which specialty

Which lever is widest

Ratio of the most to the least favorable setting of each parameter, for the selected specialty. Read this before optimizing anything.

Structure beats tenure, and it is not close. Moving from employed to partner in an office-based specialty is worth about 1.35×. Twenty-five years of experience is worth about 1.08×. Every geographic choice combined spans about 1.18×. Students spend years optimizing the parameter with the smallest coefficient.

Academic versus private: the decomposition matters

The commonly quoted "academic pay penalty" conflates two different comparisons, and only one of them is large. Against private practice pooled across employed physicians and partners, the median penalty across 25 specialties is 21%, ranging from heme/onc at 91% down to endocrinology at −1% (academic pays slightly more). But against non-partner private employees specifically, academic pay is close to parity and sometimes higher: family medicine +$11,000, nephrology +$65,000, OB/GYN +$17,000, radiology +$20,000 in academic's favor. Nearly the entire apparent penalty is the partner premium, not an academic discount. That reframes the decision: the question is not academic versus private, it is employed versus owner. The model therefore treats these as two separate modifiers and warns you when you stack both. C — the only matched academic/private dataset located is a self-reported practitioner database, not a peer-reviewed wage regression. None exists.

The second correction: the penalty is not concentrated in procedural specialties. The intuitive hypothesis fails on inspection. It holds for gastroenterology (+45%), urology (+35%), ophthalmology (+32%), and ENT (+30%) — but general surgery is +2%, radiology +8%, and anesthesiology +14%. What distinguishes the high-penalty group is not that it is procedural but that it is office-based with capturable ancillary revenue: infusion, endoscopy suites, optical, imaging, ambulatory surgery center ownership. General surgery has no ancillary stream to capture, so private practice offers it almost nothing over academia.

Academic rank progression

AAMC median total compensation, MD, clinical departments. The pattern inverts by specialty type: procedural fields front-load the entire jump at associate professor and then plateau, while cognitive fields keep climbing to full professor. Median across specialties: assistant to associate +16.4%, associate to full +7.1%, assistant to full +27.5%.

Specialty	Assistant	Associate	Professor	Asst → Assoc	Assoc → Prof	Asst → Prof
Orthopedic surgery	$511,977	$679,288	$724,956	+32.7%	+6.7%	+41.6%
General surgery	$426,104	$554,702	$651,196	+30.2%	+17.4%	+52.8%
Anesthesiology	$418,152	$451,521	$464,177	+8.0%	+2.8%	+11.0%
Diagnostic radiology	$412,356	$465,447	$495,322	+12.9%	+6.4%	+20.1%
Cardiology	$375,807	$440,449	$449,176	+17.2%	+2.0%	+19.5%
Emergency medicine	$315,013	$344,317	$356,043	+9.3%	+3.4%	+13.0%
Obstetrics & gynecology	$254,388	$305,000	$362,109	+19.9%	+18.7%	+42.4%
General internal medicine	$251,749	$290,930	$339,729	+15.6%	+16.8%	+34.9%
Family medicine	$246,009	$297,317	$349,098	+20.9%	+17.4%	+41.9%

A for the underlying survey (156 of 158 accredited schools, 99% response, 134,804 full-time faculty) but this particular table is FY2021–22 and Northeast District 1 only, because the AAMC publishes its specialty-by-rank grid behind a paywall. Use the percentages, not the levels. The dermatology row was excluded as an implausible small-cell artifact.

Locum tenens as a comparison point

Average hourly rates, self-reported, June 2025. The ratio to employed hourly pay is informative but overstated: locums rates exclude benefits, malpractice tail coverage, paid time off, and retirement contributions, and carry no income guarantee.

Specialty	Locums $/hr	Employed $/hr	Ratio	Reported annual / days worked
Gastroenterology	$367	$154	2.4×	—
Anesthesiology	$292	$133	2.2×	$163,000 / 78
Diagnostic radiology	$289	$154	1.9×	—
Pulmonary / critical care	$288	$126	2.3×	—
Cardiology	$272	$174	1.6×	—
Emergency medicine	$258	$168	1.5×	$272,000 / 85
Psychiatry	$223	$125	1.8×	$278,000 / 162
General surgery	$193	$134	1.4×	—
Internal medicine	$173	$94	1.8×	$151,000 / 88
Family medicine	$140	$96	1.5×	$139,000 / 100
Pediatrics, general	$108	$89	1.2×	$77,000 / 85
Overall average across all specialties: $215/hour. C — self-reported practitioner database.

Geography, tax, and cost of living

Physicians are one of the few high-income professions where rural pay is nominally higher. Combine that with state income tax and regional price levels and geography becomes a larger real lever than its nominal spread suggests — but the direction depends on specialty in a way most sources miss.

The rural premium is a compensation-per-unit-of-work premium, not a salary premium. A survey of about 44 rural Midwest hospitals found median compensation 10–15% above national with wRVU productivity 20–25% lower — so compensation per wRVU ran 40–50% higher. For general surgery specifically: pay +10–15%, wRVUs −35–40%, pay per wRVU +70%. Recruiters independently report rural starting salaries 9–10% above urban, with one placing the ceiling at "10% at the most." B for the survey; note the 2019 vintage.

And it inverts for referral-dependent subspecialties. MGMA first-year guaranteed compensation shows primary care at $205,588 non-metro versus $200,000 urban (+2.8%), but surgical specialists at $250,000 non-metro versus $320,000 urban, a 21.9% penalty. A rural market cannot generate the case volume an electrophysiologist, gyn-oncologist, or transplant surgeon needs. The model therefore applies opposite rural multipliers to broadly-needed specialties (1.12) and referral-dependent ones (0.92), which is a specialty interaction no single source states but which two sources jointly imply. Graded C as a construction, not a measurement.

State income tax and price level

Top marginal rate and the income at which it begins, plus BEA Regional Price Parity where retrieved. The threshold column is what matters: California's 13.30% headline rate does not touch a $600,000 income, while Oregon's 9.90% and Minnesota's 9.85% touch essentially every attending physician.

State	Top marginal rate	Threshold	Binds on physician income?	BEA RPP 2024

Two retrieved sources on state-level physician pay flatly contradict each other, and the model uses neither. Doximity's 2025 metro data places Los Angeles ($470,198), San Jose ($469,878), Sacramento ($460,671), Riverside, and San Francisco all in the national top seven metropolitan areas. A widely-cited secondary aggregator ranks California 48th of 51 states. These cannot both be true. The likely reconciliation is that Doximity's figures are cost-of-living adjusted using BEA parities while the aggregator reports nominal ranges from mixed sources — but that is an inference, not something either source states. The model therefore handles geography through three separately-sourced channels (area type, state income tax, and BEA price parity) and does not apply a nominal state compensation multiplier at all. See rejected metrics.

One pattern that does survive. BLS top-paying-area data, which is grade A and independent of both sources above, shows the highest physician wages concentrated in low-cost rural and small-metro markets: Wyoming, Mississippi, Alaska, North and South Dakota, Iowa by state, and Chattanooga, Janesville-Beloit, La Crosse, Cedar Rapids, and Eau Claire by metro. Meanwhile the lowest-paying metros in Doximity's data are academic hubs — Durham-Chapel Hill, Rochester NY, Ann Arbor. That confounds geography with practice setting and is a reason not to multiply the two modifiers naively.

Debt and the 2026 loan regime

This is the section most likely to be wrong in anything written before mid-2025. The federal student loan system for medical students changed structurally on 1 July 2026, and the change is unfavorable in one specific way that matters enormously: median debt now exceeds the borrowing cap.

Median debt, Class of 2025

$215,000

Indebted graduates. Up 5% from the Class of 2024's $205,000. 70% graduate with debt; 28% owe $300,000 or more. A

New aggregate professional cap

$200,000

Effective 1 Jul 2026, plus a $50,000 annual cap and a $257,500 lifetime federal limit. A

Grad PLUS for new borrowers

Eliminated

As of 1 Jul 2026. Existing borrowers get 3 more years or time-to-credential, whichever is less. A

Grad/professional loan rate

8.07%

Loans first disbursed 1 Jul 2026 – 30 Jun 2027. Up from 7.94% the prior year. A

The structural consequence, stated plainly. The Class of 2025 median debt of $215,000 already exceeds the new $200,000 aggregate federal cap for professional students. Four-year cost of attendance is $297,745 at a median public school in-state and $408,150 at a median private school. Students matriculating after 1 July 2026 who need to borrow the full cost of attendance will hit the federal cap and face a private-loan gap — at private-market rates, without income-driven repayment, without forgiveness, and without the interest waiver that makes RAP survivable during residency. This model's debt engine assumes all debt is federal, which will be increasingly wrong for each successive class. A on every figure; the forward implication is the model's inference.

Repayment paths available in 2026

Path	Status as of Aug 2026	Terms	Counts toward PSLF?
RAP Repayment Assistance Plan	Live, launched 1 Jul 2026	Payment is a tiered share of AGI: $10/mo flat below $10,000, then rising 1% per $10,000 bracket to 10% above $100,000, minus $50 per dependent, minimum $10. Unpaid accrued interest is waived, so the balance never negatively amortizes. A $50/month principal match applies when the payment reduces principal by less than $50. Forgiveness at 360 payments / 30 years.	Yes
Tiered / Simplified Standard	Live	Fixed 10, 15, 20, or 25-year term depending on balance.	No
IBR	Available only for loans disbursed before 1 Jul 2026	Legacy terms.	Yes
SAVE	Vacated 10 Mar 2026	Borrowers received 90-day notices to elect a new plan.	—
PAYE and ICR	Sunset by 1 Jul 2028	Existing borrowers must move to IBR, a standard plan, or RAP; auto-assigned to RAP absent an election.	—

The consequence for anyone planning on public service loan forgiveness: you must be on RAP. The Tiered Standard Plan does not generate qualifying payments. A new borrower who defaults into a standard plan out of inertia forfeits PSLF entirely.

PSLF in 2026

The 120-payment structure is intact. Only Congress can repeal it.
Employer eligibility is unchanged. The proposed rule disqualifying employers "engaged in illegal activities" was vacated on 30 June 2026, one day before its effective date, and never took effect. Government agencies and 501(c)(3) nonprofits remain qualifying. Many sources written in 2025 describe this rule as though it is in force; it is not.
Residency at a nonprofit hospital counts — but flagged as inferred. It follows from unchanged employer rules and every secondary source treats it as true, yet no authoritative confirmation was retrieved this session.
PSLF is worth much less than it used to be. One worked scenario for a 2026 matriculant with a 4-year residency and a $300,000 attending salary: about $195,000 paid over 10 years against a $230,000 starting balance, so only about $59,000 forgiven — roughly $10,000 per year of after-tax value across six attending years. The high-income RAP payment of 10% of AGI simply retires most of the balance before forgiveness arrives. Choosing a lower-paying academic job for PSLF is much harder to justify arithmetically than it was under the prior regime. C — single worked scenario from a secondary source, and the RAP tier schedule itself is secondary-sourced.
The buyback formula was revised 31 March 2026: for borrowers in SAVE on or after 18 Jul 2024, the Department looks back to the IDR plan immediately preceding SAVE rather than using SAVE's lower payments.

Resident and fellow stipends

AAMC survey as of 1 July 2025, 350 nonprofit institutions, 114,361 residents and fellows. Unweighted national averages. Nominal growth 2.2%, the slowest in four years, which is −0.48% in real terms. Regional spread from the 2024 vintage: the South runs about $62,000 at PGY-1 rising to $84,000 at PGY-8; the West about $75,000 rising to $109,000.

PGY	1	2	3	4	5	6	7	8
Stipend	$68,166	$70,499	$73,301	$77,593	$81,807	$84,744	$89,187	$94,215

One unresolved discrepancy: the AAMC debt fact card for the Class of 2025 cites a first-post-MD-year stipend of $66,986 (preliminary), against $68,166 in the stipend survey. Probably a median-versus-mean or different survey cut. The model uses the stipend survey figures throughout.

Missing-data audit

Quantified rather than described. This section is deliberately the most detailed here, because the honest answer to "what will I earn in this specialty" is partly "nobody publishes data good enough to tell you precisely, and the gaps are not random."

The structural problem: no usable physician earnings distribution exists anywhere

Source	Publishes a level?	Publishes percentiles?	Subspecialty detail?	Access
BLS OEWS, May 2025 A	Means only for 17 codes; the median is suppressed for 9 of 13 and the aggregate physician median is not published at all	P75 and P90 suppressed for 100% of physician codes at every vintage	None. 45% of physician employment sits in one residual code	Free
Doximity 2025 B	Averages, 52 specialties	No	Yes — the only free source with real subspecialty detail	Free
Medscape 2026 C	Averages, 29 specialties	No	No	Report free, detail paywalled (HTTP 402)
MGMA Provider Compensation A−	Medians — publicly only 3 rollup categories	Yes, but licensed only	Yes, licensed only	Subscription
AMGA 2025 A−	Group-level rollups plus a partial cardiology cut	No	Percent change only for interventional cardiology, EP, heart failure	Subscription
SullivanCotter 2025 A−	Percent changes only, no dollar medians	No	No	Subscription
AAMC Faculty Salary Report A	Aggregate mean only ($387,884 clinical MD, FY2024)	No	The specialty × rank grid is paywalled	Store
AMN / Merritt Hawkins 2025 B	Starting offers, 10 specialties public	No	No	Full list registration-gated

Parameter-level evidence strength

Parameter	Best available evidence	Grade	What would close the gap
Specialty compensation level, 28 major specialties	Two independent surveys agreeing on ordering, disagreeing 9.8% on level	B	Licensed MGMA or SullivanCotter payroll medians
Specialty level, 20 Doximity-only specialties	Single source, level-corrected	C	A second independent source at the same granularity
Subspecialty level, 16 specialties	Nothing published anywhere. Derived from a parent specialty by a stated rule	D	Any survey that publishes subspecialty dollar levels rather than percent changes
Earnings dispersion / percentiles	None. Federal data censored, private data publishes no percentiles	D	MGMA percentile tables, or BLS removing the top code
Training length	ACGME requirements, verified where retrievable	A/B	A consolidated fellowship-duration table; ACGME publishes only per-specialty PDFs
Resident stipends	AAMC survey, 350 institutions, 114,361 trainees	A	Nothing — this is solid
Debt, loan terms, PSLF, RAP	AAMC fact card, Dept. of Education rate announcement, statute	A	RAP tier schedule confirmed against P.L. 119-21 text rather than a secondary summary
Academic vs private differential	One self-reported practitioner database, 25 matched specialties	C	A peer-reviewed wage regression. None exists.
Partner / owner premium	Inferred from the academic-vs-private decomposition; assigned by practice-economics class	D	The associate-year to partner-year dollar step-up. Not published anywhere
Private equity effect on physician pay	Unidentified. Studies measure prices (+28% per claim in GI) and turnover (+265% in ophthalmology), not pay	—	A difference-in-differences study with physician compensation as the outcome
Rural / urban	44-hospital Midwest survey (2019) plus recruiter reports; MGMA guarantee data for the inversion	C	Current BLS nonmetropolitan tables by physician code
Suburban as a distinct tier	Nothing. Every source is binary rural/urban or metro-level	—	Any source that separates suburban from urban core
Hours worked by specialty	Medscape-derived, 2021 vintage	C	Medscape's current specialty-level hours table, which it collects but does not publish free
State-level compensation	Two sources in direct contradiction. Not used	—	A primary source with stated methodology
Cost of living	BEA Regional Price Parity, 2024 — 8 of 51 jurisdictions retrieved	C	The full 51-jurisdiction table; it exists at FRED and was not retrieved
Long-run real returns	UBS Yearbook 2026, 126-year series	A	Nothing for equities. Blended 60/40 and 80/20 real returns were not retrieved and are computed here from components
Safe withdrawal rate	Morningstar 2026 (3.9%), Bengen 2025 (4.7%), Kitces 45-year (3.5%), Trinity update to 2025	B	All four came from secondary sources; morningstar.com and kitces.com return HTTP 403 to the permitted tools
2026 IRS limits and tax brackets	Notice 2025-67 and Rev. Proc. 2025-32, verified on irs.gov	A	Nothing — independently confirmed this session
Cost of raising a child	~$299,000 (2025), built from BLS CPI. The USDA discontinued its official series after 2015 and there is no government successor	C	A federal replacement for the USDA series
Private K–12 tuition	$14,923 national average, but that universe includes parochial schools	C	NAIS independent-day-school figures; a 2026 report citing roughly $50,000 was not retrievable
Graduate / professional school cost for children	Nothing. College Board covers undergraduate only	—	Any current graduate cost-of-attendance aggregate

Specific documented voids

Sixteen subspecialties in this model have no published compensation figure anywhere. Neuroradiology, neurointerventional surgery, cardiac anesthesiology, pain medicine, surgical oncology, transplant surgery, maternal-fetal medicine, reproductive endocrinology, gynecologic oncology, orthopedic spine surgery, interventional cardiology, electrophysiology, advanced heart failure, hospitalist, palliative medicine, and pediatric hospital medicine. Each is derived from a parent specialty by an explicitly stated rule, shown in the calculator, and graded D. AMGA holds interventional cardiology and EP figures and publishes only the percent change.
No peer-reviewed multivariate model of physician earnings determinants was located at all. The only quantitative ROI analysis found is an unpublished 2009 undergraduate economics thesis.
The partner-versus-associate step-up is not published by anyone, despite being the single largest lever in the model. Years-to-partnership is published (2.0 years in orthopedics, ENT, and radiology; 3.25 in emergency medicine and pathology) but the dollar step is not.
No source separates suburban from urban. The three-tier stratification requested is only two-tier in every underlying dataset; the suburban tier here is the reference point by construction, not a measurement.
Orthopedic subspecialty compensation exists only as 2009 MGMA data. Seventeen years stale. The ranking is probably still informative; the levels are not usable, and the spine ratio was deliberately shrunk from 1.355 to 1.22 as an editorial judgment.
Fellowship durations could not be verified from ACGME for roughly 20 of 30 subspecialties. ACGME publishes duration only inside per-specialty program-requirement PDFs and AMA FREIDA does not state fellowship years. Interventional cardiology was verified verbatim at 12 months; the rest rest on two consistent secondary summaries.
Physician net worth by specialty is 2024-vintage and the 2026 edition is paywalled. The age-band data is secondary-sourced.
The nongovernmental 457(b) distinction was not verified. Nongovernmental 457(b) assets remain employer general assets subject to creditors, which materially changes the risk of the academic tax-shelter advantage. The model treats all 457(b) capacity as governmental.

Vintage lag

Source	Data vintage	Lag at Aug 2026	Effect
IRS 2026 limits and brackets	2026	0 mo	Current
Federal loan rates, caps, RAP	Jul 2026	1 mo	Current
AAMC resident stipends	Jul 2025	13 mo	Corrected forward implicitly; real growth is negative so the lag is conservative
Medscape compensation	2025	~9 mo	Trended +3% to mid-2026
Doximity compensation	2024	~20 mo	Trended +8.2% to mid-2026. No 2026 Doximity edition exists — confirmed against its own research archive
BLS OEWS	May 2025 nominal	15 mo nominal, ~32 mo effective	A three-year rolling average of panels back to Nov 2022, so effectively centered on late 2023
AAMC faculty salary	FY2024	~24 mo	Aggregate only
AAMC rank progression table	FY2021–22	~50 mo	Percentages used, levels discarded. Northeast only
MGMA ownership split	2023	~36 mo	Used as ratios only
Rural compensation survey	2019	~84 mo	Used as ratios only; the ratio is more stable than the level but 7 years is a real limitation
Hours worked by specialty	2021	~60 mo	Ordering used, levels approximate
Orthopedic subspecialty split	2009	~204 mo	Ratio only, and shrunk toward the mean

Metrics considered and rejected

Including a parameter with no evidence behind it would make the model look more capable than it is. Each decision is stated so you can disagree with it.

Metric	Disposition	Reasoning
Nominal state compensation multiplier	Rejected	Two retrieved sources contradict each other outright: Doximity places four California metros in the national top seven while a secondary aggregator ranks California 48th of 51. Neither states a methodology sufficient to adjudicate. Geography is instead handled through three separately-sourced channels — area type, state income tax, and BEA price parity — none of which requires a disputed nominal state figure.
Gender	Measured, deliberately excluded from prediction	The gap is large, well-measured, and widening: 26% in 2024, $120,917, up from 23%, exceeding 11% in every specialty examined, and driven by differential growth (men +5.7%, women +1.7%) rather than a static level difference. Medscape independently reports it crossing $100,000 for the first time and $118,000 among specialists. It is reported as a headline finding but is not a user-selectable input, because building it into a forward-looking earnings predictor for medical students would operationalize a disparity as an expectation. That is an editorial judgment, stated openly.
Private-equity ownership multiplier	Retained as a selectable setting at parity, flagged unidentified	Two grade-A difference-in-differences studies exist, and neither measures physician compensation. They establish that revenue per physician rises (GI professional fees +78%, price +28% per claim) and that retention collapses (ophthalmology resignation rate +265%, clinician headcount +46.8% over three years). Revenue rising and turnover rising are jointly consistent with pay going up or down. Assigning a multiplier either direction would be invention, so it sits at 1.00 with a wider dispersion and an explicit warning.
wRVU productivity as a user input	Rejected	The one per-specialty compensation-per-wRVU table located self-describes as "survey-style market references, not a substitute for licensed MGMA tables" and states no data year. Grade D with no vintage is not usable. wRVU context appears in the rural section, where the finding is about the ratio and comes from a survey with a stated sample.
Board certification and subspecialty certificates as a modifier	Rejected	Unlike dietetics or nursing, physician board certification is effectively mandatory for employment rather than an optional premium, so there is no meaningful uncertified comparison group. Subspecialty training is modeled as a separate specialty with its own median and its own training length, which is the correct treatment.
Rural / suburban / urban as a specialty-independent modifier	Retained, but with a specialty interaction	You asked for it and it belongs in a career-planning tool. Applying a single rural multiplier would be wrong in a way that matters: the evidence supports +10 to +15% for broadly-needed specialties and roughly −22% for referral-dependent surgical subspecialists. No single source states the interaction; two sources jointly imply it. Retained at grade C with the interaction flagged as a construction.
Subspecialties with no published figure	Retained at grade D with the derivation shown	Sixteen of them, including several a student would very plausibly be choosing between — interventional cardiology, EP, MFM, gyn-onc, surgical oncology, neuroradiology. Omitting them would be a worse answer than including them with the parent specialty, the multiplier, and the reasoning displayed in the calculator. Reproductive endocrinology carries the widest uncertainty of any value in the model.
Call burden as a compensation modifier	Retained as displayed context, not as a coefficient	Annual call days by specialty are available and useful (neurology 126, radiology 45), and per-shift call rates exist, but the published compensation figures already include call pay. Applying it as a multiplier would double-count. It is shown alongside the hourly analysis instead.
Physician net worth by specialty as a model target	Reported as context, not used	The Medscape data is real and interesting — 60% of physicians over $1M, share above $5M ranging from urology 33% to psychiatry 6% — but it is cross-sectional across ages and cohorts and cannot validate a forward projection. The 2026 edition is paywalled and the age-band figures are secondary-sourced.
Nominal-dollar projections	Rejected throughout	Every figure in this model is constant mid-2026 dollars and every return is real. A 30-year nominal projection makes late-career money look larger than it spends and inflates the apparent adequacy of any savings rate. This is the single most common error in physician financial content.

Source register

Every source, graded. A administrative census, federal statistical program, statute, or near-census payroll survey · B large survey with transparent methods, or actual contracted offers · C modeled, self-reported, small-sample, or materially stale · D derived here from a parent value, or crowd-sourced.

Source	Grade	n / basis	Used for	Verified

"Verified" means the figure was independently fetched from the primary source and confirmed, rather than accepted from a delegated research agent's report. Five consequential claims were verified this way: the 2026 IRS contribution limits against irs.gov Notice 2025-67; the Doximity subspecialty figures against Doximity's own report page; the Medscape 2026 rankings including sample size and field dates; the ACGME interventional cardiology 12-month requirement verbatim; and a two-vintage AAMC debt series (Class of 2024 fetched directly at $205,000 median / 71% indebted / 23% above $300,000, corroborating the Class of 2025 figures as a coherent +5% step).

Changelog and open questions

Version	Date	Change
1.0	15 Aug 2026	Initial build. Five parallel research agents across federal wage data, industry compensation surveys, practice-setting and geographic modifiers, the training and debt pipeline, and investing/tax/family-cost parameters. Five consequential claims independently verified. Compensation medians reconciled from two sources with a measured 9.8% source fixed effect and validated against BLS OEWS as a directional floor test, which passes at every specialty. Sixty-four specialties, sixteen of them derived from a parent by a stated rule. Financial engine built on 2026-verified IRS limits and the post-1 July 2026 federal loan regime.

Open questions carried forward

ID	Question	Status	Next action
Q1	Did BLS raise or remove the $239,200 top code for the May 2025 vintage?	Unresolved, high value	The May 2025 news-release table returned a second wage column with values up to $268.76/hour and no suppression symbol, which is impossible under a $115.00/hour top code. Either BLS lifted the code or the column was misidentified. BLS discontinued static per-occupation profile pages after May 2023 and the current profiles live in a JavaScript app the permitted tools cannot render. Resolving this requires the national XLSX at `oes/special-requests/oesm25nat.zip`, which is a ZIP archive. If the code was lifted, physician upper-tail data becomes usable for the first time and the model's entire dispersion layer should be rebuilt from measurement rather than assumption.
Q2	What is the actual associate-to-partner dollar step-up?	Resolved as a void	The largest lever in the model rests on the weakest evidence. Years-to-partnership is published; the dollar step is not, by anyone. Would need a compensation survey that separates partners from employed physicians within the same specialty and practice.
Q3	Does private equity ownership raise or lower physician take-home pay?	Unidentified in the literature	Two grade-A studies measure prices and turnover. A difference-in-differences design with physician compensation as the outcome variable does not appear to exist. Until it does, the multiplier stays at 1.00.
Q4	Can subspecialty compensation levels be obtained for the sixteen derived specialties?	Partially blocked	AMGA holds interventional cardiology, EP, and heart failure figures and publishes percent change only. MGMA and SullivanCotter hold the rest behind subscriptions. A licensed MGMA DataDive query would close most of this in one pass.
Q5	Is the true academic penalty near zero once partnership is controlled?	Suggested, not established	The one available dataset implies yes for several specialties (radiology, family medicine, OB/GYN, nephrology all favor academia against non-partner private employment). No peer-reviewed wage regression exists to confirm it. This is arguably the most consequential open question for a student choosing between tracks.
Q6	Does residency at a 501(c)(3) hospital count toward PSLF's 120 payments in 2026?	Inferred, not confirmed	Follows from unchanged employer eligibility and is treated as true by every secondary source, but no authoritative confirmation was retrieved. studentaid.gov was not reachable by the permitted tools. Materially affects the PSLF path in the model.
Q7	Are the RAP tier percentages exactly as modeled?	Secondary-sourced	The 1%-per-$10,000 schedule rising to 10% above $100,000, the $50-per-dependent reduction, the interest waiver, and the $50 principal match all come from one secondary summary. The Congressional Research Service brief is blocked by robots.txt. Should be confirmed against P.L. 119-21 or the Federal Register.
Q8	What is the real return on a blended 60/40 or 80/20 portfolio over a century?	Computed, not retrieved	UBS publishes US equities at 6.6% real, bonds 1.6%, bills 0.5% over 1900–2025. The blended figures here are computed from those components assuming no rebalancing premium and no correlation benefit, which understates a rebalanced portfolio slightly. The published blended series was not retrieved.
Q9	How much does a physician actually spend, by specialty and family structure?	Unmeasured	The model derives spending as a residual from income, savings rate, tax, debt, and stated family costs. No source measures physician household spending directly. Since the FI target is a multiple of spending, this is the largest single source of uncertainty in the financial-independence output — larger than the return assumption.
Q10	What happens to the debt engine as federal caps bite?	Will degrade over time	The model assumes all debt is federal. With median debt already above the new $200,000 aggregate cap, each successive class will carry more private debt with no interest waiver, no income-driven option, and no forgiveness. Needs a private-loan layer, which needs private-market rate data that was not gathered.

What would most improve this model, in order

A licensed MGMA DataDive or SullivanCotter query. One pass would deliver payroll-based medians and percentiles at subspecialty granularity, with academic and ownership cuts. It would upgrade the specialty level from B to A, replace all sixteen derived subspecialties with measurements, and rebuild the entire dispersion layer from data rather than assumption. Nothing else comes close.
Resolution of the BLS top-code question (Q1). Free, and potentially transformative for the upper tail.
The AAMC specialty-by-rank salary grid, currently paywalled. Would put the academic axis on grade-A footing and let the academic-versus-partner decomposition be tested rather than inferred.
Any peer-reviewed multivariate model of physician earnings. Apparently none exists. For a profession this large, this well-documented, and this consequential to individual decisions, that is remarkable.
Direct measurement of physician household spending. The financial-independence output is more sensitive to this than to the return assumption, and it is currently a residual.

Status: ACTIVE, not blocked. Version 1.0 does what was asked: 64 specialties comparable two or three at a time on a single trend line, stratified by experience, academic versus private versus partner versus locums, rural versus suburban versus urban, plus state tax, cost of living, hours worked, training length, debt under the current loan regime, employment model, and private-equity ownership — with an investing and financial-independence layer covering savings rate, tax-advantaged capacity by setting, spending multiples from 20× to 33×, Monte Carlo probability, coast-FI, and the cost of children, private school, college, and graduate school. Three things are blocked by access rather than by the absence of data (Q1, Q4, and the AAMC grid), two are genuine voids in the literature (Q2, Q9), and one will degrade predictably as federal loan caps bind (Q10). All are recorded rather than papered over.

Support this project

This application is free. Every specialty, every parameter, every chart, the whole evidence base, and the downloadable copy are available to anyone at no cost, with no account, no sign-in, no trial, no paywalled section, and no feature held back. If it is useful to you and you want to contribute toward the time it takes to keep the compensation data, tax tables, and loan rules current, there is a button below. If you do not, nothing changes.

Optional contribution

Pick an amount or enter your own. Payment is handled entirely by PayPal, which also covers Venmo, debit, and credit. Nothing about your payment is entered on this site or sent to this site’s server.

Or enter an amount (USD)

Minimum $1.00, maximum $5,000.00. The amount is re-validated on the server before any order is created.

What you are and are not agreeing to

This is a gift, not a purchase. Nothing is unlocked, upgraded, or removed by contributing or by declining. There is no supporter tier and no advertising anywhere in the application.
Not tax deductible. This is not a registered charity and no receipt for charitable deduction is issued.
This site never touches your payment details. No card number, bank account, Venmo login, or password is collected, transmitted, or stored by this site or its server. The payment window is PayPal’s, running on PayPal’s domain.
Almost nothing is stored. A random internal reference, the PayPal order and capture id, the amount, the currency, the status, and timestamps. No name, no email address, no Venmo handle, no postal address, and no payment instrument details — the database has no column in which any of those could be written.
No tracking. No analytics, no advertising pixels, no third-party scripts other than PayPal’s payment SDK, which loads only on this page section.
Refunds. Ask and it will be refunded, no reason needed.

The application works with the payment server switched off. Every calculation runs entirely in your browser. If this section shows an error, or you are offline, or you installed the downloadable single-file copy, the model still works exactly the same. That is by design, not a fallback.