How this calculator and practice test actually work.
Most SHSAT calculator sites won't tell you how their numbers are produced. We'll tell you everything. This page is the technical reference for our score conversion algorithm, our adaptive test engine, the data sources behind both, and the honest limitations of each. If you find an error, email [email protected].
- What model does the calculator use?
- Piecewise linear interpolation between 16 anchor points spanning raw 0–57 to scaled 100–400, derived from publicly available NYC DOE conversion data.
- What model does the practice test use?
- 3-Parameter Logistic Item Response Theory (3PL IRT) with Expected A Posteriori (EAP) ability estimation and Maximum Fisher Information item selection.
- How accurate are the results?
- Calculator: within ±5 scaled points of the official conversion for most raw scores. Practice test: typically ±30 points of true SHSAT composite, narrowing as the item pool grows.
- Where does the data come from?
- NYC DOE Specialized High Schools Student Handbook (public PDFs), MySchools NYC offer data, and historical cutoff scores published by the NYC DOE. Full citations below.
Part 1: The Score Calculator
What the calculator does
The calculator takes two inputs — your raw ELA score and your raw Math score — and returns three outputs: an estimated ELA scaled score, an estimated Math scaled score, and the composite (their sum). It also shows which specialized high schools your composite would qualify you for, based on the most recent published cutoff data.
The conversion model
The NYC Department of Education does not publicly release the exact raw-to-scaled conversion formula used for the SHSAT. The conversion is performed each year via a process called test equating, which adjusts the scaled scores to account for slight differences in test difficulty across forms. Because the equating tables aren't published, no third party — including us — can produce results that exactly match the official scaled scores.
What we can do is estimate the conversion based on publicly available data. We use piecewise linear interpolation between anchor points derived from:
- Sample conversion tables historically published in NYC DOE Specialized High Schools Student Handbooks (2014–2024 editions)
- Cross-referenced patterns documented in independent analyses, particularly Kenny Tan's reverse-engineering work and the published conversion guidance from established prep providers
- The general shape of equated-score curves, which is steeper near the maximum (each correct answer in the 50–57 range is worth more scaled points than each correct answer in the 30–37 range)
The anchor points used by the calculator are:
| Raw | Scaled (estimated) | Notes |
|---|---|---|
| 0 | 100 | Floor |
| 10 | 155 | Low band |
| 20 | 205 | |
| 30 | 248 | Median band |
| 40 | 300 | |
| 45 | 331 | Curve steepens |
| 50 | 365 | |
| 54 | 388 | High band |
| 57 | 400 | Maximum |
Between anchor points, the calculator linearly interpolates. For example, a raw 35 falls between anchors at raw 30 (scaled 248) and raw 40 (scaled 300), so the calculator returns a scaled score of approximately 274.
Accuracy of the calculator
Comparing our estimates to publicly reported official conversions:
- Low-to-mid raw range (10–40): typically within ±3 scaled points of the official conversion
- High raw range (45–57): typically within ±5 scaled points, with slightly more variation because the curve is steeper here and yearly equating moves it more
- Per-year drift: because the DOE re-equates each year, the same raw score can produce slightly different scaled scores from one year to the next. Our calculator estimates a typical conversion rather than year-specific.
What this means practically: if our calculator says your composite is 525, your actual SHSAT composite from the same raw scores will likely fall somewhere between 515 and 535. That range happens to straddle multiple specialized-high-school cutoffs, which is why we don't treat the calculator output as a definitive prediction.
Part 2: The Adaptive Practice Test
The practice test is more technically involved than the calculator. It uses Item Response Theory — the same family of measurement models that ETS uses for the GRE, the College Board uses for digital SAT, and the NYC DOE uses for the official SHSAT-CAT. We've built our engine from scratch and the implementation is described in full below.
The 3-Parameter Logistic model
The probability that a test-taker at ability level θ (theta) answers a given item correctly is:
P(correct | θ) = c + (1 − c) / (1 + e−1.7 · a · (θ − b))
where each item has three parameters:
- a — discrimination. How sharply the item distinguishes test-takers above and below its difficulty. Typical range 0.6–1.8. Higher means the item is more "diagnostic."
- b — difficulty. The ability level at which a non-guessing test-taker has a 50% chance of getting the item right. Typical range −2.5 to +2.5.
- c — guessing. The probability of getting the item right by random guessing alone. For our 4-option multiple choice items, this is roughly 0.20–0.25.
The constant 1.7 in the exponent is the standard logistic-to-normal scaling factor used in IRT to make the logistic curve approximate the normal ogive used in some related models.
Ability estimation: EAP, not MLE
Many introductory descriptions of IRT use Maximum Likelihood Estimation (MLE) to estimate ability. MLE has a well-known problem in adaptive testing contexts: when a test-taker has answered very few items, or has gotten all of them right or all of them wrong, the likelihood function has no interior maximum and the estimate either diverges to ±∞ or has to be artificially clamped. We've observed this firsthand during development and ruled it out.
Instead, we use Expected A Posteriori (EAP) estimation with a standard normal prior:
θ̂ = ∫ θ · posterior(θ | responses) dθ
The posterior is computed via numerical quadrature over the θ range [−4, 4] with 81 nodes. The prior is a normal distribution with mean 0 and standard deviation 1.5 (slightly wider than the standard EAP prior because SHSAT test-takers are a more variable population than the general one). The standard error of the estimate is the square root of the posterior variance.
EAP is more stable than MLE for small samples, doesn't blow up on extreme response patterns, and produces sensible estimates from the very first item — at the cost of a small Bayesian shrinkage toward the prior mean. We accept that trade-off because it's the right one for a practice test of this size.
Item selection: Maximum Fisher Information
After each ability update, the engine selects the next item by computing the Fisher Information that each available item would contribute at the current θ:
I(θ) = (1.7a)2 · (P − c)2 · (1 − P) / [(1 − c)2 · P]
The item with the highest information at the current estimate is the one that will most tighten the standard error after being answered. To prevent the same items from being served to every user (which would degrade item security over time), the engine uses randomesque selection: it picks uniformly from the top 3 information-maximizing items rather than always picking the single top item.
Termination criteria
A section ends when one of three conditions is met:
- The maximum number of items has been administered (currently 12 per section)
- The minimum has been administered (currently 6) AND the standard error has dropped below 0.32 — meaning the estimate has converged
- The item pool has been exhausted
Scaled score translation
The final ability estimate θ̂ is translated to the SHSAT scaled score range (100–400 per section) by the linear mapping:
scaled = clip(round(250 + 45 · θ), 100, 400)
The composite is the sum of the two section scaled scores, ranging from 200 to 800. This mapping is calibrated so that the median SHSAT-taker (θ ≈ 0) produces a scaled score near 250, which matches the rough center of the historical SHSAT scaled-score distribution.
Item pool and known limitations
As of May 2026, our item pool is 105 items (50 ELA and 55 Math), all reviewed by Elisa Ahmed before going live. The real SHSAT-CAT operates on hundreds of calibrated items per section; with our pool, the engine still converges on a reasonable estimate, but the standard error around your true ability is larger than the official test’s. As the pool continues to grow, that interval tightens.
What this means in practice:
- With 12 items in a section, the typical SE on θ at the end of the test is about 0.4–0.6
- That translates to a scaled score uncertainty of roughly ±18–28 points per section
- Composite score uncertainty is therefore typically ±25–40 points
- For most test-takers this is good enough to know which "tier" of specialized high schools they're competitive for, but it isn't precise enough to distinguish between adjacent schools (Bronx Science at 518 vs HSAS at 516, for example)
The pool grows as we add and review more items. Every new batch of items is reviewed by Elisa Ahmed before being added.
Part 3: Data sources
The cutoff scores, school information, and conversion anchor points used on this site come from the following primary sources. We try to link to original sources rather than aggregators.
Primary sources we use
- NYC DOE — Specialized High Schools page. The official page on SHSAT-eligible schools, registration, eligibility, dates, and accommodations.
- MySchools NYC. The official application portal where families register for the SHSAT and rank their school preferences. We do not interface with MySchools — links are for users to register directly.
- NYC DOE Specialized High Schools Student Handbook (annual PDF, published each year by the DOE). The handbook includes sample questions, scoring information, and policy documentation. Available from schools.nyc.gov.
- NYC DOE published cutoff scores (typically released each spring via the DOE communications). Our cutoff data reflects the most recent published cycle (2025–2026 admissions, results released March 2026).
Secondary references and acknowledgments
- Independent published analyses of SHSAT scoring patterns by educators and tutors, including reverse-engineering work that has been publicly shared and replicated
- Academic literature on Item Response Theory, particularly Lord (1980), Hambleton, Swaminathan & Rogers (1991), and Bock & Mislevy (1982) on EAP estimation
- Wikipedia entity references for school articles (linked individually from each school page) — used as a starting reference but cross-validated against official school websites
What we don't use
- We do not use questions from any published SHSAT prep book (Barron's, Kaplan, Princeton Review, or similar). All practice items on this site are original.
- We do not scrape questions from other SHSAT prep websites.
- We do not use unofficial or leaked DOE materials.
- We do not buy data from any private SHSAT data brokers or "answer key" services.
For a list of every page that has been corrected based on reader feedback, see the methodology changelog.
Honest limitations we want you to know about
This site is built carefully but it isn't the NYC DOE, and we don't pretend otherwise. Specific limitations:
The exact official conversion is not public. Our calculator estimates from publicly available data. Real scaled scores may differ from ours by a few points in either direction.
Cutoff scores change every year. The cutoffs displayed on the site are from the most recent published cycle. Future cycles will have different cutoffs, and meeting last year's cutoff does not guarantee admission this year.
Practice test pool is smaller than the real exam's. Our adaptive engine works correctly, but with a 105-item pool the confidence intervals on your estimated score are wider than they would be on the official test (which pulls from hundreds of items). We label confidence on every result.
We are not authorized by the NYC DOE. This is an independent resource. The NYC DOE has not reviewed, endorsed, or approved anything on this site.
For official information, always use the NYC DOE. For registration deadlines, accommodations, test dates, and your actual results, the only authoritative source is the NYC Department of Education at schools.nyc.gov and myschools.nyc.