A reproducible method for choosing a New York City elementary school — and the data behind each step. Start from the universities you care about, find the high schools that actually place students there, then the middle schools that feed those, and check the whole chain against the Department of Education's own numbers.
The idea
Almost every school comparison a parent is handed — tours, rankings, word of mouth, test scores — measures the school as it is today. None of it tells you where the child ends up. If what you care about is the far end of the pipe, the only honest way to compare schools is to start at that far end and walk backwards.
So that is what this does. Four steps, each one narrowing the last:
Then two things that turn the research into a decision: lay out the handful of possible paths and price each one in odds and dollars, and work out when each door opens and whether you can change your mind. The page ends with our own case, run through the whole thing, including the part where it fell short.
Working backwards tells you which schools. It does not tell you when to act — and that turns out to matter more. At most of these schools the entry door is widest at kindergarten and narrows sharply after, in some cases to zero. That makes the kindergarten choice both the highest-leverage decision in the chain and the hardest one to undo. Step 4 is where that gets measured.
This is not a ranking of good schools, and it is not advice about what any particular child needs. It optimises for one narrow outcome — placement into highly-ranked universities — because that outcome is measurable. Fit, temperament, commute, community and cost are all real and mostly absent here. Use this to narrow a list, then judge the short list on everything this ignores.
Step 1
Everything downstream depends on this list, so make it explicit. "Top colleges" is not a target; a named set of universities is.
Pick the definition you want to use. Your choice here propagates through the whole page — it sets which universities count as a hit, which re-ranks the high schools in step 2 and re-prices every path in step 4. So choose it deliberately:
The table shows the universities in the tier you picked, ranked by their five-year average rather than the current year — a single year moves for methodology reasons that have nothing to do with the school. The last three columns give each one's rank in 2015, 2005 and 1995, which is the cheapest available check on whether a reputation is stable or drifting. Switch to the full list to see the tier in the context of all 99 ranked national universities.
| 2026 | University | 5-yr avg | Yrs in T25 | 2015 | 2005 | 1995 |
|---|
There is no objective "top 25". The tier definitions used here came from a widely-circulated list plus subject rankings for the STEM tier, and reasonable parents will draw the line differently — some care about liberal-arts colleges that appear on no national-university list at all. That is fine. Change the list, and the rest of the method still runs. What matters is that you fix the target before you start comparing schools, rather than reverse-engineering the target from the school you already like.
Step 2
The measure is share of the graduating class, not headcount. A 3,200-student school sending more kids to Harvard than a 77-student school proves nothing on its own.
Eleven New York high schools publish, or have had published, a full matriculation list — every college, with the number of students who enrolled. Those 310 school-by-college rows are the entire evidence base for this step. Each school below is scored by the tier you selected in step 1.
A raw placement rate counts every student who enrolled at a top college, including the ones admitted through doors your child cannot use. Three of those doors matter enough to model:
So the adjustment cuts both ways: discount the private schools for admissions you cannot replicate, and uplift the free schools for admissions they won and could not afford to take. Set the weights yourself — the ranking below responds live, and zeroing all three gives you the raw published numbers back.
Applied as: private schools × (1 − legacy − sport); free public schools ÷ (1 − yield). These weights are judgement calls, not measurements — which is the point of making them editable rather than baking them in.
| High school | Top 25 | Class | Type |
|---|
Tap any school to see its full college-by-college list. Class size and years shown are whatever the published list covers — some are a single year, some pool several.
Have a list the eleven above don't cover — your own school, a newer year, a list a parent sent you? Paste it in and it is scored by the same tier and the same adjustment as every school in the table. Stored in this browser only; nothing is uploaded. The eleven published lists are fixed and cannot be edited or removed.
This is the step where the ranking stops agreeing with reputation. On raw numbers the elite privates lead and Stuyvesant sits eighth. Turn the adjustment on at the default weights and the two free schools move to the top of the table.
Whether you believe that depends entirely on whether you believe the weights, which is why they are yours to set. But three things hold up across any reasonable setting:
Stuyvesant's figure is its class of 2024 — 800 students, the one year with a complete published list, and the only school here drawn from a source that also publishes its own top-25 total. Recomputing that total from the tier table reproduces the published 20.13% exactly, which is the strongest check available anywhere on this page.
Other schools pool several years. A single cohort moves around more than a pooled one, so treat Stuyvesant's exact figure as noisier than the privates' even though it rests on far more students.
Step 3
One rung down, the data gets much wider and much thinner at the same time — 1,353 schools, but only one destination that can be measured across all of them.
The DOE publishes, for every middle and K-8 school in the city, the five high schools its graduates most commonly go on to, with percentages. That covers 1,353 schools. It is the single best dataset in this whole method, and it is free.
Its limitation is the shape of the report: only the top five destinations per school. So a school sending 2% of its class to Stuyvesant looks identical to one sending none — Stuyvesant simply doesn't crack the top five in either case.
That number is the real finding of this step. The public path to a specialized high school is not merely competitive — it is concentrated in a few dozen schools out of more than a thousand.
This is the easiest mistake to make with this data, and it will mislead you in the direction of the school you are probably considering.
A K-8 school's graduates all have to go somewhere else, so its Stuyvesant share is a share of everyone. A K-12 school keeps most of its own students — they roll into its own high school and never sit in the admissions market at all. Its published Stuyvesant share is diluted by all of them.
NEST+m is the clearest case. 47% of its graduates continue into its own high school. Its headline Stuyvesant figure is 13% — but that is 13% of everyone, including the near-half who never applied out. Among the 53% who did enter the market, 24.5% land at Stuyvesant. Against a K-8 school's 8%, the true gap is about 3×, not the 1.6× the raw figures suggest.
145 of the 1,353 schools retain some of their own graduates this way. The toggle below switches every number between the two bases; the "stays on" column shows how much dilution each school carries. Use "share of those who leave" when comparing a K-12 against a K-8.
| School | → Stuyvesant | → Big Three | → Any specialized | Stays on | DOE rank |
|---|---|---|---|---|---|
| Loading 1,353 schools… | |||||
"Stays on" is the share continuing into the school's own high school. Tap any school for its full destination list and its DOE quality history.
New York has eight specialized high schools, and treating them as one block hides a lot. Three of them — Stuyvesant, Bronx Science and Brooklyn Tech — are the original testing schools, are by far the largest, and are the ones with a genuine STEM reputation. The other five are smaller, newer, or built around different subjects.
The Big Three column counts only those three, which is the narrower and usually more meaningful measure. 115 schools feed at least one of them; 124 feed at least one specialized school of any kind. Where the two columns differ for a school, the gap is going to one of the smaller five.
Staten Island Technical High School is excluded here, so this is a "Big Three" rather than a "Big Four". It is a strong STEM school, but its intake is heavily borough-local, so including it flatters Staten Island feeders and changes little elsewhere — it would add six schools to the 115. Say the word and it is a one-line change.
A high feed rate is not proof the school causes the outcome. Citywide gifted-and-talented programmes admit by test at age four or five, so their graduates were already a selected group before the school taught them anything. The 28% and 13% at the top of this table are a mix of school effect and selection effect, and this data cannot separate them.
What it can tell you is where the door is. If a school has no measurable pipeline to the high schools from step 2, your child would be the exception rather than the rule there — which is possible, but is a plan that depends on you, not on the school.
This report covers schools that graduate students into high school — middle schools and K-8s. There is no equivalent city-wide report mapping which elementary school feeds which middle school, so the bottom rung of the ladder cannot be measured the same way.
What can be established is retention: a strong K-5 typically keeps 80–85% of its fifth graders in its own middle school if it has one, and the minority who leave go to citywide G&T programmes or private schools. So for a K-8 or K-12 school, the kindergarten choice largely is the middle school choice. For a K-5, it is a choice you will have to make again — which brings us to the part of this that actually decides things.
Step 4
Only three things are actually on the table at kindergarten. Everything else — private at 6th, Hunter at 7th, private at 9th — is not a choice you make now. It is an option your kindergarten choice leaves open, and one you would have to win years later.
Collapsing the two into one list of "paths" makes them look like alternatives you pick between today, which is the single easiest way to reason badly about this. So they are separated below: the decision, then the strategy it commits you to.
At kindergarten you hold whatever offers you hold. There is no probability to estimate here — you either have the place or you do not — so this table carries no odds of admission and no cost. It asks only: what does each option actually do, and what does it leave open?
| Option at kindergarten | Where its students go | The gate still ahead | What it leaves open |
|---|
Read the last column rather than the second. The three options are not far apart on where students end up; they are very far apart on what happens if it goes wrong.
A zoned school's recovery plan is to win a private place at 6th or 9th, from a non-selective school, against a field of families spending heavily to do the same. A K–12 gifted programme's default is to stay put — its fallback is a school, not an application. A private K–12 has nothing to recover from because there is no second gate.
That is the real content of the kindergarten decision, and it is why it deserves more research than its timing invites.
Now the long view. Every strategy is broken into three parts, because they answer different questions and get confused constantly:
Look down the "once in" column for the three private routes. The figures are identical, and they have to be: it is the same school. A child admitted to Horace Mann at 9th grade sits in the same classes, with the same teachers, applying to the same colleges as one who arrived at kindergarten. Joining later does not buy a worse outcome — it buys fewer years of the identical one.
So on price per unit of outcome, later entry wins, and by a lot:
That is the correct way to read the money, and it is the reverse of what an earlier version of this page said. The tuition is only ever paid by families who got in, so pricing it against the odds of a child who was not admitted — and therefore never paid — charges you for a branch that costs nothing.
The real trade is not cost against quality. It is cost against reliability. Kindergarten entry is expensive and, once you hold the offer, certain. Ninth-grade entry is a third of the price for the same education and roughly a one-in-twenty shot at getting it. Most families who plan on entering later never get to spend the money at all — which is a cheap outcome and a bad one.
Read the "overall" column for how likely you are to end up there, and the cost-per-point for what it costs if you do. Neither number answers the other's question.
| Strategy | Getting in | Once in → Top 25 | Overall | Tuition |
|---|
Figures in boxes are estimates you can change; bold figures are observed data. Cost per point is priced against the "once in" odds, because tuition is only paid if you get in. Tap any strategy for the derivation of every leg.
Three gates have no published admission rate anywhere: private entry at 6th grade, private entry at 9th, and the Hunter exam. Leaving them blank would be tidier but useless — you cannot compare strategies without them — so each is estimated from the best available anchor and the derivation is in the row:
These are estimates, not measurements, and they carry the most uncertainty on the page. They are also the numbers that decide whether the late-entry strategies are worth anything, so change them and watch what moves.
The useful test is how wrong an estimate would have to be to change the conclusion. For the 6th-grade route, the breakeven against a citywide G&T seat is about 24% — three times the estimate used here, and well above even the optimistic "10–20% for outside applicants" figure for private middle schools generally. So the ordering survives being wrong by a wide margin. If you think 8% is too pessimistic, put 15% or 20% in the box; the row moves, but not past the free option above it.
Ranked by overall odds, kindergarten entry wins comfortably — it is the only one of the three whose door you have already walked through. Ranked by price per unit of outcome, the order reverses exactly, because you are buying fewer years of an identical education.
Neither ranking is the answer on its own. Together they say something fairly plain: paying from kindergarten is buying certainty, not quality, and certainty at roughly $390,000 more than the same thing bought at 6th grade. Whether that premium is worth it depends entirely on how you weigh a one-in-twelve chance of the 6th-grade door opening — which is an estimate on this page, not a measurement, and the number most worth arguing with.
The free routes sit outside this trade altogether. A citywide G&T seat delivers a lower "once in" figure than any private route and costs nothing, and Hunter delivers the highest second leg on the whole page for free. Cost-per-point is undefined for both, which is the point.
Step 5
Steps 1–3 tell you which schools. This step tells you when the door is open, how wide, and what it costs to arrive late. For most families it is the step that actually determines the decision.
Every school in step 2 admits students at specific grades and essentially not at others. A K-12 private fills its 9th grade almost entirely with its own 8th graders moving up; the seats offered to outside applicants are whatever is left when families move away. Here is what that looks like in seats:
Approximate seats offered to outside applicants at each entry point. Sourced from school-published figures and admissions guidance; treat as order-of-magnitude, not exact.
For five of these six schools, kindergarten is the widest door by a large margin and it never opens that wide again. Collegiate does not admit at 6th grade at all — zero seats. Dalton goes from 65 seats to 15 to 10.
Horace Mann is the exception that proves the rule useful: it is narrower at kindergarten than at 6th grade, 30 seats versus 55. If you want a realistic non-kindergarten entry to an elite private, this data says that is where to aim — and it is the kind of thing you only see once you put seat counts next to each other.
The most common thing parents tell themselves is that if the public route doesn't work out, private school at 9th grade is the safety net. The seat counts say otherwise. Across the elite K-12 privates there are roughly 10 to 25 outside seats per school at 9th grade, city-wide, and the applicants competing for them include every family whose specialized-high-school plan just failed on the same test date.
Worse, at that size a school is not choosing the strongest applications — it is filling specific gaps in a class. A child can be an excellent candidate and still lose to institutional need. A plan whose fallback is a ten-seat lottery is a plan with no fallback.
Because each rung is conditional on the one before it, the honest way to compare two starting points is to multiply the chain through. Two public routes to the same endpoint:
The 0.68 and 0.19 are modelled from feed rates and retention, not published statistics, and the 0.30 is a conditional SHSAT success rate for an already-selected group, not a citywide admission rate. Do not take 20.4% and 5.7% as facts. Take the structure as the point: the difference between two kindergartens compounds through every year that follows, and by high school it is a multiple, not a margin.
Put the two findings together. The doors are widest at kindergarten and mostly shut afterwards. The odds multiply, so an early difference compounds. And the standard fallback — buy your way in later — is competing for ten seats.
That makes the kindergarten choice high-leverage and low-reversibility: close to the most consequential schooling decision you will make, taken with the least information about the child, at the point when they are four years old. That is uncomfortable, and it is also just true. The correct response is not to agonise — it is to notice that the decision deserves considerably more research than its timing invites, and to prefer options that keep more doors open where the placement data is close.
Where two schools are within a few points of each other on placement, the tiebreaker should be optionality: a K-8 or K-12 that removes a future admissions round is worth more than a marginally better feed rate at a K-5 that forces you to run this process again in five years.
Step 6
The first four steps rank schools on one outcome, using data schools chose to publish. This step is the independent check — the city's own measurements, 2,142 reports covering 1,984 schools across eight years.
The NYC DOE school-quality reports give every school two scores. Performance is raw student achievement, unadjusted. Impact measures the school against the outcomes expected for the students it actually enrolled — an attempt to isolate what the school contributed. Both matter, and they answer different questions.
Neither number is a percentage or a grade. For each metric the DOE takes the gap between the school's result and the benchmark — the citywide average for Performance, the school's Comparison Group for Impact — and rescales that gap onto 0.00–1.00 against the range of gaps seen across schools of the same type, capping outliers. The score you see is a weighted average of those rescaled gaps.
Three consequences worth holding onto. The scale is relative: 0.98 means the school's gap was near the top of the range its peers produced, not that it got 98% of anything. Zero gap is not 0.50 — break-even sits wherever the observed range puts it, so a school can score above 0.50 while still trailing the city. And because the rescaling is done per school type, a 0.98 elementary and a 0.98 high school sit near the top of two separate ranges, and are not the same achievement.
The metrics behind the scores differ by report too. For elementary, middle and K–8 schools it is 50% state ELA tests and 50% state math tests. For high schools it is four things at 25% each: Regents ELA, Regents Algebra I, the 4-year graduation rate, and post-secondary enrolment within six months of graduating.
Same tests does not mean the same number. Elementary, middle and K–8 schools all sit under one DOE report type, but it is scored on the grades each one actually teaches: grades 3–5 for an elementary school, grades 6–8 for a middle school, grades 3–8 for a K–8. The citywide average it is measured against is computed the same way, from schools of that type only. So a middle school at 0.96 and an elementary school at 0.96 are two different cohorts of children, tested in different years of school, each placed against a different benchmark. Reading them as one ranking is the mistake this step is supposed to prevent.
A school running both middle and high school grades is scored twice by the city under one DBN: once on the state tests its middle grades sit, once on Regents and graduation. 158 schools in this data have both. They measure different children on different things, and neither continues the other — so this page keeps them as two rows you select independently. NEST+m is the case to look at: over the same eight years its grades 3–8 report runs 0.96 to 1.04, and its high school report runs 0.89 to 0.91.
Pick as many schools as you like and they line up side by side. It starts on four Manhattan schools; remove them and add your own.
Two views of the same schools. DOE score is the dashboard number described above — adjusted for intake on the Impact side, but rescaled within school type, so it cannot answer "is this K–8 better than that K–5". State tests is the raw material underneath it: the percentage of students at Level 3 or 4 on the state exams, which the city publishes per grade. Aggregated into the grades a school actually teaches, it compares across any two schools — at the cost of telling you nothing about who the school admitted. Use the first to judge a school against its peers, the second to judge two different kinds of school against each other.
Two well-regarded District 2 middle schools, in the years around the change to how their incoming students were selected. NYC Lab is in the comparison above by default — add Salk to it and the two lines sit side by side:
Both had sat between 0.93 and 0.97 every year since 2015 — then dropped in the same year. Same buildings, same teachers, same curriculum. What changed was who walked in the door. A parent comparing schools on a single recent snapshot would read that drop as a school getting worse, when what it actually shows is how much of a screened school's measured quality was selection all along.
This is the single most useful thing in the DOE data and the reason to look at the trend rather than the current number. It applies with equal force to step 3: the schools at the top of that feed-rate table are selecting, too.
Everything in the table above is read off the DOE's own school-quality reports, published at tools.nycenet.edu/dashboard. The DOE dashboard link on each row opens that school there directly — for example NEST+m. Check any figure here against the source, and use it for the many things this page does not extract: attendance, teacher and family survey results, and the individual metrics behind each score.
Note that the series runs 2015–2019 and 2023–2025. The pandemic years are missing from the city's reporting, which means any comparison spanning that gap is comparing two different eras of testing and admissions policy.
It is worse than a gap. Both scores were rebuilt in 2023 with MIT Blueprint Labs on a smaller set of metrics, and the Comparison Group behind Impact moved from a student-matching method to a regression. The numbers either side of the gap are not the same measurement. Read the trend within 2015–2019 and within 2023–2025, and treat the step between them as unmeasured. The 2015 column predates the city's current API and could not be re-fetched per report, so it is left blank for the 158 dual-report schools rather than guessed at.
The worked example
This method was not built in the abstract. It was built to settle one real decision, in the spring of 2026, with a deadline — and it is easier to trust once you can see where it bent under a real case.
The choice was between NEST+m, the city's only K-12 citywide gifted-and-talented programme, and PS 276, an excellent zoned elementary in Battery Park City, five minutes from our door. An offer from NEST+m arrived on 31 March; the acceptance deadline was in May.
Both schools appear in step 3, so the comparison is a direct one on identical data:
Read that way it is a 1.6× gap on Stuyvesant and 2.5× across the specialized schools. But those two schools do not have the same denominator, and correcting it is the single most consequential thing we did to our own numbers.
NEST+m runs K–12, and 47% of its eighth graders simply continue into its own high school. They never take a seat in the admissions market. PS 276's graduates have no such option — every one of them applies out. Comparing 13% against 8% therefore penalises NEST+m for the students who had somewhere good to stay.
On the same basis — students who actually enter the market — it is 24.5% against 8%. About 3×, not 1.6×. We would have under-rated NEST+m by nearly half if we had taken the published figures at face value.
That comparison has a hole in it, and it is worth being honest about because your version will have one too. PS 276 sits in a wealthy neighbourhood, and a meaningful share of its families go private for middle or high school. Those children are not in the 14% — they left the public system entirely, so the DOE report cannot see them.
We never obtained PS 276's private-school placement rate. We asked; it is not published, and the school does not track it in a form anyone would share. If that number is high, the real gap between the two schools is materially smaller than 2.5×. We made the decision without it, knowing it was the one input that could have changed the answer.
Not the placement gap. The thing that decided it was the asymmetry in step 5 — which weaknesses could be fixed and which could not.
NEST+m's problems were real and we catalogued them: dirty classrooms, a science room reading 1,300 ppm CO₂, a commute across town, a pace fast enough that 10–15% of kindergarteners repeat the year. But every one of those is operational. They can be pushed on, paid around, or supplemented at home.
PS 276's weakness was structural. A standard-pace curriculum, a general peer group, and no institutional pipeline. Those cannot be fixed from the parent side — only left. And leaving means re-entering the admissions system later, at exactly the grades where step 5 shows the doors are narrowest.
So the deciding question was never "which school is better today". It was: if this turns out to be wrong, which mistake can we still recover from? That is the question the seat counts answer, and it points the same way the placement data does.
We accepted NEST+m. Theo started there. We stayed on the PS 276 waitlist while the decision was live, because accepting cost nothing and preserved the option.
The path in step 4's table is the first one — public, citywide G&T, all the way, aimed at the specialized high schools. We did not commit to it exclusively. NEST+m runs K–12, which means the 6th-grade decision point is a genuine choice rather than a forced re-entry, and the hybrid door at 6th grade stays open if the fit turns out wrong.
Three things this page cannot do, which we hit directly:
We used the data to establish that the gap was real and to understand what we could not undo. Everything after that was judgement, and it should have been.
Do it yourself
All of it is public. None of it required paid tools or an admissions consultant. The work is assembly, not access.
Write down 15–30 named universities and why. Published national rankings are a starting point; the five-year average is more stable than any single year. If your priority is a specific field, use subject rankings instead — the general list will mislead you.
Search for "[school name] matriculation list" or "[school name] college destinations". Many private schools publish several years as a PDF; some public school lists circulate through parent and student communities. Record students enrolled per college and total class size — a list without a denominator is unusable. Then compute the share of the class landing in your step-1 list.
You do not have to compute it by hand: paste the list into the box under the step-2 table — name and number, one college per line — and it is scored against the tier you picked, adjusted with the same weights, and ranked against the eleven published lists. It is saved in your browser, so it survives a reload and never leaves your machine.
The NYC school-quality reports and the MySchools directory carry the "graduates most often attend" breakdown for every middle and K-8 school. Look up any school you are considering and check whether the high schools from your step 2 appear at all.
Ask every school directly, in these words: "How many students do you admit from outside the school at kindergarten, at 6th grade, and at 9th grade?" Admissions offices answer this readily; it is rarely on the website. Those three numbers tell you more about your real chances than any tour will.
List the four or five realistic shapes — public throughout, private throughout, and the hybrids that buy in at 6th or 9th. For each, multiply the stages through and put a total tuition figure next to it. The comparison you want is odds against dollars for a whole path, not a ranking of individual schools.
For each path, write down your options if it fails at each rung — using the seat counts from step 4, not optimism. Paths whose recovery plan is a ten-seat admissions round are not recovery plans. Where two options are close on placement, take the one that leaves more doors open.
This method optimises hard for one outcome, and it does so at the age when a child is least legible. It cannot see whether a school is kind, whether your child will be happy there, or whether the pressure that produces these placement numbers is pressure you want in your house. Those questions have no dataset and they are not less important for it.
Use this to rule out schools that cannot get you where you want to go, and to know honestly what a given path costs and when it closes. Then choose among what survives on grounds this page knows nothing about.
Where a figure is an estimate rather than a published number, the page says so at the point of use. The two places that matters most are the placement adjustment in step 2 and the odds chains in step 4.