Does "Advanced" Predict College? We Checked 1,732 Schools

We went looking for a reason a bigger Advanced share would put more kids in college. Three guesses, three letdowns.

Grad rates. We checked this five ways. Four found nothing at all — a gap that size would turn up by luck roughly 4 to 7 times in 10. The fifth, on a different test year, showed about a 2-point gap and didn't repeat anywhere else. Whatever Advanced share is picking up, it doesn't look like who finishes high school.

AP pass rates. This looked real at first: a 9-point gap between higher- and lower-Advanced schools. Pull out the charter schools and it drops to under 3 points, and a gap that size shows up about 1 time in 3 by luck alone. Too weak to trust. A pass rate only counts kids who sat the exam, so pushing more kids into AP can make a school's own pass rate look worse for doing something right.

A-G courses — the high school course sequence UC and Cal State build their admission requirements around. (Coursework isn't the only route; UC says students "may meet them by completing college courses or earning certain scores on various acceptable exams.") This looked like the answer: a real gap, two years running. A third year made it vanish and flip sign. We're calling it unproven.

We got it wrong twice ourselves before we got here. An early version of this work said Advanced share added nothing. A second version said it hurt community college numbers. Neither survived matching schools that actually looked alike. We're telling you because it's what a wrong answer looks like on the way to a smaller, truer one.

This is SchoolScope's own analysis, not a CDE published figure. The underlying test scores and college destinations are California Department of Education data. The pairing, the gaps and the ranges below are ours.

A term first: a confidence interval is the range of gap sizes our data can't rule out. When we say "0.8 to 3.7 per hundred," a true gap anywhere in that stretch would be consistent with what we saw. If the range crosses zero, "no gap at all" is one of the answers still on the table.

The matching rule, and who it covers. We paired a school only with another school close to it on three things: overall test score, share of students on free or reduced-price lunch, and neighborhood median income. Just 254 of 1,732 eligible schools — about 15% — were similar enough to anything else to get matched. That isn't a random one in seven.

Matched schools are higher-scoring, less poor, and about half again as likely to be charters as the schools left out. The class-of-2023 numbers in this post describe those 254 schools. They aren't a claim about California high schools in general.

What changes when the match gets stricter. We reran the same comparison adding neighborhood education level and racial makeup as things a pair had to match on — one at a time, then together.

Read down the UC column and you can watch the finding come apart. The four rows that add the neighborhood variables all have intervals reaching past zero, and the two strictest put the gap at zero or just under it. The rows that keep a clear UC gap — no charters, bigger schools — are the ones that don't balance the neighborhood. In the four-year column, only one of the eight rows has an interval that stays above zero.

Two checks disagree about why it falls apart, and we can't referee between them at this sample size. Re-matching says the neighborhood variables were carrying most of the original gap. But a separate check — regressing each pair's UC gap directly against how unbalanced that pair was on neighborhood education — attributes only about half a point of the 2.24 to that imbalance. Some of the collapse is probably confounding. Some of it is probably that stricter matching leaves 43 pairs instead of 127, and different schools inside them.

Same-district matching found only 5 pairs. Comparing schools genuinely near each other, rather than similar on paper, would be the tightest version of this test. California doesn't have enough high schools that close on everything else to run it.

A second graduating class, and what it says. CDE hasn't published the class-of-2024 file yet — its DataQuest schedule puts the 2023–24 College-Going Rate report in fall 2026, and lists no date yet for 2024–25. We'll re-run this when they land. In the meantime we pulled a cohort we already had: the class of 2020, 1,697 schools, 128 matched pairs.

The community-college result repeated cleanly, and tighter: no shift bigger than about 3.2 students per hundred, against 4.2 in the newer class. UC came in positive again at +1.5 per hundred, and softened toward zero under the same stricter matching. Four-year came in at +1.5 with luck explaining a gap that size about 3 times in 10 — no support there.

One thing complicates the 2020 class. Those students finished high school into one of the most disrupted enrollment years California has recorded: community college headcount fell 14%, from 1,567,787 to 1,347,031. So read it as replicating the pattern, not the size.

What CDE's numbers actually count. The College-Going Rate covers high school completers, which is slightly broader than graduates — it includes CHSPE and GED completers and adult-education diploma earners. And "went to college" means enrolling anywhere in the US within 12 months of that student's own completion date, not by a particular autumn.

A note on the labels. California renamed the CAASPP reporting levels starting with 2025–26 results: Standard Exceeded became Advanced, Standard Met became Proficient. The cut scores didn't move. The test years behind this analysis (2019 and 2022) were published under the old labels; we use the new one throughout. Full mapping here.

What changed on this page. We were reading the wrong column out of the state's score file, which inflated our count of how many students sat the test at each school. We fixed it on 2026-08-15 and re-ran everything. A school had to have enough juniors tested to enter this analysis at all, so the corrected count dropped about 80 schools from each graduating class — and the results moved with them. The four-year college gap was the casualty: it was our headline effect, and it no longer clears the bar. The community-college result and the shaky UC gap both survived the re-run. The figures on this page are the ones the corrected data produces.

Want to check the numbers? The matching and the statistics run start to finish from one script with a fixed random seed, so two runs on the same data agree exactly. If you want to walk through the method or the numbers with us, get in touch.