How the Digital SAT Is Scored
The digital SAT doesn't count right answers and look up a score. It estimates your ability from which questions you got right and how hard they are. Here is what the College Board publishes about it, and what its own practice tests reveal.
What the College Board publishes
The College Board's technical manual describes the method:
- Item response theory. Each question has its own parameters (how hard it is, how well it separates stronger from weaker students, and how likely a guess is to be right). Multiple-choice questions use the three-parameter logistic model.
- Your ability is the value that best explains your pattern of right and wrong answers across both modules (a maximum likelihood estimate), converted to a 200–800 section score with a table.
- Routing. After module 1, a quick ability estimate from module 1 alone decides whether you get the harder or the easier module 2. The cut was set at the median student, so about half of test takers go each way.
- The easier route has a ceiling. In the manual's simulation, a student routed to the easier Math module who answers every question right earns at most about 580.
- Pretest questions, two in each module, don't count toward your score.
What it doesn't publish: each question's parameters, or a conversion table for the adaptive test. That is why no outside practice test can reproduce a Bluebook score exactly.
What the official practice tests reveal
The College Board's paper practice tests 4–11 come with answer keys and raw-to-scaled conversion tables. Matching their questions against the question bank by their official explanations turned up a few things:
- Almost all of their questions are in the question bank: 940 of 960 matched, and every match agreed with the official answer key. So the difficulty band of nearly every question on those tests is known.
- On paper, module 2 is no harder than module 1. The average difficulty band is 4.38 and 4.35 for the two Reading and Writing modules and 4.21 and 4.26 for Math. The paper versions aren't adaptive; only Bluebook's are.
- The difficulty bands are roughly evenly spaced. Math questions on these tests run from easiest to hardest, so the order in which the College Board placed them shows how far apart the bands are.
- The hardest questions have the longest solutions. Within the top band, the questions placed last have about twice as many math steps in their official explanations as the other band-7 questions.
Fitting a scoring model to the eight official conversion tables with these difficulties reproduces the tables to within about 10 points for scores of 700 and above. This site uses that model for its practice tests (the method and every constant are in calibrate.py).
What it means for your score
- Near the top, each missed question costs roughly 10 to 20 points. In this model, a perfect Math module 1 followed by the harder module 2 with one or two misses is about 790, three misses about 780, four about 760.
- Reaching the harder module 2 matters more than anything else in module 1: the easier route can't get you far past the 500s.
- Right answers on the harder module count for more than right answers on the easier one, because the questions are harder.
- Any practice test that isn't Bluebook, this one included, gives an estimate. Treat it as good to a few tens of points.
Sources: College Board, Digital SAT Suite of Assessments Technical Manual; Digital SAT Suite test specifications; Assessment Framework; paper practice tests 4–11 and their scoring guides.