SAT Scoring: Understand the Algorithm

What You See and What They're Hiding

Your Official Score Report will contain:

• A Math Score (200–800)
• A Reading & Writing Score (200–800)
• A Total Score (400–1600)
• Percentiles for all three scores, saying how you compared to all test takers
• Broad performance bands, which tell you very general areas of weakness but are mostly too broad to be helpful.

Your Official Score Report will NOT tell you:

• How many questions you got wrong
• Which questions you got wrong, or even which types of questions you got wrong.

Why Does The College Board Hide This Information?

• They won't share the questions you got wrong because they reuse questions from previous tests.
• They won't tell you how many questions you got wrong because it would be a nightmare to explain why Johnny did so much worse than Sally even though he answered 10 more questions correctly.

While you generally shouldn't obsess over exactly how the score will be calculated, there are some key points that impact strategy.

3 Ways to Beat the Scoring Algorithm

Even if you don't want to bother with the nitty-gritty of SAT scoring, there are three elements you should understand:

How Scoring Really Works

The SAT doesn’t just add up points. It uses two tools together:

• Item Response Theory (IRT) → Describes the questions: their difficulty, guess-ability, and how well they separate high vs. low scorers.
• Maximum Likelihood Estimation (MLE) → Looks at your answers and figures out the score level that best matches your pattern.

Think of it like this:

1. IRT: Before you ever take the test, the SAT knows how difficult each question is and how students at different levels usually respond to it.
2. MLE: After you finish, the algorithm looks at what you got right and wrong and finds the score where that exact pattern is most likely to happen.

👉 That’s why:

• Missing an easy question can hurt one student more than another.
• Getting a hard question right can help one student more than another.
• Two students can miss the same number of questions and get very different scores.

Why Module 1 is the Gatekeeper

You don’t get the harder or easier Module 2 because of some magic cutoff score. MLE — the part of the algorithm that reads your answer pattern — decides which path you take.

It looks at how you did on each Module 1 question and figures out which version of Module 2 will give it the best chance to pin down your true score.

• If your Module 1 pattern signals a higher skill level, MLE gives you a harder Module 2 to more precisely distinguish among high performing students. In effect, this opens up the highest possible scores to high-performing test takers.
• If your Module 1 pattern signals a lower skill level, MLE gives you an easier Module 2 to more precisely distinguish among low performing students. This caps your score and prevents you from achieving scores above the low 600s.

👉 So performing well on Module 1 is essential! Think of it as stepping onto one of two escalators: one goes to the top floor, the other only halfway.

Same Question, Different Impact

What's crazy is that getting the exact same question right or wrong may impact different students differently. Consider these students:

Very High Scorer (~730)

Misses an easy question: Small/no change.

• While we might think that getting an easy question wrong could totally derail a high scorer, the MLE doesn't view an easy question as holding much information for a high scorer. So this question might actually have no impact.

Misses a medium difficulty question: Significant drop.

• Getting a medium difficulty question wrong could provide significant information for MLE. Perhaps, this high scorer isn't as high scoring as initially thought.

Gets a very hard question right: Small bump

• A high scorer is expected to get hard questions right. While good, this doesn't provide MLE with significant new information.

Above-Average Scorer (~630)

Misses an easy question: Significant drop.

• This error contradicts the pattern that MLE expects. It suggests that the test taker potentially isn't as strong as initially indicated.

Misses a medium difficulty question: Small drop.

• Medium difficulty questions don't provide as much information

Gets a very hard question right: Significant increase

• This error contradicts the pattern that MLE expects. It suggests that the test taker might be stronger than initially indicated.

Average Scorer (~500)

Misses an easy question: Significant drop.

• This error contradicts the pattern that MLE expects. It suggests that the test taker potentially isn't as strong as initially indicated.

Misses a medium difficulty question: Minor decrease.

• This error conforms to the pattern that MLE expects. An average scorer is expected to get some medium difficulty questions right and some wrong.

Gets a very hard question right: Small/no change

• The MLE doesn't view hard questions as having much information for an average score. It's always good to get questions right, but the algorithm treats this as less informative because it could be luck or an outlier, so it doesn’t shift your score much.

Is the SAT Curve Always Better in Certain Months?

You’ll see endless dropdown fights about this online. Here’s the short answer:

• No. There is absolutely no reason to assume that March will have an easier "curve" than November, for example.
• First of all, technically, the SAT uses equating, not a classroom "curve." Question difficulties are pre-determined and the score is not computed by comparing you to other test-takers that day. It is true that any one month might randomly have a slightly higher number of difficult questions, and thus allow you to miss more and still achieve a high score, but there is no predictable pattern as to which months will have harder or easier "curves."

Does Everyone Get the Same Test?

Nope, not exactly.

Even two students who both get hard Module 2 won't necessarily get identical questions. There will likely be significant overlap, but questions are pulled from a broad question pool. You should expect the overall difficulties to match up with your neighbor's experience, but the details could vary, as could the order of the questions. The College Board does this in part to make it harder to cheat off your neighbor.

Does Every Question Count?

No! There are two unscored questions in each module, for a total of EIGHT unscored questions on each test.

The SAT uses these "embedded pretesting questions" to test questions for use on future tests. How you do on these questions has no impact whatsoever on your score.

Don't get too excited, though. There is NO way to tell which questions are unscored. You must do every question as if it counts. Really? No way to tell? Nope. We know from published Bluebook tests that the unscored items likely include one easier and one harder question in each module, but that's all we know. The unscored questions will be completely indistinguishable from real ones.

Guess or Leave Blank?

From the College Board’s public explanation:

"For most students who are trying their best on every question, it's better to guess than leave a question blank, especially if a student can eliminate one or two answer options before guessing."

So that seems pretty clear. But hold on! The College Board also says this:

“In the scoring model used for the digital SAT Suite, the scores students receive are a product of several factors … [including] the probability that the pattern of answers suggests they were guessing.”

So that sounds like the College Board knows whether you're guessing and penalizes you! Well not quite.

A lot of people think the SAT is judging "guessing" by looking to see if you selected all A's in a row or answered the questions in less than 10 seconds. That is NOT what they are referring to.

They just mean whether the MLE determines a correct answer to be an outlier from the overall pattern of your responses, not that they have implemented anti-guessing measures in their scoring. In other words, if you're a very low math scorer but you got one very hard multiple-choice math question right, the SAT assumes you made a lucky guess.

Practical takeaway

• There’s still no penalty for wrong answers. Always guess rather than leave blank.
• If time is short on Math, hit student-produced response questions (SPR) first—since these are largely unguessable, and the MLE system knows it—then guess on the remaining multiple choice questions. Of course, if a you have no prayer of getting the SPR question right because it's too hard for you, don't bother! Better to get a multiple-choice question right than an SPR wrong. But if you can do either and only have time for one, do the SPR.

For The Nerds

Fair warning, from here on out, we get super technical. Don't keep reading unless you're really interested.

The digital SAT uses the 3-Parameter Logistic (3PL) IRT model. Each "operational item" (in other words, each question) has three parameters estimated during embedded pretesting:

3 Parameters

• Discrimination (a): how sharply the item separates nearby ability levels—steeper slope ⇒ more diagnostic power.
• Difficulty (b): the ability level (θ) where ~50% are expected to get the item correct.
• Guessing (c): lower-bound probability of a correct response by chance (higher for 4-option MC than for SPR).

Information functions:

Each question isn’t equally “useful” at every ability level.

An item information function tells the algorithm where on the ability scale a question gives the most precise data.

• For example, an easy question gives the most “signal” about students in the lower score range.
• A tough question gives more signal about stronger students.
• A highly discriminating item (high a) has a sharper information peak, so it helps the algorithm pinpoint ability more accurately.

Pattern scoring via MLE:

Once you finish, MLE doesn’t look at how many you got right.

It looks at the entire pattern of which questions you got right or wrong and finds the single ability level (θ) that makes your pattern most probable under the IRT model.

• A student who nails a few hard questions but misses easy ones won’t get a “bonus” just for the hard hits.
• Instead, the model finds a θ where that strange pattern is most likely to have happened — which often means a lower score than expected.
• A student whose pattern cleanly matches what a 700-level test taker usually does will end up at that θ more precisely.

Guessing

The model discounts unlikely corrects from low-θ students more than from high-θ students, which limits luck-based inflation without reinstating the old wrong-answer deduction.

What We Still Don’t Know

We know the ingredients (discrimination, difficulty, guessing, item/test information, MLE, IRT, ...). We don’t know the exact recipe:

• The College Board has not published the precise weightings, the routing thresholds between modules, or the final θ→scaled score mappings.
• Because those details are undisclosed, any third-party test (including Resolve Prep) can only approximate official scoring. We use a more traditional number-correct-to-scale approach to generate useful estimates.
• Your official score may differ—and that’s expected.

Final Student Takeaways

• Do well on Module 1 to unlock the hard module and a higher ceiling.
• Answer every question—never leave blanks.
• Prioritize SPRs if you’re short on time in Math.
• Same question, different impact: the model weighs items relative to your level.
• Focus on accuracy and pacing—you can’t game the algorithm, but you can play smart.

Sources & Further Reading

• Official explanation: College Board, How Scores Are Calculated.
• ETS / College Board research primer: Basic Concepts of Item Response Theory