About 5 to 7 of the 44 Math questions, and disproportionately where careless losses gather in the 600–700 band. Not because the maths is hard — most of it is arithmetic — but because nearly every question in this domain has a wrong answer waiting for a specific, predictable slip. Learn the slips and this becomes the easiest domain on the test.
The percent-change trap
The most reliably exploited error in SAT Math.
Percent change is always measured against the original value. Not the new one.
Going from 40 to 50: the change is 10, the original is 40, so 10/40 = 25% increase.
Going from 50 to 40: the change is 10, the original is 50, so 10/50 = 20% decrease.
Same ten units, different percentages, because the base changed. And the SAT will offer you 20% in the first case and 25% in the second — the answer produced by dividing by the wrong number is on the screen, every time.
Two consequences that get tested directly:
- An increase then an equal decrease does not return you to the start. Up 20% then down 20%: 100 → 120 → 96. The second percentage acts on a bigger base.
- Successive percentages multiply, they don't add. Up 10% then up 20% is ×1.1 × 1.2 = ×1.32, a 32% increase — not 30%.
| Phrase | Means | Not |
|---|---|---|
| increased by 25% | × 1.25 | × 0.25 |
| decreased by 25% | × 0.75 | × 0.25 |
| is 25% of | × 0.25 | × 1.25 |
| is 25% more than | × 1.25 | × 0.25 |
| is 125% of | × 1.25 | × 2.25 |
| after a 25% discount | × 0.75 | × 0.25 |
Rows three and four are one word apart and mean opposite things. That word is the question.
Before computing anything, write down what the base is. Literally: "base = original price." Two seconds, and it prevents the single most common careless loss in this domain. If you're on the execution track described in 1100 to 1300 and 1200 to 1400, this one habit is worth more than a week of new content.
Rates and unit conversion
The recurring difficulty isn't the arithmetic — it's deciding whether to multiply or divide. There's a mechanical fix.
Write the units as fractions and cancel them. If you want kilometres per hour from metres per second, multiply by (1 km / 1000 m) and (3600 s / 1 hr). The metres and seconds cancel; whatever's left is right by construction.
This feels slower than reasoning it out. It's faster, because reasoning it out is where the coin-flip happens, and a coin-flip on a two-step conversion is a 25% chance of being right.
Watch for the reversed unit too — questions asking for "hours per kilometre" when you've computed kilometres per hour. That's a misread, and misreads are the second-largest cause of loss at this level.
Ratios and proportions
Set up the proportion with matching units in matching positions, then cross-multiply. The main trap is a part-to-part versus part-to-whole confusion.
If a mixture is 3 parts water to 2 parts syrup, then water is 3/5 of the total, not 3/2. Both 3/2 and 3/5 will be available as answers.
When a ratio question gives a total, convert to parts immediately: 3:2 with a total of 40 means each part is 8, so 24 and 16. Doing that conversion first turns most ratio questions into one line.
Statistics — concepts, not computation
The SAT tests statistical understanding. You will not be asked to compute a standard deviation.
| Measure | What to know | The tested idea |
|---|---|---|
| Mean | Sum ÷ count | Sensitive to outliers. One extreme value drags it. |
| Median | Middle value when ordered | Resistant to outliers. Doesn't budge. |
| Mode | Most frequent | Rarely the point of a question. |
| Range | Max − min | Entirely determined by the two extremes. |
| Standard deviation | Spread only | Tighter cluster = smaller SD. Never computed. |
The favourite question here: add an extreme value to a data set — what happens to the mean and the median? The mean moves toward the outlier; the median usually barely moves. If a question asks which measure better represents a skewed set, the answer is the median, and the reason is exactly this.
For standard deviation, questions give you two data sets and ask which has the larger spread. Look at the clustering, don't calculate. Any working that heads toward a formula means you've misread the question.
Reading the graph they gave you
Scatterplots and line of best fit. Two things get asked. What does the line predict at a given x, and how does the prediction compare to an actual plotted point? A point above the line means the actual exceeded the prediction. That gap is the residual, and questions about "how much more than predicted" are asking for it.
Also read the slope in context. In a graph of cost against units, the slope isn't a number — it's dollars per unit, and questions ask what it represents in the situation. The y-intercept is the value when x is zero, which usually has a concrete meaning like a fixed fee.
Before anything else, read the axis labels and units. A graph in thousands, or starting at 20 rather than 0, is the setup for a wrong answer that's off by a factor you'd never otherwise make.
Two-way tables: the denominator is the question
These look easy and are quietly precise. Almost every miss is a denominator error.
- "What fraction of all participants…" → denominator is the grand total.
- "What fraction of those who exercised…" → denominator is that row's total.
- "Of participants who reported improvement, what fraction…" → denominator is that column's total.
Underline the group being asked about before you touch a number. The wrong answers are the same numerator over the two other plausible denominators, so getting this right is the entire question.
Sampling and study design — two words decide it
The most learnable concept in this domain, and the one most students improvise on.
| The study had… | You may conclude |
|---|---|
| Random selection from a population | Results generalise to that population |
| Random assignment to groups | Cause and effect |
| Both | A causal claim, generalised to the population |
| Neither | Neither. Association only. |
Read the description hunting for those two words. If subjects volunteered, you can't generalise. If groups formed themselves rather than being assigned, you can't claim causation regardless of how strong the association is.
Margin of error follows one rule worth knowing: a larger sample narrows the interval. Questions asking how to reduce the margin of error are asking whether you know to increase the sample size.
Where Desmos earns its place
- Enter a data list and Desmos gives you mean and median directly — faster and safer than hand arithmetic on eight values.
- Plot points to sanity-check a line of best fit.
- Any multi-step percentage chain, to avoid a keying error mid-calculation.
What it can't do is choose the base for you, or pick the denominator. Those are reading decisions, and they're where the points actually go. Our Desmos guide covers the fast routes.
A three-day plan
- Day 1 — percent and rates. Twenty questions, writing down the base before each one. This alone tends to fix the majority of this domain's losses.
- Day 2 — graphs and tables. Twenty questions, underlining the group asked about before computing anything.
- Day 3 — statistics and study design. The selection-versus-assignment table, plus mean-versus-median under outliers.
Then re-test in a fortnight. Accuracy the next day proves nothing; accuracy after two weeks means it's actually fixed. The spacing rule is here.
Frequently asked questions
How do you calculate percent change on the SAT?
Change divided by the original value. The answer you'd get from dividing by the new value is always one of the choices.
Do you need to calculate standard deviation on the SAT?
No. It's conceptual only — tighter clustering means smaller spread. You'll never compute one.
When can a study conclude cause and effect on the SAT?
Only with random assignment. Random selection lets you generalise; it doesn't license causation.
How many data analysis questions are on the SAT?
Roughly 5 to 7 of 44 — about 15% of Math, the same share as Geometry and Trigonometry.