8 topics. Each one opens with its key idea, then the detail you need to use it.
1. Ratios, rates, proportions and unit conversion
Key idea: Set up every proportion with matching units in matching positions, and let units cancel to check your setup.
A ratio compares two quantities, and a proportion says two ratios are equal. If 3 cups of flour make 20 muffins, then 50 muffins need x cups where 3/20 = x/50. Cross-multiply to get 20x = 150, so x = 7.5 cups. Keep the same quantity on top in both fractions, cups over muffins on each side, or the equation will be wrong.
A unit rate is a ratio with 1 in the denominator, so a box of cereal that costs $4.50 for 18 ounces costs 4.50 ÷ 18 = $0.25 per ounce. Unit rates are how you compare deals, speeds or densities fairly.
For conversions, multiply by fractions that equal 1, such as 5,280 feet over 1 mile. To turn 60 miles per hour into feet per second, multiply 60 mi/hr × 5,280 ft/mi × 1 hr/3,600 s. Miles cancel and hours cancel, which leaves 316,800 ÷ 3,600 = 88 feet per second. Map scales work the same way: if 1 centimeter stands for 40 kilometers, then 2.5 centimeters stands for 2.5 × 40 = 100 kilometers.
When a problem mixes several rates, such as two prices for two items, write each rate with its own label before multiplying. Then each quantity gets matched to its own price, and you cannot swap them by accident.
2. Percentages and percent change
Key idea: Turn every percent change into a multiplier, and always divide by the original value.
A percent is a ratio out of 100, so part = percent × whole. Thirty-five percent of 260 is 0.35 × 260 = 91, and to find what percent 18 is of 72, divide 18 by 72 to get 0.25, which is 25%.
Percent change measures the change against the starting value: (new − old) ÷ old. A price rising from $24 to $30 has changed by 6 ÷ 24 = 25%. If you divide by 30 instead, you get 20%, which is a classic wrong choice.
The fastest tool is the multiplier, so a 15% increase means multiply by 1.15, and a 25% decrease means multiply by 0.75. Multipliers make chained changes easy, since a price that rises 20% and then falls 20% ends at 1.20 × 0.80 = 0.96 of the original, which is a 4% net loss, not zero. Multipliers also run backward, so if a sale price after 25% off is $60, the original was 60 ÷ 0.75 = $80. Taking 25% of $60 and adding it back gives $75, which is wrong because the discount was taken from $80, not $60.
Watch the difference between percent and percentage points. If a rate climbs from 20% to 25%, it rose 5 percentage points, but it rose 5 ÷ 20 = 25% in relative terms. Read the stem to see which one it wants.
3. Mean, median, mode and weighted averages
Key idea: Convert means to totals with sum = mean × count, and weight groups by their size.
The mean is the sum divided by the count, and the most useful version is the rearranged one: sum = mean × count. If 20 students average 70, their scores add to 1,400, so when a new student scores 91, the new mean is (1,400 + 91) ÷ 21 = 1,491 ÷ 21 = 71. Working with totals solves almost every add-a-value or remove-a-value question.
The median is the middle value once the data are sorted. With an odd count it is the single middle value, but with an even count you average the two middle values, so 3, 8, 10, 15 has a median of (8 + 10) ÷ 2 = 9. The mode is the value that appears most often.
Frequency tables hide the count, so read them carefully. Suppose 4 students scored 1, 6 scored 2, 9 scored 3 and 6 scored 4. That is 25 students, so the median is the 13th value. The running totals are 4, 10 and 19, so the 13th value falls in the group that scored 3. The mean is (4 + 12 + 27 + 24) ÷ 25 = 67 ÷ 25 = 2.68.
A weighted average combines groups of different sizes. If 10 students average 80 and 30 students average 60, the combined mean is (800 + 1,800) ÷ 40 = 65. Simply averaging 80 and 60 to get 70 ignores that the second group is three times larger, so the combined mean sits closer to 60.
4. Spread, outliers and box plots
Key idea: Extreme values pull the mean, range and standard deviation; the median and IQR barely move.
Spread measures how far data stray from the center. The range is maximum minus minimum, while the interquartile range is Q3 − Q1, the width of the middle half of the data. Standard deviation measures the typical distance of values from the mean, so data bunched near the center have a small standard deviation. You will rarely compute it on the SAT, but you must compare it, and a dot plot piled near the middle has a smaller standard deviation than one with values pushed toward both ends.
Outliers show why the measures differ, as in the set 12, 14, 15, 15, 16, 18, 60. The mean is 150 ÷ 7, about 21.4, and the median is 15. Remove the 60 and the mean drops to 90 ÷ 6 = 15, while the median stays at 15. The mean, range and standard deviation all respond strongly to extreme values, but the median and IQR are resistant because they depend on position, not size.
Shifting every value changes center but not spread: add 5 points to every score and the mean and median rise by 5, but the range and standard deviation stay the same. Doubling every value doubles both center and spread.
A box plot shows the five-number summary: minimum, Q1, median, Q3 and maximum. The box spans the IQR, but a box plot does not show the mean or the number of data points, so a choice that claims it does is wrong.
5. Standard deviation units and the normal distribution
Key idea: Measure unusual values in standard deviations, and use 68-95-99.7 for bell-shaped data.
To say how unusual a value is, count how many standard deviations it sits from the mean. The formula is (value − mean) ÷ standard deviation, so if test scores have a mean of 500 and a standard deviation of 100, a score of 700 is (700 − 500) ÷ 100 = 2 standard deviations above the mean. A score of 450 is half a standard deviation below.
Many real data sets are roughly normal, which means bell-shaped and symmetric around the mean. For these, the empirical rule says that about 68% of values fall within 1 standard deviation of the mean, about 95% within 2, and about 99.7% within 3. In the test-score example, about 95% of scores fall between 300 and 700, and the leftover 5% splits evenly between the two tails because the curve is symmetric, so about 2.5% score above 700.
In a symmetric distribution the mean and median are about equal. If a distribution has a long tail to the right, such as household incomes, the few large values pull the mean above the median. A long left tail pulls the mean below the median. So if a question says the mean is much greater than the median, you can infer a right skew or some high outliers.
6. Probability, two-way tables and conditional probability
Key idea: In conditional probability, the group you are told about is the denominator.
Basic probability is favorable outcomes divided by total outcomes, assuming each is equally likely. Expected count is probability times the number of trials, so a fair die rolled 300 times should show a 5 about (1/6) × 300 = 50 times.
Most SAT probability lives in two-way tables. Suppose 200 students are surveyed, and of 120 juniors, 50 buy lunch and 70 bring it. Of 80 seniors, 30 buy and 50 bring, which makes 80 buyers in total. The probability a random student is a junior who buys is 50/200. The probability a buyer is a junior is 50/80 = 5/8. The probability a junior is a buyer is 50/120 = 5/12. The phrase after "given" or "of the" names your denominator, which is why those three answers differ.
Two events are independent when knowing one does not change the chance of the other. Here P(buys) is 80/200 = 2/5, but P(buys given junior) is 50/120 = 5/12. Those two values are close, but independence requires them to be exactly equal, so buying lunch and being a junior are not independent.
Drawing without replacement shrinks the pool, so from 5 red and 3 blue marbles, the chance of two reds is 5/8 × 4/7 = 20/56 = 5/14.
For overlapping groups, use inclusion-exclusion: count in A or B = A + B − both. If 22 of 50 club members swim, 30 run and 9 do both, then 22 + 30 − 9 = 43 do at least one, so 7 do neither.
7. Scatterplots, lines of best fit and correlation
Key idea: Interpret slope as change in y per one unit of x, and a residual as actual minus predicted.
A scatterplot shows two variables for each data point. A line of best fit summarizes the trend, and the SAT asks you to read it, use it to predict, and interpret its parts in context. Suppose a café models daily iced-drink sales as y = 6.5x − 180, where x is the high temperature in degrees Fahrenheit. The slope means each extra degree is linked to about 6.5 more drinks sold, on average. At 80°F the model predicts 6.5 × 80 − 180 = 340 drinks. If the café actually sold 365, the residual is actual minus predicted, or +25, so the point lies above the line.
The y-intercept says the model predicts −180 drinks at 0°F, which is impossible. That shows the danger of extrapolating far beyond the observed data, so interpret an intercept only when x = 0 makes sense in the context.
Correlation describes direction and strength: points rising left to right show a positive association, and points hugging the line show a strong one. Correlation is not causation. Ice-cream sales and sunburns rise together, but neither causes the other, because sunny summer weather drives both.
Know linear and exponential models apart. Linear growth adds the same amount each period. Exponential growth multiplies by the same factor, so a town growing 3% per year is modeled by P = P₀(1.03)ᵗ, and its yearly gain gets bigger over time.
8. Sampling, margin of error and study design
Key idea: Random sampling allows generalizing to the sampled population; random assignment allows claiming cause.
A sample statistic estimates a population value, so if 90 of 400 randomly chosen residents of a 12,000-person town favor a new park, the sample proportion is 90 ÷ 400 = 22.5%. Your best estimate for the whole town is 0.225 × 12,000 = 2,700 residents.
A margin of error turns that estimate into an interval, so a poll reporting 62% with a margin of error of 4 percentage points suggests the true population value is plausibly between 58% and 66%. It does not say the value is certainly in that range, and it does not describe individual people. Larger random samples give smaller margins of error, because bigger samples vary less from the truth.
The result applies only to the population sampled. A random survey of one high school's students says nothing reliable about all teenagers in the state. Bias also breaks inference, because voluntary-response samples, such as an online poll, attract people with strong opinions. Convenience samples, such as asking only people at a gym, overrepresent one group.
Cause and effect requires an experiment with random assignment. If researchers randomly assign volunteers to a new sleep schedule or the usual one, a difference in outcomes can be credited to the schedule for people like those volunteers. If they only observe who already sleeps more, any link is an association, since some other factor might drive both.