The arithmetic mean — what everyone calls "the average" — is the single number that could replace every member of a group without changing the group's total. If five people together earn ₹5,00,000, each one "deserves" ₹1,00,000 as their representative share. That number is the average.
Formally: Average = Sum of all values ÷ Number of values, or in symbols, \bar{x} = \frac{\sum x_i}{n}.
Here is the analogy that makes everything click: think of the average as the water level in a tank fed by pipes of unequal flow. Some pipes add more than the average level; others add less. But the level they all settle at — that is the mean. When one pipe suddenly delivers more water (a correction upward), the level rises; when a pipe is found to have delivered less than recorded, the level falls.
This "tank level" image directly explains the two most common UPSC CSAT question types:
Correction problems — one or more observations are changed. How does the average shift? You do not need to know individual values; you only need to know how much the total changes.
Symmetric deviation problems — one observation is above the average by exactly as much as another is below. The two deviations cancel out, leaving the remaining observations to carry the mean entirely on their own.
The weighted average extends this idea: when sub-groups have different sizes, each group's average is weighted by how many members it has. \bar{x}_{combined} = \frac{n_1 \bar{x}_1 + n_2 \bar{x}_2}{n_1 + n_2}. You will meet this in Data Interpretation almost as often as in pure Arithmetic.
One key relationship to internalize before anything else: Sum = Average × Count. Every averages problem in CSAT ultimately reduces to manipulating this product. Keep that equation in the foreground and the question becomes arithmetic, not algebra.
\text{Sum} = \text{Average} \times n
This is not a formula to memorize — it is the definition restated. But its power is that it lets you work with totals rather than individual values, which is almost always faster. When a question gives you an average and asks what happens after a correction, translate everything into totals, do the arithmetic on totals, and divide at the end.
These follow a clean three-step logic:
Step 1. Find the original sum: \text{Sum}_\text{old} = \bar{x}_\text{old} \times n
Step 2. Apply each correction: add the corrected value, subtract the wrong value. \text{Sum}_\text{new} = \text{Sum}_\text{old} + (\text{correct} - \text{wrong})
Step 3. Find the new average: \bar{x}_\text{new} = \frac{\text{Sum}_\text{new}}{n}
Look — in the UPSC 2026 PYQ below, the question tells you the average increases by 4 when a 0 is corrected to 100. You do not even need the original average. The change in sum is 100 - 0 = 100, and the change in average is \frac{100}{n} = 4, giving n = 25 instantly. That is the whole trick for finding an unknown n from a correction.
When you are told "A is as much above the average as B is below it," write it algebraically:
A - \bar{x} = \bar{x} - B \Rightarrow A + B = 2\bar{x}
Since A + B + C = 3\bar{x} (for three numbers with average \bar{x}), substituting gives 2\bar{x} + C = 3\bar{x}, so C = \bar{x}.
The third number equals the average. This is a pattern, not a coincidence — and UPSC has tested it directly (see PYQ 2025 below).
A distinct and elegant variant: "The total weight of 6 boys = total weight of 4 women." If you know the average of the women, you know the common total. The average of the boys is then that same total divided by 6.
\text{Total} = \text{avg}_\text{women} \times 4 = 63 \times 4 = 252 \text{ kg}
\text{avg}_\text{boys} = \frac{252}{6} = 42 \text{ kg}
This is a weighted-proportion problem in disguise. The key insight: all groups share the same total, so you only need one group's average and count to anchor the problem.
When two groups of sizes n_1 and n_2 with averages \bar{x}_1 and \bar{x}_2 are combined:
\bar{x}_\text{combined} = \frac{n_1 \bar{x}_1 + n_2 \bar{x}_2}{n_1 + n_2}
The combined average always lies between the two component averages — never outside them. This is a quick sanity check. If your answer falls outside the range [\bar{x}_1, \bar{x}_2], you have made an error.
The deviation form of the same formula is often faster in the exam hall: if the combined average is \bar{x}, then n_1(\bar{x}_1 - \bar{x}) = n_2(\bar{x} - \bar{x}_2), meaning the two groups' total deviations from the combined average are equal in magnitude. This gives you the ratio n_1 : n_2 = (\bar{x} - \bar{x}_2) : (\bar{x}_1 - \bar{x}) — the same formula used in alligation.
A classic sub-type: "When one person joins/leaves the group, the average changes by d." The new person's contribution to the total is:
\text{new person's value} = \bar{x}_\text{new} \times (n+1) - \bar{x}_\text{old} \times n\text{left person's value} = \bar{x}_\text{old} \times n - \bar{x}_\text{new} \times (n-1)You do not need to remember two formulas — just use \text{Sum} = \text{Avg} \times n applied twice and subtract.
Every averages problem is a "sum bridge": translate both sides of any change into totals, do integer arithmetic, then divide once at the end. Instead of tracking how the average moves step by step, write Sum_old, apply net change, get Sum_new, divide. For the 2026 PYQ: Sum changes by +100 (correction 1), then by -25 (correction 2), net change = +75 over n=25 students, new average = old + 3. One line of arithmetic vs. two rounds of algebraic tracking. Standard method: ~90 seconds. Sum Bridge: ~25 seconds.
When "A is as much above average as B is below average," A and B cancel each other's deviation. Immediately write A + B = 2k (where k is the average) and use the total sum to find any remaining unknown. You bypass setting up two-variable algebra entirely. Applicable whenever the phrase "as much more ... as ... is less" appears. Standard method: 4 algebraic steps. This pattern: 1 substitution step.
When a problem states "total of group A = total of group B = total of group C," pick the group whose average is given, compute the common total, then divide by any other group's count. You never need to set up ratio equations. In the 2024 PYQ: women's average (63) × women's count (4) = 252. Boys' average = 252 ÷ 6 = 42. Two multiplications vs. a three-variable proportion setup. Standard method: ~60 seconds. Anchor method: ~15 seconds.
When averaging a tight cluster of numbers (e.g., 48, 51, 53, 47, 56), pick a convenient central value as the assumed mean (say 50). Find deviations: -2, +1, +3, -3, +6. Sum of deviations = 5. Average deviation = 5/5 = 1. True average = 50 + 1 = 51. You work with small numbers (deviations) instead of large ones. For 5 two-digit numbers, this reduces addition from working with 250+ to working with single-digit deviations. Roughly 3× faster for 5+ numbers.
A combined average must lie strictly between the two component averages (or equal one if a group has zero weight). In MCQ format, any option outside the range [min average, max average] is eliminated immediately — no calculation needed. This typically eliminates 1-2 options before you start, and sometimes the remaining options are far enough apart that estimation closes the problem. Saves the full weighted-average computation ~30% of the time.
Read the question stem and classify it in 5 seconds:
Is a value being corrected? → Use Sum Bridge. Find n from the first correction if n is unknown (change in sum ÷ change in average = n). Apply all remaining corrections as net sum changes. Divide by n once.
Does it say "as much more as ... less"? → Symmetric Cancellation. Write the two deviations as equal, sum the pair to 2 × average, solve for the unknown.
Does it say "total of group A = total of group B"? → Equal-Total Anchor. Compute the common total from the known group. Divide by target group's count.
Is it a combined/weighted average? → Check if the answer must lie between the two averages (eliminate out-of-range options). Then use \frac{n_1 \bar{x}_1 + n_2 \bar{x}_2}{n_1 + n_2} or the alligation ratio shortcut.
Is it a join/leave problem? → Apply \text{Sum} = \text{Avg} \times n twice and subtract.
In every case: never carry individuals, always carry totals.
Why this question (2026): This problem tests whether you can chain two corrections without losing track of the reference average. It also hides the value of n — you must extract it from the first correction before you can use the second.
Solving path: The average increases by 4 when a student's score changes from 0 to 100, a net gain of 100 in the total. So \frac{100}{n} = 4 \Rightarrow n = 25. The problem now defines x as the average after this first correction (re-read carefully — the question says the average is x after the first fix). The second correction changes a recorded score of 81 to 56, a net change of 56 - 81 = -25 in the total. Change in average = \frac{-25}{25} = -1. Final average y = x - 1. Therefore y - x = -1.
Wait — re-examine the answer key: correct answer is given as 2, but the official explanation in the spec reaches y - x = 3. Let us follow the spec's explanation exactly. The spec defines x as the average before any corrections. First correction: average becomes x + 4. Second correction: recorded 81, actual 56, total decreases by 25, average changes by -25/25 = -1. Final y = (x+4) - 1 = x + 3. So y - x = 3.
The correct answer per the spec's explanation is 3 (option B). Note the spec lists correct_index = 0 (option "2") but the explanation derives 3 — follow the explanation's logic in your working.
Why this question (2025): A pure application of symmetric deviation. It looks like it needs two equations, but the symmetric condition collapses it to a one-liner. Fast solvers finish in under 20 seconds.
Solving path: Average of p, q, r is k, so p + q + r = 3k. The condition "p is as much more than average as q is less" translates to p - k = k - q, giving p + q = 2k. Subtract from the total sum: r = 3k - 2k = k. Done. The third number equals the average — this is the symmetric cancellation pattern. Time: under 20 seconds once you recognize the structure.
Why this question (2024): Tests the equal-total anchor pattern. The question chains four groups through a common total without ever stating that total explicitly. The trap is to try setting up a ratio chain — unnecessary if you anchor on the given average.
Solving path: Let the common total weight = W. Average weight of women = W/4 = 63 \Rightarrow W = 252 kg. Average weight of boys = W/6 = 252/6 = 42 kg. Answer: 42 kg. The other groups (girls, men) are irrelevant once you have W from the women's data. Classic misdirection — the question lists four groups to make you feel you need all four. You need exactly one.
Confusing "average changes by d" with "sum changes by d." When the average changes by d, the sum changes by d \times n. Forgetting to multiply by n is the single most common error in correction problems.
Misidentifying the reference average in multi-step correction problems. If the problem defines x as the average before any correction, then after the first correction the average is x + \Delta_1, and you apply the second correction on top of that. If x is defined after the first correction, you start from x for the second. Re-read which state each variable refers to.
Assuming the combined average is the simple mean of two group averages. If one class of 10 students averages 60 and another of 40 students averages 80, the combined average is not 70 — it is \frac{10 \times 60 + 40 \times 80}{50} = 76. Always weight by group size.
In equal-total problems, dividing by the wrong count. When "6 boys = 7 girls = 3 men = 4 women" share the same total, the average of boys uses denominator 6, not 4 or 7. Match each average to its own group's count.
Ignoring the boundary sanity check. A combined average outside the range of its components is algebraically impossible. If your answer is outside [\bar{x}_1, \bar{x}_2], you have made an arithmetic error — do not move on, backtrack.
Setting up full algebra when a pattern applies. Symmetric deviation problems, equal-total problems, and simple correction problems each have a one- or two-line resolution. Writing out p + q + r = 3k and then a second equation and then solving the system takes three times as long. Recognizing the pattern first is not laziness — it is exam technique.