Statistics › Section G · The central limit theorem › Topic G.2
The central limit theorem for sums
Will the elevator overload? Can this month's receipts cover the costs? Some questions are about totals, not averages - and a sample sum is as random as a sample mean. The central limit theorem governs both.
Totals are random too
Capacity and budget questions are about totals, not averages: the sum of 16 passengers' weights, the sum of 25 days' receipts. A sample sum \(\sum X\) is as random as a sample mean - take two samples and their totals differ - and the central limit theorem governs both: for large \(n\), sums are approximately normal, with a centre and a spread that scale predictably with the sample size. Where the previous topic asked "how far will the average wander?", this one asks "how far will the total wander?" - and the answer needs only two rescalings.
The sum's centre and spread
For \(n\) independent observations from a population with mean \(\mu\) and SD \(\sigma\), the sum \(\sum X\) has mean \(n\mu\) - \(n\) copies of the centre added up - and standard deviation \(\sigma\sqrt{n}\), because independent variances add: \(n\) copies of \(\sigma^2\) total \(n\sigma^2\), whose square root is \(\sigma\sqrt{n}\). Note the direction: averages shrink as \(n\) grows (\(\sigma/\sqrt{n}\)), but sums grow (\(\sigma\sqrt{n}\)). More items mean more total - and more total spread.
The z-formula for sums
With a centre of \(n\mu\) and a spread of \(\sigma\sqrt{n}\), sums standardise like everything else:
Once converted, the standard-normal table serves as before: left-areas, complements, between-questions. And the sum machinery is the average machinery in a different jacket - since \(\sum X = n\bar{x}\), knowing one determines the other, and the CLT's normal approximation covers both.
Where sums matter
Capacity and budget questions are the natural home of sums. An elevator rated for a weight limit: what is the chance 16 randomly chosen passengers exceed it? A payroll due at month's end: what is the chance 25 business days fall short? In each, the risk lives in the total, and the total's distribution is exactly what the CLT for sums supplies - the same bell, centred and scaled to the question's units. Inventory questions join the list: a shop that sells items one at a time still orders in bulk, and the month's total demand is what decides whether shelves empty.
Worked example: the elevator
In an illustrative model, passenger weights have \(\mu = 70\) kg and \(\sigma = 12\) kg, and an elevator carries \(n = 16\) riders chosen independently. With \(n\) below 30, assume the weights are roughly normal so the sum is close to normal. 1. Total's mean: \(16 \times 70 = 1120\) kg. 2. Total's SD: \(12 \times \sqrt{16} = 12 \times 4 = 48\) kg. 3. \(z = \frac{1200-1120}{48} = 1.67\) to two decimals, so
4. Check: 1200 kg sits 80 kg over the centre - 1.67 SDs - and roughly 5% of a normal lies beyond that, so about a one-in-twenty overload risk for a full car.
Worked example: a month of receipts
A food cart takes in a mean of $1,000 per day with an SD of $200, independently across days, and must cover $23,000 of fixed costs over 25 selling days. Assume daily takings are not strongly skewed, so 25 days is enough for the CLT. 1. Total's mean: \(25 \times 1000 = 25{,}000\) dollars. 2. Total's SD: \(200 \times \sqrt{25} = 1{,}000\) dollars. 3. Standardise the shortfall:
4. Check: $23,000 lies 2 SDs below the total's centre, and only about 2.3% of a normal sits that far down - a rare short month.
Landmarks for sums
The 68-95-99.7 rule applies to totals as well as data: about 95% of sample sums fall within two \(\sigma\sqrt{n}\)'s of \(n\mu\). For the elevator, 2 SDs is \(2 \times 48 = 96\) kg, so 95% of full cars land between 1024 and 1216 kg - Only about 2.5% of cars exceed 1216 kg and about 16% exceed one SD above the centre, 1168 kg, so the chance of passing 1200 kg must lie between 2.5% and 16%: the table's 4.75% fits. Landmarks like these let sanity checks run before any table lookup.
Sums and averages, one theory
Watch the two z-formulas side by side: averages use \(\frac{\bar{x} - \mu}{\sigma/\sqrt{n}}\); sums use \(\frac{\sum x - n\mu}{\sigma\sqrt{n}}\). Same \(\mu\), same \(\sigma\), same \(n\) - only the scaling differs, because \(\sum X = n\bar{x}\). A statement about a sum converts into one about an average by dividing by \(n\), and back again by multiplying. The CLT guarantees both are approximately normal for large \(n\); choose the statistic that matches the question's units.
Common mistakes to avoid
First, using \(n\sigma\) for the sum's SD: independent variances add, so the SD grows by \(\sqrt{n}\), not \(n\). Second, using \(\sigma/\sqrt{n}\) - the average's standard error - for a sum: that is the shrinking statistic, and the sum is the growing one. Third, forgetting the mean scales as \(n\mu\), not \(\mu\). Fourth, assuming independence when days or passengers influence each other. Fifth, mixing units between the total and its z-conversion. Sixth, relying on the CLT with tiny \(n\) on very skewed data - the previous topic's caution carries over unchanged.
Key takeaways
Sample sums are approximately normal for large \(n\), by the same central limit theorem that covers averages, with mean \(n\mu\) and SD \(\sigma\sqrt{n}\) (independence required). Standardise with \(z = \frac{\sum x - n\mu}{\sigma\sqrt{n}}\) and read the standard-normal table as usual. The elevator: centre 1120, SD 48, overload risk 0.0475. The month: centre 25,000, SD 1,000, shortfall risk 0.0228. Averages shrink with \(n\); sums grow - opposite scalings, one theory.
Flashcards
Flashcard
ΣX varies from sample to sample
Test your understanding
Question 1 of 5
Passenger weights have μ = 70 kg, σ = 12 kg. For a car of n = 16, what are the total's mean and SD?
Question 2 of 5
For the elevator total (mean 1120 kg, SD 48 kg), what z-score corresponds to 1200 kg?
Question 3 of 5
For the elevator, what is P(total weight > 1200 kg)?
Question 4 of 5
For 25 days of receipts (total mean 25,000 dollars, SD 1,000), what is P(total < 23,000)?
Question 5 of 5
For n independent observations with SD σ, the standard deviation of the sum is:
