↔️ Variability

1. The big idea — center isn't enough

The mean tells you where the data piles up. Variability tells you how tightly it piles up. Picture two stats classes that both average 80 on a test. In Class A everyone scored between 78 and 82 — rock steady. In Class B half the class got 60 and half got 100 — same average, totally different vibe. The average alone can't tell them apart. Variability can. Whenever someone hands you an average, the smart follow-up question is always the same: how spread out is it?

There are three workhorse measures of spread, and they build on each other. Here's the whole unit in one table — what each captures, the units it comes out in, and how badly a single wild value (an outlier) can throw it off:

2. The range — quick but fragile

The simplest measure of spread. Subtract the smallest score from the largest: Range = max − min.

For our stress scores (column J), the highest is 38 and the lowest is 4, so the range is 38 − 4 = 34. Fast and easy — but notice what it ignores. It looks at exactly two numbers and throws away everyone in between. Change all 198 middle scores however you like and, as long as the top and bottom stay put, the range never budges. That's why one unusually high or low score — an outlier — can blow it wide open and make a tight group look wildly spread out. The range is a fine quick gut-check, but it's no foundation for serious work. For that, we want a measure that listens to every score.

3. Deviations — how far is each score from the mean?

To use all the scores, start by asking how far each one sits from the mean. That distance — score minus mean, written X − x̄ — is a deviation. A positive deviation means the score is above average; a negative one means below; and a deviation of zero means the score lands exactly on the mean. Take five simple numbers: 2, 4, 6, 8, 10. Their mean is 6. Now subtract 6 from each:

4. Variance — square the deviations, then average

The fix is to square each deviation before adding. Squaring turns every number positive (a negative times a negative is positive), so nothing cancels — and as a bonus it gives extra weight to scores that sit far from the mean, which is exactly what we want a spread measure to notice:

Add the squares: 16 + 4 + 0 + 4 + 16 = 40. That total has a name you'll see everywhere from here on: the sum of squares (often written SS) — the sum of the squared deviations. It's the raw fuel for variance, standard deviation, and nearly every test in the back half of this course. Now we average it. We have 5 numbers, but for a sample we divide by n − 1 = 4 (the reason is the next section):

5. Standard deviation — back into real units

Variance lives in squared units, so we undo the squaring: take the square root of the variance and you land back in the data's original units. Standard deviation = √10 ≈ 3.16.

So the assembly line, start to finish, is: deviations → square them → sum of squares (SS) → divide (variance) → square root (SD). Five steps, and you've turned a column of numbers into a single, readable measure of spread. Let's run that line on a fresh set of numbers, one step at a time:

6. Population vs. sample — why divide by n − 1?

You may have noticed we divided by n − 1, not n. Here's the reason, in plain terms. If you have the entire group you care about — the whole population — you divide the sum of squares by N, and the result is the population variance (σ², "sigma squared"); its square root is the population standard deviation (σ, "sigma"). But in real research we almost never have the whole population. We have a sample — a slice of a bigger group — and a sample's scores tend to cluster a little closer to their own mean than the full population's scores do to the true mean. Divide by n and you'd systematically underestimate the real spread. Dividing by n − 1 instead nudges the answer up just enough to correct that bias. You don't need to memorize the proof — just the rule.

7. Picture it — same center, different spread

This is the whole unit in one image. Both curves below are centered on the same mean — but one is tall and narrow and the other is short and wide. The mean can't tell them apart; the standard deviation is the only thing that does.

Both piles balance on the same point, so they share a mean. But the coral one spreads its scores across a much wider stretch of the number line — a larger typical distance from the center, and therefore a larger standard deviation. Tall-and-narrow is a small SD; short-and-wide is a large SD. Hold that picture; it's the intuition behind every spread number you'll ever read.

Worked example — on OUR data

Open the SJNY Student Life dataset. We want the spread of stress scores (column J). In any empty cell:

=MAX(J2:J201)-MIN(J2:J201)   → the range (38 − 4 = 34)
=VAR.S(J2:J201)              → the sample variance (≈ 34.86)
=STDEV.S(J2:J201)            → the sample standard deviation (≈ 5.9)

So stress has a mean of about 20.1 and a standard deviation of about 5.9 — meaning a typical student's stress sits roughly 6 points away from the average, in either direction. Both =VAR.S and =STDEV.S use the n − 1 (sample) version, which is what you want for our data.

Try it now — the Descriptive Statistics calculator

The same calculator from Unit 3 also reports range, variance, and standard deviation. Paste numbers or hit Load class data, and watch the spread measures appear right alongside the center. Try pasting a tight set (like 4, 5, 5, 6) and then a scattered one (like 0, 5, 5, 50) with a similar mean — and see the SD jump.

💻 In Google Sheets

You won't crunch the sum of squares by hand on exam day — Sheets does it in one cell each. Open the SJNY Student Life dataset and try these on the stress column:

=MAX(J2:J201)-MIN(J2:J201)   → the range (max − min)
=VAR.S(J2:J201)              → the sample variance (÷ n − 1)
=STDEV.S(J2:J201)            → the sample standard deviation (÷ n − 1)

The .S suffix means "sample" — these divide by n − 1, which is what you want for our data. (There are .P twins — =VAR.P and =STDEV.P — that divide by N; use those only when your numbers are the whole population.) Want to confirm the engine isn't hiding anything? Build the SD by hand in columns: put each deviation in one column with =J2-AVERAGE(J$2:J$201), square it in the next, sum that column for SS, divide by COUNT()−1, and take =SQRT(). You'll land on the same number =STDEV.S gives you — proof the shortcut is just the five-step recipe in disguise.

Key terms