1. Background — Sampling Distribution of

Theorem — Distribution of the Sample Variance

If is the variance of a random sample of size from a normal population with variance , then

and and are independent.

Idea of the proof. (sum of squared standard normals). The identity

splits it into two independent pieces; the last is , so by MGFs , giving , i.e. . One degree of freedom is “used up” by estimating with .

Worked Example

A process has . For a sample of , find :

(Table: for , so the probability is just under 0.10.)

⚠️ This result needs a normal population; it is sensitive to non-normality (unlike results for , which the CLT protects).


2. The t-Distribution

Definition — Student’s t-Distribution

If and are independent, then

has the t-distribution with degrees of freedom, with density

Theorem — The t Statistic

For a random sample from a normal population with mean ,

Proof. Write

with , , independent by Section 1. The unknown cancels.

Properties

Propertyt-distribution
Shapebell-shaped, symmetric about 0
Mean0 (for )
Variance (for ) — heavier tails than
As
N(0,1) black vs t with v = 5 (blue) and v = 2 (amber, dashed) — t has heavier tails -4 -3 -2 -1 0 1 2 3 4

Notation — Critical Values

is the value with area to its right. By symmetry .

1.7611.7141.7111.645
2.1452.0692.0641.960

Worked Example

A supplier claims mean fiber strength . A sample of gives , (population assumed normal).

From the table, , so (exact: ). A sample mean this high would be unusual if were really 500.

⚠️ Use (not ) when is unknown and replaced by . For large the two are almost identical.


3. The F-Distribution

Definition — F-Distribution

If and are independent, then

has the F-distribution with (numerator) and (denominator) degrees of freedom. It is positive and right-skewed.

Theorem — Ratio of Two Sample Variances

For independent samples of sizes , from normal populations with variances :

Proof. independently; divide each by its degrees of freedom and take the ratio.

Theorem — Reciprocal Property

Proof. If then . So , i.e. , which says .

Worked Example — Table Lookups

⚠️ Order of degrees of freedom matters: and are different distributions. Numerator df first.

Worked Example — Comparing Variances

Two machines are believed to have equal variances. Samples , . If , then and

So an observed ratio of 3.1 would cast doubt on equal variances.

F(9, 14) density; shaded right tail beyond f₀.₀₅ = 2.65 has area 0.05 2.65 0 1 2 2.65 3 4

The F-distribution is used to compare two variances and in analysis of variance (ANOVA).


4. Quantile and Probability Plots

Definition — Quantile

A quantile of a sample is a value such that a fraction of the data is . For sorted data , is taken as with

(other conventions, e.g. , are also common).

Definition — Quantile Plot

Plot of against . It shows the shape of the data (flat stretches = clusters, steep jumps = gaps/outliers).

Definition — Normal Quantile–Quantile (Q–Q) Plot

Plot the sorted data against the standard normal quantiles .

  • If the data come from a normal distribution, the points lie close to a straight line.
  • The line’s intercept ≈ and slope ≈ .
Pattern on the normal Q–Q plotMeaning
Points on a straight lineapproximately normal
Curve bending upward at the right (convex)right-skewed
Curve bending downward at the right (concave)left-skewed
S-shape, both ends flare awayheavy tails
Isolated point far off the lineoutlier

Worked Example

Ten measurements (sorted): .

14.40.061−1.547
24.70.159−1.000
34.90.256−0.655
45.10.354−0.375
55.30.451−0.123
65.50.5490.123
75.60.6460.375
85.80.7440.655
96.00.8411.000
106.20.9391.547
Normal Q–Q plot: points close to a straight line → normal model OK x(i) normal quantile z -1.5 -0.8 0.0 0.8 1.5

The points lie close to a line with intercept () and slope (compare ), so a normal model is reasonable — which justifies using procedures on these data.

⚠️ With small , Q–Q plots wiggle even for truly normal data; look for systematic curvature, not small bumps.