5. Sums, differences, averages and the CLT

All of Homework 8 and the first set of class examples.

The rules

Total or average? Read which one the question asks about. The total gets more spread out as n grows. The average gets less spread out.

Worked class examples

A door casing width is normal with mean 24 in and standard deviation 0.125 in. A door width is normal with mean 23.875 in and standard deviation 0.0625 in. They are independent. Describe the difference, casing minus door.class example

Answer

The difference D is normal. Mean = 24 − 23.875 = 0.125 in. Variance = 0.125² + 0.0625² = 0.015625 + 0.003906 = 0.01953, so the standard deviation is about 0.14 in. The variances add even though the widths are subtracted.

For the door, what event means "the door does not fit", and what is its probability?class example

Answer

The door does not fit when the casing is narrower than the door, so D < 0. z = (0 − 0.125)/0.14 = −0.89, and P(Z < −0.89) ≈ 0.19.

“The gap exceeds 0.25 in” is D > 0.25, with z = (0.25 − 0.125)/0.14 = 0.89. By symmetry it has the same probability, about 0.19.

25 people squeeze into an elevator. Weights are independent and normal with mean 160 and standard deviation 30. What limit is exceeded with probability 0.001?class example

Answer

The total is normal with mean 4000 and standard deviation √(25 × 900) = 150. The z-value with 0.001 above it is 3.09. Limit = 4000 + 3.09 × 150 = 4463.5 pounds.

Homework 8

Three flashlight batteries have independent normal lives with means 6, 7, 8 hours and variances 2, 3, 4. Describe the total life and find its 95th percentile.Homework 8, Q1A

Answer

T is normal with mean 6 + 7 + 8 = 21 and variance 2 + 3 + 4 = 9, so the standard deviation is 3. The 95th percentile is 21 + 1.645 × 3 = 25.9 hours.

For the flashlights, set up P(total < 25). Then set up the probability that the batteries last more than 25 hours on exactly 3 of 5 trips.Homework 8, Q1B and C

Answer

P(T < 25) = P(Z < (25 − 21)/3) = P(Z < 1.33) = 0.908.

So p = P(T > 25) = 0.092. The number of trips out of 5 is binomial with n = 5 and p = 0.092. P(exactly 3) = C(5,3) × 0.092³ × 0.908² = 10 × 0.092³ × 0.908² ≈ 0.0064.

A bus wait is uniform on (0, 10) minutes, so its mean is 5 and variance is 100/12. Someone catches the bus 120 times. Describe the total wait and its 95th percentile.Homework 8, Q3

Answer

The waits are not normal, but there are 120 of them, which is at least 30. By the Central Limit Theorem the total is approximately normal with mean 120 × 5 = 600 and variance 120 × 100/12 = 1000. The 95th percentile is 600 + 1.645 × √1000 ≈ 652 minutes.

A lens coating has 36 independent layers, each with mean 0.5 microns and standard deviation 0.2 microns. What is the distribution of the total thickness, and which theorem justifies it? What fraction of lenses is under 16 microns?Homework 8, Q4

Answer

By the Central Limit Theorem, with 36 layers (at least 30), the total is approximately normal with mean 36 × 0.5 = 18 microns and standard deviation √36 × 0.2 = 1.2 microns.

P(T < 16) = P(Z < (16 − 18)/1.2) = P(Z < −1.67) = 0.0475, so about 4.8% need extra layers.

Senior engineer salaries have mean 196 (thousand dollars) and variance 400. Describe the sample mean of 40 engineers.Homework 8, Q5

Answer

With n = 40, at least 30, the Central Limit Theorem says the sample mean is approximately normal. Its mean is 196 and its variance is 400/40 = 10.

The GRE class example

Verbal: mean 150.61, SD 8.57. Quantitative: mean 154.34, SD 9.58. Writing: mean 3.60, SD 0.086. Treat the sections as independent and normal.

Describe the distribution of the total score.class example

Answer

Normal, because it is a sum of independent normals. Mean = 150.61 + 154.34 + 3.60 = 308.55. Variance = 8.57² + 9.58² + 0.086² = 165.23, so the standard deviation is 12.85.

What total is the 93rd percentile? What is P(total < 320)?class example

Answer

The z-value with 0.93 to its left is about 1.48. Score = 308.55 + 1.48 × 12.85 ≈ 327.5.

P(T < 320) = P(Z < (320 − 308.55)/12.85) = P(Z < 0.89) = 0.8133.

You and five friends take the test. Set up the probability that exactly two score at least 320.class example

Answer

One person scores at least 320 with probability p = 1 − 0.8133 = 0.1867. The number out of 6 is binomial with n = 6 and p = 0.1867. P(exactly 2) = C(6,2) × 0.1867² × 0.8133⁴ = 15 × 0.1867² × 0.8133⁴ ≈ 0.229.

Describe the sample mean score of the six of you, and set up P(mean < 320).class example

Answer

The sample mean is normal with mean 308.55 and variance 165.23/6 = 27.54, so its standard deviation is 5.25. P(mean < 320) = P(Z < (320 − 308.55)/5.25) = P(Z < 2.18) = 0.985.

Compare with one person: 0.81. The average of six is much less spread out, so it is far more likely to be under 320.

The company says the Quantitative and Verbal scores have correlation 0.35. Describe the sum of just those two sections.class example

Answer

Normal with mean 150.61 + 154.34 = 304.95. Variance = 8.57² + 9.58² + 2 × 0.35 × 8.57 × 9.58 = 73.44 + 91.78 + 57.47 = 222.69. The positive correlation adds the extra term, so the sum is more spread out than it would be under independence (165.22).

Twists

Lives of 3 batteries are independent normals. Why is the total exactly normal even though n = 3 is small?

Answer

A sum of independent normal variables is itself normal for any number of terms. The Central Limit Theorem and the n ≥ 30 rule are only needed when the individual pieces are not normal.

A student computes the standard deviation of a total of three parts as 2 + 3 + 4 = 9. What is wrong?

Answer

Standard deviations do not add. Variances add. If the standard deviations are 2, 3 and 4, the variance of the total is 4 + 9 + 16 = 29 and the standard deviation is √29 ≈ 5.4.

Individual values have mean 50 and standard deviation 12. Compare the standard deviation of the total of 36 values with that of the average of 36 values.

Answer

Total: 12 × √36 = 72. Average: 12/√36 = 2. Adding values makes the total more variable. Averaging makes the result less variable.

Wait times are strongly right-skewed. A sample of 10 gives a mean. Can you treat that mean as normal?

Answer

No. The individual times are not normal and n = 10 is below 30, so the Central Limit Theorem does not apply. With a sample of 30 or more it would be approximately normal.