4. Binomial to normal

Counting successes, then using a normal curve with a half-unit correction.

The rules

The question saysCounts includedNormal version
at most 20, 20 or fewer, X ≤ 20…, 19, 20below 20.5
fewer than 20, less than 20, X < 20…, 18, 19below 19.5
at least 20, 20 or more, X ≥ 2020, 21, …above 19.5
more than 20, X > 2021, 22, …above 20.5
exactly 2020between 19.5 and 20.5
Weight picture. If you say you weigh 190 pounds, you really weigh somewhere between 189.5 and 190.5. A count of 190 on a continuous scale is that whole strip.

Worked class example: voters

500 voters are sampled and 60% of the population favours consolidation. Set up the probability that 280 or fewer in the sample favour it, using the normal approximation.class example

Answer

X is binomial with n = 500 and p = 0.6. Check: 500 × 0.6 = 300 expected successes and 200 expected failures, both at least 5.

Mean = 300. Variance = 500 × 0.6 × 0.4 = 120, so the standard deviation is √120 ≈ 10.95.

“280 or fewer” includes 280, so use 280.5. P(X ≤ 280) ≈ P(Z < (280.5 − 300)/10.95) = P(Z < −1.78) = 0.0375. The exact binomial answer is 0.0380.

For the same voters, what z-value would you use for "fewer than 280"? For "more than 280"?

Answer

Fewer than 280 leaves 280 out, so use 279.5: z = (279.5 − 300)/10.95 = −1.87.

More than 280 leaves 280 out on the other side, so use 280.5: z = (280.5 − 300)/10.95 = −1.78, and you want the area above it.

Homework version: drugs

100 drugs are in testing and 20% are eventually licensed. Find the probability that at least 15 are licensed, with and without the continuity correction.Homework 8, Q6

Answer

X is binomial with n = 100 and p = 0.2. Expected successes 20 and failures 80, both at least 5. Mean 20, variance 100 × 0.2 × 0.8 = 16, standard deviation 4.

With the correction: “at least 15” includes 15, so use 14.5. P(Z ≥ (14.5 − 20)/4) = P(Z ≥ −1.375) = 0.916.

Without: P(Z ≥ (15 − 20)/4) = P(Z ≥ −1.25) = 0.894.

The exact answer is 0.920, so the corrected version is closer.

Twists

A batch has n = 20 parts and each is defective with probability 0.05. Can you use the normal approximation for the number of defectives?

Answer

No. The expected number of defectives is 20 × 0.05 = 1, which is below 5. The binomial here is strongly skewed, so the normal curve fits it badly. Use the exact binomial formula.

Why is a continuity correction needed at all?

Answer

The binomial puts probability on whole numbers only, while the normal spreads it over every value. Each whole number k is matched with the interval from k − 0.5 to k + 0.5. Without the half-unit shift, half of the bar for the boundary count is lost or wrongly added.

X is binomial with n = 100 and p = 0.5. Using the empirical rule, about what is the chance of 60 or more successes? Ignore the half-unit correction.

Answer

Mean = 50. Variance = 100 × 0.5 × 0.5 = 25, standard deviation 5. Then 60 is 2 standard deviations above the mean, so about 2.5%.

With the correction you would use 59.5, which gives z = 1.9 and a slightly larger answer.

X is binomial with n = 400 and p = 0.2. Give the mean and standard deviation, and state the cutoff you would use for P(X > 90).

Answer

Mean = 80. Variance = 400 × 0.2 × 0.8 = 64, so the standard deviation is 8. “More than 90” leaves 90 out, so use 90.5. z = (90.5 − 80)/8 = 1.31, and you want the area above it.