2. Sample size and response rate
How many do you need for a given margin of error?
The rules
- Mean: n = (z × s / ME)². The whole bracket is squared.
- Proportion: n = z² × p(1 − p) / ME². With no earlier estimate use p = 0.5, so p(1 − p) = 0.25.
- ME is half the width. "Within 2 minutes" means ME = 2. "A width of 0.5" means ME = 0.25. "A length of 0.1" means ME = 0.05.
- Always round up, even 270.1 becomes 271. Rounding down would miss the target.
- Response rate: people to contact = n ÷ response rate, then round up again.
- Trap: the sample size uses z, not t, because you do not know n yet and so have no degrees of freedom.
| Level | 80% | 90% | 95% | 98% | 99% |
|---|---|---|---|---|---|
| z multiplier | 1.282 | 1.645 | 1.96 | 2.326 | 2.576 |
| z² | 1.64 | 2.71 | 3.84 | 5.41 | 6.64 |
What changes n
| Change | Effect on n |
|---|---|
| Smaller margin of error | Larger. Halving the margin needs 4 times the sample. |
| Higher confidence | Larger, because z grows. |
| More spread (larger s) | Larger. |
| p further from 0.5 | Smaller, because p(1 − p) drops below 0.25. |
| Lower response rate | The required n is unchanged. The number you contact goes up. |
Worked class example: the poll
With no earlier estimate, what sample size gives a 3% margin of error at 95% confidence?class example
Answer
Use p = 0.5 because it makes p(1 − p) as large as possible, so the sample is big enough whatever the true proportion is.
n = 1.96² × 0.25 / 0.03² = 3.8416 × 0.25 / 0.0009 = 0.9604 / 0.0009 = 1067.1, so round up to 1068.
Only 60% of the people sampled complete the survey. How many must be sampled?class example
Answer
1068 / 0.60 = 1780. That is already a whole number, so sample 1780. With a 50% response rate it would be 1068 / 0.5 = 2136.
Thirty earlier polls all gave a proportion near 0.4. How does that change the calculation?review class
Answer
Use 0.4 × 0.6 = 0.24 in place of 0.25. Then n = 3.8416 × 0.24 / 0.0009 = 1024.4, so 1025. The sample is a little smaller because 0.24 is less than the worst case of 0.25.
Why is p = 0.5 the safe choice?review class
Answer
The product p(1 − p) is largest at p = 0.5, where it equals 0.25. Any other value gives a smaller product and so a smaller n. Using 0.5 guarantees the margin of error is met no matter what the true proportion turns out to be.
Worked class example: olive oil and homework time
Olive oil has s = 3 ml. What sample size gives a 95% interval with a width of 0.5 ml?class example
Answer
Width 0.5 means ME = 0.25. n = (1.96 × 3 / 0.25)² = (23.52)² = 553.2, so 554 bottles.
Weekly homework time has a standard deviation of 20 minutes. What sample size puts the sample mean within 2 minutes of the population mean with probability 0.90? With 0.95?class example
Answer
90%: n = (1.645 × 20 / 2)² = (16.45)² = 270.6, so 271.
95%: n = (1.96 × 20 / 2)² = (19.6)² = 384.2, so 385. Higher confidence needs a larger sample.
Homework versions
Steel rod diameters have σ = 0.03 cm. What sample size keeps the error within 0.005 cm with probability 0.95?Homework 8, Q2
Answer
n = (1.96 × 0.03 / 0.005)² = (11.76)² = 138.3, so 139 rods.
Clay shrinkage has an estimated standard deviation of 1.2. What sample size gives an error bound of 0.2 at 95% confidence?Homework 9, Q7
Answer
n = (1.96 × 1.2 / 0.2)² = (11.76)² = 138.3, so 139.
A company wants a 90% interval of length 0.1 for the proportion who would try a new product. What is the minimum sample size? With a 50% response rate, how many must be surveyed?Homework 9, Q8
Answer
Length 0.1 means ME = 0.05. No earlier estimate, so use 0.25. n = 1.645² × 0.25 / 0.05² = 2.706 × 0.25 / 0.0025 = 270.6, so 271.
With a 50% response rate: 271 / 0.5 = 542 people surveyed.
GRE totals have a standard deviation of about 12.85. How many friends must take the test for the group mean to be within 2 points of the true mean with probability 0.95?class example
Answer
n = (1.96 × 12.854 / 2)² = (12.597)² = 158.7, so 159 people in total. You are one of them, so you need 158 friends.
Twists
A study needs n = 400 for a margin of error of 4. The client now wants a margin of 2 at the same confidence. What is the new n?
Answer
Halving the margin multiplies n by 4, because the margin is squared in the denominator. The new n is 1600.
The response rate drops from 60% to 30%. What happens to the required sample size and to the number of people contacted?
Answer
The required number of completed responses does not change. The number of people contacted doubles, because you divide by 0.30 in place of 0.60.
Your calculation gives n = 96.04. A classmate says to use 96. Who is right?
Answer
Use 97. A sample of 96 gives a margin of error slightly larger than the target. Sample sizes are always rounded up.