Standard Error and Applications
Deepen your understanding of SE and apply sampling distributions to real-world problems
Lesson Objectives
By the end of this lesson, you will be able to:
- Clearly distinguish between standard deviation and standard error
- Calculate and interpret standard error in context
- Understand how sample size affects precision of estimates
- Apply sampling distribution concepts to real-world problems
1. What is Standard Error?
Standard Error (SE)
The standard error is the standard deviation of a sampling distribution. It measures the typical distance between a sample statistic (like x̄ or p̂) and the population parameter it estimates.
We've seen two standard error formulas:
Standard Error Formulas:
Standard error tells us how much our sample statistic typically varies from sample to sample. Smaller SE means more precise estimates!
2. Standard Error vs. Standard Deviation
Students often confuse standard deviation (SD) and standard error (SE). Let's clarify:
| Aspect | Standard Deviation (σ or s) | Standard Error (SE) |
|---|---|---|
| What it measures | Variability of individual values in a population or sample | Variability of sample statistics (like x̄) across samples |
| What it describes | How spread out the data values are | How much sample statistics vary from sample to sample |
| Affected by n? | No—population SD is fixed; sample SD depends on the data | Yes—SE decreases as n increases (SE = σ/√n) |
| Typical use | Describing data spread | Measuring precision of estimates |
| Example | Heights of adults: σ = 3 inches | Mean heights from samples of 25: SE = 3/√25 = 0.6 inches |
- SD: "Standard Deviation" = variability of Data
- SE: "Standard Error" = variability of Estimates
Example 1: SD vs. SE in Context
Adult male heights have mean μ = 70 inches and standard deviation σ = 3 inches.
(a) What does σ = 3 tell us?
Answer: Individual men's heights typically vary about 3 inches from the mean. This describes the spread of the population.
(b) If we take samples of n = 36 men, what is the SE of x̄?
Answer: SE = σ/√n = 3/√36 = 3/6 = 0.5 inches
(c) What does SE = 0.5 tell us?
Answer: Sample means (average heights of 36 men) typically vary about 0.5 inches from the true population mean. This describes the precision of our estimate.
Key insight: Individual heights vary by about 3 inches, but sample means of 36 men only vary by about 0.5 inches. Sample means are much more consistent!
3. How Sample Size Affects Standard Error
One of the most important properties of SE is its relationship with sample size:
As n increases, SE decreases. Specifically: SE is inversely proportional to √n
- Larger samples → smaller SE → more precise estimates
- To cut SE in half, multiply n by 4
- To cut SE to 1/3, multiply n by 9
Example 2: Sample Size and Precision
A population has σ = 40. Compare the standard error for different sample sizes:
| Sample Size (n) | Standard Error | Interpretation |
|---|---|---|
| 25 | 40/√25 = 8 | Sample means vary by about 8 units |
| 100 | 40/√100 = 4 | Half the SE; 4× the sample size |
| 400 | 40/√400 = 2 | Half the SE again; 4× the sample size again |
| 2500 | 40/√2500 = 0.8 | Very precise; but huge sample needed! |
Key insight: Going from n = 25 to n = 100 (4× increase) cuts SE in half. But diminishing returns: going from n = 400 to n = 2500 (6.25× increase) only cuts SE from 2 to 0.8.
4. Real-World Applications
Example 3: Quality Control in Manufacturing
A factory produces bolts with target diameter 10mm. The production process has σ = 0.2mm. Quality control takes samples of 50 bolts each hour.
(a) What is the SE of the sample mean diameter?
Solution: SE = σ/√n = 0.2/√50 = 0.2/7.07 ≈ 0.0283 mm
(b) If a sample has x̄ = 10.1mm, is this concerning?
Solution:
- z = (10.1 - 10) / 0.0283 = 0.1 / 0.0283 ≈ 3.53
- P(z > 3.53) < 0.001 (very unlikely!)
Answer: Yes, very concerning! This sample mean is more than 3.5 standard errors from target. The process may be out of control.
Example 4: Political Polling
A poll of 1000 voters finds 520 support Candidate A (p̂ = 0.52). The pollster reports "52% support with margin of error ±3%". What does this mean?
Solution:
The "margin of error" is approximately 2 × SE (for 95% confidence):
- SE = √(p(1-p)/n) ≈ √(0.52 × 0.48 / 1000) ≈ √0.0002496 ≈ 0.0158
- Margin of error ≈ 2 × 0.0158 ≈ 0.0316 ≈ 3%
Interpretation: We estimate 52% support, but due to sampling variability, the true percentage could reasonably be anywhere from 49% to 55%. If Candidate A truly had exactly 50% support, getting p̂ = 0.52 in a sample would not be surprising.
Example 5: Medical Research - Drug Effectiveness
A pharmaceutical company tests a new drug. In a trial of 400 patients, 280 showed improvement (p̂ = 0.70). The placebo effect suggests 60% improvement (p = 0.60). Is the drug effective?
Solution:
Step 1: If drug is no better than placebo (p = 0.60), what's the SE?
- SE = √(0.60 × 0.40 / 400) = √(0.24/400) = √0.0006 ≈ 0.0245
Step 2: How many SE is p̂ = 0.70 from p = 0.60?
- z = (0.70 - 0.60) / 0.0245 = 0.10 / 0.0245 ≈ 4.08
Step 3: How likely is this if drug = placebo?
- P(z > 4.08) < 0.00003 (extremely unlikely)
Conclusion: If the drug were no better than placebo, getting p̂ = 0.70 would be extremely unlikely (less than 0.003% chance). Strong evidence the drug is effective!
Example 6: Education - Standardized Test Scores
A school district wants to know if a new teaching method improves math scores. Historical data shows μ = 500, σ = 100. After implementing the new method with 225 students, x̄ = 515. Is this improvement significant?
Solution:
Step 1: Find SE
- SE = σ/√n = 100/√225 = 100/15 ≈ 6.67
Step 2: Calculate z-score
- z = (515 - 500) / 6.67 = 15 / 6.67 ≈ 2.25
Step 3: Find probability
- P(x̄ ≥ 515 | μ = 500) = P(z ≥ 2.25) ≈ 0.0122 or 1.22%
Interpretation: If the new method had no effect, there's only a 1.22% chance we'd see a sample mean this high. This suggests the new method may be effective, but we'd want more evidence before declaring success.
Check Your Understanding
Question 1: A population has σ = 20. Which gives smaller SE: a sample of 50 or a sample of 200?
Answer: Sample of 200 gives smaller SE
Calculation:
- n = 50: SE = 20/√50 ≈ 2.83
- n = 200: SE = 20/√200 ≈ 1.41
Larger sample → smaller SE → more precise estimate
Question 2: True or False: Standard error measures how spread out individual data values are.
Answer: False
Explanation: Standard deviation measures how spread out individual values are. Standard error measures how much sample statistics (like x̄) vary from sample to sample.
Question 3: If SE = 4 for n = 100, what sample size is needed to achieve SE = 2?
Answer: n = 400
Explanation: To cut SE in half, we need to multiply n by 4. Since n = 100 gives SE = 4, then n = 400 gives SE = 2.
Question 4: A quality control process samples 100 items per shift. The SE is currently 0.5. Management wants SE = 0.25. What sample size is needed?
Answer: n = 400
Explanation: Current: n = 100, SE = 0.5. Want SE = 0.25 (half). Need to multiply n by 4: new n = 4 × 100 = 400.
Question 5: Why do news polls typically sample around 1000 people instead of 10,000?
Answer: Diminishing returns and cost
Explanation: To improve SE from n=1000 to n=10,000 (10× increase), SE only improves by a factor of √10 ≈ 3.16. The massive increase in cost and effort isn't worth the modest improvement in precision. A sample of 1000 already gives SE ≈ 1.6%, which is precise enough for most purposes.
Lesson Summary
- Standard Error (SE): Measures variability of sample statistics, not individual values
- SE vs. SD: SD describes data spread; SE describes estimate precision
- Sample size effect: SE = σ/√n, so larger n gives smaller SE
- Practical rule: To cut SE in half, need 4× the sample size
- Applications: Quality control, polling, medical trials, education research
- Understanding SE is crucial for interpreting study results and margins of error