How More Subjects Affect Statistical Means
What Happens to the Mean When You Add More Subjects
Short answer: the sample mean gets closer to the true population mean. That's it. That's the whole game.
But if you want to understand why this happens and what it means for your research, keep reading. Most people get this wrong, and it costs them time and bad data decisions.
The Law of Large Numbers Explained
Here's the rule: as your sample size grows, your sample mean converges toward the actual population mean. This is called the Law of Large Numbers, and it's not complicated.
Think about flipping a coin. Flip it 10 times and you might get 7 heads. That's a mean of 0.7 for heads. Flip it 10,000 times and you'll get something closer to 0.5. The extreme results even out. They're diluted by the volume.
The same principle applies to any measurement you're collecting.
Why Randomness Cancels Out
Every measurement you take has some random error attached. Individual subjects swing high or low for reasons that have nothing to do with what you're studying. Add more subjects and those random swings start canceling each other. The math is straightforward:
- Small samples amplify individual outliers
- Large samples distribute error across more observations
- The mean stabilizes because positive and negative deviations balance
Standard Error: The Real Story
Here's where people get confused. They think adding subjects always makes the mean bigger or smaller in some meaningful direction. It doesn't. What changes is the precision of your estimate.
Standard error measures how confident you can be that your sample mean reflects the true mean. The formula:
SE = σ / √n
Where σ is the standard deviation and n is your sample size. Notice that n is under a square root. This means doubling your sample doesn't cut your error in half. It cuts it by about 29%.
What This Looks Like in Practice
| Sample Size | Relative Precision | SE Reduction vs. n=10 |
|---|---|---|
| 10 | 1x (baseline) | — |
| 50 | 2.24x better | 55% smaller |
| 100 | 3.16x better | 68% smaller |
| 500 | 7.07x better | 86% smaller |
| 1000 | 10x better | 90% smaller |
You get diminishing returns fast. Going from 100 to 1000 subjects only buys you about 22% more precision. Going from 10 to 50 gets you 55%.
Central Limit Theorem: Why Normality Assumptions Work
The Central Limit Theorem says that sample means approach a normal distribution as sample size increases, regardless of the underlying population distribution. This matters because most statistical tests assume normality.
You don't need a normal population. You need enough subjects. How many? The rule of thumb is 30, but it depends on how skewed your data is. Extreme skew might need 50, 100, or more.
When Adding Subjects Changes Everything
Sometimes adding subjects doesn't just refine your estimate. It flips your conclusion.
Small samples are prone to:
- Type M errors: Magnitude errors where effect sizes look huge in small samples purely by chance
- Type S errors: Sign errors where the direction of an effect is wrong
- Publication bias amplification: Small studies with flashy results get attention, then fail to replicate
A 2016 analysis found that most published effect sizes are overestimated by at least 50% when based on small samples. Add subjects and reality sets in.
When More Subjects Won't Help
Adding subjects fixes random error. It doesn't fix systematic error. If your measurement tool is broken, your sample mean will converge to the wrong number. Faster. More confidently. Still wrong.
Common sources of systematic error that more subjects won't solve:
- Biased sampling methods
- Measurement validity issues
- Confounding variables not controlled
- Selection bias in who participates
More subjects with a flawed design just gives you a precise estimate of the wrong thing.
How to Calculate What Sample Size You Need
Power analysis tells you the minimum subjects required to detect an effect of a given size. Here's the practical process:
Step 1: Define Your Target Effect Size
What's the smallest effect you care about detecting? A 5% improvement? 10 points on a scale? This is a research decision, not a statistical one. If you don't know what effect matters, you can't calculate sample size properly.
Step 2: Set Your Error Rates
Standard conventions: α = 0.05 for Type I error, power = 0.80 for Type II error. These aren't magic numbers. They're defaults. Adjust based on your field and consequences of errors.
Step 3: Estimate Variance
You need a guess at standard deviation. Pilot data, previous studies, or domain knowledge all work. Your final sample size calculation is only as good as this estimate.
Step 4: Run the Calculation
Use G*Power, R, or any statistics software. Input your effect size, alpha, power, and variance estimate. Get your n.
Pro tip: Add 10-20% buffer for expected attrition, data cleaning exclusions, and protocol deviations.
Quick Reference: Sample Size Rules of Thumb
| Analysis Type | Minimum n | Notes |
|---|---|---|
| T-test (detecting medium effects) | 64 per group | Assumes equal groups, α=0.05, power=0.80 |
| Correlation (r = 0.3) | 85 | Minimum for reliable detection |
| Chi-square (medium effect) | 200+ | Varies heavily by expected cell counts |
| Regression (5 predictors) | 100+ | 10-20 per predictor minimum |
| ANOVA (3 groups) | 30 per group | 90 total minimum for medium effects |
These are starting points. Your specific calculation will differ based on your parameters.
The Bottom Line
More subjects = more precise estimate of the true mean. That's the mechanical relationship. What it means for your study depends entirely on whether your design, measurement, and sampling are sound.
A bad study with 500 subjects is worse than a good study with 50. The bad study gives you false confidence. The good study with 50 tells you exactly how uncertain you should be.
Fix the fundamentals first. Then add subjects.