Why we sample
We rarely see a whole population, such as every stock’s future return. Instead, we study a sample and use it to estimate population parameters. A value calculated from the sample (like its mean) is a statistic.
Sampling methods
Probability sampling gives every member a known chance of selection, which allows proper statistical inference.
- Simple random sampling: every member has an equal chance.
- Systematic sampling: every k-th member from a list.
- Stratified random sampling: split the population into groups (strata), such as sectors or bond ratings, and sample from each in proportion. It is common for building bond index funds because it keeps the sample’s mix close to the index’s.
- Cluster sampling: split into clusters, randomly pick whole clusters. Cheaper, but less precise.
Non-probability sampling relies on judgment or convenience. It is faster and cheaper but risks a sample that does not represent the population.
- Convenience sampling: whatever data is easiest to get.
- Judgmental sampling: an expert picks the sample.
Sampling error and the central limit theorem
Sampling error is the gap between a sample statistic and the true parameter: sample mean − population mean.
The central limit theorem is the key result. For a sample of size n drawn from any population with mean μ and variance σ²:
- The distribution of the sample mean is approximately normal when n is large (n ≥ 30 is the usual rule).
- Its mean is μ.
- Its standard deviation, the standard error, is σ / √n. When σ is unknown, use the sample standard deviation s.
Bigger samples give smaller standard errors and more precise estimates, but with diminishing returns: quadrupling the sample only halves the standard error.
Point estimates and confidence intervals
A point estimate is a single best guess, such as the sample mean. A confidence interval gives a range likely to contain the parameter:
Point estimate ± reliability factor × standard error
| Confidence level | z reliability factor |
|---|---|
| 90% | 1.645 |
| 95% | 1.96 |
| 99% | 2.576 |
Example. 36 monthly returns, mean 8%, s = 15%. Standard error = 15 / √36 = 2.5%. 95% interval = 8% ± 1.96 × 2.5% = 3.1% to 12.9%.
z or t? Use z when the population variance is known, or when the sample is large. Use t (with n − 1 degrees of freedom) when the variance is unknown, especially with small samples. The t-distribution has fatter tails than the normal, so it gives wider intervals; as the sample grows, it approaches the normal.
Biases in sampling
| Bias | What happens |
|---|---|
| Data snooping | Testing many patterns on the same data until something looks significant |
| Sample selection | Some data is excluded, for example because it is unavailable |
| Survivorship | Only survivors (funds, companies) are studied, inflating results |
| Look-ahead | The test uses information that was not available at the time |
| Time-period | Results depend on the specific period chosen and may not hold in others |
Exam focus
Expect to compute a standard error and a confidence interval, choose between z and t, apply the central limit theorem, and name the bias in a described study.