Sampling Methods and the Central Limit Theorem

Sampling Methods and the Central Limit Theorem

Why Sampling Matters

Data scientists rarely have access to every observation in a population. A city health department cannot interview every resident; a retail analyst cannot examine every transaction in real time. Instead, you draw a sample — a manageable subset — and use it to make inferences about the whole population. The validity of those inferences depends on two things: how you collected the sample and how large it is. This lesson addresses both.

Common Sampling Methods

Simple random sampling (SRS) gives every member of the population an equal chance of selection. A random-number generator picks IDs from a complete list. SRS is the gold standard for avoiding bias, but it requires a full list (sampling frame) of the population.

Systematic sampling selects every kth element after a random start. If you have 10,000 records and want 500, you pick one of the first 20 at random, then take every 20th record after that. It is efficient but can fail if the list has a hidden periodic pattern that aligns with your interval.

Stratified sampling divides the population into non-overlapping groups called strata (e.g., age bands or neighborhoods), then draws a random sample from each stratum. It guarantees representation of each group and usually produces more precise estimates than SRS for the same total sample size.

Cluster sampling divides the population into clusters (e.g., city blocks), randomly selects whole clusters, and surveys everyone inside the chosen clusters. It cuts travel or data-collection costs but increases sampling error because people within a cluster tend to be similar to each other.

Convenience sampling uses whoever is easiest to reach — social media followers, walk-in clinic patients, or volunteer respondents. It is fast and cheap but almost always introduces selection bias, meaning the sample systematically differs from the population. Convenience samples are acceptable for exploratory pilots but not for drawing population-level conclusions.

When you design a sampling plan, you must identify the target population, choose a method that fits your resources and research question, and document at least one bias risk. For a city-health survey, stratified sampling by neighborhood income level often outperforms SRS because health outcomes are strongly associated with socioeconomic factors.

The Central Limit Theorem (CLT)

The Central Limit Theorem is one of the most important results in statistics. In plain language it states:

> When you draw many independent random samples of size n from any population with a finite mean (μ) and finite variance (σ²), the distribution of those sample means will be approximately normal, centered at μ, with standard deviation σ/√n — and this approximation improves as n increases.

The term σ/√n is called the standard error of the mean. It tells you how much sample means vary from one sample to the next. A larger sample size shrinks the standard error, meaning your sample mean is a more precise estimate of the population mean.

The practical power of the CLT is that the population does not need to be normal. It can be skewed, bimodal, or even uniform. Given a large enough sample, the sampling distribution of the mean converges to a normal shape regardless. In practice, n ≥ 30 is the conventional threshold for most moderately skewed populations, though heavily skewed populations or those with extreme outliers may require larger samples.

Observing the CLT Through Simulation

The most convincing way to understand the CLT is to simulate it. The general approach:

1. Define a skewed population — for example, an exponential distribution, which has a long right tail.

2. Draw 1,000 independent samples at each of three sizes: n = 5, n = 30, and n = 100.

3. For each sample, compute the sample mean.

4. Plot the 1,000 sample means for each size as a histogram.

At n = 5 you will see a histogram that still shows skew inherited from the population. At n = 30 the histogram begins to look bell-shaped. At n = 100 it is clearly normal and noticeably narrower — the standard error is smaller. This visual demonstration is exactly what the CLT predicts.

Connecting CLT to Real Projects

Understanding the CLT guides practical decisions:

• Sample size planning: If your budget allows only 30 responses, you can still trust that your sample mean will behave approximately normally — but you should document that assumption and check for extreme skew or outliers.
• Confidence intervals and hypothesis tests: Most classical inference procedures assume the sampling distribution of the mean is normal. The CLT is why those procedures work even when the underlying data are not normal.
• Dashboard reporting (Power BI context): When aggregating metrics across subgroups of different sizes, subgroups with small counts will have larger standard errors and more variable means. Flagging low-count cells prevents misleading interpretations.

Data Ethics and Sampling

Convenience sampling and poorly designed stratification can systematically exclude vulnerable or underrepresented groups. A city-health survey that only reaches people with smartphones will miss elderly or low-income residents, skewing results and potentially misdirecting public health resources. Always ask: Who is missing from my sample, and how does that affect my conclusions?

Safety Note

This is a software-based lab. No physical hazards apply. Follow your program's acceptable-use policy for computing resources, keep your Jupyter Notebook files saved to approved storage, and never upload real patient or personally identifiable health data to a public notebook environment. Use only the synthetic scenario data provided in the handout.