7.5 Central Limit Theorem (Cookie Recipes)

Central Limit Theorem and Cookie Recipes

The Central Limit Theorem (CLT) explains why averages behave more predictably than individual measurements. Even when individual data points follow a weird, skewed, or non-normal distribution, the distribution of sample means will approach a normal distribution as sample size grows. This unit uses cookie recipes as the running example, but the theorem applies universally.

more resources to help you study

practice questions

Central Limit Theorem in Recipe Analysis

The CLT makes three core claims:

1. The shape becomes normal. The sampling distribution of the sample mean approaches a normal distribution as sample size increases, regardless of the population's original shape. Cookie weights from a recipe might be skewed or bimodal, but if you repeatedly pull batches of cookies and calculate each batch's mean weight, those means will form a bell curve as batch size grows.

2. The center stays the same. The mean of the sampling distribution equals the population mean:

$\mu_{\bar{x}} = \mu$

If the true average weight of a cookie from a recipe is 50 g, then the mean of all possible batch means is also 50 g. Sampling doesn't introduce bias into the center.

3. The spread shrinks. The standard deviation of the sampling distribution (called the standard error) is:

$\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}}$

This formula captures an inverse relationship between sample size and variability. As batch size $n$ increases, the standard error decreases, meaning batch means cluster more tightly around the population mean. Specifically, doubling the sample size reduces the standard error by a factor of $\sqrt{2}$, not by half.

Once you know $\mu_{\bar{x}}$ and $\sigma_{\bar{x}}$, you can compute a z-score to standardize any batch mean and compare across different batch sizes:

$z = \frac{\bar{x} - \mu}{\sigma / \sqrt{n}}$

Sampling Distributions for Cookie Batches

Smaller batches = more variability. With batches of only 10 cookies, each individual cookie has a large influence on the batch mean. One unusually heavy cookie can pull the average noticeably. The resulting distribution of batch means is wider and more spread out.

Larger batches = tighter clustering. With batches of 100 cookies, extreme values get diluted. The distribution of batch means has a much smaller standard deviation, and those means stay closer to the true population mean. This connects to the Law of Large Numbers: as the number of independent observations grows, the sample average converges toward the expected value.

The shape improves too. Even if individual cookie weights from a recipe are skewed (say, a long right tail from occasional oversized cookies), the distribution of batch means still approaches normality as batch size increases. This works for uniform, exponential, and other non-normal population distributions as well.

Sample Size Effects on Distribution Shape

How quickly the sampling distribution becomes normal depends on how non-normal the population is:

• Nearly symmetric populations: Even small samples ($n = 10$ or $15$) produce sampling distributions that look roughly normal.
• Moderately skewed populations: You'll typically need $n \geq 30$ before the sampling distribution is approximately normal. This is the common "rule of thumb" for when the CLT kicks in.
• Heavily skewed populations or those with extreme outliers: Larger samples are needed to overcome the skewness. A batch of 5 cookies from a highly skewed weight distribution might still produce a skewed distribution of means, while batches of 50+ will look much more normal.

The key takeaway: the CLT holds regardless of the population distribution shape, as long as the sample size is large enough. The $n \geq 30$ guideline works in most practical situations, but more extreme population shapes demand larger samples.

Statistical Inference and Probability

The CLT is what makes most of inferential statistics possible. Because sampling distributions become normal and predictable, you can:

• Build confidence intervals for a population mean using the normal distribution, even when the underlying data isn't normal
• Conduct hypothesis tests by calculating how likely a particular sample mean is under a claimed population parameter
• Calculate probabilities about sample means directly, since each batch mean is a random variable drawn from an approximately normal sampling distribution

Without the CLT, you'd need to know the exact shape of the population distribution before doing any of this. The theorem removes that requirement for sufficiently large samples, which is why it's so central to applied statistics.