Introduction to Sampling Distributions
Sampling Variability: This principle recognizes that different random samples from the same population will produce different values for the same statistic. This natural fluctuation is the reason why a single sample statistic is rarely exactly equal to the population parameter.
The 'All Possible Samples' Requirement: For a distribution to be a true sampling distribution, it must theoretically include the statistic from every single possible combination of individuals that could be selected for a sample of size .
Consistency of Sample Size: A sampling distribution is only valid for a specific sample size . If the sample size changes, the spread and potentially the shape of the sampling distribution will also change.
Approximation via Simulation: In practice, it is often impossible to take every possible sample from a large population. Instead, statisticians use computer simulations to take a very large number of random samples (e.g., 1,000 or 10,000) to approximate the sampling distribution.
Randomization Distributions: When a sampling distribution is created through repeated random assignment or sampling in a simulation, it is often referred to as a randomization distribution. This provides a visual and numerical estimate of the statistic's behavior.
Increasing Iterations: As the number of simulated samples increases, the resulting histogram becomes a more accurate representation of the true theoretical sampling distribution. This allows for the calculation of probabilities even when the population parameters are unknown.
| Feature | Population Distribution | Sample Distribution | Sampling Distribution |
|---|---|---|---|
| What is distributed? | Individual values in the population | Individual values in one sample | Values of a statistic from many samples |
| Center | Population Mean () | Sample Mean () | Mean of all sample statistics |
| Spread | Population Std Dev () | Sample Std Dev () | Standard Error of the statistic |
| Purpose | Describes the whole group | Describes a single subset | Describes the reliability of the statistic |
Contextual Labeling: Always define your parameters and statistics in the context of the problem. Instead of saying 'the mean,' specify 'the mean weight of all apples in the orchard' versus 'the mean weight of a sample of 10 apples.'
Check the Sample Size: When comparing two sampling distributions, always verify if the sample size is the same. A larger typically results in a sampling distribution that is more tightly clustered around the true population parameter.
Identify the Statistic: Before solving a problem, identify exactly which statistic is being tracked (mean, median, range, or proportion). The rules for the sampling distribution may vary depending on the specific statistic used.
Confusing with Iterations: Students often confuse the sample size () with the number of samples taken in a simulation. Increasing the number of samples in a simulation makes the distribution clearer, but only increasing actually changes the underlying spread of the sampling distribution.
Assuming Normality: Do not assume a sampling distribution is automatically normal. While many sampling distributions (like the mean) tend toward normality under certain conditions, others (like the range or the maximum) may remain skewed regardless of sample size.
The 'One Sample' Fallacy: A common error is thinking that a sampling distribution is just a collection of data from one large sample. It is actually a collection of single summary values from many different samples.