Parameters vs. Statistics

The Core Problem: From Sample to Population

Statistical inference addresses the problem of learning about the characteristics of a large group (the population) by studying a small, representative subset (the sample). Let's formalize these ideas.

To start, let's get a crisp, formal definition of the problem.

As the notes highlight, we are moving from a world where parameters are known to one where they are the unknown quantities we wish to find.

Population and Parameters

• A population is the complete set of all individuals, items, or data points you are interested in studying. It can be finite (e.g., all 1,200 students in a specific school) or conceptually infinite (e.g., all possible outcomes of a coin toss).
• A parameter is a numerical value that describes a characteristic of the population. It is a single, fixed number, but its value is typically unknown to us.

We use Greek letters to denote parameters. Common examples include:

• Population Mean ($\mu$): The true average of the entire population.
• Population Variance (${\sigma }^2$): The true variance of the entire population, measuring its spread.
• Population Proportion ($p$ or $\pi$): The true proportion of the population that has a certain characteristic.

For instance, in your work as an R&D engineer, the true average lifespan of every EV battery your company could ever produce is a population parameter ($\mu$). It's a fixed number, but you can never know it exactly because you can't test every battery.

Sample and Statistics

• A sample is a subset of individuals, items, or data points selected from the population.
• A statistic is any numerical value calculated from a sample. Because a different sample would likely yield a different value, a statistic is a random variable. Its value is known to us, as we calculate it directly from our data.

We typically use Roman letters for statistics.

• Sample Mean ($\bar{x}$): The average of the observations in our sample.
• Sample Variance ($s^2$): The variance of our sample data.
• Sample Proportion ($\hat{p}$): The proportion of our sample that has a certain characteristic.

A statistic that is used to guess the value of an unknown parameter is called an estimator. Let's watch a short video that introduces this concept very clearly.

The Central Idea: Using Statistics to Estimate Parameters

The fundamental goal of statistical inference is to use a sample statistic to make an intelligent guess about an unknown population parameter.

• We use the sample mean, $\bar{x}$, to estimate the population mean, $\mu$.
• We use the sample proportion, $\hat{p}$, to estimate the population proportion, $p$.
• We use the sample variance, $s^2$, to estimate the population variance, ${\sigma }^2$.

This table provides a concise summary of the notation.

A Point of Nuance: The Sample Variance $s^2$

You may have noticed something peculiar about the formula for sample variance. When calculating the variance of a full population, we divide the sum of squared deviations by the population size, $N$.
${\sigma }^2=\frac{1}{N}{\sum }_{i=1}^N(X_i-\mu {)}^2$
However, when we calculate the sample variance as an estimator for ${\sigma }^2$, we divide by $n-1$, not $n$.
$s^2=\frac{1}{n-1}{\sum }_{i=1}^n(x_i-\bar{x}{)}^2$
Why this difference? The reason is subtle but crucial. It turns out that if we divided by $n$, our sample variance would, on average, slightly underestimate the true population variance ${\sigma }^2$. Dividing by $n-1$ corrects for this tendency, making $s^2$ a "better" guess. We will prove this mathematically in the next lesson when we discuss the concept of bias.

For now, this distinction is a perfect practical example of the difference between simply describing a sample and using a sample to infer something about a population.

Self-Assessment

To solidify your understanding, consider the following scenarios. For each one, identify the population, the sample, the parameter of interest, and the corresponding statistic.

1. Manufacturing Quality Control: You want to know the true average tensile strength of a large batch of 10,000 steel rods. You randomly select 50 rods, test them, and calculate their average tensile strength.
   • Population:
   • Sample:
   • Parameter:
   • Statistic:

2. Election Polling: A news organization wants to gauge public opinion on a ballot measure. They want to know the proportion of all registered voters in a state who support it. They call 1,200 registered voters and ask for their opinion.
   • Population:
   • Sample:
   • Parameter:
   • Statistic:

3. Experimental Physics: A physicist is trying to determine a fundamental physical constant. Each measurement she takes is subject to random error. She takes 200 independent measurements and wants to describe the variability of her measurement apparatus.
   • Population:
   • Sample:
   • Parameter:
   • Statistic:

Conclusion

In this lesson, we established the fundamental vocabulary of statistical inference. This conceptual clarity is essential for everything that follows.

Key Takeaways:

• Population Parameter: A fixed, unknown numerical characteristic of an entire population (e.g., $\mu$, ${\sigma }^2$). Our goal is to estimate this.
• Sample Statistic: A known numerical value calculated from a sample (e.g., $\bar{x}$, $s^2$). It is our tool for estimation.
• Randomness: A statistic is a random variable because its value depends on the specific random sample drawn. A parameter is a constant.
• Inference: The core idea is to use the known value of a statistic to make an educated guess about the unknown value of a parameter.

Preview of the Next Lesson:

We have established that we use statistics to estimate parameters. This naturally leads to the next question: what makes a statistic a good estimator? Just because we can calculate $\bar{x}$, for example, how do we know it's a reasonable guess for $\mu$?

In the next lesson, we will begin to answer this by formalizing the properties of estimators. We will define and mathematically formulate three key characteristics: bias, variance, and consistency. These will give us a framework for evaluating and comparing different estimators.