Foundations — Confidence intervals

Before you can read the parent note's derivation, you need every letter and squiggle it uses to mean something concrete. Below we build each one from nothing, in the order they depend on each other. Nothing is used before it is drawn.

1. Population vs. Sample — the two worlds

Everything in this topic lives in a tension between two things: the whole world we cannot fully see, and the tiny scoop we actually measured.

Look at the figure: the big cloud on the left is the population (thousands of dots). The little circle on the right is your sample — just a few dots you happened to grab. The whole game of confidence intervals is: guess a fact about the big cloud using only the little circle.

Why the topic needs this: every symbol below is tagged as either a population fact (hidden, fixed, usually Greek letters) or a sample fact (measured, changes each scoop, usually Latin letters). Keeping these two worlds separate is the single most common thing beginners mix up.

2. $\mu$ — the true mean (the fish we hunt)

The picture: on the big cloud, $\mu$ is the exact centre of balance — the one spot where the cloud would perfectly balance on a pin.

Why the topic needs it: $\mu$ is the fish we are trying to catch. The entire confidence interval exists to bracket this one hidden number. Because it is fixed, we never say "the probability $\mu$ is here" — $\mu$ doesn't move.

3. $\sigma$ and ${\sigma }^2$ — how spread out the world is

The picture shows two bell-shaped clouds with the same centre $\mu$ but different widths. The narrow one has small $\sigma$ (everyone close to average); the wide one has large $\sigma$ (values scattered far). $\sigma$ is literally the typical distance from the middle.

Why the topic needs it: a confidence interval's width grows with $\sigma$. If the world is very spread out, your net must be wider to be trustworthy. We square then square-root so that distances above and below $\mu$ don't cancel out — squaring kills the minus signs.

4. $\bar{x}$ and $\bar{X}$ — the average you actually computed

Let's decode that formula symbol by symbol, because the parent note uses it constantly:

The picture for $\bar{x}$: on your little sample circle, $\bar{x}$ is its balance point — the sample's own centre. It is our best single guess for the hidden $\mu$, but it wobbles.

5. $n$ — sample size (how big your scoop is)

Why the topic needs it: $n$ is the knob you can actually control. It appears as $\sqrt{n}$ in the denominator of the interval width, so measuring more shrinks your net. This is the entire reason "more data = more confidence."

6. Random variable & distribution — why one average has a shape

The figure: imagine repeating "grab a fresh sample, compute $\bar{x}$" thousands of times. Each $\bar{x}$ is a dot. Pile them up and you get a bell shape centred on $\mu$. This pile is the sampling distribution of the mean. Crucially, it is narrower than the raw data cloud — averages jitter less than individual measurements.

Why the topic needs it: the whole derivation asks "how far can $\bar{x}$ stray from $\mu$?" That question is answered by the shape of this bell. No distribution, no interval.

This bell being normal (for large $n$) is exactly what the Central Limit Theorem guarantees — the parent leans on it, and it deserves its own study.

7. The normal distribution $N(\text{mean},\text{variance})$

Note the width ${\sigma }^2/n$: the sample-mean bell has the population spread ${\sigma }^2$ shrunk by dividing by $n$. That's the maths behind "averages jitter less" — bigger $n$, tighter bell.

8. $Z$ — the standardized ruler

Why the topic needs it: we know exact cut-off points on the one standard bell. By converting to $Z$, we borrow those universal cut-offs for any problem. This is the "standardize to control probability" step in the parent.

9. $1-\alpha$, $\alpha$, and $z_{\alpha /2}$ — confidence and its critical value

The picture: the standard bell with two small shaded tails, each holding area $\alpha /2$. The vertical lines where the shading starts are at $\pm z_{\alpha /2}$. The unshaded middle holds $1-\alpha$ of the area — that's the region we trust.

10. Standard error — the spread of the guess itself

Why $\sqrt{n}$ and not $n$? Variance divides by $n$; standard deviation is its square root, so the spread divides by $\sqrt{n}$. This is exactly why quadrupling your data only halves your error — see Standard Error for the full story.

11. $s$, $n-1$, and the t-distribution — coping when $\sigma$ is unknown

Prerequisite map

Equipment checklist

Cover the right side; you're ready when each reveal matches your own answer.

Next: return to the parent note and watch every one of these symbols slot into the derivation. Prefer Hinglish? See 1.3.21 Confidence intervals (Hinglish). Deeper roads: Central Limit Theorem, Standard Error, Hypothesis Testing, Bootstrap Methods.