Scatterplots and Relationships

Before you can build any AI system that predicts, recommends, or detects, you must see how two things move together. A scatterplot is the first tool for that: it shows you whether house price rises with size, whether a sensor reading foreshadows a failure, whether customers who buy one product tend to buy another. This topic unlocks the rest of Statistical Inference — the concepts of correlation, regression, and model diagnostics all begin with the shape of a scatterplot. If you skip this, you will later fit a straight line to data that curves, or declare a relationship that is only an outlier. Learning to read scatterplots first is what makes every later model honest.

Imagine you have two lists of numbers, each number paired with its partner: for each house, its size in square feet and its price in lakhs. You want to know if bigger houses cost more. You could look at the two lists side by side, but with fifty pairs your eyes blur. Instead, you mark a dot on a page for each pair: left-right position is size, up-down position is price. With all fifty dots down, patterns jump out. If the dots slant upward from left to right, bigger houses do tend to cost more — that is an association. If the dots form a loose cloud with no slant, size tells you little about price. If the dots curve, like a smile, then a straight-line summary would mislead you. A scatterplot is your naked-eye view of a relationship before any number tries to summarise it. The key habit: always plot first, even when you plan to compute a correlation coefficient later. The plot shows you what the number hides.

Picture yourself as a detective entering a room where the floor is a giant grid: the left edge is labelled in years of experience, the bottom edge in salary. Each person in the company places a coin on the grid at the point matching their experience and salary. You stand back and look at the pattern of coins. Do they form a rising diagonal (more experience, more salary)? Do they cluster into two separate blobs (one group well paid, another poorly)? Is there a single coin far from everyone else (an outlier, maybe the CEO, maybe an error)? That is exactly what a scatterplot does — it turns a table of numbers into a pattern of dots you can read with your eyes. The analogy works until you try to scale it: a real dataset has hundreds or thousands of points, and dots overlap. You need to add transparency or use small random jitter so you can see the density underneath. Also, the grid has no units — you must label axes carefully, or a squashed or stretched scale will make a slope look steeper or shallower than it is. The detective's eye is the point, not the coordinates.

A scatterplot is a graph that shows each observation as a dot positioned by two variables — one on the horizontal axis, one on the vertical — so that patterns of association, curvature, clustering, and outliers are visible at a glance.

You have not yet studied statistics formally, but this topic connects directly to the neighbours in your library. A scatterplot is the visual companion to Descriptive Statistics: while a mean and standard deviation summarise one variable, a scatterplot reveals how two variables move together. When you later study Correlation coefficients, you will see they turn the scatterplot's shape into a single number — and you will learn why that number can be misleading (Anscombe's quartet). When you move to Regression, the scatterplot is the first diagnostic you draw to check whether a straight line is even appropriate. And when you reach Model Evaluation, scatterplots of predicted vs actual values are how you spot bias or curvature. In short, every branch of Statistical Inference — from hypothesis testing to Gaussian Process Regression — assumes you have looked at the data first.