What Is a Scatter Plot?

A scatter plot places one numeric variable on the horizontal axis and another numeric variable on the vertical axis. Each point represents one observation with both measurements. The pattern can reveal association, clusters, gaps, and unusual observations, but it does not prove causation.

Scatter plots are useful when the question is about a relationship rather than a sequence or category total. A marketing team might compare ad spend with sign-ups for each campaign. A product analyst might compare session length with actions completed for each user cohort. A scientist might compare two measurements taken from the same specimen.

The pairing is essential. If one row contains the spend for Campaign A and the sign-ups for Campaign B, the point is meaningless even though the software can draw it. Before looking for a pattern, confirm what one row represents and how the two values were collected.

How a scatter plot works

Every observation becomes a coordinate written conceptually as (x, y). The x value determines horizontal position; the y value determines vertical position. Points farther right have larger x values, and points higher on the chart have larger y values. Unlike a line graph, a scatter plot does not normally connect independent observations.

For example, a row with Ad spend = 180 and Sign-ups = 26 becomes a point at (180, 26). A second row at (150, 19) occupies another position. Looking across all points helps you ask whether larger x values tend to occur with larger y values, whether the spread changes, and whether subgroups behave differently.

The NIST Engineering Statistics Handbook describes a scatter plot as a way to reveal relationships between two variables. It is an exploratory view, not a complete statistical conclusion.

Scatter plot example with fictional data

Download the fictional advertising CSV, or copy the table below. Each row represents one fictional campaign observation. Channel is an optional grouping field rather than one of the axes.

Ad spend,Signups,Channel
120,18,Search
180,26,Search
240,31,Search
310,42,Search
90,14,Social
150,19,Social
230,28,Social
280,35,Social

The data is fictional and is provided only for demonstration. Ad spend belongs on the x-axis and Sign-ups on the y-axis because the question is how sign-ups vary across observations with different spend. That orientation does not establish that spend caused the sign-ups; it simply makes the analytical question explicit.

What patterns can a scatter plot show?

Positive association

When points tend to rise as they move from left to right, larger x values are associated with larger y values. The fictional campaigns show this general direction. Describe it as a positive association unless a designed experiment and appropriate analysis justify a causal statement.

Negative association

When points tend to fall as they move right, larger x values are associated with smaller y values. Price versus demand can sometimes show this pattern, although many other variables may affect both measurements.

Weak or no visible association

A diffuse cloud without a clear direction may indicate a weak linear association. It does not prove that the variables are unrelated. A nonlinear curve, hidden subgroups, measurement error, or a narrow observed range can all obscure the relationship.

Clusters and groups

Points may form separate clusters. Coloring the fictional observations by Channel lets you see whether Search and Social occupy different regions. Group color should represent a real field from the data, not an arbitrary palette added for decoration.

Outliers and unusual observations

A point far from the main cloud deserves investigation. It may be a valid exceptional campaign, a data-entry error, or an observation collected under different conditions. Do not automatically delete it. Check the source and report any exclusion.

Association is not causation

A scatter plot can show that two measurements vary together. It cannot, by itself, determine why. In the advertising example, campaigns with more spend also have more sign-ups. Larger campaigns may also run longer, target larger audiences, use different creative, or occur during high-demand periods. Those variables can influence both axes.

Reverse direction is another possibility: teams may allocate more spend to campaigns that already show stronger demand. A third variable may drive both. Causal claims require a defensible study design and analysis beyond the visual pattern.

Even a fitted trend line does not solve that problem. It summarizes a form of association under assumptions. Statistical significance, uncertainty, residual patterns, and sample design need separate attention. Charttoolbox intentionally does not add regression or significance claims to the basic scatter plot.

How to read a scatter plot step by step

1. Read the title and source. Identify the population, period, and context represented.
2. Read both axes. Note the variable names, units, range, and whether either axis has been transformed.
3. Identify one point. State what a single observation means before describing the whole cloud.
4. Describe direction. Does y generally increase, decrease, or show no clear linear direction as x increases?
5. Describe form and strength cautiously. Is the pattern roughly straight, curved, clustered, or diffuse?
6. Inspect unusual points. Check outliers, gaps, overlapping observations, and boundaries.
7. Compare groups if present. Look for differences within and between groups without assuming group membership caused them.
8. Return to the data. Verify surprising points and missing values in the source table.

Scatter plot, line graph, or correlation matrix?

A line graph connects points because their sequence is meaningful. Connecting the campaign observations in the fictional table would imply an order that the data does not contain. A correlation matrix compresses several pairwise linear relationships into coefficients, but it hides the point-level shape and outliers that a scatter plot reveals.

Common scatter plot mistakes

Pairing unrelated rows

Sorting one column without the rest of the table breaks the observation pairs. Always sort the complete table so every row remains intact.

Using a category as a numeric axis

Product IDs, customer numbers, and arbitrary category codes may contain digits but do not represent measurable quantities. Use a grouping field or bar chart instead.

Hiding overlapping points

Several observations can share the same coordinate. The chart may look like one point. Smaller points, partial opacity, or a supporting data table help reveal that overlap. Do not infer sample size by counting visible marks alone.

Truncating axes to exaggerate a pattern

Scatter plots do not always require zero on both axes, but the displayed range affects perception. Use a range that shows the observed data and disclose transformations or deliberate limits. Avoid a narrow crop chosen only to make a weak pattern appear steep.

Claiming causation from direction

Write “associated with” or “tends to occur with” when that is all the graph supports. Causal language needs stronger evidence.

Make a scatter plot in Charttoolbox

Open the scatter plot maker and upload CSV, TSV, XLS/XLSX, or paste the cells. XLS and XLSX imports read the first worksheet. You can also load the fictional sample directly.

1. Review the automatically detected fields. Set X to Ad spend and Y to Signups.
2. Optionally set Group by to Channel. Use None when a single color is clearer.
3. Adjust point size and opacity, add a title, source, takeaway, and footnote, then export PNG or SVG.

Each row produces one point, including repeated coordinates. The tool accepts up to twelve groups. X and Y can use different units, but values within one selected column must use one consistent format. Blank or invalid selected cells are reported rather than silently dropped. Hovering shows full values, and the imported-data table preserves every observation.

Spreadsheet contents are processed locally in the browser. Optional analytics loads only after consent and does not receive filenames, table contents, or generated chart data. The browser workflow does not save a cloud project, so keep the source file and exported chart yourself.

Questions to ask before sharing

• Can the reader name what one point represents?
• Are both axes numeric measurements with clear units?
• Did the table preserve the correct x–y pairing?
• Are groups real and limited enough to distinguish?
• Could overlapping points or a restricted range hide important structure?
• Does the caption describe association without claiming unsupported causation?