Research Guides
Read a Website Scatter Plot Without Inventing a Ranking Rule
A website comparison shows that pages with more words also receive more search clicks. It is tempting to turn that pattern into an instruction to make every page longer. Before doing so, inspect what each plotted point represents and what other differences the comparison leaves unresolved.
This worksheet helps reviewers use a scatter plot as an investigative aid rather than a ranking recipe. It uses a fictional website example and does not describe a Colquhoun.shop charting feature.
1. Give the chart a bounded question
NIST describes scatter plots as a way to examine relationships between paired variables, including nonlinear patterns and outliers. Its guidance explicitly distinguishes association from proof of cause and effect. NIST: Scatter Plot
Write a question such as, “How do recorded word counts and search clicks vary across this fixed set of guides?” That is narrower than asking whether longer content earns better rankings. The latter requires evidence the chart alone does not provide.
2. Define one point before drawing many
For a page-level chart, pair the word count and click total for the same page. Record the text-counting method, traffic source and observation period. Do not pair a current article revision with an unexplained lifetime traffic total and then treat the comparison as a controlled test.
Choose the included page set before selecting a pattern to illustrate. Keep excluded or unavailable rows in a separate record with reasons. Otherwise, the chart can look clean because inconvenient observations quietly disappeared.
3. Investigate an attractive pattern
In an invented review of 30 guides, longer pages tend to show more clicks. The editor notices that the longest guides were also published earliest and cover the site's most familiar topics. Those facts suggest alternative explanations worth checking; they do not establish which explanation is correct.
The editor records page age and topic category in the working file, then inspects the pattern within sensible groups. If a group contains only a few pages, that limitation stays in the report. Subdividing a small sample does not magically create a reliable experiment.
The resulting next step might be a closer editorial review of specific guides, not a site-wide word-count target. A useful chart narrows the investigation rather than assigning an unsupported universal rule.
4. Keep unusual points visible
When one point sits far from the rest, inspect its source record. It might reflect a measurement error, a different page type or a genuinely unusual result. Do not remove it merely because a trend line looks tidier without it.
If a documented error is corrected, preserve the original value and correction note. If you present an additional view without a particular observation, label the exclusion and retain the full-sample view alongside it.
5. Copy this scatter-plot review card
- Question: the relationship being explored.
- Point: the entity represented by one paired observation.
- Axes: variables, units and measurement methods.
- Scope: page set, dates and exclusions.
- Pattern: what is visible without a causal interpretation.
- Alternatives: other recorded differences requiring investigation.
- Follow-up: one check that could clarify the finding.
Keep the underlying paired records with the figure so another reviewer can inspect a surprising point. Label any transformations or fitted line rather than allowing formatting choices to become hidden analytical decisions.
6. Write the limited conclusion first
A defensible summary says, “These sampled guides show an association between the measured variables; the review has not established that changing word count changes clicks.” Follow it with the specific uncertainty and next check.
Do not convert an observational pattern into a promised search result. The valuable output is a reproducible question and a clearer investigation, not a chart-shaped guarantee.