A scatter plot can show that institutions with higher spending also report higher completion. It cannot, by itself, show what would happen if a particular student paid more or if a college increased a budget. The difference matters when a descriptive relationship becomes a recommendation to buy a more expensive education.

NIST describes scatter plots as tools for examining relationships between variables, including form, variation and outliers. The institutional example below is fictional and supplies no empirical estimate. Its purpose is to identify the extra reasoning required before an observed pattern can be interpreted causally.

Draw the relationship without naming its cause

Imagine three colleges measured on the same invented expenditure definition and completion period. College A spends 10 units per student and completes 50% of entrants; B spends 20 and completes 70%; C spends 30 and completes 90%. Plotting spending horizontally and completion vertically produces a perfectly increasing straight line.

Fictional college Spending units per student Completion rate
A 10 50%
B 20 70%
C 30 90%

The descriptive slope is two percentage points of completion per spending unit. That is a property of these three supplied pairs. It does not establish that adding ten spending units to A would raise its completion by twenty points. The table has no observation of A under that intervention.

Several explanations could produce the same pattern. More spending might support completion; institutions with better-resourced entrants might also have more spending; programme duration and student mix might differ; or completion could influence resources through reputation and funding. The small table cannot separate these mechanisms.

Identify the unit of the claim

Institution-level data describe institutions. An institution with a higher average completion rate does not necessarily improve the outcome of every kind of entrant. A student-level claim needs evidence about comparable students and the pathways through which the institution could affect them. Moving between those levels without justification is a substantial analytical leap.

Also separate spending from price. Institutional expenditure per student is not the same as tuition paid by a particular household. An institution may fund spending through public support, donations or other income. A relationship involving expenditure cannot automatically be restated as a relationship between personal tuition and educational benefit.

Before interpreting a real plot, examine whether programmes, observation years, accounting categories and denominator populations match. A college with extensive research activity may not be comparable to one whose spending measure covers teaching alone. A graph does not repair mismatched variables simply by placing them on common axes.

Turn the pattern into questions worth investigating

Ask which spending categories could plausibly influence the outcome, when the expenditure occurred and whether it preceded the measured completion. Then ask what evidence compares like with like: a credible design might exploit a policy change, comparable groups or another justified source of variation. Merely adding many variables to a spreadsheet does not guarantee that selection bias disappears.

Check sensitivity to outliers and coverage. Three carefully selected institutions can create a striking line that is absent in the wider population. Report the selection rule and missing observations, and avoid extrapolating beyond the spending range actually represented.

For a personal shortlist, treat the plot as a prompt to inspect concrete services such as academic support and course access, not as a causal price calculator. The appropriate statement for the invented table is “spending and completion rise together in these three observations.” A stronger claim requires stronger evidence. That boundary preserves the useful descriptive pattern without turning an institutional association into an unsupported promise about a student's future.