Describing Direction, Form, and Strength

Introduction

A scatterplot can contain a great deal of information, but a useful description should organize what the eye sees. Three questions provide a clear starting point: Which way does the pattern go? What form does it take? How closely do the points follow it? These questions concern direction, form, and strength. They describe features of the plotted relationship; they are not interchangeable with a single numerical correlation.

For example, two scatterplots might both have an upward trend, yet one could show points close to a straight pattern while the other shows much more scatter. Reporting only that both are “positive” misses an important difference. A careful description gives the reader a compact verbal account of the plot without claiming more than the plot supports.

Definition: In a scatterplot of paired quantitative variables, direction describes whether the overall pattern tends to rise or fall as the predictor increases; form describes the pattern’s broad shape, such as roughly straight or visibly non-straight; and strength describes how closely the points cluster around that overall pattern. These are visual descriptions, not conclusions about cause and effect.

Direction: Which Way Does the Pattern Go?

A scatterplot has a positive direction when larger predictor values generally accompany larger response values. The cloud of points tends to rise from left to right. A negative direction occurs when larger predictor values generally accompany smaller response values, so the cloud tends to fall from left to right.

Sometimes the points do not show one clear overall direction. The plot may look like a broad cloud, or its pattern may go up in one region and down in another. In such a case, do not force a positive or negative label onto the entire plot. Describe what is visible, for example, “there is no clear overall upward or downward trend.” That wording does not prove that the variables are unrelated; it says only that the plot does not show a single clear direction.

Direction is about the overall tendency, not whether every point follows it. A positive association can include some points where a larger predictor value is paired with a smaller response value. The description summarizes the pattern of the cloud rather than demanding a perfect rule.

Form: What Shape Does the Pattern Take?

Form is a description of shape. A roughly linear pattern can be summarized by a straight-line trend: as the predictor changes, the response tends to change at a fairly steady rate. A nonlinear or curved pattern visibly departs from a straight-line shape. A scatterplot may also show no clear pattern at all.

Use qualified language. In real data, points rarely sit exactly on a line, so “roughly linear” is usually more appropriate than “linear” unless the plotted points truly fall on a straight line. Similarly, saying “somewhat curved” is often more faithful than suggesting a precise mathematical shape from a visual inspection alone.

Form matters because a straight-line summary can miss a systematic pattern that bends or changes direction. A plot may show a clear relationship even when a straight-line description is inadequate. For this reason, inspect the plotted points and any displayed smoother for shape rather than relying on a number alone. Describe the shape you see without inferring a specific mechanism or treating a visual pattern as a fitted model.

Strength: How Closely Do Points Follow the Pattern?

For visual description, strength concerns the scatter around the overall pattern. If points lie close to a recognizable trend, the visual association is strong. If they are spread broadly around it, the association is weaker. The relevant question is not whether the trend is steep, but whether the points cluster tightly around the pattern.

Strength is always judged relative to the pattern in the plot and its scale. A steep-looking trend is not automatically strong: a cloud can rise sharply while being widely scattered around that rise. A shallow trend can be visually strong if its points stay close to a clear pattern. The units used for the axes also affect how steep a trend appears, so steepness should not be used as a measure of strength.

A strength description should be calibrated to the evidence. Terms such as “weak,” “moderate,” and “strong” are useful summaries, not universal cutoffs. State what you mean when needed: for example, “the association is moderately strong, with noticeable scatter around a roughly straight upward pattern.” This ties the judgment to the plot instead of leaving “moderate” unexplained.

Visual Description and Numerical Correlation

For two quantitative variables, the Pearson sample correlation, denoted \(r\), is a numerical summary of the direction and strength of their linear association. Its value is between \(-1\) and \(1\). Positive values indicate a positive linear association, negative values a negative linear association, and values near zero little linear association. Values closer to \(1\) or \(-1\) indicate a stronger linear association. A correlation of \(1\) or \(-1\) corresponds to points lying exactly on a straight line with positive or negative direction, respectively, provided both variables vary.

r = \frac{\sum_{i=1}^{n}(x_i-\bar{x})(y_i-\bar{y})} {\sqrt{\sum_{i=1}^{n}(x_i-\bar{x})^2 \sum_{i=1}^{n}(y_i-\bar{y})^2}}

The formula combines the paired deviations of the observed \(x\) and \(y\) values into one number. It does not give a complete account of the plotted form, and it does not show the point cloud itself. In particular, a small correlation means little linear association; it does not establish that no relationship of any kind is present. That is why a correlation should be interpreted alongside a plot, not as a replacement for one.

Why Correlation and Visual Steepness Differ

The formula for \(r\) also helps explain why correlation is not the same as the apparent slope of a plotted trend. If every \(x\) value is shifted or multiplied by a positive constant, the centered \(x\) values in the formula are also multiplied by that constant. The numerator and denominator both acquire the same factor, which cancels. The correlation therefore stays the same, although the axis units and visual steepness can change.

x_i^* = a + b x_i,\quad b>0 \qquad\Longrightarrow\qquad x_i^*-\bar{x}^* = b(x_i-\bar{x}) $$ $$ r_{x^*,y} = \frac{b\sum_{i=1}^{n}(x_i-\bar{x})(y_i-\bar{y})} {b\sqrt{\sum_{i=1}^{n}(x_i-\bar{x})^2 \sum_{i=1}^{n}(y_i-\bar{y})^2}} = r_{x,y}

Thus, converting a predictor from hours to minutes does not change its correlation with the response, but it changes the numerical scale of the horizontal axis. This is another reason not to call a steep-looking pattern “strong” merely because it looks steep. Correlation is a standardized summary of linear association; the plotted direction, form, and scatter remain descriptions of the visible relationship.

Worked Example: A Tight Positive Linear Pattern

Hypothetical data: A tutor records hours of practice \(x\) and quiz points \(y\), out of 10, for four fictional learners: \((1,2)\), \((2,4)\), \((3,5)\), and \((4,7)\). Suppose these pairs are shown in a scatterplot. Describe its direction, form, and strength, and compare that description with its Pearson correlation.

As practice hours increase from 1 to 4, quiz points generally increase from 2 to 7. The direction is positive. The points lie close to an upward straight pattern, so the form is roughly linear and the visual association is strong. They are not exactly on one line, so “perfect” would be inaccurate.

The correlation can be checked from the definition. Here \(\bar{x}=2.5\) hours and \(\bar{y}=4.5\) points. The centered \(x\) values are \(-1.5,-0.5,0.5,1.5\), and the centered \(y\) values are \(-2.5,-0.5,0.5,2.5\). The sum of their paired products is \(3.75+0.25+0.25+3.75=8\). The sums of squared centered values are \(5\) for \(x\) and \(13\) for \(y\). Therefore,

r = \frac{8}{\sqrt{5(13)}} = \frac{8}{\sqrt{65}} \approx 0.9923

The correlation is close to \(1\), consistent with a strong positive linear association in this small hypothetical data set. The plot description still adds information: it tells us that the form is roughly straight and that the points are close to, but not exactly on, the trend. The four observations alone do not establish that practice causes higher scores.

Worked Example: A Negative Trend With Noticeable Scatter

Hypothetical data: A technician records the number of practice runs on a fictional task, \(x\), and completion time in minutes, \(y\), for five trials: \((1,13)\), \((2,11)\), \((3,12)\), \((4,7)\), and \((5,8)\). Based on the plotted points, describe the association.

The first and last observations are 4 practice runs apart and differ by \(8-13=-5\) minutes in completion time. The endpoint change per predictor unit is therefore \(-5/4=-1.25\) minutes per run. This calculation is only an endpoint summary, not a fitted slope; the points at runs 2 and 3 also show that the response does not decrease at every step.

Overall, the cloud tends to fall from left to right, so its direction is negative. Its broad pattern is roughly decreasing and approximately straight, but the points do not hug a narrow line: completion time rises from 11 to 12 minutes between the second and third runs, and the response values are noticeably spread around the downward tendency. A fair description is “a negative, roughly linear association with noticeable scatter,” rather than “a perfect decreasing relationship.”

The arithmetic supports the direction, but the plot is needed to judge the shape and scatter. The endpoint calculation alone would not tell us whether the intermediate points follow a straight pattern or how tightly the entire cloud clusters around it.

Worked Example: Similar Direction, Different Scatter

Hypothetical data: Two fictional practice groups have the same predictor values, \(x=1,2,3,4,5\), measured in sessions. Group A has response values \(12,14,16,18,20\) minutes; Group B has values \(10,13,15,19,23\) minutes. Imagine plotting each group with the same axis scales.

For both groups, the response generally rises as sessions increase, so both associations have positive direction. Group A’s points fall on a straight increasing pattern: each one-session increase accompanies a two-minute increase. Group B also tends upward overall, but its points are less tightly organized around a straight trend. For instance, from session 1 to 2, Group B rises by \(13-10=3\) minutes, while from session 4 to 5 it rises by \(23-19=4\) minutes; those changes do not follow a constant increment.

The key comparison is visual: both plots point upward, but Group A has a tighter, exactly straight pattern and Group B has more scatter and less regular form. Describing both with only the word “positive” would omit these differences. A single numerical correlation also cannot display the actual coordinates, so the scatterplots remain necessary for a full description.

A Practical Description and a Common Pitfall

A concise description can follow a consistent order: give direction, state form, then characterize scatter around the pattern. Include the variables and units so the description has meaning outside the plot.

Describe direction.
State whether the overall pattern rises, falls, or shows no single clear direction as the predictor increases.

Describe form.
Say whether the pattern is roughly straight, visibly non-straight, or not clearly patterned.

Describe strength and context.
Explain whether points cluster tightly or show substantial scatter around the pattern, naming the variables and their units.

Common mistake: “The plot has a steep positive relationship, so it is strong.” This confuses steepness with strength. A steep pattern can have substantial scatter, and a shallow pattern can have very little scatter. Correct the description by treating the features separately: direction is positive; form is roughly linear; strength is judged by how closely points cluster around that upward pattern. Then use the axis units to describe steepness, if it is relevant.

Key takeaway: Direction, form, and strength are distinct features of a scatterplot. Pearson correlation summarizes the direction and strength of linear association in one number, but it does not replace inspection of the plotted form and scatter.

Check Your Understanding

For each question, use the distinction between what the scatterplot shows and what a numerical correlation summarizes.

A scatterplot rises from left to right, but points are widely spread around the trend. Describe its direction and visual strength.
Why is “steep” not a sufficient description of association strength?
What does a Pearson correlation close to zero say about linear association, and what does it not establish about every possible relationship?
Two plots show upward trends, but one has points close to a straight pattern and the other has much more scatter. Give a description that distinguishes them.
Why should a correlation be considered alongside the scatterplot rather than used as a complete description of it?