Question

Explain How to Identify a Relationship in a Scatterplot
How is a relationship determined when looking at a scatterplot?

Answer 1

To determine whether two quantitative variables are related from a scatterplot, look for these features:

1. Direction
- Positive (direct): points trend upward left-to-right — as x increases, y tends to increase.
- Negative (inverse): points trend downward left-to-right — as x increases, y tends to decrease.
- No apparent direction: no systematic upward or downward trend.

2. Form
- Linear: points cluster around a straight line.
- Nonlinear (curved): points follow a curve (quadratic, exponential, logarithmic, etc.).
- No clear form: points are randomly scattered.

3. Strength
- Strong: points lie closely around the line or curve (little scatter).
- Moderate: noticeable but not tight clustering.
- Weak: points widely scattered with only a faint trend.
A quick rule: correlation r near ±1 indicates strong linear association; r near 0 indicates weak/none. (Thresholds are subjective.)

4. Outliers and influential points
- Look for points far from the main pattern. They can distort perceived relationship and computed statistics (slope, correlation).

5. Clusters and subgroups
- Separate groups of points may indicate different relationships for different subpopulations.

6. Other patterns to check
- Heteroscedasticity: spread of y values changes with x (unequal variance).
- Gaps or boundaries (possible measurement limits).

Practical steps
1. Plot the points.
2. Visually assess direction, form, strength, outliers, clusters.
3. If appropriate, fit a line (least squares) or curve and check residuals to confirm the chosen form.
4. Optionally compute Pearson’s r for linear strength, and r^2 to measure proportion of variance explained — but remember:
- Correlation measures linear association only.
- A high correlation does not imply causation.
- Outliers and nonlinearity can mislead r.

Examples (visualized mentally)
- Tight cloud slanting up-right: strong positive linear.
- Scattered cloud with slight downward slope: weak negative linear.
- U-shaped pattern: clear nonlinear relationship (not well summarized by correlation).
- Random cloud with no trend: no relationship.

Summary: Identify direction, form, strength, and unusual points; use a fitted line and correlation for linear patterns, but always check for nonlinearity, outliers, subgroups, and avoid inferring causation without further evidence.