PSAT 8/9 Scatterplots & Data Analysis Cheat Sheet

What You Need to Know

Scatterplots and data analysis questions test whether you can read patterns in messy real-world data, describe relationships, and make (and judge) predictions. On the PSAT $8/9$ , you’re usually doing qualitative reasoning (what the trend is, which model fits, what an outlier does) plus a little linear modeling (line of best fit, slope, intercept, residuals).

Scatterplot essentials

A scatterplot shows paired data: each point represents an input and an output.
You’re looking for association (a relationship) between the variables.
The big four features to describe are:
- Direction: positive, negative, or none
- Form: linear or nonlinear
- Strength: strong or weak
- Outliers: points far from the overall pattern

Linear model essentials

When the pattern is roughly linear, you’ll often use a line of best fit (trend line).

Slope tells the average change in output per change in input (rate).
Intercept is the predicted output when the input is zero (sometimes meaningful, sometimes not).
Predictions from the line are usually best within the data range (interpolation). Outside the range is extrapolation and is riskier.

Why this matters on the test

Common PSAT $8/9$ tasks:

Pick which scatterplot matches a description.
Decide if a linear model is reasonable.
Interpret slope/intercept in context.
Identify or reason about outliers.
Compare two data sets by center/spread or by how tight they cluster around a trend.
Use residuals to judge model fit.

Critical reminder: Correlation is not causation. A strong association does not prove one variable causes the other.

Step-by-Step Breakdown

A. How to analyze any scatterplot (fast + complete)

Check direction
- Upward trend: positive association.
- Downward trend: negative association.
- No clear trend: no association.
Check form (shape)
- Points cluster around a straight trend: linear.
- Curved pattern: nonlinear.
Check strength
- Tight cluster around the trend: strong.
- Wide scatter: weak.
Spot outliers
- Points far from the overall cluster or line.
- Ask: is it a data error, unusual case, or meaningful exception?
Decide if a linear model makes sense
- Linear is reasonable if the pattern is roughly straight and residuals would be small and patternless.

B. How to use a line of best fit (when it’s given or you draw it)

Choose two clear points on the line (not necessarily data points).
Compute slope

$m = \frac{y_2 - y_1}{x_2 - x_1}$

Write the line equation in slope-intercept form

$y = mx + b$

Interpret the slope in words: “For each increase of one unit in input, the output changes by $m$ units (on average).”
Use the line for prediction
- Plug in an input value to estimate output.
- Label answers as approximate.
Decide if prediction is interpolation or extrapolation
- Input within the observed range: interpolation (safer).
- Outside the range: extrapolation (less reliable).

C. How to work with residuals (model error)

Predicted value comes from the line.
Residual is:

$\text{residual} = y_{\text{observed}} - y_{\text{predicted}}$

Interpret sign
- Positive residual: point is above the line.
- Negative residual: point is below the line.
Judge model fit
- Smaller residuals overall = better fit.
- Residuals showing a curve/pattern = linear model might be wrong.

Quick worked micro-example (slope from two points on the line)

Suppose the trend line passes through:

$(2,\,5)$

and

$(6,\,13)$

Slope:

$m = \frac{13 - 5}{6 - 2} = \frac{8}{4} = 2$

Meaning: output increases about $2$ units per $1$ unit increase in input.

Key Formulas, Rules & Facts

Scatterplot interpretation rules

Concept	What to look for	Notes / PSAT traps
Direction	Upward = positive, downward = negative	Don’t confuse steepness with strength.
Form	Linear vs curved	If curved, a line may mislead.
Strength	Tight cluster vs wide scatter	Strength is about scatter around the trend.
Outlier	Point far from pattern	Outliers can distort slope and intercept a lot.
Interpolation vs extrapolation	Inside vs outside data range	Extrapolation is often targeted as “unreliable.”
Correlation vs causation	Association does not imply cause	Look for wording like “causes,” “leads to.”

Linear modeling essentials

Formula / idea	When to use	Notes
Slope	Rate of change between two points	$m = \frac{y_2-y_1}{x_2-x_1}$
Slope-intercept form	Write or read a line equation	$y = mx + b$
Point-slope form (optional tool)	If you know slope + one point	$y - y_1 = m(x - x_1)$
Residual	Compare observed to predicted	$y_{\text{obs}}-y_{\text{pred}}$

“Data analysis” facts that show up with scatterplots

Idea	What you need	Quick notes
Mean/median and outliers	Outliers pull the mean more than the median	Median is resistant; mean is sensitive.
Adding a constant to all values	Shifts the center	Shape/spread stays the same if every value shifts equally.
Multiplying all values by a constant	Stretches/compresses scale	Affects center and spread proportionally.
Units matter	Slope units are “output per input”	Always interpret with units/context.

Reminder about correlation values (if shown): $r$ is between $-1$ and $1$ . Values near $1$ or $-1$ mean strong linear association; values near $0$ mean weak linear association.

Examples & Applications

Example 1: Describe association (direction, form, strength, outliers)

A scatterplot shows points trending upward in a fairly tight band, except one point far below the band.

Direction: positive
Form: linear
Strength: strong (most points close to a line)
Outlier: the far-below point
Likely question twist: “What happens if you remove the outlier?”
- The line of best fit may shift (slope/intercept change), and correlation typically increases.

Example 2: Interpret slope in context

A best-fit line for “hours studied” versus “quiz score” is:

$y = 4x + 60$

Slope interpretation: each additional hour studied predicts about $4$ more points on the quiz.
Intercept interpretation: at $0$ hours, predicted score is $60$ (check if that makes sense in context).

Common question style: “What does the $4$ represent?”

Correct: average increase in quiz score per additional hour studied.

Example 3: Make and judge a prediction (interpolation vs extrapolation)

Using the model:

$y = 4x + 60$

If the data only included $x$ values from $1$ to $5$ hours, predict the score for $x = 4$ :

$y = 4(4) + 60 = 76$

That’s interpolation (safe-ish). Predicting for $x = 20$ would be extrapolation (probably unrealistic).

Example 4: Residuals and “above/below the line”

Suppose the trend line predicts $y = 50$ at a certain input, but the actual point has $y = 58$ .

$\text{residual} = 58 - 50 = 8$

Residual is positive, so the point is above the line.
Interpretation: the model underestimated by $8$ .

Common Mistakes & Traps

Mixing up direction vs strength
- Wrong move: thinking “steeper line means stronger correlation.”
- Why wrong: strength is about how tightly points cluster, not how steep the trend is.
- Fix: ask “how close are points to a line?” not “how steep is it?”
Claiming causation from a scatterplot
- Wrong move: “As $x$ increases, $y$ increases, so $x$ causes $y$ .”
- Why wrong: other variables could drive both.
- Fix: use “is associated with” unless the problem explicitly establishes causation by design.
Using two random data points instead of two points on the line of best fit
- Wrong move: picking any two plotted points to compute slope.
- Why wrong: the best-fit line is an average trend; random points include noise.
- Fix: if a line is drawn, pick two clear points on the line (often grid intersections).
Forgetting units when interpreting slope
- Wrong move: saying “slope is $2$ ” with no meaning.
- Why wrong: slope is “output units per input unit.”
- Fix: say “about $2$ (output units) per $1$ (input unit).”
Misreading the intercept when $x = 0$ isn’t meaningful
- Wrong move: treating the intercept as always realistic.
- Why wrong: some contexts don’t allow $x = 0$ (or it wasn’t observed).
- Fix: check whether $x = 0$ makes sense and whether it’s within the data range.
Trusting extrapolation too much
- Wrong move: assuming the trend continues forever.
- Why wrong: real relationships often change outside the observed range.
- Fix: if asked about reliability, say extrapolated predictions are less reliable.
Residual sign confusion
- Wrong move: thinking residual is predicted minus observed.
- Why wrong: on most PSAT-style questions, residual is defined as observed minus predicted.
- Fix: memorize:

$\text{residual} = \text{observed} - \text{predicted}$

Ignoring outliers’ impact on the line
- Wrong move: assuming one weird point “doesn’t matter.”
- Why wrong: outliers can pull the line and change predictions.
- Fix: consider whether the outlier is extreme in the horizontal direction (often most influential) or vertical direction.

Memory Aids & Quick Tricks

Trick / mnemonic	What it helps you remember	When to use
D-F-S-O	Direction, Form, Strength, Outliers	Describing any scatterplot quickly and completely
“O minus P”	Residual is Observed minus Predicted	Residual questions and above/below line
“Inside = safer”	Interpolation is more reliable than extrapolation	Any prediction reliability question
“Mean moves, median resists”	Outliers affect mean more than median	When a weird point is added/removed
“Slope = units per unit”	Always attach units to slope	Interpreting slope in context

Quick Review Checklist

You can describe a scatterplot using direction, form, strength, and outliers.
You know positive vs negative association and what “no association” looks like.
You can decide whether a linear model is reasonable (roughly straight trend, no obvious curve).
You can compute slope using:

$m = \frac{y_2-y_1}{x_2-x_1}$

You can interpret slope as an average rate of change with units.
You know the line form:

$y = mx + b$

You can label predictions as interpolation (within range) or extrapolation (outside range).
You can compute and interpret residuals:

$\text{residual} = y_{\text{obs}} - y_{\text{pred}}$

You remember: correlation does not imply causation.

You’ve got this: stay systematic (D-F-S-O) and keep your interpretations tied to context and units.