PSAT 10 Data Analysis: Scatterplots, Standard Deviation & Sample Inference

What You Need to Know

You’ll see questions where you must read data displays and make reasonable conclusions (not just compute). This mini-unit hits three big skills:

• Scatterplots & linear models: Describe relationships, interpret a line of best fit, and use/interpret residuals.
• Standard deviation: Compare spread, predict how changes to data affect spread, and recognize when data are more/less variable.
• Sample inference: Decide whether a sample result can be generalized to a population, and interpret estimates + margin of error.

Big idea: You’re often not asked to do heavy calculations; you’re asked to interpret what the numbers/graphs mean and whether a conclusion is justified.

Critical reminder: Correlation/association is not causation. A trend in a scatterplot does not prove one variable causes the other.

Step-by-Step Breakdown

A) Scatterplots: describe, model, interpret

1. Identify variables

   • $x$: explanatory (input)
   • $y$: response (output)

2. Describe the association (SOCS for scatterplots)

   • Strength: strong / moderate / weak (how tightly points cluster)
   • Outliers: points far from the pattern
   • Correlation direction: positive / negative / none
   • Shape/Form: linear vs curved

3. If the trend is roughly linear, use the line of best fit (often provided)

   • Slope $m$: “For each increase of $1$ in $x$, predicted $y$ changes by $m$.”
   • Intercept $b$: predicted $y$ when $x = 0$ (only meaningful if $x = 0$ makes sense in context).

4. Make a prediction

   • Plug into $\hat{y} = mx + b$.

5. Compute/interpret residuals

   • Residual $= y - \hat{y}$.
   • Positive residual: actual is above the line.
   • Negative residual: actual is below the line.
   • Smaller residual magnitude $|y-\hat{y}|$ means a better prediction.

6. Check for extrapolation

   • Predicting outside the observed $x$-range is risky and often labeled “not supported.”

Mini worked example (residual):

• Line: $\hat{y} = 2x + 5$
• At $x = 4$, predicted $\hat{y} = 13$.
• If actual $y = 16$, residual $= 16 - 13 = 3$ (point is above the line by $3$ units).

B) Standard deviation: compare spread and data transformations

1. Know what standard deviation measures

   • Standard deviation $\sigma$ (or $s$) describes the typical distance from the mean $\mu$ (or $\bar{x}$).

2. Compare two sets using center + spread

   • Same mean but larger $\sigma$ ⇒ more variability.
   • Larger range doesn’t always mean larger $\sigma$, but it’s a clue.

3. Recognize the effect of transformations

   • Add/subtract a constant: mean shifts, spread stays the same.
   • Multiply/divide by a constant: mean and standard deviation scale.

Mini worked example (transformation):

• If you convert inches to centimeters: $y = 2.54x$.
• Then $\sigma_y = 2.54\sigma_x$ and $\mu_y = 2.54\mu_x$.

C) Sample inference: when can you generalize?

1. Identify the population and the sample

   • Population: the full group you want conclusions about.
   • Sample: the measured subset.

2. Check sampling method

   • Best: random sample (every member has a chance).
   • Bad: convenience/voluntary response; these create bias.

3. State what can be inferred

   • If random + representative, you can generalize from sample to population.
   • If biased or non-random, conclusions should be limited to the sample.

4. Interpret a margin of error (if given)

   • If an estimate is $\hat{p}$ with margin of error $\text{MOE}$, a typical conclusion is:
     $\text{plausible population proportion} \approx \hat{p} \pm \text{MOE}$

5. Use sample size logic

   • Larger $n$ ⇒ less variability in estimates (usually smaller MOE).
   • Small $n$ ⇒ more variability; conclusions are less precise.

Key nuance: Random assignment (in experiments) supports cause-and-effect. Random sampling supports generalizing to a population. They are different.

Key Formulas, Rules & Facts

Scatterplots & linear models

Standard deviation essentials

Sample inference basics

Examples & Applications

Example 1: Describe a scatterplot (trend + outlier)

A scatterplot of $x$ = hours studied and $y$ = test score shows points rising left to right with tight clustering, but one point far below the cluster.

• Direction: positive association.
• Form: roughly linear.
• Strength: strong.
• Outlier: that low-score point (maybe illness/test anxiety).
  Exam move: Mention the outlier can weaken the model and increase prediction error.

Example 2: Interpret slope and intercept in context

Given $\hat{y} = 1.5x + 40$ where $x$ = number of practice problems and $y$ = quiz score.

• Slope $1.5$: each additional practice problem predicts about $1.5$ more points on the quiz.
• Intercept $40$: predicts $40$ points when $x = 0$ (meaningful only if taking $0$ practice problems is realistic).

Example 3: Residuals + “best fit” reasoning

Two lines are proposed for the same scatterplot.

• Line A gives residuals mostly small (like $-2$ to $2$).
• Line B gives residuals larger (like $-6$ to $6$).
  Conclusion: Line A fits better because it minimizes typical vertical error $|y-\hat{y}|$.

Example 4: Sample inference + margin of error

A random sample of $200$ students finds $\hat{p} = 0.60$ prefer online homework, with $\text{MOE} = 0.05$.

• Plausible population proportion:
  $0.60 \pm 0.05 \Rightarrow [0.55,\,0.65]$
• Interpretation: It’s reasonable to believe about $55\%$ to $65\%$ of all students in that school prefer online homework.
  Trap watch: You can generalize to that school (the population sampled from), not automatically to all students everywhere.

Common Mistakes & Traps

1. Mixing up $x$ and $y$

   • Wrong: interpreting slope as “per unit of $y$.”
   • Fix: slope is always “change in predicted $y$ per change in $x$.”

2. Assuming causation from a trend

   • Wrong: “Because $x$ and $y$ are correlated, $x$ causes $y$.”
   • Fix: correlation could be coincidence or a lurking variable.

3. Treating the intercept as automatically meaningful

   • Wrong: using $b$ when $x = 0$ is outside the data range or impossible.
   • Fix: only interpret $b$ if $x = 0$ is sensible and within/near observed data.

4. Extrapolating far beyond the data

   • Wrong: predicting at $x$ much larger/smaller than shown.
   • Fix: predictions are safest within the observed $x$ interval.

5. Residual sign confusion

   • Wrong: thinking positive residual means point is below the line.
   • Fix: residual $= y - \hat{y}$; if actual $y$ is bigger, the point is above the line.

6. Believing standard deviation measures “average” value

   • Wrong: interpreting $\sigma$ like a typical score instead of a typical distance from the mean.
   • Fix: $\sigma$ is about spread around $\mu$.

7. Forgetting how transformations affect $\sigma$

   • Wrong: saying adding $10$ increases $\sigma$.
   • Fix: adding/subtracting constant shifts center only; multiplying scales spread.

8. Overgeneralizing from a biased sample

   • Wrong: “A poll of volunteers proves the whole school thinks this.”
   • Fix: without random sampling (or at least strong representativeness), inference is weak.

Memory Aids & Quick Tricks

Quick Review Checklist

• You can describe a scatterplot using direction, form, strength, outliers.
• You can interpret $m$ in $\hat{y} = mx + b$ as “per $1$ unit increase in $x$.”
• You can compute residuals with $y - \hat{y}$ and interpret the sign.
• You avoid extrapolation outside the observed $x$-values.
• You know $\sigma$ measures typical distance from the mean, and outliers increase $\sigma$.
• You remember: $x+c$ doesn’t change $\sigma$; $ax$ makes $\sigma$ become $|a|\sigma$.
• You only generalize from sample to population when sampling is random/representative.
• You can interpret $\hat{p} \pm \text{MOE}$ as a plausible population range (if MOE is provided).

You’ve got this—focus on interpreting what the data says and what it doesn’t allow you to claim.