8.3 Data Displays: Histograms, Box Plots, & Scatter Plots

The Power of Visual Data Displays

Statistical data displays transform raw numbers into intuitive geometric representations. On the ACCUPLACER Quantitative Reasoning, Algebra, and Statistics (QAS) test, graphical reasoning questions require you to interpret frequency distributions, extract summary statistics from visual plots, compare distributions across multiple cohorts, and analyze bivariate relationships using trend lines.

---

Dot Plots & Frequency Tables

A dot plot displays individual observations along a single continuous number line, where each dot represents one occurrence of a data value. A frequency table aggregates observations into discrete categories or numerical intervals.

Dot Plot Display:
          ●
          ●       ●
  ●       ●   ●   ●
  ●   ●   ●   ●   ●   ●
--|---|---|---|---|---|---
  1   2   3   4   5   6

Analytical Features of Dot Plots & Frequency Tables

• Peaks / Modes: Value columns with the tallest stacks of dots indicate the mode(s).
• Clusters: Groupings where data points gather densely together.
• Gaps: Empty intervals along the axis with zero observations, often separating clusters from isolated outliers.
• Relative Frequency: The proportion or percentage of the total sample ($n$) represented by a given count ($f$): $\text{Relative Frequency} = \frac{f}{n} \quad \text{or} \quad \left(\frac{f}{n} \times 100\%\right)$
• Cumulative Frequency: The running sum of frequencies up to and including the current class interval.

---

Histograms: Grouped Continuous Data

A histogram is a column-based graphical display of continuous quantitative data grouped into non-overlapping, contiguous intervals called classes or bins.

Histograms vs. Bar Graphs: The Fundamental Distinction

Histogram (Touching Numerical Bins):    Bar Graph (Separated Categories):
  | [10-20][20-30][30-40]                 | [ Sedan ]   [ SUV ]   [ Truck ]

Interpreting Distribution Shapes from Histograms

1. Symmetric / Bell-Shaped (Unimodal): Single central peak with tails tapering evenly on both sides (Mean $\approx$ Median $\approx$ Mode).
2. Uniform / Rectangular: All bins have approximately equal frequencies across the entire domain.
3. Right-Skewed (Positively Skewed): Peak is located on the left at lower values; a long tail extends toward higher positive values on the right (Mean $>$ Median).
4. Left-Skewed (Negatively Skewed): Peak is located on the right at higher values; a long tail extends toward lower values on the left (Mean $<$ Median).
5. Bimodal: Two distinct, prominent peaks separated by a trough, indicating two distinct sub-populations.

---

Box-and-Whisker Plots (Box Plots)

A box plot graphically renders the Five-Number Summary $(\text{Minimum}, \; Q_1, \; \text{Median}, \; Q_3, \; \text{Maximum})$ along a single numerical axis.

Anatomy of a Box Plot:
           |------------------- [    IQR    ] -------------------|
        Minimum              Q1       Median        Q3        Maximum
           |------------------|=========|===========|------------|
           |   Bottom 25%     | Lower   | Upper     | Top 25%    |
           |                  |  25%    |  25%      |            |
           |                  |---- Middle 50% -----|            |

Key Structural Principles of Box Plots

1. The Central Box: Extends from the first quartile ($Q_1$) to the third quartile ($Q_3$). The total width of the box is the Interquartile Range ($\text{IQR} = Q_3 - Q_1$), encompassing the middle $50\%$ of the dataset.
2. The Median Line: Drawn vertically through the interior of the box at $Q_2$. If the median line sits to the left of center within the box, the distribution is right-skewed; if to the right, it is left-skewed.
3. The Whiskers: Lines extending from $Q_1$ down to the minimum non-outlier and from $Q_3$ up to the maximum non-outlier.
4. The Quarter-Rule (25% Rule): Each of the four visual sections—lower whisker, lower box segment, upper box segment, and upper whisker—contains exactly $25\%$ of all observations, regardless of physical width. A wider whisker or box segment does not contain more data points; it signifies that that $25\%$ of the data is more widely spread.

Comparing Parallel Box Plots

When analyzing parallel box plots representing two or more groups on the same scale:

• Compare Center: Contrast the positions of the median lines.
• Compare Middle Dispersion: Contrast the box lengths (IQRs).
• Compare Overall Dispersion: Contrast the total whisker-to-whisker spans (Ranges).
• Compare Skewness: Contrast whisker lengths and median positioning.

---

Scatter Plots & Bivariate Data Analysis

A scatter plot displays paired bivariate quantitative measurements $(x, y)$ on a Cartesian coordinate plane to investigate mathematical and statistical associations between two numerical variables.

• Independent (Explanatory) Variable ($x$): Plotted along the horizontal axis.
• Dependent (Response) Variable ($y$): Plotted along the vertical axis.

The Three Dimensions of Bivariate Association

Positive Linear (/)          Negative Linear (\)          No Correlation (:::)
  y                            y                            y
  |        *                   |   *                        |   *   *    *
  |     * *                    |     * *                    | *   *   *    *
  |   * *                      |       * *                  |   *   *   *
  | *                          |          *                 | *   *    *   *
  +-----------> x              +-----------> x              +-----------> x

---

The Line of Best Fit (Trend Line) & Linear Regression

A line of best fit (or trend line) is a straight line drawn through a scatter plot that minimizes the overall distances between data points and the line:

$\hat{y} = mx + b$

• $\hat{y}$ (Predicted Value): The estimated value of the response variable for a chosen input $x$.
• Slope ($m$): The predicted rate of change—the amount $\hat{y}$ increases or decreases for every $1$-unit increase in $x$.
• $y$-Intercept ($b$): The baseline value of $\hat{y}$ when $x = 0$ (meaningful only when $x = 0$ is physically realistic within the study context).

Computing Slope from a Scatter Plot

1. Select two points directly on the drawn line of best fit: $(x_1, y_1)$ and $(x_2, y_2)$. (Do not use arbitrary isolated raw data points that do not fall on the line.)
2. Calculate the slope using the standard formula: $m = \frac{y_2 - y_1}{x_2 - x_1}$.

Understanding Residuals

A residual is the vertical distance between an actual observed data point ($y$) and the predicted value ($\hat{y}$) from the line of best fit:

$\text{Residual} = y - \hat{y} = \text{Observed Value} - \text{Predicted Value}$

• A positive residual ($y > \hat{y}$) indicates the observed point lies above the trend line (model underestimated).
• A negative residual ($y < \hat{y}$) indicates the observed point lies below the trend line (model overestimated).

---

Interpolation vs. Extrapolation

• Interpolation: Estimating a response value $\hat{y}$ for an input $x$ that falls within the domain of observed data values ($x_{\text{min}} \le x \le x_{\text{max}}$). Predictions made via interpolation are generally reliable when correlation is strong.
• Extrapolation: Estimating a response value $\hat{y}$ for an input $x$ that falls outside the domain of observed data values ($x < x_{\text{min}}$ or $x > x_{\text{max}}$). Extrapolations are hazardous and often completely invalid because relationships rarely remain strictly linear across extreme ranges.

---

Crucial Conceptual Principle: Correlation Does Not Imply Causation

One of the most heavily emphasized concepts on standardized statistical exams is that statistical correlation does not establish a causal relationship.

Confounding (Lurking) Variables

Two variables may exhibit a strong positive correlation because both are simultaneously influenced by a third, unmeasured confounding variable.

• Classic Example: Monthly sales of ice cream and monthly rates of drowning incidents share a strong positive correlation. Eating ice cream does not cause drowning. The confounding variable is ambient outdoor temperature (summer heat drives both swimming activity and ice cream consumption).

Establishing Causation

Causation can only be proven through controlled, randomized experiments where an explanatory variable is actively manipulated while holding all other extraneous factors constant.

---

Step-by-Step Multi-Step Worked Examples

Worked Example 1: Analyzing Commute Times from a Frequency Histogram

A municipal transit authority surveys $80$ commuters and records daily travel times (in minutes) in a frequency table and histogram:

Question 1: What percentage of surveyed commuters travel for at least $30$ minutes?

• Qualifying bins ($t \ge 30$): $[30, 45)$, $[45, 60)$, and $[60, 75)$.
• Total qualifying count $= 28 + 12 + 4 = 44\text{ commuters}$.
• Percentage $= \frac{44}{80} \times 100\% = 55.0\%$.

Question 2: In which class interval does the median commute time fall?

• For $n = 80$, the median position is between ranks $40$ and $41$.
• The cumulative frequency reaches $36$ at the end of the second bin and $64$ at the end of the third bin.
• Therefore, ranks $40$ and $41$ both fall within the $[30, 45)$ minute interval.

---

Worked Example 2: Comparative Analysis of Parallel Factory Shift Box Plots

A quality control manager compares daily output across $100$ production days for Shift 1 and Shift 2:

• Shift 1 Five-Number Summary: $(\text{Min}=120, \; Q_1=160, \; \text{Median}=190, \; Q_3=210, \; \text{Max}=240)$
• Shift 2 Five-Number Summary: $(\text{Min}=140, \; Q_1=175, \; \text{Median}=205, \; Q_3=220, \; \text{Max}=235)$

Comparative Analysis:

1. Center: Shift 2 has a higher median daily output ($205\text{ units}$) than Shift 1 ($190\text{ units}$).
2. Middle Spread: Shift 1 has an $\text{IQR} = 210 - 160 = 50\text{ units}$, whereas Shift 2 has an $\text{IQR} = 220 - 175 = 45\text{ units}$. Shift 2 is slightly more consistent across its central $50\%$.
3. Overall Span: Shift 1 has a total range of $240 - 120 = 120\text{ units}$, while Shift 2 has a range of $235 - 140 = 95\text{ units}$.
4. Quartile Comparisons: Exactly $50\%$ of Shift 2's days produced $205\text{ units}$ or more, whereas Shift 1 only achieved $210\text{ units}$ or more on $25\%$ of its days.

---

Worked Example 3: Bivariate Scatter Plot Regression and Fuel Economy Prediction

An automotive engineer analyzes vehicle weight ($x$, in thousands of pounds) and highway fuel economy ($y$, in miles per gallon) across $30$ vehicle models. The calculated line of best fit is:

$\hat{y} = -4.2x + 48.6$

1. Interpret the Slope ($m = -4.2$): For every additional $1,000\text{ pounds}$ of vehicle weight ($1\text{ unit increase in } x$), the predicted highway fuel economy decreases by $4.2\text{ miles per gallon}$.
2. Interpret the $y$-Intercept ($b = 48.6$): A theoretical vehicle weighing $0\text{ pounds}$ would have a predicted fuel economy of $48.6\text{ mpg}$. (Note: While algebraically present, a $0\text{-pound}$ vehicle is physically impossible, so this intercept serves only as a mathematical anchor.)
3. Interpolate Fuel Economy for a $3,500\text{-pound}$ Vehicle ($x = 3.5$): $\hat{y} = -4.2(3.5) + 48.6 = -14.7 + 48.6 = 33.9\text{ mpg}$
4. Evaluate Residual for an Actual Vehicle: If an actual test vehicle weighing $3,500\text{ pounds}$ achieved $36.0\text{ mpg}$: $\text{Residual} = y - \hat{y} = 36.0 - 33.9 = +2.1\text{ mpg}$ The vehicle performed $2.1\text{ mpg}$ better than predicted by the regression line.

---

Common Pitfalls & ACCUPLACER Exam Traps

1. Equating Box Segment Width to Sample Size: A wider whisker or box quartile segment does NOT contain more observations; each segment represents exactly $25\%$ of the total count. Wider segments merely indicate higher dispersion in that quartile.
2. Calculating Slope from Raw Scattered Points: Always compute the slope of a trend line using points directly on the drawn line, not raw scatter points that happen to lie nearby.
3. Extrapolating Beyond Realistic Physical Limits: Using a linear model to predict fuel economy for an $80,000\text{-pound}$ semi-truck using passenger car regression would predict negative fuel economy, demonstrating the danger of extrapolation.
4. Concluding Causality from Correlation: Never select an answer choice claiming that an observational scatter plot proves one variable causes another.