9.4 Bivariate Data, Scatter Plots & Line of Best Fit

9.4 Bivariate Data, Scatter Plots & Line of Best Fit

Quick Answer: Bivariate data involves pairs of numerical measurements $(x, y)$ gathered from the same subjects. Graphed on a coordinate grid, a scatter plot reveals the relationship between the independent variable ($x$, horizontal axis) and the dependent variable ($y$, vertical axis). Patterns of association are described by direction (positive as $x \uparrow, y \uparrow$; negative as $x \uparrow, y \downarrow$; or no association), form (linear vs. non-linear), and strength. An informal line of best fit passes through the center of the point cloud, balancing points above and below. Its equation $y = mx + b$ allows predictions: slope ($m$) is the unit rate of change, $y$-intercept ($b$) is the initial baseline value when $x = 0$, interpolation estimates within the data range, and extrapolation predicts outside the data range.

---

Bivariate Numerical Data and the Scatter Plot

Under Florida B.E.S.T. benchmark MA.8.DP.1.1, students transition from univariate data (analyzing one variable in isolation) to bivariate numerical data—investigating the relationship between two quantitative variables measured on the same individual or object.

The Coordinate Plane Framework

To visualize bivariate relationships, data pairs $(x_i, y_i)$ are plotted as individual points on a coordinate plane:

• Independent (Explanatory) Variable ($x$): Plotted along the horizontal $x$-axis. This variable represents the input, cause, or explanatory condition (e.g., hours spent studying, ambient outdoor temperature, vehicle age).
• Dependent (Response) Variable ($y$): Plotted along the vertical $y$-axis. This variable represents the output, effect, or measured response (e.g., test score percentage, electricity consumption, vehicle resale price).

Graph Construction Best Practices

1. Axis Scaling: Each axis must have uniform, evenly spaced intervals. The scale on the $x$-axis does not need to match the scale on the $y$-axis, but each must accommodate the full range of its respective variable.
2. Axis Breaks (Truncation): If data begins far from zero, a jagged zigzag break symbol ($/ /$) may be used near the origin to avoid wasting visual space, provided it is clearly marked.

---

Patterns of Association

Under Florida B.E.S.T. benchmark MA.8.DP.1.2, students must analyze scatter plots to identify four core structural characteristics:

1. Direction of Association

• Positive Association: As the independent variable $x$ increases, the dependent variable $y$ also tends to increase. The points follow an uphill trajectory from bottom-left to top-right ($m > 0$).
  • Real-World Example: High school GPA and SAT scores; height and arm span.
• Negative Association: As the independent variable $x$ increases, the dependent variable $y$ tends to decrease. The points follow a downhill trajectory from top-left to bottom-right ($m < 0$).
  • Real-World Example: Elevation above sea level and boiling point of water; vehicle mileage and market value.
• No Association: Changes in variable $x$ show no discernible upward or downward relationship with variable $y$. Points appear as a randomly scattered cloud with roughly horizontal orientation ($m \approx 0$).
  • Real-World Example: Shoe size and scores on a statewide history examination.

2. Form of Association (Linear vs. Non-Linear)

• Linear Association: The plotted points cluster closely along the straight path of a straight line.
• Non-Linear (Curvilinear) Association: The plotted points follow a clearly defined curved pattern (e.g., parabolic curve, exponential growth/decay, or logarithmic leveling off).
  • Real-World Example: Age and reaction speed (curved: increases in youth, peaks in young adulthood, declines in senior years).

3. Strength of Association

• Strong Association: Points cluster very tightly along a linear corridor or curve with minimal scatter.
• Moderate Association: Points exhibit a clear general trend, but with moderate dispersion away from the center line.
• Weak Association: Points are widely dispersed, making the underlying trend faint and difficult to discern.

4. Structural Features: Clusters and Outliers

• Clusters: Distinct groupings or sub-clouds of points separated by wide empty spaces. Clusters often indicate that the sample contains distinct sub-populations (e.g., fuel efficiency plotted against weight for compact sedans vs. heavy pickup trucks).
• Bivariate Outliers: Points that deviate markedly from the general pattern of association established by the rest of the data. A bivariate outlier might have normal $x$ and $y$ values individually, but an atypical pairing.
  • Example: A $12\text{-year-old}$ car with only $5,000\text{ miles}$ on its odometer is a bivariate outlier in a plot of vehicle age vs. mileage.

$\begin{array}{|l|l|l|l|} \hline \textbf{Type of Association} & \textbf{Direction} & \textbf{Visual Trajectory} & \textbf{Real-World Scenario} \\ \hline \text{Positive Linear} & \text{Positive } (x \uparrow, y \uparrow) & \text{Uphill straight corridor} & \text{Study hours vs. Exam score} \\ \hline \text{Negative Linear} & \text{Negative } (x \uparrow, y \downarrow) & \text{Downhill straight corridor} & \text{Vehicle age vs. Resale value} \\ \hline \text{Non-Linear (Curved)} & \text{Variable direction} & \text{U-shaped or exponential curve} & \text{Vehicle speed vs. Braking distance} \\ \hline \text{No Association} & \text{None } (m \approx 0) & \text{Random diffuse cloud} & \text{Shoe size vs. Reading comprehension} \\ \hline \end{array}$

---

The Informal Line of Best Fit (Trend Line)

Under Florida benchmark MA.8.DP.1.3, students informally fit a straight line to bivariate measurement data that exhibits a linear association.

Golden Rules for an Informal Line of Best Fit

When evaluating or drawing a line of best fit by eye, the line must adhere to three strict criteria:

1. Follows the Central Corridor: The line must mirror the general slope, direction, and angle of the data cloud.
2. Balances the Points: The line should pass through the center of the cloud, having approximately an equal number of data points above and below the line.
3. Minimizes Distances (Residuals): The vertical distances from the data points to the line should be as small as possible.

[!IMPORTANT]

What a Line of Best Fit is NOT:

• It does NOT need to pass through the origin $(0, 0)$: Forcing a line through $(0, 0)$ when the data trend intercepts elsewhere distorts the entire model.
• It does NOT need to touch any specific raw data points: A trend line models the aggregate trend of the entire population, not individual points.
• It does NOT simply connect the first and last points: Connecting $(x_{\min}, y_{\min})$ to $(x_{\max}, y_{\max})$ creates an inaccurate line if either endpoint is an outlier.

---

Linear Model Equations ($y = mx + b$) in Context

Under Florida benchmark MA.8.DP.1.4, students write and interpret the equation of a linear trend line in slope-intercept form:

$y = mx + b$

Step 1: Identifying Two Points on the Trend Line

To write the linear equation, select two points that lie directly on the line of best fit. These do not have to be original data points; they are geometric coordinates on the line $(x_1, y_1)$ and $(x_2, y_2)$.

Step 2: Calculating and Interpreting Slope ($m$)

The slope represents the unit rate of change:

$m = \frac{y_2 - y_1}{x_2 - x_1} = \frac{\Delta y}{\Delta x} = \frac{\text{Change in Dependent Variable } y}{\text{Change in Independent Variable } x}$

• In Real-World Context: The slope tells you "for every $1\text{-unit}$ increase in the independent variable $x$, the dependent variable $y$ is predicted to change by $m$ units."
• Example: In $y = 3.5x + 20$ (where $x = \text{hours}$ and $y = \text{cost in dollars}$), the slope $m = 3.5$ means that the cost increases by $$3.50$ for each additional hour of service.

Step 3: Determining and Interpreting the $y$-Intercept ($b$)

The $y$-intercept is the point $(0, b)$ where the trend line crosses the vertical $y$-axis:

$b = y_1 - m(x_1)$

• In Real-World Context: The $y$-intercept represents the initial baseline value of the response variable $y$ when the independent variable $x = 0$.
• Example: In $y = 3.5x + 20$, the $y$-intercept $b = 20$ represents a base initial fee of $$20.00$ charged before any hours are worked.
• Contextual Validity Check: Always ask whether $x = 0$ is physically meaningful. In a scatter plot of human height vs. weight, the $y$-intercept at $x = 0\text{ cm}$ height has no physical meaning, but it mathematically anchors the linear model within the observed domain.

---

Making Predictions: Interpolation vs. Extrapolation

Once a linear equation is established, it serves as a predictive mathematical model by substituting given values for $x$ (or $y$):

1. Interpolation (Reliable)

Interpolation is the process of predicting a value within the domain of the observed data range ($x_{\min} \le x \le x_{\max}$).

• Reliability: High. Because data has already verified that a linear relationship exists throughout this interval, interpolated estimates are generally trustworthy and accurate.
• Example: If data was collected for study times between $1.0$ and $5.0\text{ hours}$, predicting the test score for a student studying $3.5\text{ hours}$ is interpolation.

2. Extrapolation (High Risk)

Extrapolation is the process of predicting a value outside the domain of the observed data range ($x < x_{\min}$ or $x > x_{\max}$).

• Reliability: Extremely Low / Hazardous. Extrapolation presumes that the linear pattern continues indefinitely into uncharted territory. In the real world, biological, mechanical, and economic limits eventually cause linear trends to plateau, bend, or reverse.
• Example: Using a child's growth rate from ages $2$ to $8$ ($y = 6x + 75\text{ cm}$) to predict their height at age $40$ yields $y = 6(40) + 75 = 315\text{ cm}$ ($10.3\text{ feet tall}$)—an absurd result caused by blind extrapolation.

---

Correlation vs. Causation

A paramount tenet of scientific data analysis under Florida B.E.S.T. standards is:

$\mathbf{\text{Association (Correlation) does NOT imply Causation!}}$

Just because two numerical variables display a strong positive or negative association does not prove that changes in variable $x$ directly cause changes in variable $y$. A strong association is frequently caused by a third, unmeasured confounding (lurking) variable.

• Famous Classic Example: There is a very strong positive correlation between monthly retail ice cream sales ($x$) and the number of swimming pool drownings ($y$).
  • Does buying ice cream cause drowning? No.
  • The Confounding Variable: Summer outdoor temperature. Hot weather causes more people to buy ice cream AND causes more people to swim in pools.

---

Step-by-Step Worked Example: Fitting and Using a Linear Trend Line

Scenario: A Florida state park ranger collects data on ambient air temperature in degrees Fahrenheit ($x$) and the daily volume of bottled water sold at the park pavilion ($y$):

Step 1: Verify the association. As temperature increases, bottled water sales increase consistently. The points form an uphill straight path: positive linear association.

Step 2: Select two points on the trend line to find slope. Suppose the informal line of best fit passes cleanly through $(72, 140)$ and $(96, 310)$: $m = \frac{y_2 - y_1}{x_2 - x_1} = \frac{310 - 140}{96 - 72} = \frac{170}{24} \approx 7.08 \approx 7.1$ Contextual Interpretation: For every $1^\circ\text{F}$ increase in daily temperature, the park pavilion is predicted to sell approximately $7.1$ additional bottles of water.

Step 3: Solve for the $y$-intercept ($b$). Using point $(72, 140)$ and $m = 7.08$: $y = mx + b \implies 140 = 7.08(72) + b \implies 140 = 509.76 + b \implies b = 140 - 509.76 = -369.8$ Model Equation: $y = 7.1x - 370$.

Step 4: Make an interpolated prediction. Predict water sales on an $84^\circ\text{F}$ afternoon ($84$ lies within the observed $72-96$ range $\implies$ interpolation): $y = 7.08(84) - 369.8 = 594.7 - 369.8 = 224.9 \approx 225\text{ bottles}$ This interpolation aligns smoothly with surrounding raw data points ($80^\circ\text{F} \implies 190$; $85^\circ\text{F} \implies 230$).

---

Common Exam Traps & Misconceptions

[!WARNING]

Exam Trap 1: Forcing the Line of Best Fit Through the Origin $(0, 0)$

A ubiquitous student error is anchoring the trend line ruler at $(0, 0)$ before drawing through the points. Unless the data specifically dictates that $y = 0$ when $x = 0$ (e.g., zero hours worked yields zero pay), forcing the line through $(0, 0)$ severely distorts the slope and invalidates all predictions.

[!WARNING]

Exam Trap 2: Inverting the Slope Ratio ($\frac{\Delta x}{\Delta y}$ Instead of $\frac{\Delta y}{\Delta x}$)

When computing rate of change from a scatter plot, students frequently compute horizontal run divided by vertical rise ($\frac{\Delta x}{\Delta y}$). Slope is universally defined as the change in the vertical dependent variable divided by the change in the horizontal independent variable: $m = \frac{\text{Rise}}{\text{Run}} = \frac{y_2 - y_1}{x_2 - x_1}$

[!WARNING]

Exam Trap 3: Confusing Association with Proven Causation

Multiple-choice test items on the FAST frequently tempt candidates with causal claims (e.g., "Because the correlation between study time and test scores is 0.92, increasing study time directly forces scores to increase.") Association proves correlation only. Causation can strictly be proven through controlled randomized scientific experiments, never through observational scatter plots alone.