Bivariate Data & Scatter Plot Interpretation
Bivariate data involves pairs of numerical observations $(x, y)$ measured on the same individual or subject to determine if a relationship exists.
- Independent (Explanatory) Variable ($x$): Plotted on the horizontal axis. Represents the controlling or predictor factor.
- Dependent (Response) Variable ($y$): Plotted on the vertical axis. Represents the outcome being measured.
- Direction:
- Positive Correlation: As $x$ increases, $y$ tends to increase (slopes upward from bottom-left to top-right).
- Negative Correlation: As $x$ increases, $y$ tends to decrease (slopes downward from top-left to bottom-right).
- Zero Correlation: Random cloud of points with no discernable linear trend.
- Form: Linear (points cluster along a straight line) vs. Non-linear (points cluster along an exponential, quadratic, or cyclical curve).
- Strength: Strong (points tightly clustered close to the trend line), Moderate, or Weak (points loosely scattered around the trend).
Pearson's Product-Moment Correlation Coefficient ($r$)
Pearson's $r$ provides a rigorous numerical metric measuring the strength and direction of linear association between two quantitative variables.
- $r = +1.0$: Perfect positive linear correlation.
- $+0.7 \le r < +1.0$: Strong positive linear correlation.
- $+0.3 \le r < +0.7$: Moderate positive linear correlation.
- $0 < r < +0.3$: Weak positive linear correlation.
- $r = 0$: No linear correlation (data points are randomly scattered or related non-linearly).
- $-0.3 < r < 0$: Weak negative linear correlation.
- $-0.7 < r \le -0.3$: Moderate negative linear correlation.
- $-1.0 < r \le -0.7$: Strong negative linear correlation.
- $r = -1.0$: Perfect negative linear correlation.
Properties of $r$: $r$ is a dimensionless number with no units. Multiplying or adding constants to either variable does not alter $r$. $r$ detects only linear relationships; a strong curved parabolic relationship will yield an $r \approx 0$!
The Centroid Mean Point $(\bar{x}, \bar{y})$ & Line of Best Fit
A line of best fit models the linear trend, enabling estimations of $y$ from values of $x$.
Every valid line of best fit MUST pass through the mean point (centroid) of the dataset: $$(\bar{x}, \bar{y}) = \left(\frac{\sum x}{n}, \; \frac{\sum y}{n}\right)$$
Bivariate scatter plot with positive linear trend, regression line of best fit, and emphasized mean point centroid $(\bar{x}, \bar{y})$.
Prediction Reliability & Correlation vs. Causation
When utilizing regression models to forecast outcomes in real-world contexts (Criterion D), mathematicians must evaluate two fundamental risks.
- Interpolation: Predicting values of $y$ for an $x$-value that lies within the range of the collected data ($x_{\min} \le x \le x_{\max}$).
Reliability: Generally reliable and valid, especially when $|r|$ is strong. - Extrapolation: Predicting values of $y$ for an $x$-value that lies outside the range of collected data ($x < x_{\min}$ or $x > x_{\max}$).
Reliability: Highly unreliable! We have no evidence that the linear relationship continues beyond the observed data range (e.g. human height vs age does not continue into adulthood).
A strong correlation ($r \approx 1$) between variable $X$ and variable $Y$ does NOT prove that $X$ causes $Y$.
- Lurking (Confounding) Variables: A third unmeasured variable drives both (e.g. sales of ice cream and rates of sunburn correlate strongly because warm sunny weather drives both).
- Spurious Correlation: Purely coincidental mathematical association without real causal link.
Worked Example 1: Line of Best Fit via Mean Point
A physics student measures the force $x$ (in Newtons) applied to a spring and the resulting extension $y$ (in mm). The summary statistics for 8 data points are:
$$\sum x = 40, \quad \sum y = 96, \quad \sum x^2 = 240, \quad \sum y^2 = 1200, \quad \sum xy = 530$$
a) Calculate the coordinates of the mean point $(\bar{x}, \bar{y})$.
b) If the gradient of the line of best fit is $m = 2.15$, find the equation of the regression line.
c) Estimate the extension of the spring under a force of $6\text{ N}$ and state whether this is interpolation or extrapolation.
Step-by-step Solution:
Part a: Mean point coordinates:
$\bar{x} = \frac{\sum x}{n} = \frac{40}{8} = 5\text{ N}$.
$\bar{y} = \frac{\sum y}{n} = \frac{96}{8} = 12\text{ mm}$.
Mean point: $(\bar{x}, \bar{y}) = (5, 12)$. [A1]
Part b: Equation of regression line:
The line passes through $(5, 12)$ with slope $m = 2.15$: [M1]
$y - \bar{y} = m(x - \bar{x}) \implies y - 12 = 2.15(x - 5)$
$y - 12 = 2.15x - 10.75 \implies y = 2.15x + 1.25$. [A1]
Part c: Prediction for $x = 6\text{ N}$:
$y = 2.15(6) + 1.25 = 12.9 + 1.25 = 14.15\text{ mm}$. [A1]
Since forces ranged from $1$ to $8\text{ N}$, $x = 6\text{ N}$ lies comfortably inside the data range. Therefore, this prediction is an interpolation and is considered reliable. [Criterion C & D: Communication & Context]
Worked Example 2: Extrapolation Critique & Causation Analysis
A biologist models tree trunk diameter $d$ (in cm) as a function of age $t$ (in years) for trees aged between $5$ and $25$ years. The regression equation is $d = 1.4t + 8.2$, with $r = 0.94$.
a) Interpret the meaning of the gradient $1.4$ in the context of the problem.
b) Predict the trunk diameter of a 60-year-old tree and evaluate the validity of this prediction.
c) Explain why strong correlation ($r = 0.94$) does not mean age is the sole cause of trunk diameter.
Step-by-step Solution:
Part a: Contextual interpretation of gradient:
The gradient $1.4$ indicates that for each additional year of growth, the tree trunk diameter increases by an average of $1.4\text{ cm}$. [A1: Criterion C]
Part b: Prediction for $t = 60$:
$d = 1.4(60) + 8.2 = 84 + 8.2 = 92.2\text{ cm}$. [A1]
Validity evaluation: This prediction is an extrapolation because $t = 60$ lies far beyond the observed upper limit ($25$ years). Biological growth typically decelerates as trees mature (following a logistic curve rather than linear growth). Therefore, this linear extrapolation is unreliable and likely overestimates true trunk diameter. [R1: Criterion D]
Part c: Correlation vs. Causation:
While age is directly related to growth, other confounding factors such as soil nutrient levels, annual rainfall, sunlight availability, and pest infestation also strongly influence growth rate. A strong correlation reflects association, but does not imply that age operates in isolation. [R1: Criterion D]
Tiered Practice Exercises
10 problems advancing from scatter plot interpretation and Pearson's r categorization to mean point regressions, extrapolation reliability, and confounding variable investigations.
Describe the direction and strength of linear correlation indicated by each value of Pearson's $r$:
a) $r = 0.88$
b) $r = -0.42$
c) $r = 0.05$
d) $r = -0.96$
View Solution & Markscheme
Method & Marks:
a) Strong positive linear correlation. [A1]
b) Moderate negative linear correlation. [A1]
c) Very weak or no linear correlation. [A1]
d) Very strong negative linear correlation. [A1]
For a dataset of 10 student study hours ($x$) and exam scores ($y$), $\sum x = 80$ and $\sum y = 650$.
a) Calculate the coordinates of the centroid mean point $(\bar{x}, \bar{y})$.
b) Explain why any valid line of best fit must pass through this point.
View Solution & Markscheme
Method & Marks:
a) $\bar{x} = \frac{80}{10} = 8\text{ hours}$. $\bar{y} = \frac{650}{10} = 65\text{ marks}$.
Centroid: $(\bar{x}, \bar{y}) = (8, 65)$. [A1]
b) The centroid represents the center of mass of the bivariate distribution. In least-squares regression, minimizing the sum of squared vertical residuals guarantees that the regression line passes through $(\bar{x}, \bar{y})$. [R1]
The regression equation relating weekly study time $x$ (hours, $2 \le x \le 12$) to exam score $y$ (%) is $y = 4.5x + 35$.
a) Predict the exam score for a student who studies $8\text{ hours}$ per week.
b) State the $y$-intercept and explain its contextual meaning.
View Solution & Markscheme
Method & Marks:
a) $y = 4.5(8) + 35 = 36 + 35 = 71\%$. [A1]
b) The $y$-intercept is $35$. In context, it represents the predicted baseline exam score of a student who studies $0\text{ hours}$. [A1]
A line of best fit passes through the mean point $(14, 52)$ and has a gradient of $-1.8$.
a) Write the equation of the line of best fit in form $y = mx + c$.
b) Estimate the value of $y$ when $x = 20$.
View Solution & Markscheme
Method & Marks:
a) Using $y - \bar{y} = m(x - \bar{x})$: [M1]
$y - 52 = -1.8(x - 14) \implies y - 52 = -1.8x + 25.2$
$y = -1.8x + 77.2$. [A1]
b) For $x = 20$: $y = -1.8(20) + 77.2 = -36 + 77.2 = 41.2$. [A1]
A scientist measures temperature $T$ ($^\circ$C) and gas pressure $P$ (kPa) for temperatures between $20^\circ\text{C}$ and $80^\circ\text{C}$. The regression line is $P = 0.35T + 95$.
a) Predict the pressure at $50^\circ\text{C}$ and classify this as interpolation or extrapolation.
b) Predict the pressure at $180^\circ\text{C}$ and evaluate the reliability of this estimate.
View Solution & Markscheme
Method & Marks:
a) $P = 0.35(50) + 95 = 17.5 + 95 = 112.5\text{ kPa}$. [A1]
This is interpolation because $50^\circ\text{C}$ lies inside the experimental range $[20, 80]$, making it reliable. [A1]
b) $P = 0.35(180) + 95 = 63 + 95 = 158\text{ kPa}$.
This is extrapolation because $180^\circ\text{C}$ is far outside the range. The gas container may rupture or the gas may undergo phase change at high temperature, making this prediction highly unreliable. [R1: Criterion D]
The heights ($x$ cm) and foot lengths ($y$ cm) of 6 adults are recorded:
$(160, 24), (165, 25), (170, 26), (175, 27), (180, 28), (190, 30)$.
a) Calculate $\bar{x}$ and $\bar{y}$.
b) Find the gradient of the line connecting $(160, 24)$ and $(190, 30)$.
c) Show that the centroid $(\bar{x}, \bar{y})$ lies exactly on this line.
View Solution & Markscheme
Method & Marks:
a) $\bar{x} = \frac{160+165+170+175+180+190}{6} = \frac{1040}{6} = 173.33\text{ cm}$.
$\bar{y} = \frac{24+25+26+27+28+30}{6} = \frac{160}{6} = 26.67\text{ cm}$. [A1]
b) Gradient $m = \frac{30 - 24}{190 - 160} = \frac{6}{30} = 0.2$. [A1]
Equation through $(160, 24)$: $y - 24 = 0.2(x - 160) \implies y = 0.2x - 8$.
c) Test centroid: $0.2(173.33) - 8 = 34.67 - 8 = 26.67 = \bar{y}$. $\blacksquare$ [R1]
A study finds a strong correlation ($r = 0.91$) between the number of firefighters sent to a fire ($x$) and the financial damage caused by the fire ($y$). The city mayor concludes: "To reduce fire damage, we must send fewer firefighters."
Identify the major flaw in the mayor's logic and name the confounding variable involved.
View Solution & Markscheme
Method & Marks:
The mayor committed the fallacy of assuming correlation implies causation. Firefighters do not cause fire damage. [R1: Criterion C]
The confounding (lurking) variable is the severity/size of the fire. Larger, more severe fires naturally result in greater financial damage AND require dispatching more firefighters. Reducing firefighters would actually increase damage. [R1: Criterion D Evaluation]
A dataset has standard deviations $s_x = 4$ and $s_y = 6$. The Pearson correlation coefficient is $r = 0.85$, and the mean point is $(10, 25)$.
a) Use the regression gradient formula $m = r\left(\frac{s_y}{s_x}\right)$ to calculate the exact slope of the line of best fit of $y$ on $x$.
b) Determine the complete linear regression equation.
c) Explain why the line of best fit of $x$ on $y$ has a different slope from the line of $y$ on $x$.
View Solution & Markscheme
Method & Marks:
a) $m = r\left(\frac{s_y}{s_x}\right) = 0.85\left(\frac{6}{4}\right) = 0.85 \times 1.5 = 1.275$. [M1][A1]
b) $y - 25 = 1.275(x - 10) \implies y - 25 = 1.275x - 12.75$
$y = 1.275x + 12.25$. [A1]
c) The regression of $y$ on $x$ minimizes vertical residuals (errors in $y$), whereas the regression of $x$ on $y$ minimizes horizontal residuals (errors in $x$, with slope $m' = r\frac{s_x}{s_y} = 0.85\times \frac{4}{6} \approx 0.567$). The two lines only coincide when correlation is perfect ($r = \pm 1$). [R1: Rigorous conceptual insight]
Consider the points: $(-2, 4), (-1, 1), (0, 0), (1, 1), (2, 4)$.
a) Show that the relationship between $x$ and $y$ is deterministic and exact: $y = x^2$.
b) Calculate $\sum x$, $\sum y$, and $\sum xy$.
c) Show that Pearson's correlation coefficient $r = 0$, and explain why this demonstrates that $r$ detects only linear relationships.
View Solution & Markscheme
Method & Marks:
a) For every point, $(-2)^2 = 4, (-1)^2 = 1, 0^2 = 0, 1^2 = 1, 2^2 = 4 \implies y = x^2$. [A1]
b) $\sum x = -2 - 1 + 0 + 1 + 2 = 0 \implies \bar{x} = 0$.
$\sum y = 4 + 1 + 0 + 1 + 4 = 10 \implies \bar{y} = 2$.
$\sum xy = (-2)(4) + (-1)(1) + (0)(0) + (1)(1) + (2)(4) = -8 - 1 + 0 + 1 + 8 = 0$. [M1][A1]
c) Covariance numerator is $\sum (x - \bar{x})(y - \bar{y}) = \sum xy - n\bar{x}\bar{y} = 0 - 0 = 0 \implies r = 0$.
Even though there is a perfect quadratic relationship ($y = x^2$), $r = 0$ because Pearson's $r$ measures strictly linear association. Symmetrical curvature cancels out positive and negative slopes, proving that $r = 0$ does not imply no relationship! [R1: Criterion C Communication]
Criterion B & D Modelling Investigation: An environmental agency tracks greenhouse gas emissions $x$ (in megatons) and mean summer temperature anomaly $y$ ($^\circ$C) over a 30-year period ($x$ ranges from $320$ to $420\text{ Mt}$). The regression model is $y = 0.015x - 4.2$, with $r = 0.88$.
a) Interpret the slope $0.015$ in terms of temperature increase per $100\text{ Mt}$ increase in emissions.
b) A climate model projects temperature anomalies if emissions reach $700\text{ Mt}$. Calculate the predicted temperature and critically evaluate two mathematical limitations of using this linear model for this prediction.
View Solution & Markscheme
Method & Marks:
Part a: An increase of $100\text{ Mt}$ corresponds to a temperature increase of $100 \times 0.015 = 1.5^\circ\text{C}$. [A1]
Part b: Prediction for $x = 700\text{ Mt}$:
$y = 0.015(700) - 4.2 = 10.5 - 4.2 = +6.3^\circ\text{C}$. [A1]
Critical evaluation of limitations:
1. Severe Extrapolation: $700\text{ Mt}$ is far beyond the observed range $[320, 420]$. Linear extrapolation assumes unchecked straight-line behavior without physical boundaries.
2. Non-linear Feedback Loops: Climate systems possess complex non-linear tipping points (e.g. melting permafrost, ocean carbon saturation, albedo loss) where temperature responses deviate sharply from simple linear regression. [R1: Criterion D Contextual Reflection]
Correlation & Linear Regression Reference Sheet
Pearson's $r$ Range
$$-1 \le r \le 1$$
Measure of linear correlation.Centroid Mean Point
$$(\bar{x}, \bar{y}) = \left(\frac{\sum x}{n}, \; \frac{\sum y}{n}\right)$$
Every line of best fit passes through here.Regression Line ($y$ on $x$)
$$y - \bar{y} = m(x - \bar{x})$$
Where $m = r\left(\frac{s_y}{s_x}\right)$.$y$-Intercept
$$c = \bar{y} - m\bar{x}$$
From point-slope form.Interpolation
$$x_{\min} \le x \le x_{\max}$$
Reliable within data range.Extrapolation
$$x < x_{\min} \text{ or } x > x_{\max}$$
Unreliable outside data range.Top 5 Common MYP Correlation Pitfalls
- Drawing a Line of Best Fit That Misses $(\bar{x}, \bar{y})$: In MYP exams, any hand-drawn line of best fit that does not pass through the plotted centroid $(\bar{x}, \bar{y})$ loses marks automatically!
- Assuming Correlation Proves Causation: A high correlation coefficient $r$ only shows that two variables move together. It does NOT prove that $X$ causes $Y$. Always consider lurking/confounding variables.
- Treating Extrapolations as Reliable Facts: Predicting outside the observed data range assumes the same linear relationship continues indefinitely, which is often biologically or physically impossible.
- Assuming $r = 0$ Means No Relationship: $r = 0$ means there is no linear relationship. A perfect curved parabola ($y = x^2$) or circle can yield $r = 0$. Always inspect the scatter plot visually!
- Switching $x$ and $y$ in Regression: The regression line of $y$ on $x$ is used to predict $y$ from $x$. It is NOT the same line as the regression of $x$ on $y$.
Self-Grading Mini Test
5 quiz questions on scatter plots, Pearson's r, centroid lines of best fit, and regression predictions.
Which value of Pearson's correlation coefficient indicates the strongest linear relationship?
Explanation: Strength depends on the absolute magnitude $|r|$. $|-0.92| = 0.92$, which is greater than $0.85$ and $0.75$. The negative sign merely indicates the direction.
Which point must every linear regression line of best fit pass through?
Explanation: By mathematical derivation of least squares, the line of best fit always passes through the mean point $(\bar{x}, \bar{y})$.
In a study with observed $x$-values ranging from $10$ to $50$, predicting $y$ for $x = 65$ is an example of:
Explanation: Because $x = 65$ lies outside the observed interval $[10, 50]$, it is extrapolation and is considered unreliable.
If two variables have a correlation of $r = 0.95$, what does this mathematically prove?
Explanation: A high value of $r$ proves strong linear association, but does NOT prove causation or rule out lurking confounding variables.
The regression equation is $y = 3.2x - 4.5$. What is the predicted value of $y$ when $x = 5$?
Explanation: $y = 3.2(5) - 4.5 = 16.0 - 4.5 = 11.5$.