Resources

An applied mathematics researcher fits a linear regression model y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x to evaluate the braking distance yy (in feet) of passenger vehicles traveling at initial speed xx (in miles per hour). (a) State the homoscedasticity assumption required for ordinary least squares (OLS) inference on this linear model. (b) Rather than evaluating residual variance directly, determine the theoretical physical dependence of braking distance on velocity under constant braking friction, and identify the resulting systematic violation in the model's residual pattern as speed xx increases.

Answer and steps

Final answer

(a) The variance of the error terms must remain constant across all levels of the explanatory variable: Var⁡(εi∣xi)=σ2\operatorname{Var}(\varepsilon_i \mid x_i) = \sigma^2 for all ii. (b) By the work-energy theorem (Wfriction=ΔKW_{\text{friction}} = \Delta K), stopping distance scales quadratically with speed (d∝v2d \propto v^2); fitting a linear model forces the residuals ei=yi−y^ie_i = y_i - \hat{y}_i to exhibit systematic non-random curvature (fanning out and quadratic bowing), violating both the linearity assumption E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0 and homoscedasticity.

Steps

01

Theoretical Relationship (Work-Energy Theorem)

Part (a): Stating the Homoscedasticity Assumption

In an ordinary least squares (OLS) linear regression model: yi=β0+β1xi+εiy_i = \beta_0 + \beta_1 x_i + \varepsilon_i the random error term εi\varepsilon_i accounts for deviations from the true regression line.

  • The Homoscedasticity Assumption: The variance of the error terms given the explanatory variable must be constant for all observations: Var⁡(εi∣xi)=σ2for all i\operatorname{Var}(\varepsilon_i \mid x_i) = \sigma^2 \quad \text{for all } i
  • This means that regardless of whether the vehicle travels at a low speed or a high speed, the spread of the random errors around the true relationship must remain the same finite value σ2\sigma^2.

Part (b): Theoretical Derivation and Residual Pattern

To find how braking distance physically depends on speed, consider a vehicle of mass mm brought to a stop from an initial speed vv by a constant frictional braking force Ffriction=μmgF_{\text{friction}} = \mu m g:

  • By the work-energy theorem, the work done by the braking force equals the change in kinetic energy: Wnet=ΔKW_{\text{net}} = \Delta K −Ffriction⋅d=0−12mv2-F_{\text{friction}} \cdot d = 0 - \frac{1}{2}m v^2
  • Substituting Ffriction=μmgF_{\text{friction}} = \mu m g: μmg⋅d=12mv2  ⟹  d=v22μg\mu m g \cdot d = \frac{1}{2}m v^2 \implies d = \frac{v^2}{2\mu g}

Thus, braking distance dd depends quadratically on initial speed: d∝v2d \propto v^2

02

Systematic Violations in the Residual Pattern

When a straight line y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x is fitted to data that inherently follows a quadratic curve (y∝x2y \propto x^2):

  • Quadratic Bowing (Non-Linearity): The residuals ei=yi−y^ie_i = y_i - \hat{y}_i do not scatter randomly around zero. Instead, they follow a systematic parabolic curve: residuals tend to be positive at low speeds, negative at intermediate speeds, and strongly positive at high speeds. This directly violates the assumption that the conditional expectation of the errors is zero: E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0
  • Fanning Out (Heteroscedasticity): In physical testing, higher absolute stopping distances compound small variations in reaction time, road conditions, and tire grip, causing errors to scale with distance. Combined with the mis-specified model, the residual spread increases markedly (fans out) as speed xx increases, violating homoscedasticity.

Pitfall: A common mistake is to treat heteroscedasticity as merely a numerical error; here, it is driven by a fundamental model misspecification—using a linear equation to represent a quadratic physical law.

Typical errors

OverviewACEpath Teacher's Diagnosis

01 The student confuses the homoscedasticity assumption with the assumption of zero conditional mean of errors, stating that the expected value rather than the variance must be constant.

Sample Student Work

Reviewed by ACEpath Teacher
Step-by-Step Review
Walk Through It with Your Teacher
Partially correct
  1. Step 1

    The expected value of the error terms given xx must be constant and equal to zero across all levels of speed: E(ε∣x)=0\mathbb{E}(\varepsilon \mid x) = 0.

    Correct

    Var⁡(εi∣xi)=σ2\operatorname{Var}(\varepsilon_i \mid x_i) = \sigma^2 for all ii (the conditional variance of the error terms is constant across all levels of xx).

    You stated the zero conditional mean assumption E(ε∣x)=0\mathbb{E}(\varepsilon \mid x) = 0, which governs model unbiasedness and exogeneity. Homoscedasticity specifically requires the conditional variance of the errors to be constant across all predictor levels, written as Var⁡(ε∣x)=σ2\operatorname{Var}(\varepsilon \mid x) = \sigma^2.

  2. Step 2

    Work done by friction equals initial kinetic energy: F⋅d=(12)m⋅v2⟹F \cdot d = \left(\frac{1}{2}\right) m \cdot v^2 ⟹ Solving for distance yields d=(m2F)⋅v2d = \left(\frac{m}{2F}\right) \cdot v^2, showing distance is proportional to v2v^2.

  3. Step 3

    Fitting a linear model instead of a quadratic model causes the residuals to exhibit a non-random U-shaped curve as speed increases.

WHAT YOU DID WELL
Compliments from Your Teacher

You correctly used the work-energy theorem to deduce that stopping distance is proportional to the square of initial velocity.

You correctly recognized that model misspecification creates a non-random, curved residual pattern across speed levels.

THE ERROR
Where Your ACEpath Teacher Spotted the Mistake
OLS Assumption Confusion (Mean vs. Variance)

no

WHY IT HAPPENED
ACEpath Teacher's Diagnosis

You confused the homoscedasticity assumption (constant conditional variance of the error terms) with the exogeneity or strict exogeneity assumption (zero conditional mean of the error terms).

REASONING PATH
Your ACEpath teacher’s suggested approach
Homoscedasticity vs. Zero Conditional Mean
  • Homoscedasticity specifies constant spread: Var⁡(ε∣x)=σ2\operatorname{Var}(\varepsilon \mid x) = \sigma^2. Zero conditional mean specifies center: E(ε∣x)=0\mathbb{E}(\varepsilon \mid x) = 0.

  • When asked for homoscedasticity, verify that you are writing a condition about Var⁡(⋅)\operatorname{Var}(\cdot), not E(⋅)\mathbb{E}(\cdot).

CORRECTED PATH
ACEpath Teacher's Demonstration
  1. Identify the root words: 'homo' means same, and 'scedastic' refers to scatter or dispersion.

  2. Express the constant dispersion mathematically using conditional variance: Var⁡(εi∣xi)=σ2\operatorname{Var}(\varepsilon_i \mid x_i) = \sigma^2 for all observations ii.

Final answer

(a) Var⁡(εi∣xi)=σ2\operatorname{Var}(\varepsilon_i \mid x_i) = \sigma^2 for all ii; (b) d∝v2d \propto v^2, leading to systematic U-shaped residual curvature and non-constant residual spread.

NEXT STEPS
Final Takeaways from Your Teacher
  • Review the classical Gauss-Markov assumptions to clearly distinguish expectations of errors from variances and covariances of errors.

  • Practice sketching residual plots that illustrate pure non-linearity versus pure heteroscedasticity (fanning).

OverviewACEpath Teacher's Diagnosis

02 The student assumes braking distance is proportional to velocity rather than velocity squared, concluding the linear model is theoretically correct and will produce purely random residuals.

Sample Student Work

Reviewed by ACEpath Teacher
Step-by-Step Review
Walk Through It with Your Teacher
Partially correct
  1. Step 1

    a) The variance of the error terms given speed xx is constant across all observations: Var⁡(ε∣x)=σ2\operatorname{Var}(\varepsilon \mid x) = \sigma^2.

  2. Step 2

    b) Braking distance is proportional to initial velocity, d∝vd \propto v, because stopping time increases linearly with speed.

    Correct

    By the work-energy theorem (Wfriction=ΔKW_{\text{friction}} = \Delta K, or μmgd=12mv2\mu m g d = \frac{1}{2}m v^2), braking distance is proportional to speed squared: d∝v2d \propto v^2.

    While stopping time is indeed proportional to initial speed under constant deceleration (t=v/at = v/a), stopping distance is the integral of velocity over time (d=vˉt=12vt=v22ad = \bar{v}t = \frac{1}{2}vt = \frac{v^2}{2a}), which makes distance scale quadratically (d∝v2d \propto v^2), not linearly.

  3. Step 3

    c) Since distance depends linearly on speed, the linear regression model correctly captures the relationship and the residuals will be randomly scattered around zero with no systematic pattern as xx increases.

    Correct

    Fitting a linear line y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x to a quadratic physical law (y∝x2y \propto x^2) causes residuals ei=yi−y^ie_i = y_i - \hat{y}_i to exhibit systematic quadratic curvature (bowing) and fanning out as xx increases, violating both linearity E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0 and homoscedasticity.

    Because this conclusion relies on your incorrect linear scaling from step 2, you missed the systematic parabolic bowing and fanning out that appear when fitting a straight line to an inherently quadratic curve.

WHAT YOU DID WELL
Compliments from Your Teacher

You correctly formulated and mathematically expressed the homoscedasticity assumption for ordinary least squares regression.

You recognized that constant braking friction implies constant deceleration.

THE ERROR
Where Your ACEpath Teacher Spotted the Mistake
Physical Law Misapplication in Regression Specification

Conflating stopping time (t∝vt \propto v) with stopping distance (d∝v2d \propto v^2).

WHY IT HAPPENED
ACEpath Teacher's Diagnosis

You related stopping distance directly to stopping time without accounting for the fact that average speed during braking also increases linearly with initial speed, or by omitting the work-energy theorem (Fd=12mv2F d = \frac{1}{2}m v^2). This led to misspecifying the underlying deterministic relationship as linear instead of quadratic.

REASONING PATH
Your ACEpath teacher’s suggested approach
Work-Energy Scaling
  • When a constant retarding force FF acts over distance dd to stop an object of mass mm with initial speed vv, the work done equals the initial kinetic energy: Fd=12mv2F d = \frac{1}{2}m v^2, so d∝v2d \propto v^2.

  • Always verify physical model specifications using conservation laws or kinematics (v2=v02+2adv^2 = v_0^2 + 2ad) before choosing a linear regression model.

Residual Bowing from Underfitting
  • Fitting a first-order linear model y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x to a quadratic process y=cx2y = c x^2 forces the residuals e=y−y^e = y - \hat{y} into a distinct parabolic arc (positive at extremes, negative in the middle).

  • Look for parabolic or curved trends in residual vs. fitted plots as an immediate indicator of missing higher-order terms.

CORRECTED PATH
ACEpath Teacher's Demonstration
  1. Apply the work-energy theorem: the work done by friction equals the change in kinetic energy: Ffriction⋅d=12mv2F_{\text{friction}} \cdot d = \frac{1}{2} m v^2.

  2. Substitute constant frictional force Ffriction=μmgF_{\text{friction}} = \mu m g: μmgd=12mv2  ⟹  d=v22μg\mu m g d = \frac{1}{2} m v^2 \implies d = \frac{v^2}{2\mu g}, meaning d∝v2d \propto v^2.

  3. Compare the quadratic relationship to the fitted linear model y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x: a straight line fitted across a parabola underestimates values at the lowest and highest speeds and overestimates values in the middle.

  4. Identify the residual pattern: the residuals ei=yi−y^ie_i = y_i - \hat{y}_i show a systematic U-shaped quadratic curve (bowing) and fanning out with speed, violating both E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0 and constant variance.

Final answer

(a) Var⁡(εi∣xi)=σ2\operatorname{Var}(\varepsilon_i \mid x_i) = \sigma^2 for all ii. (b) By the work-energy theorem, d=v22μg∝v2d = \frac{v^2}{2\mu g} \propto v^2; fitting a linear model forces residuals to exhibit systematic quadratic bowing and fanning out, violating both linearity E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0 and homoscedasticity.

NEXT STEPS
Final Takeaways from Your Teacher
  • Review kinematic equations and the work-energy theorem to verify power-law dependencies before specifying regression functions.

  • Practice sketching expected residual plots when a linear model is fitted to non-linear data (such as quadratic, exponential, or saturation curves).

OverviewACEpath Teacher's Diagnosis

03 The student confuses the violation of linearity with heteroscedasticity, incorrectly concluding that the residual variance systematically fans out rather than forming a curved pattern.

Sample Student Work

Reviewed by ACEpath Teacher
Step-by-Step Review
Walk Through It with Your Teacher
Partially correct
  1. Step 1

    The variance of the error terms is constant for all values of initial speed: Var⁡(ε∣x)=σ2\operatorname{Var}(\varepsilon \mid x) = \sigma^2.

  2. Step 2

    Work-energy theorem: F⋅d=12mv2F \cdot d = \frac{1}{2} m v^2, so braking distance dd is proportional to v2v^2.

  3. Step 3

    Because distance grows quadratically while the model is linear, the residuals fan outward with increasing variance as speed xx increases, violating homoscedasticity.

    Correct

    Fitting a linear model y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x to a quadratic physical relationship y∝x2y \propto x^2 causes the residuals ei=yi−y^ie_i = y_i - \hat{y}_i to show a systematic curved (parabolic/bowed) pattern, violating the linearity assumption E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0.

    When a linear model is fitted to quadratic data, the mean of the residuals depends systematically on xx as E[e∣x]=cx2−(β0+β1x)\mathbb{E}[e \mid x] = c x^2 - (\beta_0 + \beta_1 x). This creates a distinct U-shaped or inverted U-shaped parabolic curve (a violation of the linearity assumption), not a fan-shaped spread in variance.

WHAT YOU DID WELL
Compliments from Your Teacher

You clearly and formally stated the constant variance condition Var⁡(ε∣x)=σ2\operatorname{Var}(\varepsilon \mid x) = \sigma^2.

You used the work-energy theorem to deduce the correct quadratic relationship d∝v2d \propto v^2.

THE ERROR
Where Your ACEpath Teacher Spotted the Mistake
Confusing functional form misspecification (non-linearity) with heteroscedasticity

You identified the discrepancy between linear and quadratic growth but attributed it to changing variance rather than systematic curvature in the residual mean.

WHY IT HAPPENED
ACEpath Teacher's Diagnosis

You conflated a violation of zero conditional mean (linearity, E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0, where points follow a curved trajectory) with a violation of constant variance (homoscedasticity, Var⁡(ε∣x)=σ2\operatorname{Var}(\varepsilon \mid x) = \sigma^2, where the spread widens or narrows). Functional form misspecification primarily bends the centerline of the residuals into a curve.

REASONING PATH
Your ACEpath teacher’s suggested approach
Residual Pattern vs. Model Assumption
  • A systematic curve (U-shape or inverted U-shape) in a residual plot indicates model misspecification / non-linearity (E[ε∣x]≠0\mathbb{E}[\varepsilon \mid x] \neq 0). A funnel or fan shape indicates heteroscedasticity (Var⁡(ε∣x)\operatorname{Var}(\varepsilon \mid x) depends on xx).

  • Check the centerline of residuals first: if the average residual shifts systematically above and below zero across xx, diagnose non-linearity.

CORRECTED PATH
ACEpath Teacher's Demonstration
  1. Compare the true conditional expectation E[y∣x]=kx2\mathbb{E}[y \mid x] = k x^2 with the fitted line y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x.

  2. Calculate the expected residual: E[e∣x]=E[y−y^∣x]=kx2−(β0+β1x)\mathbb{E}[e \mid x] = \mathbb{E}[y - \hat{y} \mid x] = k x^2 - (\beta_0 + \beta_1 x), which is quadratic in xx.

  3. Conclude that the residuals exhibit a systematic parabolic bowed curve across xx, directly violating the linearity assumption E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0.

Final answer

(a) Var⁡(εi∣xi)=σ2\operatorname{Var}(\varepsilon_i \mid x_i) = \sigma^2 for all ii. (b) By the work-energy theorem, d∝v2d \propto v^2; fitting a linear model forces the residuals to exhibit a systematic parabolic curve (quadratic bowing), violating the assumption of linearity E[ε∣x]=0\mathbb{E}[\varepsilon \mid x] = 0.

NEXT STEPS
Final Takeaways from Your Teacher
  • Sketch a residual plot for a straight line fitted through a set of points on a parabola to see the resulting U-shape.

  • Practice distinguishing between patterns that indicate non-zero conditional mean (curvature) and those that indicate non-constant variance (fanning or funneling).

Lock it in

Similar question

01

A researcher fits y=β0+β1x+εy = \beta_0 + \beta_1 x + \varepsilon. State the mathematical condition for the Gauss-Markov assumption of 'no autocorrelation' among error terms.

Answer

Cov⁡(εi,εj∣xi,xj)=0\operatorname{Cov}(\varepsilon_i, \varepsilon_j \mid x_i, x_j) = 0 for all i≠ji \neq j. The error terms of distinct observations must be uncorrelated.

02

Suppose the true physical law is y=αx2y = \alpha x^2, but a researcher fits y^=β0+β1x\hat{y} = \beta_0 + \beta_1 x over x∈[0,10]x \in [0, 10]. At the endpoints (x≈0x \approx 0 and x≈10x \approx 10), will the residuals ei=yi−y^ie_i = y_i - \hat{y}_i tend to be positive or negative? Explain briefly.

Answer

Positive. A straight line fitted to an upward-opening parabola intersects the curve at two intermediate points, lying below the curve at the low and high extremes. Hence yi>y^iy_i > \hat{y}_i, making residuals positive.

03

In a physics experiment measuring the period TT of a simple pendulum against length LL, the theoretical formula is T=2πL/gT = 2\pi \sqrt{L/g}. If a student fits a simple linear regression T=β0+β1L+εT = \beta_0 + \beta_1 L + \varepsilon, describe the curvature of the residual plot eie_i versus LL.

Answer

An inverted U-shape (concave downward). Because the true curve T∝LT \propto \sqrt{L} is strictly concave, a fitted straight line underestimates TT in the middle and overestimates TT at the ends, producing negative residuals at the ends and positive residuals in the middle.

Loading workspace…