College Statistics Quiz: Residual Analysis And Influence

20 questions · exam conditions

0:00

Residual Analysis And InfluenceQuestion 1 of 20

A model predicting final exam scores from midterm scores has a standard deviation of the residuals ( $s_e$ ) of 8.5 points. Upon inspection, a data entry error is found for one student. The corrected data point has a much smaller residual than the erroneous point did. How will this correction likely affect the value of $R^2$ and the value of $s_e$ ?

R^2

will increase and

s_e

will increase.

R^2

will decrease and

s_e

will decrease.

R^2

will increase and

s_e

will decrease.

R^2

will decrease and

s_e

will increase.

← Back to quizzes

College Statistics Quiz

College Statistics Quiz: Residual Analysis And Influence

Practice Residual Analysis And Influence in College Statistics with focused quiz questions that help you check what you know, review explanations, and build confidence with test-style prompts.

What this quiz covers

This quiz focuses on Residual Analysis And Influence, giving you a quick way to practice the rules, question types, and explanations that matter most for College Statistics.

How to use this quiz

Try each quiz question before looking at the correct answer. Use the explanations to review missed ideas, then come back to similar questions until the pattern feels familiar.

All questions

Question 1

$R^2$ will increase and $s_e$ will increase.
$R^2$ will decrease and $s_e$ will decrease.
$R^2$ will increase and $s_e$ will decrease. (correct answer)
$R^2$ will decrease and $s_e$ will increase.

Explanation: The standard deviation of the residuals,

s_e

, measures the typical size of the prediction errors. Correcting a point so that it has a much smaller residual means that the overall sum of squared errors (SSE) for the model will decrease. Since

s_e = \sqrt{SSE/(n-2)}

, a smaller SSE will lead to a smaller

s_e

. The coefficient of determination,

R^2

, measures the proportion of variance in the response variable that is explained by the model. It is calculated as

1 - SSE/SST

. Since SSE decreases and the total sum of squares (SST) remains the same, the fraction SSE/SST gets smaller, and thus

R^2

increases. The model now explains the data better.

Question 2

A marketing analyst models the number of website clicks as a function of daily advertising budget. An initial linear model yields an $R^2$ value of 0.35. After investigating the data, the analyst removes a single data point that was determined to be a recording error and re-fits the model. The new model has an $R^2$ value of 0.82. Which of the following is the best description of the point that was removed?

A point with low leverage and a small residual that was close to the mean of both variables.
A point that was highly influential, likely having both high leverage and a large residual. (correct answer)
A point with low leverage but an extremely large residual, representing an unusual number of clicks for a typical ad budget.
A point with high leverage that fell very close to the original regression line, confirming the trend.

Explanation: When you encounter a dramatic change in

R^2

after removing a single data point, you're dealing with concepts of leverage, residuals, and influence in regression analysis. The key insight is understanding what could cause such a substantial improvement in model fit. The jump from

R^2 = 0.35

R^2 = 0.82

represents a massive improvement in explanatory power—from explaining 35% to 82% of the variance. For one data point to have such dramatic impact, it must be highly influential, meaning it substantially affects the regression line's slope and position. Influential points typically combine two characteristics: high leverage (unusual x-values, far from the mean) and large residuals (the point doesn't fit the pattern of other data). This describes answer B perfectly. Let's examine why the other options fail: A is incorrect because low-leverage points near the mean with small residuals have minimal impact on regression—removing such a point wouldn't dramatically change

R^2

. C describes an outlier in the y-direction only, but without high leverage, even large residuals don't drastically alter the regression line. D describes a point with high leverage that confirms the trend (small residual), which actually strengthens the original model rather than hurting it. Study tip: Remember that influence = leverage × residual effect. When you see dramatic changes in regression statistics after removing one point, look for the combination of both high leverage AND large residuals. Points that are unusual in just one dimension rarely have such extreme influence on model fit.

Question 3

A researcher is investigating a potential influential point in their dataset. They calculate two versions of the regression model: one with the point included and one with it excluded. The first model is $\hat{y} = 10.2 + 2.5x$ with $R^2 = 0.64$ . The second model is $\hat{y} = 12.1 + 1.1x$ with $R^2 = 0.78$ . Which of the following is the most valid conclusion?

The point was not influential because its removal did not greatly affect the model's predictive power.
The point was influential, and its presence in the data was masking a weaker underlying linear relationship.
The point was influential, and its presence in the data was causing the model to overestimate the strength of the linear relationship.
The point was influential, and its presence in the data changed the apparent slope by more than 100%. (correct answer)

Explanation: An influential point is one whose removal causes a large change in the regression model. Here, the slope changed from 2.5 to 1.1, a change of 1.4. Using the original slope as the base, the percentage change is

|(2.5 - 1.1) / 2.5| × 100\% = 56\%. However, this represents a change of more than half the original value, which is substantial. The y-intercept also changed from 10.2 to 12.1. These large changes in both parameters indicate the point was clearly influential. Statement (D) correctly identifies the magnitude of this change as being very large. (A) is incorrect; the changes are quite substantial. (B) and (C) are incorrect because removing the point increased \(R^2

from 0.64 to 0.78, meaning the original point was weakening the apparent relationship, not masking or overestimating it.

Question 4

After performing a linear regression, an analyst produces a normal probability plot (Q-Q plot) of the residuals. The points on the plot form a distinct S-shaped curve. What is the most appropriate interpretation of this plot?

The relationship between the variables is non-linear, and a quadratic model should be considered.
The distribution of the residuals has lighter or heavier tails than a normal distribution, violating the normality assumption. (correct answer)
The distribution of the residuals is skewed, violating the normality assumption.
The variance of the residuals is not constant, indicating that a data transformation may be necessary.

Explanation: When interpreting Q-Q plots (quantile-quantile plots) of regression residuals, you're checking whether the residuals follow a normal distribution—one of the key assumptions of linear regression. The plot compares the quantiles of your residuals against the quantiles of a theoretical normal distribution. A distinct S-shaped curve indicates that your residuals have a different tail behavior than a normal distribution. The "S" pattern occurs because either the residuals have heavier tails (more extreme values than expected) or lighter tails (fewer extreme values than expected) compared to normal. Heavy tails create an S that curves upward at the ends, while light tails create an S that curves downward at the ends. This violates the normality assumption needed for valid statistical inference. Option A is incorrect because non-linearity would show up in a residuals vs. fitted values plot, not in the pattern of a Q-Q plot. The Q-Q plot specifically tests distributional assumptions, not relationship form. Option C confuses S-shaped curves with skewness patterns—skewed distributions typically show curved but not S-shaped patterns in Q-Q plots. Option D describes heteroscedasticity (non-constant variance), which you'd diagnose from residuals vs. fitted plots or scale-location plots, not Q-Q plots. Remember this pattern: S-curves in Q-Q plots = tail problems (heavy or light tails), curved but not S-shaped = skewness, and systematic patterns in residual plots = linearity or variance issues. Focus on what each diagnostic plot is designed to detect.

Question 5

A regression model was created to predict monthly sales based on advertising spending. To check the model assumptions, the residuals were plotted against the order in which the data were collected (by month, from 1 to 36). The plot showed a cyclical pattern, with several consecutive positive residuals followed by several consecutive negative residuals. This pattern suggests a violation of the assumption of:

Normality of errors, because the residuals are not random.
Linearity, because a straight line is not the best fit for the data.
Independence of errors, because the error in one month is related to the errors in previous months. (correct answer)
Constant variance, because the spread of the residuals is not uniform.

Explanation: Plotting residuals against the time or sequence of data collection is the standard method for checking the independence of errors assumption. A distinct pattern, such as the cyclical one described (autocorrelation), indicates that the residuals are not independent. The error in one time period appears to be correlated with the errors in surrounding time periods. This is a common issue in time-series data. (A) Normality is checked with a histogram or Q-Q plot. (B) Linearity is checked with a plot of residuals versus fitted values (or the predictor variable). (D) Constant variance is also checked with a plot of residuals versus fitted values.

Question 6

A biologist is modeling the relationship between the wingspan (in cm) and weight (in g) of a species of bird. After fitting a least-squares regression line, one bird is identified as a highly influential point. If this influential point is removed from the data, which of the following changes is most likely to occur?

A large change in the slope of the regression line and a large change in the standard deviation of the residuals. (correct answer)
A small change in the slope of the regression line but a large change in the y-intercept.
A large change in the range of the wingspan values but a small change in the correlation coefficient.
A small change in both the slope and the y-intercept of the regression line, as all points contribute to the fit.

Explanation: An influential point is defined as a point that, if removed, causes a substantial change in the regression model. This includes the slope, intercept, and/or predictions. Because influential points are typically outliers in the y-direction (large residual) and/or x-direction (high leverage), removing one will dramatically alter the line's fit. This will cause a large change in the slope. Furthermore, because it has a large residual, removing it will eliminate the single largest error from the model, causing a large decrease in the standard deviation of the residuals (

s_e

). (B) is unlikely, as a large change in slope often accompanies a change in intercept. (C) is incorrect because removing an influential point often causes a large change in the correlation coefficient. (D) is the definition of a non-influential point.

Question 7

An environmental scientist studies the relationship between the concentration of a pollutant in a river (in ppm) and the population density of a certain fish species (in fish per cubic meter). The initial analysis yields a correlation coefficient of $r = -0.15$ and a regression line with a slope near zero. However, a review of the data reveals one data point that is a severe outlier: a pristine, protected area with a near-zero pollutant level and an exceptionally high fish density.

Based on the passage, if this single outlier is removed and the analysis is re-run, what is the most probable impact on the correlation coefficient and the slope of the regression line?

The correlation coefficient will become more negative (e.g., -0.60), and the slope will become more negative. (correct answer)
The correlation coefficient will move closer to 0 (e.g., -0.05), and the slope will also move closer to 0.
The correlation coefficient will become positive (e.g., 0.50), and the slope will become positive.
The correlation coefficient and the slope will remain largely unchanged because a single point cannot have a large effect.

Explanation: The outlier is described as having a very low x-value (pollutant level) and a very high y-value (fish density). Let's visualize this. If the other data points show a negative trend (more pollution, fewer fish), this point in the upper-left would pull the regression line up on the left side, making the slope flatter (closer to zero) than it should be. Removing this influential point would allow the line to fit the underlying negative trend of the remaining data more closely. This would cause the slope to become more negative. Consequently, the strength of the linear association would become more apparent, and the correlation coefficient

r

would become stronger in the negative direction (e.g., moving from -0.15 to -0.60).

Question 8

In a regression analysis of college GPA on high school GPA for a large sample of students, one data point is found to have a very high leverage value but a small standardized residual. Which of the following statements is the most accurate description of this student's data point?

The student had a high school GPA far from the average of all students, and their college GPA was much lower than predicted by the model.
The student had a high school GPA that was close to the average of all students, but their college GPA was very different from what the model predicted.
The student had a high school GPA far from the average of all students, but their college GPA was close to the value predicted by the regression line. (correct answer)
The student had both a high school GPA and a college GPA that were close to the average of all students in the sample.

Explanation: High leverage means the point has an extreme value for the explanatory variable (high school GPA), making it far from the mean of the x-values. A small residual means the point's actual y-value (college GPA) is very close to the y-value predicted by the regression line (

\hat{y}

). Therefore, the student's high school GPA was unusual, but their college GPA followed the trend established by the other data points quite well. (A) describes a high-leverage, high-residual point (influential). (B) describes a low-leverage, high-residual point (an outlier, but likely not influential). (D) describes a typical point with low leverage and a small residual.

Question 9

A regression model is used to predict fuel efficiency (MPG) from vehicle weight. The resulting residual plot (residuals vs. fitted values) shows a distinct fan shape, where the vertical spread of the residuals increases as the fitted values increase. This pattern indicates a violation of which regression assumption, and what is the primary consequence?

Linearity; the model systematically over or under-predicts for certain ranges of vehicle weight.
Independence of errors; the prediction error for one vehicle is related to the error for another.
Normality of errors; the p-values from hypothesis tests for the slope will be unreliable.
Constant variance (homoscedasticity); prediction intervals will be inaccurately narrow for heavy vehicles and wide for light vehicles. (correct answer)

Explanation: A fan or funnel shape in a residual plot indicates heteroscedasticity, meaning the variance of the errors is not constant across all levels of the predictor variable. In this case, the model's predictions are less precise (more variable) for higher fitted values (which correspond to lower MPG and heavier vehicles). A primary consequence of this violation is that standard errors are incorrect, making confidence and prediction intervals unreliable. They will be too wide for the low-variance region (light vehicles) and too narrow for the high-variance region (heavy vehicles). (A) is wrong; non-linearity shows a curved pattern. (B) is wrong; non-independence is checked by plotting residuals against time or collection order. (C) is wrong; normality is checked with a histogram or Q-Q plot of residuals, and while severe heteroscedasticity can affect inference, the most direct consequence described relates to the intervals.

Question 10

Which of the following points, if added to a dataset, would have the greatest potential to be an influential point on a least-squares regression line?

A point located at $(\bar{x}, \bar{y})$ , the mean of the explanatory and response variables.
A point with an x-value close to $\bar{x}$ but a y-value that is 3 standard deviations above $\bar{y}$ .
A point with an x-value 3 standard deviations from $\bar{x}$ and a y-value that falls directly on the original regression line.
A point with an x-value 3 standard deviations from $\bar{x}$ and a y-value that produces a large residual. (correct answer)

Explanation: Influence is a product of leverage and discrepancy (residual size). Leverage is determined by how far a point's x-value is from the mean of x (

\bar{x}

). Discrepancy is how far the point's y-value is from the pattern set by the other points (its residual). (A) is the centroid of the data; it has zero leverage and zero residual, and the line must pass through it. (B) has a large residual but low leverage because its x-value is not extreme. It is an outlier but likely not influential. (C) has high leverage but a zero residual (since it's on the line), so it is not influential. It would strengthen the existing relationship. (D) describes a point with both high leverage (extreme x-value) and a large residual (large discrepancy), the classic combination for an influential point.

Question 11

$R^2$ will increase and $s_e$ will increase.
$R^2$ will decrease and $s_e$ will decrease.
$R^2$ will increase and $s_e$ will decrease. (correct answer)
$R^2$ will decrease and $s_e$ will increase.

Explanation: The standard deviation of the residuals,

s_e

, measures the typical size of the prediction errors. Correcting a point so that it has a much smaller residual means that the overall sum of squared errors (SSE) for the model will decrease. Since

s_e = \sqrt{SSE/(n-2)}

, a smaller SSE will lead to a smaller

s_e

. The coefficient of determination,

R^2

, measures the proportion of variance in the response variable that is explained by the model. It is calculated as

1 - SSE/SST

. Since SSE decreases and the total sum of squares (SST) remains the same, the fraction SSE/SST gets smaller, and thus

R^2

increases. The model now explains the data better.

Question 12

In the context of regression diagnostics, what is the primary distinction between an outlier and a high-leverage point?

An outlier is an unusual value in the response (y) variable, whereas a high-leverage point is an unusual value in the explanatory (x) variable. (correct answer)
An outlier always has a large residual and high influence, whereas a high-leverage point always has a small residual and low influence.
An outlier is a data point that is guaranteed to be influential, whereas a high-leverage point has only the potential to be influential.
An outlier is determined by its distance from the regression line, whereas a high-leverage point is determined by its distance from the point $(\bar{x}, \bar{y})$ .

Explanation: This question tests the fundamental definitions. An outlier (in the context of regression) is a point that has a large residual, meaning its y-value is unusual given its x-value. It lies far from the regression line. A high-leverage point is one with an extreme x-value, far from the mean of the x-values. Thus, (A) is the most accurate and direct distinction. (B) is incorrect; an outlier doesn't necessarily have high influence (if its leverage is low), and a high-leverage point can have a large residual and high influence. (C) is incorrect; outliers are not guaranteed to be influential. (D) is partially correct but less precise than (A); an outlier is measured by its vertical distance from the line, while leverage is determined solely by its horizontal distance from the mean of x.

Question 13

A chemist uses linear regression to model the reaction rate (Y) based on temperature (X). The regression equation is $\widehat{\text{Rate}} = 0.5 + 2.1(\text{Temp})$ . The data used for the model had temperatures ranging from 20°C to 50°C. One observation was recorded at a temperature of 40°C with an actual reaction rate of 90.5. What is the residual for this observation, and what can be concluded about the point?

The residual is 1.0; the point lies very close to the regression line and is not an outlier.
The residual is -6.0; the point is a potential outlier but has low leverage.
The residual is 6.0; the point is a potential outlier and has high leverage.
The residual is 6.0; the point is a potential outlier but likely has low to moderate leverage. (correct answer)

Explanation: First, we must calculate the predicted value (

\hat{y}

) for a temperature of 40°C.

\hat{y} = 0.5 + 2.1(40) = 0.5 + 84 = 84.5

. The residual is the actual value minus the predicted value:

y - \hat{y} = 90.5 - 84.5 = 6.0

. A residual of 6.0 may be large enough to consider the point a potential outlier, depending on the standard deviation of the residuals. Leverage is determined by the x-value. The temperature range is 20°C to 50°C. The observation at 40°C is within the upper end of this range but not at an extreme, so it would have moderate leverage, not high leverage. An x-value of 50°C or higher would have high leverage. Therefore, the point has a large residual and low to moderate leverage. (A) miscalculates the residual. (B) gets the sign of the residual wrong. (C) incorrectly identifies the point as having high leverage.

Question 14

A researcher is concerned that the relationship between two variables, X and Y, is not linear. They fit a linear model and examine the residual plot. They then apply a logarithmic transformation to the Y variable, creating a new model of $\log(Y)$ vs. X. Which of the following would provide the strongest evidence that the transformation was appropriate and successful?

The value of the correlation coefficient, r, is closer to 1 or -1 for the transformed model.
The residual plot for the transformed model shows a more random scatter of points around the zero line. (correct answer)
The slope of the regression line for the transformed model is steeper than the slope of the original model.
The sum of the residuals for the transformed model is closer to zero than the sum for the original model.

Explanation: The primary tool for assessing the appropriateness of a regression model's form is the residual plot. If the original model was inappropriate due to non-linearity, its residual plot would show a distinct pattern (e.g., a curve). A successful transformation will correct this issue. Therefore, the strongest evidence for a successful transformation is a new residual plot (for the transformed model) that displays no discernible pattern—a random scatter of points centered on zero. This indicates that the transformed model better captures the underlying relationship. (A) While

R^2

(and thus

|r|

) will likely increase, this is a consequence of a better fit, not the primary diagnostic tool. (C) The slope's value is not a direct measure of model appropriateness. (D) The sum of residuals is always zero for any least-squares model with an intercept, so this is not a useful comparison.

Question 15

The student had a high school GPA far from the average of all students, and their college GPA was much lower than predicted by the model.
The student had a high school GPA that was close to the average of all students, but their college GPA was very different from what the model predicted.
The student had a high school GPA far from the average of all students, but their college GPA was close to the value predicted by the regression line. (correct answer)
The student had both a high school GPA and a college GPA that were close to the average of all students in the sample.

\hat{y}

Question 16

A researcher fits a linear model $\hat{y} = b_0 + b_1x$ . The sum of the residuals, $\sum(y_i - \hat{y}_i)$ , is calculated. Later, it is discovered that a single influential point was accidentally omitted from the analysis. If this point is added to the dataset and the regression is re-run, which of the following statements about the sum of the residuals for the new model is correct?

The new sum of residuals will be equal to the residual of the single point that was added.
The new sum of residuals will be non-zero, and its sign will depend on whether the added point is above or below the original line.
The new sum of residuals will be exactly zero (or reasonably close due to rounding), just as it was for the original model. (correct answer)
The new sum of residuals will be substantially larger than zero because an influential point was added to the dataset.

Explanation: A fundamental mathematical property of any least-squares regression model that includes an intercept term (

b_0

) is that the sum of the residuals is always zero (within rounding error). This property holds regardless of the data points used. When the influential point is added and the model is re-fit, a new regression line is calculated. The residuals are then calculated based on this new line. The sum of these new residuals will still be zero. The distractors prey on the misconception that adding an influential point (which has a large individual residual relative to the old line) will cause the sum of residuals to become non-zero.

Question 17

Linearity; the model systematically over or under-predicts for certain ranges of vehicle weight.
Independence of errors; the prediction error for one vehicle is related to the error for another.
Normality of errors; the p-values from hypothesis tests for the slope will be unreliable.
Constant variance (homoscedasticity); prediction intervals will be inaccurately narrow for heavy vehicles and wide for light vehicles. (correct answer)

Question 18

The residual is 1.0; the point lies very close to the regression line and is not an outlier.
The residual is -6.0; the point is a potential outlier but has low leverage.
The residual is 6.0; the point is a potential outlier and has high leverage.
The residual is 6.0; the point is a potential outlier but likely has low to moderate leverage. (correct answer)

Explanation: First, we must calculate the predicted value (

\hat{y}

) for a temperature of 40°C.

\hat{y} = 0.5 + 2.1(40) = 0.5 + 84 = 84.5

. The residual is the actual value minus the predicted value:

y - \hat{y} = 90.5 - 84.5 = 6.0

Question 19

The relationship between the variables is non-linear, and a quadratic model should be considered.
The distribution of the residuals has lighter or heavier tails than a normal distribution, violating the normality assumption. (correct answer)
The distribution of the residuals is skewed, violating the normality assumption.
The variance of the residuals is not constant, indicating that a data transformation may be necessary.

Question 20

A point with low leverage and a small residual that was close to the mean of both variables.
A point that was highly influential, likely having both high leverage and a large residual. (correct answer)
A point with low leverage but an extremely large residual, representing an unusual number of clicks for a typical ad budget.
A point with high leverage that fell very close to the original regression line, confirming the trend.

Explanation: When you encounter a dramatic change in

R^2

R^2 = 0.35

R^2 = 0.82

R^2