Math 2 Quiz: Outliers And Influential Points
15 questions · exam conditions
0:00
Outliers And Influential PointsQuestion 1 of 15

A researcher fits two regression models to predict housing prices: Model A includes all 50 observations, while Model B excludes 3 potential outliers. Model A has R2=0.71R^2 = 0.71 and RMSE = $15,400. Model B has $R2=0.89R^2 = 0.89 $ and RMSE = $8,200. The excluded points have standardized residuals of -3.2, 2.9, and 3.8. What is the most appropriate next step?

Investigate whether the excluded observations represent a different population or data collection issues before choosing between models.
Use Model A because excluding observations without theoretical justification violates the principle of using all available data for inference.
Adopt Model B immediately since it shows substantially better fit statistics and the excluded points are clear statistical outliers.
Develop Model C using robust regression techniques that can accommodate the outliers while maintaining reasonable prediction accuracy.
← Back to quizzes

Math 2 Quiz

Math 2 Quiz: Outliers And Influential Points

Practice Outliers And Influential Points in Math 2 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 Outliers And Influential Points, giving you a quick way to practice the rules, question types, and explanations that matter most for Math 2.

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

A researcher fits two regression models to predict housing prices: Model A includes all 50 observations, while Model B excludes 3 potential outliers. Model A has R2=0.71R^2 = 0.71 and RMSE = $15,400. Model B has $R2=0.89R^2 = 0.89 $ and RMSE = $8,200. The excluded points have standardized residuals of -3.2, 2.9, and 3.8. What is the most appropriate next step?

  1. Investigate whether the excluded observations represent a different population or data collection issues before choosing between models. (correct answer)
  2. Use Model A because excluding observations without theoretical justification violates the principle of using all available data for inference.
  3. Adopt Model B immediately since it shows substantially better fit statistics and the excluded points are clear statistical outliers.
  4. Develop Model C using robust regression techniques that can accommodate the outliers while maintaining reasonable prediction accuracy.
Explanation: When you encounter regression problems involving outliers and model comparison, you need to balance statistical performance with methodological rigor. The dramatic improvement from Model A to Model B (R2R^2 jumping from 0.71 to 0.89, RMSE dropping from $15,400 to $8,200) suggests the excluded points significantly influence the model, but better fit statistics alone don't justify exclusion. Option A is correct because it follows proper statistical methodology. Before excluding any observations, you must investigate whether they represent legitimate but different cases (like luxury homes in a general housing dataset) or result from data collection errors. Standardized residuals of -3.2, 2.9, and 3.8 indicate potential outliers (typically |z| > 2.5-3.0), but statistical unusualness doesn't automatically mean exclusion is warranted. Option B is wrong because blindly including all data isn't always appropriate—legitimate outliers due to measurement errors or population differences should sometimes be excluded, but only after investigation. Option C is incorrect because it prioritizes fit statistics over methodological soundness. Excluding observations solely based on statistical criteria without understanding why they're unusual can lead to biased results and poor generalizability. Option D jumps to a solution without proper diagnosis. While robust regression techniques are valuable, you first need to understand what's causing the outlying behavior before choosing an analytical approach. Remember: In regression analysis, always investigate before you eliminate. Understanding why observations are unusual is more important than achieving better fit statistics through exclusion.

Question 2

A regression analysis is performed on a dataset with 20 observations. One data point has a leverage value of 0.8, while all other points have leverage values below 0.15. The point with high leverage also has a standardized residual of -0.3. Based on the graph showing standardized residuals versus fitted values, what is the most accurate assessment of this point's influence on the regression model?

  1. The point is highly influential because its leverage exceeds the typical threshold, regardless of its small residual value.
  2. The point is not influential because its standardized residual is small, indicating it fits the overall pattern well despite high leverage.
  3. The point is moderately influential because while it has potential for influence due to extreme x-values, it aligns with the trend. (correct answer)
  4. The point cannot be assessed for influence without additional information about Cook's distance or DFBETAS values.
Explanation: A point with high leverage (0.8 is very high for n=20) has the potential to be influential, but influence depends on both leverage and how well the point fits the pattern. Since the standardized residual is small (-0.3), the point follows the general trend despite being in an extreme position in the x-direction, making it moderately influential rather than highly influential.

Question 3

A regression model predicting salary from years of experience shows that removing observation #15 changes the R2R^2 from 0.76 to 0.84 and reduces the standard error from 8200 to 6800 dollars. Observation #15 has 18 years of experience (dataset range: 1-20 years) and a salary of $45,000 (predicted value was $78,000). How should this observation be handled?

  1. Remove it permanently because it clearly improves model fit statistics and represents an obvious data entry error given the large prediction error.
  2. Retain it because the experience value is within the normal range and the salary difference might reflect legitimate individual circumstances.
  3. Transform the variables before deciding, since the large residual might indicate a non-linear relationship that could accommodate this observation.
  4. Investigate it thoroughly before deciding, as it significantly impacts model performance despite having reasonable predictor values within the data range. (correct answer)
Explanation: When analyzing influential observations in regression, you need to balance statistical impact with substantive reasoning. An influential observation is one that substantially changes your model when removed, but influence alone doesn't determine the appropriate action. Observation #15 is clearly influential—removing it improves R2R^2 from 0.76 to 0.84 and reduces standard error significantly. However, the experience value (18 years) falls well within the dataset range of 1-20 years, making it a legitimate data point rather than an obvious outlier. The large residual ($78,000 - $45,000 = $33,000) indicates this person earns much less than predicted, but this could reflect real circumstances like industry differences, geographic location, career interruptions, or other valid factors not captured in the model. Answer D is correct because thorough investigation is warranted when an observation significantly impacts model performance despite having reasonable predictor values. You should examine the data source, verify accuracy, and understand why this case deviates from the pattern. Answer A is wrong because improved fit statistics alone don't justify removal—you'd be eliminating potentially valid information just because it doesn't fit your model. Answer B is wrong because while the experience value is reasonable, you can't ignore the substantial statistical impact without investigation. Answer C is wrong because transformations address overall relationship patterns, not individual influential observations, and may not resolve this specific case's impact. Remember: influential observations require investigation, not automatic removal. Good regression analysis balances statistical performance with substantive understanding of your data.

Question 4

Two regression analyses are performed on the same dataset: a standard least squares regression and a robust regression method. The standard regression identifies 4 potential outliers with standardized residuals exceeding ±2.5, while the robust regression shows these same points have much smaller residuals (all below ±1.8). What does this pattern most likely indicate?

  1. The discrepancy suggests these points are influential outliers that substantially affect the least squares fit but represent valid data variation. (correct answer)
  2. The standard regression correctly identifies true outliers, while the robust regression inappropriately downweights legitimate extreme observations in the analysis.
  3. The robust regression is superior because it successfully accommodates the outliers and provides more reliable parameter estimates for the dataset.
  4. Both methods are flawed because consistent results across regression techniques are necessary for reliable identification of outlying observations.
Explanation: When comparing standard least squares and robust regression methods, you're examining how different techniques handle potential outliers and their influence on the fitted model. The pattern described here—where points have large standardized residuals (±2.5) in least squares but much smaller residuals (below ±1.8) in robust regression—reveals something important about these data points. In least squares regression, the model fits the majority of the data, making these four points appear as outliers with large residuals. However, when robust regression (which is designed to resist the influence of outliers) is applied, these same points show much smaller residuals, indicating they fit well with a model that isn't being pulled away by their influence. This suggests these points are influential outliers that were substantially affecting the least squares fit, but they likely represent legitimate variation in the data rather than errors or contamination. The robust method provides a fit that accommodates these points better, revealing they weren't truly aberrant—just influential. Option B is incorrect because it assumes the standard regression is the gold standard, when in fact least squares can be misleadingly influenced by leverage points. Option C overstates the case—while robust regression may provide better estimates here, we can't definitively conclude it's "superior" without more context. Option D is wrong because expecting identical results defeats the purpose of using robust methods, which are specifically designed to give different results when outliers are present. Remember: when robust and standard methods disagree about outliers, consider whether the points are influential rather than invalid—robust regression often reveals the true underlying relationship.

Question 5

In a regression analysis studying the relationship between house size (x) and price (y), removing one particular data point causes the slope to change from 150 to 95 dollars per square foot, and the correlation coefficient to change from 0.82 to 0.91. What can be concluded about this data point?

  1. The point is an outlier that weakens the linear relationship and should be removed to improve model accuracy.
  2. The point has high influence on the slope but actually represents a legitimate part of the underlying relationship pattern.
  3. The point is highly influential and removing it reveals a stronger but potentially misleading linear relationship in the remaining data. (correct answer)
  4. The point is not problematic since the correlation improvement indicates it was masking the true strength of the relationship.
Explanation: The large change in slope (from 150 to 95) indicates high influence. The increase in correlation (0.82 to 0.91) shows the point was reducing the apparent linear fit. However, this doesn't automatically justify removal - the point may contain important information about the relationship at extreme values, and removing it could lead to a model that doesn't generalize well to the full range of data.

Question 6

A study examines the relationship between study hours (x) and test scores (y) for 30 students. One student studied for 2 hours and scored 95 points, while the rest studied between 6-15 hours and scored 60-85 points. The correlation with this student included is 0.23, and without this student it becomes 0.67. What is the most accurate characterization?

  1. The student is a beneficial outlier that should be retained because they demonstrate achievable high performance with efficient study methods.
  2. The student is an influential point whose removal reveals the true positive relationship between study time and performance among typical students. (correct answer)
  3. The student represents measurement error and should be excluded because their performance contradicts the expected positive study-score relationship.
  4. The student is not influential because they represent a legitimate data point showing that individual ability can compensate for study time.
Explanation: The student is clearly influential - their inclusion dramatically reduces the correlation from 0.67 to 0.23. They have high leverage (extreme x-value of 2 hours vs 6-15 for others) and don't follow the pattern of the majority. Removing them reveals a strong positive relationship among typical students, but this doesn't necessarily mean they should be excluded from analysis without further investigation.

Question 7

A dataset of 25 observations has Cook's distances calculated for each point. Most points have Cook's D values below 0.08, but three points have values of 0.15, 0.35, and 0.95 respectively. Using the common threshold of 4n\frac{4}{n} for identifying influential points, which conclusion is most appropriate?

  1. Only the point with Cook's D = 0.95 should be considered influential since it exceeds six times the threshold value.
  2. All three points should be investigated as potentially influential since they exceed the calculated threshold of 0.16. (correct answer)
  3. The points with Cook's D = 0.35 and 0.95 are influential, but the point with 0.15 requires further investigation.
  4. None of the points are definitively influential because Cook's D thresholds are only rough guidelines for investigation.
Explanation: The threshold is 4/n = 4/25 = 0.16. All three points (0.15, 0.35, 0.95) are at or very close to this threshold, with two clearly exceeding it. While 0.15 is slightly below 0.16, it's close enough to warrant investigation along with the others. The threshold serves as a guideline for identifying points that merit closer examination.

Question 8

A time series regression model predicting monthly sales includes data from January 2020 through December 2022. April 2020 shows an unusual pattern: very low sales despite normal predictor values, resulting in a large negative residual. This point has moderate leverage (0.12) but high Cook's distance (0.45). Given the temporal context, what is the most appropriate analytical approach?

  1. Remove April 2020 from the analysis since it represents an obvious anomaly that distorts the underlying sales relationship model.
  2. Keep April 2020 unchanged since removing temporal observations violates the assumption of independent and identically distributed residuals.
  3. Investigate whether April 2020 represents a structural break requiring separate models for pre- and post-April 2020 periods.
  4. Retain April 2020 but add a dummy variable to account for the unusual circumstances that likely affected sales patterns. (correct answer)
Explanation: When analyzing influential points in time series regression, you need to balance statistical rigor with the temporal nature of your data. This situation presents a classic outlier with high Cook's distance (0.45 > 0.5 threshold) but moderate leverage, suggesting the April 2020 observation significantly influences your model fit. The correct approach is D) adding a dummy variable. April 2020 likely reflects COVID-19 lockdown effects—a real but temporary external shock that affected sales independently of your normal predictors. A dummy variable captures this systematic deviation while preserving the temporal sequence and allowing your model to account for similar future shocks. Option A is problematic because simply removing observations can bias parameter estimates and ignore valuable information about how external shocks affect sales patterns. Option B misunderstands the independence assumption—while residuals should be independent, this doesn't prohibit adjusting for known external factors that create systematic deviations. Option C suggests structural breaks requiring separate models, but this April 2020 anomaly appears to be a temporary shock rather than a permanent change in the underlying sales relationship. The key insight is distinguishing between random outliers (which might warrant removal) and systematic deviations from identifiable causes (which should be modeled). Since you can reasonably attribute April 2020's unusual pattern to external circumstances, incorporating a dummy variable preserves this information while maintaining model integrity. Study tip: In time series analysis, always consider the real-world context behind unusual observations. Temporary external shocks are best handled with indicator variables rather than data removal or model restructuring.

Question 9

A researcher collected data on the relationship between hours of study per week and final exam scores for 15 students. The correlation coefficient for all 15 students is r=0.72r = 0.72. After removing one student whose data point was (2 hours, 95 points), the correlation coefficient becomes r=0.89r = 0.89. What can be concluded about this removed data point?

  1. It is an outlier that strengthened the overall correlation when included in the dataset
  2. It is an influential point that weakened the overall correlation when included in the dataset (correct answer)
  3. It is neither an outlier nor influential since removing it improved the model fit significantly
  4. It is an outlier but not influential since the correlation remained positive in both cases
Explanation: The removed point (2 hours, 95 points) is influential because removing it substantially changed the correlation from 0.72 to 0.89 (an increase of 0.17). This point represents low study time with high performance, which goes against the general positive trend and weakens the correlation. The point is influential because it has substantial leverage (extreme x-value of 2 hours) and doesn't follow the pattern, thus affecting the regression line significantly.

Question 10

In a study of 20 students, the relationship between hours spent on social media daily and GPA showed a correlation of r=−0.43r = -0.43. When the student with the highest social media usage (8.5 hours) is removed, the correlation becomes r=−0.38r = -0.38. When the student with the lowest GPA (1.8) is removed instead, the correlation becomes r=−0.61r = -0.61. What can be concluded about these two students?

  1. The low GPA student is more influential because removing them strengthened the negative correlation more substantially (correct answer)
  2. The high social media user is more influential than the low GPA student because removing them caused a larger change in correlation
  3. Both students are equally influential since they both caused the correlation coefficient to change when removed
  4. The high social media user is an outlier while the low GPA student is influential, based on their respective impacts
Explanation: When analyzing correlation coefficients, you need to understand how removing individual data points affects the strength and direction of relationships. The key insight is that a more substantial change in correlation magnitude indicates greater influence on the overall pattern. Let's examine what happens when each student is removed. The original correlation is r=−0.43r = -0.43. Removing the high social media user changes it to r=−0.38r = -0.38, which is a change of 0.050.05 in magnitude (the correlation becomes less negative). However, removing the low GPA student changes it to r=−0.61r = -0.61, representing a change of 0.180.18 in magnitude (the correlation becomes much more negative). The student with low GPA caused a much larger change (0.180.18 vs 0.050.05), and importantly, removing them strengthened the negative correlation substantially. This suggests they were weakening the relationship between social media use and GPA in the original dataset. Option A correctly identifies that the low GPA student is more influential because their removal strengthened the correlation more substantially. Option B incorrectly focuses on the high social media user and wrongly claims they caused a larger change. Option C is wrong because the students clearly have different levels of influence based on the magnitude of correlation changes. Option D misuses terminology—both students are influential points, but to different degrees. Remember: when comparing influence on correlation, look at both the magnitude of change and whether the relationship becomes stronger or weaker. Larger magnitude changes indicate greater influence on the overall pattern.

Question 11

A dataset contains 25 observations with the regression equation y^=15+2.3x\hat{y} = 15 + 2.3x. One observation has coordinates (12, 45) and a studentized residual of 2.8. Another observation has coordinates (28, 78) and a studentized residual of 0.4. Which statement best compares the influence of these two points?

  1. The first point (12, 45) is more influential because its large residual indicates poor model fit
  2. The second point (28, 78) is more influential because its x-value gives it higher leverage despite the small residual
  3. The first point is influential due to its large residual, while the second point is influential due to high leverage (correct answer)
  4. Neither point is particularly influential since one has high leverage with low residual and the other has low leverage with high residual
Explanation: The first point (12, 45) has a large studentized residual of 2.8, indicating it doesn't fit the model well (predicted value would be 15 + 2.3×12 = 42.6), making it influential due to being an outlier. The second point (28, 78) has a high x-value giving it high leverage, and despite its small residual (it fits the line well: predicted = 15 + 2.3×28 = 79.4), high leverage points are influential because they can potentially pull the regression line. Both points are influential for different reasons - one due to large residual, the other due to high leverage.

Question 12

In a regression model with 40 data points, a statistician calculates that removing one specific point changes the slope from 0.85 to 0.92 and the R2R^2 from 0.76 to 0.81. The point has coordinates (35, 18) where 35 is well above the mean x-value and 18 is below the predicted y-value. Which characterization is most accurate?

  1. This is a high-leverage good fit point that strengthens the model when included in the analysis
  2. This is a high-leverage poor fit point that weakens the model when included in the analysis (correct answer)
  3. This is a low-leverage outlier that has minimal impact on regression coefficients but affects R-squared
  4. This is a moderate influence point that improves model fit despite having a negative residual
Explanation: The point has high leverage (x-value well above mean) and poor fit (y-value below predicted, indicating negative residual). When removed, both the slope increased (0.85 to 0.92) and R² improved (0.76 to 0.81), indicating the point was weakening the model. High leverage points with poor fit are particularly problematic because they pull the regression line toward themselves while not representing the true relationship, thus weakening the overall model fit.

Question 13

A linear regression model predicting house prices (in thousands) based on square footage has the equation y^=45+0.12x\hat{y} = 45 + 0.12x, with R2=0.78R^2 = 0.78. One house with 2000 square feet sold for $180,000. If this house is removed from the dataset and the model is recalculated as $y^=52+0.09x\hat{y} = 52 + 0.09x withwith R2=0.91R^2 = 0.91 $, which classification best describes this data point?

  1. High leverage point with low residual, making it influential to the regression parameters
  2. Low leverage point with high residual, making it an outlier but not particularly influential
  3. High leverage point with high residual, making it both an outlier and highly influential
  4. Moderate leverage point with high residual, making it an outlier that significantly impacts model fit (correct answer)
Explanation: The house has a residual of 180 - (45 + 0.12×2000) = 180 - 285 = -105 thousand, which is quite large, making it an outlier. The square footage (2000) appears to be in a moderate range, not extremely high or low, so it has moderate leverage. However, removing it substantially changed both the slope (0.12 to 0.09) and intercept (45 to 52), and dramatically improved R² (0.78 to 0.91), indicating it significantly impacts the model fit despite not having the highest possible leverage.

Question 14

A biologist studying the relationship between tree age and height notices that Cook's distance values for most trees are below 0.1, but three trees have Cook's distances of 0.8, 0.3, and 1.2 respectively. These three trees have the following characteristics: Tree X (age 45, height 12m), Tree Y (age 120, height 28m), Tree Z (age 200, height 15m). Based on this information, which tree is most likely responsible for the Cook's distance of 1.2?

  1. Tree Z, because very old trees should be much taller, making this point both high-leverage and poorly-fitting (correct answer)
  2. Tree Y, because it represents the expected pattern but with extreme values that give it high leverage
  3. Tree X, because young trees with short height represent an unusual combination that affects model stability
  4. Cannot be determined without additional information about the regression equation and residual values
Explanation: When you encounter Cook's distance problems, you're dealing with a measure that identifies influential points—observations that significantly affect the regression model when removed. Cook's distance combines two factors: leverage (how unusual the x-values are) and residual size (how poorly the point fits the predicted pattern). Tree Z is most likely responsible for the Cook's distance of 1.2 because it represents a dramatic departure from the expected biological relationship. At 200 years old, this tree should be quite tall based on normal growth patterns, yet it's only 15 meters—shorter than the 120-year-old tree that reaches 28 meters. This creates both high leverage (extreme age) and a large residual (poor fit to the age-height relationship), resulting in maximum influence on the regression line. Option B is wrong because Tree Y actually follows the expected pattern well—older trees being taller—so while it might have some leverage due to its age, it wouldn't have a large residual. Option C misunderstands the relationship; Tree X represents a normal combination of young age and modest height, which wouldn't create high influence. Option D is incorrect because Cook's distance specifically measures the combined effect of leverage and residuals, and we can reasonably infer from the biological context which tree most dramatically violates expectations. Remember that the highest Cook's distances occur when unusual x-values (leverage) combine with poor model fit (large residuals). Look for data points that are both extreme in their predictor values and inconsistent with the overall pattern.

Question 15

A researcher studying the relationship between advertising budget (x) and sales revenue (y) notices that when all 30 data points are included, the regression line has slope 2.1. When the three companies with the highest advertising budgets are removed, the slope becomes 1.8. When instead the three companies with sales revenue furthest from the regression line are removed, the slope becomes 2.4. What conclusion about influential points is most appropriate?

  1. High-budget companies are more influential than outlying companies because removing them caused a larger absolute change in slope
  2. Outlying companies are more influential because removing them caused the slope to increase rather than decrease
  3. High-budget companies have high leverage but follow the trend, while outlying companies don't follow the trend but have lower leverage (correct answer)
  4. Both groups are equally influential since they each caused a change of 0.3 units in the slope magnitude
Explanation: The high-budget companies (high x-values) have high leverage, but when removed, the slope decreased from 2.1 to 1.8, suggesting they were following and supporting the trend. The outlying companies (large residuals) don't follow the trend well, and removing them increased the slope from 2.1 to 2.4, indicating they were pulling the slope down. This pattern suggests high-budget companies have high leverage but conform to the relationship, while outlying companies have large residuals but lower leverage.