A linear model predicting marathon times has residual standard error of 18.5 minutes. Two runners have residuals of +22 minutes and -31 minutes respectively. If the model assumptions are satisfied, approximately what percentage of all residuals should have absolute values larger than the second runner's residual?
Answer: Approximately 10% because -31 minutes represents about 1.7 standard deviations, and we need both tails
The residual -31 represents 31/18.5 ≈ 1.68 standard deviations. For a normal distribution, approximately 5% of values fall beyond 1.68 standard deviations in each tail, giving 10% total with absolute values exceeding 31. Choice A only considers one tail. Choice C gives the percentage beyond 1 standard deviation. Choice D incorrectly doubles the two-tail percentage.
After fitting a regression model, an analyst creates a normal probability plot (Q-Q plot) of the residuals. The plot shows points that closely follow the diagonal line for most of the range, but the points curve away from the line at both extremes. What does this pattern suggest about the model?
Answer: The residuals have heavier tails than normal distribution, but the model may still be adequate
Points following the diagonal in the middle but curving away at extremes indicates heavier-tailed distribution than normal (more extreme residuals than expected). This is a minor assumption violation that often doesn't severely impact regression validity, especially if the middle range fits well. Choice A ignores the extreme deviations. Choice B describes bias patterns, not distributional issues. Choice D overstates the severity since most of the distribution appears normal.
A researcher fits a linear regression model to predict house prices based on square footage. The residual plot shows that most residuals are small and randomly scattered, but there are three houses with residuals of -$85,000, $92,000, and -$78,000. If the researcher removes these three houses and refits the model, which outcome is most likely?
Answer: The R2 value will increase because the outliers were inflating the sum of squared residuals
Removing outliers (points with large residuals) typically improves model fit by reducing the sum of squared residuals, which increases R2. Choice A is wrong because removing outliers often improves explanatory power. Choice C is incorrect because outliers significantly influence regression parameters. Choice D is wrong because removing outliers usually decreases residual standard error despite the smaller sample size.
A researcher compares residuals from two competing models fitted to the same dataset. Model A's residuals have interquartile range (IQR) of 8.4, while Model B's residuals have IQR of 6.7. Both models show random residual patterns with no violations. Which statement best describes the comparison?
Answer: Model B is preferable because smaller residual IQR indicates more precise predictions overall
When comparing statistical models fitted to the same dataset, residual analysis reveals how well each model performs. Residuals represent the differences between observed and predicted values, so smaller residual spread indicates more accurate predictions.
The interquartile range (IQR) measures the spread of the middle 50% of residuals, making it an excellent indicator of prediction precision. Since both models show appropriate residual patterns with no violations, you can directly compare their IQRs to assess performance. Model B's IQR of 6.7 is smaller than Model A's IQR of 8.4, meaning Model B's predictions cluster more tightly around the actual values. This makes Model B preferable because it provides more precise predictions overall, confirming answer C.
Option A incorrectly suggests that larger residual spread is beneficial. While models should capture data variability, this happens through the fitted values, not the residuals. Larger residuals actually indicate poorer fit quality.
Option B wrongly assumes that meeting basic residual assumptions makes models equivalent. Even when both models avoid violations, one can still perform significantly better than the other in terms of prediction accuracy.
Option D suggests you need mean absolute residuals for comparison, but this isn't necessary. The IQR already provides sufficient information about residual spread to make a meaningful comparison, especially when both models meet standard assumptions.
Remember: when residual patterns are appropriate for competing models, focus on measures of residual spread like IQR. Smaller spread always indicates better predictive performance.
A linear model predicting test scores has an R2=0.64 and residual standard error of 8.2 points. A student's actual test score is 78, and the model predicts 85. If this student's residual is typical for the model, what conclusion is most appropriate?
Answer: The student performed worse than expected, and this residual magnitude is consistent with the model's typical prediction accuracy
The residual is 78 - 85 = -7, indicating the student scored 7 points below prediction (worse than expected). Since the residual standard error is 8.2, a residual of -7 is within one standard deviation and thus typical. Choice B incorrectly states the student performed better. Choice C incorrectly suggests bias from one observation. Choice D incorrectly states the student performed better and misinterprets residual magnitude.
A researcher examines standardized residuals from a regression model and finds that 3 out of 60 observations have standardized residuals with absolute values greater than 2.5. The residuals appear randomly distributed with no patterns. How should the researcher interpret these findings?
Answer: The model fits adequately with a typical number of potential outliers that warrant individual examination
When analyzing standardized residuals from regression models, you're checking for outliers and model fit. Standardized residuals follow approximately a standard normal distribution, so about 95% should fall within ±2 standard deviations, and roughly 99% within ±2.5.
The correct answer is D because 3 out of 60 observations (5%) exceeding the 2.5 threshold is statistically normal. You'd expect about 1% to exceed ±2.5 in a well-fitting model, so 3 observations represents a typical amount of variation. Since the residuals appear randomly distributed without patterns, this suggests the model assumptions are reasonably met. These 3 observations should be examined individually as potential outliers, but their presence doesn't indicate model failure.
Option A misinterprets the 5% figure - this refers to the proportion of extreme residuals (3/60), not a model fit criterion. The actual percentage exceeding 2.5 is about 1.7%, which is close to the expected 1%. Option B incorrectly suggests 2.0 as an absolute cutoff. While values beyond ±2 warrant attention, exceeding this threshold doesn't automatically invalidate the model. Option C incorrectly diagnoses heteroscedasticity. Heteroscedasticity shows up as patterns in residual plots (like increasing variance), not simply from a few large residuals that are randomly distributed.
Remember: when evaluating standardized residuals, focus on both the number of extreme values and their pattern. Random extreme values are normal; systematic patterns indicate model violations that need correction.
Two models are fitted to the same dataset: Model A has R2=0.79 with residual standard error = 4.2, while Model B has R2=0.68 with residual standard error = 3.8. What is the most reasonable explanation for these results?
Answer: Model A explains more variance but Model B has better average prediction accuracy; the choice depends on the modeling goal
This scenario can occur when models are fitted to different subsets or use different approaches. R2 measures proportion of variance explained, while residual standard error measures absolute prediction accuracy. They can disagree based on data characteristics and modeling goals. Choice A ignores the meaningful difference in residual standard error. Choice B ignores the substantial R2 difference. Choice C incorrectly assumes the metrics must agree.
A regression model predicting test scores has R2=0.84. When the model is used to predict scores for a new group of students, the residuals have a standard deviation of 12 points. If the original test scores had a standard deviation of 20 points, what can be concluded about the model's performance?
Answer: The model performs worse on new data than expected, since residual standard deviation should be 8 points
With R2=0.84, we expect residual standard deviation ≈ 1−R2× original SD = 0.16×20=0.4×20=8 points. The observed 12 points suggests worse performance on new data, possibly due to overfitting. Choice B is wrong because 12 ≠ 8. Choice C incorrectly compares residual SD to original SD rather than expected residual SD. Choice D is wrong because we have sufficient information to evaluate performance.
A student calculates residuals for a linear regression and finds that 8 out of 40 data points have positive residuals, while 32 have negative residuals. The sum of all residuals is -2.4. What does this pattern most likely indicate about the regression line?
Answer: The regression line passes slightly below most data points, with a few points far above the line
Having many small negative residuals (32) and few positive residuals (8) suggests the line is slightly below most points, with some points significantly above it pulling the line down. The sum being close to zero (-2.4 ≈ 0) confirms proper calculation. Choice A is wrong because unequal numbers of positive/negative residuals doesn't indicate wrong slope calculation. Choice B incorrectly assumes systematic measurement error. Choice C is wrong because equal numbers of positive/negative residuals aren't required for proper regression.
A regression model has R2=0.58 when fitted to 80 observations. When examining leverage values, the analyst finds 6 observations with leverage greater than 0.075. Considering that high-leverage points can influence model fit, what is the most appropriate interpretation?
Answer: The model has acceptable fit, and 6 high-leverage points represent a normal proportion for this sample size
With 80 observations, having 6 high-leverage points (7.5%) is within normal range - typically expect 5-10% of points to have notable leverage. The threshold 0.075 is reasonable (≈3p/n where p=predictors, n=sample size). High leverage doesn't automatically mean bad influence. Choice B incorrectly interprets 7.5% as excessive. Choice C assumes leverage points inflate R2, but they can inflate or deflate it. Choice D confuses leverage with outliers.
A regression model predicting salary has these summary statistics: R2=0.58, residual standard error = $12,400, and sum of squared residuals = 2.84 × 109. If an employee's predicted salary is $67,500 and actual salary is $71,200, how should this residual be interpreted in context of the model's performance?
Answer: This is a moderate residual within normal model variation, representing about 0.3 standard deviations from predicted
The residual is $71,200 - $67,500 = $3,700. Comparing to the residual standard error of $12,400, this represents 3,700/12,400 ≈ 0.3 standard deviations, which is moderate and normal. Choice A incorrectly calls it small. Choice C incorrectly calls it large. Choice D incorrectly suggests more information is needed for interpretation.
Two regression models are fitted to the same dataset of 50 observations. Model A has R2=0.72 and Model B has R2=0.68. When examining residual plots, Model A shows a clear curved pattern in the residuals, while Model B shows residuals scattered randomly around zero. Which statement best describes the comparison between these models?
Answer: Model B is preferable despite lower R2 because the residual pattern suggests better model assumptions
While Model A has higher R2, the curved pattern in residuals indicates violated model assumptions (likely non-linearity), making the model unreliable. Model B's random residual pattern indicates appropriate model fit despite slightly lower R2. Choice A ignores the assumption violations. Choice C incorrectly focuses only on R2 magnitude. Choice D is wrong because R2 alone is insufficient when model assumptions are violated.
A regression model's residuals have mean = 0.08, median = -0.3, and standard deviation = 4.2. Additionally, 15% of residuals are positive while 85% are negative. What pattern is most likely present in the original data?
Answer: The data contains several extreme high outliers that pull the regression line upward
Mean > median (0.08 > -0.3) with 85% negative residuals suggests right-skewed residual distribution, typically caused by extreme high outliers. These outliers pull the regression line up, creating many negative residuals (points below the elevated line) and few large positive residuals. Choice B would show curved patterns, not skewed residuals. Choice C doesn't explain the skewed distribution. Choice D would create the opposite pattern (many positive, few negative residuals).
Two models are fitted to predict house prices. Model 1 uses only square footage (R2=0.61) and Model 2 adds number of bedrooms (R2=0.73). The residual plot for Model 1 shows random scatter, while Model 2's residuals show a slight curved pattern. For practical prediction purposes, which model should be preferred?
Answer: Model 1, because the assumption violations in Model 2 outweigh the improvement in R2
When evaluating regression models for prediction, you must balance two competing factors: explanatory power (measured by R2) and model validity (assessed through residual analysis). A model's predictions are only reliable if it meets the underlying assumptions of linear regression.
Model 2's curved residual pattern indicates a violation of the linearity assumption - the relationship between predictors and house prices isn't truly linear. This systematic pattern in residuals suggests the model is missing important nonlinear components or interactions. When regression assumptions are violated, the model's predictions become unreliable, confidence intervals become invalid, and you can't trust the statistical inferences.
Model 1, despite its lower R2 of 0.61, shows random scatter in residuals, indicating it properly meets regression assumptions. This makes its predictions trustworthy within its scope.
Choice A focuses only on the R2 improvement from 0.61 to 0.73, ignoring the assumption violations that compromise Model 2's reliability. Choice B acknowledges the curved pattern but incorrectly minimizes its importance - assumption violations aren't minor issues you can overlook for better fit statistics. Choice D makes an overly broad claim that simpler models are "always" preferable, which isn't true; the issue here is specifically about assumption violations, not complexity.
Study tip: Remember that R2 alone doesn't determine model quality. Always check residual plots first - a lower R2 with valid assumptions beats a higher R2 with violated assumptions for reliable predictions.
A linear regression model predicts house prices based on square footage. The residual plot shows points scattered randomly around zero with no clear pattern. However, three data points have residuals of +$45,000, -$52,000, and +$48,000, while most other residuals are between -$15,000 and +$15,000. What is the most appropriate conclusion about this model?
Answer: The model fits well overall, but contains outliers that may warrant further investigation or data cleaning
The random scatter around zero indicates good model fit overall, but the few extremely large residuals (much larger than the typical range) suggest outliers. These outliers don't invalidate the model but should be investigated. Choice A is wrong because random scatter indicates appropriate linear fit. Choice C is wrong because a few outliers don't necessarily shift the mean significantly from zero. Choice D is wrong because we don't know if the large residuals correspond to expensive houses, and isolated outliers don't indicate systematic bias.
A student examines residuals from a regression and calculates that the sum of squared residuals (SSR) is 847. The total sum of squares (TSS) is 2,180. However, when checking the residuals, the student finds their sum is 15.7 rather than zero. What should the student conclude?
Answer: The regression was calculated correctly; residual sums can deviate from zero due to rounding errors
A residual sum of 15.7 is very close to zero and likely due to rounding errors in calculations, especially since SSR and TSS values seem reasonable. Theoretically residuals sum to zero, but computationally small deviations are normal. Choice B incorrectly invalidates R2 over minor rounding. Choice C overstates the importance of exact zero sum. Choice D incorrectly interprets what non-zero residual sum means about the regression line.
A linear regression produces residuals with the following pattern: for the first 20 observations (collected in January), residuals average +3.2; for the next 20 observations (February), residuals average -1.8; for the final 20 observations (March), residuals average +0.4. What does this suggest about the model?
Answer: The model may be missing a time trend or seasonal component that affects the predictions
Systematic differences in residual means across time periods suggests the model is missing time-related variables (trend, seasonality). January's consistently positive residuals (+3.2) and February's negative residuals (-1.8) indicate time-dependent patterns not captured by the model. Choice A ignores the systematic time pattern. Choice C describes variance changes, not mean changes. Choice D incorrectly suggests adjusting intercept, which wouldn't fix time-dependent patterns.
A researcher reports that after removing 4 outliers from a dataset of 50 observations, the R2 increased from 0.45 to 0.72. A colleague argues this improvement is misleading. Which criticism is most statistically valid?
Answer: While R2 improved, the model may now have reduced generalizability and poorer performance on new data with similar outliers
The most valid concern is that removing outliers may create a model that doesn't generalize well to future data containing similar extreme values. Choice A is too absolute—sometimes outlier removal is justified. Choice B incorrectly suggests R2 directly depends on sample size. Choice D incorrectly cites non-existent minimum sample size rules for the percentage of data removed.
