Correlation & Simple Regression - Business Statistics

A regression model predicting monthly utility costs for a factory based on production volume has a standard error of the estimate ( $S_e$ ) of $450. The model predicts a utility cost of $12,000 for a month with a production volume of 1,000 units. Which is the best interpretation of this information?
Answer: The model's predictions have a typical error magnitude of about $450. The standard error of the estimate ( $S_e$ ), also known as the residual standard error, represents the typical or average size of the residuals (prediction errors). It is a measure of the typical distance between the observed values and the values predicted by the regression line, in the units of the dependent variable.
A company's regression model predicting monthly sales ( $y$ , in thousands) from advertising spend ( $x$ , in thousands) is $\hat{y} = 125 + 3.2x$ . If the company typically spends $8,000 on advertising monthly, what is the predicted monthly sales amount?
Answer: $150,600 When you encounter a regression equation, you're looking at a mathematical relationship that predicts one variable based on another. The equation $\hat{y} = 125 + 3.2x$ tells you that predicted sales equal 125 (thousand dollars) plus 3.2 times the advertising spend (in thousands). To solve this, substitute $x = 8$ (since $8,000 = 8 thousand dollars) into the equation: $\hat{y} = 125 + 3.2(8) = 125 + 25.6 = 150.6$ thousand dollars. Since the result is in thousands, multiply by 1,000 to get the actual dollar amount: $150.6 \times 1,000 = \$150,600$ . Looking at the wrong answers: Choice A shows $150,600 thousand, which would mean multiplying our result by 1,000 again—a common unit confusion error that would give you over $150 million in sales. Choice B gives $150.6 thousand, which treats the regression output as if it were already in actual dollars rather than thousands. Choice C shows $25,600, which represents only the variable portion ($ $3.2 \times 8 \times 1,000$ $) while ignoring the intercept term of 125. The correct answer is D: $150,600. Study tip: Always track your units carefully in regression problems. When both variables are expressed "in thousands," remember that your predicted value will also be in thousands and needs conversion to actual dollars. Write down the units at each step to avoid costly unit mix-ups.
A company models its monthly profit based on the number of salespeople it employs. The model is built using data from the past five years, during which the number of salespeople ranged from 20 to 50. The resulting equation is $\widehat{Profit} = \$10,000 + \$2,000 \times (Salespeople)$ . The company is considering reducing its sales force to 10. What is the primary concern when using this model to predict the profit with 10 salespeople?
Answer: The prediction is an extrapolation, as the model's reliability is unknown for a number of salespeople outside the range of the original data. The model was built using data where the number of salespeople (X) ranged from 20 to 50. Predicting for X=10 is outside this range. This is known as extrapolation. The linear relationship observed between 20 and 50 salespeople may not hold for a smaller sales force, making the prediction unreliable.
An economist studying income inequality calculates correlation coefficients for various relationships. Which correlation coefficient indicates the strongest linear relationship, regardless of direction?
Answer: Age vs. Technology adoption: $r = -0.82$ When evaluating correlation coefficients, you're measuring the strength of linear relationships between variables. The key insight is that correlation strength depends on the absolute value of the coefficient, not whether it's positive or negative. A correlation of -0.90 indicates a stronger relationship than +0.70 because we're looking at how closely points cluster around a straight line, regardless of the line's direction. To find the strongest relationship, compare the absolute values of all coefficients. Looking at our options: |0.76| = 0.76 for education and income, |0.69| = 0.69 for urban density and cost of living, |-0.82| = 0.82 for age and technology adoption, and |-0.71| = 0.71 for unemployment and consumer confidence. Option C shows the strongest linear relationship at 0.82. The negative sign simply tells us that as age increases, technology adoption tends to decrease, but the relationship itself is very strong. Option A ( $r = 0.76$ ) shows a strong positive relationship, but 0.76 < 0.82 in absolute terms. Option B ( $r = 0.69$ ) represents a moderate relationship that's weaker than the others. Option D ( $r = -0.71$ ) shows a strong negative relationship, but again, 0.71 < 0.82 in absolute value. Remember this key principle: when comparing correlation strengths, always use absolute values. A correlation of -0.85 beats +0.75 every time. The sign tells you direction; the absolute value tells you strength.
A regression analysis is performed on 50 data points, showing a moderate positive correlation. A 51st data point is then added. This new point has an x-value far from the mean of the other x-values and falls well below the established regression line. Which of the following is the most likely impact of adding this influential point?
Answer: It will increase the y-intercept and decrease the slope. When you encounter regression problems involving influential points or outliers, focus on how extreme data points can "pull" the regression line toward them, especially when they're far from the mean of the x-values. This new point is described as having an x-value far from the mean and falling "well below" the established regression line. Since the regression line will be pulled toward this influential point, imagine how the line must rotate and shift to get closer to this below-the-line point. The correct answer is D because the line must rotate downward (decreasing the slope) to move toward the low y-value point. When a line rotates downward around data that's mostly to the right of the y-axis, the y-intercept typically increases - think of it as the left end of the line lifting up as the right end drops down to approach the outlier. Answer A incorrectly suggests both the intercept and slope decrease. While the slope does decrease, the y-intercept actually increases due to the rotation pattern. Answer B falls into the trap of thinking one point can't matter much - but influential points with extreme x-values have disproportionate impact on regression lines, especially with moderate sample sizes like 50. Answer C correctly identifies that the slope decreases but wrongly claims the correlation coefficient increases. Adding an outlier that doesn't follow the established pattern typically weakens the correlation. Remember: influential points with extreme x-values act like leverage points that can dramatically alter regression lines. The farther a point is from the x-mean, the more it can "pull" the line toward itself.
An analyst creates a simple linear regression model. To check the model's validity, they produce a plot of the residuals versus the predicted values. The plot shows the points forming a distinct horizontal funnel shape, wider on the right than on the left. What does this pattern in the residual plot indicate?
Answer: The variance of the errors is not constant (heteroscedasticity). A plot of residuals versus predicted values is used to detect patterns in the errors. A funnel or fan shape, where the vertical spread of the residuals changes as the predicted values change, is a classic sign of heteroscedasticity, which is the violation of the constant variance assumption.
A real estate analyst uses simple linear regression to model the selling price of houses (in thousands of dollars) based on their square footage. The data includes houses ranging from 1,200 to 4,000 square feet. The resulting regression equation is $\widehat{Price} = 40.5 + 0.15 \times (Square Footage)$ . How should the intercept of 40.5 be interpreted?
Answer: The intercept lacks a practical business interpretation because a square footage of zero is outside the relevant range of the data. The intercept is the predicted value of Y when X=0. In this context, it would be the predicted price for a house with 0 square feet. Since 0 is far outside the range of the observed data (1,200 to 4,000 sq ft), this is an example of extrapolation. Therefore, the intercept has no meaningful, practical interpretation.
A company-wide analysis shows a strong positive correlation between employees' years of experience and their performance review scores. However, when the data is analyzed separately for each of the company's three divisions, each division shows a slight negative correlation. What is the best explanation for this apparent contradiction?
Answer: A confounding variable (the division) is creating an artificial positive trend when the data are aggregated. When you encounter contradictory correlation patterns between aggregated and subgroup data, you're looking at a classic case of Simpson's Paradox. This occurs when a confounding variable masks the true relationship within groups but creates a misleading pattern when data are combined. Here's what's happening: Each division likely has different baseline characteristics - perhaps Division A hires mostly entry-level employees who receive intensive training, Division B focuses on mid-career hires, and Division C recruits senior executives. Within each division, more experienced employees might actually perform slightly worse (negative correlation), possibly due to complacency or outdated skills. However, when you combine all divisions, the higher-performing divisions happen to have more experienced employees overall, creating an artificial positive correlation. Answer D correctly identifies that the division itself is the confounding variable creating this paradox. The aggregated positive correlation is misleading because it ignores important structural differences between divisions. Answer A incorrectly assumes sample size is the issue - Simpson's Paradox can occur regardless of sample size. Answer B misunderstands the relationship direction; the dependent/independent variable assignment isn't the problem. Answer C wrongly attributes the pattern to measurement error when it's actually a statistical phenomenon caused by data aggregation. Study tip: Whenever you see opposing correlation patterns between grouped and ungrouped data, immediately think Simpson's Paradox and look for a lurking confounding variable. This is a favorite exam scenario because it tests your understanding of when correlation can be misleading.
A quality control engineer studies the relationship between production temperature (°F) and defect rate (%). The correlation coefficient is $r = -0.85$ . If the regression equation is $\hat{y} = 12.5 - 0.08x$ , where $x$ is temperature and $y$ is defect rate, what can be concluded about the relationship?
Answer: Higher temperatures are associated with lower defect rates with strong negative correlation The correlation $r = -0.85$ indicates a strong negative association (since $|r| > 0.7$ ), and the negative slope confirms that higher temperatures are associated with lower defect rates. However, correlation does not imply causation. Choice A incorrectly claims causation. Choice C incorrectly describes the correlation as moderate. Choice D incorrectly describes $r = -0.85$ as weak correlation.
Two variables have a correlation coefficient of $r = 0.0$ . Which statement is most accurate about the relationship between these variables?
Answer: There is no linear relationship, but a nonlinear relationship may still exist A correlation coefficient of zero indicates no linear relationship, but nonlinear relationships (parabolic, exponential, etc.) can still exist with $r = 0$ . Choice A is too strong - it rules out all relationships. Choice B incorrectly equates zero correlation with independence. Choice D is incorrect because while the slope is zero, the line passes through $( \bar{x}, \bar{y})$ , not necessarily the origin.
A financial analyst develops a simple linear regression model to predict a company's stock price (Y) based on its quarterly earnings per share (X). The model yields a coefficient of determination ( $R^2$ ) of 0.64. Which statement correctly describes a property of this model?
Answer: The correlation coefficient between the predicted stock prices and the actual stock prices is 0.80. The coefficient of determination, $R^2$ , is the square of the correlation coefficient, $r$ , between the independent (X) and dependent (Y) variables. It is also the square of the correlation between the observed Y values and the predicted Y values ( $\hat{Y}$ ). Therefore, the correlation between predicted and actual stock prices is $\sqrt{R^2} = \sqrt{0.64} = 0.80$ .
An analyst is comparing two potential linear models to describe a dataset. Model A is the least-squares regression line, $\hat{y} = 10 + 2x$ . Model B is an alternative line, $\hat{y} = 11 + 1.9x$ , proposed by a manager based on industry experience. If both lines are applied to the same dataset, what can be definitively concluded about their sum of squared residuals (SSE)?
Answer: The SSE for Model A will be less than or equal to the SSE for Model B. When you encounter questions comparing regression models, remember that the least-squares regression line has a special mathematical property: it minimizes the sum of squared residuals (SSE) for any given dataset. The least-squares method finds the unique line that produces the smallest possible SSE by minimizing $\sum(y_i - \hat{y}_i)^2$ . This is its defining characteristic. Since Model A is the least-squares line for this dataset, it must have an SSE that is less than or equal to any other possible line fitted to the same data, including Model B. Why "less than or equal to" rather than "strictly less than"? In the extremely unlikely scenario where the manager's proposed line happens to be identical to the least-squares solution, the SSEs would be equal. Otherwise, Model A will have a smaller SSE. Option A is incorrect because it claims Model A will be "strictly" less than Model B, but we can't rule out the possibility they could be equal if the manager somehow guessed the exact least-squares line. Option B is wrong because $R^2$ isn't needed to make this comparison—the least-squares property guarantees the relationship regardless of how well either model fits. Option C reflects a fundamental misunderstanding, as a higher intercept doesn't determine SSE; the overall fit to the data points does. Study tip: Remember that "least-squares" means exactly that—it produces the least sum of squared errors. Any time you compare a least-squares model to any alternative model on the same dataset, the least-squares model will have SSE ≤ the alternative.
A researcher regresses employee job satisfaction (Y) on the number of hours worked per week (X) and finds a regression line $\hat{Y} = b_0 + b_1 X$ and a coefficient of determination $R^2$ . If the researcher decides to switch the variables, regressing hours worked (now Y') on job satisfaction (now X'), which of the following statements about the new model is true?
Answer: The new coefficient of determination, $R'^2$ , will be identical to the original $R^2$ . The coefficient of determination, $R^2$ , is symmetric. It measures the proportion of variance in one variable that is explained by the other, regardless of which is designated as the dependent variable. Correlation (r) is also symmetric. The slope is not symmetric, and the p-value for the slope will be identical because the underlying t-statistic for the relationship is the same.
A marketing analyst finds that the correlation coefficient between advertising expenditure (in thousands of dollars) and monthly sales revenue (in thousands of dollars) is $r = 0.72$ . If the standard deviation of advertising expenditure is $s_x = 15$ and the standard deviation of sales revenue is $s_y = 45$ , what is the slope of the least squares regression line for predicting sales revenue from advertising expenditure?
Answer: $2.16$ The slope of the regression line is $b_1 = r \cdot \frac{s_y}{s_x} = 0.72 \cdot \frac{45}{15} = 0.72 \cdot 3 = 2.16$ . Choice B incorrectly uses $\frac{s_x}{s_y}$ in the denominator. Choice C uses $\frac{s_y}{s_x}$ without multiplying by the correlation. Choice D incorrectly multiplies $r$ by both standard deviations.
A marketing manager models the relationship between monthly advertising spend in dollars (X) and monthly sales in dollars (Y). The resulting regression equation is $\hat{Y} = 5000 + 2.5X$ , with a correlation coefficient $r = 0.70$ . If the manager decides to re-run the analysis with advertising spend measured in thousands of dollars (let's call it $X^*$ ), what will be the new slope and correlation coefficient?
Answer: The new slope will be 2500, and the correlation will be 0.70. Correlation is a unitless measure and is unaffected by linear transformations (like changing units from dollars to thousands of dollars). Thus, the correlation remains 0.70. The slope, however, will change. Since the new predictor $X^*$ represents thousands of dollars, $X = 1000X^*$ . Substituting this into the original equation gives $\hat{Y} = 5000 + 2.5(1000X^*) = 5000 + 2500X^*$ . The new slope is 2500.
A manufacturer models the number of defective units per day (Y) as a function of the production line speed (X, measured in units per minute). The fitted regression model is $\widehat{Defects} = 3.2 + 0.05 \times (Speed)$ . How should the slope of 0.05 be interpreted?
Answer: For each one unit-per-minute increase in line speed, the model predicts an average increase of 0.05 defective units per day. The slope of a regression line represents the predicted change in the dependent variable (Y) for a one-unit increase in the independent variable (X). In this context, Y is 'defective units per day' and X is 'speed in units per minute'. Therefore, the slope of 0.05 means that for each additional unit per minute of speed, the predicted number of defects per day increases by 0.05. Choice C is a correct restatement but not the direct interpretation of the slope value itself. Choice B incorrectly interprets the slope as a percentage change.
A firm is evaluating two different simple linear regression models to forecast sales. Model A uses 'Advertising Spend' as the predictor and has an $R^2$ of 0.49 and a standard error of the estimate ( $S_e$ ) of $10,000. Model B uses 'Number of Sales Calls' as the predictor and has an $R^2$ of 0.64 and an $S_e$ of $8,000. Which statement is the most accurate comparison of the two models?
Answer: Model B is preferable as it explains more variance in sales and has smaller typical prediction errors. When comparing regression models for the same dependent variable, a better model generally has a higher $R^2$ and a lower standard error of the estimate ( $S_e$ ). Model B has a higher $R^2$ (0.64 > 0.49), meaning it explains a larger proportion of the variance in sales. It also has a lower $S_e$ ($8,000 < $10,000), meaning its predictions are typically more accurate. Therefore, Model B is the superior model based on these metrics.
A logistics company analyzes the relationship between the weight of a package (X, in pounds) and its delivery time (Y, in hours). The analysis provides the following summary statistics: Standard deviation of package weight ( $s_x$ ) = 5 pounds, Standard deviation of delivery time ( $s_y$ ) = 10 hours, and the slope of the regression line ( $b_1$ ) = 1.6. What is the coefficient of determination ( $R^2$ ) for this model?
Answer: 0.64 This is a two-step problem. First, find the correlation coefficient, $r$ , using the formula $b_1 = r \frac{s_y}{s_x}$ . Rearranging gives $r = b_1 \frac{s_x}{s_y} = 1.6 \times \frac{5}{10} = 1.6 \times 0.5 = 0.8$ . Second, find the coefficient of determination by squaring $r$ : $R^2 = r^2 = (0.8)^2 = 0.64$ .
A business analyst is preparing to run a simple linear regression. Which of the following conditions is generally considered the least critical for the primary purpose of estimating the regression coefficients (slope and intercept), although it is more important for hypothesis testing and interval estimation?
Answer: The residuals should follow a normal distribution. The normality assumption of the residuals is primarily required for the validity of hypothesis tests (t-tests) and the construction of confidence and prediction intervals. The regression line's coefficients (slope and intercept) can be calculated and used for point prediction even if normality is violated, especially in large samples. The other conditions (linearity, independence, constant variance) are more fundamental to the model being a meaningful and unbiased representation of the relationship.
A regression analysis produces the equation $\hat{y} = 15.8 + 2.4x$ where $x$ represents years of experience and $y$ represents hourly wage in dollars. The mean years of experience is $\bar{x} = 6.5$ years. What is the mean hourly wage for this dataset?
Answer: $31.40 When you encounter a regression equation with mean values, remember that the regression line always passes through the point $(\bar{x}, \bar{y})$ . This is a fundamental property of least squares regression that makes finding the mean of the dependent variable straightforward. To find the mean hourly wage, substitute the mean years of experience into the regression equation: $\hat{y} = 15.8 + 2.4x$ . With $\bar{x} = 6.5$ years, we get $\bar{y} = 15.8 + 2.4(6.5) = 15.8 + 15.6 = 31.40$ . The mean hourly wage is $31.40. Let's examine why the other answers are incorrect. Answer A ($18.20) appears to result from incorrectly subtracting the slope term: $15.8 + 2.4 = 18.2$ , treating the slope as additive rather than multiplicative. Answer B ($22.30) might come from using an incorrect x-value or making an arithmetic error in the calculation. Answer C ($15.80) represents a common misconception—thinking the y-intercept alone equals the mean wage, which ignores the effect of the mean experience level. The key insight here is recognizing that the regression line's y-intercept ( $15.80$ ) tells you the predicted wage for zero years of experience, not the mean wage for the dataset. Since the average worker has $6.5$ years of experience, the mean wage must account for this experience level through the slope coefficient. Study tip: Always remember that $(\bar{x}, \bar{y})$ lies on the regression line. When given the regression equation and $\bar{x}$ , simply plug in to find $\bar{y}$ .

Start this quiz