Regression Models - Business Calculus

Based on this statistical analysis, which conclusion is most justified?
Answer: There is a statistically significant positive association between the number of coffee shops and minor crimes, but this does not prove that new coffee shops cause crime. The statistical analysis shows a significant association (correlation), but correlation does not imply causation. There could be a lurking variable, such as population growth, that causes both the number of coffee shops and the number of minor crimes to increase. Therefore, one cannot conclude that coffee shops cause crime. Choices A and B incorrectly infer causation. Choice D is incorrect because the problem states the relationship is statistically significant.
A specific house with a living area of 2,200 square feet was recently sold for $315,000. What is the residual for this data point?
Answer: $6,000 The residual is the difference between the actual value and the predicted value (actual - predicted). First, calculate the predicted price ( $\hat{P}$ ) for a house with $A = 2,200$ square feet. Note that $P$ is in thousands of dollars. $\hat{P} = 45 + 0.12(2200) = 45 + 264 = 309$ . This corresponds to a predicted price of $309,000. The actual sales price was $315,000. The residual is: Residual = Actual Price - Predicted Price = $315,000 - $309,000 = $6,000.
A retail store's regression analysis shows that weekly profit $P$ (in dollars) is related to the number of hours of staff training $T$ by the equation $P = 2400 + 85T - 1.5T^2$ . Based on this model, what does the coefficient $-1.5$ most likely represent?
Answer: The marginal benefit of training decreases by $1.50 for each additional hour of training In the quadratic model P = 2400 + 85T - 1.5T², the coefficient -1.5 on T² indicates the rate of change of the marginal effect. Since dP/dT = 85 - 3T, the coefficient -1.5 (which is half of -3) represents how much the marginal benefit of training decreases with each additional hour. Choice A is incorrect because it misinterprets the quadratic term as a simple linear decrease. Choice C incorrectly assumes the coefficient represents cost rather than the changing rate of benefit. Choice D misunderstands the breakeven point and the meaning of the quadratic term.
According to this log-linear model, what is the approximate effect of gaining an additional 1,000 active users?
Answer: Market value is predicted to increase by approximately 0.5%. This is a log-linear model, where the dependent variable is logarithmic. In such a model, $\ln(y) = \beta_0 + \beta_1 x$ , a one-unit increase in $x$ corresponds to a multiplicative change in $y$ by a factor of $e^{\beta_1}$ . Here, the independent variable $U$ is in thousands of users, so an increase of 1,000 users means $\Delta U = 1$ . The corresponding change in $V$ is a multiplication by $e^{0.005}$ . Since $e^{0.005} \approx 1.005$ , this represents a 0.5% increase in market value. A common approximation is that the percentage change is $100 \times \beta_1 = 100 \times 0.005 = 0.5\%$ .
A restaurant chain uses the regression model $R = 8500 + 340L - 12L^2 + 180M$ to predict monthly revenue $R$ (in dollars), where $L$ is the number of lunch specials offered and $M$ is the number of menu items. The standard error for the coefficient of $M$ is 45. If the restaurant currently offers 15 lunch specials and 25 menu items, what does the coefficient 180 indicate about the relationship between menu items and revenue?
Answer: Each additional menu item increases revenue by approximately $180, holding the number of lunch specials constant When you encounter a regression model, focus on interpreting the coefficients as partial derivatives—they show how the dependent variable changes when you increase one independent variable by one unit while holding all other variables constant. In the model $R = 8500 + 340L - 12L^2 + 180M$ , the coefficient 180 represents the partial derivative $\frac{\partial R}{\partial M} = 180$ . This means that for each additional menu item (increasing $M$ by 1), revenue increases by approximately $180, assuming the number of lunch specials ($ $L$ $) remains unchanged. The word "approximately" is crucial because real-world data has variability, and regression coefficients are estimates based on sample data. Answer D correctly captures this interpretation with the key phrase "holding the number of lunch specials constant." Answer A is wrong because it claims the relationship is "exact"—regression coefficients are estimates, not precise relationships. Answer B misinterprets the coefficient as contributing to base revenue, but 180 is the marginal effect per additional item, not a base contribution. Answer C incorrectly suggests the $180 applies to every 25 menu items rather than each single additional item. Remember this pattern: in multiple regression, each coefficient tells you the marginal effect of that variable when all other variables are held constant. Watch for answer choices that ignore the "holding other variables constant" condition or misinterpret coefficients as exact rather than estimated relationships.
What would be the equation for Model 2, relating productivity $P$ to training time in minutes $M$ ?
Answer: $P = 10 + 0.025M$ When dealing with unit conversions in linear equations, you need to carefully track how changing the input variable affects both the slope and intercept of your model. In Model 1, $P = 10 + 1.5T$ where $T$ is in hours. To convert to minutes, you use the relationship $T = \frac{M}{60}$ (since there are 60 minutes in an hour). Substituting this into the original equation: $P = 10 + 1.5 \cdot \frac{M}{60} = 10 + \frac{1.5M}{60} = 10 + 0.025M$ . This gives us answer choice D. Let's examine why the other options are incorrect. Choice A ( $P = 10 + 90M$ ) mistakenly multiplies the slope by 60 instead of dividing by it - this would suggest that one minute of training equals 60 hours, which is backwards. Choice B ( $P = 600 + 1.5M$ ) correctly identifies that the slope should change but incorrectly modifies the intercept; the baseline productivity of 10 units per hour shouldn't change just because we're measuring time differently. Choice C ( $P = 600 + 90M$ ) combines both errors from A and B. The key insight is that when you convert the input variable to smaller units (hours to minutes), the coefficient must get smaller proportionally, while the intercept representing the baseline value remains unchanged. Remember: unit conversions affect slopes inversely - if you multiply the input units by a factor, divide the slope by that same factor.
The consultant is asked to use this model to predict the annual revenue for a large corporation with an annual advertising budget of $2,000,000. Which of the following statements about this prediction is most accurate?
Answer: The prediction is an extrapolation and is likely to be unreliable because the input value is far outside the range of the original data. The model was built using data for advertising budgets between $5,000 and $50,000. Using the model to make a prediction for a budget of $2,000,000 is an example of extrapolation, which means predicting outside the range of the observed data. Extrapolation is dangerous because there is no guarantee that the linear relationship observed in the sample data will hold for values so far outside that range. Therefore, the prediction is likely to be unreliable.
A consulting firm develops a model to predict project completion time $T$ (in weeks) using $T = 12 + 2.5C + 0.8E - 1.2X$ , where $C$ is project complexity score (1-10), $E$ is team experience level (1-5), and $X$ is a binary variable (1 if remote work, 0 if on-site). What is the predicted difference in completion time between an on-site project and an identical remote project?
Answer: The on-site project is predicted to take 1.2 weeks longer than the remote project When you encounter a linear model with binary variables, focus on how changing that variable from 0 to 1 affects the output. Binary variables represent "on/off" conditions, and their coefficients tell you the exact impact of switching states. In this model, $X$ represents work location: 1 for remote work and 0 for on-site work. The coefficient of $X$ is $-1.2$ , which means remote work ( $X = 1$ ) reduces completion time by 1.2 weeks compared to on-site work ( $X = 0$ ). To see this clearly, compare two identical projects differing only in location. For an on-site project: $T_{onsite} = 12 + 2.5C + 0.8E - 1.2(0) = 12 + 2.5C + 0.8E$ . For the same project done remotely: $T_{remote} = 12 + 2.5C + 0.8E - 1.2(1) = 12 + 2.5C + 0.8E - 1.2$ . The difference is $T_{onsite} - T_{remote} = 1.2$ weeks, meaning the on-site project takes 1.2 weeks longer. Choice A incorrectly reverses the relationship, suggesting remote work increases time. Choice B ignores the $X$ variable entirely, missing that location matters in this model. Choice C doubles the coefficient effect (2.4 instead of 1.2), perhaps confusing the magnitude of change. Study tip: With binary variables in linear models, the coefficient directly gives you the difference between the two states. Always check the sign carefully—negative coefficients mean the "1" state decreases the outcome compared to the "0" state.
Based on this regression model, what would be the predicted difference in annual returns between two clients who are identical except that one has a risk tolerance score of 8 while the other has a risk tolerance score of 5?
Answer: The higher risk tolerance client is predicted to have 4.2 percentage points higher annual return When you encounter a multiple regression model question asking about the difference between two scenarios, focus on how changing one variable affects the outcome while holding all other variables constant. In this regression model, $R = 2.8 + 0.12A - 0.003I + 1.4T$ , the coefficient 1.4 in front of the risk tolerance variable (T) tells you that for every 1-point increase in risk tolerance score, the predicted annual return increases by 1.4 percentage points, assuming all other factors remain the same. Since the two clients are identical except for their risk tolerance scores (8 vs. 5), you only need to calculate the impact of this 3-point difference in risk tolerance. The difference in predicted returns is: $(8-5) \times 1.4 = 3 \times 1.4 = 4.2$ percentage points. Choice A (3.6 percentage points) incorrectly multiplies the risk tolerance difference by something other than the correct coefficient. Choice B (1.4 percentage points) uses the coefficient correctly but fails to account for the 3-point difference in risk tolerance scores—this would be the answer if the difference were only 1 point. Choice D (2.8 percentage points) confuses the coefficient for risk tolerance with the constant term in the regression equation. Remember this key principle: in regression analysis, when comparing scenarios that differ by only one variable, multiply the difference in that variable by its coefficient. The other terms in the equation cancel out when calculating the difference between predictions.
Which statement provides the correct interpretation of the $R^2$ value?
Answer: 64% of the variability in stock prices can be explained by the linear relationship with quarterly earnings per share. The coefficient of determination, $R^2$ , measures the proportion of the variance in the dependent variable (stock price) that is predictable from the independent variable (earnings per share). An $R^2$ of 0.64 means that 64% of the observed variation in stock prices is accounted for by the linear model. It does not represent the percentage of accurate predictions, nor is it the correlation coefficient ( $r$ ). $r$ would be the square root of $R^2$ , which is 0.80. Finally, $R^2$ describes the strength of association, not causation or probability.
A marketing firm uses a linear regression model to predict weekly sales of a product. The fitted model is $S = 1200 + 4.5A$ , where $S$ is the number of units sold and $A$ is the advertising spending in thousands of dollars. What is the predicted increase in weekly sales if advertising spending is increased from $3,000 to $8,000?
Answer: 22.5 units The question asks for the predicted change in sales ( $S$ ) for a given change in advertising spending ( $A$ ). First, calculate the change in $A$ in the correct units. The spending increases from $3,000 to $8,000, an increase of $5,000. Since $A$ is measured in thousands of dollars, the change in $A$ is $5000 / 1000 = 5$ . The slope of the regression line, 4.5, represents the predicted change in $S$ for a one-unit increase in $A$ . Therefore, the total predicted increase in sales is the slope multiplied by the change in $A$ : $4.5 \times 5 = 22.5$ units.
If the company decides to increase its monthly advertising budget by $2,000, what is the expected increase in its monthly sales according to the model?
Answer: $50,000 When you encounter a linear regression model like $S = 50 + 2.5A$ , you're looking at a relationship where the coefficient (2.5) tells you how much the dependent variable changes for each one-unit increase in the independent variable. This coefficient is the key to solving change problems. Here, $S$ represents sales in thousands of dollars, and $A$ represents advertising budget in hundreds of dollars. The coefficient 2.5 means that for every additional hundred dollars spent on advertising, sales increase by 2.5 thousand dollars. The company wants to increase advertising by $2,000. First, convert this to the model's units: $2,000 ÷ $100 = 20 hundreds of dollars. So $\Delta A = 20$ . The expected increase in sales is: $\Delta S = 2.5 \times 20 = 50$ thousand dollars, which equals $50,000. Looking at the wrong answers: Choice A ($5,000) incorrectly uses the advertising increase of $2,000 and multiplies by 2.5 without proper unit conversion. Choice B ($50) forgets that sales are measured in thousands of dollars—this would be the answer if you ignored the units entirely. Choice C ($55,000) appears to add the y-intercept (50) to the change in sales (50), but the y-intercept represents the baseline sales level, not part of the change calculation. Study tip: In linear models, always check your units carefully and remember that the coefficient tells you the rate of change. Convert all values to match the model's units before calculating, and don't confuse the y-intercept with the change in the dependent variable.
Which statement correctly interprets the coefficient for the number of bedrooms ( $B$ )?
Answer: For apartments of the same size and distance from the city center, each additional bedroom is associated with a predicted $400 increase in monthly rent. When interpreting coefficients in multiple linear regression, you need to understand that each coefficient represents the change in the dependent variable associated with a one-unit increase in that predictor, holding all other variables constant. This "ceteris paribus" principle is crucial for proper interpretation. In the model $\hat{R} = 500 + 1.5S + 400B - 20D$ , the coefficient 400 for bedrooms means that when you compare apartments with identical size and distance from city center, each additional bedroom is associated with a $400 increase in predicted rent. This makes choice A correct. Choice B is wrong because it ignores the "all else equal" requirement. The $400 effect only applies when other factors are held constant, not "regardless of" size or distance. Choice C misinterprets what the coefficient represents—it's not a base rent for zero bedrooms, but rather the marginal effect of adding one bedroom. The base rent (when all variables equal zero) is the intercept: $500. Choice D makes the common error of comparing coefficients with different units. You can't conclude that bedrooms (measured in whole units) is more influential than size (measured per square foot) just by comparing 400 to 1.5, since they measure completely different scales. Study tip: Always remember that regression coefficients show marginal effects "holding other variables constant." When you see phrases like "regardless of other factors" or comparisons of coefficient magnitudes across different units, these are red flags for incorrect interpretations.
An e-commerce company's regression analysis reveals that monthly website conversion rate $C$ (as a percentage) follows the model $C = 3.2 + 0.15S - 0.002S^2$ , where $S$ is the number of site visits (in thousands). According to this model, what is the optimal number of site visits that maximizes the conversion rate?
Answer: The conversion rate is maximized when there are approximately 37.5 thousand site visits per month When you encounter a quadratic function in business calculus and need to find its maximum or minimum value, you're looking for the vertex of the parabola. Since this conversion rate model $C = 3.2 + 0.15S - 0.002S^2$ has a negative coefficient on the $S^2$ term, it opens downward and has a maximum point. To find where the conversion rate is maximized, you need to take the derivative and set it equal to zero. The derivative is $C'(S) = 0.15 - 0.004S$ . Setting this equal to zero: $0.15 - 0.004S = 0$ , which gives us $S = 0.15 ÷ 0.004 = 37.5$ thousand site visits. You can verify this is a maximum by checking the second derivative: $C''(S) = -0.004$ , which is negative, confirming a maximum. Answer C correctly identifies 37.5 thousand site visits as the optimal number. Answer A (150.0 thousand) appears to come from incorrectly using $0.15 ÷ 0.001$ instead of the correct $0.004$ coefficient. Answer B (75.0 thousand) doubles the correct answer, possibly from a sign error or calculation mistake. Answer D (18.75 thousand) halves the correct answer, suggesting an error in the division step. Study tip: For quadratic optimization problems, always remember the three-step process: take the derivative, set it to zero, and solve for the variable. Double-check your arithmetic carefully, as coefficient errors are common traps in multiple-choice questions.
How will the addition of this new data point, which is an outlier, most likely affect the regression model?
Answer: The slope of the regression line will decrease, and the $R^2$ value will decrease. When you encounter regression problems involving outliers, focus on how extreme data points affect both the line's direction (slope) and how well it fits the data overall ( $R^2$ ). An outlier with low experience but unusually high salary creates a "leverage point" that pulls the regression line toward it. Since this point has a low x-value (1 year) but high y-value (salary), it flattens the line's slope. Imagine the original line showing a steady upward trend from low to high experience levels—adding this outlier forces the line to accommodate a high salary at the beginning of the experience range, reducing the overall steepness. The $R^2$ value measures how much of the salary variation the experience variable explains. Adding an outlier that doesn't follow the established pattern increases unexplained variation, making the model fit worse and lowering $R^2$ . Choice B incorrectly suggests both slope and $R^2$ increase—this would only happen if the outlier reinforced the existing trend. Choice C wrongly claims the slope increases; while the $R^2$ decrease is correct, a low-experience, high-salary outlier pulls the slope down, not up. Choice D incorrectly states $R^2$ increases—outliers that don't follow the pattern always worsen model fit. Study tip: Remember that outliers typically hurt regression models in two ways: they distort the line's direction and reduce how well the line explains the data. Practice visualizing how extreme points "pull" the regression line toward them.
A logistics company finds that delivery time $D$ (in hours) is related to package weight $W$ (in pounds) and distance $M$ (in miles) by the model $D = 1.2 + 0.08W + 0.003M$ . If the coefficient of determination $R^2 = 0.76$ , which statement best interprets this regression model?
Answer: 76% of the variation in delivery time can be explained by package weight and distance combined R² = 0.76 means that 76% of the variance in the dependent variable (delivery time) is explained by the independent variables (weight and distance) in the regression model. Choice B confuses R² with the correlation coefficient r (which would be √0.76 ≈ 0.87). Choice C incorrectly interprets R² as a confidence interval coverage probability. Choice D misinterprets R² as the percentage of accurate individual predictions rather than explained variance.

Start this quiz