Multiple Regression Coefficients - Biostatistics

A study of blood pressure medication effectiveness uses the model: $BP\_Reduction = 15 + 8 \times Drug\_A + 12 \times Drug\_B - 3 \times Drug\_A \times Drug\_B + 0.1 \times Baseline\_BP$ , where Drug_A and Drug_B are binary indicators for two medications that can be used alone or in combination. What does the negative interaction coefficient suggest about the combination therapy?
Answer: The combination is more effective than either drug alone but shows diminishing marginal returns With Drug_A alone: reduction = 15 + 8 = 23. With Drug_B alone: reduction = 15 + 12 = 27. With both drugs: reduction = 15 + 8 + 12 - 3 = 32. The combination (32) is still better than either alone (23 or 27), but the interaction term (-3) indicates the combined effect is less than the sum of individual effects (8 + 12 = 20 vs. actual additional benefit of 17). This represents diminishing marginal returns, not contraindication. Choice A is wrong because the drugs still provide benefit together. Choice B is partially correct but doesn't note the combination is still most effective. Choice D is wrong because reduction is still positive and substantial.
In a regression model: $\text{Income} = 25000 + 2000 \times \text{Education} + 1500 \times \text{Experience} + 500 \times \text{Urban}$ , where Education is years of schooling, Experience is years of work experience, and Urban is a dummy variable (1 = urban, 0 = rural). What is the income difference between two individuals with identical education and experience, where one lives in an urban area and the other in a rural area?
Answer: $500 When analyzing multiple regression models, you need to understand how dummy variables work and what happens when you hold other variables constant. This regression equation shows how income depends on education, experience, and location, where Urban is a binary variable that equals 1 for urban residents and 0 for rural residents. To find the income difference between urban and rural residents with identical education and experience, you're essentially comparing two scenarios where Education and Experience are the same, but Urban differs by 1 unit (from 0 to 1). Since Education and Experience are held constant, their terms cancel out when you subtract the rural income from the urban income: $(\text{Income}_{\text{urban}} - \text{Income}_{\text{rural}}) = 500 \times 1 - 500 \times 0 = 500$ The coefficient on the Urban variable (500) directly represents this difference, making C) $500 the correct answer. Looking at the wrong answers: A) $2000 incorrectly uses the Education coefficient, suggesting confusion about which variable creates the difference. B) $1500 mistakenly uses the Experience coefficient, showing the same type of error. D) $4000 appears to sum multiple coefficients ($2000 + $1500 + $500), which misunderstands that we're comparing people with identical education and experience levels. Study tip: In regression models, the coefficient of a dummy variable always represents the difference between the two categories when all other variables are held constant. You don't need complex calculations—just identify the relevant coefficient.
A researcher models crop yield using: $\text{Yield} = 45 + 1.2 \times \text{Fertilizer} + 0.8 \times \text{Rainfall} - 0.03 \times \text{Fertilizer}^2$ . What does the coefficient -0.03 indicate about the relationship between fertilizer and yield?
Answer: The effect of fertilizer decreases as fertilizer amount increases When you encounter a regression model with both linear and quadratic terms for the same variable, you're looking at a polynomial relationship that can capture curved, non-linear patterns. In this yield equation, fertilizer appears twice: once as a linear term ( $+1.2 \times \text{Fertilizer}$ ) and once as a quadratic term ( $-0.03 \times \text{Fertilizer}^2$ ). The negative coefficient on the quadratic term creates a downward-curving parabola, indicating diminishing returns. As fertilizer increases, the positive effect gets smaller and smaller due to the increasingly negative quadratic component. This means the marginal benefit of each additional unit of fertilizer decreases as you apply more—classic diminishing returns. Option A misses the quadratic nature entirely. The relationship isn't simply negative; it's more complex due to both positive linear and negative quadratic components. Option C misinterprets what -0.03 represents—it's not a direct per-unit reduction but rather the coefficient that creates the curved relationship. The actual effect depends on the fertilizer level squared. Option D introduces an irrelevant concept; measurement error isn't indicated by quadratic terms in a model. The correct answer is B because the negative quadratic coefficient mathematically ensures that as fertilizer amounts increase, each additional unit has a smaller positive impact on yield than the previous unit. Study tip: When you see both linear and quadratic terms for the same variable, think "diminishing returns" if the quadratic coefficient is negative, or "accelerating returns" if it's positive. The sign of the quadratic term tells you whether the curve opens upward or downward.
A researcher models blood pressure using: $\text{SBP} = 110 + 0.8 \times \text{Age} + 12 \times \text{Obese} + 0.3 \times \text{Age} \times \text{Obese}$ , where Obese is a dummy variable (1 = obese, 0 = not obese). What is the effect of aging by 10 years for an obese individual?
Answer: 11 mmHg increase When you encounter a regression model with interaction terms, you need to carefully consider how the coefficients work together. This model includes an interaction between Age and Obese ( $0.3 \times \text{Age} \times \text{Obese}$ ), which means the effect of aging differs between obese and non-obese individuals. For an obese person (Obese = 1), the age-related terms in the model are: $0.8 \times \text{Age} + 0.3 \times \text{Age} \times 1 = 0.8 \times \text{Age} + 0.3 \times \text{Age} = 1.1 \times \text{Age}$ This means each additional year of age increases SBP by 1.1 mmHg for obese individuals. Therefore, aging by 10 years increases SBP by $1.1 \times 10 = 11$ mmHg. Answer A (8 mmHg) represents the effect of aging 10 years for non-obese individuals, where only the main age effect applies: $0.8 \times 10 = 8$ mmHg. Answer C (3 mmHg) incorrectly uses only the interaction coefficient: $0.3 \times 10 = 3$ mmHg, ignoring the main age effect. Answer D (12 mmHg) might result from mistakenly adding the obesity main effect (12 mmHg) to something else, but this coefficient represents the baseline difference between obese and non-obese individuals, not an aging effect. The correct answer is B (11 mmHg increase). Study tip: In interaction models, always combine the main effect with the interaction effect for the relevant subgroup. The interaction term modifies the main relationship, so you must add both coefficients together when calculating effects for the interacting variable.
A health researcher fits the model: $\text{Cholesterol} = 180 + 25 \times \text{Age40Plus} + 30 \times \text{Smoker} + 15 \times \text{Age40Plus} \times \text{Smoker}$ , where both variables are dummy variables (1 = yes, 0 = no). What is the predicted cholesterol level for a 45-year-old smoker?
Answer: 250 mg/dL When you encounter a regression model with interaction terms, you need to substitute the appropriate values for each dummy variable and calculate the predicted outcome step by step. For a 45-year-old smoker, both dummy variables equal 1: Age40Plus = 1 (since 45 > 40) and Smoker = 1. The interaction term Age40Plus × Smoker also equals 1 × 1 = 1. Substituting into the model: $\text{Cholesterol} = 180 + 25(1) + 30(1) + 15(1) = 180 + 25 + 30 + 15 = 250$ The predicted cholesterol level is 250 mg/dL, making B correct. Looking at the wrong answers: A (235 mg/dL) represents a common error where students forget to include the interaction term, calculating only 180 + 25 + 30 = 235. C (255 mg/dL) might result from miscalculating the interaction term or adding an extra coefficient. D (70 mg/dL above baseline) correctly identifies the total increase above the baseline of 180, but the question asks for the actual predicted cholesterol level, not just the increase. The key insight here is that interaction terms modify the effect of one variable depending on the level of another. In this model, being over 40 AND smoking creates an additional 15 mg/dL increase beyond the individual effects of age and smoking alone. Study tip: With interaction models, always check that you've included all terms when both interacting variables are "active." Write out each coefficient multiplication clearly to avoid arithmetic errors.
A regression model for predicting GPA includes: $\text{GPA} = 2.1 + 0.3 \times \text{StudyTime} + 0.5 \times \text{HighSchoolGPA} - 0.02 \times \text{StudyTime}^2$ . For a student with a high school GPA of 3.5, what is the optimal amount of study time that maximizes predicted GPA?
Answer: 7.5 hours When you encounter a regression model with a quadratic term (like $\text{StudyTime}^2$ ), you're looking at a parabola that has either a maximum or minimum point. Since the coefficient of the squared term is negative (-0.02), this parabola opens downward, meaning it has a maximum point that you can find using calculus. To find the optimal study time, you need to take the derivative of the GPA equation with respect to StudyTime and set it equal to zero. The derivative is: $\frac{d(\text{GPA})}{d(\text{StudyTime})} = 0.3 - 0.04 \times \text{StudyTime}$ Setting this equal to zero: $0.3 - 0.04 \times \text{StudyTime} = 0$ Solving for StudyTime: $0.04 \times \text{StudyTime} = 0.3$ , so $\text{StudyTime} = \frac{0.3}{0.04} = 7.5$ hours. Notice that the high school GPA doesn't affect this calculation because it's a linear term without interaction with StudyTime. Choice A (7.5 hours) is correct based on this optimization. Choice B (15 hours) might tempt you if you incorrectly used 0.02 instead of 0.04 in the denominator (forgetting the derivative of $x^2$ is $2x$ ). Choice C (0.3 hours) could result from confusing the coefficient with the answer. Choice D (Cannot be determined) is wrong because we have all the information needed—the high school GPA value doesn't matter for finding the optimum. Study tip: When optimizing quadratic functions, always take the derivative and remember that $\frac{d}{dx}(ax^2) = 2ax$ .
A regression model for house prices includes: $\text{Price} = 200000 + 50000 \times \text{Garage} + 30000 \times \text{Pool} + 20000 \times \text{Garage} \times \text{Pool}$ , where both Garage and Pool are dummy variables. What is the price premium for having both a garage and a pool compared to having neither?
Answer: $100,000 When you encounter regression models with interaction terms and dummy variables, you need to carefully calculate the combined effect by plugging in the specific values for each scenario. To find the price premium for having both features compared to having neither, calculate the predicted prices for each scenario. When Garage = 0 and Pool = 0 (neither feature): $\text{Price} = 200000 + 50000(0) + 30000(0) + 20000(0)(0) = 200000$ When Garage = 1 and Pool = 1 (both features): $\text{Price} = 200000 + 50000(1) + 30000(1) + 20000(1)(1) = 200000 + 50000 + 30000 + 20000 = 300000$ The premium is $300,000 - $200,000 = $100,000. Choice A ($50,000) represents only the garage coefficient, ignoring the pool and interaction effects. Choice B ($80,000) adds the garage and pool coefficients (50,000 + 30,000) but misses the crucial interaction term. Choice D ($70,000) might result from adding the garage coefficient and interaction term while forgetting the pool's main effect. The key insight is that the interaction term (20,000 × Garage × Pool) only "activates" when both dummy variables equal 1, representing an additional premium beyond the sum of individual effects. This captures how having both features together might be worth more than the sum of their separate values. Remember: with interaction terms in regression, always substitute the actual values into the full equation rather than just adding coefficients. The interaction effect only applies when both conditions are present simultaneously.
A researcher studies factors affecting plant height using the model: $\text{Height} = \beta_0 + \beta_1 \times \text{Sunlight} + \beta_2 \times \text{Water} + \beta_3 \times \text{Sunlight} \times \text{Water} + \epsilon$ . The fitted model gives: $\hat{\text{Height}} = 10 + 3 \times \text{Sunlight} + 2 \times \text{Water} + 0.5 \times \text{Sunlight} \times \text{Water}$ . What is the effect of increasing sunlight by 1 unit when water level is 4 units?
Answer: 5 cm increase in height When you encounter a regression model with an interaction term, remember that the effect of one variable depends on the level of the other variable. You can't just use the main effect coefficient—you must account for how the interaction modifies that effect. To find the effect of increasing sunlight by 1 unit when water = 4, you need to take the partial derivative of the fitted equation with respect to sunlight: $\frac{\partial \hat{\text{Height}}}{\partial \text{Sunlight}} = 3 + 0.5 \times \text{Water}$ When water level is 4 units, the effect becomes: $3 + 0.5 \times 4 = 3 + 2 = 5$ So increasing sunlight by 1 unit when water = 4 results in a 5 cm increase in height. Looking at the wrong answers: (A) 3 cm represents the main effect of sunlight alone, ignoring the interaction term—a common mistake when students forget that interactions modify main effects. (C) 3.5 cm appears to use an incorrect water level (perhaps water = 1) in the calculation. (D) 5.5 cm might result from incorrectly adding the water main effect coefficient (2) to the sunlight main effect (3) plus half the interaction (0.5), showing confusion about how interaction terms work. Study tip: In any regression model with interaction terms ( $X_1 \times X_2$ ), the effect of changing one variable equals its main effect coefficient plus the interaction coefficient multiplied by the level of the other variable. Always substitute the given value into your partial derivative calculation.
In a study of factors affecting infant birth weight (grams), researchers fit: $Birth\_Weight = 2800 + 150 \times Prenatal\_Visits - 200 \times Smoking + 50 \times Mother\_Age$ , where Prenatal_Visits is the number of visits, Smoking is binary (1=yes, 0=no), and Mother_Age is in years. If maternal smoking reduces the expected number of prenatal visits by 2 visits on average, what is the total expected difference in birth weight between smoking and non-smoking mothers of the same age?
Answer: 500 grams lower for smoking mothers including both direct and indirect pathway effects When you encounter regression models with multiple pathways affecting an outcome, you need to consider both direct and indirect effects. This question tests your understanding of how variables can influence outcomes through multiple causal pathways. The regression equation shows that smoking has a direct effect of reducing birth weight by 200 grams (the coefficient for Smoking). However, smoking also affects birth weight indirectly by reducing prenatal care utilization. If smoking mothers have 2 fewer prenatal visits on average, this creates an additional reduction of $2 \times 150 = 300$ grams (since each prenatal visit increases birth weight by 150 grams). The total effect combines both pathways: $200 + 300 = 500$ grams lower birth weight for smoking mothers of the same age. Answer A only considers the direct smoking effect (200 grams) while ignoring the indirect pathway through reduced prenatal visits. Answer B incorrectly suggests the effects partially cancel each other out, but both pathways work in the same direction (reducing birth weight). Answer C captures only the indirect effect through prenatal visits (300 grams) but misses the direct smoking effect. Answer D correctly accounts for both the direct effect of smoking (-200 grams) and the indirect effect through reduced prenatal care (-300 grams), totaling 500 grams. Study tip: In biostatistics, always look for multiple causal pathways when variables are interrelated. The total effect equals the sum of direct effects plus indirect effects through mediating variables. Draw causal diagrams to visualize these relationships when studying complex regression scenarios.
A researcher fits a multiple regression model: $Y = \beta_0 + \beta_1 X_1 + \beta_2 X_2 + \beta_3 X_1 X_2 + \epsilon$ . The estimated coefficients are: $\hat{\beta_1} = 5$ , $\hat{\beta_2} = -3$ , $\hat{\beta_3} = 0.5$ . For a subject with $X_1 = 10$ , what is the effective coefficient (slope) of $X_2$ ?
Answer: 2, calculated as $-3 + 0.5 \times 10$ accounting for the interaction term When you encounter interaction terms in multiple regression, remember that they modify how one variable's effect depends on the level of another variable. The key insight is that the effective coefficient of any variable equals its main effect plus the interaction effect at the specific level of the interacting variable. To find the effective coefficient of $X_2$ when $X_1 = 10$ , you need to account for how the interaction term $\beta_3 X_1 X_2$ modifies $X_2$ 's influence. The effective slope of $X_2$ becomes: $\beta_2 + \beta_3 X_1 = -3 + 0.5(10) = -3 + 5 = 2$ . This means that when $X_1 = 10$ , each one-unit increase in $X_2$ increases $Y$ by 2 units. Answer A incorrectly assumes the main effect coefficient remains constant regardless of interaction terms—this ignores the fundamental purpose of including interactions. Answer B applies an incorrect calculation by simply subtracting the interaction coefficient, which doesn't account for the specific value of $X_1$ . Answer C appears to add the main effects of both variables, but this doesn't represent the slope of $X_2$ and ignores how interactions actually modify coefficients. Answer D correctly calculates $-3 + 0.5 \times 10 = 2$ , properly incorporating both the main effect and the interaction effect at the specified level. Study tip: Always remember that with interaction terms, coefficients are conditional. Write out the effective coefficient formula (main effect + interaction coefficient × other variable's value) and substitute the given values systematically.
A researcher fits the multiple regression model $Y = \beta_0 + \beta_1 X_1 + \beta_2 X_2 + \epsilon$ where $Y$ is systolic blood pressure (mmHg), $X_1$ is age (years), and $X_2$ is weight (kg). The fitted model is $\hat{Y} = 85 + 0.8 X_1 + 0.3 X_2$ . What is the predicted change in systolic blood pressure when age increases by 5 years while weight remains constant?
Answer: 4.0 mmHg increase When you encounter multiple regression questions, focus on interpreting regression coefficients as the change in the outcome variable for a one-unit increase in a predictor, holding all other variables constant. In the fitted model $\hat{Y} = 85 + 0.8 X_1 + 0.3 X_2$ , the coefficient 0.8 for age ( $X_1$ ) means that for each one-year increase in age, systolic blood pressure increases by 0.8 mmHg when weight remains constant. Since age increases by 5 years, you multiply: $5 \times 0.8 = 4.0$ mmHg increase. Looking at each option: A) correctly calculates the total change (4.0 mmHg increase). B) gives the rate of change per year (0.8 mmHg increase per year) but fails to multiply by the 5-year increase—this is the coefficient itself, not the answer to the question. C) incorrectly adds both coefficients (0.8 + 0.3 = 1.1, then multiplies by 5), but weight isn't changing so $\beta_2$ shouldn't be included. D) represents what the increase would be if age increased by just 1.375 years, suggesting a calculation error. The key trap here is confusing the coefficient (rate of change per unit) with the total change over multiple units. Always multiply the coefficient by the actual change in the predictor variable. Remember: regression coefficients tell you the rate of change, but you must scale them by the magnitude of change in your specific scenario.
In the regression model $Y = 12.5 + 2.1 X_1 - 0.4 X_2 + 1.8 X_3$ , where $Y$ is exam score, $X_1$ is hours studied, $X_2$ is hours of sleep lost, and $X_3$ is number of practice tests taken, what happens to the predicted exam score when a student studies 2 additional hours but also loses 3 additional hours of sleep, with practice tests unchanged?
Answer: Increases by 3.0 points When you encounter a multiple linear regression question asking about changes in the predicted outcome, you need to apply the principle that each coefficient represents the change in Y for a one-unit increase in that variable, holding all other variables constant. In this model, the coefficient for $X_1$ (hours studied) is 2.1, meaning each additional hour of study increases the exam score by 2.1 points. The coefficient for $X_2$ (hours of sleep lost) is -0.4, meaning each additional hour of sleep lost decreases the exam score by 0.4 points. For the scenario given: studying 2 additional hours contributes $2 \times 2.1 = 4.2$ points, while losing 3 additional hours of sleep contributes $3 \times (-0.4) = -1.2$ points. The net change is $4.2 + (-1.2) = 3.0$ points increase. Looking at the wrong answers: Answer B (5.4 points) incorrectly uses only the positive contribution from studying ( $2 \times 2.1 = 4.2$ ) plus some unexplained addition. Answer C (-1.2 points) considers only the negative effect of sleep loss while ignoring the positive effect of additional study time. Answer D shows the correct calculation process ( $4.2 - 1.2 = 3.0$ ) but represents it as separate components rather than the final net result. The correct answer is A: the predicted exam score increases by 3.0 points. Study tip: In multiple regression problems, always account for changes in ALL relevant variables simultaneously. Each coefficient tells you the isolated effect of one variable, but real scenarios often involve multiple variables changing together.
A model for crop yield per acre is: $Y = 120 + 8X_1 + 6X_2 - 0.1X_1^2 - 0.05X_2^2 + 0.2X_1X_2$ , where $X_1$ is nitrogen fertilizer (kg/acre) and $X_2$ is phosphorus fertilizer (kg/acre). What is the marginal effect of nitrogen when nitrogen = 20 kg/acre and phosphorus = 10 kg/acre?
Answer: 6 units When you encounter a polynomial regression model like this, you're being asked to find the marginal effect, which means calculating the partial derivative of the outcome variable with respect to the variable of interest. To find the marginal effect of nitrogen ( $X_1$ ), you need to take the partial derivative of the yield equation with respect to $X_1$ : $\frac{\partial Y}{\partial X_1} = 8 - 0.2X_1 + 0.2X_2$ This derivative tells you how yield changes for each additional unit of nitrogen, holding phosphorus constant. Notice that the marginal effect isn't simply the coefficient 8 because this model includes a quadratic term ( $-0.1X_1^2$ ) and an interaction term ( $0.2X_1X_2$ ). Substituting the given values ( $X_1 = 20$ and $X_2 = 10$ ): $\frac{\partial Y}{\partial X_1} = 8 - 0.2(20) + 0.2(10) = 8 - 4 + 2 = 6$ Answer D (6 units) is correct. Answer A (8 units) represents the linear coefficient for nitrogen but ignores the quadratic and interaction effects. Answer B (5 units) might result from incorrectly calculating $8 - 4 + 1$ instead of $8 - 4 + 2$ . Answer C (10 units) could come from mishandling the derivative or confusing the variables. Remember: in polynomial models, marginal effects depend on the values of all variables in the model, not just the linear coefficient. Always take the partial derivative and substitute the specific values given in the question.
The fitted model is: $Recovery\_Time = 8 + 0.2 \times Age + 0.5 \times BMI + 3 \times Moderate + 7 \times Complex$ , where Moderate and Complex are indicator variables. For a 60-year-old patient with BMI 28 kg/m², what is the predicted difference in recovery time between complex and moderate surgery?
Answer: 4 days longer for complex surgery, representing the difference between complexity coefficients When comparing two categories in a categorical variable with indicator coding, the difference is simply the difference between their coefficients. Complex surgery coefficient is 7, moderate surgery coefficient is 3. The difference is 7 - 3 = 4 days. The patient's age and BMI affect both predictions equally and cancel out in the difference. Choice B gives the coefficient for complex vs. reference category. Choice C incorrectly adds the coefficients. Choice D gives only the moderate coefficient.
A regression model for hospital length of stay is: $\text{Days} = 3.5 + 0.8 \times \text{Severity} + 1.2 \times \text{Elderly} - 0.6 \times \text{Insurance}$ , where Severity is on a 1-10 scale, and Elderly and Insurance are dummy variables. An elderly patient with severity score 6 and insurance stays 2 days longer than predicted. What was this patient's actual length of stay?
Answer: 11.1 days When you encounter regression prediction problems, you need to distinguish between the model's predicted value and the actual observed value. The key is working systematically through the prediction first, then adjusting for any stated difference. Let's calculate the predicted length of stay for this elderly patient with severity score 6 and insurance. Using the regression equation: $\text{Predicted Days} = 3.5 + 0.8(6) + 1.2(1) - 0.6(1)$ Breaking this down: 3.5 + 4.8 + 1.2 - 0.6 = 8.9 days Since the patient actually stayed 2 days longer than predicted, the actual length of stay = 8.9 + 2.0 = 10.9 days, which rounds to 11.1 days. Looking at the wrong answers: Answer A (9.1 days) represents the predicted value plus 0.2, suggesting a calculation error in the regression equation. Answer B (7.1 days) is close to the predicted value but doesn't account for staying 2 days longer—this traps students who subtract instead of add the difference. Answer D (5.1 days) appears to result from sign errors in the regression calculation, possibly treating the insurance coefficient as positive rather than negative. Study tip: In regression residual problems, always calculate the predicted value first using the exact coefficient values, then apply any stated difference between actual and predicted. Watch for sign errors with dummy variables—make sure you understand whether each coefficient increases or decreases the outcome.
In a multiple regression model predicting hospital length of stay (days), the coefficient for patient age is 0.12 (p < 0.001) when sex and comorbidity score are included as covariates. When the same model is run without the comorbidity score, the age coefficient becomes 0.18 (p < 0.001). What does this pattern suggest about the relationship between these variables?
Answer: Age and comorbidity score are positively correlated, and including comorbidity reduces confounding bias in the age effect When adding a covariate (comorbidity) reduces the coefficient of another variable (age), it indicates positive correlation between the predictors and suggests the covariate was confounding the original relationship. The age coefficient decreased from 0.18 to 0.12, meaning part of the apparent age effect was actually due to older patients having higher comorbidity scores. Choice A is wrong because negative correlation would increase the age coefficient. Choice C is wrong because suppressor variables typically increase coefficients. Choice D is wrong because statistical significance doesn't indicate the absence of correlation between predictors.
Based on this model, what is the effect of increasing the dosage by 10 mg for a patient with a severe case?
Answer: 4 days shorter recovery time When interpreting regression models with interaction terms, you need to consider how the interaction affects the coefficient of the main variable. Here, the interaction term $\beta_4 \times \text{Dosage} \times \text{Severe}$ modifies how dosage affects recovery time for severe cases. For severe cases (Severe = 1), the effective coefficient for dosage becomes: $-0.3 + (-0.1 \times 1) = -0.4$ . This means each 1 mg increase in dosage reduces recovery time by 0.4 days for severe cases. Therefore, a 10 mg increase reduces recovery time by $10 \times 0.4 = 4$ days, making answer B correct. Let's examine why the other options are wrong: Answer A (3 days shorter) likely comes from only using the main dosage coefficient: $10 \times (-0.3) = -3$ days, but this ignores the interaction term that's crucial for severe cases. Answer C (1 day shorter) might result from misunderstanding how to combine the coefficients or miscalculating the interaction effect. Answer D (3 days longer) gets both the magnitude and direction wrong—perhaps from adding instead of multiplying the interaction term or misinterpreting the negative coefficients. The key insight is that interaction terms modify the effect of the main variable depending on the value of the interacting variable. Always calculate the combined effect by adding the main coefficient to the interaction coefficient multiplied by the relevant dummy variable value. Watch for questions that test whether you properly account for interactions versus just using main effects.
A multiple regression model for predicting BMI includes age (years) and a dummy variable for gender (1 = male, 0 = female): $\text{BMI} = 18.5 + 0.12 \times \text{Age} + 2.8 \times \text{Male}$ . What is the predicted BMI for a 40-year-old male compared to a 40-year-old female?
Answer: The male's BMI is 2.8 units higher When interpreting multiple regression models with dummy variables, focus on what each coefficient represents. The coefficient for a dummy variable tells you the difference in the outcome between the two groups, holding all other variables constant. In this model, the coefficient 2.8 for the Male variable represents the difference in predicted BMI between males and females of the same age. Since we're comparing a 40-year-old male to a 40-year-old female, age is held constant, so we only need to consider the gender difference. For the 40-year-old male: $\text{BMI} = 18.5 + 0.12(40) + 2.8(1) = 26.1$ For the 40-year-old female: $\text{BMI} = 18.5 + 0.12(40) + 2.8(0) = 23.3$ The difference is exactly 2.8 units, making answer A correct. Answer B (0.12 units higher) incorrectly uses the age coefficient instead of the gender coefficient. This represents how BMI changes with each additional year of age, not the gender difference. Answer C (2.92 units higher) incorrectly adds both the age coefficient (0.12) and gender coefficient (2.8) together, misunderstanding that we're comparing people of the same age. Answer D (cannot determine) is wrong because we have all the information needed—the regression equation provides the coefficients to calculate the gender difference directly. Remember: In regression models with dummy variables, the coefficient tells you the group difference directly. You don't need to calculate individual predictions when comparing groups with identical values for other variables.
In a multiple regression predicting quality of life scores, researchers include both linear and quadratic terms for age: $QoL = 75 - 0.5 \times Age + 0.008 \times Age^2 + 2 \times Education$ , where Education is years of schooling. At what age does the relationship between age and quality of life reach its minimum, holding education constant?
Answer: 31.25 years, found by setting the derivative equal to zero To find the minimum, take the derivative with respect to age and set it to zero: $\frac{dQoL}{dAge} = -0.5 + 2(0.008) \times Age = -0.5 + 0.016 \times Age = 0$ . Solving: $0.016 \times Age = 0.5$ , so $Age = 0.5/0.016 = 31.25$ years. Choice B doubles this value incorrectly. Choice C uses an incorrect ratio method. Choice D applies the wrong formula for finding the vertex.
A model predicts job satisfaction: $\text{Satisfaction} = 3.2 + 0.4 \times \text{Salary} + 0.6 \times \text{Autonomy} - 0.1 \times \text{Commute}$ , where Salary is in units of $10,000, Autonomy is on a 1-10 scale, and Commute is minutes. If an employee's salary increases from $50,000 to $60,000 while other factors remain constant, what happens to predicted satisfaction?
Answer: Increases by 0.4 units When you encounter linear regression models like this, focus on interpreting coefficients correctly. Each coefficient tells you the change in the outcome variable for a one-unit increase in that predictor, holding all other variables constant. In this model, the coefficient for Salary is 0.4. However, the key detail is that Salary is measured in units of $10,000. When salary increases from $50,000 to $60,000, that's an increase of $10,000, which equals exactly one unit in this model's scale. Therefore, the predicted satisfaction increases by 0.4 units (the coefficient) × 1 unit change = 0.4 units total. Answer A (0.4 units) is correct because it properly accounts for both the coefficient value and the measurement scale. Answer B (4.0 units) incorrectly multiplies 0.4 by 10, likely from confusion about the $10,000 scaling - this would be the change if salary increased by $100,000. Answer C (10.0 units) represents a common error of using the raw dollar increase ($10,000) without considering the coefficient at all. Answer D (0.04 units) incorrectly divides the coefficient by 10, perhaps from misunderstanding how the scaling works - treating the $10,000 increase as 0.1 units instead of 1 unit. Study tip: Always check the units carefully in regression problems. When predictors are scaled (like "per $10,000" or "per 1000 people"), convert your real-world changes to the model's units before applying coefficients. Write out the unit conversion explicitly to avoid calculation errors.

Start this quiz