Indicator Variables - Biostatistics

1. A researcher is studying the relationship between systolic blood pressure (SBP) and age, with gender as a categorical predictor. The regression model is: $SBP = 90 + 1.2 \times Age + 8 \times Gender$, where Gender = 1 for males and Gender = 0 for females. What is the predicted difference in systolic blood pressure between a 45-year-old male and a 45-year-old female?
   Answer: 8 mmHg higher for males When you encounter a multiple linear regression model with categorical variables, you're analyzing how different groups compare while controlling for other factors. Here, the model separates the effects of age and gender on blood pressure. To find the difference between a 45-year-old male and female, calculate each prediction separately. For the male (Gender = 1): $SBP = 90 + 1.2(45) + 8(1) = 90 + 54 + 8 = 152$ mmHg. For the female (Gender = 0): $SBP = 90 + 1.2(45) + 8(0) = 90 + 54 + 0 = 144$ mmHg. The difference is 152 - 144 = 8 mmHg higher for males. Actually, there's a shortcut: since both individuals are the same age, the age term cancels out when comparing them. The only difference comes from the gender coefficient, which is 8 mmHg. Answer A correctly identifies this 8 mmHg difference favoring males. Answer B gets the magnitude right but reverses the direction—males have higher predicted blood pressure, not females. Answer C confuses the gender coefficient (8) with the age coefficient (1.2), showing a misunderstanding of which variable creates the difference. Answer D incorrectly adds the age and gender coefficients (1.2 + 8 = 9.2), failing to recognize that age effects are identical for both individuals. Remember: when comparing groups in regression models, only the coefficients of variables that differ between groups contribute to the predicted difference. Identical values for other variables cancel out in the comparison.
2. In analyzing hospital length of stay, a researcher includes patient type as a predictor with three categories: Emergency (E), Elective Surgery (S), and Routine (R). Using Routine as the reference category, the model is: $Stay = 3.5 + 2.1 \times Emergency + 1.6 \times Surgery$. A patient in the Emergency category is expected to stay how many days longer than a Routine patient?
   Answer: 2.1 days longer than Routine patients When you encounter regression models with categorical predictors, you're seeing dummy variable encoding in action. The key insight is understanding what the coefficients represent when you have a reference category. In this model, "Routine" is the reference category, meaning it serves as the baseline for comparison. The equation $Stay = 3.5 + 2.1 \times Emergency + 1.6 \times Surgery$ uses dummy variables where Emergency = 1 for emergency patients (0 otherwise) and Surgery = 1 for elective surgery patients (0 otherwise). For a Routine patient: Emergency = 0 and Surgery = 0, so predicted stay = 3.5 + 2.1(0) + 1.6(0) = 3.5 days. For an Emergency patient: Emergency = 1 and Surgery = 0, so predicted stay = 3.5 + 2.1(1) + 1.6(0) = 5.6 days. The difference is 5.6 - 3.5 = 2.1 days longer for Emergency patients compared to Routine patients. This matches the coefficient directly—each coefficient in dummy variable regression represents the expected difference from the reference category. Answer A (1.6 days) confuses the Surgery coefficient with the Emergency coefficient. Answer C (3.7 days) incorrectly adds the intercept to the Emergency coefficient. Answer D (5.6 days) gives the absolute predicted stay for Emergency patients rather than the difference from Routine patients. Remember: in dummy variable regression, each coefficient tells you the expected difference from the reference category. You don't need to calculate both predicted values—the coefficient IS the difference you're looking for.
3. A researcher studying hospital length of stay creates indicator variables for admission department. There are 5 departments: Emergency, Surgery, Medicine, Pediatrics, and Psychiatry. If Medicine is chosen as the reference category, how many indicator variables should be included in the regression model to avoid multicollinearity?
   Answer: 4 indicator variables should be included in the regression model With k categories, only k-1 indicator variables are needed to avoid perfect multicollinearity. With 5 departments and Medicine as reference, we need 5-1=4 indicator variables (Emergency, Surgery, Pediatrics, Psychiatry). Choice B would create perfect multicollinearity. Choice C miscounts by excluding two categories. Choice D incorrectly suggests we need variables for combinations rather than individual categories.
4. A clinical study models blood glucose levels using: $Glucose = 95 + 25 \times Diabetic + 8 \times Age$, where Diabetic = 1 for diabetic patients (0 for non-diabetic) and Age is measured in years. Two patients are both 50 years old, but one is diabetic and one is non-diabetic. What is the predicted glucose level for the diabetic patient?
   Answer: 520 mg/dL blood glucose level When you encounter a linear regression equation in biostatistics, you're looking at a mathematical model that predicts an outcome based on input variables. This equation shows how blood glucose levels depend on diabetes status and age. To find the predicted glucose level for the diabetic patient, substitute the values into the equation: $Glucose = 95 + 25 \times Diabetic + 8 \times Age$. For a 50-year-old diabetic patient, Diabetic = 1 and Age = 50, so: $Glucose = 95 + 25(1) + 8(50) = 95 + 25 + 400 = 520$ mg/dL. Looking at the wrong answers: A) 495 mg/dL likely comes from forgetting to add the baseline intercept of 95, calculating only $25 + 8(50) = 425$, then making an arithmetic error. C) 170 mg/dL represents a calculation where you might have used Age = 50 but forgotten the diabetes coefficient, getting $95 + 8(50) = 495$, then making another error. D) 145 mg/dL could result from incorrectly calculating $95 + 25 + 8(50/10)$ or similar computational mistakes with the age coefficient. The key study tip for regression problems: always substitute values carefully and follow the order of operations. Write out each step explicitly rather than trying to do mental math, especially with multiple coefficients. Remember that each coefficient tells you how much the outcome changes per unit increase in that variable, holding other variables constant.
5. In a medical study, treatment response is modeled as: $Response = 45 + 15 \times NewDrug - 8 \times Elderly$, where NewDrug = 1 for patients receiving the new drug (0 for standard treatment) and Elderly = 1 for patients over 65 (0 for patients 65 and under). What is the expected treatment response for an elderly patient receiving the new drug?
   Answer: 52 units of treatment response When you encounter a linear regression equation in biostatistics, you're looking at a mathematical model that predicts an outcome based on specific input variables. Each coefficient tells you how much the response changes when that variable increases by one unit, holding other variables constant. To find the expected response for an elderly patient receiving the new drug, substitute the appropriate values into the equation. For this patient: NewDrug = 1 (receiving new drug) and Elderly = 1 (over 65 years old). $Response = 45 + 15(1) - 8(1) = 45 + 15 - 8 = 52$ The calculation shows that answer A (52 units) is correct. Let's examine why the other answers are wrong. Answer B (60 units) represents what you'd get if you incorrectly added both coefficients: 45 + 15 + 8 = 68, or perhaps forgot the negative sign entirely. Answer C (37 units) would result from subtracting both the drug effect and elderly effect: 45 - 15 + 8 = 38, suggesting confusion about which coefficients are positive versus negative. Answer D (45 units) is simply the baseline response when both NewDrug = 0 and Elderly = 0, ignoring the patient's actual characteristics entirely. Remember to pay careful attention to the signs of coefficients in regression equations. A negative coefficient (like -8 for Elderly) means that variable decreases the outcome, while a positive coefficient increases it. Always substitute the correct binary values (0 or 1) based on whether each condition applies to your specific case.
6. A study of academic performance models GPA using: $GPA = 2.8 + 0.6 \times Honors + 0.2 \times Tutoring$, where Honors = 1 for honors program students (0 otherwise) and Tutoring = 1 for students receiving tutoring (0 otherwise). A student not in the honors program and not receiving tutoring has a predicted GPA that differs from the model's intercept by how much?
   Answer: The predicted GPA differs from the intercept by 0.0 points When you encounter linear regression equations with binary predictor variables, remember that the intercept represents the predicted outcome when all predictor variables equal zero. Let's substitute the values for a student not in honors and not receiving tutoring into the equation. For this student, Honors = 0 and Tutoring = 0: $GPA = 2.8 + 0.6(0) + 0.2(0) = 2.8 + 0 + 0 = 2.8$ The intercept in this model is 2.8, which is the constant term. Since the predicted GPA for this student is exactly 2.8, the difference between the predicted GPA and the intercept is 2.8 - 2.8 = 0. Choice D is correct because there's no difference between the predicted value and the intercept. Choice A incorrectly suggests the difference is 0.6 points, which is actually the coefficient for the Honors variable—but this student isn't in honors, so this doesn't apply. Choice B makes the same error with the Tutoring coefficient (0.2), assuming it affects students not receiving tutoring. Choice C adds both coefficients (0.6 + 0.2 = 0.8), wrongly thinking both predictors influence the baseline prediction even when they're absent. Study tip: In regression with binary variables, the intercept is the predicted outcome for the reference group (when all binary predictors = 0). The coefficients show how much the prediction changes when moving from the reference group to each comparison group, not the absolute predicted values themselves.
7. In a regression model predicting weekly healthcare costs, insurance type is coded using two indicator variables: $I_1$ for HMO (1 if HMO, 0 otherwise) and $I_2$ for PPO (1 if PPO, 0 otherwise). The reference category is 'Uninsured'. The model is: $Cost = 450 - 120I_1 - 85I_2$. What is the predicted cost difference between HMO and PPO patients?
   Answer: HMO patients cost $35 less than PPO patients When you encounter indicator variables in regression, remember that each coefficient represents the difference between that group and the reference category. Here, you need to find the difference between two non-reference groups. To find the predicted cost difference between HMO and PPO patients, you must compare their individual predictions relative to the reference group (Uninsured). For HMO patients: $I_1 = 1, I_2 = 0$, so predicted cost = $450 - 120(1) - 85(0) = 330$. For PPO patients: $I_1 = 0, I_2 = 1$, so predicted cost = $450 - 120(0) - 85(1) = 365$. The difference is $330 - 365 = -35$, meaning HMO patients cost $35 less than PPO patients. Option A correctly identifies this $35 difference with HMO being lower. Option B incorrectly uses the HMO coefficient (-120) directly, but this represents the difference between HMO and Uninsured, not HMO and PPO. Option C makes the same error as B but assigns the wrong direction, using the PPO coefficient (-85) and incorrectly stating PPO costs less. Option D adds the absolute values of both coefficients (120 + 85 = 205), which has no meaningful interpretation in this context. The key insight is that when comparing two non-reference groups in regression with indicator variables, you must calculate each group's predicted value separately, then find their difference. Don't confuse individual coefficients with between-group comparisons that don't involve the reference category.
8. A sleep study models hours of sleep using: $Sleep = 7.5 - 0.8 \times Shift + 0.3 \times Weekend$, where Shift = 1 for night shift workers (0 for day shift) and Weekend = 1 for weekend days (0 for weekdays). On weekdays, what is the predicted difference in sleep hours between day shift and night shift workers?
   Answer: Day shift workers sleep 0.8 hours more than night shift workers on weekdays When you encounter linear regression equations with categorical variables, focus on how the coefficients represent differences between groups. This equation uses dummy variables where Shift = 1 for night workers and Weekend = 1 for weekends. To find the weekday difference between day and night shift workers, substitute the relevant values. For weekdays, Weekend = 0 for both groups. Day shift workers (Shift = 0): $Sleep = 7.5 - 0.8(0) + 0.3(0) = 7.5$ hours Night shift workers (Shift = 1): $Sleep = 7.5 - 0.8(1) + 0.3(0) = 6.7$ hours The difference is 7.5 - 6.7 = 0.8 hours, with day shift workers sleeping more. Choice A correctly identifies that day shift workers sleep 0.8 hours more than night shift workers on weekdays. Choice B incorrectly reverses the direction—the negative coefficient for Shift means night workers sleep less, not more. Choice C confuses the Shift coefficient (-0.8) with the Weekend coefficient (0.3), mixing up which variable affects the comparison. Choice D appears to add the absolute values of both coefficients (0.8 + 0.3 = 1.1), but the Weekend variable doesn't apply to weekday comparisons. Remember that in regression with dummy variables, the coefficient directly tells you the difference between groups. A negative coefficient means the group coded as 1 has lower values than the reference group (coded as 0). Always pay attention to the sign and magnitude of coefficients when interpreting group differences.
9. A pharmaceutical company models drug efficacy using: $Efficacy = 65 + 12 \times DoseHigh + 8 \times Female$, where DoseHigh = 1 for high dose (0 for standard dose) and Female = 1 for female patients (0 for male patients). For which patient group is the predicted efficacy exactly 77?
   Answer: Male patients receiving high dose medication When you encounter a multiple regression equation in biostatistics, you're looking at a model that predicts an outcome based on multiple variables. This equation shows how drug efficacy changes based on dose level and patient gender, with each coefficient representing the additional effect of that variable. To find which group has an efficacy of exactly 77, you need to substitute the appropriate values into the equation for each patient type. Let's work through each option: For option A (male, standard dose): $Efficacy = 65 + 12(0) + 8(0) = 65$. Males are coded as 0 for the Female variable, and standard dose means DoseHigh = 0. For option B (female, standard dose): $Efficacy = 65 + 12(0) + 8(1) = 73$. Females get the +8 bonus, but no dose effect. For option C (male, high dose): $Efficacy = 65 + 12(1) + 8(0) = 77$. Males get no gender effect, but the high dose adds 12 points to the baseline of 65. For option D (female, high dose): $Efficacy = 65 + 12(1) + 8(1) = 85$. This group gets both the dose effect (+12) and gender effect (+8). Only option C yields exactly 77, making it correct. Study tip: In regression problems, always identify what each dummy variable represents (0 vs 1) and systematically substitute values. Don't try to solve these mentally—write out each calculation to avoid errors with the binary coding.
10. A researcher studying depression scores creates indicator variables for marital status. With 'Single' as the reference category, the model includes: Married (M = 1 if married, 0 otherwise) and Divorced (D = 1 if divorced, 0 otherwise). The fitted equation is: $Depression = 15.2 - 4.1M + 2.8D$. Which group has the highest predicted depression score?
    Answer: Divorced individuals have the highest predicted depression scores When you encounter regression equations with indicator variables, you're working with categorical predictors that compare each group to a reference category. Here, marital status is broken into indicator variables with "Single" as the baseline. To find predicted depression scores, substitute the appropriate values into the equation $Depression = 15.2 - 4.1M + 2.8D$. For single individuals, both M and D equal 0, giving: $15.2 - 4.1(0) + 2.8(0) = 15.2$. For married individuals, M = 1 and D = 0: $15.2 - 4.1(1) + 2.8(0) = 11.1$. For divorced individuals, M = 0 and D = 1: $15.2 - 4.1(0) + 2.8(1) = 18.0$. Divorced individuals have the highest predicted depression score at 18.0, making C correct. Option A is wrong because single individuals score 15.2, which is lower than divorced individuals' 18.0. Option B is incorrect since married individuals have the lowest score at 11.1 (the negative coefficient -4.1 reduces their depression score relative to singles). Option D is false because married and divorced groups have very different scores (11.1 vs 18.0). Remember this pattern: in regression with indicator variables, the reference group's predicted value equals the intercept. Other groups' values are the intercept plus their coefficient. Negative coefficients mean that group scores lower than the reference; positive coefficients mean higher scores. Always calculate each group's predicted value to compare them directly.
11. In a study of BMI predictors, education level is coded using indicator variables: $I_{\text{HS}}$ (high school graduate), $I_{\text{College}}$ (college graduate), $I_{\text{Grad}}$ (graduate degree), with 'less than high school' as reference. The fitted model is: $\text{BMI} = 28.5 - 1.2 \cdot I_{\text{HS}} - 2.8 \cdot I_{\text{College}} - 3.1 \cdot I_{\text{Grad}}$. What is the predicted BMI for someone with a college degree?
    Answer: 25.7 is the predicted BMI for someone with a college degree When you encounter a regression model with indicator variables, you're dealing with a categorical predictor where each category gets its own binary variable. The key insight is that only one indicator variable equals 1 at a time, while all others equal 0. For someone with a college degree, the indicator variables take these values: $I_{\text{HS}} = 0$ (they're not just a high school graduate), $I_{\text{College}} = 1$ (they are a college graduate), and $I_{\text{Grad}} = 0$ (they don't have a graduate degree). Substituting into the model: $\text{BMI} = 28.5 - 1.2(0) - 2.8(1) - 3.1(0) = 28.5 - 2.8 = 25.7$ Looking at the wrong answers: Choice A (27.3) incorrectly uses only the high school coefficient, as if $28.5 - 1.2 = 27.3$. Choice C (24.6) mistakenly subtracts both the college and graduate coefficients, calculating $28.5 - 2.8 - 3.1 = 22.6$ (though even this gives the wrong number). Choice D (31.3) appears to add rather than subtract a coefficient, perhaps calculating $28.5 + 2.8 = 31.3$. The correct answer is B: 25.7 represents the predicted BMI for someone with a college degree. Study tip: With indicator variables, always identify which single category applies to your case, set that indicator to 1, and set all others to 0. The reference category (here, "less than high school") automatically has all indicators equal to 0, giving you just the intercept value.
12. In a study examining factors affecting cholesterol levels, researchers use the model: $Cholesterol = \beta_0 + \beta_1 \times Exercise + \beta_2 \times Diet$. Exercise is coded as 1 for regular exercisers and 0 for non-exercisers. Diet is coded as 1 for vegetarian and 0 for non-vegetarian. If $\beta_1 = -15$ and $\beta_2 = -22$, what does the coefficient $\beta_1$ represent?
    Answer: The difference in cholesterol between exercisers and non-exercisers, holding diet constant When interpreting coefficients in multiple linear regression, each coefficient represents the change in the outcome variable for a one-unit increase in that predictor, while holding all other variables in the model constant. This "holding constant" concept is crucial for understanding what each coefficient actually means. The coefficient $\beta_1 = -15$ tells you that cholesterol levels decrease by 15 units when Exercise changes from 0 (non-exerciser) to 1 (regular exerciser), assuming Diet stays the same. This is exactly what option B describes - the difference in cholesterol between the two exercise groups, controlling for diet. Option A is incorrect because $\beta_1$ doesn't give you the actual cholesterol level for exercisers; it gives you the difference between groups. The actual level would require adding $\beta_0$ and potentially $\beta_2$ depending on diet status. Option C confuses $\beta_1$ with $\beta_0$. The intercept $\beta_0$ represents the cholesterol level when all predictors equal zero (non-exerciser, non-vegetarian), not $\beta_1$. Option D describes an interaction effect, which would require a third term like $\beta_3 \times Exercise \times Diet$ in the model. Individual coefficients in additive models show independent effects, not combined effects. Remember: In multiple regression, each coefficient isolates the effect of one variable by "controlling for" or "holding constant" all others in the model. This allows you to estimate the independent contribution of each predictor to the outcome.
13. A clinical trial compares three treatment groups: placebo, low-dose drug, and high-dose drug. To include treatment in a regression model predicting pain scores, a researcher creates two indicator variables: $X_1$ (1 if low-dose, 0 otherwise) and $X_2$ (1 if high-dose, 0 otherwise). In the fitted model $Pain = 7.2 - 1.8X_1 - 3.1X_2$, what is the predicted pain score for the placebo group?
    Answer: 7.2 points on the pain scale When analyzing regression models with categorical variables, you need to understand how indicator (dummy) variables work. The key insight is identifying which group serves as the reference category - the baseline against which all other groups are compared. In this model, you have three treatment groups but only two indicator variables: $X_1$ for low-dose and $X_2$ for high-dose. The placebo group isn't explicitly represented by an indicator variable, making it the reference category. For placebo patients, both $X_1 = 0$ and $X_2 = 0$. Substituting these values into the equation: $Pain = 7.2 - 1.8(0) - 3.1(0) = 7.2$. The predicted pain score for the placebo group is 7.2 points, making C correct. Looking at the wrong answers: A (2.3) represents a common error where students subtract both coefficients from the intercept (7.2 - 1.8 - 3.1 = 2.3), incorrectly assuming all terms apply to placebo. B (5.4) results from subtracting only one coefficient (either 7.2 - 1.8 = 5.4), suggesting confusion about which treatment the placebo represents. D (9.0) comes from adding one coefficient (7.2 + 1.8 = 9.0), reflecting a sign error or misunderstanding of how the model works. Remember this pattern: in regression with categorical variables, the reference category's predicted value always equals the intercept alone. All other coefficients represent differences from this baseline, so they drop out when the indicator variables equal zero.
14. A study of academic performance uses the model: $GPA = 2.1 + 0.3 \times StudyHours + 0.4 \times Private$, where Private = 1 for private school students and Private = 0 for public school students. If a private school student and a public school student both study 10 hours per week, what is the ratio of their predicted GPAs?
    Answer: The private school student's GPA is 1.08 times the public school student's GPA When you encounter a linear regression model with dummy variables, you need to substitute the specific values and calculate the predicted outcomes for each group being compared. Let's calculate the predicted GPA for each student with 10 study hours per week. For the public school student (Private = 0): $GPA_{public} = 2.1 + 0.3(10) + 0.4(0) = 2.1 + 3.0 + 0 = 5.1$ For the private school student (Private = 1): $GPA_{private} = 2.1 + 0.3(10) + 0.4(1) = 2.1 + 3.0 + 0.4 = 5.5$ The ratio of their predicted GPAs is: $\frac{5.5}{5.1} = 1.078$ Since this ratio (approximately 1.08) doesn't match any of the given options A through D, the correct answer must be E (not shown but implied as "none of the above"). Looking at why each option is wrong: Option A (1.19) is too high and likely results from miscalculating the baseline or study hours effect. Option B (1.15) is closer but still overestimates the ratio. Option C (0.4) incorrectly suggests the public school student has a higher GPA, possibly confusing the coefficient with the actual ratio. Option D (1.40) drastically overestimates the difference, perhaps by misunderstanding how dummy variables work. Remember that with dummy variables, the coefficient (0.4 here) represents the additive difference between groups, not a multiplicative effect. Always calculate the full predicted values before computing ratios, and don't assume the coefficient itself is the ratio you're looking for.
15. A medical device study models patient compliance using: $Compliance = 70 + 15 \times HighTech - 10 \times Elderly$, where HighTech = 1 for high-tech devices (0 for standard) and Elderly = 1 for patients over 70 (0 otherwise). The interaction between device type and age is not included in the model. What assumption does this model make about the effect of device type across age groups?
    Answer: High-tech devices improve compliance equally for both elderly and younger patients When you encounter a linear regression model without interaction terms, the key concept being tested is how variables behave independently across different groups. In this compliance model, the absence of an interaction term between HighTech and Elderly creates a specific assumption about how device type affects different age groups. Let's examine what this model predicts. For younger patients with standard devices: Compliance = 70 + 15(0) - 10(0) = 70. For younger patients with high-tech devices: Compliance = 70 + 15(1) - 10(0) = 85. The improvement is 15 points. For elderly patients with standard devices: Compliance = 70 + 15(0) - 10(1) = 60. For elderly patients with high-tech devices: Compliance = 70 + 15(1) - 10(1) = 75. Again, the improvement is exactly 15 points. This demonstrates that answer B is correct – high-tech devices improve compliance equally (by 15 points) for both age groups. Answer A is wrong because the model shows identical improvements, not greater benefits for elderly patients. Answer C is incorrect because younger patients don't receive greater benefits – the 15-point improvement is constant. Answer D misses that high-tech devices do improve compliance for both groups, just by the same amount. The model assumes "additive effects" – each variable contributes independently without the effect of one variable changing based on the level of another. Remember: when you see a regression model without interaction terms, the effect of each predictor remains constant across all levels of other predictors.
16. A study of reaction time uses: $Time = 250 + 45 \times Caffeine - 20 \times Young + 15 \times Male$, where Caffeine = 1 for participants who consumed caffeine (0 otherwise), Young = 1 for participants under 30 (0 for 30+), and Male = 1 for male participants (0 for female). What is the predicted reaction time for a young female who consumed caffeine?
    Answer: 275 milliseconds reaction time When you encounter a multiple linear regression equation, you're working with a mathematical model that predicts an outcome based on several variables. Each coefficient tells you how much the outcome changes when that variable equals 1 (compared to 0), holding all other variables constant. To find the predicted reaction time for a young female who consumed caffeine, substitute the appropriate values into the equation. For this participant: Caffeine = 1 (consumed caffeine), Young = 1 (under 30), and Male = 0 (female). $Time = 250 + 45(1) - 20(1) + 15(0)$ $Time = 250 + 45 - 20 + 0 = 275$ The predicted reaction time is 275 milliseconds, making A correct. Looking at the wrong answers: B (290) might result from incorrectly adding all coefficients without considering the negative sign on Young or forgetting that Male = 0. C (310) could come from setting Male = 1 instead of 0, or adding all positive coefficients. D (230) might result from subtracting both the Young and Caffeine coefficients instead of recognizing that caffeine increases reaction time. The key insight is that caffeine increases reaction time by 45ms, being young decreases it by 20ms, and being female contributes 0 to the baseline. The net effect is +25ms above the 250ms baseline. Study tip: Always carefully identify which variables equal 1 versus 0 for your specific case, and pay attention to positive versus negative coefficients—they tell you whether each factor increases or decreases the outcome.
17. A researcher studies factors affecting infant mortality rates across counties. The model includes region as a categorical variable with 4 levels: Northeast, South, Midwest, and West. After fitting the initial model with West as reference, the researcher discovers that Northeast and Midwest have very similar coefficients ($\beta_{\text{Northeast}} = 2.1$, $\beta_{\text{Midwest}} = 2.3$). What is the most appropriate next analytical step?
    Answer: Test whether $\beta_{\text{Northeast}} = \beta_{\text{Midwest}}$ and consider combining these categories if not significantly different When you encounter similar regression coefficients for categorical variables, you're dealing with a potential model simplification opportunity that requires formal statistical testing rather than arbitrary decisions. The correct approach is option C: formally test whether $\beta_{\text{Northeast}} = \beta_{\text{Midwest}}$ using a statistical test (typically an F-test or Wald test). If the coefficients are not significantly different, you can combine these categories into a single "Northeast/Midwest" group, reducing model complexity while maintaining statistical validity. This creates a more parsimonious model with better interpretability and potentially improved power. Option A is wrong because arbitrarily removing variables based on coefficient magnitude ignores statistical significance and proper hypothesis testing procedures. You'd lose important information about the Northeast region entirely. Option B incorrectly assumes changing the reference category addresses the similarity issue. While this would change the coefficient values, it doesn't resolve the underlying question of whether Northeast and Midwest truly differ in their effects on infant mortality. Option D misunderstands the statistical relationship. You cannot create an interaction between two indicator variables from the same categorical predictor (Northeast and Midwest are mutually exclusive categories). This approach would be both statistically invalid and conceptually meaningless. Study tip: When you see similar coefficients in categorical variables, always think "formal testing before combining." Never make arbitrary modeling decisions based solely on coefficient magnitudes—statistical significance testing should guide your model refinement choices.
18. In analyzing hospital readmission risk, a logistic regression model includes: $\log\left(\frac{p}{1-p}\right) = -2.1 + 0.8 \times Comorbidity + 1.2 \times Emergency$, where Comorbidity = 1 for patients with comorbidities (0 otherwise) and Emergency = 1 for emergency admissions (0 for elective). For a patient with comorbidities admitted through emergency, what is the log-odds of readmission?
    Answer: -0.1 is the log-odds of readmission When you encounter a logistic regression equation, you're looking at a model that predicts the log-odds (logit) of an outcome occurring. The key insight is that you need to substitute the actual values of your predictor variables into the equation to calculate the log-odds for a specific scenario. For a patient with comorbidities admitted through emergency, both predictor variables equal 1: Comorbidity = 1 and Emergency = 1. Substituting these values into the equation: $\log\left(\frac{p}{1-p}\right) = -2.1 + 0.8(1) + 1.2(1) = -2.1 + 0.8 + 1.2 = -0.1$ Therefore, the log-odds of readmission is -0.1, making D correct. The wrong answers represent common misconceptions about logistic regression interpretation. A (-2.1) incorrectly assumes the intercept alone represents the log-odds for this patient, but the intercept only applies when all predictor variables equal zero (a patient with no comorbidities and elective admission). B (0.8) confuses the coefficient for comorbidities with the actual log-odds calculation. C (1.2) makes the same error with the emergency admission coefficient. These coefficients represent the change in log-odds associated with each variable, not the final log-odds themselves. Remember: in logistic regression, always substitute your specific variable values into the complete equation. The intercept represents the baseline log-odds when all predictors equal zero, while coefficients show how much the log-odds change when each predictor increases by one unit.
19. A health researcher models hospital readmission risk using logistic regression. The model includes hospital size as a categorical predictor: Small (< 100 beds), Medium (100-300 beds), Large (> 300 beds), with Medium as reference. The fitted logistic regression equation is: $\log\left(\frac{p}{1-p}\right) = -1.5 + 0.4 \cdot I_{\text{Small}} - 0.3 \cdot I_{\text{Large}}$. What is the odds ratio for readmission comparing Small hospitals to Large hospitals?
    Answer: The odds ratio for Small vs Large hospitals is approximately 2.01 When you encounter logistic regression with categorical predictors, you're working with indicator variables and need to understand how to compare non-reference categories. The key insight is that you can calculate odds ratios between any two categories, not just against the reference. To find the odds ratio comparing Small to Large hospitals, you need the log-odds for each group. For Small hospitals: $\log\left(\frac{p}{1-p}\right) = -1.5 + 0.4(1) + (-0.3)(0) = -1.1$. For Large hospitals: $\log\left(\frac{p}{1-p}\right) = -1.5 + 0.4(0) + (-0.3)(1) = -1.8$. The difference in log-odds is $-1.1 - (-1.8) = 0.7$. Since odds ratios are the exponential of log-odds differences, $OR = e^{0.7} \approx 2.01$. Answer A (1.91) likely results from a calculation error, possibly using an incorrect exponential approximation. Answer C (0.70) represents the odds ratio for Large vs Small (the inverse relationship) - this is $e^{-0.7}$, showing confusion about the direction of comparison. Answer D (1.49) might come from incorrectly using just one coefficient ($e^{0.4} \approx 1.49$) instead of properly calculating the difference between both non-reference categories. Remember this pattern: when comparing two non-reference categories in logistic regression, subtract their respective log-odds to find the log-odds ratio, then exponentiate. Always double-check which direction you're comparing - Small vs Large gives you the reciprocal of Large vs Small.
20. A nutrition study models daily caloric intake using: $Calories = 1800 + 300 \times Male + 150 \times Active$, where Male = 1 for male participants (0 for female) and Active = 1 for physically active individuals (0 for sedentary). Among active individuals, what is the predicted difference in daily caloric intake between males and females?
    Answer: Males consume 300 more calories than females among active individuals When you encounter a linear regression equation with multiple predictors, focus on how each coefficient represents the change in the outcome when that specific variable increases by 1 unit, holding all other variables constant. To find the difference between males and females among active individuals, you need to compare their predicted caloric intake when Active = 1 for both groups. For active males: $Calories = 1800 + 300(1) + 150(1) = 2250$. For active females: $Calories = 1800 + 300(0) + 150(1) = 1950$. The difference is 2250 - 1950 = 300 calories, with males consuming more. Looking at the answer choices: A is correct because it accurately reflects this 300-calorie difference among active individuals. B incorrectly suggests the difference is 150 calories, which is actually the coefficient for the Active variable - this would be the difference between active and sedentary individuals within the same gender. C combines both coefficients (300 + 150 = 450), but this incorrectly assumes you add the effects rather than comparing specific groups. D uses the absolute caloric intake for active males (2250) rather than the difference between groups. The key insight is that the coefficient for Male (300) directly tells you the gender difference in calories, regardless of activity level. This is because linear regression assumes additive effects - the 300-calorie gender difference remains constant whether comparing active or sedentary individuals. Always interpret coefficients as the isolated effect of that variable while holding others constant.

Start this quiz