Econometrics Practice Test (300 Questions)

Based on "Introductory Econometrics: A Modern Approach" by Jeffrey M. Wooldridge

Chapter 1: The Nature of Econometrics and Economic Data

1. Which of the following best describes "econometrics"? A. The study of economic theory using
only qualitative methods. B. The use of statistical methods to analyze economic data and estimate
economic relationships. C. The application of accounting principles to financial data. D. The study
of government policy without data. Answer: B
2. What is the primary goal of estimating an econometric model? A. To prove an economic theory
is always correct. B. To generate data that fits a perfect line. C. To estimate parameters (β ) and
test hypotheses about them. D. To maximize the number of variables in a model. Answer: C
3. Which type of data consists of a sample of individuals, households, or firms taken at a specific
point in time? A. Time series data B. Cross-sectional data C. Panel data D. Longitudinal data
Answer: B
4. Data collected on the same set of individuals or firms over multiple time periods is called: A.
Pooled cross-sectional data B. Random data C. Panel (or longitudinal) data D. Time series data
Answer: C
5. A dataset consisting of the GDP of the United States measured annually from 1950 to 2020 is
an example of: A. Cross-sectional data B. Time series data C. Panel data D. Pooled cross-
sectional data Answer: B
6. "Ceteris paribus" means: A. In the long run B. Other relevant factors being equal C. Comparing
two different populations D. Ignoring all errors Answer: B
7. Why is establishing causality difficult in econometrics? A. Because data is always scarce. B.
Because we rarely have experimental data where we can control all factors. C. Because
computers are not powerful enough. D. Because economic theory is never specific. Answer: B
8. If y is the dependent variable and x is the independent variable, correlation between x and
the error term u usually implies: A. Causality B. An unbiased estimate C. Spurious correlation or
bias D. Homoskedasticity Answer: C
9. Which of the following is NOT a step in empirical economic analysis? A. Formulating a question
of interest. B. Constructing a formal economic model. C. Collecting data. D. Guaranteeing the
results match personal beliefs. Answer: D
10. A dataset combining two different cross-sectional surveys from different years (where the
individuals are different in each year) is called: A. Panel data B. Pooled cross-sectional data C.
Time series data D. Experimental data Answer: B
11. Which variable is typically denoted by y in a regression equation? A. The explanatory variable B.
The independent variable C. The dependent (or explained) variable D. The error term Answer: C
12. In the equation wage = β0 + β1 educ + u, what is u? A. The intercept B. The slope C. The error
​ ​

term D. The fitted value Answer: C

13. Which of the following is an example of an experimental dataset? A. U.S. Census data B. Stock
market closing prices C. Data from a randomized control trial (RCT) assessing a job training
 program D. Inflation rates over the last decade Answer: C
14. Nonexperimental data is also known as: A. Observational data B. Laboratory data C. Controlled
data D. Derived data Answer: A
15. If variable A is correlated with variable B, does A cause B? A. Yes, always. B. No, correlation does
not imply causation. C. Only if the correlation is positive. D. Only if the correlation is negative.
Answer: B
16. Which of the following best describes the "counterfactual" outcome? A. The outcome that
actually occurred. B. The outcome that would have happened if a specific policy or treatment had
not been applied. C. The error term in a regression. D. The intercept of the regression line.
Answer: B
17. Retrospective data refers to: A. Data predicting the future. B. Data collected based on past
events or records. C. Data generated by a computer simulation. D. Data with no time dimension.
Answer: B
18. Which of the following is a key feature of panel data? A. It ignores time. B. It tracks the same
units over time. C. It combines completely different people from different years. D. It only looks at
aggregate macro variables. Answer: B
19. If we want to estimate the effect of class size on test scores, why might simple correlation be
misleading? A. Because test scores are always random. B. Because parents might choose schools
based on class size and quality, creating selection bias. C. Because class size is always constant.
D. Because correlation is a complex calculation. Answer: B
20. Econometric models usually include a random error term because: A. Economic relationships
are exact. B. We can measure every variable perfectly. C. Intrinsic uncertainty and unobserved
variables affect the outcome. D. It makes the math easier. Answer: C
21. In the equation y = f (x), x is the: A. Dependent variable B. Independent variable C. Parameter
D. Residual Answer: B
22. Cross-sectional data is most susceptible to which issue if the sample is not random? A. Serial
correlation B. Sample selection bias C. Unit root D. Seasonality Answer: B
23. Which field uses econometrics the most? A. Biology B. Economics C. Chemistry D. Literature
Answer: B
24. In a wage equation, "ability" is often part of the error term because: A. It is constant. B. It is
unobservable or hard to measure. C. It has no effect on wages. D. It is perfectly correlated with
education. Answer: B
25. Policy analysis involves: A. Only theoretical derivations. B. Estimating the effect of a policy
intervention on an outcome. C. ignoring data. D. assuming all variables are constant. Answer: B
26. What is the difference between a parameter and an estimator? A. They are the same. B. A
parameter is a fixed population value; an estimator is a rule/formula to find it using data. C. An
estimator is fixed; a parameter is random. D. A parameter comes from the sample; an estimator
comes from the population. Answer: B
27. Quantitative variables are: A. Categorical (e.g., Male/Female). B. Numerical (e.g., Income, Prices).
C. Only binary. D. Not useful in regression. Answer: B
28. Qualitative variables are: A. Always continuous. B. Measured in dollars. C. Categorical attributes
(e.g., Race, Gender, Region). D. Never used in econometrics. Answer: C
29. Frequency of data refers to: A. How loud the data is. B. The interval at which data is collected
(e.g., daily, monthly, annually). C. The size of the sample. D. The number of variables. Answer: B
30. If u represents "luck" in a test score regression, we typically assume the average luck in the
population is: A. 100 B. 0 C. Positive D. Negative Answer: B
31. Which of the following is NOT an example of a counterfactual reasoning question? A. How
much would a worker earn if they had joined a job training program vs if they hadn't? B. What was
the GDP of the US in 1990? C. What would the crime rate be if the police force size was doubled?
D. How would test scores change if class size was reduced by 5 students? Answer: B
32. "Observational data" implies that: A. The researcher controls the treatment assignment. B. The
researcher passively observes data generated by the real world. C. The data is generated by
computer algorithms. D. The sample size is always small. Answer: B
33. If a variable is determined within the model, it is called: A. Exogenous. B. Endogenous. C.
Deterministic. D. Stochastic. Answer: B
34. Economic models often assume individuals maximize: A. Utility or Profits. B. Taxes. C. Costs. D.
Sample size. Answer: A
35. A "spurious relationship" is one where: A. Two variables are correlated due to a third unobserved
factor, not direct causation. B. Correlation is zero. C. Regression R2 is 1.0. D. The variables are
measured in different units. Answer: A
36. If sample selection is non-random (e.g., looking only at employed women to estimate wage
offer for all women), the resulting estimates are likely: A. Unbiased. B. Biased. C. Efficient. D.
Perfect. Answer: B
37. Time series data typically exhibits: A. Independent observations. B. Serial correlation
(dependence across time). C. Random sampling properties. D. No trends. Answer: B
38. Pooled cross sections are useful for: A. Analyzing policy changes over time (e.g., before and
after). B. Tracking specific individuals for their whole lives. C. Eliminating all bias. D. Calculating
simple averages only. Answer: A
39. Which is an example of attrition in panel data? A. Adding new variables. B. People dropping out
of the survey over time. C. Increasing the sample size. D. Changing the definition of GDP. Answer:
B
40. The error term u contains: A. Only measurement error. B. Only omitted variables. C. All
unobserved factors affecting y . D. The intercept. Answer: C
41. If y is a linear function of x (e.g., y = 2x), the correlation is: A. 0 B. 1 (or -1). C. 0.5 D.
Undefined. Answer: B
42. Causal inference is most difficult with: A. Randomized experiments. B. Observational cross-
sectional data. C. Laboratory data. D. Simulated data. Answer: B
43. The law of demand is an example of: A. An econometric estimate. B. An economic model/theory.
C. A dataset. D. A statistical test. Answer: B
44. If we estimate a demand function, price is likely: A. Exogenous. B. Endogenous (determined by
supply and demand equilibrium). C. Constant. D. Zero. Answer: B
45. Which data structure orders observations chronologically? A. Cross-sectional. B. Time Series. C.
Random sample. D. Snapshot. Answer: B
46. Why do we square the residuals in OLS (Ordinary Least Squares)? A. To make them smaller. B.
To penalize larger errors more than smaller ones and ensure the sum is positive. C. To make the
math harder. D. To ignore the intercept. Answer: B
47. If the correlation between x and y is zero, the OLS slope will be: A. 1 B. 0 C. Infinite. D. -1.
Answer: B
48. A "dummy variable" is distinct because: A. It is continuous. B. It takes only discrete values 0 or 1.
C. It represents time. D. It is always the dependent variable. Answer: B
49. The phrase "holding other factors fixed" corresponds to: A. Ceteris paribus. B. Ad hominem. C.
Ipso facto. D. Carpe diem. Answer: A
50. Econometrics differs from mathematical statistics because: A. It focuses on non-experimental
(observational) data. B. It uses no math. C. It is older. D. It never uses probability. Answer: A

Chapter 2: The Simple Regression Model

51. In the simple linear regression model y = β0 + β1 x + u, β1 represents: A. The intercept. B. The
​ ​ ​

slope parameter. C. The error term. D. The predicted value. Answer: B

52. The Ordinary Least Squares (OLS) method estimates parameters by minimizing: A. The sum of
the residuals. B. The sum of the squared residuals. C. The sum of the absolute residuals. D. The
total sum of squares. Answer: B
53. The OLS residual u
^i is defined as: A. yi − y^i B. y^i − yi C. yi + y^i D. β0 + β1 xi Answer: A
​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

54. The sum of the OLS residuals ∑ u

^i is always equal to: A. 1 B. The sample size n. C. 0 (assuming
​

an intercept is included). D. The total sum of squares. Answer: C

55. The fitted regression line always passes through: A. The origin (0,0). B. The sample means of x
ˉ, yˉ). C. The point where y is maximum. D. The point where x is zero. Answer: B
and y (x ​

56. The Total Sum of Squares (SST) measures: A. The variation in x. B. The total variation in the
dependent variable y . C. The variation explained by the model. D. The variation due to error.
Answer: B
57. The R2 of a regression is defined as: A. SSR/SST B. SSE/SST C. SST /SSE D.
1 + (SSR/SST ) Answer: B
58. If R2 = 0.65, this means: A. 65% of the variation in y is explained by x. B. The slope is 0.65. C.
The correlation between x and y is 0.65. D. The error is 65%. Answer: A
59. Which assumption states E(u∣x) = 0? A. Homoskedasticity B. Zero Conditional Mean C.
Random Sampling D. No Perfect Collinearity Answer: B
60. If E(u∣x) = 0, then: A. u and x are correlated. B. u and x are uncorrelated. C. u is constant. D. x
is constant. Answer: B
61. The assumption Var(u∣x) = σ 2 is known as: A. Heteroskedasticity B. Homoskedasticity C.
Consistency D. Normality Answer: B
62. If the variance of the error term increases as income increases, the data exhibits: A.
Homoskedasticity B. Heteroskedasticity C. Perfect collinearity D. Bias Answer: B
63. The variance of the OLS slope estimator β^1 is given by: A. σ 2 / ∑(xi − x
​ ​ ˉ)2 B. σ 2 C. ​

ˉ)2 /σ 2 D. σ/n Answer: A

∑(xi − x
​

^
64. To decrease the variance of β1 , one could: A. Decrease the sample size. B. Increase the variation
​ ​

in the independent variable x. C. Increase the error variance σ 2 . D. Remove the intercept. Answer:
B
65. An estimator is "unbiased" if: A. Its variance is zero. B. It always equals the true parameter. C. Its
expected value equals the true population parameter. D. It has the smallest variance. Answer: C
66. In the regression log(wage) = β0 + β1 educ + u, the coefficient β1 is interpreted as: A. The
​ ​ ​

dollar change in wage for one year of education. B. The percentage change in wage for one year
of education (multiplied by 100). C. The percentage change in wage for a 1% change in education.
D. The elasticity of wage with respect to education. Answer: B
67. In the regression log(y) = β0 + β1 log(x) + u, β1 is: A. The semi-elasticity. B. The elasticity of
​ ​ ​

y with respect to x. C. The slope in levels. D. Meaningless. Answer: B

68. Which assumption is NOT required for OLS to be unbiased? A. Linear in parameters. B. Random
sampling. C. Zero conditional mean. D. Homoskedasticity. Answer: D
69. The standard error of the regression (σ
^ ) is estimated by: A. SSR/n B. ​ SSR/(n − 2) C. ​

SSR/n D. ^2i
∑u ​ Answer: B
70. If the correlation coefficient between x and y is r, then R2 in a simple regression is: A. r B. r2
C. r D. 1 − r Answer: B
​

71. Regression through the origin forces the intercept β0 to be: A. 1 B. yˉ C. 0 D. β1 Answer: C ​ ​ ​

72. The "Explained Sum of Squares" (SSE) is: A. ∑(yi − yˉ)2 B. ∑(y^i − yˉ)2 C. ∑ u
^2i D. ​ ​ ​ ​ ​ ​

2
∑(y^i − yi ) Answer: B
​ ​ ​

73. Assumptions SLR.1 through SLR.4 ensure that OLS estimators are: A. Normal B. Unbiased C.
Efficient D. Positive Answer: B
74. Ideally, we want the error variance σ 2 to be: A. As large as possible. B. As small as possible. C.
Equal to 1. D. Equal to the sample size. Answer: B
75. If we change the units of measurement of the dependent variable y (e.g., dollars to cents),
what happens to R2 ? A. It increases by 100. B. It decreases by 100. C. It stays the same. D. It
becomes zero. Answer: C
~
76. If we multiply the independent variable x by a constant c, the new slope estimate β1 will be: A. ​ ​

β^1 B. c ⋅ β^1 C. β^1 /c D. β^1 + c Answer: C

​ ​ ​ ​ ​ ​ ​ ​

77. The fraction of the total variation in y that is not explained by the regression is: A. R2 B.
1 − R2 C. SSR/SSE D. σ 2 Answer: B
78. SLR.3 requires that: A. x is constant. B. x varies in the sample. C. y varies in the sample. D. The
sample size is infinite. Answer: B
79. Which term represents the predicted change in y given a one-unit change in x? A. β^0 B. u
^ C. ​ ​

^ 2
β1 D. R Answer: C
​ ​

80. If the covariance between x and y is positive, the slope of the regression line will be: A.
Negative B. Zero C. Positive D. Undefined Answer: C
81. The sample average of the OLS residuals is: A. yˉ B. x ^ 2 Answer: C
ˉ C. Zero D. σ ​

^ is: A. 1 B. -1 C. Zero D. σ 2
82. The sample covariance between the regressor x and the residuals u
Answer: C
 83. When estimating β0 and β1 , we lose how many degrees of freedom? A. 0 B. 1 C. 2 D. n Answer:
​ ​

C
84. The standard error of the estimate (SER) is measured in the same units as: A. The independent
variable x. B. The dependent variable y . C. The slope β1 . D. It is unitless. Answer: B
85. Which of the following is an algebraic property of OLS? A. Unbiasedness B. Consistency C. The
regression line passes through the sample means. D. Efficiency Answer: C
86. If the true relationship is linear but you fit a curved line, you violate: A. SLR.1 (Linear in
parameters) B. SLR.2 (Random sampling) C. SLR.5 (Homoskedasticity) D. None of the above
Answer: A
87. In a wage equation, if u includes "ability" and high-ability people tend to get more education (
 0, causing bias. C. OLS is still unbiased. D. R2 will
x), then: A. E(u∣x) = 0 holds. B. E(u∣x) =
be 1. Answer: B
88. The variance of β^0 depends on: A. Only σ 2 . B. σ 2 , sample size, and sample values of x. C. Only
​ ​

sample size. D. Only R2 . Answer: B

89. A spurious correlation is: A. A true causal relationship. B. A relationship driven by a third,
unobserved factor. C. A correlation of 0. D. A negative correlation. Answer: B
90. The term "regression" was originally coined by: A. Gauss B. Markov C. Galton D. Wooldridge
Answer: C
ˉ)2 is small (little variation in x), the variance of the slope estimator will be: A. Small
91. If ∑(xi − x
​

B. Large C. Zero D. Unchanged Answer: B

92. The "residual" is the difference between: A. The true parameter and the estimated parameter. B.
The actual y and the predicted y^. C. The actual y and the mean yˉ. D. x and y . Answer: B
​ ​

93. In the model cons = β0 + β1 inc + u, β1 is: A. The Average Propensity to Consume. B. The
​ ​ ​

Marginal Propensity to Consume. C. Total consumption. D. Savings. Answer: B

94. To estimate a constant elasticity model, we use: A. Level-level model B. Log-level model C. Log-
log model D. Level-log model Answer: C
95. Which of the following is NOT a unit of measurement issue? A. Changing dollars to thousands of
dollars. B. Changing inches to feet. C. Changing y to log(y). D. Changing years to months.
Answer: C
96. The standard deviation of β^1 is called: A. The standard error of the slope. B. The root mean
​ ​

squared error. C. The t-statistic. D. The p-value. Answer: A

97. If the error term is heteroskedastic, then: A. OLS is biased. B. OLS is unbiased but variance
formulas are incorrect. C. R2 is negative. D. We cannot use OLS. Answer: B
98. Random sampling ensures that: A. The data represents the population. B. The error term is zero.
C. The regression line is flat. D. There are no outliers. Answer: A
99. If y is binary (0 or 1), the predicted value y^ is interpreted as: A. A binary outcome. B. A
​

probability. C. An integer. D. A standard deviation. Answer: B

100. The slope of the regression of y on x is the same as the regression of x on y only if: A.
R2 = 1. B. Variances of x and y are equal. C. The slope is 1. D. Never. Answer: B
101. The sample covariance between y and y^ in a simple regression is: A. 0. B. Equal to the sample
​

variance of y^. C. 1. D. Equal to the sample variance of y . Answer: B

​

^
102. If we add a constant to every xi , what happens to the slope β1 ? A. It increases. B. It decreases. ​ ​ ​

C. It stays the same. D. It becomes zero. Answer: C

103. If we add a constant to every yi , what happens to the slope β^1 ? A. It increases. B. It decreases. ​ ​ ​

C. It stays the same. D. It doubles. Answer: C

104. In the equation y = β0 + β1 x + u, if we set β0 = 0 (regression through the origin), the sum
​ ​

of OLS residuals: A. Is always zero. B. Is not necessarily zero. C. Is always 1. D. Minimizes SSE.
Answer: B
105. If the slope β^1 is positive, it means: A. x causes y . B. x and y are positively correlated in the
​ ​

sample. C. y causes x. D. The intercept must be positive. Answer: B

106. Which of the following values of R2 indicates the best fit? A. 0.05 B. -0.5 C. 0.95 D. 1.05
Answer: C
107. The units of the slope coefficient β1 are: A. The same as y . B. The same as x. C. Units of y ​

divided by units of x. D. Unitless. Answer: C

108. If SSTx = 0, then: A. The regression fits perfectly. B. We cannot compute the OLS estimator
​

(division by zero). C. The slope is zero. D. The intercept is zero. Answer: B

109. The term "linear" in "simple linear regression" refers to linearity in: A. Variables (x and y ). B.
Parameters (β0 and β1 ). C. Residuals. D. Variance. Answer: B
​ ​

110. If y is temperature in Fahrenheit and x is year, converting y to Celsius will: A. Change the R2 . B.
Change the t-statistics. C. Linearly transform the coefficients. D. Make the slope zero. Answer: C
111. In a simple regression, if the sample size n increases, the variance of β^1 : A. Increases. B. ​ ​

Decreases. C. Stays the same. D. Becomes biased. Answer: B

112. A "sample regression function" (SRF) is: A. The true relationship in the population. B. The
estimated line y^ = β^0 + β^1 x. C. The error distribution. D. The dataset. Answer: B
​ ​ ​ ​ ​

113. The "population regression function" (PRF) is: A. E(y∣x) = β0 + β1 x. B. y^ = β^0 + β^1 x. C. yˉ. ​ ​ ​ ​ ​ ​ ​ ​

D. The residuals. Answer: A

114. If β1 = 0, then: A. x and y are uncorrelated in the population. B. x causes y . C. The regression
​

line is vertical. D. R2 = 1. Answer: A

115. The estimator β^1 is a random variable because: A. It depends on β1 . B. It depends on the
​ ​ ​

random sample data. C. It is always changing over time. D. It is not calculated correctly. Answer: B
116. Which OLS algebraic property ensures the regression line passes through the means? A.
∑u ​ ^i = 0 C. The formula for β^0 = yˉ − β^1 x
^i = 0 B. ∑ xi u ​ˉ. D. SST = SSE + SSR. Answer:
​ ​ ​ ​ ​ ​

C
117. For unbiasedness, do we need the error term u to be normally distributed? A. Yes. B. No. C.
Only if n is small. D. Only if x is constant. Answer: B
118. If x is measured with error, the OLS estimator of the slope is generally: A. Unbiased. B. Biased
towards zero (attenuation bias). C. Biased away from zero. D. Consistent. Answer: B
119. The term σ (sigma) represents: A. The variance of x. B. The standard deviation of the error term
u. C. The standard error of β^1 . D. The slope. Answer: B ​ ​

120. If yi = β0 + β1 xi + ui is the true model, then u

​ ​ ​ ^i estimates: A. β0 . B. β1 . C. ui . D. yi . Answer: C
​ ​ ​ ​ ​ ​ ​
121. If the dependent variable is in logs (log(y)), we cannot have: A. Negative values for y . B.
Negative values for x. C. Negative values for β^1 . D. Negative values for β^0 . Answer: A
​ ​ ​ ​

122. The "method of moments" derivation of OLS relies on sample analogs of: A. E(u) = 0 and
E(xu) = 0. B. Var(u) = σ 2 . C. u ∼ N ormal. D. y = mx + b. Answer: A
123. If all data points lie exactly on a line: A. R2 = 0. B. SSR = 0. C. β^1 = 0. D. SST = 0. Answer:
​ ​

B
124. Which of the following increases the standard error of the slope? A. Larger n. B. Larger
variation in x. C. Larger error variance σ 2 . D. Zero covariance between x and u. Answer: C
125. Homoskedasticity means Var(u∣x) is: A. A function of x. B. Constant. C. Zero. D. Infinite.
Answer: B
126. If we run regression of y on x and get slope 2, and regression of x on y and get slope 0.5, then
R2 is: A. 0.25 B. 0.5 C. 1.0 D. Undefined. Answer: C
127. Changing the unit of x from years to months (multiplying x by 12) divides the slope coefficient
by 12. A. True. B. False. Answer: A
128. In the model y = β0 + β1 x + u, if we drop β0 , we force the line to: A. Be horizontal. B. Be
​ ​ ​

vertical. C. Go through the origin. D. Fit perfectly. Answer: C

129. If E(u∣x) = 2 (non-zero constant), the slope estimator is: A. Biased. B. Unbiased (intercept
absorbs the bias). C. Inconsistent. D. Undefined. Answer: B
130. The predicted value y^ represents: A. The error. B. The estimated conditional mean E(y∣x). C.
​

The variance. D. The residual. Answer: B

131. SST represents: A. Variability in x. B. Variability in y explained by the model. C. Total variability in
y . D. Variability in residuals. Answer: C
132. SSE represents: A. Total variability. B. Explained variability. C. Unexplained variability. D. Sample
size. Answer: B
133. SSR represents: A. Total variability. B. Explained variability. C. Unexplained (Residual) variability.
D. Slope. Answer: C
134. If R2 is low, it means: A. The model is useless. B. OLS is biased. C. x explains a small portion of
the variance in y . D. There is heteroskedasticity. Answer: C
135. The residuals u
^i are uncorrelated with the predicted values y^i in the sample. A. True. B. False.
​ ​ ​

Answer: A
136. If x is binary (0 or 1), the slope β1 is: A. The difference in means of y between the two groups. B.
​

The elasticity. C. The correlation. D. The variance. Answer: A

137. Simple regression is a special case of multiple regression with: A. k = 0. B. k = 1. C. k = 2. D.
No intercept. Answer: B
138. If yi = β0 + ui (regression on a constant only), then β^0 is: A. 0 B. yˉ C. 1 D. x
​ ​ ​ ​ ​ ˉ Answer: B
​

139. Which plot is best for detecting heteroskedasticity in simple regression? A. Histogram of x. B.
Plot of residuals vs. x. C. Plot of y vs. time. D. Pie chart. Answer: B
140. If the error term u is correlated with x, we say x is: A. Exogenous. B. Endogenous. C.
Homoskedastic. D. Significant. Answer: B
 Chapter 3: Multiple Regression Analysis: Estimation
141. The multiple regression model allows us to: A. Eliminate the error term. B. Control for multiple
factors simultaneously. C. Avoid using computers. D. Guarantee causality. Answer: B
142. In the model y = β0 + β1 x1 + β2 x2 + u, β1 measures: A. The effect of x1 on y ignoring x2 . B.
​ ​ ​ ​ ​ ​ ​ ​

The effect of x1 on y holding x2 fixed (ceteris paribus). C. The correlation between x1 and x2 . D.
​ ​ ​ ​

The intercept. Answer: B

143. Assumption MLR.3 "No Perfect Collinearity" means: A. Independent variables cannot be
correlated at all. B. Independent variables cannot be constant or exact linear combinations of
each other. C. The error term must be zero. D. The sample size must be infinite. Answer: B
144. If x1 and x2 are perfectly correlated in the sample: A. OLS estimates are unbiased. B. OLS
​ ​

estimates cannot be computed. C. R2 is 1. D. The error term is 0. Answer: B

145. Adding a new independent variable to a regression: A. Always decreases R2 . B. Always
increases (or keeps constant) R2 . C. Always increases Adjusted R2 . D. Always decreases SSR.
Answer: B
146. The "partialling out" interpretation suggests that β^1 captures the relationship between y and:
​ ​

A. The part of x1 that is uncorrelated with the other regressors. B. The total variation in x1 . C. The
​ ​

part of x1 correlated with the error term. D. The intercept. Answer: A

​

147. Omitted Variable Bias occurs when: A. We include an irrelevant variable. B. We exclude a variable
that affects y and is correlated with included regressors. C. The sample size is too small. D. The
error term is heteroskedastic. Answer: B
148. If we omit a variable x2 that has a positive effect on y (β2 > 0) and is positively correlated
​ ​

with x1 , the bias in β^1 will be: A. Positive (Upward). B. Negative (Downward). C. Zero. D.
​ ​ ​

Undefined. Answer: A
149. If x1 and x2 are uncorrelated, omitting x2 from the model will: A. Bias β^1 . B. Not bias β^1 . C.
​ ​ ​ ​ ​ ​ ​

Make β^1 zero. D. Increase R2 . Answer: B

​ ​

150. Multicollinearity refers to: A. High (but not perfect) correlation among independent variables. B.
Correlation between x and y . C. Correlation between u and y . D. Perfect correlation between x1 ​

and x2 . Answer: A
​

151. High multicollinearity leads to: A. Biased coefficients. B. Large variances (standard errors) for the
estimators. C. Small confidence intervals. D. High t-statistics. Answer: B
152. The Variance Inflation Factor (VIF) is used to detect: A. Heteroskedasticity. B. Serial correlation.
C. Multicollinearity. D. Non-normality. Answer: C
153. A VIF of 1 indicates: A. Perfect multicollinearity. B. No correlation between that regressor and
others. C. Extreme bias. D. A calculation error. Answer: B
154. The Gauss-Markov Theorem states that under assumptions MLR.1–MLR.5, OLS is: A. BLUE
(Best Linear Unbiased Estimator). B. Biased but consistent. C. Nonlinear. D. Normally distributed.
Answer: A
155. "Best" in BLUE means: A. Smallest bias. B. Smallest variance among linear unbiased estimators. C.
Highest R2 . D. Easiest to calculate. Answer: B
156. In multiple regression, the degrees of freedom is: A. n − 1 B. n − k C. n − k − 1 D. n Answer:
C

2
157. The unbiased estimator of the error variance σ 2 in multiple regression is: A. SSR/n B.
SSR/(n − k − 1) C. SSE/n D. SST /(n − 1) Answer: B
158. Including an irrelevant variable (one with β = 0) in a regression: A. Creates bias. B. Does not
create bias but increases the variance of other estimators. C. Decreases R2 . D. Violates MLR.4.
Answer: B
159. If a variable is highly significant, it means: A. It is economically important. B. The coefficient is
large. C. It is unlikely the true coefficient is zero. D. The data is perfect. Answer: C
160. The intercept β0 in a multiple regression represents: A. The value of y when all x's are zero. B.
​

The mean of y . C. The error term. D. The variation in y . Answer: A

161. Which assumption allows us to claim OLS is unbiased in multiple regression? A. MLR.1 to MLR.4
B. MLR.1 to MLR.5 C. MLR.6 only D. MLR.1 only Answer: A
162. If we scale one independent variable by multiplying by 10, the standard error of its coefficient
will: A. Multiply by 10. B. Divide by 10. C. Stay the same. D. Become negative. Answer: B
163. If we scale the dependent variable y by 10, the coefficients will: A. Multiply by 10. B. Divide by
10. C. Stay the same. D. Square. Answer: A
164. In the model y = β0 + β1 x1 + β2 x2 + u, if x1 and x2 are positively correlated, omitting x2
​ ​ ​ ​ ​ ​ ​

(where β2 < 0) causes: A. Upward bias in β^1 . B. Downward (negative) bias in β^1 . C. No bias. D.
​ ​ ​ ​ ​

Infinite variance. Answer: B

165. Micronumerosity refers to: A. Small sample size problem. B. Large error variance. C. Small
coefficients. D. High correlation. Answer: A
166. If Rj2 (the R-squared from regressing xj on other x's) is close to 1, then Var(β^j ) will be: A.
​ ​ ​ ​

Very small. B. Very large. C. Zero. D. Unaffected. Answer: B

167. Which is NOT a consequence of heteroskedasticity? A. Standard errors are invalid. B. t-statistics
are invalid. C. OLS estimators are biased. D. OLS is no longer BLUE. Answer: C
168. Standardized coefficients (Beta coefficients) measure effect sizes in units of: A. Dollars. B.
Percentages. C. Standard deviations. D. Probability. Answer: C
169. If the model is log(y) = β0 + β1 log(x) + u, β1 is: A. An elasticity. B. A semi-elasticity. C. A
​ ​ ​

linear slope. D. A dummy variable. Answer: A

170. In a level-log model y = β0 + β1 log(x) + u, a 1% increase in x leads to: A. β1 change in y . B.
​ ​ ​

β1 /100 change in y . C. 100β1 change in y . D. β1 % change in y . Answer: B

​ ​ ​

171. In a log-level model log(y) = β0 + β1 x + u, a 1 unit increase in x leads to: A. 100β1 % change
​ ​ ​

in y . B. β1 % change in y . C. β1 change in y . D. β1 /100 change in y . Answer: A

​ ​ ​

172. The Adjusted R2 penalizes for: A. Adding variables. B. Removing variables. C. High correlation. D.
Low variance. Answer: A
173. Adjusted R2 can be: A. Greater than 1. B. Negative. C. Greater than R2 . D. Equal to 100. Answer:
B
174. When choosing between non-nested models with different numbers of variables, one should
compare: A. SSR B. R2 C. Adjusted R2 D. The intercepts Answer: C
175. To minimize the variance of the OLS estimator, we prefer: A. Small sample size. B. Small variation
in x. C. Large variation in x. D. Large error variance. Answer: C

2
176. Perfect collinearity happens if: A. x2 = 2x1 . B. x2 = log(x1 ). C. x2 = x21 . D. Correlation is 0.8.
​ ​ ​ ​ ​ ​

Answer: A
177. Which of the following is an example of nonlinear relationship handled by OLS? A.
y = β0 + β1 x + β2 x2 + u B. y = β0 xβ1 C. y = 1/(β0 + β1 x) D. y = β0 + β1γ x Answer: A
​ ​ ​ ​
​

​ ​ ​ ​

178. If we run a regression on y , x1 , and x2 , and find x1 has a positive coefficient, but a simple
​ ​ ​

regression of y on x1 gives a negative coefficient, this is likely due to: A. Calculation error. B.
​

Omitted variable bias (correlation between x1 and x2 ). C. Heteroskedasticity. D. Normality ​ ​

violation. Answer: B
179. The assumption E(u) = 0 basically defines: A. The slope. B. The intercept. C. The sample size.
D. The variance. Answer: B
180. Under Gauss-Markov assumptions, Var(β^j ) decreases as: A. σ 2 increases. B. n increases. C. ​ ​

SSTj decreases. D.
​ Rj2 ​ increases. Answer: B
181. If V IFj = 10, then Rj2 is: A. 0.1 B. 0.9 C. 0.5 D. 0.0 Answer: B
​ ​

182. The error term u represents: A. Measurement error in y . B. Omitted variables. C. Random noise.
D. All of the above. Answer: D
183. If we estimate price = β0 + β1 sqrf t + β2 bdrms + u, and bdrms is highly correlated with
​ ​ ​

sqrf t, then: A. Estimates will be biased. B. Standard errors for β^1 and β^2 will be large. C. R2 will ​ ​ ​ ​

be zero. D. We must drop one variable. Answer: B

184. The term σ 2 in the variance formula refers to the variance of: A. The dependent variable y . B.
The independent variable x. C. The population error term u. D. The coefficient β . Answer: C
185. Can we compare R2 of a log-linear model and a linear-linear model directly? A. Yes. B. No,
because the dependent variables are different. C. Only if sample size is large. D. Only if Adjusted
R2 is used. Answer: B
186. If an independent variable is measured with error, this typically leads to: A. Attenuation bias
(bias toward zero). B. Upward bias. C. No bias. D. Perfect prediction. Answer: A
187. Partialling out x2 from x1 involves: A. Regressing x1 on x2 and taking the residuals. B. Dividing
​ ​ ​ ​

x1 by x2 . C. Subtracting the means. D. Adding them together. Answer: A

​ ​

188. If a variable is constant for all observations: A. It explains everything. B. It violates the "No
Perfect Collinearity" assumption (if intercept is included). C. It has a large variance. D. It should be
the dependent variable. Answer: B
189. The primary cost of including an irrelevant variable is: A. Bias. B. Inconsistency. C. Lower
efficiency (higher variances) for other estimators. D. Lower R2 . Answer: C
190. Which is true about the Intercept estimate? A. It is usually the most important parameter. B. It is
often not meaningful if x = 0 is impossible. C. It is always zero. D. It has no variance. Answer: B
191. In the partialling out process, the residuals r^i1 represent: A. The part of x1 explained by x2 . B. ​ ​ ​

The part of x1 uncorrelated with x2 . C. The error term of the main regression. D. The predicted
​ ​

values of y . Answer: B
192. If Rj2 = 0 (regressor xj is uncorrelated with all other regressors), then the VIF is: A. 0. B. 1. C.
​ ​

Infinite. D. 10. Answer: B

193. The Frisch-Waugh Theorem relates to: A. Calculating Adjusted R2 . B. The partialling out
interpretation of multiple regression. C. The F-test for joint significance. D. Heteroskedasticity
 correction. Answer: B
194. Under MLR.1-MLR.4, if we include an irrelevant variable, the estimator for its coefficient is: A.
Biased away from zero. B. Unbiased and equal to zero on average. C. Inconsistent. D. Undefined.
Answer: B
195. If we estimate a model with n = 50 and k = 5, the degrees of freedom for the error variance
estimate is: A. 45. B. 44. C. 50. D. 5. Answer: B
196. In multiple regression, the residuals u
^i sum to zero: A. Only if the intercept is included. B.
Always. C. Never. D. Only if errors are normal. Answer: A
197. The sum of squared residuals (SSR) in multiple regression is minimized by: A. Choosing β s to
make residuals zero. B. Choosing β s to make squared residuals as small as possible. C. Choosing
β s to maximize R2 . D. Both B and C. Answer: D
198. If x1 and x2 are highly correlated, it is difficult to: A. Estimate the effect of x1 holding x2
​ ​ ​ ​

constant. B. Calculate the intercept. C. Calculate SSR. D. Predict y . Answer: A

199. If we have 100 observations and 100 explanatory variables (plus intercept), R2 will be: A. 0. B.
1. C. 0.5. D. Undefined. Answer: B
200. The variance of β^j goes to infinity as: A. SSTj goes to infinity. B. Rj2 goes to 1. C. σ 2 goes to 0.
​ ​ ​ ​

D. n goes to infinity. Answer: B

201. Which assumption ensures that we can mathematically solve the OLS normal equations? A.
MLR.1 (Linearity). B. MLR.3 (No Perfect Collinearity). C. MLR.4 (Zero conditional mean). D. MLR.6
(Normality). Answer: B
202. If we omit a variable x2 and it is uncorrelated with x1 , the variance of β^1 (from the simple
​ ​ ​ ​

regression) is _____ than the variance from the multiple regression. A. Larger. B. Smaller. C. The
same. D. Unknown. Answer: B
203. When estimating the ceteris paribus effect of a variable, control variables are included to: A.
Increase R2 . B. Reduce error variance. C. Avoid omitted variable bias. D. All of the above. Answer:
D
204. In the model log(y) = β0 + β1 x1 + β2 x2 + u, if x1 is measured in logs, β1 is: A. Elasticity. B.
​ ​ ​ ​ ​ ​ ​

Semi-elasticity. C. Slope. D. Intercept. Answer: A

205. If we define prof it as revenue − cost, can we include revenue, cost, and prof it in the
same regression? A. Yes. B. No, perfect collinearity. C. Only if n is large. D. Only if intercept is
excluded. Answer: B
^ 2 is an unbiased estimator of σ 2 under assumptions: A. MLR.1-4.
206. The estimated error variance σ
B. MLR.1-5. C. MLR.1-6. D. Only MLR.2. Answer: B
207. If the dependent variable is multiplied by 2, the SSR: A. Multiplies by 2. B. Multiplies by 4. C.
Stays the same. D. Divides by 2. Answer: B
208. If the dependent variable is multiplied by 2, the standard error of the regression (SER): A.
Multiplies by 2. B. Multiplies by 4. C. Stays the same. D. Divides by 2. Answer: A
209. If we add a variable that is uncorrelated with existing variables but highly correlated with y : A.
Variances of existing coefficients increase. B. Variances of existing coefficients decrease (due to
lower σ 2 ). C. Bias increases. D. R2 decreases. Answer: B
210. Omitted variable bias formula depends on: A. The effect of the omitted variable on y (β2 ). B. The ​

correlation between included and omitted variables (δ1 ). C. Both A and B. D. Neither. Answer: C
​

211. If β^1 changes significantly when x2 is added, we conclude: A. x2 is irrelevant. B. x2 was an

​ ​ ​ ​ ​

omitted variable causing bias. C. There is multicollinearity. D. The sample size is too small. Answer:
B
212. A variable should be removed from a model if: A. It is insignificant (t < 1). B. It induces perfect
collinearity. C. It is correlated with y . D. It reduces Adjusted R2 . Answer: B
213. The "Best" in Gauss-Markov refers to: A. Efficiency (minimum variance). B. Unbiasedness. C.
Consistency. D. Linearity. Answer: A
214. If x1 and x2 are uncorrelated, omitting x2 : A. Biases β^1 . B. Does not bias β^1 . C. Makes β^1 zero.
​ ​ ​ ​ ​ ​ ​ ​ ​

2
D. Increases R . Answer: B
215. With heteroskedasticity, OLS is: A. Biased. B. Inconsistent. C. No longer BLUE. D. Impossible to
calculate. Answer: C
216. The sample regression line always passes through (x ˉ2 , ..., yˉ). A. True. B. False. Answer: A
ˉ1 , x ​ ​ ​

217. If the intercept is excluded, R2 : A. Is always higher. B. Is always lower. C. Can be negative or
misleading. D. Is unchanged. Answer: C
218. The variance of the intercept β^0 is minimized when: A. x
​ ​ ˉ = 0. B. x
ˉ is large. C. Sample size is
2
small. D. σ is large. Answer: A
219. Standard errors of coefficients measure: A. Sampling variability of the estimates. B. The error
term variance. C. The spread of x. D. The bias. Answer: A
220. If x1 is measured in thousands and x2 in units, this causes: A. Multicollinearity. B. Bias. C.
​ ​

Scaling differences in coefficients (no problem). D. Heteroskedasticity. Answer: C

221. Including a variable that is a linear function of others (e.g., x3 = x1 + x2 ) violates: A. MLR.1 B. ​ ​ ​

MLR.2 C. MLR.3 D. MLR.4 Answer: C

222. Which of the following is NOT held constant when interpreting β1 in multiple regression? A. x2 ​ ​

B. x3 C. u D. Parameters. Answer: C
​

223. If we estimate y = β0 + β1 x1 and then y = β0 + β1 x1 + β2 x2 , the sum of squared residuals

​ ​ ​ ​ ​ ​ ​

(SSR): A. Increases or stays same. B. Decreases or stays same. C. Must stay same. D. Becomes
zero. Answer: B
224. The term "linear" in multiple linear regression means: A. Linear in variables. B. Linear in
parameters. C. Straight line graph. D. No exponentials. Answer: B
225. If we use Generalized Least Squares (GLS) instead of OLS when there is heteroskedasticity: A.
We get biased estimates. B. We get BLUE estimates. C. We get inconsistent estimates. D. It makes
no difference. Answer: B
226. In a model with k regressors and intercept, the number of estimated parameters is: A. k B.
k + 1 C. n D. n − k Answer: B
227. The sample average of fitted values yˉ^ equals: A. 0. B. yˉ. C. x
​ ^ . Answer: B
ˉ. D. σ ​

228. If V IF > 10, it is a common rule of thumb for: A. High heteroskedasticity. B. High
multicollinearity. C. Significance. D. Autocorrelation. Answer: B
229. Dropping a collinear variable to fix multicollinearity might lead to: A. Omitted variable bias. B.
Higher variance. C. Lower R2 . D. Nonlinearity. Answer: A
230. The variance of β^j is proportional to: A. The sample size n. B. The inverse of the sample size 1/n
​ ​

(roughly, via SST). C. n2 . D. n. Answer: B

​

Chapter 4: Multiple Regression Analysis: Inference

231. Inference (testing hypotheses) requires which additional assumption? A. MLR.1 B. MLR.4 C.
MLR.6 (Normality of errors) D. MLR.2 Answer: C
232. Under MLR.1–MLR.6, the OLS estimators follow which distribution? A. Uniform B. Normal C. Chi-
square D. Poisson Answer: B
233. The t-statistic is calculated as: A. (β^j − hypothesized val)/se(β^j ) B. β^j /σ C. (β^j − βj )2 D.
​ ​ ​ ​ ​ ​ ​ ​ ​

se(β^j )/β^j Answer: A

​ ​ ​ ​

234. If the absolute value of the t-statistic is greater than the critical value, we: A. Fail to reject the
null hypothesis. B. Reject the null hypothesis. C. Accept the null hypothesis. D. Change the model.
Answer: B
235. The null hypothesis H0 : βj = 0 typically implies: A. The variable xj has a large effect on y . B.
​ ​ ​

The variable xj has no ceteris paribus effect on y . C. The intercept is zero. D. The variable xj is
​ ​

negative. Answer: B
236. A p-value is: A. The probability the null hypothesis is true. B. The probability of observing a test
statistic as extreme as the one computed, assuming the null is true. C. The significance level α. D.
The probability the alternative is true. Answer: B
237. If the p-value is 0.03 and the significance level is 0.05, we: A. Reject the null. B. Fail to reject the
null. C. Accept the null. D. Do nothing. Answer: A
238. The 95% confidence interval roughly corresponds to: A. β^ ± 1se(β^) B. β^ ± 2se(β^) C. ​ ​ ​ ​

β^ ± 3se(β^) D. β^ ± 0.5se(β^) Answer: B

​ ​ ​ ​

239. To test joint significance of multiple variables (e.g., β1 = 0 and β2 = 0), we use: A. The t-test. ​ ​

B. The F-test. C. The Z-test. D. The Chi-square test. Answer: B

(SSRr −SSRur )/q (SSRur −SSRr )/q
240. The formula for the F-statistic using SSR is: A. B. C. SSRr
D.
​ ​ ​ ​
​

SSRur /(n−k−1) SSRr /(n−k−1) SSRur

​ ​ ​

​ ​ ​

SSRur
SSRr ​
​

​
Answer: A
241. In the F-test, q represents: A. The sample size. B. The number of regressors. C. The number of
restrictions (hypotheses). D. The error variance. Answer: C
242. The "Restricted Model" in an F-test is the model where: A. All variables are included. B. The
restrictions imposed by the null hypothesis are assumed true (variables removed). C. Only the
intercept is included. D. No intercept is included. Answer: B
243. If the F-statistic is very large, we likely: A. Reject the joint null hypothesis. B. Fail to reject the
null. C. Have a calculation error. D. Have multicollinearity. Answer: A
244. Testing the "Overall Significance" of the regression means testing: A. β0 = 0 B. ​

2 2
β1 = β2 = ... = βk = 0 C. R = 1 D. σ = 0 Answer: B
​ ​ ​

245. Which relationship holds between the t-stat and F-stat for a single restriction (q = 1)? A.
F = t B. F = t2 C. F = t D. F = 2t Answer: B
​
246. A "statistically significant" coefficient means: A. It is economically large. B. It is not zero. C. We
have strong evidence to reject the hypothesis that it is zero. D. It explains all variation. Answer: C
247. Economic significance refers to: A. The size of the t-statistic. B. The magnitude and practical
importance of the coefficient estimate. C. The p-value. D. The F-statistic. Answer: B
248. If a confidence interval includes zero, we usually: A. Reject the null hypothesis that β = 0. B.
Fail to reject the null hypothesis that β = 0. C. Conclude the variable is highly significant. D.
Assume the data is bad. Answer: B
249. To test H0 : β1 = β2 , we can rewrite the model using: A. θ = β1 − β2 B. θ = β1 + β2 C.
​ ​ ​ ​ ​ ​ ​

Interaction terms. D. Logs. Answer: A

250. The critical value for a two-tailed t-test at 5% significance with large n is approximately: A.
1.645 B. 1.96 C. 2.58 D. 1.28 Answer: B
251. Which assumption allows us to claim t-statistics follow the t-distribution in small samples? A.
Homoskedasticity B. Normality of errors C. Random sampling D. Linearity Answer: B
252. If we exclude a relevant variable, the variances of the remaining coefficients usually: A.
Increase. B. Decrease (become more precise). C. Stay the same. D. Become infinite. Answer: B
253. Reporting regression results usually involves listing: A. Coefficient, Standard Error, R2 , Sample
Size. B. Coefficient and SSR only. C. Just the t-statistics. D. Just the p-values. Answer: A
254. A one-tailed test is used when: A. We don't know the sign of the effect. B. We have a prior
expectation about the direction (sign) of the effect. C. The sample is small. D. The variable is
binary. Answer: B
255. If the F-test for overall significance has a p-value of 0.0001, we conclude: A. The model
explains nothing. B. The model explains a significant amount of variation. C. All coefficients are
zero. D. The intercept is zero. Answer: B
256. The denominator degrees of freedom for the F-test is: A. n − 1 B. q C. n − k − 1 D. k Answer:
C
257. If the confidence interval is

0.5, 2.5

, the point estimate is likely: A. 1.0 B. 1.5 C. 2.0 D. 0.0 Answer: B

258. The standard error of the difference (β^1 − β^2 ) involves: A. Only the variances of β^1 and β^2 . B.
​ ​ ​ ​ ​ ​ ​ ​

Variances and the covariance between β^1 and β^2 . C. The sum of standard errors. D. The
​ ​ ​ ​

difference of standard errors. Answer: B

259. Which test statistic is always positive? A. t-statistic B. F-statistic C. Z-statistic D. Residual
Answer: B
260. If we test H0 : β1 = 1, the t-statistic is: A. β^1 /se(β^1 ) B. (β^1 − 1)/se(β^1 ) C. (β^1 + 1)/se(β^1 )
​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

D. 1/se(β^1 ) Answer: B
​ ​

261. Non-normality of errors in large samples: A. Invalidates t-tests. B. Does not matter for t-tests
(due to CLT). C. Makes OLS biased. D. Requires non-parametric methods. Answer: B
262. The p-value for a one-sided test is _____ the p-value for a two-sided test (assuming the sign
matches). A. Half B. Double C. The same as D. Square root of Answer: A
263. If we reject the null hypothesis, we prove the alternative is true. A. True. B. False (we only find
evidence for it). C. True, if p < 0.01. D. True, if n > 1000. Answer: B
264. The power of a test is: A. Probability of rejecting a true null (Type I error). B. Probability of
rejecting a false null. C. Probability of accepting a true null. D. Probability of accepting a false null.
Answer: B
265. Standard errors reported by software assume: A. Homoskedasticity. B. Heteroskedasticity. C.
Multicollinearity. D. Zero mean. Answer: A
266. A variable can be statistically significant but economically insignificant if: A. Sample size is very
small. B. Sample size is very large. C. Variance is high. D. It is never possible. Answer: B
267. A variable can be economically significant but statistically insignificant if: A. Sample size is very
large. B. Sample size is small (large standard error). C. The coefficient is zero. D. t-stat is 10.
Answer: B
268. The confidence interval formula β^ ± c ⋅ se(β^) assumes: A. A symmetric distribution (like t or
​ ​

normal). B. A skewed distribution. C. Bias. D. Binary data. Answer: A

269. Testing H0 : β1 + β2 = 1 is an example of: A. An exclusion restriction. B. A linear combination
​ ​ ​

hypothesis. C. An F-test for overall significance. D. Testing for heteroskedasticity. Answer: B

270. When q = 1, the F-statistic equals: A. The t-statistic. B. t2 . C. t. D. 1. Answer: B
​

271. When testing H0 : β1 = 0 against H1 : β1 

​ ​ = 0 at the 5% level, we reject the null if the 95%
​ ​

confidence interval: A. Contains 0. B. Does not contain 0. C. Contains 1. D. Is very wide. Answer: B
272. If we test 20 independent hypotheses at the 5% level, and all nulls are true, we expect to
reject roughly: A. 0. B. 1. C. 5. D. 20. Answer: B
273. The F-distribution has _____ degrees of freedom parameters. A. 1 B. 2 C. 3 D. n Answer: B
274. If the standard error of a coefficient doubles, the t-statistic: A. Doubles. B. Halves. C. Squares.
D. Stays the same. Answer: B
275. If the sample size n increases, the critical value for the t-test (at the same α): A. Increases. B.
Decreases (converges to Normal critical value). C. Stays constant. D. Goes to zero. Answer: B
276. Hypothesis testing is about making inferences on: A. Sample statistics. B. Population
parameters. C. Residuals. D. Data points. Answer: B
277. A Type I error occurs when: A. We reject a true null hypothesis. B. We fail to reject a false null
hypothesis. C. We reject a false null hypothesis. D. We fail to reject a true null hypothesis. Answer:
A
278. A Type II error occurs when: A. We reject a true null hypothesis. B. We fail to reject a false null
hypothesis. C. We reject a false null hypothesis. D. We fail to reject a true null hypothesis. Answer:
B
279. Increasing the significance level (α) from 0.01 to 0.05: A. Increases the probability of Type I
error. B. Decreases the probability of Type I error. C. Increases the probability of Type II error. D.
Has no effect. Answer: A
280. The p-value depends on: A. The sample data. B. The significance level. C. The population
parameter (true value). D. The alternative hypothesis only. Answer: A
281. Testing H0 : β1 = β2 = β3 = 0 requires: A. Three separate t-tests. B. One F-test with q = 3. C.
​ ​ ​ ​

One t-test. D. One F-test with q = 1. Answer: B

282. In the F-statistic formula, if SSRr = SSRur , then: A. F = 0. B. F = 1. C. F is infinite. D. The
​ ​

null is rejected. Answer: A

283. We use the "Restricted Model" to find: A. The error variance. B. The increase in SSR when
variables are removed. C. The intercepts. D. The predicted values. Answer: B
284. To test if x1 has a positive effect on y , we use: A. A two-tailed test. B. A right-tailed t-test. C. A
​

left-tailed t-test. D. An F-test. Answer: B

285. If the 99% confidence interval is [2.5, 4.5], what can we say about the 95% CI? A. It is wider
(e.g., [2.0, 5.0]). B. It is narrower (e.g., [2.8, 4.2]). C. It is the same. D. It depends on n. Answer: B
286. If we fail to reject the null hypothesis, it means: A. The null is definitely true. B. The null is
definitely false. C. There is insufficient evidence to conclude the null is false. D. The data is flawed.
Answer: C
287. Which statistic is used to test if two coefficients sum to 1 (H0 : β1 + β2 = 1)? A. F-statistic
​ ​ ​

only. B. t-statistic (or F-statistic with q = 1). C. Chi-square. D. R-squared. Answer: B

288. The standard error of β^j depends on σ . In practice, we use: A. σ . B. σ
​ ​ ^ (SER). C. 0. D. 1. Answer:
B
289. If a variable is statistically significant, the confidence interval for its coefficient: A. Includes
zero. B. Does not include zero. C. Is very wide. D. Is very narrow. Answer: B
290. When n is very large, the t-distribution converges to: A. Standard Normal (Z ). B. F-distribution.
C. Uniform. D. Chi-square. Answer: A
291. If Fstat < CriticalValue, we: A. Reject the null. B. Fail to reject the null. C. Conclude variables
​

are significant. D. Increase sample size. Answer: B

292. The F-test is valid only if errors are: A. Heteroskedastic. B. Homoskedastic and Normal (or large
sample). C. Correlated. D. Zero. Answer: B
293. Can we use an F-test for a single restriction? A. No. B. Yes, and F = t2 . C. Yes, and F = t. D.
Only if n > 100. Answer: B
294. In the regression output, "P>|t|" typically refers to: A. The one-tailed p-value. B. The two-tailed
p-value. C. The critical value. D. The standard error. Answer: B
295. If we test H0 : β1 = 0 and H1 : β1 < 0, and get t = 2.5: A. Reject the null. B. Fail to reject
​ ​ ​ ​

(wrong sign). C. Reject if α = 0.01. D. Data error. Answer: B

296. Confidence intervals are derived from: A. The point estimate and the standard error. B. The R2 .
C. The F-statistic. D. The residuals. Answer: A
297. A "joint" hypothesis test involves: A. Only one parameter. B. Multiple parameters simultaneously.
C. Only the intercept. D. Only the error term. Answer: B
298. If variables are jointly significant, it implies: A. All of them are individually significant. B. At least
one of them is significant (or they work together). C. None are significant. D. The model is perfect.
Answer: B
299. The critical value for F depends on: A. Numerator degrees of freedom only. B. Denominator
degrees of freedom only. C. Both numerator and denominator df. D. Sample size only. Answer: C
300. If we change the units of y , the t-statistic: A. Changes. B. Remains unchanged. C. Doubles. D.
Becomes zero. Answer: B
 Chapter 5: OLS Asymptotics
301. Asymptotic properties refer to properties that hold when: A. Sample size n is small. B. Sample
size n approaches infinity. C. Number of variables k is large. D. R2 is large. Answer: B
302. Consistency means that as n → ∞, the estimator: A. Converges in probability to the true
parameter. B. Becomes normally distributed. C. Has zero variance. D. Equals zero. Answer: A
303. OLS is consistent under assumptions: A. MLR.1 through MLR.4. B. MLR.1 through MLR.6. C. Only
MLR.2. D. Only MLR.6. Answer: A
304. Inconsistency (Asymptotic Bias) persists even if: A. We remove variables. B. We increase the
sample size. C. We use t-tests. D. We check for errors. Answer: B
305. If x is correlated with u, the OLS estimator is: A. Unbiased and Consistent. B. Biased but
Consistent. C. Biased and Inconsistent. D. Unbiased but Inconsistent. Answer: C
306. Asymptotic Normality implies that in large samples, the distribution of β^j is approximately: A.
​ ​

Uniform. B. Normal. C. T-distributed (with low df). D. F-distributed. Answer: B

307. The "square root n" rule implies that standard errors decrease at a rate of: A. 1/n B. 1/ n C. ​

n D. n2 Answer: B
​

308. The Lagrange Multiplier (LM) test is an alternative to the F-test that requires estimating: A.
Only the unrestricted model. B. Only the restricted model. C. Both models. D. No models. Answer:
B
2
309. The LM statistic is calculated as: A. SSR ⋅ n B. n ⋅ Raux (from auxiliary regression of residuals).
​

2
C. F ⋅ q . D. t . Answer: B
310. The LM statistic follows which distribution asymptotically? A. Normal B. t C. Chi-square (χ2q ) D.
​

F Answer: C
311. Which assumption is NOT needed for consistency? A. Zero conditional mean (E(u∣x) = 0). B.
Normality of errors (u ∼ N ormal). C. Random sampling. D. No perfect collinearity. Answer: B
312. Ideally, for large sample inference, we want the error variance σ 2 to be: A. Small. B. Large. C.
Infinite. D. Negative. Answer: A
313. If plim(β^) = β , the estimator is: A. Unbiased. B. Efficient. C. Consistent. D. Normal. Answer: C
​

314. The term "plim" stands for: A. Probability Limit. B. Polynomial Limit. C. Partial Limit. D. Parameter
Linear. Answer: A
315. Asymptotic efficiency of OLS requires: A. Only MLR.1-4. B. MLR.1-5 (Gauss-Markov assumptions).
C. Normality. D. Small samples. Answer: B
316. In large samples, the t-distribution looks like: A. The Chi-square distribution. B. The Standard
Normal (Z) distribution. C. The F distribution. D. The Uniform distribution. Answer: B
317. Does heteroskedasticity affect consistency? A. Yes, it makes OLS inconsistent. B. No, OLS
remains consistent. C. Yes, if sample is small. D. Only if errors are non-normal. Answer: B
318. The auxiliary regression for the LM test regresses: A. y on x. B. Residuals from the restricted
model on all independent variables. C. Residuals on predicted values. D. x1 on x2 . Answer: B
​ ​

319. For the LM test, if n ⋅ R2 exceeds the critical value, we: A. Reject the null hypothesis. B. Fail to
reject the null. C. Accept the null. D. Re-calculate. Answer: A
320. Large sample properties justify using OLS even when: A. Errors are not normal. B. Estimators are
biased in finite samples. C. Variables are correlated with errors. D. Both A and B. Answer: D
321. If correlation between x and u decreases, inconsistency: A. Increases. B. Decreases. C. Stays
same. D. Becomes zero. Answer: B
322. The "asymptotic variance" is: A. The variance limit as n → 0. B. The variance approximation for
large n. C. Always smaller than finite sample variance. D. The variance of the population. Answer:
B
323. Which test is typically easier to compute if the unrestricted model is hard to estimate? A. F-
test B. LM test C. t-test D. Wald test Answer: B
324. If a variable is omitted that is correlated with x, β^1 converges to: A. β1 B. β1 + Bias term C. 0
​ ​ ​ ​

D. ∞ Answer: B
325. The Law of Large Numbers (LLN) generally supports: A. Unbiasedness. B. Consistency. C.
Homoskedasticity. D. Multicollinearity. Answer: B
326. The Central Limit Theorem (CLT) generally supports: A. Consistency. B. Asymptotic Normality. C.
Zero mean. D. Efficiency. Answer: B
327. If OLS is inconsistent, collecting more data: A. Fixes the problem. B. Makes the estimate
converge to the wrong value more tightly. C. Increases variance. D. Makes errors normal. Answer:
B
328. Asymptotic standard errors are obtained by plugging σ
^ into the variance formulas: A. True. B.
False. Answer: A
329. The degrees of freedom adjustment (n − k − 1) matters less as: A. n gets larger. B. n gets
smaller. C. k gets larger. D. R2 gets larger. Answer: A
330. Which assumption allows us to treat regressors as fixed in repeated samples for finite sample
analysis? A. Random sampling. B. Zero conditional mean. C. Fixed regressors (alternative to
random sampling). D. Normality. Answer: C

Chapter 6: Further Issues

331. Beta coefficients are obtained by: A. Multiplying variables by 100. B. Standardizing variables
(subtract mean, divide by SD) before regression. C. Taking logs. D. Dividing by n. Answer: B
332. Beta coefficients measure effect sizes in units of: A. Dollars. B. Standard deviations. C.
Percentages. D. Probability. Answer: B
333. If we multiply the dependent variable y by a constant c, the t-statistics: A. Multiply by c. B.
Divide by c. C. Stay the same. D. Square. Answer: C
334. If we multiply an independent variable x by c, its coefficient β^: A. Is multiplied by c. B. Is divided
​

by c. C. Stays the same. D. Becomes negative. Answer: B

ˉ 2 ) is preferred to R2 when: A. Comparing models with different numbers of
335. Adjusted R2 (R
variables. B. The sample size is small. C. There is heteroskedasticity. D. The data is time series.
Answer: A
336. Adjusted R2 increases if: A. Any variable is added. B. A new variable's t-statistic is greater than 1
(in absolute value). C. SSR increases. D. SST decreases. Answer: B

2
337. A quadratic model y = β0 + β1 x + β2 x2 + u allows for: A. Constant marginal effects. B.
​ ​ ​

Increasing or decreasing marginal effects (turning points). C. Percentage changes. D. Binary

outcomes. Answer: B
338. In the model y = β0 + β1 x + β2 x2 , the marginal effect of x is: A. β1 B. β2 C. β1 + 2β2 x D.
​ ​ ​ ​ ​ ​ ​

β1 + β2 x Answer: C
​ ​

339. The turning point of the quadratic y = β0 + β1 x + β2 x2 occurs at: A. x = −β1 /(2β2 ) B. ​ ​ ​ ​ ​

x = −β1 /β2 C. x = 0 D. x = β2 /β1 Answer: A

​ ​ ​ ​

340. An interaction term β3 (x1 ⋅ x2 ) allows: A. The effect of x1 to depend on x2 . B. The effect of x1
​ ​ ​ ​ ​ ​

to be constant. C. The intercept to change. D. The error to be reduced. Answer: A

341. In y = β0 + β1 x1 + β2 x2 + β3 (x1 x2 ), the partial effect of x1 is: A. β1 B. β3 C. β1 + β3 x2 D.
​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

β1 + β2 Answer: C
​ ​

342. To compare a log-linear model (logy ) and a linear model (y ), we can use: A. Standard R2 . B.
Adjusted R2 . C. The squared correlation between y and y^ (after exponentiating predicted log ​

values). D. We cannot compare them. Answer: C

343. When predicting y from a log model log(y) = …, simply calculating exp(logy ): A. ​

Systematically underestimates y . B. Systematically overestimates y . C. Is unbiased. D. Is

impossible. Answer: A
344. The correction factor for predicting y from a log model (assuming normal errors) is: A.
^ 2 /2) B. σ
exp(σ ^ 2 C. 1/n D. 0.5 Answer: A
345. Duan's smearing estimate is used when: A. Errors are normal. B. Errors are not normal (for
predicting y from log model). C. Sample size is small. D. There is heteroskedasticity. Answer: B
346. A prediction interval for y 0 is _____ a confidence interval for E(y 0 ∣x). A. Wider than B. Narrower
than C. The same size as D. Unrelated to Answer: A
347. The variance of the prediction error is: A. Var(y^0 ) B. σ 2 C. Var(y^0 ) + σ 2 D. 0 Answer: C ​ ​

348. Residual analysis helps to: A. Identify outliers or influential observations. B. Prove causality. C.
Increase R2 . D. Calculate coefficients. Answer: A
349. If a variable is "over-controlled" (e.g., controlling for beer consumption when testing beer tax
effects on fatalities), we mask: A. The total effect of the policy. B. The error term. C. The
intercept. D. The sample size. Answer: A
350. Centering variables around their means before creating interactions: A. Changes the model fit (
R2 ). B. Makes the coefficients on the main effects interpretable as effects at the mean. C. Biases
the estimates. D. Removes multicollinearity completely. Answer: B
351. Can Adjusted R2 be negative? A. Yes. B. No. C. Only in time series. D. Only in simple regression.
Answer: A
352. Log-log models yield coefficients interpreted as: A. Slopes. B. Elasticities. C. Semi-elasticities. D.
Probabilities. Answer: B
353. Semi-elasticity is the interpretation for: A. Log-Log models. B. Level-Level models. C. Log-Level
or Level-Log models. D. Quadratic models. Answer: C
354. Using log(1 + y) allows us to: A. Handle cases where y = 0. B. Increase R2 . C. Reduce sample
size. D. Eliminate bias. Answer: A
355. "Goodness of fit" should not be the only criterion for model selection because: A. It ignores
statistical significance. B. It encourages overfitting (including irrelevant variables). C. It is hard to
calculate. D. It is always low. Answer: B
356. Confidence intervals for predictions are narrowest at: A. The extreme values of x. B. The sample
mean of x. C. x = 0. D. x = ∞. Answer: B
357. A standardized variable has a mean of __ and variance of __. A. 0, 1 B. 1, 0 C. 0, 0 D. 1, 1 Answer:
A
358. If the interaction term β3 in y = β0 + β1 x1 + β2 x2 + β3 x1 x2 is statistically significant: A. The
​ ​ ​ ​ ​ ​ ​ ​ ​

effect of x1 depends on x2 . B. The model is linear. C. There is multicollinearity. D. x1 and x2 are

​ ​ ​ ​

correlated. Answer: A
359. Non-nested models are models where: A. One is a special case of the other. B. Neither is a
special case of the other. C. They have the same variables. D. They have no error term. Answer: B
360. When comparing non-nested models, we prefer the one with: A. Higher Adjusted R2 . B. Lower
Adjusted R2 . C. More variables. D. Fewer variables. Answer: A
361. Polynomial regression involves: A. Powers of x (e.g., x2 , x3 ). B. Logs of x. C. Lags of x. D.
Dummy variables. Answer: A
362. If the marginal effect of x diminishes as x increases, the coefficient on x2 should be: A.
Positive. B. Negative. C. Zero. D. One. Answer: B
363. The "optimal" level of x in a quadratic model is found by: A. Setting the derivative w.r.t x to zero.
B. Setting x = 0. C. Setting y = 0. D. Maximizing R2 . Answer: A
364. Standard errors of predictions _____ as we move away from the data center. A. Decrease B.
Increase C. Stay constant D. Become zero Answer: B
365. The 95% prediction interval contains approximately 95% of: A. The true population means. B.
The future observations. C. The coefficients. D. The residuals. Answer: B
366. Data scaling affects: A. t-statistics. B. F-statistics. C. Coefficient estimates. D. R2 . Answer: C
367. Beta coefficients are useful for: A. Comparing relative strength of regressors. B. Fixing
heteroskedasticity. C. Removing outliers. D. Increasing n. Answer: A
368. If y is salary and x is sales, log(salary) = 4 + 0.3log(sales) implies a 1% increase in sales
leads to: A. 0.3% increase in salary. B. 30% increase in salary. C. $0.3 increase in salary. D. 3%
increase in salary. Answer: A
369. The adjusted R2 formula is: A. 1 − (1 − R2 )(n − 1)/(n − k − 1) B. 1 − R2 C. R2 /k D.
(SSR/n)/(SST /n) Answer: A
370. Choosing variables solely to maximize Adjusted R2 : A. Is the best strategy. B. Can lead to bias if
theoretical controls are omitted. C. Guarantees causality. D. Eliminates errors. Answer: B

Chapter 7: Qualitative Information (Dummy Variables)

371. A dummy variable takes on values: A. 0 and 1. B. 1 and 2. C. -1 and 1. D. Any integer. Answer: A
372. In y = β0 + δ0 f emale + β1 educ + u, δ0 measures: A. The slope of education. B. The
​ ​ ​ ​

difference in intercepts between females and males (base group). C. The predicted wage for
males. D. The error variance. Answer: B
373. The "Dummy Variable Trap" occurs when: A. You use too many dummies. B. You include dummies
for all groups AND the intercept (perfect collinearity). C. You use binary variables. D. The
coefficients are zero. Answer: B
374. To avoid the dummy variable trap with g groups, you should include: A. g dummies. B. g − 1
dummies (plus intercept). C. 1 dummy. D. No dummies. Answer: B
375. The "Base Group" (or Benchmark Group) is the group: A. Represented by a dummy variable. B.
Without a dummy variable (captured by the intercept). C. With the highest value. D. With the
lowest value. Answer: B
376. If the dependent variable is log(y) and the dummy coeff is 0.20, the approximate percentage
difference is: A. 20% B. 0.2% C. 2% D. 200% Answer: A
377. The exact percentage difference for a dummy coefficient β in a log-linear model is: A. 100β B.
100(eβ − 1) C. eβ D. β 2 Answer: B
378. An ordinal variable (e.g., credit rating 1 to 5) should typically be entered as: A. A single
continuous variable. B. A set of dummy variables (e.g., rate2, rate3...). C. A log variable. D. It
cannot be used. Answer: B
379. An interaction term between a dummy D and continuous variable x (D ⋅ x) allows for: A.
Different intercepts. B. Different slopes for the groups. C. Constant slopes. D. Reduced variance.
Answer: B
380. The Chow Test is used to test: A. Heteroskedasticity. B. Whether regression coefficients differ
across two groups (structural break). C. Normality. D. Serial correlation. Answer: B
381. The Chow Test is essentially an: A. F-test. B. t-test. C. LM test. D. Z-test. Answer: A
382. In the Linear Probability Model (LPM), the dependent variable is: A. Continuous. B. Binary (0 or
1). C. Logged. D. Always positive. Answer: B
383. In an LPM, y^ is interpreted as: A. The predicted probability of success (P (y = 1∣x)). B. The log
​

odds. C. The odds ratio. D. A certain outcome. Answer: A

384. A major flaw of the LPM is: A. Coefficients are hard to interpret. B. Predictions can be < 0 or
> 1. C. It is biased. D. It requires complex software. Answer: B
385. Another flaw of the LPM is that the error term is always: A. Heteroskedastic. B. Homoskedastic.
C. Normal. D. Zero. Answer: A
386. The coefficients in an LPM measure: A. Change in log odds. B. Change in probability. C. Elasticity.
D. Variance. Answer: B
387. If we interact all variables with a group dummy, this is equivalent to: A. Running separate
regressions for each group. B. Using the Chow test. C. Adding one dummy. D. Doing nothing.
Answer: A
388. Self-selection bias in program evaluation (e.g., job training) implies: A. Participation is random.
B. Participants differ systematically from non-participants. C. The treatment effect is zero. D. We
need more data. Answer: B
389. The "Average Treatment Effect" (ATE) measures: A. The effect on the treated only. B. The effect
averaged over the population. C. The effect on the controls. D. The maximum effect. Answer: B
390. Regression adjustment helps control for: A. Observed differences (covariates). B. Unobserved
differences. C. Random error. D. Calculation mistakes. Answer: A
391. If y is discrete (e.g., number of kids 0, 1, 2...), OLS coefficients measure: A. Effect on the
expected value (mean). B. Effect on the probability of having 1 kid. C. Nothing useful. D. Integer
changes only. Answer: A
392. If we define M ale = 1 if male, and F emale = 1 if female, and include both plus an intercept,
we have: A. A good model. B. Perfect multicollinearity. C. A nonlinear model. D. Heteroskedasticity.
Answer: B
393. To test if returns to education differ by gender, we test the coefficient on: A. F emale. B.
Educ. C. F emale ⋅ Educ. D. The intercept. Answer: C
394. If the confidence interval for the dummy coefficient includes 0: A. The groups are significantly
different. B. The groups are not significantly different. C. There is bias. D. The dummy variable trap
occurred. Answer: B
395. The "percent correctly predicted" is a goodness-of-fit measure for: A. Simple regression. B.
Binary dependent variable models (LPM). C. Time series. D. Log models. Answer: B
396. In program evaluation, the "control group": A. Receives the treatment. B. Does not receive the
treatment. C. Is excluded from analysis. D. Is the base year. Answer: B
397. A "difference-in-means" statistic is equivalent to the slope in a regression of y on: A. A single
dummy variable. B. A continuous variable. C. A log variable. D. Nothing. Answer: A
398. If a program is randomly assigned: A. Selection bias is minimized/eliminated. B. Selection bias is
high. C. We cannot use OLS. D. We need many controls. Answer: A
399. Covariates in program evaluation are used to: A. Increase bias. B. Adjust for confounding factors.
C. Reduce sample size. D. Make the model linear. Answer: B
400. The interpretation of δ0 in log(wage) = β0 + δ0 f emale + u is approximately: A. The dollar
​ ​ ​

difference in wages. B. The percentage difference in wages. C. The slope of wages. D. The
intercept for men. Answer: B
401. If we have 4 regions (North, South, East, West), how many dummies do we include in the
regression (with intercept)? A. 4 B. 3 C. 2 D. 1 Answer: B
402. If we use "South" as the base group, the coefficient on "North" represents: A. The average for
North. B. The difference between North and South. C. The difference between North and the
national average. D. The sum of North and South. Answer: B
403. In the LPM, the variance of the error term depends on: A. y . B. The values of the independent
variables (probabilities). C. It is constant. D. The sample size. Answer: B
404. To fix heteroskedasticity in LPM, we often use: A. Robust standard errors. B. OLS. C. Larger
sample. D. Fewer variables. Answer: A
405. Which model ensures predicted probabilities are between 0 and 1? A. LPM B. Logit or Probit. C.
OLS. D. WLS. Answer: B
406. The Chow test follows which distribution? A. t B. F C. Chi-square D. Normal Answer: B
407. If the interaction term D ⋅ x is insignificant: A. The slopes are different. B. The slopes are not
statistically different. C. The intercepts are different. D. The model is invalid. Answer: B
408. Using a dummy dependent variable with OLS is called: A. Probit. B. Logit. C. Linear Probability
Model. D. Tobit. Answer: C
409. If a dummy variable is 1 for 50% of the sample, its variance is: A. 0.25 B. 0.50 C. 1.00 D. 0.00
Answer: A
410. In a study of discrimination, if the dummy for "minority" is negative and significant after
controlling for qualifications: A. There is evidence of potential discrimination. B. There is no
discrimination. C. The model is wrong. D. Qualifications don't matter. Answer: A
411. Which is a valid reason to use LPM over Logit/Probit? A. It is easier to interpret coefficients
directly. B. It always gives better predictions. C. It has no heteroskedasticity. D. It fits the data
perfectly. Answer: A
412. If we want to test if a program had ANY effect (intercept or slope), we test joint significance of:
A. The dummy and all its interactions. B. Just the dummy. C. Just the interactions. D. The intercept.
Answer: A
413. Intercept shifts allow for: A. Different slopes. B. Parallel regression lines with different heights. C.
Curved lines. D. Zero error. Answer: B
414. Can we use R2 to compare an LPM and a Logit model? A. Yes directly. B. No, they are estimated
differently. C. Only if N is large. D. Always. Answer: B
415. With panel data, a dummy variable for "Year" captures: A. Differences across units. B. Time-
specific effects common to all units. C. Measurement error. D. Slope changes. Answer: B
416. Policy analysis often uses "Difference-in-Differences" which relies on: A. Dummy variables for
time and treatment group. B. Only continuous variables. C. Only one period. D. Random guessing.
Answer: A
417. The estimated propensity score is often used to: A. Replace OLS. B. Match control and
treatment units. C. Calculate R2 . D. Test for normality. Answer: B
418. If we reject the null in a Chow test, it implies: A. There is a structural break (groups are different).
B. Groups are identical. C. There is heteroskedasticity. D. We need more data. Answer: A
419. Usually, the group with Dummy = 0 is called the: A. Reference category. B. Treatment group.
C. Outlier. D. Variable. Answer: A
420. If t stat for a dummy is 0.5, the difference between groups is: A. Statistically Significant. B. Not
Statistically Significant. C. Huge. D. Negative. Answer: B

Mixed Review
421. Bias vanishes as n → ∞ for: A. Unbiased estimators. B. Consistent estimators. C. All estimators.
D. Biased estimators. Answer: B
422. Which assumption allows us to separate the effect of x from the error u? A. Zero conditional
mean (E(u∣x) = 0). B. Homoskedasticity. C. Normality. D. Linear parameters. Answer: A