Department of Economics Spring 2009

University of California Prof. Woroch

Economics 140: MIDTERM #1 (Exam Code “A”)

SAMPLE ANSWERS
I. TRUE/FALSE/UNCERTAIN: Below are statements that may be true or false, or possibly
ambiguous. State which one you believe to be the case, and more importantly, give a detailed
but concise explanation for your answer. Each question is worth 10 points for a total of 20.

1. If Y1 ,..., Yn represent n i.i.d. draws of some random variable, then the sample “geometric mean”
~
Y = (Y1 × ... × Yn )1 / n will have a higher variance than the familiar sample arithmetic average,
1
Y = (Y1 + ... + Yn ) .
n
1
Answer: False/Uncertain. We know that with i.i.d. draw, Y = (Y1 + ... + Yn ) is BLUE, the Best
n
Linear Unbiased Estimator, which means that among the linear and unbiased estimators, the sample
~
average is the most efficient (i.e. minimum variance). However, Y = (Y1 × ... × Yn )1 / n is not linear in the
draws from the r.v. Nor is it necessarily unbiased. Thus we cannot compare variances. The
geometric average could have a lower or a higher variance than the arithmetic average.

2. If your dataset is such that both the independent ( X i ) and dependent ( Yi ) variables in a bivariate
linear regression have zero sample means, then the OLS estimator of the intercept is also zero.
)
Answer: True. The OLS estimator of the intercept in a bivariate linear regression is: β 0 = Y −
)
β̂1 * X . When ( X , Y ) = (0,0) as claimed in the statement, β 0 = 0 + β̂1 *0 = 0.

II. SHORT PROBLEMS: Answer each of the following 2 questions. Each is worth 16 points.

1. New Hampshire (NH) does not have a state income tax and yet NH MA
its neighbor Massachusetts (MA) has, at times, had one of the Sample average 37.5 36.0
highest income tax rates in the nation. To analyze the possible hours per week
effect of income tax on labor supply, you employ a dataset Sample variance 250 400
containing a random sample of NH and MA employees that is Sample size 1,500 1,200
summarized in the adjacent table.

a) Allowing for different variances in hours worked in the two states, perform a test at the 1% level
of whether employees in the two states work the same number of hours per week.
Answer:
H o : μ NH = μ MA
H 1 : μ NH ≠ μ MA
37.5 − 36 1 .5
t= = = 2 .1
250 400 1/ 2
+
1500 1200
As the critical value is 2.5, we cannot reject the null at 1% level.
 b) If you had access the original individual employee dataset for the two states, describe how you
would use regression analysis to test the same hypothesis. Give details of the steps of your work.
Answer:
Hrs = α + β NH + u where NH is a dummy variable (NH=1 for New Hampshire and 0 for MA) and
Hrs is the dependent variable, Hrs is the number of hours worked by individual in both states. Test the
null hypothesis that β =0 vs. the alternative β is different from zero. Because the OLS coefficient
represents the difference in means between state NH and MA.

2. Seeking to understand the relationship between music file sharing and CD sales, a researcher regresses
CD sales in a metro area (CDSALESi) on the fraction of that metro area’s population who are active
users of peer-to-peer services (P2PUSERSi): CDSALES i = β 0 + β1 P 2 PUSERS i + u i , for metro area i .
a) Would you expect the OLS assumption #1 (i.e. E (u i | X i ) = 0 ) to hold in this application of
regression? Explain.
Answer:
E (u i | X i ) = 0 is likely to be violated. Metro areas differ in population, more population increases the
chances of more sales and more users. Metro areas are different in terms of economic growth and
development, for example richer metro areas have more access to internet and also buy more CDs.

b) For this regression, describe how heteroskedasticity could arise in the regression, and how that
heteroskedasticity would affect homoskedasticity-only OLS estimate of the slope.
Answer:
Heteroskedasticity is likely to arise because of the wide differences in the populations of the metro
areas. Notice that CD sales are an aggregate amount for the metro area and so its average (and
variance) will be directly related to metro size, and so too will be the variance of the error term. The
fact that P2PUSERS is expressed as a fraction of the metro population will not change that
conclusion. Heteroskedasticity will not affect expressions for OLS estimator of the slope—the
formula is the same regardless. But if the researcher uses homoskedasticity-only formulas for the
standard errors (SEs) of the OLSE of the slope, they will be biased, not consistent and not distributed
normally even in large samples. Tests performed on those SEs are misleading.

III. MULTI-PART QUESTIONS: Answer each of the following 2 questions. Together, they are
worth 48 points. Point assignments appear in [square brackets] in each subpart.

1. You consider enrolling in a master’s program in economics after you graduate from Berkeley. Using
solid economic reasoning, you weigh the benefits, measured by the incremental discounted lifetime
earnings (IDLE), and the costs, in terms of tuition and reduced income while in the program. A study
conducted by the Carnegie Council on Higher Education reports that the IDLE from an economics
master varies with the economy-wide unemployment rate (UR). Not knowing the unemployment rate
that will prevail when you finish the master’s program, you place equal probability (i.e., 0.2) on each
of the five possible outcomes.

a) [4] What is the expected IDLE from graduating with a masters conditional on unemployment rate
being 8% or higher?
Answer: $233.33k

b) You are not convinced that the sample average of $300k IDLE from the Carnegie study is accurate
Economics 140 Page 2 Midterm #1 Answers
 and so you gather earnings data from the most recent Current Population Survey to test whether
the population is really $300k. You randomly select 100 graduates of masters programs and, with
some clever use of the CPS data, you find a sample average IDLE of $400k with a standard error
of $50k.
i) [4] Test the null hypothesis that the population mean is $300k against the alternative that is
different from $300k at a significance level of 5%.
Answer: Incremental
H o : μ = 300 U.S. Discounted
H 1 : μ ≠ 300 Unemployment Lifetime
Rate (UR) Earnings (IDLE)
400 − 300
t= =2 6% $450k
50 7% $350k
Reject the Null given the critical value is 1.96. 8% $275k
ii) [4] What is the p-value of the test? 9% $225k
Answer: 2Φ (−2) = 0.0456
10% $200k
2. The national debate over the design of a stimulus package has drawn attention to the role of the
concept of marginal propensity to consume. The “MPC” measures the portion of disposable income
(wage and investment income after taxes and transfers, Y) that households will spend on goods and
services (consumption, C). Below is the Excel output from an OLS regression using the Stock and
Watson add-in that attempts to estimate the MPC with a random sample of 34 households. Notice the
“missing numbers” that are indicated by the symbols: <1> through <5>.

Robust Regression Output: Dependent Variable = C

Regression Statistics
R-Squared <5> F statistic = 2361.140
Adjusted R-Squared 0.985 Prob > F = 1.000
Standard Error 345.196
Number of Obs 34 Mean (C) = 1935.294
Std Dev (C) = 2860.729

Coeff. Std. Error t stat p > |t| Lower 95% Upper 95%
Intercept 215.368 <1> 3.283 0.001 <3> 343.944
Y <2> 0.018 48.592 0.000 0.830 <4>

a) [4] Write down the theoretical regression model estimated in the table.
Answer:
C i = β 0 + β1 * Yi + u i . The question is clear about the theoretical regression model, not the
estimated regression line.

b) [12] Using information in the table, compute the four missing numbers: <1>, <2>, <3> and <4>.

Answer:
( )
< 1 >= SE βˆ0 =
215.368
3.283
= 65.601

< 2 >= βˆ1 = 0.018 * 48.592 = 0.875

< 3 >= Lower _ CI 95% (β 0 ) = 215.368 − 1.96 * 65.601 = 86.790
< 4 >= Upper _ CI 95% (β 1 ) = 0.875 + 1.96 * 0.018 = 0.910

Economics 140 Page 3 Midterm #1 Answers

 c) [5] Give the economic interpretation of the least squares estimates of the intercept and slope,
indicating how they relate to MPC.
Answer:
βˆ0 : If an individual has no disposable income, then his/her estimated consumption is $215.37, on
average. (Think this estimated coefficient as subsistence consumption provided by the
government).
β̂1 : For every one dollar increase in the individual’s disposable income, his/her consumption
increases $0.87, on average. This is the estimated MPC.

d) [5] Find the R2 of this regression (i.e., <5>) and state in words what is the meaning of that value in
the context of this application of regression analysis.
Answer:
TSS TSS
2860.729 2 = S C2 = = ⇒ TSS = 270,064,423.6
N − 1 33
SSR SSR
345.196 2 = SER 2 = = ⇒ SSR = 3,813,128.909
N −2 32
SSR
R2 = 1− ≅ 1 − 0.014 = 0.986
TSS
Then, disposable income explains 98.6% of the variation in consumption

e) [5] Construct a 95% confidence interval for the average increase in consumption associated with a
100 increase in the average disposable income.
Answer:
Given the lower and upper bounds found in the table and on b),
respectively, CI 95% (100 * β1 ) = 100 * [0.830;0.910] = [$83;$91] .

f) [5] If low-income households are believed to consume all of the income they earn, state the null
hypothesis regarding their MPC. Perform a two-sided test of this hypothesis where the “size” of
the test is 5%.
Answer:
Individuals consuming all income they earn implies β 1 = 1 . So, the test is H 0 : β1 = 1 vs.
H 1 : β1 ≠ 1 .
βˆ − 1 − 0.125
t − stat = 1 = = −6.94
( )
SE βˆ1 0.018
The critical value for a two-sided alternative is 1.96. Since |t-stat| > 1.96, we reject the null.

Economics 140 Page 4 Midterm #1 Answers