DEPARTMENT OF STATISTICS AND DATA SCIENCE

The Wharton School

University of Pennsylvania

Statistics 6130 Fall 2022

Practice Test 2.2

IMPORTANT: You have a single attempt for this test. Make sure you do not submit your
test until you are completely settled on your final answers.

Make sure you have scrolled carefully through the output to make sure you did not miss
any questions.

There are 26 multiple choice questions.

This is a closed book/closed notes test. You are allowed to use a calculator.

You can use a 1 page cheat sheet, both sides. It can be handwritten or typed.

You have one hour and twenty minutes to complete the exam.

The exam is to be completed individually. No interaction with other students is permitted.

The computer output associated with the questions should be considered an essential part
of the questions. The multiple-choice questions are equally weighted; the number of
correct answers determines your grade.

Unless otherwise stated prediction intervals should use a 95% confidence level.
All logarithms are natural logs (that is, ln or loge) unless otherwise noted.
You can use scratch paper for any calculations.

STOP
DO NOT TURN THE PAGE UNTIL YOU ARE INSTRUCTED TO PROCEED.
Stat 6130, Practice Test 2.2 -2 of 15-

1. In a multiple regression with 5 continuous predictors, the estimated equation was 𝑦𝑦� =
2 − 3𝑥𝑥1 + 4.2𝑥𝑥2 + 0.2log (𝑥𝑥3 ) + 14𝑥𝑥4 − 1.1𝑥𝑥5 . It was found that x1 and x2 had
significant slopes, while the other three predictor variables had insignificant slopes.
Which of the following is a sound next step?
A. Remove the most insignificant variable and rerun the regression with the 4
remaining variables.
B. Report as a final model 𝑦𝑦� = 2 − 3𝑥𝑥1 + 4.2𝑥𝑥2 .
C. Remove x3, x4, and x5, and rerun the regression with x1 and x2 as the
predictors.
D. Add the interaction between 𝑥𝑥2 and 𝑥𝑥5 and rerun the regression.
E. None of the above.

2. An analyst ran a multiple regression where the y-variable was the cost of a trip in an
Uber type taxi service and the two x-variables were the distance of the trip in miles
and the time of the trip in minutes. The observations were a random sample from all
trips taken in a city on a given day. Even before looking at the data, which of the
following issues can you be almost certain will arise in the regression analysis?
I. A lack of normality in the residuals.
II. Autocorrelation.
III. Collinearity.
A. III.
B. I and II.
C. I, II and III.
D. II.
E. I.

3. The probability that the Board fires the CEO of Wells Fargo in October is 0.2. If the
Board fires the CEO, the probability that the stock price falls by more than 3% is 0.7.
What is the chance that the Board fires the CEO AND the stock does not fall by more
than 3%?
A. 0.14
B. 0.06
C. 0.24
D. 0.5
E. None of the above
Stat 6130, Practice Test 2.2 -3 of 15-

4. If you play a standard roulette wheel and bet on a single number, the probability that
you win is 1/37 = 0.027. If you make this bet three times in a row and you assume
that spins are independent, what is the probability that you win at least once?
(A) 0.027.
(B) -0.921.
(C) 0.081.
(D) 0.072.
(E) 0.079.

5. The number of times an individual goes to the movies in a month is modeled as random
variable X, with the following probability distribution:

# times 0 1 2
p(x) 0.7 0.2 0.1

Given a town of 10,000 people what is the expected number of visits a month for the
whole town?
(A) 5000.
(B) 4000.
(C) 0.3.
(D) 0.4.
(E) 40.

6. The addition of a two-level categorical variable (without an interaction) to a simple

regression that contains a continuous explanatory variable

(A) Allows the slope of the continuous variable to differ across the two levels.
(B) Increases the coefficient of the continuous explanatory variable.
(C) Reduces collinearity in the fitted model.
(D) Forces the slope of the continuous variable to be the same for the two levels of the
categorical variable.
(E) Always adds significant predictive power to the model.
Stat 6130, Practice Test 2.2 -4 of 15-

Questions 7-13

A medical device company is planning on launching a new device in the US in the next 18
months. As a part of their initial market research effort, they have conducted a pricing study
to get a sense of a realistic price for the product. The product will be the first in a new
product category in the country so there is no information on competitive products.

During the study, the product was fully described to the respondents, and they were simply
asked what price they would be willing to pay for the product. So, in this question the
variable “Price” is really “what price would you be willing to buy/prescribe this product at.”

The 180 respondents to the survey were medical professionals who would be able to
prescribe the device. The following summary statistics describe the price in US dollars each
respondent stated that they would be willing to pay for the product.

90
-2.33 -1.64 -1.28 -0.67 0.0 0.67 1.28 1.64 2.33 Summary Statistics

80
Mean 57.1
Std Dev 10.8

70 N 180.0

Quantiles
60

100.0% maximum 87
50 99.5% 87
97.5% 78.9
40 90.0% 70
75.0% quartile 65
30 50.0% median 57.5
0.005 0.06 0.18 0.35 0.55 0.75 0.92
25.0% quartile 49.25
Normal Quantile Plot 10.0% 42
2.5% 36
0.5% 32
0.0% minimum 32

7. If the data was converted from $US to Japanese Yen, then which of the following
sample quantities would not change?
(A) The mean of the price.
(B) The standard deviation of the price.
(C) The standard error of the mean of price.
(D) The inter quartile range (IQR) of the price.
(E) None of the above. They would all change.
Stat 6130, Practice Test 2.2 -5 of 15-

8. If the company wanted the device to be affordable to 95% of the population, then
based on this data what approximate price would you suggest they set?
(A) $32.
(B) $30.
(C) $75.
(D) $36.
(E) $39.

9. Which of the following provides an approximate 95% confidence interval for the
population mean?
(A) $(55.5, 58.7).
(B) $(57.0, 57.2).
(C) $(35.5, 78.7).
(D) $(54.8, 59.4).
(E) None of the above.

10. If the researchers had wanted a 95% confidence interval with a width of $2, then what
would have been an appropriate sample size for the study?
(A) 233.
(B) 180.
(C) 22.
(D) 467.
(E) 43.

Questions 11-13

The researchers who had done the study described in Q7-10 had in fact implemented the
study so that there were 100 and 80 of each of two types of prescribers, Nurse Practitioners
(NPWH) and Midwives (CNM) respectively.

This stage of the study concerns comparisons between these two groups about their
willingness to pay. The outcome variable is still price. Summary statistics for these
comparisons follow. Assume that the samples are legitimate simple random samples from
their respective populations.
Stat 6130, Practice Test 2.2 -6 of 15-

90

80

70

60

50

40

30 CNM NPWH

Specialty

Quantiles

Level Minimum 10% 25% Median 75% 90% Maximum

CNM 32.0 39.2 46.5 54.5 65.8 70.0 87.0
NPWH 33.0 45.0 52.3 59.0 65.0 70.9 85.0

Means and Std Deviations

Std Err
Level Number Mean Std Dev Mean Lower 95% Upper 95%
CNM 80.0 55.6 11.8 1.3 53.0 58.3
NPWH 100.0 58.3 9.7 1.0 56.4 60.3

t Test

NPWH-CNM
Assuming unequal variances
Difference 2.705 t Ratio 1.647
Std Err Dif 1.642 DF 152.007
Upper CL Dif 5.950 Prob > |t| 0.102
Lower CL Dif -0.540 Prob > t 0.051
-4 -2 0 2 4
Confidence 0.950 Prob < t 0.949
Stat 6130, Practice Test 2.2 -7 of 15-

11. Assuming that a two-sided alternative was used in the two-sample t-test what do you
learn from the output?
(A) Nurse practitioners in the sample are willing to pay a higher average price, but
the difference is not significant.
(B) Nurse practitioners in the sample are willing to pay a higher average price,
and the difference is significant.
(C) That the population medians of the two distributions are indistinguishable.
(D) That the population means are less than 2 standard errors of the difference
between means, apart.
(E) None of the above.

12. Based on the two-sided p-value and a hypothesis test that uses α = 0.05, what sort of
error might your conclusion be subject to?
(A) A Type I error.
(B) A Type II error.
(C) Either a Type I or a Type II error, it is impossible to say.
(D) With a sample size this large (> 30) the test will not make an error.
(E) None of the above.

13. Which of the following provides an appropriate interpretation of the 95% confidence
interval for the difference in means?
(A) There is a 95% chance that the population means are equal.
(B) 95% of samples will have a confidence interval equal to (-0.540, 5.950).
(C) 95% of the complete sample of 180 data points should lie within the range
(-0.540, 5.950).
(D) There is less than a 5% chance that the means are equal.
(E) None of the above.

Questions 14-17

A financial derivative is based on the next day’s return of the S&P500 stock market index. If
the S&P500 falls the derivative pays back nothing. If the S&P500 increases by between 0
and 1% it pays back $10,000. If the S&P500 increases by more than 1% it pays back
$20,000. The probabilities of thee three events are 0.42, 0.49 and 0.09 respectively.
Stat 6130, Practice Test 2.2 -8 of 15-

14. What is the expected payout, μ, of the derivative?

(A) 15,000.
(B) 4,900.
(C) 6,700.
(D) 10,000.
(E) None of the above.

15. What is the standard deviation, σ, of the payout of the derivative?

(A) 21,256,200.
(B) 6333.246.
(C) 21,256.
(D) -5,124.367.
(E) None of the above.

16. If the derivative costs $5,000 to purchase, what is the expected profit (payout – cost) of
the derivative?

(A) 10,000.
(B) 5,000.
(C) 1,700.
(D) -1,700.
(E) None of the above.

17. Based on the probability model above, if someone claimed that the probability of two
down days in a row for the S&P500 was 0.1764, then which of the following
assumptions are they likely to be implicitly making?

(A) That returns were approximately normal.

(B) That returns were independent from day to day.
(C) That the returns on the two days were disjoint events.
(D) That Bonferroni’s inequality is in fact an equality.
(E) No assumptions need to be made except that the probability is between 0 and 1.

Questions 18 – 21
An internet streaming service was interested in understanding potential drivers of user
engagement. The company has an internal metric for individual user engagement that is
made up of a number of underlying features, for example, frequency and length of visits
to the site, interactivity on the site and so on. It is not known exactly how they calculate
the metric but that is not important for this question. Each subscriber has a monthly
Stat 6130, Practice Test 2.2 -9 of 15-

engagement number and that will be the y-variable in the following questions.
Engagement potentially takes on the values 1 through 100, with higher numbers
indicating more engagement.

There were three initial potential drivers studied:

x1 = the number of minutes in each hour devoted to commercials.
x2 = the number of times each hour the service made a recommendation to the
user of new music to sample.
x3 = the time in minutes until a user inactivity alert was pushed to the
application and the music stopped playing (no-one wants to play music to an empty
room).

The data was collected on 1000 randomly selected subscribers and all the metrics were
measured over the same month. Because the company had been performing a lot of A/B
testing, there was plenty of variation in all of the x-variables.

Graphical summaries and output from the multiple regression follows.

Correlations

Engagement Commercials Num recs Inactivity alert

Engagement 1.0000 -0.5144 0.0045 -0.4269
Commercials -0.5144 1.0000 0.0244 0.8837
Num recs 0.0045 0.0244 1.0000 0.0342
Inactivity alert -0.4269 0.8837 0.0342 1.0000
Stat 6130, Practice Test 2.2 -10 of 15-

Scatterplot Matrix

75

60

45

13
11
9
7
5

2.5
2
1.5
1
0.5
0

40
35
30
25
20
15
10
45 60 75 5 7 9 11 0 1 2 15 25 35

Mahalanobis Distances

4.0

3.0 UCL=3.076

2.0

1.0

0.0
0 100 200 300 400 500 600 700 800 900

Row Number
α = 0.05

90
85
80
75
70
65
60
55 Summary of Fit
50
45
RSquare 0.268341
40
35 RSquare Adj 0.266138
35 40 45 50 55 60 65 70 75 80 85 90 Root Mean Square Error 6.863945
Engagement Predicted RMSE=6.8639 Mean of Response 57.746
RSq=0.27 PValue<.0001 Observations (or Sum Wgts) 1000
Stat 6130, Practice Test 2.2 -11 of 15-

Parameter Estimates

Term Estimate Std Error t Ratio Prob>|t|

Intercept 70.440153 1.090706 64.58 <.0001 *
Commercials -1.719011 0.159095 -10.80 <.0001 *
Num recs 0.1118473 0.196427 0.57 0.5692
Inactivity alert 0.1524998 0.0704 2.17 0.0305 *

18. Generally speaking (that is, not specifically related to this dataset) if you only insisted
on reviewing the correlation matrix and not the scatterplot matrix, then what important
regression-related features could you potentially miss?
i. A non-linear relationship between y and one of the x’s.
ii. Leveraged data points.
iii. Potential heteroscedasticity in the regression.

(A) i.
(B) i and ii.
(C) i, ii and iii.
(D) ii and iii.
(E) ii.

19. One of the analysts in the company had been pushing recommendations as a way of
increasing user engagement. What does the above model suggest about the value of this
idea, given that Commercials and Inactivity don’t change?
(A) There is strong evidence that increasing the recommendations will increase
engagement.
(B) Increasing recommendations will decrease engagement and the change is
significant.
(C) There is no evidence that increasing recommendations will improve
engagement.
(D) The model does not provide output to address this question.
(E) None of the above.
Stat 6130, Practice Test 2.2 -12 of 15-

20. Notice that the correlation between Inactivity alert and Engagement is negative,
whereas in the multiple regression model the coefficient for Inactivity is both positive
and significant. What explains the opposite signs?
(A) This is a chance happening so there is no explanation needed.
(B) A negative interaction between the Inactivity and Commercials variables.
(C) No correlation between the Inactivity and Commercials variables.
(D) Strong correlation between the Inactivity and Commercials variables.
(E) None of the above.

21. One of the analysts had used a rule of thumb in the past which was “a 5-minute increase
in commercials per hour would be expected to result in a 5-point drop in engagement
on average”. After controlling for Num recs and Inactivity alert, what does the fitted
model suggest about this rule of thumb?
(A) It suggests that the rule of thumb is back-to-front, increasing commercials is
expected to increase engagement.
(B) There is strong evidence that the commercial’s effect is more negative and
severe than the rule of thumb suggests.
(C) It suggests that the rule of thumb is accurate.
(D) The model does not in fact address the rule of thumb in any way.
(E) There would need to be an interaction in the model, between Inactivity and
Commercials to be able to address the validity of the rule of thumb.

Questions 22 – 26
One of the questions that the company had been considering was whether there could
be “genre” effects in play. The basic idea is that people who listen to different types of
music (e.g. contemporary v. classical) might have different sensitivities to the amount
of time devoted to commercials and so on. To examine this, they classified each of their
subscribers into a predominant use bucket, which initially was in indeed contemporary
and classical. This is included as a two-level categorical variable in the regression.

Output from this new regression follows (the Num recs variable has been dropped from
the model). The leverage plot for Commercials is also included in the output.

Summary of Fit
RSquare 0.367132
RSquare Adj 0.365226
Root Mean Square Error 6.383753
Mean of Response 57.746
Observations (or Sum Wgts) 1000
Stat 6130, Practice Test 2.2 -13 of 15-

Least Squares Means Table

Least
Level Sq Mean Std Error Mean
Classical 62.905368 0.45994698 62.8446
Contemporary 56.512099 0.22476958 56.5266

Commercials

Leverage Plot
90
85
80
75
70
65
60
55
50
45
40
35
5 6 7 8 9 10 11 12 13 14 15

Commercials Leverage, P<.0001

22. Describe the Genre effect.

(A) After controlling for Commercials and Inactivity, classical listeners seem to be
more engaged on average, but the effect is not significant.
(B) After controlling for Commercials and Inactivity, classical listeners seem to be
more engaged on average, and the effect is significant.
(C) After controlling for Commercials and Inactivity, classical listeners seem to be
less engaged on average, but the effect is not significant.
(D) After controlling for Commercials and Inactivity, classical listeners seem to be
less engaged on average, and the effect is significant.
Stat 6130, Practice Test 2.2 -14 of 15-

(E) None of the above answers are reasonable interpretations of a potential Genre
effect.

23. Notice that the RMSE is much lower in this new model than the previous one (the one
used for Q18-21). What does this lower RMSE indicate about the new regression?
(A) It is an unfortunate event, but the increased R2 in the new model compensates
for it.
(B) It is a positive development because the prediction intervals from the new
regression will be narrower.
(C) It is an unfortunate event because the prediction intervals from the new
regression will be wider.
(D) RMSE should not be used to judge the quality of a regression in the first place, so
the change in RMSE is uninformative.
(E) It indicates that outliers have been excluded from the new analysis.

24. On inspecting the leverage plot for Commercials do you see cause for concern regarding
the appropriateness of this new multiple regression model?
(A) The leverage plot shows clear curvature, invalidating the linear regression.
(B) The leverage plot shows strong heteroscedasticity which should be addressed
through a log transform of y.
(C) The leverage plot shows that there are some extremely leveraged points that
should be removed from the analysis.
(D) The leverage plot is encouraging and offers no evidence against the validity of
this new model.
(E) None of the above.

25. For a new Classical subscriber who has values for Commercials and Inactivity alert
exactly equal to the mean of each these two variables as calculated over the entire
dataset (mean of Commercials = 10.17982, mean of Inactivity alert = 30.436563), which
of the following provides an approximate 95% prediction interval for their
Engagement?
(A) (50.14, 75.67)
(B) (56.52, 69.29)
(C) (44.98, 70.51)
(D) (61.99, 63.83)
(E) (59.38, 84.92)
Stat 6130, Practice Test 2.2 -15 of 15-

26. If you wanted to examine whether Classical and Contemporary listeners were
differentially sensitive to Commercials with regard to Engagement, how would you
suggest refining this current model?
(A) Remove Inactivity alert from the model and refit.
(B) Increase the size of the data set to at least 10,000 observations and refit the
current model.
(C) Add an interaction term between Genre and Commercials to this model and
refit.
(D) The model already addresses this question, so needs no refinement.
(E) None of the above.