关键词 > STAT411/616
STAT 411/616 Exam 2-Spring 2022
发布时间:2022-04-29
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
STAT 411/616 Exam 2-Spring 2022
Instructions:
1. OPEN NOTES/OPEN BOOK/INTERNET PERMITTED. Calculators are permitted under Honor Code conditions. You may not speak about this exam with any other student during the exam period.
2. This is a task-based exam. It is designed to equip you with real-world problem-solving experience. As such, it will present various vignettes to you for your solution and explanation. You may use whichever methods you can justify to analyze the data.
This instrument will test your ability to follow instructions, and to make progress with lack of explicit direction. You will pursue research/application ideas. You will work with varied data and formats in a minimum of time. You will demonstrate your ability to use appropriate statistical software test procedures. You will learn to use appropriate statistical results reporting phraseology.
For each task you should:
Summarize your data in verbal, tabular or graphic form.
Perform EDA on the data.
Explain what the testing problem is about.
You should clearly state the null hypothesis and what its rejection means.
When presenting software output, define terms or methods, such as “Pairwise comparisons using t tests with pooled SD,” explain why we pool or do not pool the data. When using a specific method, briefly explain its usage, assumptions, and interpretation. For instance, in discussing Kruskal-Wallace technique, one could describe it as “one-way ANOVA on ranks,” or ”a nonparametric test that can be used to determine if there are statistically significant differences between two or more groups of an independent variable on a continuous or ordinal dependent variable.” (For example, see #Failed ANOVA assumptions in code/Non-parm.Examples.r)
Use of more powerful parametric method is preferred if the data is truly known parametrically. But, do your methods assume certain distribution or moment conditions? If these are not, fix, or use different methods with fewer population assumptions.
The well written report is subject to Professional Standards (see website), and must be turned in on Canvas NLT 0600 on May 4, 2022. The exam will be available starting 0600 CST on 28 April. You do not have to typeset mathematical derivations unless you so choose. NOTE: If the solution is correct, typeset math is
easier to read, and therefore easier to score higher.
Please exhibit a respectful attitude on the test, flippant answers will be penalized.
This is a pledged exam. You may NOT work with any other student or person on this exam. Each student will submit their own report. It is envisioned that you will use the internet for parts of this exam. Please try and use the notation in the course notes if possible, it will help you avoid confusion.
You may pose questions to the moderator(s) only in the form of a discussion on Canvas. The moderator will attempt to answer your question in a timely fashion pending availability. As always, in the absence of direct guidance, pursue your path and justify your assumptions and conclusions.
Additional requirements:
The report for the exam be in a SINGLE .pdf file, uploaded to Canvas at or before the due date and time. Your report must include:
A cover page with name, class (STAT 411 or STAT 616), assignment number, and date;
All of the code you used to solve each problem, along with RELEVANT output. You can include this as part of the body of the report, or you can put all of the code in an appendix of the report. Do not include incorrect code or error messages, unless they are related to some point you are raising in your discussion. All code/output must be in a monospace font (not proportional);
Handwritten or typed mathematic derivations or proofs if needed;
Any graphs that you used in answering the question. See R's help facility or visit https://statistics.berkeley.edu/computing/saving-plots-r for information on saving graphs from R in a
form that can be incorporated into your report;
Page Limit: Please only include the documentation necessary to establish your conclusions and justify your assertions. Superfluous printouts and inclusion of error messages will hurt your Q:Q ratio (quantity to quality). It is expected that no problem should take more than about 7 pages (including graphs) to adequately respond.
3. There are 7 problems on the test, totaling 255 points. You must show all work for full credit. Simplify your answers as much as possible. Please write legibly.
4. Under the Honor Code, you must NOT DISCUSS this test with anyone who hasn’t taken it yet. Note that points will be deducted for incorrect pledge statement/signature. Be calm, think, remember what you know. Try and work the problems in the easiest way possible.
State and Sign the Pledge: (10 Points)
_____________________________
Signature
______________________________411/616
Print Your Name (circle one)
________________________________________
Date Exam Taken Start/End Time
1. (50 points) Returns on the Major U.S. Stock Exchanges (New York Stock Exchange (NYSE), American Stock Exchange (AMEX) and NASDAQ) for the period 12/31/1926 through 12/31/2018 are in the file exam2.indexReturns.csv in this Exam's data directory which is on Canvas in Canvas/Files/Data/_Exam 2 Data. NOTE: For your convenience and support, we also provide the long-form dataset that R requires, that filename is exam2.indexReturns.long.csv.
Universe refers to the major U.S. stock exchanges, (New York Stock Exchange (NYSE), American Stock Exchange (AMEX) and NASDAQ). Each of these indexes quote and trade thousands of securities. The return data is the annual percent return in an index (NYSE, AMEX, NASDAQ) for the year end date indicated.
There are four ways to calculate an index returns, these are the "type" factor. Each level corresponds to a different constituent stock weighting scheme in producing the index level, for which the annual returns (in percent) are found. These levels are
vwretd Market capitalization-weighted return with dividend
vwretx Market capitalization-weighted return without dividend
ewretd Equal-weighted return with dividend
ewretx Equal-weighted return without dividend
The questions we seek to answer include:
• If there a difference in the universe considered?
• Is there a difference between types of market returns, i.e., EW and MW, with and without dividends?
• Are there interactions between the universe and the return type, and what is their meaning?
In addition, please answer the following questions (10 points each):
a. Obtain the mean, median, and geometric mean (CAGR) annual returns for each universe and return type. Discuss these results.
b. Describe the data: time range, frequency, summary statistics, etc. Note that R's Anova and other functions require long-form data, which has been provided to you.
c. Check all major parametric Anova assumptions. You are familiar with normality and HOV diagnostics; independence can be checked with a runs test. Be sure and order your universe factors to NYSE, AMEX and NASDAQ, otherwise R assumes they are alphabetical, and we want to make comparisons relative to the NYSE.
d. Assuming you reject the omnibus hypothesis (be sure and state it), perform post-hoc testing to determine which regions of the market are significantly different.
e. Using R's {pwr} package or an online power program or GPower (freely available for Win or Mac OS-X; see tutorial provided in the Exam data directory), perform a power analysis for the problem. For this data, what power did we achieve? Determine the sample size required to detect a ± 4% difference in mean returns with a probability (power) of 80%.
2. (40 points) An observational study obtained decibel sound pressure level (SPL) data for an exponential horn (type of loudspeaker), as a function of distance from the audio source. The data is available in Canvas/Files/Data/_Exam 2 Data/spl.txt.
a. Prepare a comparative analysis of different
regression models for this data. You should
consider the linear model, and local (curvilinear)
models, such as polynomial, spline and LOESS
regression. Evaluate the model fits and provide
your conclusion about which models best get at
the data generating process. In your quantitative
criteria be sure and include MAE and RMS error,
as well as your recommendations for final
model. Devise a tabular comparison to facilitate
review of your models.
b. Unfortunately, the SPL sensor is subject to
detecting random bursts of energy, resulting in
occasional abnormally high readings. The data
including these bad readings are in the file
spl.contaminated.txt. Based on visual
inspection, you would repeat your previous
models and also include robust repression, such
as quantile regression (at least try the median), and Kendall-Theil regression. Be sure an provide pseudo R2 (one can use the nagelkerke function in the {rcompanion} package). Perform a complete analysis and comparison, make your conclusions and recommendation for the best model to use for this system.
3. (40 points). Maggie Simpson speaker ratings. Maggie gave a speech and her delivery was rated by
10 different instructors on a 5-point Likert scale (ordinal). Using the dataset exam2.maggie.txt in the exam data directory, answer the following questions:
a. What was her median score?
b. What were the first and third quartiles for her scores?
c. Are the data reasonably symmetric about their median?
d. Based on this, what is null hypothesis of the test?
e. According to the one-sample Wilcoxon signed-rank test, are her scores significantly different from a neutral score of 3?
f. Is the confidence interval output from the test useful in answering the previous question?
g. Overall, how would you summarize her results? Be sure to address the practical implication of her scores compared with a neutral score of 3.
h. Do these results reflect what you would expect from looking at the bar plot?
4. (50 points) Football Kickers Analytics. (Former athlete's company). One of our STAT 616 alumni and Rice football players started a company which provides training and technology (Pro-Posts) to improve the accuracy of field goals and extra points made by kickers.
The data is in simpleKicking.1718dataraw.xlsx. You will find that the number of teams using the Pro-Posts system in competitive use during the 2017 and 2018 seasons is 22 out of 413. Your assignment is to evaluate the Simple Kicking system's efficacy, that is, is the Simple Kicking system an effective training tool? As always, motivate and justify your conclusions.
You will find the necessary data dictionary in the notes tab in the spreadsheet. There are also links to some sites which have data for field goals to provide context if needed.
In addition to field goal performance, your report should include treatment of extra points. In accordance with Professional Standards, you should prepare your analysis as if it were a white paper used by Simple Kicking in a sales or venture capital solicitation situation.
You have much flexibility in how you define efficacy, obviously it revolves around is there a difference between the kickers using Pro-Posts and those not? The more interesting papers will consider the improvements in the various levels of field goal distance
5. (30 points) Piglet and Pooh ratings. This test is designed to tell if there is a significant difference in Pooh's speaking scores over Piglet's. The (two-sample) Mann-Whitney U test is conducted with the wilcox.test function, which produces a p-value for the hypothesis. First the data should be summarized and examined using bar plots for each group. If the bar plots show that the distributions of scores for Pooh and Piglet are relatively similar in shape, the Mann-Whitney U test can be interpreted as a test of medians.
Using the dataset exam2.pooh.txt in the exam data directory, answer the following questions:
a. What was the median score for each instructor?
b. What were the first and third quartiles for each instructor’s scores?
c. Are the data for both instructors reasonably similar in shape and spread?
d. Based on your previous answer, what is the null hypothesis for the Mann–Whitney test?
e. According to the Mann–Whitney test, is there a difference in scores between the instructors?
f. How would you summarize the results of the descriptive statistics and tests? Include practical considerations of any differences.
6. (20 points) This question has to do with expected mean squares. In Exam 1 we asked about the logic of Anova. One way to state it is in terms of expected mean squares of treatment and errors. Although you will rarely need to know the expected value of MST or MSE, it is important to see that both expected values are the same when the null hypothesis is true and that the expected value of MST is larger when the null hypothesis is false.
Under H0 , E(MStmt) = E(MSerr) = σ2 , and under H1 , E(MStmt) > σ2 so that the resulting F ratio can increase. You may recall that SStmt = ∑ ni (xi − x )2 = ∑ nixi − nx 2 .
a. For a single factor Anova, find/derive E(MSerr).
b. Find/derive E(MStmt) for H0 when it is both true and false.
7. (25 points) Polynomial regression with interactions. This problem uses the pollution dataset exam2.pollute.txt in the exam data directory.
Find a parsimonious model for this data. Be sure and write the estimating equation for your final model. You should evaluate your stopping point with R2 and AIC. Do not mindlessly use stepwise search. You should consider including polynomial predictors. Be sure and consider interactions. Fully interpret your resulting model. You should be able to obtain an R2 of at least .76 with an AIC of 326 or better.