ESE 302 Home Page

Instructor: Tony E. Smith
274 Towne (898-9647)
tesmith@seas.upenn.edu
Office Hours: By Appointment

Teaching Assistant: 
Dafeng Xu

dafeng.xu.pku@gmail.com       
Office Hours:   Fridays: 10-11 AM

                          Towne Library (Group Study Area)

 
• Devore is now available in Bookstore
  

COURSE DESCRIPTION

This course builds on ESE 301 (Engineering Probability), and introduces students to the basic methods of statistical estimation, hypothesis testing, and regression. The emphasis is on practical applications of these tools, including the analysis of a variety of real-world data sets using standard statistical software. The capstone of the course is a small-team project, typically involving pairs of individuals. Each team is expected to formulate a problem of interest, gather relevant data pertaining to the problem, and analyze this data using multiple regression techniques. The project culminates in a written report that is designed to strengthen the students’ technical writing skills.

EDUCATIONAL OBJECTIVES.   This course will introduce students to:

1. Formal statistical methods for engineering applications.
2. Practical analysis and interpretation of engineering data.
3. Methods for formulating problems in statistical terms.
4. Use of JMP sofware for regression and tests of hypotheses.
5. Technical writing and communication of analytical results.  

PREREQUISITES

(return to contents)

REQUIRED MATERIALS

 
• TEXT: Devore [D] Probability and Statistics, Seventh or Eighth Edition.

RECOMMENDED MATERIALS

 
• SOFTWARE: JMP Start Statistics , 3rd Ed. [JMP], J. Sall, A. Lehman & L. Creighton

COURSE TOPICS

(return to contents)

COURSE GRADING

Code of Academic Integrity

(return to contents)

COURSE TIME AND LOCATION
 

• Tuesdays and Thursdays, 12:00 to 1:30 PM
  
 
• Room 313, Towne Building

(return to contents)

CETS Labs
 

• JMP  is available on the machines in the CETS Labs located in M62 Towne and 100 Moore Suite.
  

• All students need ENIAC accounts to use the labs, which can be obtained from the CETS office (164 Levine, 9-5, M-F). Information on lab hours is also available there.

(return to contents)

TENTATIVE SCHEDULE FOR SPRING 2013

(return to contents)

HOMEWORK ASSIGNMENTS

HOMEWORK POLICY:

• Homework can be handed in at the end of class on the due date, or in the box outside my office (274 Towne) by 5 PM on the due date.

• Late homework assignments will be penalized. But be sure to hand them in. Missing homeworks will be dealt with severely in determining final course grades.

(return to contents)
 

PRACTICE EXAM PROBLEMS

(return to contents)
 

EXAM POLICY:

• Both exams are open book and open notes (NO LAPTOPS!). Students are expected
  
to bring a calculator and their copy of Devore to the exams.
 
• Makeup exams will only be given only in cases of (documented)
  
illness or extenuating circumstances.

(return to contents)
 

EXAM SOLUTIONS

(return to contents)
 
 

DATA SETS FOR CLASS

       All class data sets can be downloaded from the web site: http://www.seas.upenn.edu/~ese302/lab-content/

       In addition, the following homework data sets can be accessed directly::

• BLADES. This is the data set for problem 1 in Problem Set 1. The first variable is blade lifetime and the second is the blade type.

 
• POPULATION . This is the data set for problem 4 in Problem Set 2. The variable is population size of 100 midwest farming towns.

 
• BRIDGE . This is the data set for problem 1 in Problem Set 3. The variable is the number of passengers in each of 500 cars sampled during rush hour on the Ben Franklin bridge.

 
• LUMBER . This is the data set for problem 6 in Problem Set 3. The first variable is the diameter of each tree (in inches) and the second variable is volume of lumber produced from each tree (in cubic feet).

 
• HOUSING. This is the data set for problem 7 in Problem Set 3. The variables correspond to the selling price, number of bedrooms, total floor space, presence of storm windows, lot frontage, garage size, and location for each of the 26 houses.

 
• LUNGS. This is the data set for problem 2 in Problem Set 4. The variable is the pulminary compliance level for each of 100 subjects
  

 
• BATTING. This is the data set for problem 5 in Problem Set 4. The variables correspond to the batting average during the regular season and world series for each of the 11 players listed.

 
• FOOTBALL. This is the data set for problem 6 in Problem Set 4. The variables correspond to the year and average weight of the Texas football team.

 
• SPEED . This is the data set for problem 1 in Problem Set 5. The variable is the speedometer reading for each of 20 speedometers.
  

• DIET. This is the data set for problem 5 in Problem Set 5. The variables correspond to the weight before and after the diet for each of the 20 women tested.

 
• CARS. This is the data set for problem 7 in Problem Set 5. The variables correspond to the city mpg, horsepower, wheel base , and weight of each car.

• SIMPLE REGRESSION.   These notes supplement the class material on the simple regression model, and in particular, include a more detailed development of the Gauss-Markov Theorem.

• AUTOCORRELATION. These notes illustrate the problem of residual autocorrelation in regression, and also develop a simple two-stage estimation procedure which can be implemented in JMPIN to reduce the effects of autocorrelation.

• HETEROSCEDASTICITY. These notes illustrate the problem of heteroscedasticity in regression, and develop two procedures for reducing this effect. These procedures, Variance-Stabilizing Transformations and Weighted Least Squares, are each illustrated with examples.
  

• REGRESSION OUTLIERS. These notes discuss the identification and treatment of outliers in data sets. The key method for identifying outliers in JMPIN is Cook's D measure of influence. This measure is discussed, along with the issue of whether or not to omit those outliers that are identified.

• STEPWISE_REGRESSION . These notes develop the procedure of Stepwise Regression for choosing a best subset of explanatory variables from a large set of potential explanatory variables. The discussion focuses on the specific procedural steps and command options available in JMPIN.

• LOGISTIC_REGRESSION. These notes develop the method of Logistic Regression for analyzing data with dichotomous dependent variables, such as voting outcomes ("Yes" vs "No") and test results ("Pass" vs "Fail"). This estimation procedure is available in JMPIN.

• REGRESSION WITH PANEL DATA. These notes illustrate the special problem of data involving multiple sample units measured at consecutive points in time. Here the problem of residual autocorrelation is more complex that in the AUTOCORRELATION notes above.

• MULTIPLE CATEGORICAL VARIABLES. This note attempts to clarify the interpretation of the beta coefficients obtained for categorical variables when more than one categorical explanatory variable is used in the regression.  

• COMPARING RELATED REGRESSIONS. This note develops a procedure for statistically comparing coefficient estimates in two related regressions.  The method is illustrated in terms of a two-city comparison of factors affecting housing prices. 

•  MULTIPLE REGRESSION (k = 2). This note develops least-squares estimation for the case of 2 regressors, and uses these results to illustrate the main properties of both Partial Regression Plots and Partial Residual Plots.

• MULTIPLE REGRESSION PROCEDURE. These notes outline a suggested procedure for performing multiple regression with many potential explanatory variables. This procedure is not meant to be a rigid recipe for regression. Rather, it is meant to illustrate the types of considerations which might be relevant in any given situation.

1. PROJECT DESCRIPTION

During the first few weeks of class, you should choose a partner to work with. Projects are expected to involve teams of two individuals. Individual projects are permitted. Teams of three are also permitted, but not encouraged -- and are expected to do more work. 

Each team is expected to undertake a case study involving a statistical analysis of some data set.  The only substantive requirement is that your analysis should focus on multiple regression.  This analysis should demonstrate a sound statistical knowledge of regression (including goodness of fit and significance tests of coefficients).  The report is to be typed double-spaced and is expected to be on the order of 15 to 20 pages in length (this is not a rigid requirement).  The first page should contain an introduction which (i) motivates the problem, (ii) states all of the main assumptions [without mathematics], and (iii) briefly summarizes your findings. The main body of the report should contain a detailed development of your statistical analysis, including a mathematical formulation of both the problem studied and the analytical methods employed.  Use plots and graphs wherever possible to illustrate your results (preferably in JMP). But be sure to back these up with appropriate discussion and analyses. [Do not include graphs or tables that are not discussed in the text.] All source material (including software packages used) should be cited explicitly. The last page should summarize your findings and conclusions in detail. Finally, be sure to include page numbers in your report. (I write comments on every project, and am very unhappy when I have no page numbers to refer to!).

Along with the hard copy of your project, you must send me an email attachment (preferably on the same day you turn in your project) including the following items:

• A JMP file of your data ( with columns labeled). This data should include all variables used in your analysis, and any transformed variables that appear in the regressions (so that I can easily repeat these regressions myself). If the variables in each column are not self explanatory, include a short description in the NOTES section of each column information window.

• A WORD file of  your complete written project (including all tables and figures) . If your project is written in some format other than Microsoft Word, then please submit it as an Acrobat (.pdf) file.

There are no constraints on the subject of your case study.  You might start by looking through the set of  projects that are included in this web page.  (These projects are presented in their original form -- including possible errors. So don't assume that everything in them is correct. They  intended mainly to suggest possible topic areas and data sources. )  With respect to data, it is preferable to use real data from an experiment or survey that you or someone else has performed. For example, sports fans may wish to consider published data on their favorite players or teams.  (A variety of interesting data sources can also be found by ‘web surfing’.)  In any case, you must clearly specify the source of your data.

Students often find it difficult to obtain the data sets they want to study. So it is advisable to start looking as soon as possible. There is a list of web sites given below where you can start to search for existing data.

The final grade will be based on several factors: the appropriateness and sophistication of the analytical methods employed, the correctness of the analysis carried out, the logic and perceptiveness of the conclusions drawn, and the overall clarity of the presentation.

2. DATA FOR REGRESSIONS

One key point to remember in gathering data is that your data must involve properties of well-defined sampling units . For example, to study the relation between income and years of education, it would be ideal to have data on individual workers (sample units) with both the income and years of education for each individual. This is usually not possible. But often such data exists at, say, the state level. So here you could do a regression by taking states as sampling units and regressing per capita income of states against average years of education of state residents.

A particularly vexing problem here is the preponderance of data in the form of summary tables. For example, if you only have a summary table listing average income for various education categories in the US, it is very difficult to run a regression on this data --because there is no clear sampling unit. Since most data you find will be in the form of summary tables, it is very difficult to use such data in regressions (without a host of additional assumptions). However, if you were able to find summary tables for each of a number of countries, then you could use 'country' as a meaningful sample unit in regression, and examine the relation between education and income across countries.

So in short, you should try to find data for which the sampling unit is well defined, and hopefully for which there are sufficiently many samples to allow an interesting regression. A common rule-of-thumb here is to have at least 10 samples for every beta parameter estimated. So a simple regression (two beta parameters) should ideally have at least 20 samples. This does not mean that you shouldn't consider a wide range of possible explanatory variables. It only means that your final regression should have enough samples to allow reasonable estimation of each parameter. (See the notes on Stepwise Regression above for further discussion.)
 

3. SELECTED WEB SITES FOR DATA SOURCES
 

PENN CAMPUS RESOURCES

http://data.library.upenn.edu/index.html
http://www.cml.upenn.edu/
 

GENERAL DATASET COLLECTIONS

http://lib.stat.cmu.edu/datasets/
http://www.stat.ucla.edu/cases/
https://www.cia.gov/library/publications/the-world-factbook/index.html/
http://genderstats.worldbank.org   
http://www.icpsr.umich.edu/
http://web.lexis-nexis.com/statuniv/
http://www.lib.umich.edu/libhome/Documents.center/stats.html
 

CENSUS DATA

http://www.census.gov
http://www.census.gov/DES/www/welcome.html
http://dataferrett.census.gov/TheDataWeb/
http://www.census.gov/apsd/www/statbrief/
 

COMMODITIES

http://www.carprices.com/
http://www.consumerreports.org/
http://www.diamonds.com/
http://www.diamondfinder.com/
 

CRIME

http://www.albany.edu/sourcebook/
http://www.ojp.usdoj.gov/bjs/dtdata.htm#crime
http://bjsdata.ojp.usdoj.gov/dataonline/
http://www.fbi.gov/ucr/ucr.htm#nibrs
 

ENVIRONMENTAL

http://www.epa.gov/enviro/html/ef_overview.html
http://www.eia.doe.gov/
http://www.pasda.psu.edu/
 

INTERNATIONAL DATA

http://www.geographic.org/
http://www.un.org/databases/
http://unstats.un.org/unsd/default.htm
http://www.worldbank.org/data/

MEDICAL

http://www.cdc.gov/nchs/fastats/
http://www.cdc.gov/scientific.htm
http://www.nci.nih.gov/public/factbk95/index.htm
http://www.who.ch/hst/hsp/a/country.htm
http://www.lungusa.org/
http://seer.cancer.gov/

http://www.cdc.gov/brfss/smart/2002/summary_matrix_02.htm
 

NATIONAL DATA

http://www.bls.gov
http://www.stat-usa.gov/econtest.nsf
 

SPORTS

http://www.sportstalk.com
http://sportsillustrated.cnn.com/
http://www.baseballprospectus.com/
http://www.hockeyguide.com/
http://www.nhl.com
http://www.pgatour.com
 

SURVEYS AND POLLS

http://www.nua.ie/surveys/
http://www.gallup.com/poll/releases/
http://www.cnn.com/ALLPOLITICS/
 

TRANSPORTATION

http://www.bts.gov/ntda/
http://www.apta.com/research/stats/
http://www.njtide.org/links/index.html
http://www.nhtsa.dot.gov/people/ncsa/
http://www.ntsb.gov/Aviation/Stats.htm
http://www.nhtsa.dot.gov/people/ncsa/fars.html
 

(return to contents)

EXAMPLE PROJECTS

  The following  projects have been selected as examples of the level and quality of analysis that I am looking for. However, please be aware that none

  these projects is free of errors (i.e., none have been "corrected"). So please don't think that because something appears in an example project that it is

  automatically "OK". If you are not sure about something, please ask me.

• Project 1 : Airline Flight Delays
  

• Project 2 : Life Expectancies across Countries  

• Project 3 : Factors influencing the Dow Index  

• Project 4 : Economic Analysis of Stock Prices

• Project 5 : Factors influencing Student GPAs

• Project 6 : Graduate School Rankings

• Project 7 : Analysis of Baseball Pitching

• Project 8 : Factors influencing SAT Scores

• Project 9 : Analysis of Changing Religious Perspectives

• Project 10 : Analysis of Auto Crash Fatalaties

• Project 11 : CO2 Emmissions per Capita

• Introductory Lecture

• Lecture 1
  

• Lecture 2
  

• Lecture 3
  

• Lecture 4
  

• Lecture 5

• Lecture 6

• Lecture 7
  

• Lecture 8
  

• Lecture 9
  

• Lecture 10
  

• Lecture 11
  

• Lecture 12
  

• Lecture 13
  

• Lecture 14
  

• Lecture 15
  

• Lecture 16
  

• Lecture 17
  

• Lecture 18

• Lecture 19

• Lecture 20

• Lecture 21

• Lecture 22

• Lecture 23

• Lecture 24