IEOR 5550 "Design and Analysis of Experiments" Syllabus - Summer Session II 1996

Objectives:

The student will be able to: outline the basic steps of an industrial experiment; solve basic statistical word problems; design experiments using the concepts of randomization and blocking; perform t-test and ANOVA analysis; contrast confidence intervals with significance levels; perform diagnostics and suggest solutions; outline the basic assumptions in model building; design and analyze two level factorial designs; contrast Taguchi's methods with classical methods; design an experiment to analyze robustness; perform the steps of response surface methodology; demonstrate ability to recognize examples of poor statistical statements and graphics.

Prerequisites:

ME 3900 or equivalent; e-mail and world-wide web access

Texts:

Box, G.E.P., Hunter, W.G., Hunter, J.S., Statistics for Experimenters, New York, Wiley 1978.
Huff, D.W., How to Lie with Statistics, New York, Norton, 1982 (older editions okay).

Class Days, Time & Location:

MWThF, 10:25am-1:05pm. ME 202

Instructor, Office and Office Hours:

Michael W. Usrey, ME 111, (612) 626-8391, 624-1398 FAX, musrey@me.umn.edu, Office Hours TBD and by appointment.

Teaching Assistant:

Dinesh Wadhwani, Management and Economics 332, 624-7010, 624-1316 FAX, wadhwani@msi.umn.edu.

Course Schedule:

            BHH    Assignment
  Date    Chapter     Due      Topic
  ----    -------  ----------  -----
07/18/96     1                 Science and Statistics; Statistics on Internet
07/19/96     2                 Comparing Two Treatment Means

07/22/96     3                 Random Sampling and Independence
07/24/96    4-5        1       Randomization and Blocking; Significance Tests and Confidence Intervals
07/26/96     6                 Experiments to Compare k Treatment Means

07/29/96     7                 Randomized Blocks and Two-Way Factorial Designs
07/31/96    9,10       2       Empirical Modeling; Factorial Designs at Two Levels
08/02/96                       MIDTERM

08/05/96   11-12               Full and Fractional Factorial Designs at Two Levels
08/07/96    13                 More Applications of Fractional Factorial Designs; Guest Speaker - Perry Parendo
08/09/96    14         3       Least Squares Regression Analysis

08/12/96    15                 Response Surface Methods
08/14/96           Ind. Study  Independent Study Presentations; Surveys; Review
08/16/96           4; Project  FINAL

Grading:

40%    Homework
10%    Independent Study
20%    Project
10%    Midterm Exam
20%    Final Exam

Classes:

Lecture notes are under construction at:
  • http://www.me.umn.edu/home/musrey/5550/96ssii.html (this document)

    The students are expected to: read appropriate material before class; participate in all class discussions; locate and read other external material; be open minded; challenge assumptions; ask questions; show respect for classmates and instructor.

    Homework:

    Numerical answers should always be followed with appropriate discussion and interpretation of the numbers. Individual assignments are to be handed in; you may confer with others about the assignments. The instructor will compile an e-mail roster to facilitate teamwork. Homework is due at the beginning of class on the due date listed on the Course Schedule. Late homework will be penalized 10% for each school day it is late. No homework will be accepted beyond 3 school days after the due date.

    Independent Study:

    Students should collect examples of each of the 9 ways to lie with statistics. Examples can be collected from books, journals, newspapers, magazines, on-line services, and other sources. For each of the examples, students will submit: a copy of the source material; a paragraph description of how the example qualifies as a particular type of lie; 1-3 rules of thumb for the proper way to communicate that type of statistical information; and a specific recommendation for the source author to improve the presentation of their data. Students will select the "best" of their nine examples, create a transparency, and briefly present their example to the class on the date identified on the Course Schedule. All nine examples will be turned in at the beginning of class on that date.

    Project:

    Students will plan, execute and report upon an experimental investigation of their own design. I strongly recommend that the student projects follow the structure of:
  • Coleman and Montgomery, "A Systematic Approach to Planning for a Designed Industrial Experiment," Technometrics, Feb 1993, pp 1-27, Rochester: NY, American Society for Quality Control,
    for both the planning of your project and the structure of your report. This paper is on reserve at Walter Library.

    The main objective of the project is for you to show that you understand the concepts and techniques taught to you throughout the course, and at the same time allow you to investigate something of interest to you. The project write-up is due at the beginning of class on the date listed on the Course Schedule; no late projects will be accepted.

    The subject/hypotheses that you are to investigate is entirely up to you. The only stipulation is that it should be a controlled experiment, i.e. you may not simply analyze previously compiled data. In general your project should consist of a single-factor experiment, with multiple treatments. This corresponds to the designs discussed in BHH Chapters 1-7, and the course notes pages 1-71. You will be analyzing your data via either an independent t-test, a paired t-test, or ANOVA.

    You may work alone, or with at most 1 other person. It would be highly recommended that you give me, at some unspecified date before experimentation starts, a short written proposal on your plan, although this is not required. After seeing such a proposal I can give you verbal feedback concerning what might be the potential pitfalls/obstacles that you may encounter. Such a proposal would save the embarrassing situation of being at the end of the quarter with no hope of finishing the experiment.

    Grading will be both objective, on such elements as technique, assumptions, etc., and subjective, on such elements as organization, planning, thoroughness, imagination, etc. The following items are on the check-list I will use for grading. In general, each of these items are graded as "-" (inadequate), "0" (adequate), or "+" (exceeded requirements). After tallying these, a numerical grade is assigned. As a benchmark, a paper with all "0"'s would score an 80:

  • Problem definition
  • Selection of variables
  • Blocking and other strategies for non-design variables
  • Choice of measurement(s)
  • Measurement system
  • Assumptions
  • Stated hypothesis
  • Correct analysis
  • Residual analysis
  • Conclusions
  • Tie to theory

    In summary, the problem should be explained in detail, all assumptions stated, relevant supporting theory, all experimental strategies explained, all techniques and data analysis shown and explained (i.e. don't just show results), conclusions given, and recommendations for further study given. If you have a particular question whether XYZ should be in the report, ask. Otherwise, assume that the reader knows nothing of the problem discussed and needs to know all details. Description of the theory of the statistical tests used is not needed, unless something novel is attempted. An example report which was graded "excellent" is given in the class notes.

    Some previous IEOR 5550 experiments...

  • medical device strength
  • kite flight duration
  • particle board construction
  • popcorn yield
  • brownie taste
  • picking up nickels vs. pennies
  • biased dice roller
  • survey
  • auto resale value
  • play-dough strength
  • medical data base
  • drive time to work
  • service time at restaurant
  • response to mail request
  • comparison of methods to test chip speed
  • effects of floor burnishing pads on gloss
  • heating liquids in a microwave
  • analysis of surgeon gown liquid repellency
  • postal experiment
  • effect of pan materials on boiling times
  • analysis of plasma spraying process
  • effects of stress and strain on vulcanized rubber
  • distribution of pennies in circulation
  • effectiveness of antacids
  • how many green M&Ms are there?
  • shuffling a deck of cards
  • Kelloggs vs Post in raisin bran war
  • fastest roller skates
  • absorption by commercial paper towels

    Exams:

    Both tests will be administered during class time. Both tests will be open reference.

    Software:

    Do NOT turn in raw computer output! You should answer the question as if you were doing the work by hand, showing which quantities are sequentially calculated in order to arrive at the final answer, with appropriate discussion of the final answer.

    MULTREG is a public-domain program which does basic statistics, t-tests, ANOVA, and regression. The regression component can be "tricked" into analyzing factorial designs. The documentation is available at the east bank bookstore and you can down load an MS-DOS copy of the software from: ftp://merlin.add.uni-frankfurt.de/MedArchiv/multreg.zip

    DESIGN-EASE is a commercial package which we have a site license for; it can be found in the IT Computer Labs in room EECS 3-170. Several copies of the manual should also be there. DESIGN-EASE does ANOVA, full and fractional factorial designs.

    ------------------------------------------------------------------
    The views and opinions expressed in this page are strictly those of the page author. The contents of this page have not been reviewed or approved by the University of Minnesota.
    Michael W. Usrey
    musrey@me.umn.edu
    http://www.me.umn.edu/home/musrey/