STAT 5303: Designing Experiments
Spring Semester 2013

Basic Information

Official University policies on many issues are available here.

Finals Week Office Hours

Monday (5/13):		11:00 a.m. to 2:00 p.m.
Tuesday (5/14):		10:30 a.m. to 12:30 p.m.
Wednesday (5/15):		11:00 a.m. to 12:00 p.m.
Thursday (5/16):		10:30 a.m. to 12:00 p.m
Friday (5/17):		10:00 a.m. to 11:30 a.m
Saturday (5/18):		approximately 2:30 p.m. to 3:30 p.m. in Ford 367 and 3:40 p.m. to 4:00 p.m. in Rapson 45

Final Exam is 4:00 p.m. to 6:00 p.m. on Saturday May 18 in Rapson Hall Room 45.

The exam is open book and open notes, but only physical books or notes are permitted. No electronic books or notes are allowed.

Bring your printed project report to hand in with your final exam.

Textbook

Some good news and some bad news:

The bad news is that the textbook is out of print and hence not readily available from bookstores, although you may find a used copy somewhere.
The good news is that the author (Prof. Oehlert) has generously made the book available as a no-cost download. Go to his page for links to the PDF file for the book itself and datafiles for examples and exercises from that book.

Homework

Assignment #12 (due Thursday, May 9):

P5.2
P5.3

Assignment #11 (due Thursday, May 2):

E17.1
P17.1
P17.4
E18.1
E18.2
P18.2

Assignment #10 (due Thursday, April 25):

E15.2
E15.3
P15.1
P15.6
P16.1
P16.4
P16.8
P16.9

Assignment #9 (due Thursday, April 11):

P13.1
P13.4
P13.12
E10.3
P10.5
E14.2
E14.4
P14.3
P14.5
P14.6 Just do parts b, c, and d, but omit part a

Assignment #8 (due Thursday, March 28):

P11.2
E12.1 and E12.2: Just do the first of the "standard five questions"---draw the Hasse diagrams. Show model terms, indicating which are fixed and which are random, and which are nested and which are crossed; you don't need to find test denominators or expected mean squares.
E12.4
P12.1
P12.2

Assignment #7 (due Thursday, March 14):

E11.1 find point estimates, but do not find confidence intervals,
E11.2 (a) and (b), but for part (b) just find point estimates, but do not find confidence intervals,
E11.5 find point estimates for both "between" and "within" variance components, but don't find confidence intervals.

Assignment #6 (due Thursday, March 7):

E10.4 and P10.7.

Assignment #5 (due Thursday, February 28):

E8.1, P8.2, P8.3, and P8.7.

Assignment #4 (due Thursday, February 21):

E7.2, E7.3, and P7.2.

Assignment #3 (due Thursday, February 14):

E4.2, E4.4, P4.2, P4.3, E6.2, E6.3, E6.5, and P6.1.

Assignment #2 (due Tuesday, February 5):

E3.3, E3.4, and P3.2.

Assignment #1 (due Tuesday, January 26):

E2.1, E2.5, and P3.1 (For this problem don't do the analysis---just identify the experimental units, treatments, and measurement units).

Computing

We will use R. You may download R for Macintosh, Windows, and Linux from the "CRAN" link on the R-Project home page.

Basic R information

Here is information about R and here is more information about R.

R package Stat5303

We will use Prof. Oehlert's R package called Stat5303, which adds some extra commands that we will need.

The Stat5303 package depends on certain other R packages:
car (version 2.0 or later), effects, FrF2, mvtnorm, perm, RLRsim, and tseries.
You should install these and all the packages they depend on.

Stat5303 R package (0.7-1) for Mac

Stat5303 R package (0.7-1) for Windows

Download the package and save the file into a place where R can find it (e.g., your home directory or the desktop). Start R, set the working directory to that location (e.g., use setwd(), and then use

 install.packages("Stat5303_0.7-1.XXX",repos=NULL)

Data

Because typing in data is tiresome and can be a source of errors, the data from the homework problems and examples (from the text) are available as files that R can read. There are a couple of ways to get data into R.

Individual data files

Data sets for examples are named exmpl6.3 for example 3 from chapter 6.
Data sets for exercises are named ex3.1 for chapter 3, exercise 1, and
data sets for problems are named pr3.3 for chapter 3, problem 3.

Here are dataset files for R that you can download. So, for example---provided your machine is connected to the Internet---from within R you can use

 resindata <- read.table("http://www.stat.umn.edu/~corbett/classes/5303/RDataFiles/exmpl3.2",header=TRUE) 

When you are not connected to the Internet, you could use

 resindata <- read.table(file="exmpl3.2",header=TRUE) 

One package

Russ Lenth at the University of Iowa has also provided two R packages that include the data sets from the book.
For Macintosh/Linux oehlert_1.02.tar.gz
For Windows oehlert_1.02.zip
Download the package and save the file into a place where R can find it (e.g., your home directory or the desktop).
Start R, set the working directory to that location (e.g., use setwd(), and then use

 install.packages("oehlert_1.02.XXX",repos=NULL)

 library(oehlert)

 pr17.4 

Note that the data set names, variable names, and variable codings in the oehlert data package and the direct-web-accessible data files may not be the same.

Computer handouts

These handouts (written by Prof. Oehlert) demonstrate using R to work through examples from the text. As you work through them, remember you want to see both (1) how to get the computer to perform certain tasks for you, and (2) why you want the computer to do those tasks. In other words, you should learn something about the underlying statistical concepts in addition to learning how to use the computer.

• Here is a handout with examples showing how to use R to perform randomization tests.
• This R handout uses the resin lifetime data set (Example 3.2 in the text) to demonstrate analysis of a completely randomized design, and also discusses dose-response polynomial modeling.
• Our analysis procedures depend on certain assumptions. Some violations of those assumptions are not too serious, but others can cause problems. This handout discusses how you can detect problems and what you can do about them.
• This handout shows how to use R to find power and sample size for the ANOVA F test.
• This handout shows how to use R to analyze data from experiments where the treatments are combinations of factors at various levels.
• R can also deal with two-series factorials.
• Here is a random effects handout. The usual R analysis for these models uses lmer() (pronounced "elmer") which is part of the "lme4" package. That is not part of the Stat5303 package, so if you want to use lmer() you will need to install the lme4 package. We will also use some functions from the Stat5303 package to perform additional analyses after running lmer().
  Here are data files for the calves and resistors examples discussed in the random effects handout.
  
• Here is information about nesting and mixed effects in R and here is a data file for the particle counts example discussed in this handout.
  
• Here is information about restricted mixed models in R and here is background information about REML.
• Here is an example of a randomized complete block design in R.
• We can analyze Latin square designs in R.
  This handout also shows how to deal with an outlier in the data by using a dummy variable to effectively remove that observation from the dataset. However, doing that also destroys balance, so we need to consider the order in which terms are entered in the model in R. Examples are presented in the handout.
• Here is information on unbalanced factorial designs.
• Of course, there is a handout for balanced incomplete block designs.
• Naturally, there is a handout for confounding in factorial designs.
• Yes, R knows about split-plot and split-split-plot designs.
• R knows about models with covariates.
• R knows about fractional factorial designs.
• Here is some information about how you can use Tukey's method to deal with interactions.
  
• Here is a worked example from the book showing how to use R to look for Tukey's one-degree-of-freedom-transformable-nonadditivity.
• Here is a handout on response surface designs and here is more material on response surface designs.
  Here is some R code for the cake-baking example (19.1) in the text.

Project

• There is more information in the text---see chapter 20.
• The project involves a designed experiment, not an observational study or a survey.
• You need to clearly identify experimental units and treatments and, if appropriate, blocking factors and measurement units.
• Describe how you use randomization in your experiment.
• You should analyze the data using methods we have covered in this course, such as ANOVA for a continuous response.
• Remember that we've studied comparative experiments, so you probably want to examine differences between treatments.
• No human subjects, nor anything else that would require review board approval.
• You need to propose a question, plan an experiment to answer that question, perform that experiment, collect and analyze the data, and write a final report covering all of that.
• Remember to include in both the proposal and the final report enough background information so someone who is not trained in your area of expertise can understand what you're trying to do and why someone would want to do it.
• This has to be a new experiment; you cannot just take some old data from work you did before and reanalyze that.
• This is to be an actual experiment, and not just a computer simulation.
• The proposal is not like a contract, but it needs to be specific enough so I can tell tell if what you're planning fits within these guidelines.

Comments? Questions? Send me an e-mail note: corbett@stat.umn.edu