Team C

100/100 for I3 and I4.

Emily Boncek, Christopher Chang, Kelly Chang, and Stephanie Sindler

General comments:
----------------

You are good to go on your project.  I look forward to seeing II.5.

Specific comments:
-----------------

* I think it will be important to keep track of which subject or
  department each course is in, since there ought to be differences
  across subject in how important attendance is for learning.

* You have said that the sampling frame "is the list of faculty whose
  email addresses are in the Carnegie Mellon directory or listed on
  departmental websites."  You will want to make sure you cross out
  any duplicates before constructing an SRS w/o replacement from this
  frame.

* You are concerned about sampling proportional to the size of each
  college.  This is very much like stratified sampling which we will
  discuss soon in class.  An alternative to proportionate sampling is
  the use of post-stratification weights (which we will also discuss).

* Under "mode of data collection" you state, "Email online survey,
  face to face is also possible. Email is most convenient, if
  professors prefer we can accommodate for face-to-face interviews."

  I am concerned that faculty will ignore email requests.  I think FtF
  will have better response rate.  Perhaps you should plan to do a FtF
  followup as a last resort for non-responders, or something like
  that.

* In I.4 I would describe your sampling plan like this: You are
    * stratifying ("blocking") by college, 
    * sampling departments as clusters
    * stratifying by attendance requirement and level (1xx vs 2xx etc)
      to sample within classes
 
  This is actually a potentially very good design (and one that is
  used in some national surveys: e.g. stratify by state, sample
  congressional districts as clusters, and then do some sort of
  stratified sample within clusters). 

  However, you may wish to consider a simpler (and simpler to analyze)
  design, such as 

  (a) stratify by college; and then take an SRS, w/o replacement, of
      classes within each college.  Take as your faculty respondent
      whoever is teaching the classes in your sample.

  (b) Stratify by college; then take an SRS, w/o replacement of
      *departments*, then take *every* instructor who is teaching an 
      undergraduate course in that department for the current
      semester.  [this samples departments as clusters within college
       strata, btu simplifies the sampling within departments]
  
  These are not the only schemes possible.  The main idea is keep to one
  level of stratifying and/or one level of clustering, and take all
  respondents within a cluster.  

  You might find the list of classes being taught this spring at
 
      https://enr-apps.as.cmu.edu/open/SOC/web/images/documents.htm

  useful for this purpose.  [or were you thinking of doing fall 2010
  classes, so there are final grade distributions?]

  For the purpose of sample size calculation (for II.5) pretend that
  you will do a simple random sample of classrooms (even though your
  design will be more complicated than that!).

* How will you "measure" representativeness of the target population?
  Think of a few descriptive variables that you can get the values of
  for the population, and collect data on these variables for your
  sample.