good to keep notes and send the client (and your mentor!) meeting
summaries

PHASES OF CONSULTING

1. Establish rapport

2. Identifying the research problem

   Tell me about your study

   learn about the data

      * What does one (single, complete) observation look like?
      
      * If you measure this again, will you get the same answer?
      
      * If I can (solve this research problem), then what would you do
        next?

   Clients often have a very specific idea of what their research
   problem is (i.e. what they want), but they may not be able to
   express cleanly.

   Part of your job is to figure out what they want, without changing
   it into something more familiar or easier to you.

3. Setting goals

   in my experience goals evolve over time, and you will have to
   circle back to the goals several times during a project

   there is a tradeoff between

   * being sufficiently flexible about this, vs.

   * not so flexible that you lose control of the feasible scope of
     the project

4. Agreeing on a division of responsibility

   the advice here is more germaine to a consulting center

     (some projects are for a fee, others not, but treat every project
     as if for a fee!)

   generally speaking

     - set up regular meeting schedule that works for the client
     - similarly for your faculty mentor
     - client is responsible for
       o getting you non-statistical information about the project
       o procuring data (unless the project is explicitly for you to
         do that)
     - you are responsible for
       o learning about the problem
       o learning what you don't already know about stat/datascience
       o offering advice and analysis (quant  qual) that is meaningful
         to the client
	 
5. Reviewing what has occured / summing up
 
   at the end of every meeting; and followup with a written summary to
   the client!


CONSULTING ROLES

* Helper Role
  - lower-level "do'er" role

* Leader Role
  - client wants you to lead the whole research project (not recommended!)

* Data-blesser Role (or conclusion blesser)

  How does this table look?

  I want to conclude X.  Does my analysis allow me to do that?  Or --
  what analysis should I do to reach this conclusion?

* Collaborator Role

  This is the most fun.  you pool talents and different knowledge
  bases/experiences with the client's and jointly work out approach,
  analysis, interpretations, etc.

* Teaching Role

  You are always a teacher, whether or not you think you are!

  You *have* to teach the client about what you're doing - this is
  part of explaining and reporting

  But you will also teach about what statisticians consider important,
  about the phases of consulting, about what it's like to work with
  someone from Carnegie Mellon, etc.

HUMAN SIDE OF CONSULTING

* Negotiating for a desired role

* Some negotiating principles

  - Never attack a client's position clients have multiple interests

  - use common (project-related) interests to form strong rapport;
    address divergent interests later

  - clients are more likely to accept a role for yoy if it seems
    "right" -- fair, reasonable, honorable

  - suggestions with precedent (objective standard and persuasive
    argument) are easier for clients to accept

  - if you are on the short end of a power differential, try
    negotiating on the basis of principle

    o you have the advantage here of 3 backups: me, your faculty
      mentor, and Jamie.  So power differentials should be easier to
      manage in 726 projects

* Influencing the direction of a consultatiobn

  - good tools (positive)
    o clarification
    o approval
    o general leads

  - good tools (changing the direction of client's thinking)
    o information gathering
    o rejection (paired with a positive alternative)
    o introduction of new information

* Consulting on a wide range of problems

  - always be a student of the subject area
  - there are no dumb questions

  You many also be consulting using a wide variety of quantitative
  tools

  - always be a student of statsitics and data science
  - there are no dumb questions

(the first time a question is asked it is not dumb, by definition.  If
you repeatedly ask the same questions, that *is* dumb)

(don't be afraid to ask for more time to think about [or do research on]
really tough problems)

* working with clients with varied statistical backgrounds

  - work at the client's level of expertise; don't try to dazzle

  - you may end up teaching the client something new, to taise their
    level of expertise

  - clients sometimes do not know what they do not know.  Be gentle
    and graceful in helping them to understand better

  - you will occasionally run into a client who wants you to do
    procedure X against your best judgement (e.g. use support vector
    machines when a simple logistic regression would suffice)

    o this requires grace and flexibiity on your part, and maybe some
      of the negotiating tools we mentioned above!

    o if it is not so time consuming that it ruins the project, a good
      approach can be to try the client's suggestion and compare it
      with simpler methods -- either you will see that svm's are
      helpful, or the client will see that logistic regression is just
      fine.

* cleaning up the mess

  - the data or the constraints are so inappropriate that the client's
    quesions cannot be answered

  - patience and teaching are required

  - also refocus the project on (planning for) better data collection
    and variable storage

* Surviving (in academia)

  - some statistics departments undervalue consulting
  - some consulting units undervalue statistical sophistcation

  - you have to find / negotiate / build the happy medium

  - in general good strategies are;
    o mounting an agressive ongoing education campaign
    o energetically pursuing collaborative roles
    o be assertive regarding joint authorship