Wednesday, June 4, 2014

The Away Mission Approach to Validating a New Model

I found this gem of writing in my notes today...I'd like to share it.  If you are ever tasked to look into or validate a new model, this is a concept approach I created for that exercise.  And yes, I am a Star Trek fan.  You should have guessed that with a name like 'Kirk'.  Enjoy!


The Away Mission approach to Validating a New Model

Your mission:  To seek out and understand a new model.  To boldly go where you have not gone before.

Enter the planet’s orbit and do a scan…

A.       Determine what the planet (the model) is like and if it in habitable

a.       What is the name of the model?

b.      Is it similar to something I am familiar with?

c.       Search the databanks (any databases you can access) to find research on it

                                                               i.      If possible, use academic sources (over say Wikipedia or blogs--except this one of course!)

                                                             ii.      Send communications to analysts that may have worked with those models—In your communiques, ask if they are familiar with them and ask if they would be willing to consult with you when you are on the planet’s surface (or doing more detailed analysis of the model)

                                                            iii.      In your study, you may come across new words and methodologies you are not familiar with—takes note of these and take advantage of the time to learn

1.       If its helpful, create a flow chart as you study the model, for example…


d.      Know what you are dealing with prior to beaming to the surface and pack accordingly (make sure you pack a phaser!)

                                                               i.      What assumptions does the model have?

                                                             ii.      Are there conditions that are built in?

                                                            iii.      In what situations is the model specifically used for?

                                                           iv.      Obtain any formulas that are associated with the model (i.e. structures seen through the database research)

After beaming down…

B.      Break out your tricorder and do more thorough scans & tests

a.       Look at how the model was used in the context of what you are validating (and compare to how it is used in the contexts you studied)—is it used properly in your context?

b.      Were the assumptions of the model taken into account?

                                                               i.      Here, it may be helpful to test data based on the assumption of the model.  For  example, if a technique is based on data being cointegrated, do a cointegrated test

c.       Are the conditions (if any) reasonable based on the context you are looking at?

d.      Analyze the dependent and independent variables

                                                               i.      Do they make sense to use in the context of your model (and its type)?

e.      Were there tests performed to find the best-fitting model (i.e. using AIC or BIC for example)?  If the best-fitting model was not used, what was the reasoning behind it?

f.        Were their sensitivity and back-tests performed on the model?  Do the results show any weakness and is it explained?  Do the sensitivity tests make sense given their respective scenarios? 

g.       What are the tests associated with the model?  Do the results say anything about the model’s weakness or strength?

Make a report to your commanding officer…

C.      This is the most important part, as your report will be viewed by people who may or may not know the model you studied

a.       Make sure you show that you took the time to understand the model and its context and can relay if it was used appropriately

                                                               i.      Highlight important components

                                                             ii.      Note any assumptions of the model and if they were properly checked

                                                            iii.      Present a summary review of the model strength results and note any strengths and weaknesses of the model—so they are clear to decision makers that use it

                                                           iv.      Note any strengths or weaknesses discovered when analyzing the back-tests and sensitivity checks

                                                             v.      Present results of any tests you ran (i.e. for assumptions)

                                                           vi.      Make a statement about the validity of the model…its structure, its use being appropriate (or not), and any comments about how the model can be improved in the future