In this two-page .pdf file, see how puzzling patterns in student test scores become abundantly clear with the use of data visualization.

You’ve designed a program or intervention to try to influence outcomes for a large number of people.  Whom should you target:  those closest to deciding the desirable way, to nudge them “over the line”?  Those farthest away?  Or those on the fence?  Which strategy will have the greatest impact?
 
An analyst and VP at Fidelity Investments, Victor S. Y. Lo, has devised and evaluated uplift models  for topics including election email campaigns, personalized medicine, credit card marketing, and supply-chain modeling.  I thought Victor might have an answer to the question, "at what place along the likelihood spectrum is a person most influenceable by an intervention?”  That is, for greatest “lift,” is it the high-likelihood people who should be targeted, or the low, or those on the fence? 
 
His ultimate answer was that it’s context-dependent.  No research has shown that it's generally a certain sort of person for whom an intervention will have the most effect.  It could be one answer for getting young second-generation Americans in the Northeast to apply for credit cards, and another answer for getting middle-aged California diabetics to avoid hospital readmission.  In each context the uplift research will point your way forward.  This lesson has been borne out in my own work over the years in higher education and health care.
 
The takeaway here is to distrust any blanket statement claiming that you should always give the greatest focus to those people at a certain place along the probability spectrum.