I know we spent millions every year on ours ~ and a monthly sample selection would have over 1.5 million elements.
The really big bucks were spent on the sampling universe ~ we pre-identified the destination types we wanted to include and they were always in a constant flux AND there'd be changes in mail classifications, and every time that happened the whole shebang had to be reworked.
There aren't even private sector marketing firms with the kind of money and personnel to do a job like that.
Our planning organization had a predictive system that would estimate, based on sample reports from the Cost Ascertainment System what mail volume might be the next month.
I could take the same information and make the same prediction using an algorithm I"d developed.
One day I noticed that not once in 100 months/accounting periods had the planning organization ever correctly predicted whether or not mail volume would increase or decline.
Thought that fascinating since the statistical group stood on their heads to get the basic information for them to come up with that particular notation ~ mightily important to postal management I"ll tell you!
My algorithm, ALWAYS predicted whether or not volume in each successive period was going up, or going down!
Others have seen this before ~ but for those of you who haven't, I submitted my algorithm as one of those BENEFICIAL EMPLOYEE SUGGESTIONS.
They (top management) didn't accept my suggestion ~ didn't even send me a note rejecting it ~ but they relieved the guys responsible for running the competition ~ the prediction system that was ALWAYS wrong.
That system was, as I understand it, designed by the former Chief Postal Economist, Charles Guy ~ here's an article by him; >http://www.lexingtoninstitute.org/post-office-may-not-last-forever-an-interview-with-charles-guy?a=1&c=1139
I've probably never agreed with him on anything ~
Nice story... it’s unbelievable the faith some people place in numbers, and how easily management is wooed by numbers (even if they are never right).
Numbers are the best way to represent things empirically, yes, but they are not magical, and one should resist the temptation to be easily impressed by them. The more I read peer-reviewed papers, the more I find that they’re basically a load. The numbers are only as good as the person, methodology, and agenda used to represent them.
This isn’t to say that 2012 polls so far are inaccurate. We have little empirical reason to believe that they are. But you can’t just find a pattern, and say, this is a good model. Nate Silver decided that the “third year” of a president’s term was a good prediction for re-election. How much arbitrary can you get?