Re: ???Statistical Measurement; A Related Question (Metrics)

[sandy_perle@email.fpl.com]

For starters, the most important factors to be considered include:

1. What are the main "outcome indicators" and,

2. What are the process indicators that inform you whether or not you have any 
reasonable chance to meet the "valid expectations" of the outcome indicator.

For instance, an outcome indicator could be an overall manrem goal, or target. 
Goals and targets are NOT interchangeable. The goal is often more liberal in 
nature. The target is an interim step to achieve the goal.

Anyway, for rad protection an outcome indicator could be a manrem target 
reduction program. One process indicator could be manrem/man-hr worked. Process 
indicators could be setup for various workgroups, different shift workers and 
different types of areas in the facility.

Bottom line, the indicators need to be measurable and be specific. They should 
be few in number, and they should be evaluated periodically to ensure that 
progress is being made in meeting the outcome indicator. Countermeasures should 
be established when the target goes awry, as they often do. that means 
contingency plans should be developed prior to a problem being identified. Be 
proactive. Sort of like a FMEA (failure modes and effect analysis) or an EMEA 
(error modes and effect analysis). Use FTA (fault tree analysis) to identify 
potential failures in the system, build a process and then collect data, lots of
data, and analyze .. in detail.

Anyway, just a few of my thoughts. All part of TQM .. Total Quality Management.

Sandy Perle
Supervisor Health Physics
Florida Power and Light Company

(407) 694-4219 office
(407) 694-3706 fax

sandy_perle@email.fpl.com

______________________________ Reply Separator _________________________________
Subject: ???Statistical Measurement; A Related Question (Metrics)
Author:  radsafe@romulus.ehs.uiuc.edu at Internet-Mail
Date:    11/8/95 7:44 PM


Now, for a branch in the Statistical Measurement Road. . . .Let's talk METRICS!
     
If statistical measurement requires that I "sample" my program, what categories 
of data 1) are manageable and 2) actually reflect the overall performance of a 
program.
     
This raises the ugly spectre of HP ***metrics***.  I have seen a number of 
programs built to "measure" the overall effectiveness of a program, but they all
     
seemed to degenerate into an excuse to buy more computers and color printers so 
we could line the walls of the HP break room.  Added to that, some of the 
numbers 
were a real pain to compile and maintain.
     
My dim understanding of statistical management says that the measurement should 
be:
     
    1)  easily obtained
     
    2)  easily tabulated
     
    3)  at a glance, reflect the "true" performance of a process or system
     
As RP tends to deal with "behavior" parameters and boundaries of operation (we 
set rules, but don't do the work) rather than repeatable assembly line 
production 
processes, what statistic would one use to measure the performance of a program.
     
So, I'd like to see some statistics proposed that would "test" for the 
performance of the following ("typical") facets of an RP program that meet the 
above requirements.  I'd prefer to see ONE measurement that tells the story; not
     
5 or 10 separate measurements that I have to spend a couple of hours 
interpreting 
(my feeling is that a ratio of stats may be the ticket here, but I'm 
speculating):
     
    1)  How effective is your external dose control?
     
    2)  How effective is your internal dose control?
     
    3)  How obedient are your users/workers to RP requirements?
     
    4)  How responsive is your program to off-normal events?
     
    5)  How good is your program at predicting off-normal events?
     
Thanks,
     
Jim Barnes, CHP
RSO
Rockwell Aerospace; Rocketdyne Division