Transcript Document

Measurement and System
Thinking
E Rothman
Professor of Statistics
Director of CSCAR
January 29, 2009
Sense Perception ..
• There are many ideas regarding
management that are inconsistent with the
use of a system model.
• These include—
–
–
–
–
High Efficiency for all components leads to…
Numerical Goals
Incentive Pay
Merit Systems of Evaluation
How do we learn?
• Represent the ‘object’ we want to say
something about with a model
• Manipulate the model—perhaps with data
• Use the prediction of the model
Examples of Models
•
•
•
•
•
Photograph
Verbal description
System Representation
Measurement System
Senge Diagram, Flow charts etc
A System
• A collection of components—methods,
equipment, materials, environments, and
people—that come together repeatedly for
a purpose.
• The people have individual objectives and
aims---aligned with the purpose.
• The test of the purpose—Does everyone
win if we attain the purpose?
High Efficiency may not be best
• Purpose –High quality –high throughput--low cost—and little WIP
• Component 1: Capability 500 per hour
• Component II: Capability: 200 per hour
• InputComponent ICompon.IIOutput
• If component I operates at 100% efficiency,
throughput will not increase yet WIP will
grow.
Measurement System
• Key Questions: Why? What? And How?
• Why? This defines the purpose of the
system
• What should we do to achieve the
purpose?
• How do we do it?
Purpose
• To make a clear and concise statement --that will create action---regarding the
status of the system
What should we measure?
• Outcomes—responses—as broad as possible
to represent the state of the system.
• Predictors—of the outcomes—as far upstream
so as to be useful. (Outcomes are too late and
too expensive to manage.)
• Stratifying Variables—Related to the
individuals or groups you want to compare
and the outcomes
Simpson’s Paradox
• Consider two basketball players C and G.
The percentage of shots from the field that
score: 60 and 50% respectively.
• Who is a better shooter from the field?
Distance from the Hoop
C
G
70
90
0.75
0.2
Far
C
G
30
40
0.25
0.8
Merit Pay
• Do we know all differences—stratifying
variables with values– identifying each of
the employees?
• Are there potential “lurking” variables?
Time Plot
• Key issue involves system stability. A stable system
–with respect to a measurement system—yields a
sequence of identically distributed outcomes. The
probability of a 5, say, on the first run of the
system is the same as the probability of a 5 on any
other run of the system. When a system is stable
ask about the common cause of the variation.
• This idea comes from W Shewhart—circa 1935
• And is an essential element of system
management—W Edwards Deming (1900-1993)
ALL Dimensions
• In the Goal E. Goldratt, in his critique of
high efficiency, identifies the key measures:
throughput, W.I.P., and cost.
• What aspects of the operation are not
reflected in these three measures?
No True Value of a Measurement
• Operational definitions are required
Choose between Two Salary
Options
1. Start at $20,000 per year and increases in
your annual salary by $2000 each year.
2. Start at $10,000 per six months and
increases in the six-month salary by $500
each six-month period.
Gross Salary for each Six-month Period
•
•
•
•
•
•
$10,000
$10,000
$11,000
$11,000
$12,000
$12,000
•
•
•
•
•
•
$10,000
$10,500
$11,000
$11,500
$12,000
$12,500
It’s the Harmonic Mean not the
Arithmetic Mean
• If you invest $D each month you have spent
2xD dollars in two months, and
• You’ll have D/10 shares the first month and
• D/20 shares the second month.
• Average price per share =2D/{D/10 +D/20}
• =$13.33