You run a manufacturing plant. Every day, your operators take
measurements. Dimensions, weights, temperatures, hardness values,
surface finishes. Those numbers flow into your quality system, your SPC
charts, your Cpk calculations, your acceptance decisions. You base scrap
decisions on them. You base process adjustments on them. You base
customer shipment approvals on them.
And almost none of you have ever asked the fundamental question:
Is your measurement system even capable of telling you the
truth?
This is not a hypothetical problem. In my 25 years of auditing
manufacturing quality systems across automotive, aerospace, medical
devices, and electronics, I have found that roughly 60% of the
measurement systems on factory floors produce data that is, to varying
degrees, unreliable. The gages are wrong. The operators are
inconsistent. The fixtures introduce variation. The environmental
conditions shift. And nobody knows — because nobody ever checked.
This article is about Measurement System Analysis (MSA), and more
specifically, why ignoring it is one of the most expensive mistakes a
manufacturer can make.
What Is Measurement System
Analysis?
Measurement System Analysis is exactly what it sounds like: a
structured methodology for evaluating whether your measurement system —
the gage, the operator, the method, the environment, the entire act of
measuring — produces data that is trustworthy enough to use for making
decisions.
The core idea is simple but profound: every measured value is
a combination of the true part variation and the measurement
error. What your gage reads is not what the part actually is.
It is what the part is, plus or minus the noise introduced by the
measurement process itself.
Mathematically:
Observed Variation = True Part Variation + Measurement System
Variation
If your measurement system variation is large relative to your part
variation, your data is mostly noise. You are making decisions based on
static. You are adjusting a process based on measurement error, not
actual process shift. You are scrapping good parts and shipping bad
ones, and you don’t even know which is which.
MSA gives you the tools to quantify this. To put a number on how much
of your observed variation is real and how much is your measurement
system lying to you.
The Gage R&R Study:
The Foundation of MSA
The most common MSA tool is the Gage Repeatability and
Reproducibility (Gage R&R) study. It is the backbone of
measurement system validation, and it works like this:
You select a number of parts (typically 10) that represent the
expected range of your process variation. You select a number of
operators (typically 3) who normally perform the measurement. Each
operator measures each part multiple times (typically 2-3 trials). Then
you decompose the total observed variation into its components:
-
Repeatability (Equipment Variation): Can the
same operator, measuring the same part, with the same gage, get the same
answer each time? If not, your gage itself is introducing noise. This
could be due to gage wear, poor design, inadequate resolution, or
environmental instability. -
Reproducibility (Appraiser Variation): Can
different operators, measuring the same parts, with the same gage, agree
with each other? If not, your method is operator-dependent. This could
be due to inadequate training, subjective measurement techniques, or
inconsistent fixture usage. -
Part-to-Part Variation: This is the real
variation in your parts — the signal you’re actually trying to
measure.
The results are expressed as percentages of total variation (or
tolerance, depending on the approach):
- Gage R&R < 10%: Acceptable. Your measurement
system is trustworthy. - Gage R&R 10-30%: Marginal. Acceptable depending
on the application, but you should be working to improve it. - Gage R&R > 30%: Unacceptable. Your
measurement system is producing more noise than signal. Any decisions
based on this data are suspect.
Here is what I see in the field: companies running Gage R&R
studies that come back at 40%, 50%, even 60% — and doing nothing about
it. The study gets filed away in a quality manual, and the factory keeps
using the same gage, the same method, the same untrained operators,
producing the same unreliable data, day after day.
The
Real-World Consequences of Bad Measurement Data
Let me paint a picture that I have seen play out dozens of times.
An automotive supplier is machining precision shafts. The critical
dimension has a tolerance of ±0.025 mm. They are using a digital caliper
with 0.01 mm resolution to measure this dimension. Their SPC chart shows
frequent out-of-control points. The process appears unstable. Operators
are constantly making adjustments — tweaking offsets, changing tooling,
slowing down cycles. Scrap rate is running at 4.5%.
Someone finally decides to run a Gage R&R. The result:
Gage R&R = 52% of tolerance. The measurement system
is consuming more than half the tolerance band. The “out-of-control”
points on the SPC chart are not real process shifts — they are
measurement noise. The “unstable” process is actually quite stable. The
scrap is being generated not by a bad process, but by a bad measurement
system that is causing operators to adjust a process that doesn’t need
adjusting, overcorrecting real variation, and scrapping parts that are
actually within specification.
This is not rare. This is the norm in plants that have never done
MSA.
Here are the specific ways bad measurement data destroys
manufacturing operations:
1. False Rejects
(Scrapping Good Parts)
When your gage reads high or low due to measurement error, you reject
parts that are actually good. This is pure waste. I audited a medical
device manufacturer that was scrapping 7% of its injection-molded
housings based on dimensional inspection. A Gage R&R revealed that
their CMM fixture was flexing under clamping pressure, introducing 0.015
mm of error into every measurement. After fixturing redesign, the scrap
rate dropped to 0.8%. That 6.2% reduction represented $340,000 per year
in recovered material and labor.
2. False Accepts (Shipping
Bad Parts)
This is the dangerous side. Your gage reads a bad part as good, and
it goes to your customer. In regulated industries (automotive,
aerospace, medical), this can trigger recalls, plant shutdowns, and
regulatory action. In any industry, it destroys customer trust. The cost
of one customer quality escape typically exceeds the cost of an entire
MSA program by orders of magnitude.
3. Phantom Process
Instability
Your SPC chart shows out-of-control signals that aren’t real. Your
engineers spend hours investigating “special causes” that don’t exist.
Your operators chase ghosts. Meanwhile, real process shifts go
undetected because they’re buried in the measurement noise.
4. Incorrect Process
Capability Estimates
Your Cpk is calculated from observed data. If your measurement system
contributes 30% of the observed variation, your Cpk of 1.33 is actually
a Cpk of something closer to 1.55 once measurement error is removed. But
you’ll never know — and you’ll never optimize the right thing. You’ll be
trying to improve a process that’s already capable while your
measurement system drags the numbers down.
5. Wrong Root Cause Analysis
When a defect investigation relies on measurement data that’s
unreliable, you end up fixing the wrong things. You change tooling when
the gage was the problem. You adjust temperatures when the operator
technique was the problem. You replace machines when the fixture was the
problem. I have seen companies spend hundreds of thousands of dollars on
equipment upgrades that solved nothing because the root cause was a
measurement system nobody ever validated.
The Five Types of
Measurement System Error
MSA is not just about Gage R&R. A complete measurement system
analysis evaluates five characteristics:
1. Bias
The difference between the observed average measurement and a
reference value (master or standard). If your gage consistently reads
0.003 mm high, it has a bias of +0.003 mm. Bias is systematic error — it
shifts all your data in one direction. You can often compensate for bias
with calibration, but only if you know it exists.
2. Linearity
Does the bias change across the measurement range? A gage might be
perfectly accurate at 25.000 mm but read 0.005 mm high at 50.000 mm. If
you only calibrate at one point, you’ll never catch this. Linearity
studies measure bias at multiple points across the gage’s operating
range.
3. Stability
Does the measurement system drift over time? A gage that was accurate
on Monday might be off by Thursday. Thermal expansion, wear, spring
fatigue, electronic drift — all of these introduce stability problems.
Stability is verified through regular calibration checks with a known
standard, plotted on a control chart over time.
4. Repeatability
As discussed above: can the same operator with the same gage get the
same result on the same part repeatedly?
5. Reproducibility
Also discussed above: do different operators agree?
All five must be evaluated for a complete picture. Most companies
stop at repeatability and reproducibility. Many don’t even get that
far.
Why MSA Gets Ignored
If MSA is so important, why do so many organizations skip it?
First, it feels like overhead. You’re not producing
parts while you’re doing a Gage R&R study. You’re pulling operators
off the line, tying up gages, taking time that could be spent making
product. In a plant running at 95% utilization, there’s never a “good
time” to stop and study measurement systems.
Second, the results are uncomfortable. Nobody wants
to discover that the data they’ve been basing decisions on for months or
years is unreliable. It raises questions about past shipments, past
scrap decisions, past process changes. It’s easier not to look.
Third, the fix is often expensive. If your Gage
R&R comes back at 45%, you might need to replace gages, redesign
fixtures, retrain operators, or move measurements to a
climate-controlled environment. These are real costs, and they’re costs
that don’t produce visible output — so they get deferred.
Fourth, many quality professionals don’t fully understand MSA
themselves. They’ve read about it. They know the AIAG MSA
manual exists. They might even have run a study once. But they don’t
internalize the fact that every measurement they take is potentially
contaminated by measurement error, and that this contamination has real
financial and quality consequences.
Practical MSA: How to
Actually Do This
Here is a practical framework for implementing MSA in a manufacturing
environment:
Step 1: Inventory
Your Critical Measurements
Not every dimension needs a formal Gage R&R. Start with the
critical ones — the dimensions that appear on your control plans, the
ones tied to customer specifications, safety requirements, or regulatory
compliance. List the characteristic, the gage used, the operator, and
the method.
Step 2: Prioritize by Risk
Which measurements carry the highest risk if they’re wrong?
Customer-facing dimensions on tight tolerances go first. Safety-critical
measurements go first. Measurements driving SPC charts go first.
Step 3: Conduct Gage R&R
Studies
Use the standard cross-tabulated approach: 10 parts, 3 operators, 2-3
trials. Use the ANOVA method if you have the statistical software (it’s
more accurate than the range method). Calculate %GRR against both total
variation and tolerance.
Step 4: Act on the Results
This is where most companies fail. They run the study and file it.
The value of MSA is in the actions it drives:
- High repeatability error (equipment issue): Replace
or repair the gage. Improve fixturing. Control the environment. Increase
gage resolution. - High reproducibility error (operator issue):
Standardize the method. Write clear, visual work instructions. Train and
recertify operators. Use go/no-go fixtures to eliminate
subjectivity. - High interaction (operator × part): Some operators
handle certain part types differently. Investigate and standardize.
Step 5: Make MSA Recurring
Measurement systems degrade. Gages wear. Operators change. Methods
drift. Run Gage R&R studies annually at minimum, or whenever you
change gages, operators, methods, or environments.
Attribute MSA:
Don’t Forget Visual Inspection
Everything I’ve discussed so far applies to variable measurements —
dimensions, weights, forces. But in most manufacturing plants, the most
common inspection method is visual: operators looking at parts and
making pass/fail judgments. Weld quality, surface defects, color
matching, assembly completeness.
Attribute measurement systems need MSA too. The tool for this is the
Attribute Agreement Analysis (or Kappa study). You give
multiple operators the same set of parts (some known good, some known
bad, some borderline) and have them classify each one. Then you
measure:
- Within-operator agreement: Does the same operator
give the same answer for the same part on repeated trials? - Between-operator agreement: Do different operators
agree with each other? - Agreement with standard: Do the operators’
decisions match the known truth?
I have run attribute studies where operators agreed with themselves
less than 80% of the time. Where different operators agreed with each
other less than 60% of the time. Where agreement with the known standard
was essentially a coin flip. These operators were making pass/fail
decisions on products going to customers every single day.
The fix for attribute measurement error is usually better visual
standards: boundary samples, photographs with annotations, physical
reference specimens, and clearer defect definitions. But you’ll never
know you need these fixes until you measure the measurement system.
The ROI of MSA
Let me put numbers on this, because quality professionals often
struggle to justify MSA to management.
Consider a typical mid-size manufacturer producing 500,000 parts per
year with a 3% scrap rate on a critical dimension. The cost per scrapped
part is $12. Annual scrap cost: $180,000.
A Gage R&R study reveals that 40% of observed variation is
measurement error. Analysis shows that approximately one-third of
scrapped parts were actually within specification — they were false
rejects driven by gage error. That’s $60,000 per year in unnecessary
scrap.
The fix: a $4,000 gage upgrade and $2,000 in operator training.
Payback period: one month.
Now consider the flip side: the cost of a false accept. A bad part
reaches a customer. The customer issues a corrective action request. You
conduct a containment, sort, root cause investigation, and corrective
action implementation. In the automotive industry, the average cost of a
customer quality escape is $25,000 to $250,000 depending on severity. A
single escape can cost more than your entire MSA program.
MSA is not overhead. It is insurance. It is also intelligence — the
kind that prevents you from making expensive decisions based on faulty
data.
The Deeper
Lesson: Trust Nothing, Verify Everything
The philosophy behind MSA extends beyond measurement systems. It is a
mindset that every manufacturing organization should adopt: the
data you trust is only as good as the system that produced
it.
Your SPC charts are only valid if your measurement system is capable.
Your Cpk values are only meaningful if your gage error is small. Your
scrap decisions are only correct if your inspection data is reliable.
Your process adjustments are only appropriate if the variation you’re
observing is real.
Most organizations treat measurement as a given — a transparent
window into their process reality. It is not. Measurement is a process
in itself, subject to variation, error, and failure, just like any other
manufacturing process. And it deserves the same level of scrutiny,
control, and improvement.
The factories that understand this — the ones that validate their
measurement systems before trusting their data — make better decisions.
They scrap less. They ship fewer defects. They investigate real problems
instead of phantom ones. They optimize the right variables. They sleep
better at night.
The ones that don’t? They keep making decisions based on numbers they
never verified, using gages they never validated, operated by people
they never trained, in environments they never controlled. And they
wonder why their quality never seems to improve.
Measure your measurement system. The data you save may be
your own.
Peter Stasko is a Quality Architect with over 25
years of experience in manufacturing quality management across
automotive, aerospace, medical device, and electronics industries. He
specializes in transforming quality systems from compliance-driven
paperwork into operational advantages that directly impact the bottom
line. His approach combines deep technical expertise in statistical
methods with practical, shop-floor-level implementation that real
factories can actually sustain.
