Modified RSS (MRSS): The Middle Ground Between RSS and Worst-Case
Pure RSS assumes processes stay centered. Real processes drift. Modified RSS (MRSS) applies a correction factor — typically 1.5× — to RSS results, accounting for mean shifts without going all the way to worst-case. This guide explains when MRSS is the right choice.
Worst-Case
All at limits simultaneously.
MRSS
RSS with mean-shift buffer.
RSS
Pure statistical, centered.
Why RSS Isn't Always Enough
RSS analysis assumes every dimension is centered on its nominal value. In reality:
- Tool wear causes dimensions to drift over time
- Setup variations shift the mean between batches
- Thermal changes affect machine accuracy throughout the day
- Operator differences introduce systematic biases
These "mean shifts" don't increase random variation (σ) but do move the distribution center away from nominal. RSS misses this entirely.
⚠️ The RSS Blind Spot
A process with Cpk = 1.33 (±4σ = tolerance) can still drift 1.5σ from nominal while staying "in control." RSS assumes the mean never moves. MRSS accounts for this realistic scenario.
What MRSS Does
Modified RSS multiplies the RSS tolerance by a correction factor:
MRSS Tolerance = RSS Tolerance × Factor
Common factor: 1.5 (accounts for ~1.5σ mean shift)
This gives you a range that's more conservative than RSS but still tighter than worst-case — typically 40–50% smaller than WC.
The 1.5 Factor: Historical Origin and Reality
Why 1.5? The factor has its historical origin in Motorola's Six Sigma methodology:
- A "Six Sigma" process has ±6σ within tolerance
- Motorola observed that processes typically drift ±1.5σ over time
- So a "6σ process" actually delivers 4.5σ performance in practice
- The factor 1.5 emerged from this observation
Important: MRSS Is a Convention, Not Physics
The 1.5× factor is not a universal physical law for tolerance stacks. It's an engineering convention adopted by many companies based on Motorola's Six Sigma observations. Different organizations use different factors (1.2–2.0) based on their process experience. Document your factor and its rationale. If you need physics-based mean-shift modeling, use Monte Carlo simulation with explicit per-dimension mean shifts.
For many tolerance stacks in manufacturing environments with moderate drift, the 1.5× factor provides a reasonable margin. But it's a rule of thumb, not a derived physical relationship.
MRSS Formula
Standard RSS (Baseline)
RSS = √(Σ ti²)
Where ti is the tolerance of dimension i.
Modified RSS
MRSS = RSS × k
Where k = correction factor (typically 1.5).
Alternative MRSS Forms
Some organizations use more complex formulas that apply the factor differently:
Bender formula:
MRSS = √(Σ ti²) + k × (WC − RSS)
Adds a fraction of the WC-RSS gap. When k=0.5, this approximates ×1.5.
When to Use MRSS
Choose MRSS when:
- Processes drift — Tool wear, thermal effects, or setup variation are known issues
- RSS is too optimistic — Past assemblies showed more variation than RSS predicted
- WC is too pessimistic — Functional analysis shows WC drives unreasonable tolerances
- Cpk data is unavailable — You can't verify that processes stay centered
- Company standard requires it — Many automotive and aerospace firms mandate MRSS
When NOT to Use MRSS
Stick with RSS when:
- Processes are SPC-controlled — Active monitoring keeps means centered
- Per-dimension Cpk is high — Cpk ≥ 1.67 means little room to drift
- Monte Carlo is available — MC can model mean shifts explicitly
Use Worst-Case when:
- Safety-critical — 100% conformance is required
- Low volume — Statistics don't apply to 10-unit runs
- MRSS still fails spec — If MRSS doesn't pass, WC definitely won't
Worked Example
Consider a 4-dimension stack:
| Dimension | Nominal | Tolerance (±) | t² |
|---|---|---|---|
| Housing | 50.00 | 0.10 | 0.0100 |
| Bearing | 25.00 | 0.05 | 0.0025 |
| Shaft | 24.90 | 0.08 | 0.0064 |
| Retainer | 0.05 | 0.02 | 0.0004 |
| Sum of t² | 0.0193 | ||
Step 1: Calculate RSS
RSS = √0.0193 = ±0.139 mm
Step 2: Calculate MRSS (factor = 1.5)
MRSS = 0.139 × 1.5 = ±0.209 mm
Step 3: Compare to Worst-Case
WC = 0.10 + 0.05 + 0.08 + 0.02 = ±0.25 mm
Results Comparison
| Method | Tolerance | vs RSS |
|---|---|---|
| RSS | ±0.139 mm | 1.0× |
| MRSS (1.5) | ±0.209 mm | 1.5× |
| Worst-Case | ±0.250 mm | 1.8× |
MRSS (±0.209 mm) is 50% more conservative than RSS but 16% tighter than WC. If your spec allows ±0.22 mm, MRSS passes while RSS might be optimistic.
Choosing Your MRSS Factor
The 1.5× factor is the most common convention, but your organization should define and document its MRSS method. Different factors represent different assumptions about process drift:
| Factor | Typical Use (Company-Defined) |
|---|---|
| 1.2–1.3 | Well-controlled processes, frequent SPC, documented low-drift history |
| 1.5 | Common convention — moderate drift assumed (Six Sigma origin) |
| 1.7–2.0 | Conservative margin for unknown suppliers or high-drift processes |
Best practice: Document which factor your organization uses and why. If process data exists (Cpk, historical drift), use that to justify your choice. If you need physics-based modeling rather than a convention, use Monte Carlo with explicit mean shifts per dimension.
MRSS vs. Monte Carlo: Convention vs. Physics-Based Modeling
MRSS applies a global correction factor based on convention. Monte Carlo lets you model explicit, per-dimension mean shifts when you have the data. Here's how they compare:
| Aspect | MRSS | Monte Carlo |
|---|---|---|
| Speed | Instant (formula) | ~1 second (simulation) |
| Mean-shift handling | Global factor (1.5×) | Per-dimension explicit shifts |
| Distributions | Assumes normal | Any distribution |
| Output | Single tolerance number | Full histogram, percentiles |
| Best for | Quick estimates, standards compliance | Complex stacks, detailed analysis |
Use MRSS when you need a quick, convention-based estimate that's more conservative than RSS. Use Monte Carlo when you need physics-based modeling: explicit per-dimension mean shifts, non-normal distributions, or detailed insight into distribution shape and tail risks. Monte Carlo with measured process data gives you actual predictive power; MRSS gives you a standardized margin.
Common Mistakes
Applying Factor to Wrong Quantity
MRSS factor multiplies the RSS tolerance, not individual tolerances. Don't apply 1.5× to each dimension — that gives a much larger (incorrect) result.
Double-Counting Conservatism
If you already use inflated Cpk assumptions (Cpk < 1.0 per dimension), applying MRSS on top is overly conservative. Use one method to account for process drift, not both.
Using MRSS When WC Is Required
For safety-critical or 100% interchangeability requirements, MRSS is not sufficient. It still assumes statistical averaging — some assemblies will exceed the MRSS range.
Industry Adoption
MRSS is widely used in industries where RSS is too optimistic but WC drives impractical tolerances:
- Automotive — Many OEMs specify MRSS (factor 1.4–1.5) for powertrain tolerances
- Aerospace — Used for non-safety-critical assemblies where WC isn't mandated
- Consumer electronics — Balances yield against tight packaging constraints
- Medical devices — Applied to non-critical interfaces; safety items use WC
Summary: Decision Matrix
| Situation | Method |
|---|---|
| Safety-critical, 100% conformance | Worst-Case |
| High volume, SPC-controlled, centered processes | RSS |
| Process drift expected, need margin over RSS | MRSS (1.5×) |
| Complex distributions, need histogram/percentiles | Monte Carlo |
| Unknown process capability | MRSS or WC |