What's the Difference Between Cpk and Ppk? —— The Same Process, Cpk 1.35 and Ppk 0.93, What Does the 0.42 Difference Indicate?

1. Reviewing the Scene: Two Conflicting Numbers

A process capability review at a precision stamping factory. On the process engineer's screen, two numbers are displayed side by side:

Cpk = 1.35, Ppk = 0.93

The same characteristic, the same batch of data, the same specification, yet the two numbers differ by 0.42, and they straddle the 1.33 threshold—one just meets the standard, the other does not.

In the meeting room, three opinions emerged:

• Process: "Cpk 1.35, the process capability is sufficient, the equipment is fine."
• Quality: "Ppk 0.93, there is a risk of exceeding the specification in actual customer measurements, so it should be judged as nonconforming."
• The plant manager's assistant suggested a compromise: "Can we report both and let the customer decide?"

I've seen this scenario too many times. Both sides are not wrong, but they are only telling half the story. Cpk and Ppk are not two indicators that can be "picked" at will—they answer two different questions, and the difference between them is the most valuable information.

This article aims to explain this thoroughly: why there are two indices, what the exact difference is, what the 0.42 difference means, and how to use this difference to directly identify improvement directions.

2. Let's Be Clear: The Numerator is Exactly the Same, Only the Denominator Differs

Placing the two formulas side by side is the foundation for understanding everything:

Cpk = min(USL − μ, μ − LSL) / (3 × σ_within)
Ppk = min(USL − μ, μ − LSL) / (3 × σ_overall)
      └──────── Exactly the Same ────────┘

The numerator is identical: the same upper specification limit USL, the same lower specification limit LSL, and the same process mean μ.

The only difference is the σ in the denominator. Therefore, there is an identity:

Cpk ÷ Ppk = σ_overall ÷ σ_within

This is not a coincidence; it is directly derived from the definitions. Remember the significance of this ratio—it measures how much the process center has drifted.

3. What Do the Two σs Measure?

To determine which of Cpk and Ppk is "more correct," we need to understand what each σ represents.

σ_within (within-subgroup standard deviation): This measures the variation within the same subgroup. A subgroup is a set of samples taken under almost identical conditions and in a continuous manner (e.g., 5 pieces taken continuously from the same shift). Therefore, within-subgroup variation represents the inherent instantaneous variation of the process under stable conditions—equipment precision, material variation within a batch, and operator technique variation.

σ_overall (overall standard deviation): This measures the variation by mixing all individual values. It is equal to:

Overall variation = Within-subgroup variation + Subgroup-to-subgroup "process center drift"

Mathematically, this is strictly true: Total sum of squares = Within-subgroup sum of squares + m × Subgroup mean sum of squares (where m is the subgroup size). The differences between subgroups are not averaged out but are fully included in the overall variation.

A set of actual measurements (25 subgroups × 5 pieces = 125 measurement values):

σ_within (R̄/d2)          = 1.7885   ← Used for Cpk
σ_between (between subgroup means) = 1.9485   ← Subgroup drift amount, almost as large as within-subgroup
σ_overall (all 125 values)   = 2.4217   ← Used for Ppk, 35.4% larger than within-subgroup

Cpk = 1.252      Ppk = 0.925
1.252 ÷ 0.925 = 1.354 = 2.4217 ÷ 1.7885   ✓ The identity holds

Notice the most glaring point here: the subgroup-to-subgroup variation (1.95) is almost as large as the within-subgroup variation (1.79). This is not due to equipment instability but the process center drifting.

4. "Short-term/Long-term" is the Most Misleading Statement

Almost all textbooks say, "Cpk is short-term capability, Ppk is long-term performance." This statement is not wrong, but it misleads 90% of readers because it naturally leads to the interpretation of "data collection time span."

This is not the case.

• Collecting 25 subgroups over 20 minutes—as long as you mix all individual values, the result is still Ppk.
• Collecting data over 3 months—as long as there is a reasonable subgroup structure, the result is still Cpk.

"Short-term" in "Cpk = short-term" refers to the time scale of the sources of variation: within-subgroup variation represents the best case when conditions are held constant, which is a theoretical instant; overall variation includes "how much the center has drifted over time." The time span is a result, not a criterion.

The correct terminology should be:

Cpk = Within-subgroup capability index   (within-subgroup capability)
Ppk = Overall performance index           (overall performance)

The criterion is only one: whether the σ in your denominator is calculated within subgroups or across all data.

5. What Questions Do They Answer?

This is the key to choosing between them:

So, returning to the initial scenario:

• The process engineer says, "Cpk 1.35, the equipment is fine" — correct, the equipment can indeed produce good results.
• The quality engineer says, "Ppk 0.93, there is a risk of exceeding the specification" — also correct, the customer indeed received something with risk.

The complete conclusion from both perspectives is: The equipment is sufficient, but the process center is drifting.

6. A Comparative Experiment: Same Within-subgroup Variation, Center Drifts, Ppk Collapses

To make this concept tangible, let's use two sets of simulated data. Both have the same within-subgroup variation (σ ≈ 1.0), with the only difference being whether the "center drifts":

A · Stable Process with a Fixed Center

σ_within = 0.963    σ_overall = 0.958    σ_overall/σ_within = 0.995
Cp = 1.04   Cpk = 1.03      Pp = 1.04   Ppk = 1.04
→ Cpk ≈ Ppk, the process is stable

B · Process with a Center Drifting Between ±2σ

σ_within = 1.088    σ_overall = 1.671    σ_overall/σ_within = 1.536
Cp = 0.92   Cpk = 0.87      Pp = 0.60   Ppk = 0.56
→ Cpk is much greater than Ppk, the center is drifting

The within-subgroup variation of both processes is almost the same, but B's Ppk collapses to 0.56.

In one sentence: The difference between Cpk and Ppk is the amount of process center drift.

7. Deliverable: Cpk/Ppk Ratio Diagnostic Table

This ratio can be used directly as a diagnostic tool. The table below provides interpretations based on the assumption of a two-sided specification and approximately normal data:

What happens after identifying the drift source? Once the center stabilizes, σ_overall will converge towards σ_within, and Ppk will automatically catch up—only then will the two indices converge, and the capability index truly represent the process.

Three specific troubleshooting actions:

1. Check the drift-related rules on the control chart: Rule 2 (9 consecutive points on the same side), Rule 3 (6 consecutive points increasing or decreasing), Rule 5, Rule 6—these rules indicate "center drift."
2. Layered comparison of subgroup means: Calculate the mean for each shift, batch, equipment, and mold (one AVERAGEIFS function can do this), and see which layer's mean is skewed.
3. Align the time axis: Overlay the Ppk variation cycle with the moments of shift changes, batch changes, and model changes—the drift often aligns with a specific operation.

8. The Same Cpk Can Have Two Different Values

This is another source of confusion and the real reason why many people argue, "Your Cpk is wrong."

σ_within can be estimated in several ways, with different results:

For the same data, Cpk can be 1.252 or 1.359, a difference of 8.5%. Neither number is wrong; they are different estimates.

R̄/d2 is the classic estimation method from the control chart era (only the range is needed, can be calculated manually), but it only uses the two extreme values within the subgroup, discarding a lot of information; pooled s uses every data point within the subgroup, making it statistically more effective. This is why different statistical software can give different Cpk values for the same data.

Therefore, the reporting standard for capability indices should be:

Not recommended: Cpk = 1.25
Recommended:     Cpk = 1.252 (σwithin = R̄/d2, n=5, 25 subgroups)
Recommended:     Ppk = 0.925 (σoverall = sample standard deviation s, n=125)

A capability report without specifying the estimation method and sample size cannot be verified.

9. How Are They Used in Standards?

The common practice in the automotive industry (AIAG, IATF 16949 system):

Initial process study (PPAP submission)  →  Use Ppk ≥ 1.67
                                         Require ≥100 data points, ≥25 subgroups
Production process capability            →  Use Cpk ≥ 1.33
Special characteristics CC / SC          →  Require ≥ 1.67

The logic behind this specification aligns with the definitions provided earlier: a new process has not yet proven its stability, so only the actual performance (Ppk) can be considered; once the process is in production, controlled, and continuously validated, capability (Cpk) can be discussed.

Therefore, "using Ppk for PPAP and Cpk for production" is not an administrative regulation but a necessary result of statistical logic. Conversely, if a new process has a good Cpk but a poor Ppk, it means the process is not stable and should not be approved.

10. Three Prerequisites (Lack of Any One Makes the Index Misleading)

① The measurement system must be qualified. GR&R ≤ 10% (or as per customer requirements), and the measurement tool resolution should reach 1/10 of the tolerance band. The noise from the measurement system is fully included in σ—if the measurement is unstable, the capability index is false.

② Cpk must be calculated in a controlled state. When the process has out-of-control points, the "within-subgroup variation" itself does not represent any stable condition; the Cpk calculated at this time is merely a decoration. Ppk can still be calculated even if the process is not in control, but it is only a snapshot and cannot be used as a capability measure.

③ Normality assumption. The formulas for Cp, Cpk, Pp, and Ppk all assume that the data is approximately normal. Directly applying these formulas to count data or one-sided truncated data (e.g., positional tolerances with only an upper limit) will lead to incorrect conclusions—such cases should use non-normal methods or transform the data before calculation.

Cpk is "whether the equipment can achieve it," Ppk is "how much it actually achieved over a period"; the difference between the two is the amount of process center drift. When the indices conflict, first identify the action causing the center to drift.

Knowledge code: 6.3.2

Version: v20261017

Author: QTank QTank is dedicated to providing systematic professional knowledge, methodologies, and practical tools to quality management practitioners, helping enterprises continuously improve their quality capabilities.