Scenario
"What's going on with this process capability report? The yield clearly dropped to 95%, and you're showing me a CPK = 1.5? The customer wants yield, not just a good-looking number! Did you mess up the data?" The PM's voice echoed through the office. The engineer looked at the report in his hand; he had entered all the data, and the software had processed it, so how could this be? He felt he hadn't made a mistake, yet couldn't explain why, leaving him looking utterly embarrassed.
Plain English Explanation
In the factory, we deal with data every day, but did you know? Not all data are alike. There are two basic types, and getting them wrong is like using a weighing scale to measure height – the result will naturally be incorrect. These two types of data are: Variable Data and Attribute Data.
Think of them as two different ways of "measuring things":
- Variable Data:
* For example: wafer thickness, voltage, temperature, line width, product weight. These are values that can be precisely measured.
* This type of data can tell you "how good" the process is, not just "whether it's good or not."
- Attribute Data:
* For example: number of defective products, number of defect points, quantity of pass/fail items, number of customer complaints. You can only count how many there are; you cannot say there is "half" a defect.
* This type of data primarily tells you "whether the process is good or not," meaning whether it meets standards.
Why is it important to understand this clearly? Because they require different analytical tools! You cannot use tools for analyzing variable data (e.g., Cp/Cpk) to analyze attribute data, and vice versa. It's like trying to count apples with a measuring tape; the result will be completely misleading. The engineer's problem lies precisely here: he likely misused a measurement-based CPK to analyze attribute-type yield data, which naturally led to being scolded by the PM.
Practical Judgment
To correctly interpret data, the first step is to clarify which type your data belongs to. Below is a simple judgment criterion and practical advice:
| Data Type | Characteristics | Applicable Control Chart | Applicable Process Capability Metric | Examples |
|---|---|---|---|---|
| Variable Data | Continuous, has units, quantifiable, can have decimal places | X-bar & R/S Charts | Cp, Cpk, Pp, Ppk | Wafer thickness, voltage, temperature, line width, product strength |
| Attribute Data | Discrete, usually integers, counts or proportions | p/np/c/u Charts | Yield, Defect Rate, DPMO, Defects Per Unit (DPU) | Number of defective products, number of defect points, quantity of pass/fail items |
Practical Advice:
- Define Data Source: Before starting any data analysis, you must first clarify whether your data comes from "measurement" or "counting." This is the foundation for selecting the correct analysis tool and the first step to avoid mistakes.
- Goal-Oriented Selection:
* If you want to monitor "defect rates" or "flaw rates," and the data is attribute-type (e.g., number of defective items per batch, number of defects per wafer), then you should choose p/np/c/u Charts.
- Limitations of Cpk: Cpk is a process capability metric specifically for variable data. It evaluates the degree of deviation between the process center and the specification center, as well as the magnitude of process variation. For attribute data, such as yield, directly calculating Cpk is meaningless because yield itself is already a result; it does not possess the concept of a continuous distribution with upper and lower specification limits. High yield does not necessarily mean good process capability, nor does low yield necessarily mean poor Cpk. The two are indicators of different dimensions and should not be confused.
How InsightFab Does It
In InsightFab, you simply upload your data, and the system automatically determines the data type and recommends the most suitable control chart and analysis method, allowing you to easily choose the right tool and avoid misinterpretations.
Golden Rule
"Understand whether your data is 'measured' or 'counted' to choose the right tools and let the data speak the truth."