Which control charts to use?
Quality Analyst 651A A common question is "Which control chart should I use?" Although the answer can become deep and complex, here are some simple recommendations. Which Control Charts to Use? A common question is "Which control chart should I use?" Although the answer can become deep and complex, here are some simple recommendations. First decide what type of data you're dealing with. A. Variable data takes on a measurable, numeric value. There are many possible values. B. Attribute data consists of categories. There are only a few (usually two) discrete values. With variable data, decide how large subgroups are. If the subgroup size is one, then use an Individual measurements chart, with or without a moving R chart. If the subgroup size is from two to ten (or possibly twelve), then use an X and R chart. If the subgroup size is over ten (or twelve), then use an X and S chart. There are also some specialized charts for variable data. They can be helpful, but are more difficult to use and explain. Here's a quick list: 1. The Cumulative Sum (CuSum) chart is more sensitive to small, sustained changes in level than the standard control charts. 2. The Exponentially Weighted Moving Average (EWMA) chart is also sensitive to small, sustained changes in level. It smoothes out noisy data sets and is sometimes used with autocorrelated data. 3. The Median and Individual measurements (MI) chart can control an entire family of processes in one chart. It's useful for multi-cavity molds and multi-head fillers. With Attribute data, decide on what type of distribution the data follows. Binomial data takes on two values, usually "good" or "bad". If the sample size is constant, use an np-chart. If the sample size changes, use a p-chart. Poisson data is a count of infrequent events, usually defects. If the sample size is constant, use a c-chart. If the sample size changes, use a u-chart. BEWARE! The p-, np-, c-, and u-charts assume that the likelihood for each event or count is the same (or proportionally the same) for each sample. In addition, c- and u-charts require that the event be "rare". If the data follows the theoretical model, attribute charts can offer advantages. If the data violates theory, the attribute charts generally fail. Try using an Individual measurements and moving Range (IR) chart instead. The attached table summarises the charting choices. QualityAnalyst All en Y Thomas Harding/Users/AdeptScientific