GlossaryKind of term: Manufacturing termOutlier
What is an outlier?
An outlier is a value that stands clearly apart from the rest of the same series, such as one part that took far longer than the batch.
An outlier is a value so far from the other values in the same series that you have reason to doubt it arose under the same conditions. It can be a time per part, but also a measured dimension. If twenty flanges turn in 11 to 13 minutes each and one takes 40, that one is an outlier. An outlier is a signal, not an error in itself: the value can be genuine and show something unusual, or come from a measuring or booking error.
What does an outlier depend on?
Whether a value counts as an outlier follows from a chosen criterion, not from a law of nature. Common rules of thumb are more than one and a half times the interquartile range beyond the first or third quartile, or more than about three standard deviations from the mean; formal outlier tests usually assume a normal distribution. In a short series the outlier itself inflates the standard deviation and so stands out less; the median and quartiles are more robust there. Times are also skewed: a breakdown makes a part take longer, not shorter. A strikingly short time points rather to a booking error or a skipped operation.
How does an outlier work in practice?
Average an outlier in and the standard for that operation shifts; throw every odd value away and you may miss a tool wearing too fast in that material. A common approach is to trace the cause on the job card and with the operator, flag the value with that cause, and keep it out of the standard but in the data. The criterion you flag by is a convention; set it in advance. A first-off that takes longer every batch then counts as setup time, not as an outlier. A measured dimension is different: if it falls outside the tolerance, the part is nonconforming however exceptional the value, and a value leaves a capability study only with an assignable cause, such as a measuring error.
What is an outlier often confused with?
An outlier is not variation: variation is the normal band the values of a batch fall in, the outlier sits outside it. And an outlier that returns on the same operation is no longer chance but a signal about the route, the tooling or the program.
Frequently asked questions
Should I remove an outlier from the actuals?
Do not delete it; flag it. Write the cause next to it and keep it out of the standard you adjust. If it comes back on the same operation in later batches, it belongs in the analysis instead.
How do I tell whether a time is an outlier?
Set the per-part times of the batch side by side and compare them with the median and the band most of them fall in. Use a criterion agreed in advance, then check the job card for what happened to that part.