Calculating Standard Deviations

Most of you know how to calculate a standard deviation or obtain it using Excel. The basic formula may not immediately come to mind, but you know what it represents in relation to a population based on a set of results: “the dispersion of those results.”

In fact, depending on how it is intended to be used, there are several types of standard deviation. They all have the same name, but they are calculated in different ways.

If we consider three methods that provide the following input data:

Method 1: A mean and an R-bar control chart (XmR chart)
Method 2: An X control chart based on individual values
Method 3: Random sampling of values from a population

All three methods provide genuine potential sources of information. However, none of these methods is truly appropriate for every possible situation when estimating the standard deviation.

Long-term estimate of a standard deviation
The most commonly used formula is given by the following relationship:

where x̄ is the mean of the results, x represents the individual values, and n is the number of values.

Some statistical software programs use c4(n−1) instead of n−1, with c4 = 4(n−1)/(4n−3). If n = 5, then c4 = 0.94.

Method 1: When the data come from an x̄ and R-bar control chart, this usual estimate of the standard deviation is valid only if the process is stable, even though some people use it when processes are not stable.

This method overestimates the dispersion if the process is affected by a special cause. This estimate should never be used to calculate the limits of a control chart. Control limits are calculated using sampling distributions.

Method 2: When the data come from a chart with individual values, this approach can provide an estimate of process variability.

Method 3: For a random sample of data from a population, this is the only method that makes sense. It does not require determining the median or the average moving range.

Short-term estimation of a standard deviation

A standard method for estimating the standard deviation from an x̄ and R-bar control chart is given by the following relationship:

Short-Term Estimation of a Standard Deviation
A standard method for estimating the standard deviation from an X-bar and R-bar control chart is given by the following equation:

where R is the average moving range used in the control chart, and d₂ is a value from the table below that depends on the sample size used. For n = 5, d₂ = 2.326.

Method 1: When the data are obtained from x̄ and an R̄ control chart, this estimator may understate the value of the standard deviation because R does not take into account the variations, or special causes, that may occur in the x values.

Method 2: When the data are obtained from a chart with individual values, using the median moving range, the calculation of R is not possible.

Method 3: For a random sample of data from a population, this calculation is not possible because the sequence of the variation is not known.

Short-Term Estimation of a Standard Deviation

The following relationship is taken from one of the equation options used to determine the median centerline of an individual values chart:

with R = median moving range

Method 1: When the data come from x̄ and an R-bar control chart, this estimator is not directly applicable.

Method 2: When the data come from an individual values chart, this calculation is an alternative.

Method 3: For a random sample of data from a population, this calculation is not possible because the sequence of variation is unknown.

Short-term estimation of a standard deviation
The following equation comes from one of the equations used to determine the center line of an individual values chart:

Where MR is the moving range between two consecutive points, and d₂ is a value from the table below that depends on the sample size used; for n = 2, d₂ = 1.128.

Method 1: When the data come from x̄ and an Rbar control chart, this estimator is not directly applicable.

Method 2: When the data come from a chart with individual values, this calculation is an alternative.

Method 3: For a random sample of data from a population, this calculation is not possible because the sequence of variation is not known.

Short-term estimation of a standard deviation

The following relationship is sometimes used by software programs to pool standard deviations when there are m subgroups of n samples:

“where Sp is the pooled/average standard deviation based on m series as a function of n, and c4(d) is a constant depending on m and n.”

The purpose of c4(d) is to reduce the bias of the estimate.

Method 1: When the data come from an x̄ and R̄ control chart, this estimator may understate the value of the standard deviation because R does not account for variations, or special causes, that may occur in the x inputs.

Method 2: When the data come from a chart with individual values, this calculation is not directly possible because the calculation of R̄ for a subgroup size of one is not possible.

Method 3: For a random sample of data from a population, this calculation is not possible because the sequence of variation is not known.