There are five (5) basic steps in preparing for a capability study:
Step 1: Define the Response Parameter of Interest
The response parameter of interest (Y) is the dependent variable. This is
the product characteristic of interest. Note that any given characterization
study may consider more than one Y simultaneously, as will be the case in Part
B of the discussion.
It is important to ensure that the method of measurement related to each of
the selected parameters is valid and that the related measurement apparatus is
reliable. A measurement scale is valid if it reports what it purports to
measure. Scale reliability is connected with the notions of repeatability such
as the degree to which the apparatus displays error which significantly
inhibits repeatability. If in doubt, the measurement scale should be studied
and the measurement apparatus calibrated and subjected to a separate
capability analysis. This is particularly true in the instance of sensory
scales of measure, which should be avoided if possible.
Furthermore, it is very important to prepare a stepbystep instructional
plan to assist those individuals tasked with implementing the measurement
scheme.
Step 2: Identify the Measurement Vehicle
To gather data, there must be a source of measurement referred to as the
measurement vehicle. In some instances, the vehicle may be the actual product;
in others, it may be a contrived substitute. When destructive testing is
involved and/or the costs associated with sample preparation are high, the use
of a contrived vehicle should be considered. It is generally considered
preferable to use the actual product since inferences resulting from
statistical analysis are more likely to be valid.
If a contrived vehicle is used, care must be used to ensure its
representativeness in terms of the target population to which it is applied.
In some instances, this implies a sound understanding and definition of the
population. In many situations, this requires empirical testing and evaluation
of a statistical nature. The use of a contrived vehicle should always be well
documented and undergo substantive technical review by experienced
practitioners prior to implementation.
Step 3: Determine the Basis of Subgrouping
A rational subgroup is a sample in which all items are produced under
conditions that only allow random effects to influence the measurements. The
special, or "assignable," nonrandom causes of extraneous variability such as
differing raw materials, personnel, or test conditions will then occur between
subgroups rather than within them. The intent is to ensure minimum
withingroup variation of a random nature.
Subgrouping is generally considered "rational" if the resultant subsets of
data exhibit only random variations and are also homogeneous with respect to
the conditions under which the data was gathered such as the same design,
process, and material conditions being present at the time of data capture.
With such rationality, an unbiased estimate of the standard deviation can be
made. Time is a useful, but not unique, classification parameter for the
development of rational subgroups. Other classes of variables include but are
not limited to design, process, personnel, environment, and material.
A crucial point to remember is that in many cases, the bias resulting from
an inadequate sampling strategy may be large enough to preclude the researcher
from recognizing the process' "true" capability. Therefore, extreme care
should be taken when establishing the basis for rational subgroups; however,
common principles often do exist. In this context, a discussion of these
general principles would be relative. It may be necessary to test the basis of
subgrouping to ensure an optimum level of compliance to the latter
guidelines.
Step 4: Establish the Sampling Methodology
Before presenting the details of this step, it is important to recognize
its relative nature. A sampling strategy applicable to one situation may not
be generalized in some other situation. This portion of the discussion
provides basic direction for a less experienced researcher by highlighting
more common concerns, issues, and practices. Therefore, the following
information should not be strictly classified as hard rules. Sampling
methodology is essentially the procedure by which measurements are gathered.
There are two basic types of sampling methods  sequential and random. The
type of method used depend's upon, but is not limited to:
 The objectives of data collection
 The basis by which rational subgroups are formed
 The basic nature of the manufacturing process
 Production volume
 The time interval between observations
 The type of data being gathered
Sequential sampling is often considered the preferred method because it has
the greatest potential for realizing the intent of "rationality" where the
formation of subgroups is concerned. If sequential sampling is used, it is
generally recommended that the subgroup size (n) be targeted at n = 5 where
appropriate.
Again, the basic intent is to minimize systematic, nonrandom error in
light of Equation B1 and Equation B2.
Therefore, sampling methodology plays a critical role
when estimating capability. It is also generally recommended that when making
an estimate of shortterm capability, the total number of subgroups (N) be N =
6 for reasons of statistical precision. Some consider the ideal number of
subgroups to be 25<N<50; however, it must be stressed that such a
recommendation is usually made in the context of statistical process control
(SPC) charting activities. Since the first steps in a process characterization
study involve the analysis of shortterm capability, the level of estimate
precision afforded by such a large number of subgroups is not necessarily
required  hence the lower recommended number. If during the course of
sampling, any of the subgroups should prove to be biased, then all such
subgroups should be discarded from the final data analysis. If a subgroup is
discarded, a replacement subgroup of the same size should be gathered under
the same subgrouping and sampling scheme.
As a general rule, random sampling should be used if
it is not possible to retain the order of production. The sample would
represent all the production over a period of time, batch, or lot, and
therefore, bias may be present. The rule of minimizing within subgroup
variation would again apply when attempting to estimate instantaneous
reproducibility. If multiple random samples are used, the subgroup size (n)
should be n = 5, and the total number of subgroups (N) should be N = 6. If a
single random sample must be used because of practical considerations
associated with taking the measurements, the subgroup size should generally be
n = 30 and the number of subgroups should be N = 1.
Other combinations of N and n may be considered should
the manufacturing or sampling circumstances require; however, all such
sampling schemes should be constructed so that the total number of samples (N
x n) is around 30 for statistical precision. (Recall the relative nature of
this discussion. As a consequence, it must not be taken out of its
illustrative context and used as a basis for statistical debate.)
It should also be pointed out that no general rule or
guidance can be given relative to the frequency of subgroups. Each case must
be decided on its own unique merits. While it is recognized that production
and cost constraints significantly affect decisions regarding sampling
frequency, the guidelines listed in Step 3 must also be considered. This
essentially represents the tradeoff between data integrity, statistical
precision, and cost.
Justifiable deviations from these guidelines are
numerous and should be allowed if the manufacturing or data analysis
circumstances require; however, such deviations should be subject to
documentation and undergo a substantive technical review by an experienced
practitioner prior to implementation.
Step 5: Gather Data Related to the Response Parameter
In most situations, a data collection sheet should be
used to record data. The data collection sheet should capture the following
information:

Date(s) of data capture

Product characteristic being
investigated

Units of measure related to the product
characteristic

Type of measurement apparatus

Date of last calibration

Method of sampling

Production sequence (if
applicable)

Sampling interval

Time of data capture

Perturbing circumstances during data
capture

Subgroup size (n)

Total number of subgroups (N)

Name of person responsible for data
capture

Engineering or product specification (if
applicable).