178 J. Rosenberg
There are also statistical methods for the optimization
of process metrics, such as Evolutionary Operation (Box and Draper, 1969), response surface methodology Montgomery and Myers, 2002), and data envelopment analysis/stochastic frontier analysis (Jacobs et alb.
Data QualityAt this point, it is appropriate to return to the context of measurement and the dependence of statistical analysis on the quality of the underlying data collection process.
Data quality is a critical problem in industrial management, yet one often only vaguely recognized by decision makers who consume the ultimate endproducts of those data. This problem has come to light with the
development of data warehouses, as warehouse developers discover that bad data can turn a data warehouse into a data garbage dump. The first step, then, in using measurements is ensuring that those measurements are of sufficient validity and accuracy to enable conclusions to be drawn from them.
The sources of data quality problems are manifold (apart from the question of bad metrics, dealt within Sect. 3).
Chief among them are●
Organizational problems
●
Lack of precise definitions
●
Lack
of data validation●
Missing data
●
Sampling bias
6.1. Organizational ProblemsIt is common for metrics to be defined and collected by people other than those to whom the metrics apply this a recipe for trouble. The problem is exacerbated when a process is evaluated by management on the basis of metrics that the people carrying out the process find irrelevant or misguided the inevitable result is distortion of
Fig. 8A Process drifting slowly out of control as shown in (
a)
A standard control chart, (
b) A cusum chart
6 Statistical Methods and Measurement the work process to produce acceptable numbers, rather than valid or meaningful ones. Fora metrics
program to be successful, all parts of the organization involved need to be in agreement on the meaningfulness of the metrics and their role in the organization’s effective functioning.
Share with your friends: 
