Meaning
Count data models in quality control often encounter a higher frequency of zero defects than is predicted by a standard Poisson distribution. This statistical model, known as zero inflated Poisson, addresses this issue by combining a binary distribution for the zero-defect state with a standard Poisson distribution for the remaining states. It is particularly useful for modeling processes that are highly stable and produce zero defects most of the time but can experience random failures due to external factors.
By accounting for this excess of zeros, the model provides a more accurate estimate of process quality.
Model Selection
Choosing between a standard Poisson model and the zero inflated Poisson model depends on the nature of the zero counts. If the zeros are generated by two distinct physical processes, such as a machine being turned off versus the machine operating perfectly, the zero inflated model is preferred. In pharmaceutical packaging, for instance, a seal is either perfectly formed (generating zero leaks) or it is subject to standard random variations that may occasionally produce leaks.
Distinguishing these two sources of zeros is essential for correct risk assessment. It allows engineers to identify whether a problem is systemic and requires a redesign of the equipment or if it is a random variation that can be handled through standard process controls.
Parameter Estimation
Fitting this model requires estimating both the probability of the perfect state and the mean rate of the Poisson process. Statisticians use maximum likelihood estimation or Bayesian methods to find these parameters from the observed defect counts. Because the model has two components, the calculations are more complex than those for a standard Poisson distribution.
However, the resulting estimates are much more reliable, especially when the process has a low average defect rate.
Operational Application
Quality managers use the results of this model to optimize their audit schedules and allocate inspection resources. If the model indicates that the probability of the perfect state is very high, the company can safely reduce the frequency of audits. This reduction lowers labor costs and minimizes disruptions to the production line.
It also allows the quality team to focus their attention on lines that are more prone to variation and failure.