Sampling

Proper sampling is a kind of art. We need to guarantee that we collect data from all parts of the population is represented but none is overrepresented. Even the simplest poll creates lots of questions. If you are fortunate to measure variables with measurable physical properties, we face similar problems. To meet these requirements, the sampling is either

·        representative: the sample must come from all the subgroups according to their weight. We must have exact knowledge about the driving factors (for example at an election: age, gender, highest degree, etc.). See the US Election 1948 and how it can go wrong. Or

·        random: we try to find a scheme that is close to random. For example, if we have a batch of 1000 pieces and need to measure 50 pieces. If we select every 20th, may hit a pattern. If we have a 4-cavity moulding tool, there is a high chance of getting parts only from one cavity. Or due to easier accessibility, we do not take parts from the middle of the box. To avoid this, we may select the closest prime number (19) and pick every 19th.

Additionally, we must ensure that the measurement is free from bias. It is challenging to formulate neutral questions for a poll but we must work on it. A calibrated device with acceptable measurement system analysis results serves this purpose for dimensional characteristics.

A critical part of sampling is the sample size. Minitab has an adequate sample size calculator, although this area is not handled as its importance. Before starting sampling, we must define what differences we want to detect. If we select too few, the analysis gives no adequate answer. If we choose too many, the time and costs are unnecessarily high.

Discrete variable requires at least one order more (starting in the magnitude of 1000) data show a difference from continuous variables (pretty acceptable from 30).

#projectmanagement #interim #sixsigma #medicaldevices #operationalexcellence