Who could ever be against having good measurements? Good measurements are like apple pie and motherhood. Since we all want good measurements, it sounds reasonable when people are told to check out the quality of their measurement system before doing an experiment or putting their data on a process behavior chart. In this column I shall consider what properties your measurement system does and does not need in order to be useful in experimental or observational studies.
|
ADVERTISEMENT |
The structure of an observation
Any observation can be visualized as having two parts: One component will represent the value of the thing being measured, and the other will consist of the combined errors of the measurement operation.
Observation = Product Value + Measurement Error
Once we have this model for an observation, we can begin to discuss the properties of the measurement system by considering the properties of the measurement errors themselves. In most cases the properties of greatest interest will be consistency, precision, and bias. As always, the question of consistency (or homogeneity) is primary.
…
Comments
Temperature - weather/climate
Are temperature measurements of the weather at one place regarded in the context of "the same thing" - "If repeated measurements of the same thing are placed on an XmR chart they should display a reasonable degree of homogeneity" ? Air temperature is never "the same thing" because the air is always moving. Seasons, night and day, cause the temperature to rise and fall, making XmR charts useless. However Jones et al claim that the accuracy of global average temperature is known to 1/1000th deg C, because of the number of readings taken, despite most temperature data until recently having a recording accuracy of +/- 0.5 deg C, and despite huge changes in the number of readings used for averages over a number of years, and despite changes in global coverage.
I'd be interested in a discussion of the accuracy of such data.
Add new comment