Where actual measurement is not appropriate or it is difficult to achieve, data can be obtained by counting the number of occurrences of the characteristic under investigation (e.g. the number of black ones in a packet of jelly babies, or the number of light bulbs that don't work).
In these circumstances, sample sizes are generally much larger. They depend on the proportions of the yes/no, pass/fail population. For example, a failure rate of 1% would require about 2500 items to be sampled before we could estimate failure outcomes with confidence.
This type of counted data is known as attribute data.
It is generally recommended that the outputs selected for the sample be produced consecutively by the process, as this gives the best picture at the time of the sample.