Contents - Index

Representative Sample

All sampling plans assume that a representative sample is selected from the lot being inspected.  One way of selecting a representative sample is a random sample.  However, this type of sample is frequently difficult to implement.  Two equally valid methods are stratified sampling and periodic sampling.

A stratified sample is one where it is specified that equal number of samples should come from different parts of the lots.  For example, suppose the product is contained in 10 tote pans and that 20 samples are required.  It might be specified that one unit is to be selected from each end of each tote pan.

A periodic sampling is one where a frequency of selecting samples is specified.  For example, suppose 10 samples are to be selected from a lot of 1000 units.  It might be specified that every hundredth unit that passes by be included in the sample.  When performing periodic sampling it is important to make sure the sampling period doesn't align with any process period.  For, example, don't select every 40th unit from a 40 well tabletting process because all the sample will come from one well.

A representative sample is important to assure a run or cluster of defects is not missed.  If the defects where to occur at random, any method of selecting samples would be fine.  Two methods of sampling that should not be used are handful sampling and carton sampling.

Handful sampling is where multiple units are selected at once in a single handful.  This reduces the chances of detecting a run or cluster because the samples all come from the same portion of the lot.  Never select more than one sample at a time.  Handful sampling should always be avoided.

Carton sampling is where, in order to save time, multiple samples are selected from a small number of cartons.  For example, suppose there are 50 cartons and 20 samples are required.  Ideally, 20 cartons should be selected and one unit removed from each.  However, instead 5 samples are selected from each of 4 cartons.  This reduces the chances of detecting a run or cluster because the samples all come from the same portion of the lot.  Carton sampling should be avoided if possible.