Internal inspection of sockets and switches.

I think Graham's comment is not intended to be take literally (well I hope not). Clearly as the sample size is reduced, the risk of picking up an unrepresentative number of good or bad examples rises, and the indeterminacy  of the sample result versus what you would have seen with 100% sampling increases. (the so called 'law of small numbers')

This is not to say that you cannot do a sampled test on a small number of items, just that you need to be aware of the perils of doing so.

Mathematics only really comes to the rescue properly in the case of a supposedly identical set of samples, telling you your confidence/uncertainty  in relation to sampled fraction say the output of a production line.

 ( the canonical example has to be Bruceton Analysis often applied where sampling is destructive so you cannot afford too much of it- imagine testing the yield of a hand grenade factory, or closer to home the correct blowing of fuses. Neyer d-optimal test  may be better for tests where the output is a number rather than a simple yes/no such as determining breakdown voltages with a given confidence for as few tests as possible.


This is all very pretty but really so much hot air for an EICR/PIR situation.


The problem is in applying the formal statistics to inspecting something like the set of all 13A  sockets on the ring is that they are not really a set of identical objects.  The one most used for vacuum cleaning is far more likely to be externally damaged, the ones in the kitchen are more likely to have been overloaded or wet, etc.


And even if that was not the case, we are not after a yield fraction in the stats sense, we want to find the failures, so we should look first at those locations most likely to be problematic- which I suspect experienced folk can do, because in practice once you find one serious problem, the 'how many more should I look at ?' answer changes (upwards), or at least it should.

After all, to condemn the circuit, you only need one killer fault.

M.

I think Graham's comment is not intended to be take literally (well I hope not). Clearly as the sample size is reduced, the risk of picking up an unrepresentative number of good or bad examples rises, and the indeterminacy  of the sample result versus what you would have seen with 100% sampling increases. (the so called 'law of small numbers')

This is not to say that you cannot do a sampled test on a small number of items, just that you need to be aware of the perils of doing so.

Mathematics only really comes to the rescue properly in the case of a supposedly identical set of samples, telling you your confidence/uncertainty  in relation to sampled fraction say the output of a production line.

 ( the canonical example has to be Bruceton Analysis often applied where sampling is destructive so you cannot afford too much of it- imagine testing the yield of a hand grenade factory, or closer to home the correct blowing of fuses. Neyer d-optimal test  may be better for tests where the output is a number rather than a simple yes/no such as determining breakdown voltages with a given confidence for as few tests as possible.


This is all very pretty but really so much hot air for an EICR/PIR situation.


The problem is in applying the formal statistics to inspecting something like the set of all 13A  sockets on the ring is that they are not really a set of identical objects.  The one most used for vacuum cleaning is far more likely to be externally damaged, the ones in the kitchen are more likely to have been overloaded or wet, etc.


And even if that was not the case, we are not after a yield fraction in the stats sense, we want to find the failures, so we should look first at those locations most likely to be problematic- which I suspect experienced folk can do, because in practice once you find one serious problem, the 'how many more should I look at ?' answer changes (upwards), or at least it should.

After all, to condemn the circuit, you only need one killer fault.

M.