Enforcing Random Sampling

Ilya S.

Registered
Hi,

We are currently experiencing issues with our random sampling. We have the same sample sizes as our customer, however we don't find any bad parts, while they are finding rejects. Have any of you guys faced this kind of issue? Any ideas on how to enforce random sampling? Or do you have a good method for picking out parts for a random sampling?
 

Miner

Forum Moderator
Leader
Admin
Do you have first-hand evidence that the samples are not random, or is this an assumption? Are your customer's samples random? While convenience sampling is definitely a possibility, there may also be:
  • Differences in the measurement devices (e.g., calipers vs. micrometers, vs. pin/ring gages)
  • Differences in how the measurements are made (e.g., random diameter vs. max diameter vs. mean diameter)
  • Differences in how the requirements are interpreted, particularly with visual criteria
Most samples taken at a final inspection after product has been packaged are typically convenience samples because it is usually difficult or disruptive to unpack then repack the product. Early in my career, I worked for a tier 1 supplier of automotive weatherstripping. The product was coiled into a large box, which took two people to lift. These boxes were then stacked three high on a pallet. One inspector could only sample parts from the top box, and only from the top layer because they could not lift a box and could not pull a sample from lower layers without removing all of the layers above then repacking them all.
 

japayson

Involved In Discussions
Why do you think there is anything wrong with your sampling? What you described can certainly happen. I don't think there should be panic on this occurrence, but I have seen it happen (panic that is)., then someone comes to the quality manager and asks "how could we send bad parts?" Because sampling only shows a small portion of the lot. If you make any bad parts some will get through your sampling plan. This might be picked up in a second (the customer's) sampling. Or not. You might look into the "red beads" demonstration presented by Deming and others. One thing that could be wrong is that your sampling and inspection plan may not be adequate for the results you want. This depends upon your actual defect rate versus the operating curve of your sampling plan.
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
If you are concerned, you could raise your sample sizes. You should at least take a look at if the "true" failure rate is what the customer is getting, what is the probability with your sample size that you would not detect a failure.

But it is important to communicate with the customer. Is the customer using the same "opertational definition" of a "failure" as you are? If there is a difference there, then you will get different results.

Perhaps the customer us doing some form of "Smart Sampling" that is picking out items likely to be failures. Maybe they have detected something in screening the items that you have not. Has the customer sent back the defective items (or can you obtain them) to see what the failures were.
 

SeanN

Involved In Discussions
Hi,

We are currently experiencing issues with our random sampling. We have the same sample sizes as our customer, however we don't find any bad parts, while they are finding rejects. Have any of you guys faced this kind of issue? Any ideas on how to enforce random sampling? Or do you have a good method for picking out parts for a random sampling?
It could be that you use simple random sampling while they use stratified sampling. there is no right or wrong here. It depends on the specific objectives and characteristics of the population you are sampling from. But stratified sampling is beneficial when the population exhibits heterogeneity, then stratified sampling ensures that all subgroups are represented in the sample. This approach can be more efficient in capturing variations or rare occurrences in specific subgroups.
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
Good point on the stratified sampling, which could be considered a form of "smart sampling" when it is effective.
 

John Predmore

Trusted Information Resource
I have seen cases where parts become defective after inspection, in transportation for example if the temperature in the railcar gets too hot or the truck hits a bad pothole. One idea would be have your inspectors mark parts they inspect in an inconspicuous manner, a green dot in the corner of the label or a fillet radius, perhaps. Then, when your customer returns defective parts they find, you would know whether those returns were previously inspected by your people.
 
Top Bottom