P Values - In understandable terms what does the P Value represent?

[cpine]

Looking for information on P values. As every SPC program I have been involved in does not reflect a P value, I am a little concerned that I just had a PPAP rejected for a low P value. In understandable terms what does it represent?

[Jim Wynne]

Sounds like someone was doing some (perhaps inappropriate) hypothesis testing on your PPAP data. The p-value is a probability estimation, given as a number between 0 and 1, which essentially tells the likelihood of the means of two sets of data being different due to chance alone. Values closer to 0 indicate smaller probabilities. If your customer is complaining about a low p-value, it could mean that he took two different sets of data from your PPAP submission (say, your SPC data and your MSA data) and thinks that the difference in the means between the two datasets is significant--that is to say, that they probably didn't occur due to just random sampling differences. It's another way of saying (and I'm still surmising here) that they might think that one or both of your datasets were fudged, or were otherwise funky.

[Benjamin28]

We have some statistics wizards on this site so I'm sure you'll find a pretty solid explanation. I wanted to clarify though, you're talking about a p-chart for spc and you want to know what the P value stands for?

"A p chart plots the portion of each sample that is defective. For example, if 25 out of 100 parts are defective, the p value would be 25/100=0.25. If we consider that individual parts are measured as zero for defective or one for ok, then those measurements follow a binary distribution. The p values in effect are means of averages of individual part measurements, and thus follow a binomial distribution."

However, if that's what your customer is refering to I don't know why you'd get rejected for having a low value as low would be the desirable result.

I wonder if you could explain further what portion of your PPAP your customer was referring to when they suggested a low " P -value" ?

[cpine]

In Reply to Parent Post by Jim Wynne

Sounds like someone was doing some (perhaps inappropriate) hypothesis testing on your PPAP data. The p-value is a probability estimation, given as a number between 0 and 1, which essentially tells the likelihood of the means of two sets of data being different due to chance alone. Values closer to 0 indicate smaller probabilities. If your customer is complaining about a low p-value, it could mean that he took two different sets of data from your PPAP submission (say, your SPC data and your MSA data) and thinks that the difference in the means between the two datasets is significant--that is to say, that they probably didn't occur due to just random sampling differences. It's another way of saying (and I'm still surmising here) that they might think that one or both of your datasets were fudged, or were otherwise funky.

Thank you so much for getting back to me so fast.

The customer is saying just the opposite. his statement is that the lower the P value the higher the probability my reported values are .005. The customer states that anything less than a .05 is suspect. I'm incline to agree with you however I have no formal reference to argue.

To make matters worse he is also suggesting that are gaging method is inadequate. I know the rule of thumb is to use a gage that has a resolution 10 X the part feature and thats all and well however when your dealing with tenths and millions it is near impossible to find a gage that has a .00001 resolution. Are there any acceptions to this rule?

Once again thank you so much for your help.

[Miner]

I think that your customer is saying the same thing as what Jim said. In hypothesis testing, a p-value less than 0.05 is significant. However, we do not have any information to put this into context.

If you can provide additional information and actual data, we can provide more specific help. Right now, we are speculating and winging our responses.

[Jim Wynne]

In Reply to Parent Post by cpine

Thank you so much for getting back to me so fast.

The customer is saying just the opposite. his statement is that the lower the P value the higher the probability my reported values are .005. The customer states that anything less than a .05 is suspect. I'm incline to agree with you however I have no formal reference to argue.

Perhaps my response wasn't clear; what I meant to say was that lower p-values indicate a higher probability that there is something non-random going on. A p-value of 0 would mean, theoretically, that there is no chance that some phenomenon x was due to random causes, while a p-value of 1 would indicate that it was certain that only random causes were at work. In actual practice the number will almost always be >0 and <1, however.

While I don't know anything about the data your customer was evaluating, or whether his methods were kosher, if his hypothesis testing were on the up-and-up and resulted in a p-value of .005, it's pretty clear that something is wrong with your results. If you can give more information about the data being evaluated, we can be more helpful.

In Reply to Parent Post by cpine

To make matters worse he is also suggesting that are gaging method is inadequate. I know the rule of thumb is to use a gage that has a resolution 10 X the part feature and thats all and well however when your dealing with tenths and millions it is near impossible to find a gage that has a .00001 resolution. Are there any acceptions to this rule?

If there are exceptions, the customer makes them--it's his rules you're dealing with. What's the specification, and what are you using to measure it? This is another of those instances where contract review would have been helpful; part of the APQP process is reviewing the specifications to make sure that you have the proper gaging in place. In a case like this, where you're not sure whether what you're using is adequate, it's best to ask the customer what they use to measure the feature in question, and to duplicate that method.

In Reply to Parent Post by cpine

Once again thank you so much for your help.

Glad to help--that's why we're here.

[cpine]

Please find attached for your review and discussions a copy of the file sent to the customer (minitab).

I hope this helps.

Again,Thank you.

Chuck

[BradM]

Are you talking about where it states:

AD 5.107 P<.005?

I am not familar with Minitab, so I am unfamilar with the AD test. However, generally the p value along side a test is like Jim states. The test statistic and the p value go together. The test statistic will show a significant (or non-significant) finding utilizing that test, and the p value is the probability of that significance being random, and not due to the measured influence of interest.

But, this is a P, and not a p. So I am not sure if they are the same.

As far as the gauging resolution, Jim is correct. Does your customer have a specific requirement for that?



I would think your customer would be more concerned with your outliers on your R-Chart