Originally Posted By: FutureDoc
Originally Posted By: sayjac
Says the poster that attempts to make the claim that the tears seen on this board are the results of, or close enough to a random sample to make a statistically valid and reliable conclusion regarding actual tear numbers. Laughable. Can't have it both ways and the Bitog anecdotes can never be classified as a random sample, let alone a large enough sample size.
Sampling a large fleet as suggested would be a much 'closer to' a random sample than the Bitog anecdotes you so passionately try to argue as being random 'enough'. Your now obvious bias against Purolator in this matter make your statements regarding random sampling in the matter highly suspect. It's true that one can make inferences on tear numbers from the Bitog anecdotes, but that is all. Inferences though are not conclusive to actual numbers.
As for the topic, Purolator took the time to respond and are aware of the issue and state they are attempting to make changes to address the issue(s). Clearly as shown in this thread for some that is not and will not be enough. Suggestion to those folks would be simply, change brands and move on.
I never thought I would be discussing statistical and research methods on a oil forum... oh well, here we go.
I never said that BITOG reporting would be a random sample. I have made it clear that we do not need a true random sample. Anyone with any background in applied statistical research or bio-stats would see the rational behind this.
We are dealing with reported problems and like my food poisoning example, we can compared reported cases to the expected number of cases of a given issue. The issue at hand is that the true number of failure is not going to be reported. Most folks do not cut open cans. Thus, when you see a string of reported failures (Say dozens over the course of a month) compared to maybe 1-3 over the course of the same time, THAT is the test you implement. All you need to do is test the number of reported failures against the number of expected failures in normal operations and that will be your valid statistical test.
To say that BITOG is not a valid data-source is kinda like saying that the people who bother going to the hospital is not valid for food poisoning. Some don't open cans, some people after eating rotten fish just chug a fifth a pepto and grunt it out. If you see a spike in the reported cases against the typical trend, then there is a issue at hand. You do not need to take a survey of the neighborhood. It is wasteful especially if the neighborhood is woefully ignorant of the issue.
If you "sample" a fleet, then all you can generalize about will be "for fleet use". A fleet vehicle does have different characteristics than a typical private automobile. It will skew towards models specific for that fleet. Now, if there is a problem, then that problem will show up but if there is no tearing, then... you lack the data. You can't conform a null.
FutureDoc, Ph.D. Former TA of Quantitative Analysis (applied statistics in a social science capacity)
I generally agree with this. When actual results spike against expected results, then it prompts an investigation.
I believe that manufacturers would apply the same testing to batches before they are released for sale. But oil filters are going to be harder to test since real operating conditions may not be adequately simulatable.
