Originally Posted By: FutureDoc
Ok, lets talk experimental research statistics (really empirical research design). If you wanted to ask "does Purolator tear more than BrandX or x, y, and/or z, then yes, you would want to do a controlled experiment with the same conditions. That is because the brand itself would be the independent variable we want to deem significant. However, because this is not a "comparison between brands" but rather an issue of Purolator quality control/application. Is the Hudson more polluted that other rivers is a different question and design that asking if the Hudson River is polluted. You just test the item in question. Furthermore, the same "controlled" method might not return as rich of a dataset. "Fieldwork" is useful. You can't always do some work in controlled labs and return the same data as compared to the field. Think about running 70-100 sample test of all Purolator stock (including jobbers) on all vehicle makes from the last 25 years, on all motor oil weights and brands. Not really feasible. Worse if you do a dozen or so driving types, climates, commutes, etc. Thus when ask "are tears still and issue, small data-sets are still useful. You kinda answered the issue yourself.
Well, I think you have to compare it to different filters on equal grounds, don't you? If you see 100/1000 Purolators tear (all of these numbers are completely arbitrary for discussion purposes), then sure, you can say that Purolators tear. You have observed that happening. But what about the other brands? If failure rates are equal in Wix then who cares about Purolator? If you observe the same trend in each brand then you would just have to conclude that it is the nature of the beast when it comes to oil filters and tears. As far as I can tell it would be the only conclusive way to determine if Purolator has a brand-specific issue, or if tears in media are par for the course. At that point even small sample sizes (such as those reported here) would be useful in establishing a trend provided each brand has been observed equally.
Originally Posted By: FutureDoc
So if a few dozen that actually open the filter notice an issue (with some having back-to-back filter failures), there is a problem. From a stat standpoint, I might have an equal chance of winning the lottery than getting back-to-back filter failures. If it was 1 in 10,000, then back to back filters with different production dates should be 1 in 100,000,000. Getting multiple failure by such a small subset (BITOG), that 1 in 10,000 failure rate means that of the BITOG population, would have to had purchased and opened 610,000 Purolator filter to reach that mark in 6 or so months. If it was even 1 in 100, you should see 549 "everything is ok" filter threads for the 61 reported failure in the last year.
Is there something of a known rate for the failure of filters in the industry? I'm legitimately asking as I was under the impression that it was unknown. If that's the case then at least there is a baseline to compare to, in which case I would agree with you that seeing so many failures on BITOG is abnormal.
And just to clarify, I didn't think media tears were even a slightly common occurrence so there probably IS something going on at Purolator (poor media, quality control, whatever). But it's still just an educated guess at this point without any real concrete numbers. This is a bit of a reach, but if you flip a coin 100 times and it comes up heads 75 times, it doesn't mean that heads is more/less likely to occur on the next flip. It just means that you observed heads 50 times more than tails. After 1,000 flips it could be 75 heads and 925 tails. Unlikely, sure, but not impossible. Same thing with the filter tears being observed here. It could just be a coincidence.
And with that said now I feel like kind of a nitwit for going this in depth on the subject, haha. Filters aren't supposed to tear and if they are, it probably means something is wrong. Which is enough for me because I don't really care... I'll just use M1. It has been a bit of a slow day today