My wife and I were in Firestone-Walker Brewery (Buellton, California) after Solvang Danish Days. (That’s me playing in the Solvang Village band.) My wife was comparing an Adam Firestone photo on the wall with a man at a table. I was admiring a woman seated near the bar with balletic posture. The balletic woman picked up a pizza and delivered it to the man and sat with him. My wife went over and asked the man if he was Adam Firestone? He was, with his sister Polly. While my wife chatted with them, I did not engage, because I was responsible for FORD recalling the Firestone tire sizes that Firestone did NOT recall.
August 2, 2000: Firestone recalled “…all ATX and ATXII tires of the P235/75R15 size (Ford Explorer) manufactured since 1991 and all Wilderness AT tires of that same size manufactured at Firestone’s Decatur, Illinois, plant.”
August 30, 2000: NHTSA recommended that Firestone expand the recall. Firestone declined. Investigation continued.
February 2, 2001: NHTSA was aware of 174 fatalities “alleged to be related to a tire failure.”
November 2000: Congress passed the TREAD Act [Public Law 106-414, HR 5164], requiring that NHTSA obtain and use insurance company reports. NHTSA asked insurance companies for “full VIN, personal identifiers, and specific crash information.” The insurance companies replied that providing such information would violate personal privacy. Stalemate.
2001: A friend survived a Firestone tire blowout in her Ford Bronco on highway 5 at 75 mph. I suspected the design defect that caused tire failures extended to other tire sizes. My tire reliability estimates showed that Firestone and Ford could have suspected tire problems in 1997. I shared my reliability estimates with a Ford statistician working on the Firestone tire problems.
May 2002: FORD recalled the other Firestone tire sizes!
What Data are Necessary?
The NHTSA Firestone complaints database included the tire failure date and the vehicle make, model, year, and personal identification (plus typos). Their database also indicated whether the tire was original equipment. Reliability analysis is easy, if you have life data.
I proposed that NHTSA provide early warning of automotive defects by using statistics that DO NOT require vehicle and personal identification. Generally accepted accounting principles (GAAP) require statistically sufficient data to estimate age-specific field reliability, expected complaints, and upper prediction limits. That data is vehicle counts, by year, make, model, and complaint counts by type. NHTSA’s TREAD Act insurance study found that insurance companies have such data.
Maximum likelihood and least squares nonparametric reliability estimates are available, from vehicle sales and complaints. Age-specific reliability estimates help detect exceptions, process shifts, improvements, or deteriorations attributable to calendar intervals. They help separate the effects of vehicle, make, model, tire type and size, or plant.
Imagine working for Firestone in the 1990s. In 1991 you would have had data from 1991; in 1992, you would have had data from 1992; etc. Figure 1 shows Kaplan-Meier reliability estimates, computed each year through 2000, aka a “broom chart”. Ward’s supplies vehicle counts and the NHTSA published annual complaint counts, for OEM tires.
Figure 1. Broom chart of tire reliability estimates from ships and complaints shows decreasing reliability. Longer lines are from older data.
Imagine working for Ford in the 1990s. You would have known Ford sales and repairs by VIN and symptom, which yield life data for the estimates in figure 1. But suppose some repairs were tire related problems? You could have made the reliability estimates in figure 2, from Ford sales and NHTSA complaints, by model (tire size). Figure 2 shows the least squares estimator for renewal processes [Chapter 6 of “Random-Tandem Queues…”]. Ask Fred or email@example.com if you want software in workbook or R-Script.
Figure 2. Age-specific reliability estimates by model from sales and complaints data in year 2001.
Estimate age-specific field reliability by calendar intervals, even if you don’t have life data: read the next Accendo Weekly Update for explanation. Make actuarial forecasts of failures and estimate upper prediction limits (UCL) as with control charts. If failures in some calendar interval exceed the UCL, take a sample, estimate reliability from age-at-failure data, search for root causes, and evaluate process improvements. Then revise forecasts, estimate risk under alternatives, act accordingly, and verify the improvements.
“Could Firestone and Ford Have Known?” ERI News, April 2001, Equipment Reliability Institute and SRE Lambda Notes, Vol. 34, Dec. 2001, pp 28-34
NHTSA TREAD Act Section 3(d) Insurance Study, http://www.nhtsa.dot.gov/cars/problems/studies/insurance/insreport4final.htm
Ward’s Automotive Yearbook, publication of wardcc.com
“Random-Tandem Queues and Reliability Estimation, Without Life Data,” https://docs.google.com/viewer?a=v&pid=sites&srcid=ZGVmYXVsdGRvbWFpbnxmaWVsZHJlbGlhYmlsaXR5fGd4OjU1NTQwNTJhNDkxNWJlNGM