Could Firestone and Ford Have Known?

by Larry George, March 30, 2001

http://www.equipment-reliability.com newsletter 2, March 2001 and Society of Reliability Engineers Lambda Notes

On August 2, 2000, Firestone recalled “…all ATX and ATXII tires of the P235/75R15 size manufactured since 1991 and all Wilderness AT tires of that same size manufactured at Firestone’s Decatur, Illinois, plant.” On August 30, 2000, the NHTSA recommended that Firestone expand the recall. Firestone declined. Investigation continues. As of February 2, 2001, the NHTSA was aware of 174 fatalities “alleged to be related to a tire failure.” Firestone tires were original equipment on some Ford and Mazda SUVs, light trucks, and pickups.

The NHTSA Firestone complaints database [1] includes the tire failure date and the vehicle make, model, and year. The difference between failure date and vehicle year is the age of the vehicle at tire failure. The complaints database also indicates whether the failed tire was an original one. Using vehicle production data [2], I made the nonparametric estimates of age-specific tire reliability in figure 1. The last data point for each year is the estimated reliability in the year 2000.

Figure 1. Age-specific tire reliability estimates of recalled tires by year of production

Reliability got worse in tires manufactured in 1996, but you’d have to wait to see the evidence of that. That is because you can’t estimate reliability at age five until tires are at least five years old. The NHTSA collected almost all the complaints in the year 2000, so the NHTSA couldn’t have made the estimates in figure 1 until late in the year 2000.

Table 1. Fatalities by Year of Tire Failure

Year	Fatalities
1991	0
1992	2
1993	0
1994	4
1995	2
1996	7
1997	9
1998	15
1999	34
2000	39

Presumably, lawyers contacted Firestone and Ford regarding these fatalities, so both companies would have had this data, perhaps in the years in which the tire failures occurred. Shouldn’t that have prompted corporate concern?

The risk management department of a major medical company told me, “Just don’t kill anybody.” They explained that their concern was risk, the number of opportunities for failure times the probability of failure per opportunity (unreliability) times expected cost per failure.

Firestone and Ford were probably concerned about the ominous trend in fatalities, at least in 1999. Would it have been possible to recognize problems earlier?

What If You Had Been Working for Firestone?

Imagine that you were working for Firestone in the 1990s. Firestone knows its ships and warranty returns, because generally accepted accounting principles require that data. Ships and returns are statistically sufficient for estimating age-specific field reliability [3], forecasting returns, and providing early warning.

In 1991 you would have had the data from 1991; in 1992, you would have had data from 1991 and 1992; and so on. You could have estimated reliability each year. Figure 2 shows the nonparametric least squares estimates, as if they had been computed each year from 1994 through 2000. (I used Ward’s vehicle ships and the NHTSA’s annual failure counts, for original equipment tires.)

Figure 2. Age-specific tire reliability estimates by year from ships and returns data

In the year 2000, problems were obvious, but could they have been recognized earlier? In 1997, the figure would have ended at the sixth year, with a reliability decrease from six nines (0.999999) to five nines (0.99999). That would have gotten my attention. That decrease in reliability could have been because

· Tires produced in 1991 started failing in 1996

· Tires produced in 1996 started failing in 1996

· Something in between occurred

That reliability decrease should have called for sampling, more analysis, and possible corrective action, if Firestone had been estimating tire reliability from ships and returns data. Firestone would have saved half their costs if the recall had been made in early 1997.

What If You Had Been Working for Ford?

Imagine that you were working for Ford in the 1990s. Ford knows its own production figures, and it tracks warranty repairs by VIN and symptom, which yields age data sufficient to make the estimates in figure 1. It is not necessary to track warranty repairs by VIN. That requires 1000 times as much data and incurs at least 1000 times as many errors.

Suppose some warranty repairs were due to tire related problems. You could have made the estimates in figure 3, which shows estimates made from ships and NHTSA complaints, year by year.

Figure 3. Age specific tire reliability estimates by vehicle make and model from ships and returns data

The last data point for each vehicle model is the estimated reliability in the year 2000. The Mountaineer went into production in 1996, and tire problems became evident in its second year, 1997. The Explorer began production in 1990, and tire problems became evident in its sixth year, 1997. These reliability estimates should have encouraged Ford to investigate tire related problems in 1997, if Ford had been estimating age-specific reliability, even from ships and returns data.

Conclusions and Recommendations

Age-specific field reliability can be estimated without tracking parts and products by serial number. It also helps detect exceptions, process shifts, improvement, or deterioration attributable to calendar intervals. It helps separate the effects of vehicle, make, model, tire type, or plant.

If you’re trying to control risk like that experienced by Firestone and Ford, estimate age-specific field reliability by calendar intervals. Make actuarial forecasts of returns and put upper confidence limits (UCL) on the forecasts. Use them like control charts. If returns 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. Revise forecasts, estimate risk under alternatives, and act accordingly.

References

· Http://www.nhtsa.dot.gov/hot/Firestone/complaints.xls, Feb. 2, 2001 version. The analyses reported in this article are based on the Dec. 6, 2000, version.

· Ward’s Automotive Yearbook, annual publication of Ward’s Communications.

· L. L. George, “Measure Field Reliability With Statistics,” Equipment-Reliability Newsletter, Jan. 2001.

Mark Felthauser, CCI/Triad, helped with the statistical analyses. Eva Langfeldt, Text Support, edited the article. I am grateful to them for their thoughtful contributions. Contact them through Larry, pstlarry@yahoo.com or eva@textsupport.net.