06 August 2014

The GM switch decision

On May 29, Anton R. Valukas, a partner at the law firm Jenner & Block and a former federal prosecutor, delivered a 325-page report on the Chevy Cobalt ignition switch recalls to General Motors’ Board of Directors. The defective switches were apparently involved in at least 13 fatal crashes, yet it took GM almost 11 years to initiate a recall. (Some other GM models, such as the Saturn Ion, had the same defective switch, but I’ll refer to them all as “Cobalts.”)
The Valukas report is likely to be the best description we’ll have for some time on what went wrong at GM. Valukas’s team interviewed almost all the important players and read thousands of documents, and the story they tells is fascinating. Yet the report suffers from a number of flaws, most of which stem from its being put together by lawyers, for whom the main question seems to have been: Who was responsible? (The report resulted in the dismissal of 11 GM employees.)
The ignition switches installed in the Chevy Cobalt from its introduction in model year (MY) 2003 into MY 2007 were defective; they allowed the key to be turned on and off with so little force (torque) that the key could be moved to the accessory (ACC) position by a very light force, as when the key fob was brushed by the driver’s knee. When the key moved to ACC while the car was moving, the motor shut down and the steering and brakes, denied power assist, became heavy. Most critically, after approximately 0.15 seconds power was lost to the sensor for the airbags, so that they would not deploy in a crash. (Shutting off power to the sensor was a deliberate design choice, to prevent airbags going off while the car is parked.)
The GM engineer mainly responsible for the switch, Ray DeGiorgio, knew about the low-torque condition, which was outside GM’s established parameters, but approved the switch anyway. (Why the switch, manufactured by Delphi Mechatronics, had such low torque, and why DeGiorgio approved it, remain a mystery.)
Complaints about stalling caused by the key turning to ACC began to reach GM soon after the MY 2003 Cobalts hit the streets, but DeGiorgio and the other GM engineers working on the switch apparently did not know that when the key was in the ACC position the airbags would not deploy. They did know, however, that when a car loses power it remains controllable, so the engineers considered the ignition switch problem a “customer convenience” rather than a safety issue; they felt no pressure to solve the problem quickly.
That a car remains controllable when power is lost seems undeniable: Drivers lose power all the time—for example, when they inadvertently run out of gas—without the car becoming uncontrollable. (Of course, some Cobalt accidents apparently involved young women who had been drinking, for whom the heavier steering and braking might have been more problematic). So the switch didn’t cause the accidents, though it made them much worse by disabling a key safety feature. (The press and the public occasionally seem confused on this point; for example, the August 1, 2014 New York Times refers to “accidents caused by [GM’s] vehicles’ faulty ignition switches.”)
Soon other GM engineers began investigating reports of Cobalt airbags failing to deploy in crashes. There were at least 54 such incidents prior to the recall. However, these engineers were not aware for some time of the ignition switch connection, for two reasons.
First, the ignition switch was changed during MY 2007 to solve the low-torque problem, but the responsible engineer (DeGiorgio again) did not change the part number, contrary to GM policy. So in 2009 when the engineers looking at the airbag non-deployments noticed that they ceased for MY 2008 and later Cobalts, they reasonably concluded that the switch—which they thought had remained unchanged—couldn’t be the cause. Worse, when they asked DeGiorgio whether the switch had been changed, he replied that it had not. It is unclear why DeGiorgio (who was one of the dismissed employees) took these actions. (Interestingly, he bought a 2007 Cobalt for his son, but we don’t know whether it had the defective switch.)
The second reason the engineers missed the ignition switch connection was that the airbag sensor in some of the crashed Cobalts retained a record of the key position when power was lost, and in almost half of these cases the sensor showed the key to be in the RUN, not the ACC, position. In the report the engineers only offer one explanation for this anomaly, but the discussion is confusing, and no clear answer emerges.
If the sudden loss of power didn’t cause the crash—that is, the car remained controllable—then something happening between the driver’s losing control and the crash must have caused the key position to change to ACC. It was only in late 2013 that GM’s engineers finally concluded that the airbags would not deploy in a crash where the vehicle had several impacts before the final crash, as when it goes over a curb or rough terrain or hits small objects before hitting a larger object, with the first bumps causing the key to turn to ACC (a conclusion the report accepts; Valukas’s team did not commission any independent engineering analyses).
GM engineers had noticed that many of the airbag non-deployments involved off-road crashes, and by 2009 had come up with a theory—“contact bounce”—that jarring the ignition switch could “open up” the switch so that the signaling mechanism for the airbags would think the key was in the ACC position. The contact bounce theory was similar to the final theory, except it sought the cause in the internals of the switch rather than in the key position. The engineers conducted “abusive and teeth-chattering tests” in 2009 “in which the car was driven through steep ditches and deep potholes” without ever getting the switch to “open up.” The report doesn’t say, but presumably the engineers were testing Cobalts built after the ignition switch had been changed; otherwise the bouncing would likely have caused the key to rotate to ACC. Once more the failure to change the part number, and reassurances that the switch had not changed, seem to have misled GM’s investigators.
The report only discusses 13 non-deployment cases (not all fatal), presumably because they were the only ones with written evaluations from counsel. It’s not a representative sample, but for what it’s worth, three of the cases don’t fit the report’s theory, and a fourth isn’t described in enough detail to know if it fits. In two cases, there was no series of impacts; in one, a car was hit from the side in an intersection, in another the car rear-ended a tractor-trailer. In a third case, the car left the road going backwards; it’s not clear whether the airbag should have deployed or if it would have made any difference if it had. Finally, in a fourth case there is no indication whether the car left the road or had a second impact.
(One of the 13 fatalities described in The New York Times but not in the Valukas report also fails to fit the theory: a head-on crash with a drunk driver who crossed the highway center line. The air bag failed to deploy, but it’s hard to see how this could have resulted from the ignition switch problem.)
If the story really is as simple as “defective ignition switch disables airbags,” then by 2007 two outsiders had come up with the correct diagnosis: A Wisconsin state trooper at the site of a fatal crash noticed that the ignition key was in the ACC position and surmised that the airbags had failed to deploy for that reason, and an Indiana University Transportation Research Center report on that accident reached the same conclusion. Valukas faults the GM engineers for not being aware of the outsiders’ reports, which were in the public record, but it seems likely that even if the GM engineers had read the reports, they would have had little effect. For one thing, the outsiders got it right because, unlike GM’s engineers, they weren’t privy to the misinformation that the switch hadn’t been changed or to the fact that the sensors often showed the key to be in RUN position. The outsiders were only able to connect the dots because they had fewer dots to connect.
There’s another reason, barely alluded to in the Valukas report, why the GM engineers would be likely to discount information provided by outsiders: Many of these outsider were plaintiffs or their lawyers, who had a great deal to gain if an accident’s cause could be traced to a GM design defect. This is not new territory. The MY82-MY87 Audi 5000 model and various MY2000-MY2010 Toyotas were claimed by persons involved in accidents to have had “sudden acceleration.” Audi sales suffered, and Toyota initiated massive recalls and a $1.2 billion settlement, as a result of these claims. The National Highway Traffic Safety Administration later concluded that most, if not all, the sudden-acceleration crashes resulted from driver error. Outside information frequently fails to be objective.
In March 2012 GM engineers examining a crashed Cobalt at a junkyard noticed that the ignition switch turned extraordinarily easily. The engineers had not brought any tools with them to measure the torque, but using a fish scale purchased from a nearby bait and tackle shop (you can’t make this stuff up), they measured the torque for a number of Cobalt ignition switches in the yard. The torque on many switches was so low that they concluded that the key could turn to ACC if the car hit a pothole. The next day one of the engineers searched the Cobalt warranty database and discovered the numerous complaints about the ignition switch turning to ACC. He then elevated his concerns to more senior management.
But it would still be almost 20 months before GM finally decided to replace the defective switches. No doubt part of the problem was the inertia built into getting any large organization to take an expensive action. But perhaps a more important difficulty was that it remained unclear to several decision makers that the low-torque switch was the cause of the problem. And they had reasons.
In May 2012 the GM engineers visited the junkyard again and tested some 40 Cobalts. (This time they brought a torque wrench.) They found that earlier Cobalts required lower torque, but this was also true of some MY 2007 and MY 2008 models. And in spring or summer of 2012, DeGiorgio and his supervisor again stated that there had been no changes in the switch that would affect the torque.
Then, in April 2013, plaintiff’s attorneys took apart pre- and post-2007 switches and showed GM lawyers just how they had been changed: a plunger in the later switches was longer by about a millimeter, just enough to significantly increase the torque. But it was only on October 29, 2013 (one month before GM initiated the formal recall process) that Delphi confirmed that in 2006 DeGiorgio had approved the change to the plunger.
The report characterizes the ignition-switch saga as a “history of failures”: GM personnel “failed” to understand the connection between the switch position and airbag deployment, to search for GM or publicly available documents that would have pointed to the connection, to take apart the pre-and post-2008 switches to compare them, to alert key decision makers that the airbag nondeployments were causing fatalities, to demand action in the face of mounting fatalities, to make themselves or others accountable, and to marshal the expertise and information available to solve the problem.
The search for the cause of the non-deploying airbags involved many persons in many areas. All of these people had many other tasks to perform, and given their various backgrounds and responsibilities, it’s not surprising that the delay in organizing a recall stretched over many years. Leaving aside Ray DeGiorgio, whose actions arguably both created the problem and delayed its solution, it’s not obvious that anyone “failed.” Suppose that you could at any time have initiated a recall of the defective switches. Would you have done it much before GM did? Remember, there was a good deal of information and misinformation pointing to some other cause than low torque.
Valukas heaps scorn on the GM engineers’ search for the “root cause,” but without knowing the root cause, you couldn’t know that replacing the switches would work. Clearly, once you saw that the airbag non-deployments ceased in models with the new switch, you didn’t need to know the root cause (presumably, the too-short plunger) or why the sensor occasionally showed the key position to be RUN (an anomaly that is still unexplained). But to see that, you have to know that the key was changed in 2007, something the engineers couldn’t confirm until one month before the recall was initiated. (True, a plaintiff’s lawyer had shown them the two plungers some six months earlier, but you can understand the GM engineers discounting explanations from such a source.)
The actions of GM’s personnel might look like “failures” to a litigator, but for anyone familiar with large-organization decision processes when faced with technical problems, their actions were to be expected (again leaving aside DeGiorgio’s actions, which remain inexplicable). For example, we have the sociologist Dianne Vaughan’s book on the decision process leading to the 1986 loss of the space shuttle Challenger (The Challenger Launch Decision: Risky Technology, Culture, and Deviance at NASA). Vaughan presents the narrative twice, first as a journalist (or a lawyer) might give it, as a series of stupid and/or venal decisions that any sensible person would have seen as likely to lead to disaster. But the second account, told in greater detail, shows how the launch decision reflected years of experience with a process that everyone acknowledged was fraught with risk. Let me assure you that nothing you’ve ever done in your life is likely to have been as carefully thought through as NASA’s decision to launch Challenger. Then, at the end, Vaughan shows how the experience that the NASA engineers relied on, when organized in a different way, might have told them not to launch.
For those not up to Vaughan’s exhausting 592-page account, the lessons are summarized in the “Blowup” chapter of Malcolm Gladwell’s What the Dog Saw: And Other Adventures. The chapter’s tagline says it all: “Who can be blamed for a disaster like the Challenger explosion? No one, and we’d better get used to it.”
The Valukas report concludes with 18 pages of recommendations dealing with organizational structure, culture, training, and other matters intended to make a recurrence of these kinds of safety problems less likely. Most of them sound reasonable, but it’s not evident that they were reviewed by anyone with an expertise in management or organizational structure. They are, in fact, typical lawyers’ products, with an emphasis on written reports, formal meetings, and other methods familiar to persons involved with the court system. I doubt if many people on the Valukas team ever worked for a large for-profit company. Of course, Valukas’s firm, Jenner & Block, is a large organization, with 450 lawyers and perhaps 1,500 employees overall. The very largest law firms might have three times that number. But General Motors has 210,000 employees. Moreover, law firms are semi-feudal, with each partner having his own fiefdom built around his or her clients, arrangements that differ markedly from most other forms of enterprise. Jenner & Block lawyers may work with large clients, but their day-to-day law-firm experience doesn’t make them experts on the organization of large for-profit companies.
It’s not that nothing can be done, but doing something requires a lot of hard thought by people who understand the organization. What we tend to get instead are rituals like the Valukas report—a Who-Is-To-Blame? orgy of hindsight, concluding with untested sounds-good recommendations. The report is duly accepted, higher-ups make fulsome apologies, lower-level people are fired. Almost everybody feels better.

No comments:

Post a Comment