Strikes, Scabs, and Tread Separations: Labor Strife and the Production of Defective
Bridgestone/Firestone Tires
Author(s): Alan B. Krueger and Alexandre Mas
Source: The Journal of Political Economy, Vol. 112, No. 2 (April 2004), pp. 253-289
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/10.1086/381479 .
Accessed: 28/01/2011 14:15
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=ucpress. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Political Economy.
http://www.jstor.org
253
[Journal of Political Economy, 2004, vol. 112, no. 2]
᭧ 2004 by The University of Chicago. All rights reserved. 0022-3808/2004/11202-0006$10.00
Strikes, Scabs, and Tread Separations: Labor
Strife and the Production of Defective
Bridgestone/Firestone Tires
Alan B. Krueger
Princeton University and National Bureau of Economic Research
Alexandre Mas
Princeton University
This paper provides a case study of the effect of labor relations on
product quality. We consider whether a long, contentious strike and
the hiring of replacement workers at Bridgestone/Firestone’s Decatur,
Illinois, plant in the mid-1990s contributed to the production of defective
tires. Using several independent data sources and looking before
and after the strike and across plants, we find that labor strife at
the Decatur plant closely coincided with lower product quality.
Monthly data suggest that defects were particularly high around the
time concessions were demanded and when large numbers of replacement
workers and returning strikers worked side by side.
Do workers provide more effort and due diligence if they feel they are
treated better? Does the labor relations climate affect the quality of
production? Economic models of fairness (e.g., Rabin 1993; Fehr and
Ga¨chter 2000) suggest that workers are more cooperative and less prone
We are grateful to Colin Cameron, David Card, Henry Farber, Seth Harris, Alain Kornhauser,
Paul Rosenbaum, Chris Sims, Jim Sturm, and seminar participants at Princeton
University, Stanford Law School, University of California at Berkeley, Massachusetts Institute
of Technology Sloan School, the NBER, the biennial Hong Kong Economic Association
meeting, the University of Pennsylvania, North Carolina State University, the Stockholm
University Institute for International Economic Studies, and the Society of Labor
Economists for helpful comments.
254 journal of political economy
to commit sabotage if they feel they are treated well and in good faith.
In addition, the dependence of worker effort on pay is an essential
feature of efficiency wage models. Yet a relationship between worker
treatment and the quality of production has proved difficult to establish.
Quality is often unobserved or hard to measure. This paper provides
new evidence on the impact of labor strife on the quality of production
at the plant level by examining the incidence of defective Bridgestone/
Firestone tires.
In August 2000, Firestone and Ford jointly announced the recall of
14.4 million size P235/75R15 ATX, ATX II, and Wilderness AT tires,
some 6.5 million of them still on the road, mostly on Ford Explorers.1
The National Highway Traffic and Safety Administration (NHTSA) issued
an advisory concerning several other sizes and models of Firestone
tires in September 2000. The NHTSA reported that the Firestone tires
under investigation were related to 271 fatalities and more than 800
injuries. The most common source of failure of the recalled tires was
tread separation, that is, a sudden detachment of the rubber tread from
the steel belts, causing the tire to blow out. One of every 400 tires
produced in the Decatur, Illinois, plant in 1995 was returned under
warranty because of a tread separation by 2000.2
As described in detail in Section I, three of Firestone’s 11 North
American tire production plants, including its Decatur plant, which
manufactured a large number of P235/75R15 tires, underwent a severe
strike beginning in July 1994, after Bridgestone/Firestone proposed
deviating from the industrywide pattern bargain by moving from an
eight- to a 12-hour shift that would rotate between days and nights, as
well as cutting pay for new hires by 30 percent. Almost immediately
after 4,200 workers walked out on strike, the company hired replacement
workers. A final contract, which included provisions to recall all
strikers, was not settled until December 1996.
A unique vista into the possible effect of labor relations on product
quality is possible in this instance because the recall of Firestone tires,
congressional hearings, and scores of liability lawsuits have made confidential,
proprietary data publicly available. Tires are still made in large
part by hand, so there is scope for human error in producing this
1
The recall pertained to size P235/75R15 Firestone ATX and ATX II tires made in all
Firestone plants from 1991 to 2000 and P235/75R15 Wilderness AT tires produced in the
Decatur, Ill., plant from 1996 to 2000. On May 22, 2001, Ford announced plans to replace
all Wilderness AT tires on certain Ford vehicles. In October 2001, Firestone recalled
additional P235/75R15 tires and P255/70R16 Wilderness AT tires.
2
This figure is based on our calculations using warranty data provided by Bridgestone/
Firestone to the NHTSA. It is an underestimate of the overall incidence of tread separations
because not all tires were covered by the warranty program at the time of a tread separation,
and some owners may not have returned covered tires under the warranty program despite
having a tread separation.
labor strife 255
product. In addition, because millions of tires are produced and are in
service each year, failure rates can be calculated for an enormous sample.
Laboratory tests by Firestone and Ford have been unable to identify
a single manufacturing or design defect that is responsible for the high
incidence of defective tires. A variety of different types of defects were
probably at work. Is it possible that the labor dispute affected the safety
of the tires? A number of observers—congressmen, plaintiffs’ attorneys,
and reporters—have hypothesized that undertrained replacement workers
or lax supervision during the strike contributed to the tire defects.
It is also possible that discord among replacement workers, union members
who crossed the picket line, and returning strikers contributed to
the production defects. And workers may have been fatigued and more
prone to errors because Firestone introduced a 12-hour rotating shift
to operate the plant 24 hours a day during the strike.
At least five other hypotheses have been proposed as the source of
defective tires. First, Bridgestone/Firestone executives blamed the defects
in part on the design of the Ford Explorer, which they argued was
prone to roll over. Second, Bridgestone/Firestone also argued that Ford
recommended that the air pressure of the tires be set at 26 pounds per
square inch (psi), whereas the tire manufacturer recommended 30 psi.
At lower pressures, tires become hotter and are more prone to blow
out. Third, others have conjectured that features of the manufacturing
process in the Decatur plant increased the risk of safety defects, including
a shortening of the vulcanization process (time in which tires
are cooked to imprint the tread) and plant conditions that may have
allowed moisture to seep into the rubber linings. Fourth, some tire
experts and workers have conjectured that faulty material inputs, such
as outdated rubber, weakened tires manufactured in Decatur. Finally,
some have claimed that the design of the tires was responsible for the
high rate of tread separations.
Anecdotal evidence is available for all these explanations. In Sections
III–V, we try to systematically sort through the possible reasons for the
safety-related defects, focusing closely on whether the production of
defective tires coincided with the labor dispute. We examine three different
data sets and provide five types of evidence. We focus mostly on
P235/75R15 tires because they have more data available, but we look
at other tires as well. Almost all the P235 tires were produced in three
plants: Decatur, Illinois; Joliette, Quebec; and Wilson, North Carolina.
For nearly three years—from April 1994 to December 1996—union
workers at the Decatur plant either were on strike or were working
without a contract; tires were produced by 1,048 replacement workers,
union members who crossed the picket line, management, and recalled
strikers in this period. The Wilson plant was nonunion, so it did not
experience a strike. A Canadian union represents the Joliette plant, but
256 journal of political economy
labor relations there were much less contentious. Joliette had a sixmonth
strike over fringe benefits at the end of 1995, but the plant did
not hire replacement workers (which are illegal in Quebec).
The evidence that we have assembled suggests that the strike and
associated labor strife in Decatur were major contributing factors to the
production of defective tires. First, descriptive analysis and count regression
models using data on claims for compensation for property
damage or personal injuries show significantly higher defect rates for
tires manufactured in Decatur during the labor dispute (defined as
1994–96) than in Decatur at other times or in other plants, conditional
on production and other variables. Second, NHTSA data on complaints
involving Firestone P235/75R15 tires indicate a similar pattern: a significantly
higher incidence of complaints filed concerning tires manufactured
in Decatur during the strike period than at other times or at
other plants. Third, engineering tests conducted by Firestone—which
hold constant conditions such as speed, load, tire pressure, and ambient
temperature—indicate a higher failure rate for tires manufactured in
Decatur during the strike compared with tires manufactured in Decatur
during non–strike years or in other plants.
Fourth, we more closely examine the timing of the production of
defective tires using data on the number of claims disaggregated by the
month the tires were produced. This analysis finds an excess number of claims
for tires produced in Decatur in the few months before the contract
expired, when Bridgestone/Firestone demanded concessions, and in
the period in which many replacement workers and recalled union
workers worked side by side. This finding leads us to a somewhat nuanced
conclusion on the role of replacement workers: it is not simply
that undertrained or poorly supervised replacement workers produced
defective tires. Instead, the timing suggests that the concurrence of
replacement workers and union members working side by side before
the contract was settled, as well as labor strife in the months leading
up to the strike, coincided with the production of a high number of
defective tires.
Fifth, we examine other models of Firestone tires, which have not
been recalled by Firestone. The same workforce produced these tires,
so labor strife might have affected their quality. Results for these models
also indicate a higher defect rate for tires produced in Decatur during
the strike than in Decatur before or after the strike or in other plants.
Although these findings suggest that labor strife contributed to the
problems with Bridgestone/Firestone tires, it is unclear whether one
can generalize from our results to other settings because tire manufacturing
may be different from other industries and because the investigation
into the problems with P235 tires made our case study possible
in the first place, so “selection bias” may be an issue. Our findings were
labor strife 257
not preordained, however, because at the start of our work we were
unaware that an excess number of defective tires were produced during
the dispute period in the Decatur plant, that the precise timing would
indicate a major role for union workers as opposed to replacement
workers, or that there was an unusual number of problems with nonP235
tires manufactured in the Decatur plant during the dispute. Still,
we would recommend that the reader exercise caution in extending
our results to other settings; our paper provides a detailed case study
of only one firm in one unique period of its history.
I. Background and Industrial Relations Environment
In 1988, the Japanese tire manufacturer Bridgestone purchased Firestone,
making Bridgestone/Firestone the largest tire manufacturer in
the world. Initially, labor relations went smoothly. For example, Bridgestone
upheld Firestone’s agreement to allow its new plant in Warren
County, Tennessee, to organize by means of a card sign-up, forgoing a
secret ballot election. In 1991, the three-year contract negotiated between
Bridgestone and the United Rubber Workers (URW) became the
pattern for the tire industry, and the company and union established a
much ballyhooed “Partnership for Involvement” program.3
When negotiations
began in January 1994, however, Bridgestone/Firestone
sought to deviate from the industrywide practice of pattern bargaining,
asking for what a company negotiator called “a turnaround agreement.”
Since the 1940s, Firestone plants had adhered to the industrywide pattern
bargain. Peter Schofield, Firestone’s former director of labor relations,
said that the January 1994 bargaining session “certainly didn’t
make the union happy” (Meyer 2002, p. 337).
Bridgestone/Firestone demanded that the union move from eighthour
to 12-hour shifts that rotated between day and night, that it operate
the plant seven days a week, that the pay of new hires be reduced by
30 percent (although the cut would be made up by faster seniority
increases over a three-year period), that hourly workers contribute to
their health care costs, that the piece rate system be moved to a plantwide
performance pay system, and that vacations be cut by two weeks
for senior workers. This request for concessions came at a time of falling
national unemployment and just after Bridgestone/Firestone’s first profitable
year since Bridgestone acquired Firestone (Whitford 1994). The
URW proposed that Bridgestone/Firestone follow the master pattern
3
The material in this paragraph draws heavily from Franklin (2001) and Meyer (2002).
258 journal of political economy
agreement set with Goodyear, which called for no wage increases other
than cost-of-living adjustments.4
After the contract expired on April 1, 1994, union members continued
working without an agreement. On July 12, 1994, URW locals representing
4,200 workers in three of Bridgestone/Firestone’s production
plants (Des Moines, Decatur, and Oklahoma City), the Akron, Ohio,
research center, and the Noblesville, Indiana, air springs plant went on
strike. Bridgestone almost immediately began to hire temporary replacement
workers and operate the plants with managerial and supervisory
workers. Replacement workers were paid 30 percent less than the
union rate (Franklin 2001, p. 140). By January 1995, when the company
converted temporary replacement workers to permanent replacement
workers, it had hired a total of 2,300 replacement workers (p. 140).
The union set up pickets outside the struck plants. In Decatur, the
union posted the names of each member who crossed the picket line
under the title “Hall of Shame.” The list included a previous president
of the local union (Franklin 2001, p. 127). More than 300 of the 1,209
members of Local 713 in Decatur eventually crossed the picket line. By
May 1995, the Decatur plant employed 1,048 replacement workers and
371 permanent workers.5
The company initiated 12-hour rotating shifts
and operated the plant 24 hours a day.
On May 7, 1995, members of the URW Local 713 in Decatur voted
by a 2–1 margin to officially end their strike against Bridgestone/Firestone.
The other plants quickly followed suit.6
On May 22, the URW
unconditionally offered to return to work. According to Local 713 President
Roger Gates, the union offered to unconditionally end the strike
as a strategy to prevent Bridgestone/Firestone from hiring additional
replacement workers and to forestall a union decertification election
(Cimini 1996). The strike was the longest in the history of the URW;
its strike fund was depleted. Bridgestone/Firestone unilaterally imposed
its last offer from July 1994, which cut wages for some job classifications
by $5.34 an hour, to about $12 per hour; moved piece rate pay to a flat
hourly rate plus plantwide performance pay; paid new hires 30 percent
less than before the strike; froze pension benefits; and required workers
to contribute a portion of their health care premiums (pp. 26–27). The
company also continued operating 12-hour shifts.
Although the strike officially ended in May 1995, the labor dispute
continued. The company notified the union that it intended to per-
4
The previous agreement at Goodyear also expired in April 1994, and the new contract
was settled as soon as the old one expired.
5
Statistics included in a letter submitted by Theodore Hester on behalf of Bridgestone/
Firestone, contained in U.S. House (2000, p. 107).
6
In January 1995 the local representing the research facility in Akron had voted to end
the strike.
labor strife 259
manently retain the replacement workers and that it would “advise the
URW of the number of employees for whom we have work” (statement
of Bridgestone/Firestone spokesperson Trevor Hoskins [Bureau of National
Affairs 1995, pp. A6–A7]). In July 1995, the financially bankrupt
URW voted to merge into the United Steel Workers of America (USWA).
The URW ceased to exist. The USWA made the rehiring of all striking
workers a priority for future negotiations, but Bridgestone/Firestone
and the USWA did not reach a new agreement until December 1996.
On July 18, 1995, the USWA announced a worldwide boycott of Bridgestone/Firestone
and Sears Roebuck, its largest domestic retailer. The
union said it initiated the boycott because the company insisted on
retaining replacement workers instead of recalling some 2,000 strikers
who were willing to return to work (Cimini 1996).7
The USWA broke
off contract talks in January 1996 because Bridgestone/Firestone would
not agree to a system of pattern bargaining.
A final settlement, which included an agreement to recall all strikers,
was ratified in December 1996. The new contract gave an across-theboard
wage increase of 35 cents compared to the 1994 levels, gave a
$750 bonus, and restored pension benefits. Hourly wages were about
$1.00 below the rate at Goodyear (Sabath 1997). A 12-hour shift remained
in place, but it no longer rotated between day and night for
the same worker.
Most strikers had been recalled prior to the final contract agreement.8
By early November 1996, all but 40 strikers were recalled. Bridgestone/
Firestone increased the operation of its plants to 24 hours a day, seven
days a week, during the strike, so total employment increased. Because
Bridgestone/Firestone gradually recalled strikers after the union unconditionally
agreed to return to work, replacement workers and returning
strikers worked side by side for several months before a new
contract was reached. From the union’s perspective, working conditions
in this period were described as “brutal.”
Greenhouse (1996, p. A16) observed that “organized labor made the
dispute one of its crusades largely because it involved one of the largest
uses of permanent replacement workers in the nation’s history.”9
7
Among other activities, the USWA staged an anti-Firestone protest at the Indianapolis
500.
8
John Lampe, Bridgestone/Firestone’s executive vice president, testified that “by 1996,
over half of our work force in the Decatur plant was unionized workers that had gone on
strike and come back, and less than half were then replacement or temporary workers”
(U.S. House 2000, p. 1391).
9
There are many other notable examples in which labor strife has arisen during strikes
in which replacement workers were hired. Examples include the Hormel strike of 1985–
86 in Austin, Minn. (Green 1990); the Iowa Beef Packers strike of 1987 in Dakota City,
Iowa (Craypo 1994); the Phelps Dodge Corp. strike of 1983 in Phelps Dodge, Ariz. (Rosenblum
1995); and the International Paper Strike of 1988 in Jay, Maine, which coincided
with toxic gas leaks leading to the evacuation of the town (Bierecree 1991).
260 journal of political economy
TABLE 1
Total Production of Size P235/75R15 Tires, Bridgestone/Firestone Plants,
January 1991–March 2000
Plant
Number of
ATX and
Wilderness
Tires
Percentage
of Total
Claims per
Million ATX
Tires Produced Notes
Aiken, S.C. 1,442,115 6.9 NA Plant began operation
in 1999
Decatur, Ill. 6,408,584 30.5 356.0 Strike from July 1994
to December 1996;
1,048 replacement
workers
Joliette, Que. 5,638,302 26.9 58.9 Strike from August
1995 to February
1996; no replacement
workers
La Vergne, Tenn. 455,157 2.2 NA No strike
Oklahoma City, Okla. 530,515 2.5 44.1 Strike from July 1994
to December 1996
Wilson, N.C. 6,503,642 31.0 142.0 Nonunion; no strike
Total 20,978,315 100.0
Source.—Production is derived from a Safety Forum spreadsheet. Claims per million are taken from a chart submitted
by Firestone to the House Committee on Commerce; they pertain to property damage and personal injury claims for
ATX size P235/75R15 tires.
Note.—P235/75R15 tires include ATX, ATX II, and Wilderness AT tires. Aiken produced only Wilderness tires.
Tire Production
Column 1 of table 1 reports the number of ATX, ATX II, and Wilderness
size P235/75R15 tires produced in each of the six Firestone plants that
manufactured such tires from 1991 to 2000. (Henceforth, we use P235
to refer to size P235/75R15 tires and ATX to refer to both ATX and
ATX II models.) The Decatur, Joliette, and Wilson plants produced the
vast majority of these tires. Wilderness tires were not produced until
1996. A smaller number of ATX tires continued to be produced after
Wilderness tires were introduced. Column 4 summarizes the labor relations
situation in the 1990s.
Tire manufacturing remains a highly complex, labor-intensive task.
According to Firestone’s former chief executive officer, Masatoshi Ono,
“A typical tire can have more than 26 components, 14 different rubber
compounds and require 29 separate steps to manufacture” (quoted in
an August 23, 2000, press release [http://www.bridgestone-firestone
.com/news/mediacenter/news/000823a.htm]). The Decatur plant is
Firestone’s oldest manufacturing facility. The plant was originally used
by Caterpillar to build large engines for the U.S. military during World
War II. It was next used as a U.S. Army signal depot until 1961. In 1962
Firestone purchased the 160-acre site and converted it to produce tires.
Local 713 of the URW immediately organized it. Franklin (2001, p. 117)
describes the plant as “dark, brooding, and smoky in places, a noisy
labor strife 261
monument to the almost unchanging way most tires are still made: with
much reaching and lifting and cutting, a lot of human muscle power,
and an endless demand for unflinching robot-like concentration.” Particularly
in the Decatur plant, tire production was not automated. For
example, workers had much discretion over the amount of effort exerted
to wrap the steel belts, the condition of the rubber used to produce the
tread and sidewall, and the implementation of the vulcanization process.
There is much anecdotal evidence that replacement workers were less
skilled tire makers and were perhaps more prone to build defective
tires. According to William Newton, a retired tire builder from Decatur,
“It takes two years to become a good tire builder.” When Mr. Newton
returned to work after the strike, he said he “saw a lot of people who
didn’t know how to build tires” (quoted in Barboza [2000, p. C1]). Joe
Roundtree, another retired worker from Decatur, reportedly testified
in a civil case that he doubted the ability of replacement workers to
match the quality of experienced union workers and that he thought
moving to 12-hour shifts compromised product quality (Fogarty 2000).
Other employees have testified that they were ordered to remove
blisters from tire sidewalls by piercing them with an awl, a small, handheld
pointed instrument, and that production pressures during the labor
dispute impaired quality. “I protested it, but I done it,” Jan S. Wagoner
said. “I didn’t like sticking an awl all the way through the sidewall
of a tire” (Skertic 2000). The Decatur plant manager, Harry McMillan,
testified that the plant “was not producing the volume of tires it was
expected to produce [during the dispute], and waste and scrap levels
were higher than the company expected them to be” (quoted in Pinkerton
[2001b]). Tires are scrapped if a defect is detected. “They
preached quality,” Lonnie Dart, a retired Decatur worker testified, “but
if you didn’t make the numbers, you was in trouble” (Skertic 2000).
According to one news account, production quotas made thorough
inspections nearly impossible (Barboza 2000).
Tire Safety
Tires are more likely to fail if they run at higher speeds or at higher
temperatures or carry heavier loads. Older tires are also more likely to
suffer a tread separation. A large majority of tread separations occurred
in the southwestern part of the United States, almost certainly because
of the higher average temperature there. We are not aware of any evidence
indicating a change in the regional distribution of Firestone tires
produced in different plants, but the location of tires that are installed
as original equipment, as opposed to replacement tires, is determined
by the auto manufacturer.
One possibly relevant development was that Ford began producing
262 journal of political economy
Explorers at a new plant that opened in St. Louis, Missouri, in 1995,
and many of these vehicles were sent to dealers in the Southwest. The
other Explorer plant was in Louisville, Kentucky. Firestone’s Decatur
plant likely supplied tires to both plants because of its proximity. There
are several reasons why we doubt that the temporal pattern of defective
tires originating from Decatur that we document below was caused by
a change in distribution to warmer states during the dispute, however.
First, the St. Louis plant began producing in 1995, but there was already
an excess number of defective tires produced in Decatur in early 1994.
Second, our results are robust if we limit the analysis to replacement
tires, which are unaffected by the new Ford plant. Third, the excess
number of defects from Decatur abated after 1996, while the St. Louis
plant continued to operate. Fourth, and perhaps most important, engineering
tests that hold constant temperature and other environmental
factors point in the same direction as the field data.
Historically, Firestone has monitored tire safety by running product
tests, studying warranty adjustment information, and analyzing failed
tires that were returned from the field. Another source of information
is claims paid for property damages and personal injuries. According
to congressional testimony (U.S. House 2000, p. 137), the company did
not closely monitor these data for safety purposes until July 2000, in a
joint effort with Ford.10
Column 3 of table 1 reports the number of claims for property damages
or personal injuries involving Firestone ATX tires. It is clear that
the Decatur plant has a much higher rate of claims per million tires
produced. A key question is, Was the claims rate elevated for tires made
during the labor dispute?
II. Firestone Claims and NHTSA Complaints Data Sets and
Descriptive Analysis
We obtained Firestone’s internal claims data from the NHTSA. These
data consist of a brief description of claims filed before 2000 for property
damage or personal injuries stemming from tire failures. About 70 percent
of the claims involved problems related to tread separations. The
data we analyze include claims concerning 2,597 P235 tires produced
between 1991 and 1999. The data set includes micro data on tire model
and size, date of failure, and the 10-digit Department of Transportation
(DOT) tire code, from which plant and week of production can be
inferred.
10
In a deposition for a civil suit in Texas, Allen Rauner, a Ford engineer, testified that
he requested claims data in 1999 from Firestone but was told that the company did not
have such data (Pinkerton 2001a).
labor strife 263
We use these data to construct a cell-level data set, where the cell is
a production year by age by plant. We also have added information for
each cell on the number of P235 tires produced (by model), the share
of tires that were used as original equipment, and the share of tires that
are ATX model, which were all provided by Safety Forum, a nonprofit
organization that works with trial attorneys involved in lawsuits against
Firestone and Ford.
The second set of data on failures was collected by the NHTSA.11
Briefly, the NHTSA recorded information reported by consumers, police
authorities, insurance companies, and other organizations on Firestone
tires that failed. The NHTSA data also include information on the tire
model and size, date of failure, and DOT tire code. We use these data
to construct a cell-level data set on the number of complaints by year
of production, plant, and age. One advantage of the NHTSA data set
is that it indicates whether the tire failure resulted in a fatality or injury,
allowing us to calculate failure rates depending on the severity of the
incident. Another advantage is that we can analyze data on P235 and
non-P235 tires. And unlike the claims data, the complaints data indicate
whether the tire was installed as original equipment or as a replacement
tire.
There are three major shortcomings with the NHTSA complaints data,
however. First, the records contain many missing values for the variables
of interest. Of 5,189 reported complaints (on all tire sizes), only 1,076
have complete data, 3,524 have partial data, and 586 are missing data
on the production year, plant, and calendar year of failure of the tire.
Second, many complaints were filed after the recall was announced and
may involve routine use-related failures, such as running over a nail.
Third, there is no financial incentive to report complaints to the
NHTSA, as opposed to claims data, so underreporting is probably common.
We limited the data to complaints filed before the August 2000
recall to avoid an avalanche of frivolous complaints or complaints that
were sparked because the recall singled out Wilderness tires made in
Decatur. Before the recall, it was not publicly known that Decatur tires
might have a higher defect rate.
To handle the missing values, we multiply imputed missing data using
the conditional distribution for the available data using a technique
based on Rubin (1987). If, say, a randomly chosen half of all complaints
had missing data, we could just double the number of complaints with
complete data. But this would not make use of the partially reported
data, and we could probably do better filling in the missing data by
11
The complaints micro data set is publicly available from the NHTSA. It began compiling
the complaints regarding Firestone tires in February 2000 and ceased in June 2001.
The data are described in detail in the data appendix to Krueger and Mas (2002).
264 journal of political economy
conditioning on the information that we do have. For example, if a
tire’s age was missing, we randomly assigned the age from the set of
tires with the same year of production and vehicle miles.
Although it is possible to utilize a more sophisticated technique to
impute missing values, given the strong likelihood that data are randomly
missing (e.g., a tire was so badly damaged that the DOT code
was not visible), we implemented a simple, transparent hot-decking technique
that relied on a priori expectations of which variables contain
information on the missing values. We created 10 micro complaints data
sets imputing missing values in this way and then aggregated each micro
data set into plant by year of production by age cells. We report summary
statistics for the average of the 10 cell-level data sets. The advantage of
proceeding in this way (as opposed to assigning conditional expected
values to missing data) is that the standard errors that we compute take
into account sampling variability introduced by the imputation
method.12
The assumption that is required for the consistency of the
estimates using both imputed and nonimputed data is that the missing
data are missing at random conditional on the variables used in the
imputation. It is reassuring to note that none of our conclusions is
qualitatively altered if we use only the nonmissing observations.13
Although both the claims and complaints data sets have limitations,
they complement each other and have the virtue of being independently
collected. For example, while the claims data are more complete and
claimants have a financial incentive to file a claim, they lack information
on the severity of resulting injuries. Moreover, there is no industrywide
standard definition of a claim, so the claims data rely on Firestone’s
interpretation and recording of claims.
Descriptive Analysis
Figure 1 reports the cumulative number of claims against Bridgestone/
Firestone per million ATX tires produced by the year the tires were
produced and the age of the tires. The left panel displays data for the
Decatur plant and the right panel for Joliette and Wilson, the two other
main producers.14
The three years of the labor dispute—1994, 1995,
and 1996—stand out in the figure for Decatur. Four years after pro-
12
Specifically, standard errors are calculated by taking the square root of the sum of
the average sampling variance of the estimates in each of the data sets and the variance
of the estimates across the 10 imputed data sets (to account for variability due to imputation);
see the data appendix in Krueger and Mas (2002).
13
Estimates using the sample of complaints without imputed data are available on
request.
14
Ford submitted a similar graph in congressional testimony, although it pertained just
to tread separations and combined all plants other than Decatur (see U.S. House 2000,
p. 153).
labor strife 265
Fig. 1.—Claims per million Firestone ATX tires produced, by age of tire at failure,
production year, and plant. Source: Authors’ calculations based on Firestone claims data
submitted to the NHTSA. Claims pertain to personal injury or property damage involving
Firestone ATX tires prior to 2000.
duction, tires that were made in Decatur during the labor dispute were
at least 15 times more likely to have resulted in a claim than tires manufactured
in other plants. Such a large discrepancy in failure rates is
not evident for other production years, although the claims rate for
Decatur tires was about double that for other tires for the 1993 production
year.
Figures 2 and 3 report similar figures using the complaints data for
ATX and Wilderness size P235 tires combined. Figure 2 uses only the
subsample with complete data for year, plant, and age, and figure 3 uses
the average of 10 data sets with allocations for missing data, as described
previously. The number of complaints is about half as large as the number
of claims, even after imputations are made for missing data. Nonetheless,
both figures tell a similar story, and a story similar to the claims
data: tires produced during the labor dispute (1994–96) at the Decatur
plant have a much higher failure rate than those produced at Joliette
or Wilson, although before and after the dispute period the rate of
complaints is similar for tires manufactured in Decatur and in the other
plants.
Figure 4 displays cumulative incidence rates for cases in the NHTSA
database involving at least one fatality.15
That is, we report the cumulative
15
The figure uses imputed data for missing values. The results are quite similar if records
with incomplete data are eliminated.
Fig. 2.—Complaints registered by the NHTSA per million P235/75R15 Firestone tires
produced, by age of tire, production year, and plant (nonimputed data). Source: Authors’
calculations based on NHTSA and Safety Forum data. Complaints concern defects in 1991–
99, reported prior to August 2000. Observations with missing data are excluded.
Fig. 3.—Complaints registered by the NHTSA per million P235/75R15 Firestone tires
produced, by age of tire, production year, and plant (average of 10 imputed data sets).
Source: Authors’ calculations based on NHTSA and Safety Forum data. Complaints concern
defects in 1991–99, reported prior to August 2000.
labor strife 267
Fig. 4.—Number of accidents with fatalities per million Firestone P235/75R15 tires
produced, by age of tire, production year, and plant (average of 10 imputed data sets).
Source: Authors’ calculations based on NHTSA and Safety Forum data. The sample includes
ATX and Wilderness P235/75R15 tires. Accidents occurred between 1991 and 1999
and could have been reported at any time after 1991.
number of recorded tire failures linked to a fatality per million tires
produced, by year of production, plant, and age of tire. This sample
was selected because fatalities are unlikely to be undercounted in the
NHTSA data and because records for cases involving fatalities were less
likely to have missing data. Even with a small number of counts, these
graphs convey a similar message: a higher rate of fatal accidents involving
tires produced in Decatur during the labor dispute than tires produced
in other plants or tires produced in Decatur during a more tranquil
period of labor relations. To put the numbers in perspective, note that
before the recall, tires made in Decatur during the labor dispute had
a fatal accident rate of 10–30 per million tires produced. As a benchmark,
we note that the risk of death from parachuting is estimated at
13 per million jumps (Poynter 1992).16
16
Because tires are driven on frequently over several years, these statistics admittedly
compare apples and oranges to dramatize the risk from tire failure. A common but still
unsatisfactory alternative is to put both risks per mile traveled. One way to put the figures
reported in the text in a comparable context, however, is to note that the risk of a fatal
tire blowout is reduced by changing one’s tires, and each car has four tires with more or
less mutually exclusive risks, so the reduction in risk for a driver from replacing P235 tires
manufactured during the labor dispute surely exceeded the risk reduction from refraining
from one parachute jump.
268 journal of political economy
III. Count Data Models
To model tire failures conditional on other variables, we estimate a series
of count data regression models. Specifically, we model , the numberyijt
of claims (or complaints) for tires produced in plant i in year j that
failed in calendar year t, as
E(y Fx ) p exp (x b), (1)ijt ijt ijt
where the x-vector of explanatory variables includes the logarithm of
tires produced in plant i in year j, a dummy variable indicating tires
manufactured in Decatur, a dummy indicating tires manufactured during
the labor dispute (defined as 1994–96), an interaction between the
Decatur dummy and labor dispute dummy, a quadratic in tire age, and
possibly other variables such as the fraction of tires in the cell that were
installed as original equipment on new vehicles. The vector of parameters
is denoted b.
The “law of rare events” holds that the total number of events of some
kind that occur in a time interval will approximately follow a Poisson
distribution if a single event may occur in any of a large number of
trials, but the probability that the event occurs in any one trial is small.
Tire failures would seem a natural to model as a Poisson process because
they are rare events and there are a large number of trials, with hundreds
of thousands of cars driving on tires made in the same plant and year.
A testable restriction of the Poisson model is that the conditional mean,
, equals the conditional variance, . Even if this re-E(y Fx ) Var (y Fx )ijt ijt ijt ijt
striction does not hold, however, Poisson maximum likelihood estimates
of b are still consistent if equation (1) is a correct specification, although
the standard errors are inconsistent.
A common feature of count data—and one that we find in our tire
failure data—is that the conditional variance exceeds the conditional
mean. As a consequence, to avoid understating the standard errors, we
compute robust standard errors that are consistent in the presence of
overdispersion (see Cameron and Trivedi [1998, pp. 63–65] on robust
Poisson). Specifically, we allow
Var (yFx ) p a exp (x b), a 1 1,i i i
where a is an overdispersion parameter. The robust standard errors thus
equal the usual Poisson standard errors multiplied by the square root
of a. We estimate a by
n 2
ˆ(y Ϫ y )i iϪ1
ˆa p (n Ϫ k) .͸ ˆyip1 i
In addition, we present estimates of negative binomial models for our
first set of specifications. The negative binomial model does not impose
labor strife 269
TABLE 2
Means of Data Used in Aggregate Analysis of ATX and Wilderness AT P235/
75R15 Tires (Np135)
Claims Complaints*
Unweighted
Means
(1)
Weighted
Means
(2)
Unweighted
Means
(3)
Weighted
Means
(4)
Number of claims
or complaints
12.10
[29.37]
… 5.81
[10.88]
…
Log production 13.11
[.71]
13.37
[.56]
13.11
[.70]
13.35
[.55]
Proportion ATX .84
[.32]
.90
[.19]
.84
[.32]
.89
[.21]
Proportion original
equipment
.62
[.24]
.52
[.28]
.62
[.24]
.53
[.28]
Age 2.67
[2.22]
3.83
[1.40]
2.67
[2.22]
3.55
[1.54]
Age2
12
[15.47]
16.63
[12.02]
12
[15.47]
15.24
[12.21]
Decatur dummy
(1pyes)
.33
[.47]
.57
[.50]
.33
[.47]
.53
[.50]
Dispute period
dummy, defined
as 1994–96
(1pyes)
.33
[.47]
.62
[.49]
.33
[.47]
.63
[.49]
Decatur#dispute
period
.11
[.32]
.50
[.50]
.11
[.32]
.47
[.50]
Source.—Authors’ calculations.
Note.—Standard deviations are in brackets. Weighted means are weighted by the number of claims (col. 2) or
complaints (col. 4), thus giving the average characteristics of tires that failed. In col. 2 there were 24 cells with a weight
of zero (i.e., no claims), and in col. 4 there were 43 cells with a weight of zero (i.e., no complaints).
* Contains imputations for missing values.
the restriction that the mean equal the variance and is also consistent
if equation (1) holds.
Table 2 reports means of the two cell-level data sets we use to estimate
count models. Columns 1 and 2 present summary statistics for the Firestone
claims data set and columns 3 and 4 for the NHTSA complaints
data set. In both data sets, we have combined data on ATX and Wilderness
tires. The dependent variable , shown in the first row, is the(y )ijt
total number of failures in either ATX or Wilderness P235 tires in a
particular plant by year of production by calendar year. Cells in the
claims and complaints data sets pertain to three plants (Decatur, Joliette,
and Wilson) by nine production years (1991–99) and nine calendar
years (1991–99). Columns 1 and 3 present unweighted means, and columns
2 and 4 use as weights the number of claims or complaints in the
cell. Thus columns 2 and 4 provide information on the characteristics
of tires that failed.
Complaints are only about half as common per cell as claims. This
finding probably reflects underreporting of complaints in the NHTSA
270 journal of political economy
data. Nevertheless, the characteristics of the failed tires are quite similar.
In both data sets, for example, more than 90 percent of the failed tires
are ATX models, reflecting the fact that ATX tires have a higher failure
rate and are older (and therefore more likely to fail) in these samples.
The average failed tire is around 3.5 years old at the time of the incident.
About half of the failed P235 tires were produced in Decatur during
the period of the labor dispute (1994–96); by comparison, only 9 percent
of all P235 tires manufactured from 1991 to 1999 were produced in
Decatur in this period.
Table 3 presents Poisson models for the Firestone claims data (cols.
1–3) and for the NHTSA complaints data (cols. 4–6). The first model
conditions on log production and includes dummy variables indicating
tires made in Decatur, tires made during the labor dispute in any plant,
and the interaction between the two. The second model also conditions
for the age and age squared of the tire. The third model adds controls
for the proportion of tires produced in the plant that year that are ATX
model and the fraction of tires that were installed as original equipment.
In all the models, tires produced in Decatur during the labor dispute
are found to have a statistically significantly higher rate of claims or
complaints. First consider the results for claims. In column 3, for example,
the Decatur-dispute interaction has a coefficient of 1.48, with a
t-ratio of 4.93. This coefficient is a “difference-in-differences” estimate
in that it measures the difference in claims between the dispute period
and nondispute period in Decatur less the difference over the corresponding
time periods in the other plants, conditional on the other
variables. To interpret the magnitude of the interaction effect, note that
, which implies that the incidence of claims is 339exp (1.48) p 4.39
percent higher for tires produced in Decatur during the labor dispute,
other things held constant. The Decatur main effect is on the margin
of being significant at the .05 level in column 3, suggesting that, before
and after the dispute, the claims rate was somewhat higher for tires
manufactured in the Decatur plant than for those manufactured in
Joliette and Wilson. This differential is driven by a higher claims rate
in Decatur than in the other plants after the dispute, which is possibly
a lingering effect of the dispute. The dispute period dummy is insignificant,
suggesting that tires from Joliette and Wilson did not result in
an increase in claims if they were manufactured during the labor
dispute.
The coefficient on the log production variable can be interpreted as
a “defect scale” elasticity. Interestingly, in the models in columns 2 and
3, the number of claims rises more than in proportion with the number
of tires produced, suggesting that failures exhibit “increasing returns to
scale,” perhaps because of (accelerating) fatigue in the production
process.
labor strife 271
TABLE 3
Poisson Regression Models for Number of Defective P235/75R15 Tires by Plant,
Production Year, and Calendar Year: Aggregate Analysis of ATX and
Wilderness AT Tires (Np135)
Explanatory
Variable
Dependent Variable: Claims
Dependent Variable:
Complaints*
(1) (2) (3) (4) (5) (6)
Constant Ϫ8.71
(2.69)
Ϫ22.45
(1.73)
Ϫ18.69
(2.06)
Ϫ8.26
(2.34)
Ϫ15.60
(1.78)
Ϫ12.78
(2.17)
Log production .80
(.20)
1.49
(.11)
1.27
(.13)
.72
(.18)
1.08
(.12)
.92
(.14)
Proportion ATX .19
(.48)
.52
(.49)
Proportion original
equipment
Ϫ1.40
(.56)
Ϫ1.58
(.65)
Age 2.09
(.17)
2.16
(.17)
1.17
(.13)
1.22
(.13)
Age2
Ϫ.21
(.02)
Ϫ.21
(.02)
Ϫ.11
(.02)
Ϫ.11
(.02)
Decatur (1pyes) Ϫ.14
(.45)
1.31
(.26)
.65
(.33)
Ϫ.36
(.43)
.43
(.30)
Ϫ.15
(.38)
Labor dispute
period
(1pyes)
Ϫ.10
(.37)
.25
(.19)
.07
(.19)
.13
(.29)
.38
(.19)
.20
(.19)
Decatur#dispute
period
2.40
(.57)
1.39
(.29)
1.48
(.30)
2.29
(.50)
1.73
(.33)
1.71
(.35)
Mean dependent
variable
12.10
[29.37]
12.10
[29.37]
12.10
[29.37]
5.81
[10.88]
5.81
[10.88]
5.81
[10.88]
Pseudo 2
R .40 .81 .82 .42 .68 .69
Log likelihood Ϫ1,417.21 Ϫ452.82 Ϫ436.66 Ϫ559.28 Ϫ304.93 Ϫ294.43
Note.—Robust standard errors are in parentheses. Standard deviations of the dependent variables are in brackets.
The dependent variable corresponds to incidents per year of production, plant of production, and calendar year of
failure. Incidents in the NHTSA data set (cols. 4–6) correspond to reporting periods prior to August 2000 for failures
occurring between January 1991 and December 1999. Incidents in the Firestone claims data set (cols. 1–3) occurred
between January 1991 and December 1999, and claims were made before January 2000. Proportion original equipment
denotes the proportion of tires produced in the cell that were original equipment for Ford, Mercury, or Mazda vehicles.
Proportion ATX denotes the proportion of tires produced in the cell that were ATX tires. To adjust for incomplete
reporting in the NHTSA data set, missing values were imputed by assigning values from other observations with complete
data, conditional on available data. Ten imputed data sets were generated. Reported estimates were obtained by averaging
the estimates derived from each imputed data set. Standard errors were calculated by taking the square root of the
sum of the average sampling variance across data sets and the variance of the estimates across the 10 imputed data sets
(to account for variability due to imputation). Other summary statistics reported for cols. 4–6 are the average of the
estimates across the imputed data sets.
* Contains imputations for missing values.
Surprisingly, the proportion of tires in a cell that are ATX models
has a small and statistically insignificant, though positive, effect on the
incidence of claims. The proportion of tires installed as original equipment
has a negative coefficient, which in this case is statistically
significant.
Insofar as the labor dispute is concerned, the results for complaints
are similar to those for claims. The Decatur-dispute interaction is highly
statistically significant, with a t-ratio of 4.89 in column 6. The model in
column 6 indicates that tires made in Decatur during the labor dispute
were associated with times more complaintsexp (1.71 ϩ .20) p 6.75
272 journal of political economy
than tires made in Decatur before or after the dispute. Similarly, tires
manufactured in Decatur during the strike were associated with
times more complaints than tires manu-exp (1.71 ϩ .20 Ϫ .15) p 5.81
factured in Wilson or Joliette in nondispute years. The number of complaints
concerning tires produced during the dispute in Joliette and
Wilson is insignificantly higher than that for tires produced there in
other years. And it is worth noting that we cannot reject that the plant
main effect and dispute interaction are equal for Wilson and Joliette if
we estimate a model that allows their coefficients to differ.
The production variable has close to a unit elasticity in the complaints
models in columns 5 and 6. The coefficient on proportion new equipment
is negative in the complaints equation and statistically significant.
For both dependent variables, the coefficients on the age variables
display an inverted U–shaped pattern, with the number of failures peaking
when tires are around five years old. Note that we have made no
adjustment to production for tires that are no longer in service. Older
tires are more likely to be removed from the road, so the age variables
in part reflect obsolescence. We could adjust production by assuming
some depreciation rate each year. If we include unrestricted age dummies,
however, they would completely absorb any (cross-plant) age-specific
obsolescence rate because of the logarithmic specification of production.
When we included unrestricted age dummies, the coefficients
on production and the Decatur and labor dispute dummies were hardly
changed, and the pattern of age effects indicated that the quadratic
specification of age fit the data reasonably well. For example, if the
model in column 3 of table 3 is estimated with unrestricted age dummies,
the coefficient on production is 1.28 (vs. 1.27) and the coefficient
on the Decatur-dispute interaction is 1.45 (vs. 1.48). Because the results
are so similar if age dummies are included, we have used the more
parsimonious quadratic specification.
If we limit the sample of complaints to those involving replacement
tires—which are not subject to the regional distribution decisions of the
auto manufacturer—the results are very similar. For example, the coefficient
on the Decatur-dispute interaction in the model in column 5
of table 3 rises from 1.73 to 2.30 (with a t-ratio of 4.6) if the analysis is
limited to spare tires. The other coefficients are also qualitatively similar.
This finding suggests that possible changes in Ford’s distribution practices
are unlikely to account for the results in table 3.
As mentioned previously, the Poisson models impose the constraint
. A generalization of the Poisson, the neg-Var [yFx] p E[yFx] p exp (x b)
ative binomial II model, specifies the conditional variance as
while maintaining equation (1). If ,Var [yFx] p (1 ϩ l) exp (x b) l p 0
the negative binomial is equivalent to a Poisson. A common test for
overdispersion is a likelihood ratio test of whether . Table 4 presentsl 1 0
labor strife 273
TABLE 4
Negative Binomial Regression Models for Number of Defective P235/75R15
Tires by Plant, Production Year, and Calendar Year: Aggregate Analysis of
ATX and Wilderness AT Tires (Np135)
Explanatory
Variable
Dependent Variable: Claims
Dependent Variable:
Complaints*
(1) (2) (3) (4) (5) (6)
Constant Ϫ2.31
(2.40)
Ϫ19.95
(2.11)
Ϫ16.55
(2.19)
Ϫ6.50
(2.24)
Ϫ14.86
(1.83)
Ϫ11.62
(2.12)
Log production .35
(.18)
1.34
(.14)
1.15
(.14)
.59
(.17)
1.04
(.13)
.86
(.14)
Proportion ATX .27
(.55)
.42
(.51)
Proportion original
equipment
Ϫ1.71
(.72)
Ϫ1.77
(.72)
Age 1.91
(.17)
2.01
(.18)
1.07
(.13)
1.14
(.13)
Age2
Ϫ.19
(.02)
Ϫ.19
(.02)
Ϫ.10
(.02)
Ϫ.10
(.02)
Decatur (1pyes) Ϫ.13
(.33)
1.12
(.31)
.37
(.38)
Ϫ.04
(.33)
.43
(.30)
Ϫ.26
(.38)
Labor dispute
period
(1pyes)
Ϫ.17
(.29)
.24
(.21)
Ϫ.03
(.23)
.23
(.26)
.45
(.19)
.23
(.20)
Decatur#dispute
period
1.32
(.48)
1.46
(.34)
1.60
(.36)
1.62
(.41)
1.69
(.33)
1.74
(.36)
Mean dependent
variable
12.10
[29.37]
12.10
[29.37]
12.10
[29.37]
5.81
[10.88]
5.81
[10.88]
5.81
[10.88]
Pseudo 2
R .02 .23 .24 .07 .23 .25
l 43.49†
6.30†
5.81†
8.07†
1.61†
1.47†
Log likelihood Ϫ391.44 Ϫ307.89 Ϫ303.40 Ϫ335.29 Ϫ277.30 Ϫ270.45
Note.—Standard errors are in parentheses. Standard deviations of the dependent variables are in brackets. The
dependent variable corresponds to incidents per year of production, plant of production, and calendar year of failure.
Incidents in the NHTSA data set (cols. 4–6) correspond to reporting periods prior to August 2000, for failures occurring
between January 1991 and December 1999. Incidents in the Firestone claims data set (cols. 1–3) occurred between
January 1991 and December 1999, and claims were made before January 2000. Proportion original equipment denotes
the proportion of tires produced in the cell that were original equipment for Ford, Mercury, or Mazda vehicles.
Proportion ATX denotes the proportion of tires produced in the cell that were ATX tires. To adjust for incomplete
reporting in the NHTSA data set, missing values were imputed by assigning values from other observations with complete
data, conditional on available data. Ten imputed data sets were generated. Reported estimates were obtained by averaging
the estimates derived from each imputed data set. Standard errors were calculated by taking the square root of the
sum of the average sampling variance across data sets and the variance of the estimates across the 10 imputed data sets
(to account for variability due to imputation). Other summary statistics reported for cols. 4–6 are the average of the
estimates across the imputed data sets. The symbol l represents the scaling parameter for the conditional variance in
the negative binomial model. The conditional variance of the negative binomial model equals the conditional variance
of the Poisson model when .l p 0
* Contains imputations for missing values.
†
We can reject the null hypothesis that at the .05 level with the likelihood ratio test.l p 0
maximum likelihood estimates of negative binomial regression models
corresponding to those in table 3. Both the claims and complaints data
exhibit overdispersion. This causes the standard errors on many of the
coefficients to roughly double in the negative binomial model compared
to a conventional Poisson model. The parameter estimates are fairly
similar in the negative binomial regressions, however, as one would
expect because both models are consistent if equation (1) holds.
274 journal of political economy
TABLE 5
Poisson Regression Models for Number of Defective P235/75R15 Tires by Tire
Model (ATX or Wilderness), Plant, Production Year, and Calendar Year; Pooled
Sample (Np164)
Explanatory
Variable
Dependent Variable: Claims
Dependent Variable: NHTSA
Complaints*
(1) (2) (3) (4) (5) (6)
Constant Ϫ13.36
(3.11)
Ϫ25.57
(2.20)
Ϫ21.09
(2.22)
Ϫ9.38
(2.52)
Ϫ13.61
(2.39)
Ϫ11.35
(2.01)
Log production 1.15
(.23)
1.72
(.15)
1.44
(.15)
.79
(.19)
.94
(.17)
.80
(.14)
ATX (1pyes) .57
(.27)
.60
(.29)
Proportion original
equipment
Ϫ1.64
(.47)
Ϫ1.48
(.51)
Age 2.07
(.18)
2.12
(.18)
1.17
(.16)
1.22
(.15)
Age2
Ϫ.20
(.02)
Ϫ.21
(.02)
Ϫ.11
(.02)
Ϫ.11
(.02)
Decatur (1pyes) Ϫ.01
(.47)
1.63
(.31)
.87
(.36)
Ϫ.36
(.45)
.29
(.39)
Ϫ.26
(.39)
Labor dispute
period
(1pyes)
Ϫ.02
(.39)
.32
(.21)
.12
(.20)
.13
(.31)
.34
(.25)
.17
(.22)
Decatur#dispute
period
2.32
(.58)
1.26
(.32)
1.25
(.30)
2.27
(.52)
1.82
(.41)
1.81
(.37)
Mean dependent
variable
9.96
[25.62]
9.96
[25.62]
9.96
[25.62]
4.61
[8.68]
4.61
[8.68]
4.61
[8.68]
Pseudo 2
R .42 .81 .82 .40 .66 .67
Log likelihood Ϫ1,483.96 Ϫ490.20 Ϫ461.93 Ϫ591.45 Ϫ339.36 Ϫ323.08
Note.–Robust standard errors are in parentheses. Standard deviations of the dependent variables are in brackets.
The dependent variable corresponds to incidents per year of production, plant of production, tire model, and calendar
year of failure. Cells corresponding to fewer than 200 units of production were dropped. Incidents in the NHTSA data
set (cols. 4–6) correspond to reporting periods prior to August 2000, for failures occurring between January 1991 and
December 1999. Incidents in the Firestone claims data set (cols. 1–3) occurred between January 1991 and December
1999, and claims were made before January 2000. Proportion original equipment denotes the proportion of tires
produced in the cell that were original equipment for Ford, Mercury, or Mazda vehicles. To adjust for incomplete
reporting in the NHTSA data set, missing values were imputed by assigning values from other observations with complete
data, conditional on available data. Ten imputed data sets were generated. Reported estimates were obtained by averaging
the estimates derived from each imputed data set. Standard errors were calculated by taking the square root of the
sum of the average sampling variance across data sets and the variance of the estimates across the 10 imputed data sets
(to account for variability due to imputation). Other summary statistics reported for cols. 4–6 are the average of the
estimates across the imputed data sets.
* Contains imputations for missing values.
In table 5 we present additional Poisson models in which the unit of
observation is disaggregated into finer cells, delineated by ATX or Wilderness
tires as well as year of production, plant, and calendar year of
failure. Again we adjust the standard errors for overdispersion. We pool
the ATX and Wilderness cells together in the sample and include a
dummy variable indicating whether the cell pertains to ATX or Wilderness
tires. These models provide a more precise assessment of
whether ATX tires have a higher failure rate than Wilderness tires.
Indeed, the results for both claims and complaints indicate a substantially
higher failure rate for ATX tires, other things equal. The other
labor strife 275
results are fairly similar to those for the more aggregate cells in tables
3 and 4.
Incidents with Fatal and Nonfatal Injuries: NHTSA Data
Table 6 presents estimates of Poisson models using the NHTSA complaints
database; we provide separate estimates for the number of cases
involving at least one fatal accident, number of cases involving a nonfatal
injury, and number of cases without a reported injury. The unit of
observation is a cell defined by plant by year by age, as in tables 3 and
4. Because we are less concerned about reporting biases in cases involving
injuries or fatalities, we use the full reporting period to calculate
counts in this table. We continue to limit the sample to tires that failed
in 1999 or earlier, however. Results for the unallocated data, which are
quite similar, are in Krueger and Mas (2002).
Interestingly, the coefficient on the Decatur–dispute period interaction
is positive and statistically significant—and of roughly equal magnitude—in
all the models. Likewise, the Decatur and dispute period
main effects are statistically insignificant in all the models. To estimate
the excess number of fatal incidents involving tires manufactured in
Decatur during the labor dispute, denoted D, we calculated
D p S [exp (x b ϩ S a ϩ Dg ϩ S # D d) Ϫ exp (x b ϩ S a ϩ Dg)],ijt ijt j i j ij ijt j i
(2)
where is a vector of explanatory variables such as log productionxijt
and proportion original equipment, Sj is a dummy variable indicating
production during the dispute period (1994–96), Di is a dummy variable
indicating production in Decatur, and is the Decatur–disputeS # Dj ij
period interaction.17
In our hypothetical calculation, we set the Decaturdispute
interaction to zero but otherwise keep production and the other
variables the same. The results suggest that 35.2, or 53 percent, of the
67 fatal accidents linked to P235 tires were due to the excess number
of defects occurring on tires manufactured in Decatur during the labor
dispute. The number of fatalities is higher because fatal accidents had
1.30 victims, on average. Thus these results suggest that about 46 lives
were lost because of the excessive number of defects occurring in P235
tires manufactured in the Decatur plant during the labor dispute.
Using the model in column 2, we can also predict the number of
fatalities that were avoided because of the recall of all P235 Firestone
17
Notice that eq. (2) also equals because the exponen-S y Ϫ S exp(x b ϩ Sa ϩ D g)ijt ijt ijt ijt j i
tial of the fitted Poisson values replicates the total count.
276
TABLE 6
Poisson Regression Models for Number of Incidents with Defective Tires, by Type of Accident, NHTSA Complaints Data (Full
Reporting Period): Aggregate Analysis of ATX and Wilderness AT Tires (Np135)
Dependent Variable
Accident with Fatality Nonfatal Injuries
Noninjurious
Complaints All Complaints
Explanatory Variable (1) (2) (3) (4) (5) (6) (7) (8)
Constant Ϫ17.33
(4.06)
Ϫ13.56
(4.94)
Ϫ17.98
(3.59)
Ϫ11.02
(3.28)
Ϫ13.55
(2.04)
Ϫ8.45
(1.64)
Ϫ13.61
(2.01)
Ϫ8.40
(1.61)
Log production 1.05
(.29)
.84
(.33)
1.15
(.25)
.76
(.22)
1.02
(.14)
.74
(.11)
1.03
(.14)
.75
(.11)
Proportion ATX Ϫ.74
(1.25)
Ϫ1.31
(1.12)
Ϫ.87
(.41)
Ϫ.91
(.40)
Proportion original equipment Ϫ.77
(1.71)
Ϫ1.72
(1.53)
Ϫ1.58
(.59)
Ϫ1.54
(.58)
Age .98
(.30)
1.08
(.32)
.95
(.26)
1.21
(.25)
.89
(.14)
1.11
(.11)
.90
(.14)
1.11
(.11)
277
Age2
Ϫ.09
(.04)
Ϫ.10
(.04)
Ϫ.09
(.04)
Ϫ.10
(.03)
Ϫ.08
(.02)
Ϫ.10
(.01)
Ϫ.09
(.02)
Ϫ.10
(.01)
Decatur Ϫ.07
(1.08)
Ϫ.62
(1.20)
.82
(.55)
Ϫ.22
(.60)
.31
(.33)
Ϫ.52
(.28)
.35
(.32)
Ϫ.48
(.27)
Labor dispute period .42
(.56)
.27
(.60)
.17
(.55)
Ϫ.13
(.54)
.26
(.23)
.02
(.19)
.26
(.22)
.02
(.18)
Decatur#dispute 2.39
(1.15)
2.71
(1.15)
1.62
(.73)
2.22
(.70)
1.67
(.38)
2.05
(.31)
1.68
(.37)
2.08
(.30)
Mean dependent variable .49
[1.21]
.49
[1.21]
.88
[2.08]
.88
[2.08]
9.35
[16.57]
9.35
[16.57]
10.73
[19.39]
10.73
[19.39]
Pseudo 2
R .41 .43 .42 .48 .63 .70 .65 .73
Log likelihood Ϫ88.26 Ϫ85.89 Ϫ135.85 Ϫ120.84 Ϫ495.21 Ϫ392.18 Ϫ534.38 Ϫ416.42
Note.—Robust standard errors are in parentheses. Standard deviations of the dependent variables are in brackets. All specifications correspond to NHTSA complaints data. The dependent variable
corresponds to incidents per year of production, plant of production, and calendar year of failure. Incidents correspond to reporting periods prior to June 2001, for failures occurring between
January 1991 and December 1999. Proportion original equipment denotes the proportion of tires produced in the cell that were original equipment for Ford, Mercury, or Mazda vehicles. Proportion
ATX denotes the proportion of tires produced in the cell that were ATX tires. To adjust for incomplete reporting in the NHTSA data set, missing values were imputed by assigning values from other
observations with complete data, conditional on available data. Ten imputed data sets were generated. Reported estimates were obtained by averaging the estimates derived from each imputed data
set. Standard errors were calculated by taking the square root of the sum of the average sampling variance across data sets and the variance of the estimates across the 10 imputed data sets (to
account for variability due to imputation). Other summary statistics reported are the average of the estimates across the imputed data sets.
278 journal of political economy
tires. Specifically, we predicted the number of fatal accidents that would
occur over the period 2000–2004 by
2
S [exp (x b ϩ S a ϩ Dg ϩ S # D d) ϩ (Aj ϩ q)k ϩ (Aj ϩ q) k ]ijq ij j i ij 0 0 2
for q p 1, … , 5, (3)
where A0 is tire age in 1999, and k1 and k2 are the coefficients on age
and age squared in the Poisson model. This calculation gives an estimate
that 191.4 additional fatal accidents would have occurred on the recalled
P235 tires over five years.18
With the assumption of 1.30 deaths per fatal
accident, these results suggest that 249 deaths were avoided as a result
of the recall.
IV. Engineering Tests
Firestone performs “advanced aging” tests on tires that are selected off
the production line. These stress tests are designed to detect production
defects before tires are in the field and are also required for certain
tire ratings. We obtained data on 106 high-speed stress tests Firestone
performed on size P235 tires and two identical tests on size P255 tires
that Firestone submitted to Congress.19
The high-speed stress test is used
by tire manufacturers to obtain a Society of Automotive Engineers “S”
speed rating. The test data pertain to tires produced from 1989 to 1998,
although most of the data pertain to 1995–98; there are only two tests
available for Decatur prior to 1995, both conducted in 1990. Three of
the 108 tests either had missing variables or were unreadable and were
therefore dropped from the analysis.
According to the protocols, the high-speed test is conducted by overinflating
a tire to 38 psi and running it on a lab machine at specified
(accelerating) speeds for prescribed lengths of time, at 100 degrees
Fahrenheit (plus or minus 5 degrees), with a load of 80 percent of the
maximum load stamped on the tire sidewall. For the first five minutes,
the tire is run at 50 miles per hour (mph), then the speed increases to
75 mph for the next five minutes, then to 87 mph for 10 minutes, then
to 93 mph for 10 minutes; thereafter the speed increases in increments
of 6 mph after each 10-minute period. The tire undergoes this test until
it bursts. The most common type of failure was a tread separation. To
18
Of course, it is possible that the replacement tires could also contribute to fatalities.
We have no way of quantifying this effect and suspect that it is small in any event.
19
Firestone refers to these tests as “Indoor Test U1.” It is possible that more than 108
test results were submitted to the congressional record, but we could obtain only 108
results from the information the committee made publicly available. Results are robust if
we exclude the two tests on P255 tires.
labor strife 279
Fig. 5.—High-speed stress tests: box plots of miles traveled before burst. Source: Authors’
calculations from Firestone U1 test data reported to the House Committee on Commerce
(U.S. House 2000). Tests in this sample were conducted in 1995–98. The bottom of each
box represents the value at the twenty-fifth percentile, and the top represents the value
at the seventy-fifth percentile. The middle line indicates the distance traveled before the
median tire burst. The vertical lines (known as upper and lower adjacent hinges) extend
to the first data point encountered within 1.5 times the interquartile range from the edge
of the box. Observations that lie beyond the upper and lower adjacent hinges are indicated
by a circle.
receive an S speed rating, the tire must exceed 112 mph for 10 minutes
before breaking.
The advantage of analyzing these test results is that they hold many
factors constant, including the load, temperature, tire pressure, speed,
and driving conditions. A disadvantage of this engineering test (and
others) is that it is unclear whether failures caused by the extreme
conditions in laboratory settings can be extrapolated to everyday, normal
use; the types of conditions (e.g., driving speeds) encountered in practice
are different. Nevertheless, the laboratory tests provide another
check on the quality of the tires produced during the labor dispute visa`-vis
other tires.
Figure 5 presents box plots for the high-speed test results. The outcome
measure used to summarize test performance is distance traveled
before the tire developed a problem, since this reflects both the amount
of time the tire ran and its speed. The figure indicates that tires made
in Decatur during the labor dispute had lower performance on the
280 journal of political economy
high-speed stress test than tires made in Decatur after the labor dispute
or tires made in Joliette or Wilson either during or after the dispute.
The median tire produced in Decatur during the dispute traveled less
distance before bursting than the tire ranked at the bottom twenty-fifth
percentile from the Decatur plant after the dispute or the other plants
either during or after the dispute. Indeed, the tire that performed at
the top seventy-fifth percentile from Decatur during the dispute performed
about as well as the twenty-fifth percentile tire produced at other
plants at the same time. Also, tires produced in Decatur during the
dispute have a positively skewed distribution, whereas the other tires
are more symmetric, which tends to exaggerate the performance of
Decatur tires in least-squares compared to median regression models.
Table 7 presents regression estimates in which the dependent variable
is the number of miles traveled on the stress test before failing, and the
key explanatory variables are a Decatur dummy, dispute period dummy,
and Decatur–dispute period interaction. We also control for preproduction
model tires; for example, the Wilderness tire was a preproduction
tire before it was marketed in 1996. Although one could argue that
the preproduction tires should be excluded because they have a different
design, they do provide an indication of the skill and care of the
employees who built the tires at the various plants. We vary the years
covered by the sample, as well as present results in columns 4 and 5
excluding the preproduction tires. In all cases, however, the results are
similar to the box plots: tires made in Decatur did not perform as well
on the stress tests as tires produced in Decatur after the dispute or tires
produced in the other plants. For example, in column 3, the Decaturdispute
interaction has a coefficient of Ϫ17 miles, with a t-ratio of 3.18.
Interestingly, these tires all passed the standard for an S speed rating,
so one could reasonably question whether the standard is set at an
appropriately stringent level.
V. Precise Timing
The NHTSA complaints data set can be used to determine the particular
week in which tires that resulted in a claim were produced. Although
we do not have weekly production figures, we can examine whether
there was an excess number of complaints generated by tires around
key dates, such as when replacement workers were hired. In addition,
we can make a crude adjustment for scale by interpolating annual production
figures. Because the data are noisy, we examine the month in
which tires were produced instead of the week, but the weekly data
suggest the same patterns.
Figure 6 displays the number of complaints concerning tires produced
each month in Decatur. Figure 7 makes an attempt to scale complaints
TABLE 7
High-Speed Stress Test Results: Ordinary Least Squares Estimates
Dependent Variable: Miles Traveled to Burst
Explanatory Variable
Means:
Full Sample
(1)
Full Sample
(2)
1994–98
(3)
1995–98
(4)
No Preproduction
Tires
(5)
No Preproduction
Tires, 1995–98
(6)
Constant 108.83
(2.30)
110.16
(2.65)
110.16
(2.66)
108.83
(2.37)
110.16
(2.86)
Decatur (1pyes) .47
[.50]
2.13
(3.72)
1.57
(4.18)
1.99
(4.22)
Ϫ.63
(4.37)
Ϫ1.41
(5.25)
Dispute period (1pyes) .61
[.49]
4.93
(3.36)
3.74
(3.66)
4.61
(3.92)
4.88
(3.91)
5.34
(4.87)
Decatur#dispute period .30
[.46]
Ϫ16.85
(4.80)
Ϫ16.34
(5.22)
Ϫ17.29
(5.43)
Ϫ12.63
(5.96)
Ϫ13.64
(7.13)
Preproduction models .35
[.48]
Ϫ1.19
(2.55)
Ϫ1.43
(2.64)
Ϫ2.31
(2.81)
Mean dependent variable 107.28
[12.57]
107.28
[12.57]
107.48
[12.97]
107.07
[13.08]
107.28
[12.35]
107.49
[13.37]
Observations 105 98 90 68 57
2
R .23 .24 .25 .15 .18
F-statistic 7.40 7.31 7.00 3.90 3.82
Note.—Standard errors are in parentheses. Standard deviations are in brackets. Years refer to years of production.
Fig. 6.—Number of complaints concerning P235 tires produced in Decatur, by month.
Source: Authors’ calculations based on NHTSA complaints data. Records with missing
data are excluded.
Fig. 7.—Number of permanent workers, replacement workers, and estimated complaints
per million tires produced, by month (Decatur plant). Source: Authors’ calculations based
on NHTSA complaints data, estimated monthly production (see Krueger and Mas 2002,
app.), press reports, and Firestone documents submitted to the House Committee on
Commerce (U.S. House 2000). Dashed lines indicate estimates of employment; blocks
indicate specific data points.
labor strife 283
by monthly production.20
The figure displays the number of complaints
relative to estimated monthly production (in millions) as well as our
best estimate of the number of replacement workers and permanent
workers working in the Decatur plant each month. Interestingly, there
was a spike in the number of tires that generated complaints in the first
half of 1994, around the time in which concessions were demanded and
the old contract expired. Tires produced early in the strike were cited
in a relatively small number of complaints, even when we scale by production.
Through early 1995, when a large number of replacement
workers were building tires, the number of complaints remained relatively
low. It was not until the end of 1995, when large numbers of
replacement workers and returning strikers worked side by side, that
the complaints rate reached its highest level.21
Complaints remained
high for tires made through the first half of 1996.
For figure 8 we constructed monthly production estimates for Joliette
and Wilson tires combined and then subtracted their monthly complaints
rate from the Decatur rate to net out any temporal companywide
effects that may have existed. Figure 8 displays the difference in the
complaints rate by month of production between the Decatur plant and
Joliette and Wilson plants. This figure displays a pattern similar to that
in figure 7.
The monthly complaints data are somewhat noisy and sparse and are
possibly affected by retrospective reporting biases. One reason we have
confidence that the pattern displayed in figures 6–8 is not spurious,
however, is that the claims data by production month show a strikingly
similar pattern: an excess number of complaints in the first half of 1994
and an even larger peak in the second quarter of 1996. Figure 9, for
example, displays the claims rate by month for the Decatur plant minus
the Joliette and Wilson plants.
These results suggest that faulty workmanship by novice replacement
workers is not the sole source of problem tires. A more complicated
dynamic is required to explain the timing of the production of defective
tires. It appears likely to us that something about the chemistry between
the replacement workers and recalled strikers, or the cumulative impact
20
We have annual P235 production by plant, but not monthly data. We used the annual
data and other sources to estimate monthly production for the Decatur, Joliette, and
Wilson plants. The procedures for estimating monthly production are described in detail
in the data appendix of Krueger and Mas (2002).
21
Roger Gates, former president of the URW local in Decatur, has argued that most
workers had not come back to work until December 1996, when a new contract was settled,
so there was no period of time in which a significant number of returning strikers worked
with replacement workers (Ginsbach 2002). This claim, however, is at odds with press
releases from the AFL-CIO, which report that of the 4,200 Firestone workers who went
on strike companywide, all but 700–800 had been recalled by March 1996, and all but
150 had been recalled by October 1996 (see Parks 1996a, 1996b).
Fig. 8.—Difference in the number of complaints per million tires produced, by month
(Decatur plant minus Joliette and Wilson plants). Source: Authors’ calculations based on
NHTSA complaints data. Records with missing data are excluded.
Fig. 9.—Difference in the number of claims per million tires produced, by month
(Decatur plant minus Joliette and Wilson plants). Source: Authors’ calculations based on
Firestone claims data.
labor strife 285
of labor strife in general, created the conditions that led to the production
of many defective tires. Frictions among production workers,
supervisors, and inspectors engendered by the strike and request for
labor concessions might also have played a role. It is also possible that
the Decatur plant became less vigilant at monitoring and scrapping
defective tires in the second half of 1995, which may have contributed
to the dissemination of faulty tires. The fact that many different types
of defects appear to have been responsible for the tire problems is
consistent with our interpretation that the human element was im-
portant.
VI. Other Tire Models and Sizes
If labor strife and ill will among replacement workers, recalled strikers,
and management contributed to the production of defective P235/
75R15 tires, then one would expect these factors to have compromised
the quality of other Firestone tires as well. The NHTSA database includes
complaints concerning non-P235 tires covered by their EA00–023 investigation.
These data are probably not as complete as the P235 sample
but nonetheless provide information as to whether the number of complaints
for other tires produced during the labor dispute was elevated.22
Table 8 presents estimates of Poisson regression models in which the
dependent variable is the number of occurrences of complaints concerning
non-P235 Firestone tires. The sample consists of three plants
by nine production years by nine calendar year cells. As explained in
the data appendix of Krueger and Mas (2002), we use a variety of sources
to derive the total number of non-P235 tires produced in each plant
each year. This variable is likely to be measured with error, so we constrain
the coefficient to one in column 3. In view of the likely impact
of measurement error, it is surprising that the unconstrained production
elasticity exceeds one (see col. 1). The estimated effect for the Decatur–
dispute period interaction is reduced somewhat if we constrain the coefficient
on log production to one.
Most important for our purposes, in all the models the coefficient
on the Decatur plant–dispute period interaction term indicates that
Decatur had an elevated number of complaints involving tires produced
during the labor dispute, other things equal. These results suggest that
the high incidence of defective tires in the Decatur plant during the
labor dispute was not unique to P235, as one would expect if labor strife
was an important reason for the production of defective tires.
22
To avoid contaminating the sample by including P235 tires because of the allocation
of missing values, when we constructed the sample we eliminated all tires that might have
been P235 tires and then imputed the missing values for this subsample.
286 journal of political economy
TABLE 8
Poisson Regression Model for Non-P235/75R15 Firestone Tires under NHTSA
EA00–023 Investigation (Np135)
Dependent Variable: Number of Complaints (with Imputations for Missing Values)
Explanatory Variable
Means
(1) (2) (3)
Constant Ϫ38.37
(9.21)
Ϫ14.60
(.31)
Log total non-P235 tires
produced
15.49
[.45]
2.49
(.58)
1*
Age 2.67
[2.22]
.35
(.26)
.28
(.26)
Age2
12
[15.47]
Ϫ.08
(.05)
Ϫ.08
(.05)
Decatur .33
[.47]
Ϫ.87
(.57)
Ϫ1.12
(.54)
Dispute .33
[.47]
Ϫ.48
(.39)
Ϫ.19
(.35)
Decatur#dispute .11
[.32]
1.98
(.80)
1.34
(.74)
Mean dependent variable 2.12
[5.14]
2.12
[5.14]
2.12
[5.14]
Pseudo 2
R .33 .26
Log likelihood Ϫ339.44 Ϫ376.79
Note.—Robust standard errors are in parentheses. Standard deviations of the dependent variable are in brackets.
All specifications use NHTSA complaints data. The dependent variable corresponds to incidents per year of production,
plant of production, and calendar year of failure. Incidents correspond to reporting periods prior to August 2000, for
failures occurring between January 1991 and December 1999. To adjust for incomplete reporting in the NHTSA data
set, missing values were imputed by assigning values from other observations with complete data, conditional on available
data; tire size was not imputed, however. Ten imputed data sets were generated. Reported estimates were obtained by
averaging the estimates derived from each imputed data set. Standard errors were calculated by taking the square root
of the sum of the average sampling variance across data sets and the variance of the estimates across the 10 imputed
data sets (to account for variability due to imputation). Other summary statistics reported are the average of the estimates
across the imputed data sets. The specifications in col. 3 constrain the coefficient of log production to one.
* The coefficient is constrained to one.
VII. Conclusion
The evidence we have assembled suggesting that the labor dispute contributed
to the production of many defective Firestone tires in the 1990s
is circumstantial but broad and consistent; we think that the evidence
pointing toward the role of labor strife is more persuasive and comprehensive
than that in favor of the alternative hypotheses that have
been proposed. The temporal patterns in all the data sets we have
examined indicate that a higher than expected rate of problems occurred
in tires produced in Decatur, the plant with the most contentious
labor relations, during the plant’s labor dispute. We can also rule out
many other explanations—for example, the fact that high-speed stress
tests conducted under uniform conditions indicated that tires produced
in Decatur during the labor dispute performed less well weighs against
the hypothesis that changes in regional distribution patterns or the Ford
Explorer was the root cause of the elevated defect rate during the labor
dispute—but we cannot be certain that some unobserved factor did not
labor strife 287
change, causing defect rates to rise in the struck Decatur plant in the
1994–96 period, and then changed back. This would be a remarkable
coincidence; no such factor has yet been identified, but it is a possibility.
Unless another factor can be found that explains the sudden rises in
defects in tires produced when Bridgestone/Firestone demanded contract
concessions from the union in stalled negotiations and again when
replacement workers and recalled strikers worked side by side, we think
that the weight of the evidence points to labor strife as being at the
root of many of the defective tires.
Our results complement other findings concerning labor-management
conflict. Freeman and Medoff (1984, p. 167) cite evidence suggesting
that the union productivity effect in the underground bituminous
coal industry varies sharply over time with the state of labor
relations and management policies. Kleiner, Leonard, and Pilarski
(2002) study the impact of the industrial relations climate on productivity
at a plant that assembles large aircraft. The evidence they present
suggests that industrial relations events had short-term effects on productivity,
which dissipated in the longer term. Katz, Kochan, and Gobeille
(1983) study the relationship between various plant-level measures
of industrial relations and economic performance at General Motors
and find strong correlations between the industrial relations climate
and productivity. Additionally, they find a strong correlation between
some measures of industrial relations performance and the quality of
production. Interestingly, the industrial relations measure with the
strongest negative correlation with production of automobiles that did
not pass inspection is the negotiation time required to reach a new
contract agreement.
Ironically, an internal Bridgestone document obtained by the URW
reportedly stated that “while it was nice to share a good relationship
[with the union], it would no longer be in the company’s interest” (see
Franklin 2001, p. 122).23
The stock market valuation of Bridgestone/
Firestone fell from $16.7 billion to $7.5 billion in the four months after
the recall was announced, and the top management of Bridgestone/
Firestone has been replaced. The company also closed the Decatur plant
in December 2001 and considered abandoning the Firestone brand
name. If antagonistic labor relations were responsible for many of the
defects, even indirectly, this episode would serve as a useful reminder
that a good relationship between labor and management can be in both
the company’s and the union’s interests.
More speculatively, our results raise the possibility that there may be
costs associated with hiring replacement workers and labor strife that
23
This document was obtained by the union from an anonymous source, so it is unclear
how much stock to put in it.
288 journal of political economy
are not internalized or anticipated by labor or management, especially
in industries that affect the public safety. We estimate that more than
40 lives were lost as a result of the excessive number of problem tires
produced in Decatur during the labor dispute, and that number probably
would have been more than twice as high if it were not for the tire
recall. If the external costs were not taken properly into account by
labor and management, then public policy could possibly play a valuable
role by requiring more safety inspections for products manufactured
during a strike or period of labor strife.
References
Barboza, David. 2000. “Firestone Workers Cite Lax Quality Control.” New York
Times (September 15), p. C1.
Bierecree, Adrienne M. 1991. “Capital Restructuring and Labour Relations: The
International Paper Company Strike.” In International Contributions to Labour
Studies, vol. 1, edited by Gerald Berk et al. San Diego, Calif.: Academic Press
(for Notre Dame Univ., Labor Studies Center).
Bureau of National Affairs. 1995. “Rubber Workers Call End to Strike against
Bridgestone/Firestone Plants.” Daily Labor Report, no. 99 (May 23), pp. A6–
A7.
Cameron, A. Colin, and Pravin K. Trivedi. 1998. Regression Analysis of Count Data.
Econometric Society Monograph no. 30. Cambridge: Cambridge Univ. Press.
Cimini, Michael H. 1996. “Labor-Management Bargaining in 1995.” Monthly Labor
Rev. 119 (January/February): 25–46.
Craypo, Charles. 1994. “Meatpacking: Industry Restructuring and Union Decline.”
In Contemporary Collective Bargaining in the Private Sector, edited by Paula
B. Voos. Ithaca, N.Y.: ILR Press (for Indus. Relations Res. Assoc.).
Fehr, Ernst, and Simon Ga¨chter. 2000. “Fairness and Retaliation: The Economics
of Reciprocity.” J. Econ. Perspectives 14 (Summer): 159–81.
Fogarty, Thomas. 2000. “Retirees Cite Production Practices: Depositions in ’92
Case May Shed Light on Recent Problems, but Company, Unions Dispute
Claims.” USA Today (August 24).
Franklin, Stephen. 2001. Three Strikes: Labor’s Heartland Losses and What They Mean
for Working Americans. New York: Guilford.
Freeman, Richard B., and James L. Medoff. 1984. What Do Unions Do? New York:
Basic Books.
Ginsbach, Pam. 2002. “Labor Strife Found Key Factor in Production of Defective
Tires at Bridgestone/Firestone.” Daily Labor Report [Bur. Nat. Affairs] (January
11), p. AA-1.
Green, Hardy. 1990. On Strike at Hormel: The Struggle for a Democratic Labor Movement.
Philadelphia: Temple Univ. Press.
Greenhouse, Steven. 1996. “Accord Reached in Dispute at Tire Company.” New
York Times (November 7), p. A16.
Katz, Harry C., Thomas A. Kochan, and Kenneth R. Gobeille. 1983. “Industrial
Relations Performance, Economic Performance, and QWL Programs: An Interplant
Analysis.” Indus. and Labor Relations Rev. 37 (October): 3–17.
Kleiner, Morris M., Jonathan S. Leonard, and Adam M. Pilarski. 2002. “How
Industrial Relations Affect Plant Performance: The Case of Commercial Aircraft
Manufacturing.” Indus. and Labor Relations Rev. 55 (January): 195–218.
labor strife 289
Krueger, Alan B., and Alexandre Mas. 2002. “Strikes, Scabs and Tread Separations:
Labor Strife and the Production of Defective Bridgestone/Firestone
Tires.” Working Paper no. 461 (January). Princeton, N.J.: Princeton Univ.,
Indus. Relations Sec. http://www.irs.princeton.edu/wpframe.html.
Meyer, Bruce M. 2002. The Once and Future Union: The Rise and Fall of the United
Rubber Workers, 1935–1995. Akron, Ohio: Univ. Akron Press.
Parks, James B. 1996a. “Bridgestone Calls Most Back; USWA Fight Will Continue
on Fair Labor Practices.” AFL-CIO News (October 11).
———. 1996b. “Bridgestone Faces Worldwide Worker Solidarity.” AFL-CIO News
(March 22).
Pinkerton, James. 2001a. “Doctor Tells Jurors How Firestone Tires Failed.” Houston
Chronicle (August 15), p. A21.
———. 2001b. “Firestone Attorneys Put Blame on Ford in Rollover Lawsuit.”
Houston Chronicle (August 14), Business section, p. 1.
Poynter, Dan. 1992. Parachuting: A Skydiver’s Handbook. 6th ed. Santa Barbara,
Calif.: Para Publishing.
Rabin, Matthew. 1993. “Incorporating Fairness into Game Theory and Economics.”
A.E.R. 83 (December): 1281–1302.
Rosenblum, Jonathan D. 1995. Copper Crucible: How the Arizona Miners’ Strike of
1983 Recast Labor-Management Relations in America. Ithaca, N.Y.: ILR Press.
Rubin, Donald B. 1987. Multiple Imputation for Nonresponse in Surveys. New York:
Wiley, 1987.
Sabath, Donald. 1997. “Goodyear Workers Stand Firm.” Cleveland Plain Dealer
(April 22), p. C1.
Skertic, Mark. 2000. “Firestone Retirees Cite Pressure to Produce.” Chicago SunTimes
(October 27).
U.S. House. 2000. Committee on Commerce. The Recent Firestone Tire Recall Action,
Focusing on the Action as It Pertains to Relevant Ford Vehicles. Hearings before the
Subcommittee on Telecommunications, Trade and Consumer Protection and the Subcommittee
on Oversight and Investigations. 106th Cong., 2d sess.
Whitford, Marty. 1994. “Bridgestone Unit to Show First Full-Year Profit.” Rubber
& Plastics News (January 17), p. 6.