1. Introduction
Nearly 900,000 organizations in 170 countries have adopted the ISO 9001 Quality Management
System standard,1 a remarkable figure given the lack of rigorous evidence regarding how the standard
actually affects organizational practices and performance. Implementing a quality management system
* Quality Management and Job Quality
2
that conforms to ISO 9001 requires companies to document operating procedures, training, internal auditing, and corrective action procedures. It also requires companies to implement procedures to improve existing procedures. Proponents claim quality programs such as ISO 9001 improve both management practices and production processes and that these improvements, in turn, will increase both sales and employment (unless productivity gains outweigh sales increases). The latter benefits are magnified if customers use the adoption of ISO 9001 or other quality programs as a signal of high quality products or services. To the extent that greater employee skill and training are required to develop and implement procedures to
improve procedures, the theory of human capital suggests that employees’ earnings should rise as well. Finally, ISO 9001 can improve worker safety through the identification and elimination of potentially hazardous practices, development of a formal corrective action process, and institutionalization of routine audits and management reviews. Some critics point to the potential for quality programs such as ISO 9001 to harm employees by formalizing and documenting work practices. Such routinization may reduce skill requirements and increase cumulative trauma disorders (e.g., Brenner, Fairris, and Ruser 2004).
We need to examine several outcomes of vital importance to company managers and owners, including whether ISO 9001 is associated with subsequent sales growth and longer company survival. We examined single-plant firms across an array of industries in California, comparing ISO 9001 adopters to
comparison groups of non-adopters matched on industry, location, size (baseline sales, employment, and
total payroll), and pre-adoption injury rates. We employed a difference-in-differences approach that
accommodates common shocks that affect each industry. When appropriate, we also control for company
characteristics that vary over time such as size and changes in occupational mix.
We find ISO adopters to have higher rates of corporate survival, sales and employment growth, and
wage increases than the matched control groups of non-adopters. While we find that adopters become
more likely to report no injury rates (as measured by workers’ compensation claims) in the years
Quality Management and Job Quality
3
following adoption, we find no evidence that a firm’s total or average injury costs improved or worsened
subsequent to adoption.
2. Literature
In this section, we review the literature that has addressed how ISO 9001 predicts changes in
outcomes of interest to owners and managers, such as profitability, and then discuss the much smaller
literature on how ISO 9001 affects changes in outcomes of interest to employees, such as employment
and injury rates.
2.1 Organizational outcomes
A few careful empirical studies have examined how implementing the ISO 9001 quality management
standard affects employers’ outcomes and practice.2 Most of these studies examine the impact of ISO
9001 on manufacturers in the United States. Terlaak and King (2006) find that plants that adopt ISO 9001
typically increase their rate of production growth. Others find ISO 9001 certification to be associated with
subsequent abnormal returns along a host of financial metrics including stock prices (Corbett, Luca, and
Pan 2003; Sharma 2005). Various studies find benefits strongest among small firms (Docking and Dowen
1999; McGuire and Dilts 2008) and among those with a modest level of technological diversity, and/or
early adopters (Benner and Veloso 2008). King and Lenox’s (2001) finding that adopting ISO 9001 leads
plants to reduce waste generation and toxic chemical emissions suggests that implementing the quality
management standard has positive spillover effects that can improve environmental management
practices. Naveh and Erez (2006) deduce from survey data that ISO 9001 adoption enhances worker
productivity and workers’ attention to detail, but impedes worker innovation. Interestingly, we found no
prior research that examines how the ISO 9001 quality standard affects product or process quality, nor
how it affects employees. Similarly, albeit outside the realm of ISO 9001, a number of event studies find
that financial performance, as measured by stock price and operating income, improves after firms win a
2 A much larger literature examines why firms adopt the ISO 9000 standard. See Corbett, Montes-Sancho and
Kirsch (2005) for a comprehensive review.
Quality Management and Job Quality
4
quality award and implement Total Quality Management programs (Easton and Jarrell 1998; Hendricks
and Singhal 1996, 1997, 2001a, 2001b).3
Several studies have found ISO 9001 adopters’ financial performance to be superior to that of peers
prior to, but not after, registration (see Heras, Gavin, and Dick 2002 for evidence from Spain; Häversjö
2000 for Denmark; and Simmons and White 1999 for the United States). These studies are consistent with
a positive selection effect, but do not suggest that any causal benefits are associated with ISO registration.
We return to the distinction between correlation and causality in the subsequent analyses.
2.2 Employee outcomes
We found almost no prior research that examines how the ISO 9001 quality management standard
affects key outcomes of interest to employees, such as employment and earnings.4 Among the few studies
that have examined how other quality programs affect occupational health and safety is Adler, Goldoftas,
and Levine’s (1997) case study of a General Motors automobile plant. They found that the plant’s
suspension of job rotation subsequent to the adoption and implementation of the Toyota Production
System’s quality principles precipitated a dramatic rise in cumulative trauma disorders (CTDs), leading
the authors to conclude that Toyota Production System principles such as short cycle times, standardized
work methods, and minimizing worker idle time increase the risk of CTDs. These findings are consistent
with those of Wokutch’s (1992) study of Japanese auto transplants in the United States. Adler et al.
(1997) further found that other Toyota Production System principles such as employee focus on
continuous improvement could help reduce injury rates, provided employees were empowered to focus on
safety and health issues.
A few larger-scale studies have examined the relationship between quality management practices and
worker injuries. Lean production practices such as faster work pace and reduced cycle time have been
3 See Wayhan and Balderson (2007) for a comprehensive review of studies that examine how implementing Total
Quality Management affects financial performance.
4 In a suggestive small study (45 ISO adopters), O’Connor (2005) finds that Oregon employers with ISO
certification increased employment more rapidly than peers in their industry. That analysis did not control for
employer size and did not test for statistical significance.
Quality Management and Job Quality
5
found to be positively associated with worker stress (Conti et al. 2006) and quality circles and job rotation
with CTDs (Brenner et al. 2004). But these studies’ reliance on data collected simultaneously for
workplace practices and illness/injury rates precludes distinguishing selection from causal effects.5
Naveh and Marcus’s (2007) analysis of the road safety experience of 40 ISO 9002 adopters in the
trucking industry found the number of accidents post adoption to be reduced faster by certified firms than
by their peers. Presumably due to the modest sample size, the results are not consistently statistically
significant.
This paper reports the results of the first large-scale study to examine the effects of ISO 9001 on
employee outcomes. Among the worker-level outcomes on which we focus are wages and the frequency
and magnitude of worker injuries. We also examine the effects of ISO 9001 on a number of
organizational outcomes including sales, employment, and company survival. Our empirical approach of
constructing and analyzing a panel dataset of plant-level data on a wide array of worker outcomes
overcomes the small sample sizes and/or lack of longitudinal data that have plagued previous studies.
Because ISO 9001 is typically adopted at the plant-level and outcome data typically available only at the
firm level, we focus on single-plant firms. Our data and methods enable us to clearly distinguish the
effects of selection on observables from causal effects. Moreover, whereas prior studies of safety and
health have largely relied on self-reported survey data, we measure the frequency and cost of injury using
workers’ compensation data, which render our measures of occupational health and safety outcomes less
susceptible to measurement error and potential bias arising from ex post rationalization by the managers
who decided to invest in the quality management practices being evaluated.
5 For example, Brenner, Fairris, and Ruser (2004) interpret the positive association of unions with cumulative
trauma disorders as selection effects of unionization, the positive association of job rotation with cumulative trauma
disorders as reverse causality (whereby high rates of repetitive motion injuries lead plants to introduce job rotation),
and the positive association of quality circles and just-in-time production with cumulative trauma disorders as causal
effects of the work practices. Attempting to address causality by analyzing panel data, Fairris and Brenner (2001)
found that CTD rates in their industry declined after plants adopted self-directed work teams or Total Quality
Management, but average industry CTD rates increased after plants implemented quality circles or job rotation.
Because they could link workplace practices only to industry-level (3-digit SIC codes) CTD rates, it is difficult to
interpret these results.
Quality Management and Job Quality
6
3. Theory and Hypotheses
Companies that implement a quality management system that conforms to ISO 9001 typically
improve the documentation of operating procedures, training, and procedures for corrective action. To
become certified to the ISO 9001 standard, a plant hires an accredited third-party auditor to certify that
the plant has (1) written procedures for all significant operations, (2) training, monitoring, and other
procedures in place to ensure that written procedures are followed, and (3) implemented procedures for
continuously improving its other procedures. The latter requirement has implications for training,
decision-making, and incentives with respect to low-level employees. The cost of implementing ISO
9001, which includes developing procedures, documentation, and training and hiring a third-party auditor,
range from $97,000 to $560,000, depending on the size and complexity of the operation (Docking and
Dowen 1999).6
3.1 ISO 9001 and changes in plant scale
The value of ISO 9001 certification lies in a combination of learning, incentives, and signaling. The
learning channel operates if the ISO 9001 certification process teaches managers how to reduce costs or
cost-effectively improve quality. The incentives channel operates if ISO 9001 certification increases
customers’ willingness to pay for quality, which, in turn, is an incentive for managers to improve product
quality. The signaling channel, like the incentives channel, operates if, in the absence of ISO 9001, many
customers cannot detect (and thus are unwilling to pay for) improvements in product or service quality. In
such cases, certification could be a useful signal that enables buyers to distinguish higher- from lowerquality
firms (Spence 1973). ISO 9001 can play this signaling role as long as adoption is more often
profitable for firms that already had higher quality, as Terlaak and King (2006) argue.
All of these channels yield higher unit sales and/or higher prices. There is considerable evidence,
moreover, consistent with these channels, that many industrial buyers use ISO 9001 certification to screen
6 Docking and Dowen’s (1999) reported average cost estimates range from $71,000 to $409,000 in 1996 dollars,
which we adjusted for inflation using the Inflation Calculator created by the U.S. Department of Labor’s Bureau of
Labor Statistics (available at http://www.bls.gov/; accessed June 7, 2008).
Quality Management and Job Quality
7
potential suppliers (e.g., Ferguson 1996).
All of these channels give rise to the same predictions.
HYPOTHESIS 1a: ISO 9001 certification leads to higher rates of firm survival.
HYPOTHESIS 1b: ISO 9001 certification leads to higher sales.
If the hypothesized increase in sales is due to increased unit sales (rather than merely increased
prices) that cannot be accommodated by existing worker capacity, ISO 9001 certification would increase
employment.
HYPOTHESIS 2a: ISO 9001 certification leads to higher employment, but by less than sales
increases.
Survey data further reveals that ISO 9001 can enhance worker productivity (Naveh and Erez 2006). If
so, employment growth will be proportionately less than sales growth, leading to:7
HYPOTHESIS 2b: ISO 9001 certification leads to higher labor productivity.
3.2 ISO 9001 and wages
ISO 9001 can have positive or negative effects on wages.8 Helper, Levine, and Bendoly (2002) found
that attempts to foster employee involvement led to higher wages because companies sought to
compensate employees for exerting the incremental effort to achieve the requisite higher skill levels.
Employees of firms that adopt ISO 9001 are often asked to perform many discretionary tasks such as
documenting new procedures and offering quality improvement ideas. ISO 9001 plants must develop and
deploy quality-related training to ensure that employees properly implement new procedures and develop
the skills required to conduct internal audits and root-cause analyses and continuously improve the plant’s
other procedures. These tasks require specific skills and imply increased reliance on employees’
discretionary efforts. The higher discretionary effort might require more skills (as in theories of human
capital), lead firms to pay higher wages to induce higher effort (as in efficiency wage theories [Levine
7 If the productivity gain is greater than the increase in unit sales, employment can fall, a case we do not consider
further.
8 These hypotheses are elaborated in Helper et al. (2002).
Quality Management and Job Quality
8
1992]), or increase employees’ bargaining power (Lindbeck and Snower 1986). If greater human capital,
efficiency wages, and/or bargaining power are important, we have:
HYPOTHESIS 3: ISO 9001 certification leads to higher wages.
Ensuring that written procedures are present and followed typically implies a fairly routinized
workplace. Such routinization can reduce frontline workers’ skills, discretion, and bargaining power. We
discovered in the course of our field research, for example, a plant that was implementing ISO 9001 with
the express purpose of documenting workers’ tacit knowledge and procedures so that the plant could be
replicated overseas using lower cost labor. When these forces prevail, we have:
HYPOTHESIS 3′: ISO 9001 certification leads to lower wages.
3.3 ISO 9001 and occupational health and safety
Adopting ISO 9001 can improve occupational health and safety in a variety of ways. In the process of
formally documenting procedures, managers can identify and eliminate hazardous practices and add
safety precautions. Moreover, by fostering more focused attention to detail (Naveh and Erez 2006), ISO
9001 adoption can reveal new “win-win” opportunities for improving quality or efficiency and
occupational health and safety that were previously obscured by indirect and distributed costs and benefits
(King and Lenox 2001). Additionally, serious accidents can be avoided by organizations that have
processes in place that provide warning signals and prompt corrective action (Marcus and Nichols 1999).
Finally, routine auditing and corrective action procedures required by ISO 9001 for addressing
management system failures encourage root-cause analysis, which can identify problematic work
practices that would otherwise lead not only to quality failures, but also to occupational health and safety
concerns.
Indeed, departments charged with managing quality sometimes also manage health and safety, and
companies are increasingly implementing integrated management systems that incorporate all these areas
(Toffel 2000; Barbeau et al. 2004). Implementing ISO 9001 can improve occupational health and safety if
the tools of continuous improvement that often accompany certification are applied to problems in this
area. Employees who know how to identify root causes of quality problems, for example, also have the
Quality Management and Job Quality
9
skills to identify root causes of safety problems. Exploiting these opportunities yields:
HYPOTHESIS 4: Adopting ISO 9001 reduces the number and cost of occupational injuries.
ISO 9001 emphasizes routinization and standardization of tasks, but high rates of repetition and
increased monitoring can increase stress and repetitive motion injuries, potentially worsening the safety
records of plants with quality programs (as argued by Brenner et al. 2004). Additionally, a number of
studies have found ISO 9001 adopters to have higher equipment utilization (Koc 2007; Huarng, Horng,
and Chen 1999). To the extent that this translates into reduced employee downtime, this could increase
employee fatigue, a major cause of injuries (Williamson and Boufous 2007). New quality management
procedures implemented in association with ISO 9001 also sometimes add inspection tasks to work
processes optimized for production rather than inspection, which can result in poor ergonomic conditions
that leave employees susceptible to injuries (Landau and Peters 2006). In the presence of these forces, we
have:
HYPOTHESIS 4′: Adopting ISO 9001 increases the number and cost of occupational injuries.
4. Data
4.1 Sample
Because the typical scope of an ISO 9001 certification is a single plant, but injury data from the
workers’ compensation system are available primarily at the company level, we facilitate the linking of
ISO certification data to injury data by restricting our sample to single-plant firms. Roughly 80% of
manufacturing plants in California are single-plant firms, according to 2005 Dun & Bradstreet data.
We obtained the identity and certification dates of ISO 9001 adopters from the ISO 9000 Registered
Company Directory produced by QSU Publishing Company. According to this source, 5,995 companies
in California were certified to ISO 9001 at the end of 2005. Linking this list of company names and
addresses with Dun & Bradstreet data yielded 1,846 single-plant firms in California that had adopted ISO
9001.
We obtained annual workers’ compensation and payroll data for 1993 through 2003 (the latest year
then available due to reporting lags) from the Uniform Statistical Reporting Plan database of the Workers’
Quality Management and Job Quality
10
Compensation Insurance Rating Board (WCIRB), a nonprofit association of all firms licensed to provide
workers’ compensation insurance in California. WCIRB, which collects and analyzes workers’
compensation claims for all employers covered by worker’s compensation insurance in California, linked
77% (1,418) of our single-plant firms that had adopted ISO 9001 to its database. This proportion is
similar to the proportion of California firms from which WCIRB gathers workers’ compensation data
(i.e., firms that obtain workers’ compensation insurance rather than self-insure).
WCIRB provided us the names and addresses of the 116,389 non-adopting firms that shared the same
region-industry combinations as the adopters. We then linked as many as possible to Dun & Bradstreet
data and were able to confirm that 20,777 of the non-adopters were single-plant firms.
We then obtained annual employment and sales data for 1993-2005 from the National Establishment
Time-Series (NETS) database, a compendium of Dun & Bradstreet data, for most of the ISO adopters and
potential matches: 1,079 adopters and 18,480 non-adopters. Cleaning the data to eliminate firms with
missing values and outliers (see details in the next section) reduced our sample to 916 adopters and
17,849 non-adopters. We used this sample for our selection analysis. Sample characteristics are provided
in Table 1. As described below, we identified subsets of these firms to create matched samples for the
analyses of the effects of ISO registration.
4.2 Measures
We measure each company’s annual injury rate as the number of injuries it reported to claim
workers’ compensation, using WCIRB data. In our models, we employ the log of one plus the injury rate.
We also obtained from WCIRB data each company’s total annual workers’ compensation injury costs (in
dollars) and annual total payroll9 (in dollars). To reduce the effect of outliers, we took the log of these
injury costs after adding $1,000. We also obtained from WCIRB data each company’s location in one of
15 industries and 8 California region (both of which are listed in Table 1).
For each firm-year, we calculated average occupation riskiness as a weighted average of workers’
9 Our payroll measure is what WCIRB calls “exposure,” which equals total payroll after subtracting overtime pay,
shift premiums, and a few other minor adjustments for each of 500 occupational class codes.
Quality Management and Job Quality
11
compensation Pure Premium Rates across the firm’s employment across 500 occupation codes.10 To
understand the intuition behind this measure, if an employer in a given year has a workforce that is in
occupations that, on average, have twice the state-wide workers’ compensation costs, average
occupational riskiness for that firm-year would be twice the state average. For each year, we calculated
each firm’s average wage as annual total payroll (from WCIRB) divided by annual establishment
employment (from Dun & Bradstreet).
We then cleaned the dataset as follows. We recoded sales of zero to missing. We omitted firm-years
with less than $5,000 in payroll. To avoid confounding our analysis with rapidly growing or shrinking
firms, we included only observations in which a firm’s payroll in a given year was between half and twice
its previous year payroll (provided the previous year’s payroll data existed), and did the same for
employment. We also sought to exclude firms that operated only part of the year by omitting firm-years
for which the average wage was below $7,020 (what a half-time worker would earn at California’s
minimum wage in 2002). To reduce the effect of outliers, we analyzed the log of payroll, average
occupational riskiness, employment, average wage, and sales after adding a small amount to deal with
zeros and other small values.
Summary statistics are provided in Table 2.
5. Analysis and Results
5.1 Selection model
The main goal of this paper is to identify any causal effects of ISO 9001 adoption for employees and
employers. ISO 9001 adoption could correlate with outcomes, but not cause them, if a factor such as good
management led to both ISO adoption and good outcomes. If this form of self-selection is important, then
10 Pure Premium Rates are established by WCIRB based on historical workers’ compensation costs for each
occupation.
We calculated a firm f’s average occupational hazardousness in a given year t as:
average occupational hazardousnessft =
ft payroll
Σ payroll × Pure Premium Rate
c
cft c
where Pure Premium Ratec is the 2007 workers’ compensation Pure Premium Rate per $100 of payroll for
occupation class code c and payroll is the measure of total payroll defined as in the previous footnote.
Quality Management and Job Quality
12
we anticipate ISO adopters had good outcomes prior to ISO adoption as well as after. To understand the
self-selection process, we estimate the selection model:
1) ISOit = F(Yit-1&-2 , yeart , industryi , regionj , uit )
where F(.) is the probit function, Yit-1&-2 is the average of 1- and 2-year lagged levels of injury rates and
costs, payroll, employment, wages, sales, and average occupational riskiness, yeart is a complete set of
year dummies, industryi is a set of 15 industry dummies, and regionj is a set of eight California region
dummies. Because we are interested in the determinants of adoption, we drop adopters from the sample
after their adoption year. We report robust standard errors clustered by firm to account for
heteroscedasticity and non-independence among a firm’s observations across years.
5.2 Results of selection analysis
The main result on selection is that larger firms adopt ISO 9001 more often. For example, median
sales and payroll of adopting firms are $3.48 million and $1.04 million, whereas these figures are $0.78
million and $0.16 million, respectively, for non-adopters. We examine this issue more carefully with the
adoption probit regression. Larger firms (in terms of sales and total payroll) adopted ISO 9001 at a higher
rate than other firms in their industry, year, and region (see Table 3). If one employer has one log point
(roughly 1 standard deviation) higher sales and payroll than average, for example, the coefficients predict
a 0.054 percentage point above-average probability of ISO 9001 adoption per year, with most of that
effect due to higher payroll. This increase is roughly equal to 10% of the mean of the sample, which is 5.4
out of a thousand firms adopting each year. Because adoption requires fixed costs—including learning
about the standard’s requirements and developing policies, procedures, and training programs—it is not
surprising that larger firms are more likely to make this investment.
The sales coefficient is statistically significant and economically meaningful after controlling for both
payroll and employment. Thus, ISO adopters had above-average labor productivity prior to adopting ISO
9001. Note that the marginal effect of payroll (0.046, p < 0.01) is higher than the marginal effect of
employment (0.009, n.s.), suggesting that ISO 9001 adopters paid above-average annual wages prior to
ISO adoption (or that WCIRB data on payroll is more precisely measured than D&B data on
Quality Management and Job Quality
13
employment). If higher wages correlate with higher skills and higher skills with higher quality, this result
suggests that ISO adopters may already have been producing above-average quality prior to ISO adoption
(compared to their industry and region).
ISO 9001 adopters also had slightly lower workers’ compensation injury costs, so that 1 log point
lower injury costs (a bit less than one standard deviation) predicts 0.005% higher adoption of ISO 9001 in
the next year (p < 0.05). Our statistically insignificant coefficient on injury rates, and its tiny marginal
effect, provides no evidence that adopters differed from non-adopters in annual number of injuries prior to
adoption.
5.3 Estimating the effects of ISO 9001 certification
To examine the causal effects of ISO 9001, we conduct a difference-in-differences analysis whereby
we compare the changes in payroll, employment, wages, sales, average occupational riskiness, and injury
rates and costs among ISO-certified firms relative to those of a matched set of control firms. This method
permits each firm to have its own baseline level of each outcome. To ensure a valid comparison, we
developed a matched control group.
5.3.1 Developing matched samples
Matching is widely used to construct a quasi-control group based on similar characteristics to those of
the treatment group (Heckman, Ichimura, and Todd 1998). Intuitively, we want to compare companies
that adopt ISO to peers in their industry that, prior to adoption, had similar sales, employment, payroll,
injury rates, and other observable factors.
Matching on the propensity score, the probability of receiving the treatment conditional on covariates,
is as valid as matching on a series of individual covariates (Rosenbaum and Rubin 1983). The identifying
assumption is that the assignment to the treatment group is associated only with observable “pre-period”
variables, and that all remaining variation across the groups is random. This assumption is often referred
to as the “ignorable treatment assignment” or “selection on observables.”
When used to evaluate job training programs, propensity score matching methods have performed
well in replicating the results of randomized experiments under three conditions: (1) the same data
Quality Management and Job Quality
14
sources are used for participants and non-participants; (2) an extensive set of covariates is employed in
the program-participation model used to estimate propensity scores; and (3) participants are matched with
non-participants in the same local labor market (Smith and Todd 2005). Conversely, Heckman , Ichimura,
and Todd (1997) note that substantial bias can result if: (4) controls are included for which propensity
scores are off the support of the participants’ propensity scores (that is, if some of the controls differ
substantially from any of the treatments); (5) the distributions of the participants and non-participants’
propensity scores differ; or (6) unobservable factors influence both participation and outcomes.
We address these six potential sources of bias as follows. First, we use identical data sources for all
facilities (adopters and non-adopters). Second, we include an extensive set of adoption covariates. Third,
we ensure that adopters and non-adopters operate within the same markets by including industry and
region as matching criteria. We address the fourth and fifth concerns by implementing nearest neighbor
matching with a “caliper” restriction to preclude matching when the propensity scores differ by more than
a fixed threshold (as explained below).
The sixth concern addresses selection on unobservables. For example, in the context of ISO 9001, it is
possible that managers in facilities with a “safety culture” (which we do not observe in our data) might be
both more likely to insist upon strong safety performance and more inclined to adopt ISO 9001. We
address this concern in two ways: (1) we control for such unobserved factors that are stable over time by
including a fixed effect for each employer; and (2) we control for differences in levels and trends by
including in our matching criteria lagged levels and trends of sales, employment, earnings, and injury
rates and costs.
We implemented propensity score matching in three steps. In the first, we generated propensity scores
by estimating a probit model for adoption status during 1994-2005. We included a full set of year
dummies, 15 industry dummies, and eight California geographic region dummies. We also included
lagged levels of each of our outcome variables: injury rates and costs, payroll, employment, wages, sales,
and average occupational riskiness. We employed a highly flexible functional form by including the
average level of the prior two years as well as the log and square of this average. Table A1 in the On-line
Quality Management and Job Quality
15
Appendix contains descriptions and summary statistics of the variables we used in the matching process.
Because we are interested in the determinants of adoption, we drop adopters from the sample after their
adoption year. The results of this probit are reported in Table A2 in the On-line Appendix. The predicted
probability of adoption estimated from this probit model is our estimated propensity score.
In the second step, we matched each adopter during its certification year to the non-adopter in the
same industry with the most similar (nearest) propensity score, and that had at least one year of postadoption
data. We refer to the absolute difference between the two propensity scores in a matched pair of
firms as the “match distance.” Of the 892 adopters for which we had estimated propensity scores, we
successfully matched 550 adopters to 550 controls.
In the third step, we assessed the quality of these matches and refined our matched sample. To
confirm whether our matching process resulted in a highly comparable set of adopters and non-adopters,
we compared the two groups’ lagged levels of seven outcome variables: number of injuries, injury costs,
payroll, employment, wages, sales, and average occupational riskiness. The identifying assumption of the
difference-in-differences approach is that the treatment group’s trend during the post-period would have
been indistinguishable from the control group’s trend, had treatment not occurred.
To examine the plausibility of this assumption, we compared the two groups’ performance trends in
the period prior to ISO adoption (as in Barber and Lyon 1996; Dehejia and Wahba 1999; Eichler and
Lechner 2002). We did so using two measures of pre-adoption trends, (1) the percent change in lagged
performance comparing three-and-four-year lagged average to a one-and-two-year-lagged average, and
(2) the difference in the log of these two averages (we added a small constant before taking the log). Ttest
results indicated statistically significant differences at the 10% level for nine of the 21 comparisons
we made for either levels or trends, far more than would be expected by chance. Thus, the first stage of
matching does not yield a very credible comparison group.
To improve the match quality, we dropped pairs for which the propensity scores of the ISO adopters
and potential matches exceeded a given caliper size. A smaller caliper means ISO adopters and their
matches are more similar, but reduces the sample size. After trying various values, we settled on a caliper
Quality Management and Job Quality
16
of 0.07, which yielded a closely matched (but slightly smaller) set of 471 pairs of adopters and controls.
In the caliper-restricted matched sample, t-tests indicated that the groups differed along only 1 of the 21
metrics at the 5% level and 4 of the 21 metrics at the 10% level (see Table A3 in the On-line Appendix).
We ran our evaluation model on this matched sample (see Table 4 for summary statistics).
We followed a slightly modified version of the matching process described above to develop the
matched sample for the survival analysis. When generating propensity scores and matching firms, we no
longer excluded firms that lacked data after the match year, and also matched firms that had adopted as
late as 2003. These modifications resulted in 622 pairs of adopters and non-adopters being included in our
matched sample for the survival analysis.
5.3.2 Organizational survival
We use both nonparametric methods and duration models to examine the survival rate of our matched
pairs of adopter and non-adopter firms. Because our dataset extends through 2003 and the matching was
done in a specific year for the ISO adopter and its matched firm, data on both firms in the pair were right
censored after the same number of years at risk of firm death. We employ a conservative definition of
“firm death” by counting a firm as dead only if it disappears from both the D&B and WCIRB datasets.
Among the 622 pairs of firms that constitute our matched sample for the survival analysis, 0.5% of
the adopters and 7.1% of the controls had disappeared from both our Worker’s Compensation (WCIRB)
and Dun & Bradstreet (D&B) datasets by 2003. A t-test confirmed that survival rates of adopters were
statistically significantly higher (p<0.01).
Although the analyses of raw survival rates use a matched sample, there are still small differences
within each pair in observable factors. In Table A4 in the On-line Appendix, we present the results of a
cross-sectional logit model, a conditional logit (with a conditional fixed effect for each pair), and
stratified Cox proportional hazard models (with each pair its own strata). These models enabled us to
condition on numerous observable factors such as pre-adoption sales and employment. As expected, the
large survival advantage of ISO adopters persists in these specifications. These results provide robust
support for Hypothesis 1a that ISO adoption increases a company’s survival rate.
Quality Management and Job Quality
17
5.3.3 Evaluation model
To assess the impact of adoption, we conduct a difference-in-differences analysis by estimating the
following model for each outcome Yit at firm i in year t:
(2) Yit = αi + β · ISOit + Σj γj·Xjit + δt · yeart + εit ,
where αi is a complete set of firm-specific intercepts. The variable ISOit is an indicator variable coded one
in years after a firm is ISO-certified. Of primary interest is its coefficient β, the estimated effect of
achieving certification. We also include a full set of year dummies (yeart). Depending on the outcome
variable being estimated, we include in Xjit controls for the firm’s current log of: payroll, employment,
sales, average occupational riskiness, or number of injuries (see Tables 5 and 6).
We also employ a variation of this specification that refines our ability to measure the effects of
certification. In Equation 2, the single post-certification dummy variable estimates an overall average
change in outcome levels, pre- to post-adoption. But such improvements might be large in the first few
years after adoption and then attenuate, or take several years to emerge. To estimate potential effects each
period after certification, we include dummies coded 1 for adopters “one to three years after
certification,” “four to six years after certification,” and “seven to nine years after certification.”
Employer outcomes. The results for the difference-in-differences evaluation model of employment
and sales are presented in Table 5. All results employ the matched sample, using years in which both the
ISO adopter and match survived. Employment is about 10 percentage points higher in ISO-certified
workplaces after certification (column 1, b= 10.3%, p < 0.01) than at the comparison firms. These results
strongly support Hypothesis 2a, which predicted higher employment growth at ISO 9001 firms. Column 2
reveals that this increase appears to grow over time, from 6.1% (SE = 1.8%, p < 0.01) in years 1-3, to
22.5% in years 4-6 (SE = 2.9%), to a quite large 32.5% in years 7-9 (SE = 6.3%). Because only 62 of our
471 matched ISO 9001 adopters (13%) adopted seven or more years prior to the end of the matched
dataset used for these models, results in this category are typically not precisely estimated. Nevertheless,
the high employment growth in years 7-9 is statistically significant from the average effect (p < 0.01) and
from the effect in years 1-3 (Wald test F=17.70, p<0.01), though not from the effect in years 4-6.
Quality Management and Job Quality
18
Sales are almost 9 percentage points higher at ISO 9001 firms than at comparison firms (column 3),
which supports Hypothesis 1b. Column 4 reveals that this advantage does not show up until year 4 and
beyond (effect size in years 4-6 = 25%, SE = 4.0%; which is statistically indistinguishable from the effect
size in years 7-9 of 17%, SE = 8.8%, p < 0.10).
This increase in sales is roughly what would be expected from the higher payroll at ISO-certified
firms (column 5). At the same time, there is suggestive evidence of growth in sales conditional on payroll
in years 4-6 after certification (column 6, b = +13.2%, p < 0.01). In contrast, the coefficient is small, less
than 2% in absolute value, and not statistically significant in years 1-3 or years 7-9. Thus, Hypothesis 2b,
predicting gains in labor productivity, is not supported. In results not shown, sales also did not increase
statistically significantly faster at ISO adopters if we controlled for employment instead of payroll; that is,
total sales rose at adopters relative to matched non-adopters, but not sales per employee.
Employee outcomes. The results of our evaluation model for employee outcomes are presented in
Table 6. Column 1 reveals that total payroll at ISO firms grew about 17.7% more than at our matched
control firms (SE = 1.7%, p < 0.01). This increase grew steadily over time from 14% in years 1-3 to 36%
in years 7-9 (column 2). Nearly a third of this increase in payroll (measured using workers’ compensation
records) appears to be due to higher employment (measured using D&B data). That is, conditioning on
employment, payroll at ISO-certified firms grew about 13.5% more than at comparison firms (SE = 1.6%,
p < 0.01, column 3). This growth increased steadily over time from 11.7% in years 1-3 to 22% in years 7-
9. It is plausible that the correlation of total payroll and employment would be higher if they were
measured from a common data source.11
Recall that we measured wages as total payroll from workers’ compensation records divided by
employment data from D&B. Keeping in mind all the caveats necessary for such a measure, we see that
annual wages grew to be about 7.5% higher at ISO firms than at their matches (column 5), which supports
Hypothesis 3 (and thus refutes Hypothesis 3′ that ISO adoption lowers wages). Although the point
11 In addition, because ISO adopters had lower death rates, the results on sales, employment, and total payroll would
be slightly more positive for ISO adopters if we included non-surviving firms in the analysis.
Quality Management and Job Quality
19
estimates decline over time, we cannot reject that the coefficients are constant over time (column 6).
Unlike the previous outcomes, our estimates of the effect of ISO 9001 on injury rates and costs are
conditioned on employment, payroll, sales, and average occupational riskiness. We find that trends in the
total value of injury costs (columns 7 and 8) and average cost per injury (columns 9 and 10) are unrelated
to ISO 9001 certification (among pairs of firms where both had at least one injury).
The number of injuries, a count variable, is analyzed with a negative binomial regression model with
conditional firm effects. For technical reasons, the estimation sample of this negative binomial model uses
only firms with a positive number of injuries in at least one year. To ensure an appropriate comparison,
we included in this sample only those pairs of employers where the adopter and matched non-adopter
both reported at least one injury. The results indicate that adoption did not predict a higher or lower
number of injuries (Table 6, columns 11 and 12, conditioning on employment and other control
variables).
As an extension, we also ran a probit model to predict which employers reported zero injuries
(technically, zero workers compensation claims) in all years after the match year, using the full set of 471
matched pairs of firms. To do so, we collapsed our panel data into a cross section, coding the dependent
variable “1” for firms that reported no injuries all years after the match year, and “0” for firms that
reported at least one injury after the match year. We included the following as controls: the log of each
firm’s average employment and average payroll for post-match years, region dummies, and industry
dummies. Recall that the matched set of adopters and non-adopters used in this regression (and all
regressions reported in Tables 5 and 6) had similar (and statistically indistinguishable) injury rates during
the years prior to the match. In contrast to the negative binomial results discussed above, the results of
this cross-sectional probit analysis indicate that ISO 9001 adopters were subsequently 5 percentage points
more likely to report no injuries (Table 7), a large effect given that the mean of the dependent variable is
28%. Taken as a whole, these three sets of results provide limited support for Hypothesis 4 that ISO
adoption lowers injury costs and rates, but no support for Hypothesis 4’ that ISO worsens these measures
of occupational safety and health.
Quality Management and Job Quality
20
As an extension, we examined whether adopters and non-adopters subsequently differed in two
particular types of injury rates: serious sudden-onset injuries (as typically occur following industrial
accidents) and, separately, serious cumulative injuries (which could result from more repetitive work).
This sub-analysis is motivated by the concern that ISO adoption may lead to more repetitive work and,
thus, more cumulative injuries. WCIRB data only categorizes “serious” injuries, which are those
associated with at least $2,000 in costs. These conditional fixed effects negative binomial models include
controls for employment, payroll, sales, average occupational riskiness, calendar years, and dummies to
denote the number of years until or since the match year. The results (not shown to conserve space)
indicate no change in sudden-onset injury rates (b = -0.064; SE = 0.087) between adopters and their
matched controls. However, the regression predicting the number of serious cumulative injuries yields an
incident rate ratio of 0.61 (b = -0.495; SE = 0.314) for has adopted ISO 9001, which is economically
meaningful decline among adopters, but not statistically significant. Further research is warranted to
explore the circumstances under which ISO 9001 adoption may help reduce serious cumulative injuries.
The workers’ compensation data show a small shift to safer occupations at firms that become ISO
9001 certified. That is, the average worker in a post-certification firm works in an occupation for which
workers’ compensation costs are almost 5% lower than at the comparison firm (b = -0.047, SE = 0.009, p
< 0.01) (column 13 of Table 6).
6. Conclusion
Our results are readily summarized:
• ISO adopters had far lower organizational death rates than matched firms within their industries.
• Sales and employment grew substantially more rapidly post certification at ISO 9001 adopting
firms than at matched firms.
• Total payroll and (to a lesser extent) annual earnings per employee grew substantially more
rapidly post certification at ISO 9001 adopting firms than at matched firms.
• ISO 9001 adopters already had slightly lower than average injury costs at the time of adoption,
Quality Management and Job Quality
21
and we found no evidence that this gap widened or narrowed after adoption. Adopters were more
likely to report no injuries for workers’ compensation at all in the years following adoption.
When comparing pairs of adopters and matched comparison firms that each had a positive
number of injuries, we found no differences in their number of injuries.
Our finding that ISO 9001 certification benefits employers bolsters prior research that reported other
such benefits associated with implementing ISO 9001 (e.g., Corbett et al. 2005; Terlaak and King 2006)
and TQM as well as winning quality awards (e.g., Hendricks and Singhal 1996, 1997, 2001b). Our results
are particularly credible because we analyze a larger sample of ISO certifications than almost any
previous study, we have performance data at the workplace level (unlike many previous studies that study
how ISO certification at a single plant affects financial performance at a multi-plant organization), we
measures performance using third-party data (rather than self-reported data), and we develop carefully
matched sets of non-adopters. Our results regarding the benefits of ISO 9001 certification for
employment, payroll, and average annual earnings are new.
A concern with the causal interpretation of these results is that employers with better growth
prospects might both adopt ISO 9001 and have higher post-adoption growth rates. We have two reasons
to doubt the importance of this alternative causality. First, we control for employer fixed effects, industryspecific
time trends, and a host of observable characteristics (via the matched comparison group). Second,
although we do not match on pre-adoption trends in sales in ISO adopters and comparison firms, the preadoption
trends are quite similar. Thus, we have no evidence that adopting firms had better growth
prospects prior to ISO 9001 adoption.
Some critics of ISO 9001 and related programs have expressed concern that benefits to employers
derive largely from the deskilling and routinizing of tasks. They hypothesize that employer gains come at
the expense of employees’ earnings. Our results showing that total payroll rises even faster than
employment, which implies an increase in average annual earnings, do not provide evidence of deskilling.
Our results have implications for managers, organizational scholars, and public policy. For managers,
the lessons are that the process of ISO 9001 certification appears to be valuable to most adopters. We
Quality Management and Job Quality
22
cannot be sure how broadly these lessons apply to non-adopters, but the extremely large benefits of
adoption (e.g., roughly 10% increases in sales) suggest that far more employers could benefit from ISO
9001 adoption than currently have.
The extremely large benefits of adoption also have lessons for organizational scholars. Some
academics have criticized quality programs such as ISO 9000 as management fads that are unlikely to
help the employer or employees (see Abrahamson 1996 and the citations in Staw and Epstein 2000).
Fashion may well play a role in the adoption of many management practices, but our results indicate that
ISO 9000 appears to deliver value for many organizational stakeholders.
We would not anticipate the large benefits we measure if potential customers could already see
product or service quality, if managers already understood how to achieve higher productivity and quality
cost-effectively, and if managers could capture the returns to any improvements in quality or productivity
(Levine 1995, ch. 3). The large increases in employment, total payroll, and sales we estimate are
consistent with at least one of these market imperfections slowing the spread of quality programs. As
such, these results also support arguments that public policy should promote quality programs; for
example, by subsidizing employee training or educating managers about the value of quality programs
(e.g., Helper and Levine 1995).
7. References
Abrahamson, E. 1996. Management fashions. Academy of Management Review 21(1): 254-285.
Adler, P. S., B. Goldoftas, D. Levine. 1997. Ergonomics, employee involvement, and the Toyota
Production System: A case study of NUMMI's 1993 model introduction. Industrial and Labor
Relations Review 50(3): 416-437.
Barbeau, E., C. Roelofs, R. Youngstrom, G. Sorensen, A. Stoddard, A. D. LaMontagne. 2004.
Assessment of occupational safety and health programs in small businesses. American Journal of
Industrial Medicine 45:371-379.
Barber, B. M., J. D. Lyon. 1996. Detecting abnormal operating performance: The empirical power and
specification of test statistics. Journal of Financial Economics 41(3): 359-399.
Benner, M.J., F.M. Veloso. 2008. ISO 9000 practices and financial performance: A technology coherence
perspective. Journal of Operations Management 26: 611–629.
Brenner, M. D., D. Fairris, J. W. Ruser. 2004. ‘Flexible’ work practices and occupational safety and
health: Exploring the relationship between cumulative trauma disorders and workplace
Quality Management and Job Quality
23
transformation. Industrial Relations 43(1): 242-266.
Conti, R., J. Angelis, C. Cooper, B. Faragher, C. Gill. 2006. The effects of lean production on worker job
stress. International Journal of Operations & Production Management 26(9): 1013-1038.
Corbett, C. J., A. M. Luca, H-N Pan. 2003. Global perspectives on global standards: A 15-economy
survey of ISO 9000 and ISO 14000. ISO Management Systems (Jan-Feb).
http://www.iso.org/iso/en/iso9000-14000/articles/surveysandoverviews.html.
Corbett, C. J., M. J. Montes-Sancho, D. A. Kirsch. 2005. The financial impact of ISO 9000 certification
in the United States: An empirical analysis. Management Science 51(7): 1046-1059.
Dehejia, R. H., S. Wahba. 1999. Causal effects in nonexperimental studies: Reevaluating the evaluation
of training programs. Journal of the American Statistical Association 94(448): 1053-1062.
Docking, D. S., R. J. Dowen. 1999. Market interpretation of ISO 9000 registration. Journal of Financial
Research 22(2): 147-160.
Easton, G. S., S. L. Jarrell. 1998. The effects of Total Quality Management on corporate performance: An
empirical investigation. Journal of Business 71(2): 253-307.
Eichler, M., M. Lechner. 2002. An evaluation of public employment programmes in the East German
state of Sachsen-Anhalt. Labour Economics 9: 143-186.
Fairris, D., M. D. Brenner. 2001. Workplace transformation and the rise in cumulative trauma disorders:
Is there a connection? Journal of Labor Research 22(1): 15-28.
Ferguson, W. 1996. Impact of the ISO 9000 series standards on industrial marketing. Industrial
Marketing Management 25(4): 305-310.
Häversjö, T. 2000. The financial effects of ISO 9000 registration for Danish companies. Managerial
Auditing Journal 15(1/2): 47-52.
Heckman, J. J., H. Ichimura, P. E. Todd. 1997. Matching as an econometric evaluation estimator:
Evidence from evaluating a job training programme. The Review of Economic Studies 64(4): 605-
654.
Heckman, J. J., H. Ichimura, P. Todd. 1998. Matching as an econometric evaluation estimator. The
Review of Economic Studies 65(2): 261-294.
Helper, S., D. I. Levine. 1995. A quality policy for America. Contemporary Policy Issues 13(2): 26-37.
Helper, S., D. I. Levine, E. Bendoly. 2002. Employee involvement and pay at American auto suppliers.
Journal of Economics & Management Strategy 11(2): 329-377.
Hendricks, K. B., V. R. Singhal, 1996. Quality awards and the market value of the firm: An empirical
investigation. Management Science 42(3): 415-436.
Hendricks, K. B., V. R. Singhal. 1997. Does implementing an effective TQM program actually improve
operating performance? Empirical evidence from firms that have won quality awards.
Management Science 43(9): 1258-1274.
Hendricks, K. B., V. R. Singhal. 2001a. Firm characteristics, Total Quality Management, and financial
performance: An empirical investigation. Journal of Operations Management 19: 269-285.
Hendricks, K. B., V. R. Singhal. 2001b. The long-run stock price performance of firms with effective
Quality Management and Job Quality
24
TQM programs as proxied by quality award winners. Management Science 47(3): 359-368.
Heras, I., P. M. Gavin, M. C. Dick. 2002. ISO 9000 registration’s impact on sales and profitability: A
longitudinal analysis of performance before and after certification. International Journal of
Quality Reliability Management 19(6): 774-791.
Huarng, E., C. Horng, C. Chen. 1999. A study of ISO 9000 process, motivation, and performance. Total
Quality Management 10(7): 1009-1025.
King, A. A., M. J. Lenox. 2001. Lean and green? An empirical examination of the relationship between
lean production and environmental performance. Production and Operations Management 10(3):
244-256.
Koc, T. 2007. The impact of IS0 9000 quality management systems on manufacturing. Journal of
Materials Processing Technology 186: 207-213.
Landau, K., H. Peters. 2006. Ergonomic demands in automotive component inspection tasks.
Occupational Ergonomics 6: 95-105.
Levine, D. I. 1992. Can wage increases pay for themselves? Tests with a productive function. Economic
Journal 102: 1102-1115
Levine, D. I. 1995. Reinventing the Workplace. Washington DC: The Brookings Institute.
Lindbeck, A., D. A. Snower. 1986. Wage setting, unemployment and insider-outsider relations. American
Economic Review. 76: 235-239.
Marcus, A. A., M. L. Nichols. 1999. On the edge: Heeding the warnings of unusual events. Organization
Science 10(4): 482-499.
McGuire, S. J., D. M. Dilts. 2008. The financial impact of standard stringency: An event study of
successive generations of the ISO 9000 standard. International Journal of Production Economics
113: 3-22.
Naveh, E., M. Erez. 2006. Innovation and attention to detail in the quality improvement paradigm.
Management Science 50(11): 1576-1586.
Naveh, E., A. Marcus. 2007. Financial performance, ISO 9000 standard, and safe driving practices effects
on accident rate in the U.S. motor carrier industry. Accident Analysis & Prevention 39(4): 731-
742.
O’Connor, P. 2005. Do management standards bring success? ISO 9000 firms grew rapidly in Oregon.
Oregon Labor Market Information System, October 27.
http://www.qualityinfo.org/olmisj/ArticleReader?itemid=00004604#seg0004, accessed March 16,
2008.
Rosenbaum, P. R., D. B. Rubin. 1983. The central role of the propensity score in observational studies for
causal effects. Biometrika 70(1): 41-55.
Sharma, D. S. 2005. The association between ISO 9000 certification and financial performance. The
International Journal of Accounting 40: 151-172.
Simmons, B. L., M. A. White. 1999. The relationship between ISO 9000 and business performance: Does
registration really matter? Journal of Managerial Issues 11(3): 330-343.
Smith, J. A., P. E. Todd. 2005. Does matching overcome LaLonde's critique of nonexperimental
Quality Management and Job Quality
25
estimators? Journal of Econometrics 125(1-2): 305-353.
Spence, M. 1973. Job market signaling. Quarterly Journal of Economics 87: 355-374.
Staw, B. M. and L. D. Epstein. 2000. What bandwagons bring: effects of popular management techniques
on corporate performance, reputation, and CEO pay. Administrative Science Quarterly 45 (3):
523-556
Terlaak, A. A., A. A. King. 2006. The effect of certification with the ISO 9000 Quality Management
Standard: A signaling approach. Journal of Economic Behavior & Organization 60: 579-602.
Toffel, M. W. 2000. Anticipating greener supply chain demands: One Singapore company's journey to
ISO 14001. In R. Hillary (Ed.), ISO 14001: Case Studies and Practical Experiences. Sheffield,
UK: Greenleaf Publishing.
Wayhan, V. B., E. L. Balderson. 2007. TQM and financial performance: What has empirical research
discovered? Total Quality Management 18(4): 403-412.
Williamson, A., S. Boufous. 2007. A data-matching study of the role of fatigue in work-related crashes.
Transportation Research: Part F 10(3): 242-253.
Wokutch, R. E. 1992. Worker Protection, Japanese Style. Ithaca, NY: ILR Press
