A LONGITUDINAL ANALYSIS OF THE MOTOR CARRIER INDUSTRY—THREE
ESSAYS
By
William A. Muir

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Business Administration – Logistics – Doctor of Philosophy
2017

ABSTRACT
A LONGITUDINAL ANALYSIS OF THE MOTOR CARRIER INDUSTRY—THREE
ESSAYS
By
William A. Muir
Motor carriage is the backbone of supply chains and a major component of the U.S. economy.
The industry has endured remarkable levels of external pressure over the last three decades,
including major shifts in shipper expectations, volatility and uncertainty in factor markets
and changing governmental regulations. These pressures continue to transform the industry.
This dissertation examines the motor carrier industry from three distinct perspectives: as
an economic agent, as a productive entity and as a principal to public safety.
Essay 1 examines longitudinal change in industry-level aggregate inventories, linking
inventory investment to prior-period changes in freight cost. A dynamic model is specified
and, using economic indicators and time-series inventory data on forty-five industry sectors,
polynomial distributed lags are estimated. The results show that aggregate inventories
adjust in the directions expected under optimal lot size ordering, however these adjustments
are protracted over several quarters.
Essay 2 links the diversification of motor carrier production activities within the truckload
and less-than-truckload segments to operational efficiency. The effect of output diversification
on carrier utilization of productive resources is theorized to depend on the degree of
interdependence among work flows. A panel is developed with five years of archival data
and two-stage data envelopment analysis is used for estimation and theory testing. Results
are consistent with theory but also suggest structural differences between carrier classes.
Essay 3 draws on theories of organizational routines to link variability in motor carrier safety performance over time to longitudinal change in performance. A metaroutine,
International Organization for Standardization 9000, is hypothesized to simultaneously
reduce variability and catalyze change. A regulating effect on variability is supported by a
four-year panel of federal safety data and is conditioned by a carrier’s average level of safety
performance, such that unsafe carriers appear to benefit most from the metaroutine.

Copyright by
WILLIAM A. MUIR
2017

TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vi

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii

KEY TO ABBREVIATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

viii

ESSAY 1

LOT SIZE ADJUSTMENTS AND THE DYNAMIC
AGGREGATE INVENTORIES . . . . . . . . . . .
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Literature Review . . . . . . . . . . . . . . . . . . . . .
1.3 Theory and Hypotheses . . . . . . . . . . . . . . . . . .
1.3.1 Economic Order Quantity . . . . . . . . . . . . .
1.3.1.1 Demand for Inventory . . . . . . . . . .
1.3.1.2 Inventory Re-ordering Costs . . . . . .
1.3.1.3 Inventory Holding Costs . . . . . . . . .
1.3.2 Inventory Adjustment Speed . . . . . . . . . . .
1.4 Data Sources and Variable Operationalization . . . . . .
1.4.1 Overview of Modeling Approach . . . . . . . . .
1.4.2 Sample Selection . . . . . . . . . . . . . . . . . .
1.4.3 Variable Operationalization . . . . . . . . . . . .
1.4.3.1 Dependent Variable . . . . . . . . . . .
1.4.3.2 Focal Predictors . . . . . . . . . . . . .
1.4.3.3 Covariates . . . . . . . . . . . . . . . .
1.5 Model Specification . . . . . . . . . . . . . . . . . . . . .
1.5.1 Contemporaneous Model . . . . . . . . . . . . .
1.5.2 Dynamic Model . . . . . . . . . . . . . . . . . . .
1.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6.1 Model Testing . . . . . . . . . . . . . . . . . . .
1.6.2 Magnitude and Pattern of Adjustment . . . . . .
1.7 Managerial Implications . . . . . . . . . . . . . . . . . .
1.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . .
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . .

UPDATING OF
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

1
2
4
7
7
8
8
9
10
13
13
13
14
14
16
17
19
20
20
21
23
23
25
27
30

PRODUCTION POSSIBILITIES, OUTPUT DIVERSIFICATION
AND OPERATIONAL EFFICIENCY IN MOTOR CARRIAGE . . .
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Industry Deregulation . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.2 The ‘Returns to Scale’ Controversy . . . . . . . . . . . . . . . . . .
2.2.3 Operational inefficiency and the Role of Carrier Segment Focus . .
Theory and Hypothesis Development . . . . . . . . . . . . . . . . . . . . .
Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1 Data and Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.2 Variable Operationalization . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.

35
36
37
38
39
41
43
46
46
49

ESSAY 2
2.1
2.2

2.3
2.4

iv

2.4.2.1 Dependent Variable . . . . . . . . . .
2.4.2.2 Focal Predictor . . . . . . . . . . . . .
2.4.2.3 Covariates . . . . . . . . . . . . . . .
2.4.3 Statistical Analysis . . . . . . . . . . . . . . . .
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.1 Regions of Significance . . . . . . . . . . . . . .
2.5.2 Alternative Specifications . . . . . . . . . . . .
2.5.3 Robustness Tests . . . . . . . . . . . . . . . . .
2.5.3.1 Subsamples . . . . . . . . . . . . . . .
2.5.3.2 Censored Regression . . . . . . . . . .
2.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.1 Theoretical Contributions . . . . . . . . . . . .
2.6.2 Managerial Implications . . . . . . . . . . . . .
2.6.3 Limitations and Directions for Future Research
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . .
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

49
51
52
53
54
57
58
60
60
61
63
66
68
70
70
71

.
.
.
.
.
.
.

78
79
81
81
84
86
87

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

88
89
89
91
91
94
94
96
96
97
97
98
101
102
104
106
109

ESSAY 3

CONSISTENTLY SAFE: HOW ISO 9000 CERTIFICATION PROMOTES CONSISTENCY IN MOTOR CARRIER SAFETY . . . . .
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Motor Carrier Safety . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.2 ISO 9000 Series Quality Assurance . . . . . . . . . . . . . . . . . .
3.3 Theory and Hypothesis Development . . . . . . . . . . . . . . . . . . . . .
3.3.1 Structure: How ISO 9000 Promotes Consistency in Carrier Safety
3.3.2 Agency: How ISO 9000 Promotes Continuous Improvement in Carrier Safety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.1 Data and Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.2 Variable Operationalization . . . . . . . . . . . . . . . . . . . . . .
3.4.2.1 Dependent Variables . . . . . . . . . . . . . . . . . . . . .
3.4.2.2 Focal Predictor . . . . . . . . . . . . . . . . . . . . . . . .
3.4.2.3 Covariates . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.3 Model Specification . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.3.1 Contextual Factors . . . . . . . . . . . . . . . . . . . . .
3.4.3.2 Regression Equations . . . . . . . . . . . . . . . . . . . .
3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.1 Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.2 Post-Hoc Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.1 Managerial and Policy Implications . . . . . . . . . . . . . . . . . .
3.6.2 Limitations and Future Research . . . . . . . . . . . . . . . . . . .
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v

LIST OF TABLES

Table 1.1: Descriptive statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

Table 1.2: Industry subsectors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

Table 1.3: Results of AR(1) mixed effect model estimation. . . . . . . . . . . . . . .

22

Table 2.1: Descriptive statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

Table 2.2: Results of mixed effect model estimation. . . . . . . . . . . . . . . . . . .

55

Table 2.3: Monte Carlo simulation results. . . . . . . . . . . . . . . . . . . . . . . .

57

Table 2.4: Results of estimation for alternative models. . . . . . . . . . . . . . . . .

60

Table 2.5: Results of estimation for subsamples. . . . . . . . . . . . . . . . . . . . .

63

Table 2.6: Results of estimation for random-effects Tobit regression models. . . . .

64

Table 3.1: Summary of the motor carrier safety literature, 2007–2017. . . . . . . . .

83

Table 3.2: Descriptive statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

92

Table 3.3: Regression results for inconsistency and change in motor carrier safety
performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

99

Table 3.4: Bootstrapped intra-class correlations of annualized safety performance
measures. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

101

vi

LIST OF FIGURES

Figure 1.1: Impulse response for a one percent shock to demand (a), reorder cost
(b) and holding cost (c). . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

Figure 2.1: Conceptual model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

46

Figure 2.2: Distribution of estimated output-oriented DEA efficiency (operational
inefficiency) scores. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

50

Figure 2.3: Histogram of carrier output concentration within the less-than-truckload
segment (LTL concentration). . . . . . . . . . . . . . . . . . . . . . . . .

52

Figure 2.4: Model-estimated relationship between LTL concentration and production inefficiency, with 95% confidence bands. . . . . . . . . . . . . . . .

56

Figure 2.5: Johnson-Neyman plot of the regions of significance for the simple slope
of LTL concentration on operational inefficiency. . . . . . . . . . . . . .

58

Figure 2.6: Plots of the effects of LTL concentration on operational inefficiency for
the full sample (a) and subsamples of Class I (b) and Class II (c) motor
carriers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

62

Figure 3.1: Comparison of alternative consistency measures using unsafe driving
data for two carriers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

94

vii

KEY TO ABBREVIATIONS

AIC

Akaike Information Criterion

BASIC

Behavior Analysis and Safety Improvement Category

CFR

Code of Federal Regulations

CPI

Consumer Price Index

CSA

Compliance, Safety, Accountability

DEA

Data Envelopment Analysis

FMCSA

Federal Motor Carrier Safety Administration

HOS

Hours of Service

ICC

Intraclass Correlation Coefficient

ISO

International Organization for Standardization

L-Q

Linear-Quadratic

LTL

Less than Truckload

M3

Manufacturers’ Shipments, Inventories and Orders Survey

MCA

Motor Carrier Act of 1980

MRP

Material Requirements Planning

MRTS

Monthly Retail Trade Survey

MWTS

Monthly Wholesale Trade Survey

NAICS

North American Industry Classification System

OMCI

Office of Motor Carrier Information

PPI

Producer Price Index

RFID

Radio-Frequency Identification

SE

Standard Error

TL

Truckload

UD

Unsafe Driving

VM

Vehicle Maintenance

viii

ESSAY 1
LOT SIZE ADJUSTMENTS AND THE DYNAMIC UPDATING OF
AGGREGATE INVENTORIES

1

1.1

Introduction
Business inventories are a critical component of the economy, serving as a lagging

indicator of macroeconomic activity (The Conference Board, 2017) and moving procyclically
with economic cycles (Khan and Thomas, 2007; Kesavan and Kushwaha, 2014). In the U.S.,
total business inventories exceeded 1.8 trillion dollars in 2016 (U.S. Census Bureau, 2017),
a 30 percent increase from the October, 2006, levels reported by Shah and Shin (2007).
At the firm-level, inventory performance is a key driver of customer-focused and financial
outcomes. When compared against their peers, firms with poorer than average inventory
performance exhibit inferior stock market performance (Chen et al., 2007; Alan et al., 2014).
Excess inventories inflate holding expenses, risk product obsolescence and prompt price
discounting (Kesavan and Kushwaha, 2014). Out-of-stocks impact service levels directly,
resulting in sale deference, product substitution or defection (Musalem et al., 2010).
While normative inventory theory is frequently developed at the stock-keeping-unit level,
recent empirical inventory research has greatly improved our understanding of the aggregate
behaviors of inventories at the firm and industry levels. Empirical inventory models offer
utility as a means to reconcile mathematical models with real-world behavior (Rumyantsev
and Netessine, 2007), to investigate between-industry and between-firm heterogeneity in
response to phenomena (Bray and Mendelson, 2012; Kesavan and Kushwaha, 2014) and to
elucidate indicators of practical and theoretical importance (e.g., Gaur et al., 2005; Eroglu
and Hofer, 2011; Kesavan and Mani, 2013). As Rumyantsev and Netessine (2007) note,
empirical models of aggregate inventory can directly benefit firm managers and industry
analysts since these models may relax modeling assumptions frequently made in analytical
models, such as stationarity and exogeneity.
In a recent empirical study, Larson et al. (2015) explore the dynamics of firm-level
inventories and observe that current-period inventory investments are a function of priorperiod sales growth. This observation is directly at odds with theory underlying the adaptive
base stock policy which the authors seek to test (Larson et al., 2015). The delay that
they observe in the adjustment of inventory investment to demand has, in fact, been noted

2

elsewhere in the literature (see Bils and Kahn, 2000), but remains understudied. Indeed,
researchers have yet to widely incorporate ordering dynamics within empirical inventory
models. Nor has a dynamic response to many other key determinants of inventory investment
been investigated within the literature.
The purpose of this paper is to further investigate the phenomena identified by Larson
et al. (2015). Using a panel of aggregate inventories at the industry-level, we show that
business inventories are elastic not only to prior-period changes in sales, but also to priorperiod changes in costs associated with placing re-orders and holding inventory. Our results
show that aggregate inventories adjust in the directions expected under optimal lot size
ordering, however these adjustments are protracted over several quarters. For adjustments
resulting from changes to physical inventory holding costs, the updating process is delayed for
up to three months. Taken together, these results suggest that institutional and individual
processes and biases attenuate timeliness in the acquistion and processing of information
and the execution of replenishment orders. We posit that many of the biases underlying this
protraction are the same as those identified in laboratory settings by Schweitzer and Cachon
(2000) and subsequent studies of single-period newsvendor ordering behavior.
Further, this study shows that the predictive power of a model accounting for dynamic
updating is remarkably stronger than that of a model considering only contemporaneous
effects of regressors. Competing AR(1) models are estimated using a panel of time-series
data from business inventories in the manufacturing, wholesaling and retailing sectors. After
removal of seasonal effects, and after controlling for macroeconomic factors, contemporaneous
effects of focal regressors (demand, re-order cost, inventory holding cost) explain less than onetenth of the variation in year-over-year changes in aggregate inventory investments between
1993 and 2016. In contrast, under a dynamic specification with polynomial distributed lags
up to twelve months, these same focal predictors explain nearly half of the variation in
year-over-year changes in aggregate inventory investments during this same period. Impulse
response functions suggest that inventory updating concludes within ten months; we theorize
that strategic responses, which are not captured by the model, may extend further in time.

3

Implications are drawn for inventory managers, policy makers and researchers.
The remainder of the paper is structured as follows. First, the empirical inventory
management literature is reviewed. A subsequent section introduces a normative theory of
inventory updating behavior, which is followed by a discussion of organizational and individual
factors that might attenuate behavior under the normative model. From these, research
hypotheses are drawn. Section 4 provides an overview of the data and operationalization of
variables. Econometric specifications are developed in Section 5 and results are presented in
Section 6. The paper closes with implications for managers and a general discussion.

1.2

Literature Review
Explaining fluctuations in inventory investments remains critical to understanding

the business cycle as well as firm behavior. Within the classical production-smoothing
theory of inventory (Holt et al., 1960), inventories are held to smooth production under
a concave short-run production function (i.e., rising marginal costs), thereby improving
average productivity and reducing average production costs. Arising from this theory is the
linear-quadratic (L-Q) model, which has been regarded as the workhorse of the inventory
literature (Banerjee and Mizen, 2006), especially as it applies to the finished goods inventories
of manufacturers (Ramey and West, 1999). The L-Q model assumes that demand is linear and
that the costs of production and holding inventory are quadratic. Adjustments to inventories
are protracted because production costs are convex. In contrast, the flexible-accelerator
variant of the L-Q model, developed by Lovell (1961), assumes that inventory adjustments
incur some cost and, thus, firms close only a fraction of the gap between target and actual
inventory holdings in any period. Estimates of this fraction suggest that inventories adjust
quite slowly to target levels (Blinder, 1986; Bils and Kahn, 2000; Khan and Thomas, 2007);
adjustment rates reported in the literature range from three months (Maccini and Rossana,
1981) to more than one year (Feldstein et al., 1976). Speculation as to the cause of slow
adjustment under these models include the presence of institutional processes, warehousing
and personnel constraints, risk-aversion under uncertainty, contractual inflexibilities, long
lead times and the omission of relevant regressors (a thorough discussion is provided by

4

Feldstein et al., 1976).
Over time, many of the core tenants of production-smoothing theory have been shown to
be untenable (Cachon et al., 2007). Under production-smoothing theory and the L-Q model,
inventories should respond inversely to demand shocks (i.e., covary negatively) and move
countercyclically with the business cycle. These basic assumptions have been called into
question. For instance, Blinder (1986) and Khan and Thomas (2007) show that variation in
production exceeds that of sales and that the covariance between inventory investments and
sales is, in-fact, positive—findings which are directly at odds with a production-smoothing
motive. Further, numerous empirical studies have shown that inventory investments move
procyclically with the business cycle and not countercyclically as would be expected by
a pure production-smoothing motive (e.g., Wen, 2005; Khan and Thomas, 2007; Kesavan
and Kushwaha, 2014). In light of these shortcomings, significant efforts have been made
to better align the L-Q model and production-smoothing theory with the so-called stylized
facts of aggregate inventory behavior in practice. For example, Blinder (1986) argues that
supply shocks might account for observations of greater volatility in inventories than in
demand (c.f., Wen, 2005). Banerjee and Mizen (2006) relax the assumption of the L-Q
model that inventory and sales are linearly cointegrated and propose an I(2) model under the
presence of polynomial cointegration. However, given the lack of evidence for a productionsmoothing motive, and its narrow application, the extant empirical inventory research has
tended to abandon the L-Q model in favor of alternative frameworks. Seminal work is
summarized below.
Khan and Thomas (2007) consider inventory investment under generalized (S, s) inventory
policies, in which nonconvex delivery (fixed) costs give rise to inventories. Application of
this policy assumes that firms seek to minimize inventory costs, ordering some optimal lot
size S − s to return an inventory to S only when inventories are at a sufficiently low level
(s). A key feature of this theory is that its application is not limited to a narrow subset
of manufacturing inventories, as the production-smoothing motive is (Ramey and West,
1999), but may offer a general motive for business inventories including trade inventories

5

held by wholesale and retail firms (Mosser, 1991). In their model economy, Khan and
Thomas (2007) are able to reproduce many of the empirical findings that confound the
production-smoothing motive, including procyclical inventory investments and positive
covariation between inventories and demand. Importantly, Khan and Thomas (2007) provide
further evidence that inventories adjust slowly to targets, taking multiple quarters to reach
equilibrium in their model.
Gaur et al. (2005) and successive work (e.g., Rumyantsev and Netessine, 2007; Cachon
and Olivares, 2010; Larson et al., 2015) appeal to a broader class of management and
inventory theory to develop empirical models of inventory behavior. Accordingly, this
stream of research departs sharply from the comparatively narrow focus of the L-Q model,
investigating a wider set of response variates and covariates. Gaur et al. (2005) demonstrate
that a majority of variability in retail inventory turnover can be explained by between-firm
differences and within-firm changes over time in three factors: gross margin, capital intensity
and deviations of actual sales from sales forecasts. In explaining fluctuations in inventory
investments, Rumyantsev and Netessine (2007) consider firm size differences as well as
differences in margins, sales growth and demand uncertainty on inventory investments. They
further consider inventory holding costs in their model using the three-month treasury bill
rate as a proxy to represent the opportunity cost of holding capital, but find the effects of
holding cost to be confounded by trend. Cachon and Olivares (2010) show that finished goods
inventories of vehicle manufacturers scale with the increases to the downstream (dealership)
network but decline with gains in production flexibility of plants. A commonality among
this stream of literature is that, with only a few exceptions (e.g., Kesavan and Kushwaha,
2014), inventory adjustments are modeled contemporaneously to covariate shocks.
This study extends the findings of Larson et al. (2015), who examine the behavior of
aggregate inventories under an adaptive base stock policy. This policy relaxes an (S, s) policy
assumption of independent and identically distributed demands. To examine the degree
to which an adaptive base stock policy might explain inventory behavior in the practice,
the authors model quarterly inventory investments—purchases scaled by same-period cost

6

of goods sold—as a function of forecasted changes in sales. Underlying this model is a
key assumption that inventory investments are not a function of sales growth from prior
periods. Instead, Larson et al. (2015) find that greater sales growth in the previous quarter
attenuates the relationship between demand and inventory investment. This finding strongly
suggests that frictions exist over and above delivery lags that condition the timeliness of
inventory investments. We extend the findings of Larson et al. (2015) by explicitely modeling
the dynamics of inventory investment. Unlike previous work, we do not limit our dynamic
analysis to the demand distribution or assume that the rate of inventory updating is constant
over time (c.f., flexible-accelerator model). Instead, we rely on normative theory of optimal
ordering behavior and utilize a flexible functional form to investigate non-linear dynamic
inventory response to multiple predictors.

1.3
1.3.1

Theory and Hypotheses
Economic Order Quantity

Like the adaptive base stock policy research of Larson et al. (2015) and the (S, s) policy
research of Khan and Thomas (2007), we rely on normative theory of optimal ordering
behavior to examine inventories in aggregate. Within classical inventory theory, the economic
order quantity for an inventory minimizes the costs of re-ordering and holding inventory
given some level of demand. For a lot of Q units and demand rate D, the total of these costs
per unit of time (C) can be given as the sum of average costs of re-ordering (K) inventory
per unit of time (KSQ−1 ) and the average cost of holding (H) inventory per unit of time
(HQ2−1 ). An optimal order quantity (Q∗ ) is obtained by setting equal to zero the partial
derivative of this cost function with respect to quantity, which yields Q∗ = 2DKh−1


1/2

.

In this formulation, demand, re-order costs and holding costs are assumed stationary and
exogenous. Given these assumptions, the expected quantity of on-hand cycle stock (Qc ) for
an inventory observed at an arbitrary point in time (E [Qc ]) is .5Q∗ . For [D, K, h] ⊆ R+ ,
E [Qc ] can be rewritten in a logarithmic specification—in a form that is linear in parameters
and suggests optimal partial elasticities of inventory with respect to demand, transportation

7

cost and holding cost:
√
ln (E [Qc ]) = −ln( 2) + .5ln(D) + .5ln(K) − .5ln(h)

1.3.1.1

(1.1)

Demand for Inventory

According to Equation 1.1, a one percent increase in demand should result in an approximate1
one-half percent increase in the expected quantity of cycle stock, holding fixed re-order cost
and inventory holding cost constant. The association between demand and inventories has
been the subject of intense investigation, however the relationship has been specified in
log-log form only in a number of studies (Ballou, 1981; Eroglu and Hofer, 2011; Shan and
Zhu, 2013; Kesavan and Kushwaha, 2014). The partial elasticity of total inventories with
respect to demand depends on the magnitude of aggregate safety stocks held, but recent
empirical estimates of this parameter compare quite favorably to the coefficient provided in
Equation 1.1. For instance, the average partial elasticity reported by Kesavan and Kushwaha
(2014) is .45 for inventories in the United States and Shan and Zhu (2013) report a partial
elasticity of .65 for inventories in China.
H1 : The partial elasticity of aggregate inventories with respect to aggregate demand is
greater than zero.

1.3.1.2

Inventory Re-ordering Costs

Similarly, a one percent increase in re-order costs should result in an approximate one-half
percent increase in the expected quantity of cycle stock, holding demand and inventory
holding cost constant. Blinder and Maccini (1991) cite the existence of fixed re-order costs
as a primary piece of evidence against production-smoothing theory, but note a lack of
research emphasis on models considering these costs. Indeed, this lack of empirical evidence
remains almost three decades later, especially as it pertains to the cost of transporting
a data generating process given by Y = X β , β serves as a reasonable approximation
for the percent change in Y associated with a one percent change in X if β is small, as it
is here. An exact calculation of the percent change in Y given θ percent change in X is
eβln(1+.01θ) .
1 For

8

inventory, arguably the most volatile cost component of re-ordering. The scant evidence
available into this relationship can be found within the transportation literature and generally
supports a positive association between transportation costs and inventories (Swanson et al.,
2016), as would be expected under Equation 1.1. Swanson et al. (2016) demonstrate
that a cointegrating relationship exists between aggregate transportation expenditures and
aggregate inventory expenditures. They find that changes in transportation expenditures
positively influence inventory expenditures, but find no support for a reciprocal relationship,
suggesting exogeneity of the transportation cost component within Equation 1.1.
H2 : The partial elasticity of aggregate inventories with respect to re-order costs is greater
than zero.

1.3.1.3

Inventory Holding Costs

Lastly, Equation 1.1 suggests a negative partial elasticity of inventories with respect to
inventory holding cost. When demand and transportation cost are held constant, a one
percent increase in holding cost should result in an approximate one-half percent decrease
in the expected quantity of cycle stock. Rumyantsev and Netessine (2007) identify two
components of a firm’s inventory holding cost: a physical cost of carrying inventory and an
opportunity cost of committing capital. Firms with inadequate financial resources to carry
inventory may also incur financing charges (increasing holding costs) or may place orders at
less-than-optimal lot sizes (Buzacott and Zhang, 2004). The relationship between holding
cost and inventories has received mixed support in the literature. Overwhelmingly, the real
interest rate has been used to proxy for the cost of capital, albeit unsuccessfully (Jones
and Tuzel, 2013). This lack of association is perplexing since standard theory of monetary
policy predicts that changes to short-term interest rates should impact inventory investment
spending (Maccini et al., 2004). In one of these few studies that find evidence of this
relationship, Akhtar (1983) reports elasticities of inventory investments with respect to the
real interest rate ranging from −.09 to −.11, directionally correct but lower in magnitude
than would be expected under Equation 1.1. Chen et al. (2007) and Rumyantsev and

9

Netessine (2007) find an overall negative relationship, although neither study is able to
produce consistent results across industry sectors. Maccini et al. (2004) find no short-term
relationship and conclude instead that only a long-run relationship exists between inventories
and the real interest rate, which they argue occurs between changes of persistent regimes.
Larson et al. (2015) measure holding cost by risk of product obsolescence and demonstrate
that firms alter their inventory ordering behavior when faced with greater obsolescence risk.
Several studies, such as Rajagopalan (2013), have been unable to link retail inventories to
cost of capital. A search of the inventory literature does not reveal an empirical test of
the relationship between the physical cost of holding inventory and inventory investment
behavior. For this reason, and because of a lack of consistent findings on other measures,
the true overall effect of inventory holding cost on inventory behavior remains unclear.
H3 : The partial elasticity of aggregate inventories with respect to holding costs is less than
zero.

1.3.2

Inventory Adjustment Speed

Normative theories of inventory management largely assume that decision makers act
rationally to economize on cost (Equation 1.1) and to maximize the sum of discounted
future cash flows (as seen in the objective function of the L-Q model2 ) thus maximizing
profit (Ren and Croson, 2013). In reality, decision makers operate in uncertain and changing
environments and their decisions are frequently influenced by the presence of institutional
2 The

L-Q model also assumes that economic agents will economize on cost, given that
there is a trade-off between the costs of holding inventory and moving to a new level of
production. This can be seen in the following objective function, where t is time, b is the
discount rate, C is cost, Q is a product quantity, H is inventories, P is price, S is sales and
U is a cost shock, a0 , a1 , a2 are costs associated with changing production, producing and
holding inventory, and a3 is the inventory accelerator:
max limT →∞ Et

T
X

bj (Pt+j St+j − Ct+j )

j=0

(

s.t.

Ct
Qt

= 0.5a0 ∆Q2t + 0.5a1 Q2t + 0.5a2 (Ht−1 − a3 St )2 + Uct
= St + Ht − Ht−1

10

and individual processes and biases. For example, when placing orders in the single-period
newsvendor problem, decision makers frequently exhibit pull-to-center bias, anchoring order
quantity on prior demand distributions (Schweitzer and Cachon, 2000; Bolton et al., 2012).
One manifestation of this bias is a tendency to order at some level between that of mean
demand and the optimal, profit-maximizing quantity. In multi-period settings, the L-Q
model and flexible-accelerator variant assume that inventories adjust gradually toward target
levels, making only a partial adjustment in any period. Under an (S,s) policy an adjustment
is delayed at least until an inventory reaches its re-order point. While a certain degree of
gradual adjustment may be theoretically admissible, the sluggishness with which aggregate
inventories adjust in practice cannot be accounted for by normative models (Bils and Kahn,
2000). Further, the collective evidence suggests that influential processes and biases drive
inventory behaviors that deviate from those expected of optimal-seeking organizations and
agents.
The speed with which an individual or organization responds to new, important external
information depends on their ability to perform a series of tasks including scanning the
environment, processing or analyzing information and taking action (Thomas et al., 1993).
For instance, an individual’s cognitive inability to process information may account for some
of the pull-to-center bias observed in newsvendor experiments (Bolton et al., 2012). For
complex organizations, however, acting on new information typically infers that organizational
mechanisms are in-place and available to transfer information from scanners to analyzers
and from analyzers to action takers (Smith et al., 1991). These steps frequently involve
multiple organizational actors, each of whom may bring their own set of individual biases.
Inventory managers, for instance, may exhibit seemingly-irrational behavior such as avoiding
adverse information, procrastinating, interpreting or acting on information inconsistently or
overestimating their span of control (Gino and Pisano, 2008). These behaviors may slow
or stop the acquisition and processing of information or delay action. It is not surprising,
then, that speed of organizational response is not homogeneous across firms or within
industries—factors such as firm size (Chen and Hambrick, 1995), external orientation (Smith

11

et al., 1991) and capability to interpret information (Daft and Weick, 1984) may constrain
the timeliness of response.
Despite significant advances made in supply chain management over recent decades, it
remains unclear if the phenomenon of slow inventory updating has changed much, if at all,
since it was popularized by Feldstein et al. (1976). For instance, inventory record inaccuracies continue to persist well-beyond the advent of computerized inventory management
(DeHoratius and Raman, 2008). Musalem et al. (2010) contend that these innaccuracies have
resulting in defacto-periodic review systems where accurate inventory information is available
to firms only at discrete points on time. Managers may also choose not to follow automated
ordering systems, due to perceived or actual inadequacies in the systems or a misalignment
of incentives (van Donselaar et al., 2010). Further, the complexity of inventory management
has increased—proliferation of stocking-keeping-units along with increased demands from
businesses (e.g., just-in-time deliveries, real-time tracking) and consumers (e.g., service
levels) have placed immense pressure on inventories and the personnel who manage them.
Meanwhile, the repercussions of contractual inflexibilities with suppliers and third-party
logistics firms are magnified under global sourcing arrangements. Domestically, increased
fuel volatility in a now-deregulated trucking industry has led to speculation and risk-taking.
Fuel surcharges visible to carriers on invoices lag diesel prices by several weeks (Smith, 2016),
presenting conflicting informational clues to inventory managers. Taken together, evidence
of individual biases in policy updating and the continued presence of these institutional
factors suggest that the updating process should be protracted for aggregate inventories
across industries.
H4 : Updates of aggregate inventories to target levels will be protracted over multiple periods.

12

1.4
1.4.1

Data Sources and Variable Operationalization
Overview of Modeling Approach

An overview of the modeling approach is as follows. Contemporaneous and dynamic
specifications of Equation 1.1 are modeled within the general linear mixed effect modeling
framework. Dynamic effects are specified by constructing polynomial distributed lags up to
twelve months. An industry-level panel of aggregate inventories and sales is developed using
data from the U.S. Department of Commerce, Bureau of Economic Analysis. Re-ordering
costs and inventory holding costs are measured using producer price indices. Controls are
introduced to account for the common business cycle and other macroeconomic effects. After
removing seasonality and time-invariant effects through seasonal differencing, year-over-year
inventory investments are modeled as an AR(1) process.

1.4.2

Sample Selection

Consistent with prior research (e.g., Rajagopalan and Malhotra, 2001; Cachon et al., 2007;
Chen et al., 2007; Shah and Shin, 2007), we examine inventories aggregated to the industrylevel. Among the advantages of industry-level data is generalizability of findings across
industry sectors and the avoidance of small sample bias or idiosyncrasies that might accompany stock-keeping-unit level or firm-level panels. A potential disadvantage to the use
of industry-level inventory data is the possbility that certain inventory-level effects might
be masked or attenuated (Bray and Mendelson, 2012). For example, it is reasonable to
assume that inventories will face both idiosyncratic shocks and structural (aggregate) shocks.
Chen et al. (2007) compare industry-level inventory data gathered and reported by the
U.S. Census Bureau, like that used in this study, to firm-level inventory data obtained
through COMPUSTAT. They find inventory patterns between the two data sources to be
“reassuringly similar” (Chen et al., 2007, p. 435). Industries were selected at the three and
four digit levels of the North American Industry Classification System (NAICS), resulting in
forty-four industry subsectors (Table 1.2). Wholesalers were defined at the four digit level

13

Table 1.1: Descriptive statistics.
Variable‡

Mean S.D.

(1) Inventories
(2) Recession
probability
(3) Consumer price
index
(4) Trade balance
(5) Establishments
(6) Demand
(7) Reordering cost
(8) Holding cost

(1)

(2)

.03
.00

.09
.29

1.00
.24∗∗

.02

.01

.32∗∗

.32∗∗

.07
.00
.03
.04
.00

.20
.25∗∗
.02
.27∗∗
.11
.66∗∗
.03
.32∗∗
.02 −.04∗∗

−.09∗∗
.12∗∗
−.01
.28∗∗
.25∗∗

(3)

(4)

(5)

(6)

(7)

1.00
1.00
.47∗∗
−.01
.36∗∗
.69∗∗
.21∗∗

1.00
.02†
.45∗∗
.52∗∗
.02†

1.00
.17∗∗
.04∗∗
.04∗∗

1.00
.33∗∗
−.10∗∗

1.00
−.06∗∗

Notes: † p < .10, ∗ p < .05, ∗∗ p < .01 (two-tailed); ‡ all variables except recession probabilities log-transformed
and all variables seasonally-differenced

for consistency to manufacturers and retailers; three-digits wholesaling NAICS aggregate
across product types for merchant wholesalers of durable and non-durable goods. All data
were obtained in unadjusted form, as recommended by Enders (2014).

1.4.3

Variable Operationalization

This subsection provides an overview of the operationalization of variables used in the study.
Descriptive statistics for all variables are provided in Table 1.1.

1.4.3.1

Dependent Variable

Monthly inventory data were obtained from the United States Department of Commerce
(Census Bureau) for a twenty-three year period beginning June 1993, and ending June 2016.
As this study seeks to test for the presence of a common phenomenon across inventories, observations were collected on total inventories within industries (i.e., manufacturing inventories
included finished goods inventories as well as raw materials inventories and work-in-process
inventories; see also Cachon et al., 2007). Data on manufacturing inventories were collected
from the Census Bureau’s Manufacturers’ Shipments, Inventories and Orders Survey (M3).
Data on wholesale inventories were collected from the Census Bureau’s Monthly Wholesale
Trade Survey (MWTS). Data on retail inventories were collected from the Census Bureau’s

14

Table 1.2: Industry subsectors.
NAICS

Subsector

NAICS

Subsector

326xxx
327xxx
331xxx
332xxx
333xxx
334xxx
335xxx
336xxx
337xxx
339xxx

Plastics and rubber products
Nonmetallic mineral products
Primary metals
Fabricated metal products
Machinery
Computers and electronic products
Electrical equipment/appliances/components
Transportation equipment
Furniture and related goods
Miscellaneous durable goods

Wholesaling
4231xx Motor vehicle and motor vehicle parts/supplies
4232xx Furniture and home furnishings
4233xx Lumber and other construction materials
4234xx Professional and commercial equipment/supplies
4235xx Metals and minerals, except petroleum
4236xx Electrical and electronic goods
4237xx Hardware, plumbing and heating equipment/supplies
4238xx Machinery, equipment, and supplies
4239xx Miscellaneous durable goods

4241xx
4242xx
4243xx
4244xx
4245xx
4246xx
4247xx
4248xx
4249xx

Paper and paper products
Drugs and druggists’ sundries
Apparel, piece goods, and notions
Grocery and related products
Farm product raw materials
Chemicals and allied products
Petroleum and petroleum products
Beer, Wine, and distilled alcoholic beverages
Miscellaneous nondurable goods

Retailing
441xxx Motor vehicles and motor vehicle parts
4423 xxx Furniture, furnishings, electronics, and appliances
444xxx Building materials and garden equipment/supplies

445xxx
448xxx
452xxx

Food and beverage stores
Clothing and clothing accessory stores
General merchandise stores

Manufacturing
311xxx Food products
312xxx Beverage and tobacco products
313xxx Textile mills
314xxx Textile products
315xxx Apparel
316xxx Leather and allied products
321xxx Wood products
322xxx Paper products
323xxx Printing
324xxx Petroleum and coal products
325xxx Chemical products

15

and Monthly Retail Trade Survey (MRTS). Collectively, inventories held by these forty-four
subsectors were valued at 1.22 trillion dollars in June 2016, compared to total U.S. business
inventories of 1.81 trillion dollars for this same month.

1.4.3.2

Focal Predictors

Demand. Monthly sales data were obtained for each subsector in their unajusted form.
Consistent with previous studies, sales were used to proxy for demand (e.g., Rumyantsev
and Netessine, 2007; Bray and Mendelson, 2012; Kesavan and Kushwaha, 2014).
Re-ordering cost. The fixed portion of costs incurred when re-ordering inventory is predominantly comprised of two parts: personnel-related costs associated with preparing,
administering and inspecting re-orders and costs associated with the movement of inventory.
Recognizing that the latter of these two costs is the most volatile component of re-order
costs,3 we proxy transportation costs for re-ordering costs. We use the producer price
index (PPI) for less-than-truckload shipments of general freight. PPIs have previously been
utilized in the empirical inventory work (e.g., Chen et al., 2005) and measure the average
change over time in market prices observed by domestic producers for their output. The
less-than-truckload PPI was selected as carriers specializing in less-than-truckload shipments
face costs in addition to those faced by full truckload carriers (e.g., terminal operations,
break-bulk handling, operation of pickup and delivery vehicles). We later replace this PPI
with the index for full-truckload shipments and find the statistical results to be qualitatively
similar. The 2017 NAICS defines the less-than-truckload industry as follows:
This U.S. industry comprises establishments primarily engaged in providing longdistance, general freight, less than truckload (LTL) trucking. LTL carriage is characterized as multiple shipments combined onto a single truck for multiple deliveries within
a network. These establishments are generally characterized by the following network
activities: local pick-up, local sorting and terminal operations, line-haul, destination
sorting and terminal operations, and local delivery (U.S. Census Bureau, 2017).
3 In

an ordinary least squares regression, linear trend accounts for 99.39 percent of the
variability in the U.S. Social Security Administration’s National Average Wage Index (NAWI)
between calendar years 1992 and 2015, F (1, 22) = 3067, p < .01.

16

Inventory holding cost.

Despite decades of investigation, researchers have been unable

to consistently link fluctuations in the cost to finance inventory to aggregate inventory
investment behavior (Jones and Tuzel, 2013). Instead, we focus on the physical cost of
carrying inventory, a component of holding cost that is distinct from financing cost and an area
that remains understudied in the empirical inventory management literature (Rumyantsev
and Netessine, 2007). We measure changes in this cost, as faced by the average firm, using
the PPI for warehousing and storage services. The 2017 NAICS defines this industry as
follows:
Industries in the Warehousing and Storage subsector are primarily engaged in operating
warehousing and storage facilities for general merchandise, refrigerated goods, and
other warehouse products. These establishments provide facilities to store goods. They
do not sell the goods they handle. These establishments take responsibility for storing
the goods and keeping them secure. They may also provide a range of services, often
referred to as logistics services, related to the distribution of goods. Logistics services
can include labeling, breaking bulk, inventory control and management, light assembly,
order entry and fulfillment, packaging, pick and pack, price marking and ticketing,
and transportation arrangement. However, establishments in this industry group
always provide warehousing or storage services in addition to any logistic services.
Furthermore, the warehousing or storage of goods must be more than incidental to
the performance of services, such as price marking (U.S. Census Bureau, 2017).

1.4.3.3

Covariates

We introduce covariates into the model to control for potential confounds (e.g., common
trend, seasonal patterns, underlying processes and macroeconomic cycles) and to improve
statistical efficiency in estimation. Coviariates were selected based on a review of empirical
inventory models within the literature and are described in the following section.
Trend. Linear trend is included in the model to control for potential confounds associated
with time that might bias parameter estimates. Chen et al. (2005, 2007) and Rajagopalan
and Malhotra (2001) find evidence of a long-term decline in U.S. inventories that they
attribute to the adoption of new technology and inventory practices, such as just-in-time
production.

17

Seasonality. Month effects are mathematically removed through seasonal differencing. The
specification of the pre-differenced model includes month dummy variables with random
slopes for exposition.
Business Cycles.

Firms actively adjust inventory investments in response to economic

shocks (Kesavan and Kushwaha, 2014). We control for macroeconomic conditions using
smoothed continuous recession probabilities (Piger and Chauvet, 2017), which were obtained
from the U.S. Federal Reserve. Recession probabilities are developed using a Markovswitching dynamic factor model and indicate the probability that the U.S. economy is in a
recession or expansion at a given time (Chauvet, 1998).
Inflation. The Consumer Price Index (CPI) is included as an inflationary control since
currency inflation may confound the relationship between aggregate inventories and other
variables that are measured in dollars. The use of a price index is consistent with previous
studies (e.g., Chen et al., 2007; Kesavan and Kushwaha, 2014). Monthly CPI observations
were obtained from the U.S. Bureau of Labor Statistics.
Lead Time. Inventory theory suggests that safety stock levels depend on re-order lead time,
thus impacting aggregate inventory levels. Global sourcing typically increases re-order lead
times by 20 to 50 days, due to increased transit time and customs clearance (Jain et al.,
2014). Rajagopalan and Malhotra (2001) argue that, in the long-run, increased importation
has not only increased lead times but may prompt firms to add inventory as a buffer against
fluctuations in currency and other uncertainties associated with global sourcing. Indeed,
evidence suggests that firms importing goods observe longer lead times and hold more
inventory than firms who source domestically (Alessandria et al., 2010). To control for these
effects, we obtained data on the monthly U.S. trade balance of goods in its unadjusted form
from the U.S. Department of Commerce.
Industry Consolidation.

Lastly, we control for the level of consolidation within each

industry subsector using establishment data obtained from the U.S. Department of Labor’s
Quarterly Census of Wages and Employment. Firms should enjoy economies of scale in
inventory management due to decreased lot sizes (as seen in Equation 1.1), decreased

18

demand variability (Cachon and Olivares, 2010) and network advantages such as inventory
pooling (Lim et al., 2016). Establishment data were obtained in unadjusted form and monthly
estimates were interpolated separately for each subsector using cubic splines (Forsythe et al.,
1977).

1.5

Model Specification
This section specifies mixed-effect models used in hypothesis testing. A initial model

(not tested) is described in Equation 1.2, where the inventory level for subsector i at time t
depends on a subsector-specific deviation from its sector-level intercept; subsector-specific
effects of time and seasonality; and population-average effects of the business cycle (CYCLE),
the consumer price index (CPI), the trade balance (TRADE), the number of establishments
in the subsector (EST), the level of demand in the subsector (DEMAND), re-ordering
cost (REORDER) and holding cost (HOLDING). All continuous variables in the model
are log-transformed except time and CYCLE, the latter of which is in probability form
and contains non-positive observations. We scale CYCLE by .01 for interpretability in the
regression equation.

Level 1
Yit = β0i + β1i × t +

12
X

ψji × Mt +

j=1

3
X

πk × Si

(1.2)

k=2

+ β2 × CYCLEt + β3 × ln(CPI)t + β4 × ln(TRADE)t
+ β5 × ln(EST)it + β6 × ln(DEMAND)it
+ β7 × ln(REORDER)t + β8 × ln(HOLDING)t + Rit
Level 2
β0i = γ00 + U0i

(1.3)

β1i = γ10 + U1i

(1.4)

ψji = γj0 + Uji , ∀j = Feb, . . . , Dec

(1.5)

19

1.5.1

Contemporaneous Model

A contemporaneous model (Equation 1.7) is specified by taking a seasonal difference of
both sides of Equation 1.2. Stationarity is not a mathematical necessity in longitudinal
analysis of panel data since covariances in the residual variance-covariance matrix can be
estimated (Newsom, 2015). Nonetheless, the removal of seasonal components in time series
data is often desirable to eliminate common covariation and autocorrelations that may persist
at seasonal lags (Enders, 2014). After seasonal differencing, time-invariant effects—those
of month (ψji ), sector (πk ) and starting position (β0i )—are removed from the regression
equation. Subsector-specific effects of time (linear trend) are estimated as random intercepts
in the differenced equation. Since inventories are log-transformed (from Equation 1.1), the
seasonally differenced equation explains year-over-year percent changes in inventories.

Level 1
∆12 Yit = 12 × β1i + β2 × ∆12 CYCLEt + β3 × ∆12 ln(CPI)t

(1.6)

+ β4 × ∆12 ln(TRADE)t + β5 × ∆12 ln(EST)it
+ β6 × ∆12 ln(DEMAND)it + β7 × ∆12 ln(REORDER)t
+ β8 × ∆12 ln(HOLDING)t + Rit
Level 2
β1i = γ10 + U1i

1.5.2

(1.7)

Dynamic Model

To estimate the competing, dynamic specification (Equation 1.9), the contemporaneous
effects in Equation 1.7 of demand (β6 ), reorder cost(β7 ) and holding cost (β8 ) are replaced
by polynomial distributed lags. Lags up to twelve months are specified, following the advice
of Lovell (1961) that “the planning horizon of the firm is surely shorter than a year for
decisions involving output adjustment and inventory” (p. 294). Polynomials up to the second

20

degree where constructed as a balance between making potentially limiting assumptions
about the distribution of lagged effects and overfitting the model.

Level 1
∆12 Yit = 12 × β1i + β2 × ∆12 CYCLEt + β3 × ∆12 ln(CPI)t

(1.8)

+ β4 × ∆12 ln(TRADE)t + β5 × ∆12 ln(EST)it
+

2 X
12
X

αp × z p × ∆12 ln(DEMAND)it−z

p=0 z=0

+

2 X
12
X

δp × z p × ∆12 ln(REORDER)it−z

p=0 z=0

+

2 X
12
X

ωp × z p × ∆12 ln(HOLDING)it−z

p=0 z=0

+ Rit
Level 2
β1i = γ10 + U1i

1.6

(1.9)

Results
Models of year-over-year inventory investment were estimated using maximum-likelihood

with AR(1) residual structures. Four models were estimated: a model with only control
variables, the contemporaneous model, the dynamic model and a parsimonious dynamic
model. Parameter estimates and measures of model fit are listed in Table 1.3. Prior to
estimation, we tested for the presence of a unit-root in the response variable. The Levin-LinChu panel unit-root test (Levin et al., 2002) rejects the null hypothesis that the response
panel contains unit roots before subtracting out cross-sectional means (t∗ = −10.41, p < .01)
and after subtracting out cross-sectional means (t∗ = −11.97, p < .01), when lag lengths for
augmented Dickey-Fuller tests are selected using the Akaike information criterion (AIC).
As seen in Table 1.3, estimates of the AR(1) parameter range from .92 to .95; positive
log-likelihood values were estimated for all models.

21

Table 1.3: Results of AR(1) mixed effect model estimation.
Control
Response= ∆12 ln(Inventories)
Fixed effects
Intercept †
∆12 CYCLE
∆12 ln(CPI)
∆12 ln(TRADE)
∆12 ln(EST)
∆12 ln(DEMAND)

∆12 ln(REORDER)

∆12 ln(HOLDING)

Variance Components
Variance (eit )
Autoregressive (AR1)
Measures of fit
−2 Log likelihood
AIC
pR2 (Level 1)
Notes: ∗ p < .05,

∗∗

Contemporary

Dynamic (A)

Dynamic (B)

Est.

S.E.

Est.

S.E.

Est.

S.E.

Est.

S.E.

γ10
β2
β3
β4
β5
β6
α0
α1
α2
β7
δ0
δ1
δ2
β8
ω0
ω1
ω2

.015∗∗
.043∗∗
.576∗∗
.008∗∗
.404∗∗

.005
.003
.063
.002
.087

.014∗∗
.044∗∗
.452∗∗
−.002
.412∗∗
.073∗∗

.005
.003
.070
.002
.084
.005

.002
.028∗∗
.179∗
−.002
.334∗∗

.004
.004
.070
.002
.074

.002
.028∗∗
.180∗∗
−.002
.334∗∗

.004
.004
.068
.002
.074

.096∗∗
−.013∗∗
<.001∗∗

.004
.002
<.001

.096∗∗
−.013∗∗
<.001∗∗

.004
.002
.000

.057∗∗
−.006
<.001

.017
.007
.001

.057∗∗
−.006∗∗

.012
.002

−.020
−.015
.002∗

.026
.011
.001

−.020
−.015
.002∗

.025
.011
.001

σe2
ϕ

.008∗∗
.952∗∗

.057∗∗

−.075∗

<.001
.003

−48, 841.9
−48, 827.9
.13

.007∗∗
.947∗∗

.021

.030

<.001
.003

−49, 069.5
−49, 049.5
.22

.004∗∗
.920∗∗

<.001
.004

−49, 861.3
−49, 829.3
.52

.004∗∗
.920∗∗

<.001
.004

−49, 861.3
−49, 831.3
.52

2
2
2
p < .01 (two-tailed); † intercept scaled by 12; R2 for Level 1 approximated as (σnull
− σfull
)/σnull
(Bryk and Raudenbush, 2002)

22

1.6.1

Model Testing

The contemporaneous model exhibits improved fit over the control model (χ2∆ (3) = 227.6,
p < .01) and explains 22 percent of the Level 1 variability in year-over-year inventory
investments, when approximated using the method proposed by Bryk and Raudenbush
(2002). Coefficients for the effects of demand, re-ordering cost and inventory holding cost
are statistically significant and are in their hypothesized directions.
Fit of the dynamic model (A) is superior to that of the contemporaneous model, when
compared by AIC (∆ = −779.8). This model explains an additional 30 percent of the Level
1 variability in year-over-year inventory investments, for a total of 52 percent. Thus, the
data support Hypothesis 4 which suggest that aggregate inventories exhibit a protracted
adjustment to target levels. Prior to calculating partial elasticities, we prune the seconddegree polynomial of re-ordering cost since the coefficient (δ2 ) is not statistically different
than zero. This change results in a 2.0 point decrease in AIC without a change to loglikelihood (χ2∆ < .01, p > .05), suggesting improved fit. Model fit of this more parsimonious
dynamic model (B) is otherwise qualitatively identical to that dynamic model (A).

1.6.2

Magnitude and Pattern of Adjustment

Estimates and confidence intervals for the overall magnitude of adjustment (as partial
elasticities) and pattern of adjustment (impulse response) are constructed using the Monte
Carlo method. 10, 000 draws were taken from a multivariate normal distribution using
the model-produced mean vector and variance-covariance matrix of parameter estimates.
Equation 1.10 contains the formula used to recover point estimates of the partial elasticities
with respect to demand (α), re-order cost (δ) and holding cost (ω). Corresponding impulse
responses with 95% confidence intervals are depicted in Figure 1.1.
εi =

2 X
12
X

ip × z p , ∀i = α, δ, ω

(1.10)

p=0 z=0

Hypothesis 1 suggests that the partial elasticity of aggregate inventories with respect
to aggregate demand is greater than zero. Consistent with this hypothesis, a one percent

23

∆12 ln(Inventories)

0.12
0.09
0.06
0.03
Lag

0.00
0

1

2

3

4

5

6

7

8

9

10

11

12

7

8

9

10

11

12

7

8

9

10

11

12

−0.03
(a) Demand

∆12 ln(Inventories)

0.12
0.09
0.06
0.03
Lag

0.00
0

1

2

3

4

5

6

−0.03
(b) Reorder cost

∆12 ln(Inventories)

0.12
0.08
0.04
Lag

0.00
0

1

2

3

4

5

6

−0.04
−0.08
(c) Holding cost

Figure 1.1: Impulse response for a one percent shock to demand (a),
reorder cost (b) and holding cost (c).

24

change in aggregate demand is associated with an estimated .44 percent adjustment to
aggregate inventories [CIL = .40, CIU = .48] over the subsequent year. As shown in Figure
1.1, the impulse response to a demand shock is slightly convex, tapering off to zero at the
ten month mark.
Hypothesis 2 suggests that the partial elasticity of aggregate inventories with respect
to re-order costs is greater than zero. we also find support for this hypothesis. A one
percent change in re-order costs, as proxied by the PPI for less-than-truckload freight
shipments, is associated with an estimated .26 percent adjustment to aggregate inventories
[CIL = .11, CIU = .41] over the subsequent year. The impulse response to a re-order cost
shocks declines linearly over time and does not differ from zero after the eight months.
Lastly, Hypothesis 3 suggests that the partial elasticity of aggregate inventories with
respect to inventory holding costs is less than zero. This hypothesis was also supported
by the data. A one percent change in re-order costs, as proxied by PPI for warehousing
and storage services, is associated with an estimated −.31 percent adjustment to aggregate
inventories [CIL = −.52, CIU = −.10] over the subsequent year. The impulse response to a
holding cost shock is convex, with significant effects between two and eight months after the
shock.

1.7

Managerial Implications
These results carry important implications for policy makers, firm-level managers and

market analysts. At current inventory levels, a one percent increase in reorder costs results
in an additional 4.75 billion dollar investment in U.S. inventories within the following twelve
months. Our analysis focuses on the transportation component of these costs, of which fuel
costs accounted for 34 percent of the average marginal trucking cost per mile in 2014 (Torrey
and Murray, 2015). Public-sector officials considering legislation or policies that may impact
the transportation industry (i.e., fuel taxes, adjustments to strategic oil reserves, highway
tolls) should be aware of, and consider the potential effects on private inventories. Since
inventory is an investment of firm resources, both firms and economies experience opportunity
costs related to foregoing alternative resource allocations, such as investments in human

25

captial, physical capital or technology. As an example, a well-intentioned but short-sighted
increase to fuel tax by a local or state government to fund infrastructure projects may
channel firm resources into inventories and away from the local labor market.
The estimation of a convex impulse response to shocks in physical inventory holding
costs also carries significant implications for providers of transportation and warehousing
services. First, these results show that inventories decrease by nearly a third of one percent
in response to a one percent increase in the physical costs of holding inventory. However, the
effects of this price sensitivity may not be immediately apparent to service providers; it is
likely that contractual inflexibilities constrain the inventory updating process (Larson et al.,
2015). Thus, service providers should be careful not to draw conclusions from initial trends
or responses to rate changes until clients have had sufficient time to adjust inventories to a
new equilibrium. Further, our results suggest that these service providers must not insulate
themselves from rate information in the transportation and warehousing markets, even if
they do not provide a full suite of logistics services. Normative theory of optimal ordering
behavior posits that client firms seek to jointly economize on the costs of re-ordering and
holding inventory, implying a trade-off between these costs. Indeed, the findings of this
study are consistent with those expected under trade-off behavior. Thus, the quality of a
service provider’s forecast for client demand of logistics services may hinge, in part, on the
service provider’s ability to broadly forecast future rates in markets for logistics services.
Service providers offering integrated logistics services may hold an advantage in their ability
to control rates on both sides of the trade-off. For instance, an integrated service provider
desiring to maintain some maximum level of warehouse utilization may choose to buffer or
attenuate rising freight costs through pricing of their services.
A recent stream of research has explored the relationship between a firm’s inventory
leaness and financial and market performance (Eroglu and Hofer, 2011; Steinker and Hoberg,
2013; Alan et al., 2014). Our findings suggest that inventory leaness should not be viewed
narrowly as a relationship between inventory levels and demand, as has frequently been
done in the past. For instance, Eroglu and Hofer (2011) model firm’s inventory leanness to

26

sales and estimate an average elasticity of .91, nearly twice the value estimated in this study
and in other recent studies that consider covariates (e.g., Kesavan and Kushwaha, 2014)
and far greater than what would be expected under optimal ordering behavior. Indeed,
inventory theory strongly suggests that firms who respond myopically to changes in demand
will order at suboptimal quantities if other important factors do not remain stationary. Thus
it is possible that many of the firm-level inventory abnormalities found in previous studies
(i.e., deviations from inventory productivity curves) can be accounted for by variability
in re-order and holding costs—or these abnormalities may simply be transient artifacts of
protracted updating processes. This should give caution to firm managers, market analysts
and investors who are using firm-level inventory data to compare performance against peer
firms. It also suggests that firms who can react quickly to changing market conditions
may gain a cost advantage over other firms who are slower to acquire, process and act on
information from the external environment.

1.8

Discussion
This research makes several important contributions to the literature. This study extends

the finding of Larson et al. (2015) that inventory investments are a function of prior-period
sales. In line with the macroeconomic literature, we view this phenomenon as the protraction
of inventory updating processes. However, we show that protraction occurs across the broad
range of business inventories. Further, the phenomenon is not only driven by the demand
distribution, which might be inferred from Schweitzer and Cachon (2000) and subsequent
laboratory research in single-period problems—inventory updates are also protracted in
response to changes in re-order costs and inventory holding costs. This may explain why
previous studies have struggled to find a significant contemperaneous effect of holding cost on
inventories: effects on aggregate inventories are not felt until at least two months after a cost
shock. New micro-level theororizing is needed to account for these facts since the underlying
forces appear to be both cognitive and institutional in nature and are not well-known.
We believe this is the first study to test the partial elasticity of aggregate inventories
with respect to transportation costs. While the optimization literature is replete with

27

studies into the relationship between transportation cost and inventories (Ali and O’Connor,
2013), very little empirical evidence exists into the true nature of this relationship. We
treat transportation cost as a component of fixed reordering cost and demonstrate that
inventories respond strongly to variability in the freight market. These results suggest that
transportation costs should no longer be assumed away in inventory models; rather, there
may be a need to review and update existing empirical models to account for this effect. This
study is limited in that we only examine operational changes to inventories out to twelve
months and do not account for strategic change which may have occured at longer intervals
(i.e., due to sustained shifts in the transportation market), although Lovell’s (1961) cautions
on lags of greater than 12 months suggest we are safe in our assumption. Nonetheless, future
research should investigate changes to supply and distibution networks in response to the
price swings and general volatility in the post-recessionary market for freight tranportation
and its major cost components, such as fuel.
The literature has been keenly interested in assessing aggregate inventories for the
presence of long-term trend. Managerial practices, such as just-in-time production, and the
proliferation of technology (e.g., MRP, RFID) have enabled leaner inventories (Rajagopalan
and Malhotra, 2001). However, demands for greater product variety and service levels,
decreased product life cycles and globalization have placed conflicting pressures on firms to
bolster inventory holdings (Chen et al., 2007). Gaur et al. (2005) find an overall decrease in
inventory turnover within the retail sector between 1987 and 2000, after controlling for gross
margin, capital intensity and sales surprise. They find this effect to be more pronounced for
retailers with low capital intensity. In contrast, Chen et al. (2007) conclude that inventory
turnover improved in the retail sector beginning in mid-1990s, and document a steady decline
in inventories within other sectors (i.e., manufacturing, wholesaling) since the early 1980s.
Rajagopalan and Malhotra (2001) show that, at least for manufacturers, this trend may date
as far back as the 1960s. Our results indicate that an overall, long-term trend in aggregate
inventories is not present between 1993 and 2016. And, while we do not test individually
for the presence of trend within sectors and subsectors, subsector-level variability of linear

28

trend (modeled as a random intercept in the differenced equation) was tested for, and was
not found to be statistically significant. These findings do not rule out the presence of
non-linear trend, such as quadtratic trend or one or more stuctural breaks as may have
occurred during the 2007–2009 recession. However, after macroeconomic factors and the
order quantity regressors are held constant, there is no evidence of a level shift over time.
Lastly, these results strongly suggest, as others have noted, that economies of scale exist
in inventory management. Holding demand and other factors in the regression equation
constant, a one percent consolidation within an industry subsector is associated with a
one-third percent decrease in inventories. This is not entirely surprising given the vast
amount of research into optimization of supply chain networks. As firms grow, so should
their ability to pool inventories (Lim et al., 2016). However, the empirical evidence on scale
economies in inventory management is still in its infancy. Future research should consider
estimating inventory productivity functions to assess the extent to which economies are
enjoyed across product types (e.g., durables, non-durables) and between industry sectors.

29

REFERENCES

30

REFERENCES
Akhtar, M. (1983). Effects of interest rates and inflation on aggregate inventory investment
in the United States. The American Economic Review, 73(3):319–328.
Alan, Y., Gao, G. P., and Gaur, V. (2014). Does inventory productivity predict future
stock returns? A retailing industry perspective. Management Science, 60(10):2416–2434.
Alessandria, G., Kaboski, J. P., and Midrigan, V. (2010). Inventories, lumpy trade, and
large devaluations. The American Economic Review, 100(5):2304–2339.
Ali, A. I. and O’Connor, D. J. (2013). Using truck-inventory-cost to obtain solutions to
multi-period logistics models. International Journal of Production Economics, 143(1):144–
150.
Ballou, R. H. (1981). Estimating and auditing aggregate inventory levels at multiple
stocking points. Journal of Operations Management, 1(3):143–153.
Banerjee, A. and Mizen, P. (2006). A re-interpretation of the linear quadratic model when
inventories and sales are polynomially cointegrated. Journal of Applied Econometrics,
21(8):1249–1264.
Bils, M. and Kahn, J. A. (2000). What inventory behavior tells us about business cycles.
American Economic Review, 90(3):458–481.
Blinder, A. S. (1986). Can the production smoothing model of inventory behavior be saved?
The Quarterly Journal of Economics, 101(3):431–453.
Blinder, A. S. and Maccini, L. J. (1991). Taking stock: A critical assessment of recent
research on inventories. The Journal of Economic Perspectives, 5(1):73–96.
Bolton, G. E., Ockenfels, A., and Thonemann, U. W. (2012). Managers and students as
newsvendors. Management Science, 58(12):2225–2233.
Bray, R. L. and Mendelson, H. (2012). Information transmission and the bullwhip effect:
An empirical investigation. Management Science, 58(5):860–875.
Bryk, A. S. and Raudenbush, S. W. (2002). Hierarchical linear models for social and
behavioral research: Applications and data analysis methods. Sage, Newbury Park, CA.
Buzacott, J. A. and Zhang, R. Q. (2004). Inventory management with asset-based financing.
Management Science, 50(9):1274–1292.
Cachon, G. P. and Olivares, M. (2010). Drivers of finished-goods inventory in the U.S.
automobile industry. Management Science, 56(1):202–216.
Cachon, G. P., Randall, T., and Schmidt, G. M. (2007). In search of the bullwhip effect.
Manufacturing & Service Operations Management, 9(4):457–479.
Chauvet, M. (1998). An econometric characterization of business cycle dynamics with
factor structure and regime switching. International Economic Review, 39(4):969–996.

31

Chen, H., Frank, M. Z., and Wu, O. Q. (2005). What actually happened to the inventories
of American companies between 1981 and 2000? Management Science, 51(7):1015–1031.
Chen, H., Frank, M. Z., and Wu, O. Q. (2007). U.S. retail and wholesale inventory
performance from 1981 to 2004. Manufacturing & Service Operations Management, 9(4):430–
456.
Chen, M.-J. and Hambrick, D. C. (1995). Speed, stealth, and selective attack: How small
firms differ from large firms in competitive behavior. Academy of Management Journal,
38(2):453–482.
Daft, R. L. and Weick, K. E. (1984). Toward a model of organizations as interpretation
systems. Academy of Management Review, 9(2):284–295.
DeHoratius, N. and Raman, A. (2008). Inventory record inaccuracy: An empirical analysis.
Management Science, 54(4):627–641.
Enders, W. (2014). Applied econometeric time series. John Wiley & Sons, Hobuken, NJ.
Eroglu, C. and Hofer, C. (2011). Lean, leaner, too lean? The inventory-performance link
revisited. Journal of Operations Management, 29(4):356–369.
Feldstein, M., Auerbach, A., Hall, R. E., and Lovell, M. C. (1976). Inventory behavior
in durable-goods manufacturing: The target-adjustment model. Brookings Papers on
Economic Activity, 1976(2):351–408.
Forsythe, G. E., Moler, C. B., and Malcolm, M. A. (1977). Computer methods for
mathematical computations. Prentice-Hall, Englewood Cliffs, NJ.
Gaur, V., Fisher, M. L., and Raman, A. (2005). An econometric analysis of inventory
turnover performance in retail services. Management Science, 51(2):181–194.
Gino, F. and Pisano, G. (2008). Toward a theory of behavioral operations. Manufacturing
& Service Operations Management, 10(4):676–691.
Holt, C. C., Modigliani, F., Muth, J. F., and Simon, H. A. (1960). Production planning,
inventories, and workforce. Prentice-Hall, Englewood Cliffs, NJ.
Jain, N., Girotra, K., and Netessine, S. (2014). Managing global sourcing: Inventory
performance. Management Science, 60(5):1202–1222.
Jones, C. S. and Tuzel, S. (2013). Inventory investment and the cost of capital. Journal of
Financial Economics, 107(3):557–579.
Kesavan, S. and Kushwaha, T. (2014). Differences in retail inventory investment behavior
during macroeconomic shocks: Role of service level. Production and Operations Management,
23(12):2118–2136.
Kesavan, S. and Mani, V. (2013). The relationship between abnormal inventory growth and
future earnings for U.S. public retailers. Manufacturing & Service Operations Management,
15(1):6–23.

32

Khan, A. and Thomas, J. K. (2007). Inventories and the business cycle: An equilibrium
analysis of (S, s) policies. The American Economic Review, 97(4):1165–1188.
Larson, C. R., Turcic, D., and Zhang, F. (2015). An empirical investigation of dynamic
ordering policies. Management Science, 61(9):2118–2138.
Levin, A., Lin, C.-F., and Chu, C.-S. J. (2002). Unit root tests in panel data: Asymptotic
and finite-sample properties. Journal of Econometrics, 108(1):1–24.
Lim, M. K., Mak, H.-Y., and Shen, Z.-J. M. (2016). Agility and proximity considerations
in supply chain design. Management Science. Forthcoming.
Lovell, M. (1961). Manufacturers’ inventories, sales expectations, and the acceleration
principle. Econometrica: Journal of the Econometric Society, 29(3):293–314.
Maccini, L. J., Moore, B. J., and Schaller, H. (2004). The interest rate, learning, and
inventory investment. The American Economic Review, 94(5):1303–1327.
Maccini, L. J. and Rossana, R. J. (1981). Investment in finished goods inventories: An
analysis of adjustment speeds. American Economic Review Proceedings, 71(2):17–22.
Mosser, P. C. (1991). Trade inventories and (S, s). The Quarterly Journal of Economics,
106(4):1267–1286.
Musalem, A., Olivares, M., Bradlow, E. T., Terwiesch, C., and Corsten, D. (2010). Structural
estimation of the effect of out-of-stocks. Management Science, 56(7):1180–1197.
Newsom, J. T. (2015). Longitudinal structural equation modeling: A comprehensive introduction. Routledge, New York.
Piger, J. M. and Chauvet, M. (2017).
Smoothed U.S. recession probabilities
[RECPROUSM156N]. Retrieved from FRED, Federal Reserve Bank of St. Louis;
https://fred.stlouisfed.org/series/RECPROUSM156N, February 8, 2017.
Rajagopalan, S. (2013). Impact of variety and distribution system characteristics on
inventory levels at U.S. retailers. Manufacturing & Service Operations Management,
15(2):191–204.
Rajagopalan, S. and Malhotra, A. (2001). Have U.S. manufacturing inventories really
decreased? An empirical study. Manufacturing & Service Operations Management, 3(1):14–
24.
Ramey, V. A. and West, K. D. (1999). Inventories. In Taylor, J. B. and Woodford,
M., editors, Handbook of Macroeconomics Volume 1b, pages 863–923. Elsevier Science,
North-Holland, Amsterdam.
Ren, Y. and Croson, R. (2013). Overconfidence in newsvendor orders: An experimental
study. Management Science, 59(11):2502–2517.
Rumyantsev, S. and Netessine, S. (2007). What can be learned from classical inventory
models? A cross-industry exploratory investigation. Manufacturing & Service Operations
Management, 9(4):409–429.

33

Schweitzer, M. E. and Cachon, G. P. (2000). Decision bias in the newsvendor problem with
a known demand distribution: Experimental evidence. Management Science, 46(3):404–420.
Shah, R. and Shin, H. (2007). Relationships among information technology, inventory,
and profitability: An investigation of level invariance using sector level data. Journal of
Operations Management, 25(4):768–784.
Shan, J. and Zhu, K. (2013). Inventory management in China: An empirical study.
Production and Operations Management, 22(2):302–313.
Smith, J. (2016). Trucking industry hits road bump with rising diesel prices. Wall Street
Journal. Retrieved from https://www.wsj.com.
Smith, K. G., Grimm, C. M., Gannon, M. J., and Chen, M.-J. (1991). Organizational
information processing, competitive responses, and performance in the U.S. domestic airline
industry. Academy of Management Journal, 34(1):60–85.
Steinker, S. and Hoberg, K. (2013). The impact of inventory dynamics on long-term stock
returns–an empirical investigation of U.S. manufacturing companies. Journal of Operations
Management, 31(5):250–261.
Swanson, D., Williams, B. D., Gu, J., and Waller, M. A. (2016). Full steam ahead: Firms in
the U.S. economy adjust inventory for changes in transportation costs but not the reverse.
Transportation Journal, 55(3):282–295.
The Conference Board (2017). Economics Watch: Monthly Economic Series Report
(January, 2017). New York City, NY.
Thomas, J. B., Clark, S. M., and Gioia, D. A. (1993). Strategic sensemaking and organizational performance: Linkages among scanning, interpretation, action, and outcomes.
Academy of Management Journal, 36(2):239–270.
Torrey, W. F. and Murray, D. (2015). An Analysis of the Operational Costs of Trucking:
2015 Update. American Transportation Research Institute, Arlington, VA.
U.S. Census Bureau (2017). Manufacturing and trade inventories and sales: November,
2016 (Release Number CB17-08). Suitland, MD.
van Donselaar, K. H., Gaur, V., van Woensel, T., Broekmeulen, R. A., and Fransoo, J. C.
(2010). Ordering behavior in retail stores and implications for automated replenishment.
Management Science, 56(5):766–784.
Wen, Y. (2005). Understanding the inventory cycle. Journal of Monetary Economics,
52(8):1533–1555.

34

ESSAY 2
PRODUCTION POSSIBILITIES, OUTPUT DIVERSIFICATION AND
OPERATIONAL EFFICIENCY IN MOTOR CARRIAGE

35

2.1

Introduction
Corporate diversification has been regarded as one of the most important areas for

industrial research (Wan et al., 2011), with strong links to both accounting-based and
market-based measures of firm performance (Palich et al., 2000). Corporate incentives for
diversification include reduced risk exposure within external markets, presence of anti-trust
policies and curtailment of executive employment risk (Hoskisson and Hitt, 1990). Resourcebased theorists view managerial incentives to improve the utilization of productive services of
resources under the firm’s control as the most prominent motive for corporate diversification,
arguing that firm resources are only rarely fully utilized due to their indivisibility (Kor
et al., 2016). However, market failures (e.g., transaction costs, imperfect mobility) restrict
the firm’s ability to sell the unused services of resources (Peteraf, 1993). Thus, firms are
motivated to diversify into related business activities where resources and the services
they provide can be spread across multiple businesses (i.e., productive activities), thereby
improving resource utilization and the firm’s overall level of efficiency (Wan et al., 2011).
This research examines corporate diversification and its relationship to operational
efficiency within the U.S. motor carrier industry. Only a handful of studies have examined
motor carriers’ diversification of production activities (e.g., Stock, 1988; Smith et al., 1990;
Feitler et al., 1997; Hanna and Maltz, 1998; Pettus, 2001); similarly, its linkages to carrier
efficiency have received only passing attention (e.g., Silverman et al., 1997; Scheraga, 2011).
Yet, operational efficiency is a matter of prominent importance to motor carriers, having
been linked to firm survival in both the truckload (TL) segment (Fawcett et al., 2016, p. 23)
and in the less-than-truckload (LTL) segment (Nebesky et al., 1995, p. 571). Operational
efficiency of firms in the motor carrier industry is also a matter of economic importance.
Freight bills in the U.S. for over-the-road shipments totaled more than 700 billion dollars in
2015 (ATA, 2017), accounting for nearly four percent of the nation’s gross domestic product.
The industry is also one of the nation’s largest employers, with more than 1.4 million workers
employed in the U.S. alone, of which approximately 900 thousand workers are drivers (BLS,
2017).

36

Thus, the purpose of this study is to propose and test theory relating a motor carrier’s
corporate diversification to efficiency in carrier operations. We examine motor carriers that
have expanded from one of the two primary market segments for general freight carriage
(TL and LTL) into the other segment. We postulate that this diversification is driven
by managerial desire to improve the utilization of productive resources, but also suggest
that the success of this strategy is contingent on the direction of shifts in organizational
interdependence. We test our theory using a panel of 120 medium and large U.S. motor
carriers of general freight, with data obtained from governmental archives. Our methodology
consists of a two-stage analysis, whereby data envelopment analysis (DEA) is employed
in a first stage to estimate annual carrier technical efficiency scores over five measurement
occasions and, in a second stage, efficiency scores are regressed on contextual factors in
a mixed effects modeling framework. Our statistical model estimates our theorized cubic
polynomial relationship between a carrier’s segment focus and operational efficiency. Thus,
we produce estimates of operational efficiency across the production possibilities curve and
use these estimates to assess marginal effects from changes to segment focus.
The remainder of this essay is structured as follows. First, we review the literature on
motor carrier productivity, segment focus and operational efficiency. Following this review,
we present a theoretical framework for our study and formulate research hypotheses. Our
two-stage methodology is then presented in the subsequent section, along with the results
from hypothesis testing. Next, we provide a discussion of our findings and implications for
theory and motor carrier management. We close by addressing limitations of our study and
proposing several potential avenues for future research.

2.2

Literature Review
Our research examines the composition of a motor carrier’s core service offering (pro-

duction of TL and LTL carriage) and the impact of this diversification or focus strategy
on carrier efficiency. While issues adjacent to our topic such as motor carrier productivity
and returns (economies) of scale have received extensive treatment in the transportation
literature, attention to the issue of production efficiency has been very limited (Scheraga,

37

2011).1 Further, the literature has tended to bifurcate the motor carrier population such that
it has only been in rare instances that carrier production of TL and LTL service has been
treated as a continuum—indeed, we will later show that a non-trivial portion of large U.S.
motor carriers concurrently produce both TL and LTL carriage. To gain an understanding
of these two market segments and the differences between them, we review the literature on
the deregulated market for TL and LTL movements of general freight and strategic focus
by motor carriers on one or both of these sectors. In doing so, we pay special attention to
literature pertaining to output production and carrier utilization of productive resources
(i.e., labor, capital assets), since these areas form the basis for our research.

2.2.1

Industry Deregulation

Regulatory reform under the Motor Carrier Act of 1980 (MCA) sharply changed the U.S.
trucking industry, eliminating pricing controls, decreasing organization of labor (unionization)
among drivers, easing entry for new carriers and greatly increasing competition between
carriers (Belzer, 2000). Industry deregulation also reduced or eliminated many regulatory
roadblocks and restrictions that hindered carriers from further developing their resource
bases (Pettus, 2001) and entering new markets (Silverman et al., 1997). In doing so, a
greater degree of polarization emerged in the deregulated marketplace between carriers
offering truckload (TL) service–carriage of shipments generally weighing 10,000 or more
pounds directly from origin to destination—and those offering less-than-truckload (LTL)
service—carriage of shipments of less than 10,000 pounds typically via a hub-and-spoke
1 The

terms productivity and efficiency are frequently conflated in the popular press
but are distinct concepts (Coelli et al., 2005). Consistent with the literature, we define
productivity as the ratio of outputs produced to inputs used in production (McMullen
and Okuyama, 2000). Our use of the term efficiency refers to the technical efficiency—not
allocative efficiency—in production, represented by distance of some observed production
level from an efficient frontier (Farrell, 1957). If scale economies exist, a firm that is perfectly
technically efficient may be able to improve productivity by increasing scale. Conversely, if
diseconomies of scale exist then a firm that is technically efficient and increases scale may
realize a lower level of productivity while remaining technically efficient.

38

network.2 For instance, while the total number of motor carriers nearly tripled within the ten
years following industry deregulation (Silverman et al., 1997), the LTL segment experienced
a sizeable consolidation. For the portion of the LTL market comprised of medium and large
sized carriers, estimates placed the level of consolidation as high as 60 percent (Smith et al.,
1990; Nebesky et al., 1995). This LTL segment consolidation came in stark contrast to the
expectations of deregulation advocates, many of whom had anticipated market entry and
proliferation by small trucking firms within both industry segments (Rakowski, 1988; Belzer,
2000).

2.2.2

The ‘Returns to Scale’ Controversy

A common explanation for diverging market structures post-regulatory reform has been
the differences between the two segments in returns to scale (Rakowski, 1988; Belzer, 2000;
Giordano, 2008). Returns to scale refers to economies gained (or, if diseconomies exist,
relinquished) as carriers grow their operations. If economies exist (i.e., marginal productivity
increases and marginal cost decreases with size) then carriers are incentivized to grow and a
consolidation would be expected in the market (Belzer, 2000). Alternatively, if diseconomies
exist then carriers will have fewer incentives to scale their operations. The presence or
absence of scale economies in the trucking industry has been a long-standing matter of debate
among practitioners, industry analysts, regulators and academic researchers (Rakowski, 1988;
Belzer, 2000; Fawcett et al., 2016). Beyond the differences that exist between the two industry
segments, the discussion is further complicated by an often-implicit assumption that any
economies or diseconomies might be classified as being linearly homogeneous across levels of
carrier size—an assumption inconsistent with economic theory (Johnston and Ozment, 2013).
Unfortunately, findings from the estimation of economic production and cost functions have
2 Prior

to the MCA, the U.S. Interstate Commerce Commission avoided classifying carriers
between TL and LTL and instead categorized individual shipments (McMullen and Tanaka,
1995). The U.S. Department of Transportation, which now regulates the U.S. trucking
industry, still categorizes intercity shipments of general freight as TL or LTL but also
classifies carriers by their majority revenue commodity group, which includes TL or LTL
classifications for carriers of general freight (DOT, 2017).

39

been mixed even when this assumption is relaxed, as is most commonly accomplished within
a transcendental logarithm form or first-order Taylor series expansion of the Cobb-Douglas
production function.3 For example, in the LTL segment, researchers have reached differing
conclusions that carriers enjoy scale economies (e.g., Rakowski, 1988; McMullen and Tanaka,
1995; Giordano, 2008), are faced with diseconomies (e.g., Xu et al., 1994) or that neither
economies nor diseconomies exist (e.g., Nebesky et al., 1995). Conclusions from the TL
segment offer a similarly mixed picture (Fawcett et al., 2016).
Over time, research into scale economies in transport became increasingly nuanced;
analysts began fixing regressors such as operational and network characteristics and redefined
economies using terms such as returns to density (economies gained when operations are
scaled by networks remain fixed; Caves et al., 1984), economies of scope (economies gained
from network expansion; Jara-Díaz and Basso, 2003) and economies of integration (economies
of density combined with economies from vertical integration; Keeler, 1989). Given that the
underlying issue of scale returns has been (and remains) controversial and unresolved, the
motor carrier literature has occasionally disregarded these nuances as irrelevant (Rakowski,
1988) or simply confusing (Johnston and Ozment, 2013) to the core issue at hand.
Despite this long-standing lack of consensus on the topic, recent studies such as Fawcett
et al. (2016) and Jin et al. (2017) have tended to yield to the positive evidence that
scale economies might be available to at least some carriers within each of the segments.
Anecdotally, a recent flurry of mergers and acquisitions within the trucking industry also
seems to support this. For instance, in April 2017, the boards of directors for Knight
Transportation and Swift Transportation unanimously approved a merger of the two carriers,
creating the industry’s largest long-haul TL carrier (Tyler, 2017). Swift Transportation’s
Chairman stated that the merger will allow the two companies to “capitalize on economies
of scale to achieve substantial synergies” (Business Wire, 2017). A band of other mergers,
such as XPO Logistics’s 2015 acquisition of Con-way Freight, suggests that similar size
3 We

refer readers interested in the properties and restrictions of this function to Beattie et al. (2009) or, alternatively, to Creel and Farell (2001) for a discussion within the
transportation literature of the function’s assumptions and limitations.

40

advantages might also be present in the LTL segment.

2.2.3

Operational inefficiency and the Role of Carrier Segment Focus

One potential explanation for the mixed results reported in the returns-to-scale literature
lies in the literature’s pre-occupation with assessing scale economies using econometric
functions that estimate a conditional mean response. There are several strong, underlying
assumptions which, if violated, bring into question the validity of statistical conclusions
drawn from estimation of these models. First, evidence suggests that exogeneity assumptions
may not hold for input quantities (Creel and Farell, 2001) or for output quantities (Nebesky
et al., 1995) in these models, suggesting the potential for bias in estimation. Second, the
duality theorom used to relate cost and production functions relies on an assumption of
cost minimization under perfect competition, which is a weak, if not unrealistic, assumption
in many markets (Bairam and Kahya, 1998). Most importantly, econometric production
and cost functions, such as the commonly-employed transcendental logarithm, implicitly
assume that all firms are technically efficient (Coelli et al., 2005). Scheraga (2011) shows
that not all carriers in the trucking industry are efficient; in fact, more than 90 percent
of the sampled carriers exhibited some degree of inefficiency. Thus, estimates from prior
production and cost functions may not have presented sound evidence regarding the presence
returns to scale exist in trucking. Rather, these models have provided an estimate of the
average economies or diseconomies realized when a firm at one level of output adjusts to
some new level of output. Accordingly, the presence of scale economies could readily be
masked within a sample of carriers who are, on average, inefficient and unable to effectively
leverage their scale.
If this is true then the issue of carrier inefficiency is arguably more relevant to the
trucking industry (across the population of carriers) than the presence or absence of scale
economies which might only be accessible to some small proportion of efficient firms. Indeed,
in their analysis of the TL segment, Jin et al. (2017, p. 23) contend that “operating at an
exceptional level of efficiency has allowed modern TL carriers to survive decades of intense

41

competition post-industry deregulation.” Nebesky et al. (1995) make a similar argument for
carriers in the LTL segment, asserting that superior efficiency of surviving firms may have
driven much of the large-scale consolidation seen in the segment after regulatory reform. If
carriers are to succeed in the hyper-competitive trucking market then they must identify
and address sources of inefficiency in their operations and exploit potential opportunities for
increased efficiency.
Unfortunately, only a handful of studies have examined operational (in)efficiency in
trucking. McMullen and Okuyama (2000) examine changes in technical efficiency of motor
carriers before and after the MCA using Malmquist productivity indices. Prior to the
MCA, proponents of regulatory reform argued that deregulation of the trucking industry
would move carriers closer to production frontiers, since institutional rules enacted by the
Interstate Commerce Commission to govern the trucking industry imposed certain technical
inefficiencies and, in some cases, impaired carrier adoption of new technologies (Belzer,
2000; McMullen and Okuyama, 2000). McMullen and Okuyama (2000) indeed find evidence
for an overall increase in technical efficiency after regulatory reform, but also identified
a technological regress which they interpret as an increased focus by surviving firms on
delivering quality service. Like Jin et al. (2017), McMullen and Okuyama (2000, p. 343)
postulate that surviving carriers were those who were “best able to increase technical
efficiency and productivity in response to regulatory reform,” forcing out less efficient
carriers who were unable to adapt to the post-MCA environment.
Examining the motor carrier industry, Scheraga (2011) extends the concept of production
competence (Cleveland et al., 1989) to carrier operations and utilizes Malmquist productivity
indices to examine changes in carrier efficiency over time based on the seven dimensions of
strategic orientation for motor carriers (see Feitler et al., 1997): cost, efficiency, price, risk,
service, size and niche. The last of these dimensions, niche or product niche, refers to the
degree to which a motor carrier focuses production within the LTL segment (Scheraga et al.,
1994; Feitler et al., 1997). For clarity, we adopt an alternative term for this focus dimension,
LTL Concentration, as proposed by Smith et al. (1990). In Scheraga’s (2011) analysis only

42

the LTL concentration dimension was found to influence the Malmquist index scores; motor
carriers with higher LTL concentration (greater production in the LTL segment) showed
improvements in operating efficiency. In contrast, the other six dimensions appeared to
offer little explanatory power in Scheraga’s (2011) model or, in the case of the carrier size
dimension, served primarily to establish boundary conditions around the effects of LTL
concentration. Accordingly, LTL concentration of carriers appears to be uniquely influential
among the seven dimensions of carrier strategic orientation, holding a critical association
with efficiency in production.
Our study extends the results of Scheraga’s (2011) analysis, along with findings by Fawcett
et al. (2016) who show, using data solely from the TL segment, that asset leanness is positively
associated with performance. We rely on theories of firm growth through diversification (Penrose, 1959) and organizational interdependence (Thompson, 1967) to argue that effective
utilization of carrier assets in multiple markets underlies the positive relationship between
LTL concentration and operational efficiency that was found by Scheraga (2011). In the
following section, we present our reasoning for this mechanism along with research hypotheses
and a conceptual model of our hypothesized relationship between LTL concentration and a
carrier’s operational (in)efficiency.

2.3

Theory and Hypothesis Development
Firms are collections of productive resources of which managers are incentivized to utilize

fully through expansion within existing markets and diversification into new markets (Penrose,
1959). Firm growth and a firm’s “productive opportunity” (Penrose, 1959, p. 31) are functions
of an existing resource base and are limited by managerial ability; external markets for
a firm’s products and markets for the factors used in production; and uncertainty and
risk (Kor and Mahoney, 2000). Desirable firm outcomes (e.g., profitability, efficiency) are
achieved through effectively managing, bundling and deploying firm resources to render
valuable services, such as those used as inputs in production (Sirmon et al., 2007; Kor
et al., 2016). However, only in rare and transitory cases might a firm’s resources and the
services they produce be fully utilized (Penrose, 1959)—underutilization is a product of

43

the indivisibility of resources as well as continuous organizational learning (Teece, 1982).
Thus, firm managers remain perpetually motivated to more effectively employ unused and
underutilized productive services of resources toward improving firm outcomes (Pettus, 2001;
Kor et al., 2016). A firm’s diversification of production activities is one approach that can
be used toward this end.
In a production context, diversification refers to the addition of new productive activities
which differ sufficiently from existing activities (Penrose, 1959). Diversification need not
occur within a new area of specialization; firms may diversify by increasing the variety of
items produced or may diversify through vertical integration (Penrose, 1959). Resource-based
theorists widely recognize the advantages of diversifying into related areas, since firms can
realize economies of scope from sharing of resources across operations and from combining
related production activities (Teece, 1982; Palich et al., 2000; Wan et al., 2011). Diversification
may not always be advantageous for firms, however, such as when existing productive
resources are transferred to highly dissimilar markets (Wernerfelt and Montgomery, 1988),
when a firm enters a market where they hold a technological inferior position (Penrose,
1959) or when a firm’s business portfolio becomes so diverse that opportunities for resource
sharing become limited or managerial capability is strained (Palich et al., 2000).
Organizational theory further suggests that a firm’s efficient utilization of existing productive resources within a new activity depends on complementarity of the new activity’s
work flow to that of existing activities. Thompson (1967) considers forms of organizational
work flow by intensity of linkages between constituent units and identifies three generic
patterns of organizational interdependence: pooled interdependence (low intensity or absence
of work flow between units), sequential interdependence (moderate intensity of work flow)
and reciprocal interdependence (high intensity and dependence of work flow). As intensity
and/or complexity of work flow increases, so do requirements for coordination and communication among constituent units. Thus, effective resource-sharing and coordination of
activities become increasingly difficult as an organization transitions into a higher degree of
interdependence (Thompson, 1967), suggesting that diversification’s benefits (i.e, improved

44

employment of unused or underutilized productive resources) might be best realized when
new firm activities do not drastically increase the level of intensity in organizational work
flows.
A motor carrier’s expansion and diversification into additional service areas may benefit
the firm by improving market presence, stabilizing output, improving customer service levels
and reducing the risks associated with single-product production (Stock, 1988). Indeed, after
regulatory reform the trucking industry saw a significant increase in the number of carriers
offering expanded services (Hanna and Maltz, 1998; Feitler et al., 1997; Scheraga, 2011). In
many cases, motor carriers have diversified production activities by offering transportation
service in both the TL and LTL segments (Smith et al., 1990; Ying and Keeler, 1991;
McMullen and Tanaka, 1995). However, the transportation literature has noted significant
differences between these segments in the level of coordination required for production.
Freight flows in an LTL network are highly interdependent (Silverman et al., 1997); the
effects of a single failure (e.g., breakdown of a truck) can ripple through a LTL carrier’s
entire network (Giordano, 2008; Han et al., 2008). Thus, the efficient completion of any
single shipment is predicated upon smooth operations in the carrier’s hub and spoke network,
termed here as intra-haul interdependency. In contrast, efficient completion of truckload
shipments necessitates that TL carriers schedule drivers and equipment against individual
shipments in such a way as to meet shipper requirements while minimizing deadheading
(empty hauls). We refer to this lower level of interdependency seen in the TL segment as
inter-haul sequential interdependency.
Figure 2.1 depicts our hypothesized relationship between a carrier’s segment focus (LTL
concentration) and operational inefficiency. Consistent with the arguments of Thompson
(1967), we postulate that a carrier’s movement in segment focus toward (i.e., diversification
or expansion into) a higher level of interdependence should make the coordination of
resources and production activities increasingly difficult (Thompson, 1967). Following the
resource-based perspective of diversification as a corporate mechanism to improve resource
utilization, resource coordination challenges across a business portfolio should translate

45

Operational inefficiency
Pure TL

LTL Concentration

Pure LTL

Figure 2.1: Conceptual model.
into lower overall levels of operating efficiency. This relationship can be seen in Figure
2.1 where, for pure TL carriers, the marginal effect of LTL concentration on inefficiency is
positive (Hypothesis 1). Conversely, we hypothesize that pure LTL carriers who diversify
into the TL segment—moving to a lower overall level of interdependence—will be able to
better employ underutilized productive resources, resulting in a lower degree of operational
inefficiency (Hypothesis 2). Accordingly, we theorize the overall relationship between LTL
concentration and operational inefficiency as a polynomial with two inflection points.

2.4
2.4.1

Methodology
Data and Sample

We test our research hypotheses using a sample obtained from governmental archives of
motor carrier production, operating and financial statistics over the five-year period of 1999
through 2003. The use of archival data from this period is appropriate to our test of theory.
Most importantly, we seek to examine the motor carrier industry during a period of relative
stability, void of much of the transient—but still lingering—volatility that has occurred in
factor markets since 2003. For instance, escalation of the well-publicized driver shortage began

46

in 2005 and, with some ebbs and flows in the trucking labor market over the past decade, has
artificially constrained motor carriers’ ability to grow and expand (Costello and Suarez, 2015)
while also impacting carrier productivity (Min and Lambert, 2002; Suzuki, 2007; Cantor et al.,
2011). Additionally, transient shifts and volatility in the price of fuel have, to some extent,
influenced the operational decisions of motor carriers (Winebrake et al., 2015) and have led
to speculation, hedging or even bankruptcy filings in the industry (Smith, 2016). Further,
carriers continue to adjust to governmental regulations imposed over the past two decades,
such as changes to hour-of-service requirements, and these regulations continue to influence
carrier productivity (Costello and Suarez, 2015). Nonetheless, both the fundamental nature
of the TL and LTL segments (e.g., levels of market competition, productive resources in use,
contrast in coordination requirements and complexity in operations) and our underlying
theory remain durable over time. Our use of historical data allows us to research corporate
diversity by motor carriers in a more ‘controlled’ setting—one where threats to the validity
of our findings from transient environmental factors are significantly reduced.
Data used in this study were obtained from archives published by the U.S. Department
of Transportation, Office of Motor Carrier Information (OMCI). These data contain Form
M filings required by carriers under 49 U.S.C. § 14123. Consistent with previous studies
into the industry (e.g. Xu et al., 1994; McMullen and Okuyama, 2000; Giordano, 2008), we
limit our sample to large (Class I) and medium (Class II) carriers of general freight. Thus,
annual revenues for all carriers in our sample exceed three million dollars after adjustment
to 1994 dollars using the procedure within 49 C.F.R. § 369. To ensure that our analysis
includes only established carriers in the industry we limit our sample to those motor carriers
who filed a Form M during each of the five years. Motor carriers are uniquely identified in
our dataset by their motor carrier number.
A central premise of our theorizing is that carrier management is incentivized to improve
the utilization of productive resources under their control (Penrose, 1959). Managers have
limited control over owner-operators (Corsi and Grimm, 1987)—drivers that move freight
under contract with carriers using vehicles that the drivers own and maintain (Cantor

47

et al., 2013). The use of owner-operators has been ubiquitous in the U.S. trucking industry
since contractual relationships with these drivers can offer carriers operational flexibility
(i.e., in level of output) while reducing carrier investments in employee labor and capital
equipment (Corsi and Grimm, 1987). Han et al. (2008) found that, around the time of our
period of analysis, nearly 70 percent of all general freight carriers used owner-operators to
some extent. For these carriers, expenses attributed to hiring owner-operators accounted for
an average of 30 percent of annual operating expenses (Han et al., 2008).
Previous studies have linked the use of owner-operators to key carrier outcomes. For
instance, Miller and Saldanha (2016) find that motor carrier safety declines as carriers rely
more heavily on owner-operators. While empirical evidence is lacking on the relationship
between owner-operator use and carrier operational efficiency, Scheraga (2011) notes that a
carrier’s use of owner-operators may bias input measures in studies of carrier productivity.
We adopt two approaches to minimize our potential for exposure to this bias. First, we
restrict our sample to carriers whose owner-operator expenses accounted for no more than
15 percent of total operating expenses, thus removing carriers who rely more heavily on
owner-operators from our sample. A balance between sample size and internal validity
were primary criteria in establishing this threshold, which is half that of the industry
average. For instance, lowering the threshold to zero percent of total operating expenses
(no owner-operator use) would result in a sample consisting of only one carrier. As an
additional safeguard, we employ a statistical control for owner-operator use in our analysis;
our operationalization of this variable is later discussed.
We performed an audit of the resulting sample to identify and filter anomalous observations, as is typical when working with transportation archives (e.g., Han et al., 2008; Fawcett
et al., 2016). For 10 of the 130 carrier in our dataset, Form M filings contained reporting
errors (e.g., implausible or non-positive values) and, out of caution and a desire to maintain
balanced data, we filtered these carriers from our sample. Thus, our final sample consisted
of five years of annual observations on 120 carriers, for a total sample size of 600.

48

Table 2.1: Descriptive statistics.
Variable

Mean

S.D.

ICC†

(1)
(2)
(3)
(4)
(5)
(6)
(7)

.56
.74
.54
.04
.29
.05
1.85

.50
2.23
1.48
.11
.41
.04
1.09

1.00
.98
1.00
.99
.94
.81
.74

Carrier class
Drivers (×103 )
Power units (×103 )
Total miles (×109 )
LTL concentration
Outsourcing
Operational
inefficiency

Notes: ∗ p < .05 (two-tailed); † ICC calculated as

2.4.2

(1)

Pearson Product-Moment Correlation
(2)
(3)
(4)
(5)
(6)

−.34∗
−.37∗
−.38∗
−.17∗
−.25∗
.19∗

σb2/ σ 2 + σ 2
b
w

.97∗
.95∗
.27∗
.19∗
−.16∗

.97∗
.29∗
.20∗
−.18∗

.23∗
.22∗
−.21∗

.09∗
.16∗

−.14∗




Variable Operationalization

This subsection provides an overview of the operationalization of variables used in the study.
Descriptive statistics for all variables are provided in Table 2.1.

2.4.2.1

Dependent Variable

We use data envelopment analysis (DEA) to estimate technical efficiency scores for each
carrier at each of the five measurement occasions. DEA is an appropriate analytic strategy
to assess relative efficiency of productive entities (i.e., decision-making units) over time
in longitudinal settings (Bogetoft and Otto, 2010). Since its introduction to the logistics
literature (see Clarke and Gourdin, 1991; Kleinsorge et al., 1991), DEA has been widely
employed to assess efficiency of productive entities in the transportation sector (e.g., Mejza
and Corsi, 1999; Scheraga, 2004; Weber and Weber, 2004; Ha et al., 2010; Rogers and
Weber, 2011; Wanke, 2014; Min and Ahn, 2017). We select DEA because it is suited to
multiple-input, multiple-output models and can estimate efficiency scores without requiring
distributional assumptions. Further, commonly used econometric production or cost models
(e.g., those based on least-squares estimation) assume that all productive entities are
technically efficient (Coelli et al., 2005). In contrast, this paper seeks to directly examine
a determinant of technical inefficiency, necessitating the use of a benchmarking technique

49

148

150
130

Frequency

125
101

100
75

54
50
37
30 27 27
25
12
4
0

1

2

9

3

11
3

2
4

3

2
5

>5

Figure 2.2: Distribution of estimated output-oriented DEA efficiency
(operational inefficiency) scores.
to derive efficiency scores. Malmquist productivity indices are not created, as has often
been done in prior carrier efficiency research (e.g., McMullen and Okuyama, 2000; Scheraga,
2011), since we do not seek to examine changes in productivity between time periods. A
histogram of estimated efficiency scores is provided in Figure 2.2; we discuss our specification
of the DEA model which produced these scores next.
We specify an output-oriented DEA model since the focus of our theorizing is on
improvements to the utilization of productive resources and, thus, output efficiency or
enhancement (Cook et al., 2014). Our use of an output-oriented model is consistent with
previous work on motor carrier efficiency (e.g., McMullen and Okuyama, 2000; Scheraga,
2011). We view carriers as having a common production possibility set T t = {(xt , y t ) ∈
n
t
t
Rm
+ × R+ | x can produce y }, where m is a vector of inputs consisting of the number

of power units (x1 ) and drivers (x2 ) for carrier i during period t and n is a vector of
outputs consisting of the miles produced4 of TL service (y1 ) and LTL service (y2 ) for the
4 The

use of annual miles is consistent with Feitler et al. (1998), who view efficiency
for less-than-truckload firms as a (ratio) relationship between annual miles and number of
power units. However, we also consider a labor input, recognizing that classical economic
theory of productivity (e.g., the Cobb-Douglas production function) and the literature on
transportation economics in the motor carrier industry considers both labor and capital
assets as inputs to production.

50

same carrier and period. TL and LTL miles are treated separately given differences in the
nature of output between the two segments (McMullen and Tanaka, 1995; McMullen and
Okuyama, 2000); solving for a frontier using aggregates of these outputs would be clearly
inappropriate given our hypotheses. Thus, given the two outputs, a production possibilities
frontier (concave output isoquant) is solved, representing the various combinations of outputs
that a carrier might efficiently produce given some level of input. We allow for variable
returns to scale (Banker et al., 1984), an assumption that is appropriate for most production
contexts (Talluri et al., 2013) and that is consistent with recent literature on carrier efficiency.
For instance, Fawcett et al. (2016, p. 48) state, “it is clear that it would be overly simplistic
to assume the complete absence of economies of scale in the motor carrier industry.” The
distance function for our output-oriented model is defined in Equation 2.1. Output-based
distances for each carrier at each year are evaluated against the efficiency frontier and these
observations are assigned a Farrell output efficiency score θ which is lower-bounded at a value
of one (corresponding to efficient carriers). Thus, we refer to this variable as operational
inefficiency.
maxθ,λ θ,

s.t.

2.4.2.2

(2.1)

n
P



λi xji ≤ xjp ∀j



i=1



n

P


λi yki ≥ θykp ∀k

i=1


n
P



λi = 1



i=1





λi ≥ 0 ∀i

(2.2)

Focal Predictor

We operationalize our focal predictor, LTL concentration, as the ratio of LTL miles produced
by a carrier in a given year to total miles (LTL + TL) produced by that carrier in the
same year. Thus, possible values of LTL concentration range from a zero, corresponding
to occasions in which a carrier produced only TL carriage, to one, which corresponds to
occasions in which a carrier produced only LTL carriage. The observed distribution of LTL

51

450

379

Frequency

350
250
90

74

66

60
38

30

20

17
6

0
0
(TL)

0 to .2

.2 to .4

.4 to .6

.6 to .8

.8 to 1

1
(LTL)

Figure 2.3: Histogram of carrier output concentration within the
less-than-truckload segment (LTL concentration).
concentration is shown in Figure 2.3. The transportation literature has tended to bifurcate
carriers between the two segments (Scheraga, 2011); 31 of the 120 carriers in our sample
(25.8%) concurrently produced both TL carriage and LTL carriage, thus participating in both
segments. Further, carrier LTL concentration was fairly stable over our period of analysis with
an intraclass correlation (ICC) of .94, indicating that the overwhelming majority of variability
in LTL concentration exists between (and not within) carriers. The sample data does not
suggest that average levels of LTL concentration differ in the population between Class I
carriers (m = .37, sd = .43) and Class II carriers (m = .23, sd = .38), based on estimation
within a linear mixed model (χ2∆ (1) = 3.74, p > .05). Nor is carrier class (I or II) a significant
predictor of a carrier’s participation in both market segments; the odds ratio associated with
carrier class did not differ significantly from a ratio of one when estimated within a binomial
(logit link) generalized estimating equation (OR = .72, SE(robust) = 1.21, p > .05).

2.4.2.3

Covariates

Carrier Class.

We control for carrier class to hold constant any residual, conditional

differences in carrier inefficiency due to size (i.e., after variable returns to scale is accounted
for in the DEA model). Carrier class is assigned using the procedures at 49 C.F.R. § 369.
Class I carriers are defined as those carriers having adjusted annual carrier operating revenues

52

(after revenue deflation to 1994 dollars) of more than 10 million dollars. Class II carriers are
defined as those carriers having adjusted annual carrier operating revenues of more than 3
million dollars but less than 10 million dollars. Our sample of 120 carriers is comprised of
53 Class I carriers and 67 Class II carriers; there were no instances of class promotion or
demotion during our sampling period. We code carrier class using weighted effects coding
such that groups are compared with the aggregate population mean (Cohen et al., 2003);
Class I carriers are used as the base group.
Year. We include year fixed effects to control for unobserved, carrier-invariant sources of
variability in production inefficiency that occur over time (e.g., macroeconomic conditions,
regulatory changes). Year dummy variables were constructed using unweighted effects coding
such that measurement occasions (years) are compared against the overall unweighted mean.
1999 is used as the base year.
Outsourcing. We operationalize outsourcing as the proportion of a carrier’s total operating
expenses at a measurement occasion arising from equipment rentals (with and without
drivers) and carrier-procured transportation services. Thus, this variable accounts for carrier
expenses incurred under contract with owner-operators and other carriers for the movement
of shipments, as well as contracts with third-party firms for the use of revenue vehicles. In
each of these instances, the carrier has limited control over drivers and/or capital equipment
used in production. In their analysis of carrier utilization of owner-operators, Han et al.
(2008) show that carriers who participate in the LTL market segment are less likely to
outsource freight movements. Thus, if outsourcing is also related to operational inefficiency,
its omission from our regression equation may bias coefficient estimates for LTL concentration.
We include outsourcing as a covariate on these grounds and mean-center the variable prior
to entry into the regression equation for interpretability of the intercept.

2.4.3

Statistical Analysis

Statistical analysis is performed using the mixed effect modeling framework. A mixed
effects model is selected since it offers flexibility in the specification of regression models

53

for longitudinal data, accounting for both population characteristics and subject-specific
effects (Fitzmaurice et al., 2012). Our model is defined in Equation 2.3. In this model, outputoriented DEA efficiency scores (operational inefficiency; Yit ) for carrier i at a measurement
occasion t are a function of the carrier’s class, the year, the extent of outsourcing and a
cubic polynomial of LTL concentration. We allow for a randomly varying subject effect
and specify three models: a model which includes only covariates while excluding the LTL
concentration polynomial (Control Model); a model which includes all variables and assumes
no within-group residual correlations (Model A); and a model with includes all variables and
allows for an AR(1) autoregressive process in the within-group residual correlation structure
(Model B). Estimation is performed by maximum likelihood.
Level 1
Yit = β0i + β1 × CLASSi +

5
X

βj × YEARt + β6 × OUTSOURCINGit

(2.3)

j=2

+ β7 × LTL_CONCENTRATIONit + β8 × (LTL_CONCENTRATIONit )2
+ β9 × (LTL_CONCENTRATIONit )3 + Rit
Level 2
β0i = γ00 + U0i

2.5

(2.4)

Results
The results of estimation for the three models are presented in Table 2.2. Model B, the

AR(1) model, offered superior fit to the data over the control model (χ2∆ (4) = 41.11, p < .01)
and over Model A, the variance-components model (χ2∆ (1) = 22.96, p < .01). The effects
from the cubic polynomial were jointly significant (χ2∆ (3) = 16.89, p < .01), and predictions
of operational inefficiency in Model B explained 81.34 percent of the total variability in
DEA-estimated operational inefficiency scores and 15 percent of Level 1 (within carrier)
variability in DEA-estimated operational inefficiency scores when approximated using Bryk
and Raudenbush’s (2002) proposed technique.

54

Table 2.2: Results of mixed effect model estimation.

Response=Operational inefficiency
Fixed effects
Intercept
CLASS:II
YEAR:2000
YEAR:2001
YEAR:2002
YEAR:2003
OUTSOURCING
LTL_CONCENTRATION
LTL_CONCENTRATION2
LTL_CONCENTRATION3

γ10
β1
β2
β3
β4
β5
β6
β7
β8
β9

Variance Components
Standard Deviation (b0i )
Standard Deviation (eit )
Autoregressive (AR1)

σb
σe
ϕ

Measures of fit
−2 Log likelihood
AIC
AICc
pR2 (Level 1)

Control Model
(R = VC)
Est.
S.E.

Model A
(R = VC)
Est.
S.E.

Model B
(R = AR(1))
Est.
S.E.

1.85∗∗
.19∗
−.01
.01
−.06
−.07
.54

.09
.08
.05
.05
.05
.05
1.19

1.98∗∗
.15†
<.01
<.01
−.06
−.06
.92
4.03†
−15.91∗∗
11.93∗∗

.11
.10
.04
.04
.04
.04
1.17
2.12
5.32
3.38

1.98∗∗
.15
<.01
<.01
−.06
−.06
.48
3.88†
−15.46∗∗
11.62∗∗

.12
.09
.04
.04
.04
.05
1.25
2.33
5.77
3.62

.92∗∗
.55∗∗

.07
.02

1.04∗∗
.52∗∗

.10
.02

1.05∗∗
.60∗∗
.41∗∗

.11
.04
.08

1, 307.00
1, 325.00
1, 325.31
.02

1, 288.85
1, 312.85
1, 313.38
.11

1, 265.89
1, 291.89
1, 292.52
.15

2
2
2
Notes: † p < .10, ∗ p < .05, ∗∗ p < .01 (two-tailed); R2 for Level 1 approximated as (σnull
− σfull
)/σnull
(Bryk
and Raudenbush, 2002), where the covariance structure of the R matrix in the null model is of the same type
(e.g., VC, AR(1)) as the model being tested.

The model-estimated cubic relationship between LTL concentration and operational
inefficiency is shown in Figure 2.4. Polynomial inflections estimated by the model are
comparable to those theorized within Figure 2.1. Hypothesis 1 suggested that, for pure TL
carriers, LTL concentration will have a positive marginal effect on operational inefficiency.
Indeed, a positive marginal effect was estimated at this level of LTL concentration, but was
significant under a two-tailed hypothesis test only at an α of .10 (β = 3.88, t = 1.66, p < .10).
Hypothesis 2 suggested that, for pure LTL carriers, LTL concentration will also have a
positive marginal effect on operational inefficiency (thus, a decrease LTL concentration will
be associated with greater operational efficiency). Consistent with this hypothesis, a positive
marginal effect of LTL concentration was estimated for pure LTL carriers (β = 7.81, t =

55

3.00
2.75

Operational inefficiency

2.50
B
2.25
2.00

D

A

1.75
1.50
1.25

C

1.00

Perfectly efficient

0.75
0.50
0

.1

.2

.3

.4

.5

(TL)

.6

.7

.8

.9

1
(LTL)

Figure 2.4: Model-estimated relationship between LTL concentration and
production inefficiency, with 95% confidence bands.
4.11, p < .01).
Prior to estimating regions of significance for the estimated polynomial, we test for mean
differences between the four points identified in Figure 2.4 (pure TL, local maximum, local
minimum, pure LTL) within a Monte Carlo simulation. 10, 000 draws were taken from
a multivariate normal distribution using the model-produced mean vector and variancecovariance matrix of parameter estimates; Bonferroni corrections are made to maintain
a family-wise error rate of .05. From Model B, a local maximum in the polynomial was
found at an LTL concentration of .15 (85% truckload), where operational inefficiency was
estimated to be the highest across all levels of LTL concentration at a value of 2.26. A local
minimum was found at an LTL concentration of .74, where operational inefficiency was
estimated to be the lowest across all levels of LTL concentration at a value of 1.09. The
results of comparisons between model-derived estimates at these two points and those at
levels of LTL concentration corresponding to Pure TL and Pure LTL are listed in Table 2.3.
Notably estimates of operational inefficiency do not differ significantly between pure TL
carriers (m = 1.98) and pure LTL carriers (m = 2.02).

56

Table 2.3: Monte Carlo simulation results.
Point
(A)
(B)
(C)
(D)

LTL Concentration

Est.

S.E.

Pure Truckload
85% Truckload†
74% Less-than-truckload‡
Pure Less-than-truckload

1.98
2.26
1.09
2.02

.12
.24
.21
.18

Absolute Mean Difference
(A)
(B)
(C)
.28
.89∗
.04

1.17∗
.24

.93∗

Notes: ∗ mean difference differs from zero at α = .05/6 (two-tailed); † local maximum θ̂∗ for ≥50% TL; † local
minimum θ̂∗ for ≥50% LTL.

2.5.1

Regions of Significance

We estimate regions of significance of the simple slope (rate of change) of the cubic polynomial
LTL concentration on operational inefficiency using the Johnson-Neyman Technique (see
Miller et al., 2013). We provide a derivation for regions of significance of a cubic polynomial
here. Given the linear combination of regression coefficients, an estimate of the variance of
a simple slope is given by:
s2b = w0 Sb w

(2.5)

The weight column vector w is the first partial derivative of the design matrix with
respect to the explanatory variable of interest (X), here

w0



= 1 2X



3X 2

. Sb is the

estimated 3 × 3 variance-covariance matrix of regression coefficients. In this case of a
curvilinear relationship between X and Y , where raw polynomials up the third degree (cubic)
are included in the regression equation and where the effect of X on Y does not depend on
the value of some other variable(s), the standard error for the simple slope of X on Ŷ is:
sb =

q

s11 + 4Xs12 + 6X 2 s13 + 4X 2 s22 + 12X 3 s23 + 9X 4 s33

(2.6)

The confidence bands of the simple slope are then solved for zero, representing those
values of the explanatory variable where the slope crosses the threshold of significance (Miller
et al., 2013). This can be simplified to solving the absolute value of the t-value for the
critical value from the t distribution:
β1 + 2β2 X + 3β3 X 2
tcrit = ± r
s11 + 4Xs12 + 6X 2 s13 + 4X 2 s22 + 12X 3 s23 + 9X 4 s33

57

(2.7)

12.50

Negative rate of change
(decreasing inefficiency)

n.s.

n.s.

10.00

Positive rate of
change (increasing
inefficiency)

7.50
5.00
2.50
0.00
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

−2.50
−5.00

Figure 2.5: Johnson-Neyman plot of the regions of significance for the
simple slope of LTL concentration on operational inefficiency.
The resulting Johnson-Neyman plot is shown in Figure 2.5. Two regions of significance
were found. The marginal effect of LTL concentration on operational inefficiency is significantly negative between LTL concentration values of .22 and .65. This effect turns
significantly positive between LTL concentration values of .79 and 1.00. The marginal effect
of LTL concentration on operational inefficiency is nonsignificant at all other values of LTL
concentration.

2.5.2

Alternative Specifications

Two alternative specifications to the AR(1) cubic model (Model B) are considered. First,
we reduce the degree of the polynomial for LTL concentration to a quadratic (Model C),
thus testing for significance of a second inflection.5 Next we impose a harmonic form on
the non-linear relationship between LTL concentration and operational inefficiency using
5 An

AR(1) model including controls and only a linear effect of LTL concentration on
operational inefficiency (i.e., polynomial degree
= 1) does not differ significantly from an


2
AR(1) control model χ∆ (1) = 0.60, p = .44 .

58

sinusoidal regression (Model D), in order to estimate a more parsimonious model with
symmetry in the curvatures. This is accomplished first using a single sine wave (Equations
2.8 and 2.9), matching the harmonic in Figure 2.1 with respect to phase (ϕ, implicitly zero)
and frequency (f , implicitly one), but leaving amplitude (β10 ) free for estimation. Estimates
of amplitude in harmonic regression are known to be asymptotically efficient (e.g., under
ordinary least squares estimator; see Durbin, 1960); we continue to allow AR(1) covariances
in the R-matrix, recognizing that these within-carrier residuals are correlated and may
otherwise bias standard errors.
Level 1
Yit = β0i + β1 × CLASSi +

5
X

βj × YEARt + β6 × OUTSOURCINGit

(2.8)

j=2

+ β10 × sin (2π × LTL_CONCENTRATIONit ) + Rit
Level 2
β0i = γ00 + U0i

(2.9)

The results of estimation for the two alternative specifications are presented in Table
2.4. Estimation of Model C resulted in a convex quadratic relationship between LTL
concentration and operational inefficiency; coefficents of the polynomial are jointly significant
(χ2∆ (2) = 6.82, p = .03). However, consistent with our theoretical reasoning for a cubic
relationship, the quadratic model offers inferior fit to the sample (χ2∆ (1) = 10.07, p < .01).
Similarly, the less flexible (but more parsimonious) sinusoidal model (Model D) improved fit
over an AR(1) control model (χ2∆ (1) = 6.50, p = .01), with an estimated sine wave amplitude
of 0.42 (t(475) = 2.64, p < .01). Comparisons between non-nested models B and D are made
using Akaike information criterion (AIC), as recommended by Singer and Willett (2003)
and Fitzmaurice et al. (2012). Both AIC and second-order AIC (AICc) give preference to
Model B, the cubic model.

59

Table 2.4: Results of estimation for alternative models.

Response=Operational inefficiency

Model B
(R = AR(1))
Est.
S.E.

Fixed effects
Intercept
CLASS:II
YEAR:2000
YEAR:2001
YEAR:2002
YEAR:2003
OUTSOURCING
LTL_CONCENTRATION
LTL_CONCENTRATION2
LTL_CONCENTRATION3
sin (2π × LTL_CONCENT)

γ10
1.98∗∗
β1
.15
β2
<.01
β3
<.01
β4
−.06
β5
−.06
β6
.48
β7
3.88†
β8 −15.46∗∗
β9
11.62∗∗
β10

.12
.09
.04
.04
.04
.05
1.25
2.33
5.77
3.62

Variance Components
Standard Deviation (b0i )
Standard Deviation (eit )
Autoregressive (AR1)

σb
σe
ϕ

1.05∗∗
.60∗∗
.41∗∗

.11
.04
.08

Measures of fit
−2 Log likelihood
AIC
AICc
pR2 (Level 1)
Notes: † p < .10, ∗ p < .05,
and Raudenbush, 2002).

2.5.3
2.5.3.1

1, 265.89
1, 291.89
1, 292.52
.15
∗∗

Model C
(R = AR(1))
Est.
S.E.
1.99∗∗
.16†
−.01
<.01
−.05
−.06
.31
−2.77∗∗
2.63∗∗

.12
.09
.04
.04
.04
.05
1.27
.95
.96

1.03∗∗
.62∗∗
.45∗∗
1, 275.97
1, 299.97
1, 300.50
.09

.09
.05
.08

Model D
(R = AR(1))
Est.
S.E.
1.90∗∗
.17†
−.01
<.01
−.06
−.06
.25

.10
.09
.04
.04
.04
.05
1.26

.42∗∗

.16

.97∗∗
.62∗∗
.42∗∗

.08
.04
.08

1, 276.29
1, 298.29
1, 298.74
.10

2
2
2
− σfull
)/σnull
(Bryk
p < .01 (two-tailed); R2 for Level 1 approximated as (σnull

Robustness Tests
Subsamples

We first test for robustness of the regression results using subsamples by carrier class. The
full sample of carrier data (n = 600) was divided into two subsamples consisting of Class I
motor carriers (n = 265) and Class II motor carriers (n = 335). Model B (the cubic model)
was estimated separately for each subsample; estimation results are presented in Table 2.5.
The analyses suggest a structural difference between carrier classes; a χ2 (11) difference of
116.63 between the aggregated model (Model B) and disagreggated (subsample) models is
statistically significant at α = .05 (p < .01). Parameter estimates of the cubic function in the

60

subsample models were jointly significant for Class I carriers (χ2∆ (3) = 19.25, p < .01) and
Class II carriers (χ2∆ (3) = 17.22, p < .01). For pure truckload carriers, the instantaneous rate
of change in operational inefficiency for a unit increase in LTL concentration is significantly
positive at α = .05 for both Class I carriers (m = 3.54, t = 1.98, p = .05) and Class II carriers
(m = 8.61, t = 2.14, p = .03). Thus, while Hypothesis 1 was only supported at α = .10
when carriers classes were aggregated within Model B, disaggregation to subsamples offers
stronger statistical evidence. Lastly, as with the aggregated model (Model B), Hypothesis
2 was supported at α = .05 for Class II carriers, where the instantaneous rate of change
in operational inefficiency for a unit decrease in LTL concentration is significantly positive
(m = 13.20, t = 4.27, p < .01); however, this effect is weaker and nonsignificant for Class I
carriers (m = 1.99, t = 1.25, p = .21).

2.5.3.2

Censored Regression

Previous motor carrier efficiency studies using two-stage DEA analysis (e.g., McMullen
and Okuyama, 2000; Scheraga, 2011) have argued that DEA efficiency scores are censored
and, as such, have employed Tobit regression to estimate the second-stage regression model.
The appropriateness of this statistical strategy (and other statistical strategies when DEA
efficiencies are regressed) is the subject of intense debate within the literature (Liu et al.,
2016). For instance, McDonald (2009) contends that the use of Tobit models is inappropriate
since DEA efficiency scores are not generated by a censoring process and, instead, represent
fractional data. Nonetheless, recognizing that our data contains a lower bound, we estimate
a random-effects Tobit regression to assess the extent to which our results might be impacted
by any censoring in the data generating process. We estimate two models—one with time
fixed effects (Tobit A) and one without time fixed effects (Tobit B)—since the inclusion
of fixed effects within non-linear panel data models using finite samples may produce
biased and inconsistent estimates under the maximum likelihood estimator (Greene, 2004).
The results of estimation are presented in Table 2.6; within each model, coefficients for
the cubic polynomial are comparable to those previously estimated within a linear mixed

61

Operational inefficiency

3.00
2.50
2.00
1.50
1.00
0.50
0

.1

.2

.3

.4

.5

.6

.7

.8

.9

1
(LTL)

(TL)

(a) Full sample

Operational inefficiency

3.00
2.50
2.00
1.50
1.00
0.50
0

.1

.2

.3

.4

.5

.6

.7

.8

.9

(TL)

1
(LTL)

(b) Class I Motor Carriers

Operational inefficiency

4.00
3.00
2.00
1.00
0.00
0

.1

.2

.3

.4

.5

.6

(TL)

.7

.8

.9

1
(LTL)

(c) Class II Motor Carriers

Figure 2.6: Plots of the effects of LTL concentration on operational inefficiency for
the full sample (a) and subsamples of Class I (b) and Class II (c) motor carriers.

62

Table 2.5: Results of estimation for subsamples.

Response=Operational inefficiency

Full Sample
(R = AR(1))
Est.
S.E.

Fixed effects
Intercept
CLASS:II
YEAR:2000
YEAR:2001
YEAR:2002
YEAR:2003
OUTSOURCING
LTL_CONCENTRATION
LTL_CONCENTRATION2
LTL_CONCENTRATION3

γ10
1.98∗∗
β1
.15
β2
<.01
β3
<.01
β4
−.06
β5
−.06
β6
.48
β7
3.88†
β8 −15.46∗∗
β9
11.62∗∗

.12
.09
.04
.04
.04
.05
1.25
2.33
5.77
3.62

Variance Components
Standard Deviation (b0i )
Standard Deviation (eit )
Autoregressive (AR1)

σb
σe
ϕ

1.05∗∗
.60∗∗
.41∗∗

.11
.04
.08

Measures of fit
−2 Log likelihood
AIC
AICc
pR2 (Level 1)
Notes: † p < .10, ∗ p < .05,
and Raudenbush, 2002).

1, 265.89
1, 291.89
1, 292.52
.15
∗∗

Class I
(R = AR(1))
Est.
S.E.

Class II
(R = AR(1))
Est.
S.E.

1.79∗∗

.12

2.02∗∗

.19

.05
.05
.05
.04
1.10
1.79
4.33
2.75

<.01
−.01
−.08
−.07
.91
8.61∗
−28.61∗∗
20.70∗∗

.06
.06
.06
.09
1.99
4.02
10.07
6.30

.09
.02
.10

1.22∗∗
.80∗∗
.60∗∗

.17
.11
.08

−.02
.01
−.02
−.05
−.11
3.54∗
−10.28∗
6.34∗
.71∗∗
.34∗∗
−.33∗∗
351.52
375.51
376.75
.12

797.75
821.75
822.72
.17

2
2
2
− σfull
)/σnull
(Bryk
p < .01 (two-tailed); R2 for Level 1 approximated as (σnull

model (Model B). Consistent with Hypothesis 1, the marginal effect of LTL concentration
on operational inefficiency for pure TL carriers was positive and significant for Tobit A
(m = 4.53, t = 2.08, p = .03) and for Tobit B (m = 4.46, t = 1.88, p = .02). Similarly,
a positive marginal effect of LTL concentration on operational inefficiency for pure LTL
carriers was positive and significant in Tobit A (m = 7.56, t = 1.17, p < .01) and Tobit B
(m = 8.07, t = 1.14, p < .01), as predicted under Hypothesis 2.

2.6

Discussion
Our study bridges Thompson’s (1967) conceptualization of interdependence in organiza-

tional workflows with the Penrosian theoretical perspective of firm growth and utilization of
productive resources. Using archival data from the motor carrier industry, we find evidence

63

Table 2.6: Results of estimation for random-effects Tobit regression models.
Tobit A
(Year Effects)
Est.
S.E.

Response=Operational inefficiency
Fixed effects
Intercept
CLASS:II
YEAR:2000
YEAR:2001
YEAR:2002
YEAR:2003
OUTSOURCING
LTL_CONCENTRATION
LTL_CONCENTRATION2
LTL_CONCENTRATION3

γ10
β1
β2
β3
β4
β5
β6
β7
β8
β9

Variance Components
Standard Deviation (b0i )
Standard Deviation (eit )

σb
σe

1.96∗∗
.15
<.01
.01
−.06
−.06
.72
4.53∗
−16.58∗∗
12.06∗∗

.18
.11
.10
.09
.13
.11
1.88
2.08
4.30
2.41

1.03∗∗
.53∗∗

.13
.03

−2 Log likelihood
†

∗

Notes: p < .10, p < .05,

1, 141.33
∗∗

Tobit B
(No Year Effects)
Est.
S.E.
1.97∗∗
.14

.54
4.46∗
−16.93∗
12.49∗
1.05∗∗
.54∗∗

.18
.11

1.94
1.88
3.98
2.27
.13
.02

1, 149.56

p < .01 (two-tailed).

that carrier diversification into a market segment with a higher degree of internal interdependence reduces operational efficiency. We also find evidence that carrier diversification
into a market segment with a lower degree of internal interdependence increases operational
efficiency. Relying on the work of Penrose (1959), we postulate that the mechanism underlying these relationships is carrier management’s ability (or inability) to improve the
utilization of unused and underutilized productive resources in the face of changing levels of
interdependence among organization workflows. For motor carriers participating only in
the TL market segment, our results suggest that diversification of production into the LTL
market segment may lead to a decrease in overall operational efficiency. Production in the
LTL market segment requires a substantially greater degree of coordination among carrier
activities than does production in the TL market segment.
In LTL carriage, workflows exhibit a reciprocal pattern of interdependence, which we
refer to as intra-haul interdependency. The results of our subgroup analysis suggest that
this decrease in overall operational efficiency as carriers shift from inter-haul sequential

64

interdependency is stronger for Class II (medium-sized) carriers than it is for Class I (large)
carriers. When data across the two classes were aggregated, we found mixed support for
this effect, which was statistically significant at α of .05 in a censored regression but was
only statistically significant at α of .10 when data was viewed as fractional (not censored)
and within-group residuals were allowed to covary as an AR(1) process.
We contend that this evidence regarding Hypothesis 1, although mixed in level of
significance across statistical techniques, is compelling. Specifically, economic theory and
management theory jointly suggest that firms should seek efficiency in their operations.
Penrose (1959), for instance, argues that firm management will strive to most efficiently utilize
productive resources. Our hypotheses suggested that, for a pure TL carrier, the marginal
effect of LTL output production would be an increase in overall operational inefficiency. Thus,
consistent with observed frequencies of carrier concentration in our dataset (Figure 2.3),
one should expect to find, and does find, few carriers operating in this region of inefficient
production. The lackluster level of statistical support offered by the AR(1) mixed effect
model of aggregated carrier data may simply be a reflection of consistency between carrier
behavior and theory. Given an absence of carrier concentration at this point of inefficiency,
wider confidence intervals were estimated by the statistical model.
Our second Hypothesis suggested that motor carriers participating solely in the LTL
market segment could better employ underutilized productive resources and, thus, improve
overall operating efficiency, by diversifying production into the TL market segment. This
hypothesis was broadly supported in our analyses; while it did not hold for one of our
subsamples, the effect was prominent in the full data set. Our analysis of regions of
significant for the simple slope of the effect of LTL concentration on operational inefficiency
suggests that carriers producing between 79 percent and 100 percent of their output as
LTL carriage might achieve additional gains to overall operational efficiency by increasing
production in the TL market segment. As before, we rely on resource-based theory (Penrose,
1959) to argue that the underlying mechanism is carrier management’s ability to improve
the employment of unused and underutilized productive resources.

65

Our analysis carries important insights into the role of carrier size on operational efficiency.
Mean levels of carrier operational efficiency only vary weakly, if at all, across the two carrier
classes. This suggests that, in the motor carrier population, large carriers appear no more
efficient than medium-sized carriers. However, our subgroup analysis demonstrates that
structural differences do exist between these two classes. Specifically, while average efficiency
levels may not differ between the classes, operational efficiency for large (Class I) carriers
appears to be more robust than that of medium (Class II) carriers to changes in the focus
of production between the two market segments. However, this also suggests that medium
size carriers may be better able to capitalize on opportunities to improve the utilization
of productive resources through diversification, but may also be more susceptible to the
potentially damaging effects of increased interdependence in operations. Thus, our findings
are consistent with those of Scheraga (2011), showing that carrier size attenuates the effects
of LTL concentration (concentration of production on LTL carriage) on operational efficiency,
while also providing a more fine-grained view into the nature of the moderated relationship.

2.6.1

Theoretical Contributions

A primary theoretical contribution of our study is the link between interdependence and a
firm’s ability to drive improvements in resource utilization through diversification. In the
context of the motor carrier industry, we delineate between two types of interdependence:
intra-haul interdependency and, for the TL market segment where coordination requirements
are substantially lower, inter-haul sequential interdependency. This distinction in level of
interdependence between segments, though subtle, has important risk-sharing implications
for carriers, in addition to the production efficiency implications examined in this study.
For instance, Silverman et al. (1997, p. 36) state that “the hub-and-spoke nature of LTL
carriage makes drivers throughout the network highly interdependent, requiring a degree of
coordination for which hierarchy (company driver) offers a lower cost than market (independent owner-operators). TL carriage requires far less coordination, making independents
a more feasible contractual form.” However, the implications of increased interdependence

66

for LTL carriage extend well beyond the decision to outsource some portion of shipments
to owner-operators. For instance, higher interdependence may impact capital investment
decisions (e.g., equipment maintenance, lease/purchase tradeoffs, investments in monitoring
technologies) and the nature of shipper-carrier relationships.
For researchers, we believe that our study provides strong evidence against bifurcation
of motor carriers by market segment. Clearly the market segments for TL carriage and
LTL carriage are distinct, but many large and medium-sized motor carriages—more than
25 percent in our sample—participate concurrently within both segments. Our data also
suggests that carriers of both classes (I and II) are equally likely to concurrently participate
in both segments. The literature has viewed the output decision as a critical dimension of
a motor carrier’s strategic orientation (Feitler et al., 1998). We demonstrate the complex,
non-linear relationship between this dimension and an important firm outcome—operational
efficiency. Bifurcation of carriers into these segments forces an unnatural dichotimization
on the industry and may significantly impair the ability to detect associations with critical
carrier outcomes.
Lastly, our analysis paints a rich picture of the nature of variability in operational efficiency
that exists within the motor carrier industry. Our DEA estimates of output-oriented efficiency
indicate that carriers in our sample were perfectly efficient (on the efficiency frontier) during
only four percent of the measurement occasions. While operational efficiency estimates
were generally high across the sample (Figure 2.2), nearly 75 percent of the variability in
operational efficiency was due to between-carrier differences—suggesting stability in efficiency
levels over time. Much of the research into the motor carrier industry after regulatory reform
has been focused on the issue of economies of scale; most frequently, this research has utilized
econometric functions which assume firms to be perfectly efficient (Coelli et al., 2005). We
provide evidence that this assumption my be unrealistic. Further, given that inefficient firms
may be unable to capitalize on economies of scale (if economies do exist), we contend that
researchers should first seek to better explain the stable, between-carrier differences that
exist in operational efficiency. Thus, new theory is needed to explain the determinants of

67

carrier inefficiency, along with empirics to asses its impact to operations and key business
outcomes.

2.6.2

Managerial Implications

A carrier’s decision of how much carriage to produce in each segment (TL and LTL) is an
important, potentially even vital matter of strategy (Feitler et al., 1998), impacting the
utilization of productive resources, altering interdependencies among workflows, changing coordination needs and coordination-related investments and influencing risk-sharing decisions.
Market-driven rewards may even exist for carriers who can offer shippers a wider range of
transportation services. Our study shows that the diversification decision impacts a carrier’s
overall level of operational efficiency. Our analysis provides managers with direct estimates
of industry average levels of efficiency across the production possibilities curve (Figure 2.4)
as well as estimated effects of altering the production decision (Figure 2.5). Our results offer
separate conclusions for carriers that predominantly participate in the LTL segment and for
those that predominantly participate in the TL segment.
Core LTL carriers committing sufficient production capacity (i.e., resources) toward the
TL carriage may be able to realize sizable gains in efficiency. Among the carriers in our sample,
operational efficiency was greatest for those allocating 74 percent of production output to
LTL carriage (where output is measured in miles) and 26 of production output to TL carriage.
We theorize that, at this point, carriers are best able to leverage underutilized productive
resources (e.g., labor, capital equipment) toward TL production. An overcommitment
of production capacity to TL carriage may have undesirable consequences; operational
inefficiency increases, on average, once TL production exceeds 26 percent. Thus, managers
seeking to improve resource utilization and overall efficiency through a diversification strategy
into the TL segment should be mindful of the balance between a sufficient commitment and
over-commitment of productive resources and production capacity.
Core TL carriers diversifying (by entering the LTL segment) will not experience the
immediate efficiency gain their LTL peers diversifying into TL enjoy. Instead, these core TL

68

carriers appear to face significant coordination challenges associated with interdependence in
the production of LTL carriage. Specifically, core TL carriers who fail to sufficiently penetrate
the LTL market—those who allocate less than 35 percent of production output toward
the LTL carriage—risk a substantial increase in production inefficiency. McMullen and
Okuyama (2000) suggest that carriers who expand into the LTL segment may do so to pursue
a differentiation strategy. While differentiation or even full-line diversification (Penrose, 1959)
may indeed be the underlying impetus for core TL carriers to expand into the LTL market,
managers should be cognizant of the level of investment required to remain competitive.
Barring sufficient resource investment and market penetration, expansion challenges are
likely to manifest themselves in the form of operational inefficiency.
Accordingly, it is imperative to note that achieving a high level of production efficiency
(i.e., technical efficiency) is not an end in itself. Cost reductions realized from gains in
production efficiency may be overshadowed by revenues forgone when a carrier’s product mix
is suboptimal in the marketplace. Thus, carriers must jointly consider allocative efficiency,
including shipper preferences and marginal revenues in product lines, within their production
decisions. As one example, it may be the case that carriers offering a complete range of
services can demand a price premium from certain classes of shippers, such those who prefer
fewer but closer relationships with their transportation service providers. Given the general
competitiveness of the industry it is unclear how widespread these premiums are in the
transportation marketplace. Existing evidence has instead emphasized the importance of
lean and efficient production. For example, recent research by Fawcett et al. (2016) provides
insight into the relationship between resource leanness, resource slack and financial outcomes
for motor carriers operating in the TL segment. Their research indicates that leanness
is a key driver of financial performance for motor carriers, as has been found in several
other industries. Thus, Fawcett et al.’s (2016) findings provide underlying support to the
importance of fully utilizing carrier resources, as we argue here, while drawing a link to firm
financial performance. We caution managers, however, not to neglect the marketplace and
the need for allocative efficiency within their production decisions.

69

2.6.3

Limitations and Directions for Future Research

We identify two primary limitations of our study. First, our sample, while advantageous
to our test of theory, may not provide an accurate representation of the current or future
population of the U.S. motor carrier industry. Thus, caution should be exercised in the
generalization of our findings. Second, we utilized a two-stage data analysis strategy where
we estimated DEA efficiency scores in a first stage and then, in a second stage, regressed
these estimates on a set of predictors. DEA scores represents benchmarks of efficiency and
may contain error, since they are only estimates of a productive unit’s level of efficiency
relative to a sampled population. While we believe that our sample of 600 observations is
sufficiently large to avoid severe errors in our solution of a production possibilities frontier,
we recognize that non-random error in the resulting efficiency estimates might bias statistical
results of our second-stage analysis.
We contend that explaining stable, between-carrier differences in production efficiency
remains a fruitful avenue for future research. Such research would not only increase our
theoretical understanding of the motor carriers, but could be impactful to the motor carrier
industry, shippers and larger economies. Additionally, we believe that greater insight is
warranted into the role of interdependence in the motor carrier industry. For instance, future
research could examine implications for risk-sharing, as well as coordination challenges
associated with greater interdependence.

2.7

Conclusion
Our study examined the relationship between a motor carrier’s diversification of produc-

tion into a new market segment and operational efficiency. Using panel data obtained from
transportation archives, we provide evidence that the effects of diversification on operational
efficiency depend on shifts in organizational interdependence, impacting a carrier’s ability to
efficiently employ existing, underutilized resources in the new segment.

70

REFERENCES

71

REFERENCES
American Trucking Association (ATA) (2017). Reports, Trends & Statistics. Retrieved
from http://www.trucking.org/News_and_Information_Reports_Industry_Data.aspx [Accessed 5-May-2017].
Bairam, E. I. and Kahya, E. (1998). Production versus cost functions: Unreliability of the
duality theorem in accounting and economics. In Bairam, E. I., editor, Production and
Cost Functions, chapter 3, pages 42–53. Ashgate, Brookfield, VT.
Banker, R. D., Charnes, A., and Cooper, W. W. (1984). Some models for estimating
technical and scale inefficiencies in data envelopment analysis. Management Science,
30(9):1078–1092.
Beattie, B. R., Taylor, C. R., and Watts, M. J. (2009). The economics of production.
Krieger Publishing Company, Malabar, Florida.
Belzer, M. H. (2000). Sweatshops on wheels: Winners and losers in trucking deregulation.
Oxford University Press, New York, NY.
Bogetoft, P. and Otto, L. (2010). Benchmarking with DEA, SFA, and R, volume 157 of
Internation Sereis in Operations Research & Management Science. Springer Science &
Business Media, New York, NY.
Bryk, A. S. and Raudenbush, S. W. (2002). Hierarchical linear models for social and
behavioral research: Applications and data analysis methods. Sage, Newbury Park, CA.
Bureau of Labor Statistics (BLS) (2017). Industries at a glance: Truck transportation:
NAICS 484. Retrieved from https://www.bls.gov/iag/tgs/iag484.htm [Accessed 5-May2017].
Business Wire (2017).
Knight Transportation and Swift Transportation announce all stock transaction with a combined enterprise value of $6 billion.
Retrieved from http://www.businesswire.com/news/home/20170410005557/en/KnightTransportation-Swift-Transportation-Announce-Stock-Transaction [Accessed 5-May-2017].
Cantor, D. E., Celebi, H., Corsi, T. M., and Grimm, C. M. (2013). Do owner–operators
pose a safety risk on the nation’s highways? Transportation Research Part E: Logistics
and Transportation Review, 59:34–47.
Cantor, D. E., Macdonald, J. R., and Crum, M. R. (2011). The influence of workplace
justice perceptions on commercial driver turnover intentions. Journal of Business Logistics,
32(3):274–286.
Caves, D. W., Christensen, L. R., and Tretheway, M. W. (1984). Economies of density
versus economies of scale: Why trunk and local service airline costs differ. The RAND
Journal of Economics, 15(4):471–489.
Clarke, R. L. and Gourdin, K. N. (1991). Measuring the efficiency of the logistics process.
Journal of Business Logistics, 12(2):17.

72

Cleveland, G., Schroeder, R. G., and Anderson, J. C. (1989). A theory of production
competence. Decision Sciences, 20(4):655–668.
Coelli, T. J., Rao, D. S. P., O’Donnell, C. J., and Battese, G. E. (2005). An introduction
to efficiency and productivity analysis. Springer, New York, NY, 2 edition.
Cohen, J., Cohen, P., West, S. G., and Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences. Routledge, New York, NY, 3 edition.
Cook, W. D., Tone, K., and Zhu, J. (2014). Data envelopment analysis: Prior to choosing
a model. Omega, 44:1–4.
Corsi, T. M. and Grimm, C. M. (1987). Changes in owner-operator use, 1977–1985:
Implications for management strategy. Transportation Journal, 26(3):4–16.
Costello, B. and Suarez, R. (2015). Truck driver shortage analysis. Technical report,
American Trucking Associations: Arlington, VA.
Creel, M. and Farell, M. (2001). Economies of scale in the U.S. airline industry after
deregulation: a Fourier series approximation. Transportation Research Part E: Logistics
and Transportation Review, 37(5):321–336.
Department of Transportation (DOT) (2017). Form M: Annual Report Form for
Class 1 & Class 2 Motor Carriers of Property and Household Goods. Retrieved from
https://www.fmcsa.dot.gov/mission/form-m [Accessed 5-May-2017].
Durbin, J. (1960). The fitting of time-series models. Revue de l’Institut International de
Statistique, 28(3):233–244.
Farrell, M. J. (1957). The measurement of productive efficiency. Journal of the Royal
Statistical Society, 120(3):253–290.
Fawcett, A. M., Jin, Y. H., Hofer, C., Waller, M. A., and Brazhkin, V. (2016). Sweating
the assets: Asset leanness and financial performance in the motor carrier industry. Journal
of Business Logistics, 37(1):43–58.
Feitler, J. N., Corsi, T. M., and Grimm, C. M. (1997). Measuring firm strategic change in
the regulated and deregulated motor carrier industry: An 18 year evaluation. Transportation
Research Part E: Logistics and Transportation Review, 33(3):159–169.
Feitler, J. N., Corsi, T. M., and Grimm, C. M. (1998). Strategic and performance changes
among LTL motor carriers: 1976-1993. Transportation journal, 37(4):5–12.
Fitzmaurice, G. M., Laird, N. M., and Ware, J. H. (2012). Applied longitudinal analysis.
John Wiley & Sons, Hoboken, NJ.
Giordano, J. N. (2008). Economies of scale after deregulation in LTL trucking: A test case
for the survivor technique. Managerial and Decision Economics, 29(4):357–370.
Greene, W. (2004). The behaviour of the maximum likelihood estimator of limited dependent
variable models in the presence of fixed effects. The Econometrics Journal, 7(1):98–119.

73

Ha, H.-K., Yoshida, Y., and Zhang, A. (2010). Comparative analysis of efficiency for major
Northeast Asia airports. Transportation Journal, 49(4):9–23.
Han, C., Corsi, T. M., and Grimm, C. M. (2008). Why do carriers use owner operators in
the U.S. for-hire trucking industry? Transportation Journal, 47(3):22–35.
Hanna, J. B. and Maltz, A. (1998). LTL expansion into warehousing: A transaction cost
analysis. Transportation Journal, 38(2):5–17.
Hoskisson, R. E. and Hitt, M. A. (1990). Antecedents and performance outcomes of
diversification: A review and critique of theoretical perspectives. Journal of Management,
16(2):461–509.
Jara-Díaz, S. R. and Basso, L. J. (2003). Transport cost functions, network expansion and
economies of scope. Transportation Research Part E: Logistics and Transportation Review,
39(4):271–288.
Jin, Y. H., Swanson, D., Waller, M., and Ozment, J. (2017). To survive and thrive under
hypercompetition: An exploratory analysis of the influence of strategic purity on truckload
motor-carrier financial performance. Transportation Journal, 56(1):1–34.
Johnston, A. and Ozment, J. (2013). Economies of scale in the U.S. airline industry.
Transportation Research Part E: Logistics and Transportation Review, 51:95–108.
Keeler, T. E. (1989). Deregulation and scale economies in the U.S. trucking industry:
An econometric extension of the survivor principle. The Journal of Law and Economics,
32(2):229–253.
Kleinsorge, I. K., Schary, P. B., and Tanner, R. D. (1991). The shipper-carrier partnership:
A new tool for performance. Journal of Business Logistics, 12(2):35.
Kor, Y. Y. and Mahoney, J. T. (2000). Penrose’s resource-based approach: The process
and product of research creativity. Journal of Management Studies, 37(1).
Kor, Y. Y., Mahoney, J. T., Siemsen, E., and Tan, D. (2016). Penrose’s ‘The Theory of
the Growth of the Firm’: An exemplar of engaged scholarship. Production and Operations
Management, 25(10):1727–1744.
Liu, J. S., Lu, L. Y., and Lu, W.-M. (2016). Research fronts in data envelopment analysis.
Omega, 58:33–45.
McDonald, J. (2009). Using least squares and tobit in second stage DEA efficiency analyses.
European Journal of Operational Research, 197(2):792–798.
McMullen, B. S. and Okuyama, K. (2000). Productivity changes in the U.S. motor carrier
industry following deregulation: a Malmquist index approach. International Journal of
Transport Economics, 27(3):335–354.
McMullen, B. S. and Tanaka, H. (1995). An econometric analysis of differences between
motor carriers: Implications for market structure. Quarterly Journal of Business and
Economics, 34(4):16–29.

74

Mejza, M. M. and Corsi, T. M. (1999). Assessing motor carrier potential for improving
safety processes. Transportation Journal, 38(4):36–50.
Miller, J. W. and Saldanha, J. P. (2016). A new look at the longitudinal relationship
between motor carrier financial performance and safety. Journal of Business Logistics,
37(3):284–306.
Miller, J. W., Stromeyer, W. R., and Schwieterman, M. A. (2013). Extensions of the
Johnson-Neyman technique to linear models with curvilinear effects: Derivations and
analytical tools. Multivariate Behavioral Research, 48(2):267–300.
Min, H. and Ahn, Y.-H. (2017). Dynamic benchmarking of mass transit systems in the
United States using data envelopment analysis and the Malmquist productivity index.
Journal of Business Logistics, 38(1):55–73.
Min, H. and Lambert, T. (2002). Truck driver shortage revisited. Transportation journal,
42(2):5–16.
Nebesky, W., McMullen, B. S., and Lee, M.-K. (1995). Testing for market power in the
U.S. motor carrier industry. Review of Industrial Organization, 10(5):559–576.
Palich, L. E., Cardinal, L. B., and Miller, C. C. (2000). Curvilinearity in the diversification–
performance linkage: An examination of over three decades of research. Strategic Management Journal, 21(2):155–174.
Penrose, E. T. (1959). The theory of the growth of the firm. John Wiley, New York, NY.
Peteraf, M. A. (1993). The cornerstones of competitive advantage: A resource-based view.
Strategic management journal, 14(3):179–191.
Pettus, M. L. (2001). The resource-based view as a developmental growth process: Evidence
from the deregulated trucking industry. Academy of Management Journal, 44(4):878–896.
Rakowski, J. P. (1988). Marketing economies and the results of trucking deregulation in
the less-than-truckload sector. Transportation Journal, 27(3):11–22.
Rogers, M. M. and Weber, W. L. (2011). Evaluating CO2 emissions and fatalities tradeoffs
in truck transport. International Journal of Physical Distribution & Logistics Management,
41(8):750–767.
Scheraga, C. A. (2004). The relationship between operational efficiency and customer
service: A global study of thirty-eight large international airlines. Transportation journal,
43(3):48–58.
Scheraga, C. A. (2011). Strategic fit in the general freight motor carrier industry. Transportation Research Part E: Logistics and Transportation Review, 47(4):490–506.
Scheraga, C. A., Haslem, J. A., and Corsi, T. M. (1994). The financial mobility of motor
carriers: Pre-and post-deregulation. Transportation Research Part A: Policy and Practice,
28(5):401–414.

75

Silverman, B. S., Nickerson, J. A., and Freeman, J. (1997). Profitability, transactional
alignment, and organizational mortality in the U.S. trucking industry. Strategic Management
Journal, 18(Summer):31–52.
Singer, J. D. and Willett, J. B. (2003). Applied longitudinal data analysis: Modeling change
and event occurrence. Oxford University Press, New York, NY.
Sirmon, D. G., Hitt, M. A., and Ireland, R. D. (2007). Managing firm resources in dynamic
environments to create value: Looking inside the black box. Academy of Management
Review, 32(1):273–292.
Smith, J. (2016). Trucking industry hits road bump with rising diesel prices. Wall Street
Journal. Retrieved from https://www.wsj.com.
Smith, R. D., Corsi, T. M., and Grimm, C. M. (1990). Motor carrier strategies and
performance. Transportation Research Part A: General, 24(3):201–210.
Stock, J. R. (1988). The maturing of transportation: An expanded role for freight carriers.
Journal of Business Logistics, 9(2):15–31.
Suzuki, Y. (2007). Truck driver turnover: What rate is good enough? International Journal
of Physical Distribution & Logistics Management, 37(8):612–630.
Talluri, S. S., Kull, T. J., Yildiz, H., and Yoon, J. (2013). Assessing the efficiency of risk
mitigation strategies in supply chains. Journal of Business Logistics, 34(4):253–269.
Teece, D. J. (1982). Towards an economic theory of the multiproduct firm. Journal of
Economic Behavior & Organization, 3(1):39–63.
Thompson, J. D. (1967). Organizations in action: Social science bases of administrative
theory. Transaction Publishers, New Brunswick, NJ.
Tyler, E. (2017). Knight and Swift get speedy antitrust clearance to merge. Bloomberg:
The Bureau of National Affairs. Retrieved from https://www.bna.com/knight-swift-speedyn57982087453/ [Accessed 5-May-2017].
Wan, W. P., Hoskisson, R. E., Short, J. C., and Yiu, D. W. (2011). Resource-based theory
and corporate diversification: Accomplishments and opportunities. Journal of Management,
37(5):1335–1368.
Wanke, P. F. (2014). Efficiency drivers in the Brazilian trucking industry: A longitudinal study from 2002-2010. International Journal of Physical Distribution & Logistics
Management, 44(7):540–558.
Weber, M. M. and Weber, W. L. (2004). Productivity and efficiency in the trucking
industry: Accounting for traffic fatalities. International Journal of Physical Distribution &
Logistics Management, 34(1):39–61.
Wernerfelt, B. and Montgomery, C. A. (1988). Tobin’s q and the importance of focus in
firm performance. The American Economic Review, 78(1):246–250.

76

Winebrake, J. J., Green, E. H., Comer, B., Li, C., Froman, S., and Shelby, M. (2015). Fuel
price elasticities in the U.S. combination trucking sector. Transportation Research Part D:
Transport and Environment, 38:166–177.
Xu, K., Windle, R., Grimm, C., and Corsi, T. (1994). Re-evaluating returns to scale in
transport. Journal of Transport Economics and Policy, 28(3):275–286.
Ying, J. S. and Keeler, T. E. (1991). Pricing in a deregulated environment: The motor
carrier experience. The Rand Journal of Economics, 22(2):264–273.

77

ESSAY 3
CONSISTENTLY SAFE: HOW ISO 9000 CERTIFICATION PROMOTES
CONSISTENCY IN MOTOR CARRIER SAFETY

78

3.1

Introduction
The safe operation of large trucks on public roads is a matter of importance to motor

carriers, to shippers and to the welfare of the general public (Cantor et al., 2017; Miller,
2017). Vehicle accidents involving large trucks are inherently unpredictable and account
for a disproportionate share of total accidents on U.S. roads (Monaco and Redmon, 2012).
These accidents are also costly. The economic cost of fatal large truck accidents in the U.S.
exceeded 15 billion dollars in 2014—nearly 4.5 million dollars per fatal accident (Cantor
et al., 2017). Because of the financial and safety-related implications of these accidents,
motor carrier safety is routinely viewed as an issue of public policy (Cantor et al., 2009).
In the U.S., the Federal Motor Carrier Safety Administration (FMCSA) is charged
with reducing the number of crashes involving large, commercial vehicles. The FMCSA
collects and compiles data pertaining to crash investigations, moving violations and carrier
performance during roadside inspections as part of the agency’s safety compliance and
enforcement program, titled Compliance, Safety, Accountability (CSA). These carrier safety
data are used to “identify unsafe motor carriers and prioritize FMCSA enforcement resources
on those that pose the greatest safety risk” (FMCSA, 2017, p. 9). A central assumption of
the CSA program and, more broadly, public policy regarding motor carrier safety, is that a
carrier’s safety performance is influenced by its behavior and its management practices (Mejza
et al., 2003; Miller and Saldanha, 2016). This assumption has been widely supported in
the literature: some carriers tend to be safer than others and a carrier’s level of safety
performance tends to persist over time (Miller et al., 2017b).
Accordingly, a primary focus of the motor carrier safety literature has been to identify
systematic explanations for the presence of these stable, between-carrier differences in
safety performance. Identifying factors that contribute to motor carrier safety is important.
Regulatory agencies such as the FMCSA can utilize these factors when formulating new
policy and when evaluating existing policy, such as those pertaining to driver logs (Cantor
et al., 2009) or the effectiveness of the FMCSA’s new entrant program (Cantor et al., 2017).
Regulatory agencies can also use these factors to identify high-risk carriers and to prioritize

79

enforcement resources. For example, motor carrier safety tends to decline as turnover
increases in the driver workforce (Miller et al., 2017c) and when carriers experience greater
financial strain (Miller and Saldanha, 2016). Identifying factors that contribute to safety
also remains important to carriers—who risk human, equipment and financial loss—as well
as shippers, brokers and numerous other stakeholders.
Our research examines the relationship between motor carrier adoption of an organizational routine for quality management, International Organization for Standardization (ISO)
9000, and consistency in a carrier’s safety performance. The ISO 9000 series is a family
of international quality standards and quality guidelines that specifies what organizations
should do to produce products and services that consistently meet the expectations of the
organization’s customers as well as statutory and regulatory requirements (ISO, 2016). Prior
research has linked motor carrier adoption of ISO 9000 to reductions in the number of vehicle
accidents as well as improvements in FMCSA-collected safety performance measures (Naveh
and Marcus, 2007). Our research focuses instead on variability that occurs in a carrier’s safety
performance over time. The quality literature refers to a state of low outcome variability
as consistency and acknowledges that attainment of a high level of performance does not
imply consistency in performance (Su and Linderman, 2016). Consistency has long been
considered to be among the most critical of motor carrier attributes (Stephenson and Willett,
1969; Bardi et al., 1989; Morash and Clinton, 1997; Premeaux, 2002; Dobie, 2005; Voss et al.,
2006; Robinson et al., 2013). Yet, to date, the concept of consistent performance remains
unexplored within the motor carrier safety literature.
It is important to identify factors that promote consistency in the safety performance
of motor carriers, just as it is important to identify factors that promote a high level of
safety performance. First, the organizational competencies required to achieve a high level
of performance differ from those required to achieve consistency (Su and Linderman, 2016).
Second, the concept of consistency in safety performance carries unique implications for
carriers and policy makers. Intuitively, a carrier’s latent profile of safety performance might
be influenced by the carrier’s history regarding consistency in safety performance over time,

80

since variability indicates predictability at any given measurement occasion. From this
perspective, minimizing variability (maximizing consistency) in safety performance would be
beneficial to carriers and stakeholders—if carriers differ systematically in the variability of
their safety performance, uncovering the systemic causes of this variability can assist carriers
in improving the predictability of safety outcomes, such as vehicle accidents and safety
violations. These factors may also identify to regulators carrier-level traits or competencies
that reduce the likelihood of extreme instances of safety-related events, such as carrier
involvement in a large number of vehicle accidents within a given measurement occasion.
A less intuitive implication of safety consistency, and one that we also advance here, is
that some level of variability in a carrier’s safety performance might be beneficial, since
variability may catalyze change. To relate consistency and change we rely on the theory of
routine dynamics (Feldman and Pentland, 2003). We envision a motor’s carrier behavior as
being a function of its organizational routines, as has been posited regarding the behavior of
organizations in the general case (Knott, 2001). We then theorize the effects of ISO 9000 on
consistency and change given ISO 9000’s role as a higher-level organizational routine, or
metaroutine (Pentland et al., 2012).
The remainder of this essay proceeds as follows. First, we review the literature on motor
carrier safety and ISO 9000 quality assurance. Following these reviews, we present our
theoretical framework and formulate research hypotheses relating ISO 9000 to consistency
and change in motor carrier safety. We then provide our methodology and present the results
of hypothesis testing. The essay closes with a discussion of our findings and implications for
theory and managers.

3.2
3.2.1

Literature Review
Motor Carrier Safety

Table 3.1 summarizes recent motor carrier safety research. Existing work is organized around
several research streams including carrier adoption of monitoring technology (e.g., Cantor
et al., 2006, 2008, 2009; Hickman and Hanowski, 2011); labor topics, such as the use of

81

owner-operators, unionization and the effects of driver turnover on carrier safety (e.g., Corsi
et al., 2012; Monaco and Redmon, 2012; Miller et al., 2017c); truck driver behavior (e.g.,
Braver et al., 1992; Häkkänen and Summala, 2001; Morrow and Crum, 2004; Cantor et al.,
2010; Swartz et al., 2017); regulatory effectiveness and carrier behavior under governmental
regulation (e.g., Moses and Savage, 1992; Saltzman and Belzer, 2002; Chen, 2008; Cantor
et al., 2017; Miller, 2017; Miller et al., 2017b); and, lastly, the relationship between carrier
safety and firm-level characteristics such as firm size and financial performance (e.g., Naveh
and Marcus, 2007; Britto et al., 2010; Cantor et al., 2016; Miller and Saldanha, 2016).
However, as seen in Table 3.1, carrier consistency in safety performance has not been
incorporated into recent causal models of motor carrier safety. Indeed, a review of the motor
carrier safety literature across multiple disciplines (see Miller, 2017, for a discussion) did
not reveal any prior investigation into the topic of consistency in safety over time. This
absence may not be surprising, however, given that data on motor carriers’ longitudinal
safety performance was far less granular until only recently, in late 2010, when the FMCSA
launched its CSA program (Miller et al., 2017a). Since this time, research into carrier-level
safety performance has shifted toward longitudinal frameworks.
When viewed longitudinally, total variability in the safety performance of the motor
carrier population appears to dominated by time invariant, between-carrier differences.
In Miller and Saldanha (2016), between-carrier differences accounted for more than fifty
percent of the variation in each of three CSA Behavior Analysis and Safety Improvement
Category (BASIC) safety measures, vehicle maintenance (VM), hours of service compliance
(HOS) and unsafe driving (UD). However, carrier safety performance is not static; the
nature of longitudinal change in safety performance has also been the subject of research.
Hickman and Hanowski (2011) employ a difference-in-difference quasi-experimental design to
examine the effects of in-cab video monitoring on safety-related driver behaviors, finding that
adoption led to a significant decrease in at-risk behavior. Miller et al. (2017a) examine carrier
improvement in safety after implementation of the CSA program and ensuing disclosure of
carrier safety information. Miller et al. (2017b) estimate autoregressive relationships and

82

Table 3.1: Summary of the motor carrier safety literature, 2007–2017.

Carrier-level study

Motor Carrier Safety Performance Aspect
Performance
Performance Performance
Change
Consistency
Level

Key finding(s)

Naveh and Marcus (2007)

Accident rates are lower for ISO 9000 certified motor carriers
than for non-ISO certified carriers
Cantor et al. (2008)
Motor carrier adoption of safety technologies is highest among
carriers with poor records of safety performance
Cantor et al. (2009)
Electronic logbook adoption is associated with fewer hours
of service violations and fewer crashes
Carrier financial performance is positively associated with
Britto et al. (2010)
carrier safety
Hickman and Hanowski (2011) Carrier implementation of in-cab video monitoring technology
decreases at-risk driving behaviors
Corsi et al. (2012)
Motor carriers who employ unionized drivers tend to be safer
than those utilize non-union drivers
Monaco and Redmon (2012)
Owner-operator use leads to improved carrier safety outcomes, but is unrelated to average crash severity
Firm size is positively related to carrier safety, with varying
Cantor et al. (2016)
effect sizes across industry segments
Miller and Saldanha (2016)
Carrier safety is positively associated with both financial
performance improvement and static financial performance
Cantor et al. (2017)
Motor carriers who successfully complete the FMCSA’s new
entrant program are safer than those carriers who fail
Miller (2017)
Likelihood of a carrier exiting warning flag status for HOS
compliance is a function of elapsed time and carrier size
Safety improvement tapered over time in response to CSA
Miller et al. (2017a)
information disclosure; tapering influenced by carrier traits
Miller et al. (2017b)
FMCSA carrier safety measures (BASICs) serve as stable,
leading indicators of other safety measures
Miller et al. (2017c)
Carrier safety declines with increases in driver turnover; effect
is stronger for carriers with low activity control
This study
ISO certification attenuates carrier safety performancevariability relationship among large U.S. motor carriers

R/E

R

.

E

.

.

R/E

.

.

R

.

.

.

R

.

R

.

.

R

.

.

R

.

.

R

R

.

R

.

.

E

R

.

E

R

.

R/E

R

.

R

.

.

E

R

R/E

Notes: R = modeled as a response variable; E = modeled as a explanatory variable; R/E = modeled as both response and explanatory

83

longitudinal inter-relationships among CSA measures, identifying unique (but temporally
stable) reflexive changes in the measures in response to prior-period changes in other CSA
measures. Our research argues that temporally stable differences exist between carriers in
the consistency of their safety performance, and that these differences can be explained
by carrier-level factors (i.e., traits, competencies). We examine the effects of one such
factor, carrier certification to ISO 9000 quality assurance standards, since the principal
purpose of these standards is to promote uniformity and consistency in product and service
quality (Anderson et al., 1999; Benner and Veloso, 2008).

3.2.2

ISO 9000 Series Quality Assurance

The ISO 9000 series is a family of international quality standards and quality guidelines that
specifies what organizations should do to produce products and services that consistently
meet the expectations of the organization’s customers as well as statutory and regulatory
requirements (ISO, 2016). Early versions of the series (prior to 2000) were applicable only
to the manufacturing sector, however the scope of the ISO 9000 series was later expanded
to address quality assurance for service providers (Martínez-Costa et al., 2009). Since its
inception, companies across a wide variety of industries have sought to obtain certification of
compliance to ISO standards (Gray et al., 2015). To become ISO 9000 certified, organizations
must map their operating processes associated with quality production, implement replicable
sets of routines and procedures to assure a consistent level of quality and demonstrate
conformance to documented processes (Naveh and Marcus, 2005; Benner and Veloso, 2008).
In addition to promoting an improved organizational focus on quality, ISO certification
serves as a public signal of effective quality management (Anderson et al., 1999) and may
provide competitive benefits to firms, such as improved financial performance (Corbett
et al., 2005; Levine and Toffel, 2010; Lo et al., 2013). It is not uncommon for industrial
purchasers to require suppliers to become ISO 9000 certified, since a supplier’s ISO 9000
compliance should reduce the buyer’s need to monitor supplier behavior (Terlaak and King,
2006) while promoting uniformity in product or service quality across multiple, ISO 9000

84

certified suppliers (Su et al., 2015).
ISO 9000 series certification has been linked to improvements in the management of
workplace safety as well as improved safety outcomes for employees. Employees of ISO 9000
certified organizations report higher levels of employee involvement in safety management,
enhanced effectiveness of safety communication and feedback within their organizations
and increased comprehensiveness of safety-related training (Vinodkumar and Bhasi, 2011).
Levine and Toffel (2010) find that ISO certified employers experience lower on-the-job
death rates than non-ISO certified employers and are more likely to have zero filings for
injury-related workers’ compensation than their non-certified counterparts. The existence
of ties between ISO 9000 and workplace safety are not entirely surprising. Veltri et al.
(2013, p. 121) note that ISO 9000 series standards “point to the same basic set of best
practices” as other leading sets of formal standards used for safety management. Further,
ISO 9000 certified organizations tend to be more compliant with documented organizational
processes (Gray et al., 2015) and ISO 9000 quality management tools such as those used for
root cause analysis and continuous improvement can readily be applied to manage health
and safety in the workplace (Levine and Toffel, 2010).
Initial evidence suggests a positive link between ISO 9000 certification and safety
performance in the motor carrier industry. Naveh and Marcus (2007) contend that a motor
carrier’s adoption of ISO 9000 standards can lead to increased safety performance, primarily
through improvements in three areas: carrier driving procedures, maintenance of vehicle and
overall safety management. Using matched samples of certified and non-certified carriers,
Naveh and Marcus (2007) associate ISO 9000 certification with improved safety performance.
However, further evidence is needed to corroborate, update and extend their findings. First,
the ISO 9000 series in use by their sample of carriers—the 1994 version of the ISO series—has
since undergone three iterations of revisions (2000, 2008, 2015). Additionally, this 1994
ISO 9000 series was directed at quality assurance in manufacturing (Martínez-Costa et al.,
2009) as the series was not expanded for applicability to service providers until the 2000
revision. More recent updates include changes related to safety. For instance, the current

85

ISO 9000 series (2015) takes a more systematic approach to risk management, focusing
on preventative action through risk-based thinking. Next, the SafeStat indicators utilized
in Naveh and Marcus’s (2007) analysis are now obsoleted under the CSA program, having
been replaced by more granular measures of carrier safety performance (Miller et al., 2017a).
Lastly, as Miller and Saldanha (2016) note, Naveh and Marcus (2007) leave several lingering
questions regarding longitudinal components of their data collection and analysis. One such
question is if ISO 9000 series certification by motor carriers is associated with improved
consistency in safety performance, as would be expected under a management system focused
upon controlling quality. We present theory for this association next.

3.3

Theory and Hypothesis Development
An organization’s behavior is defined by its routines (Nelson and Winter, 1982). Organi-

zational routines are “repetitive, recognizable patterns of interdependent actions, carried
out by multiple actors” (Feldman and Pentland, 2003, p. 95). Theories of organizational
behavior commonly distinguish between two classes of routines: operational routines and
metaroutines (Knott, 2001). Operational routines guide day-to-day organizational behavior.
Metaroutines are higher-level routines—routines to govern and modify routines (Nelson and
Winter, 1982). The ISO 9000 series is regarded as an organizational metaroutine (Pentland
et al., 2012). Metaroutines such as ISO 9000 serve as integrating frameworks between
operational routines and the environment, regulating changes in operational routines (Lewin
et al., 2011), systematizing creative processes (Adler et al., 1999), identifying the need
for new routines (Schroeder et al., 2008) and providing a conduit for managerial intervention (Knott, 2001). The theory of routine dynamics (Feldman and Pentland, 2003) posits
that organizational routines can promote stability through structure and, simultaneously,
promote change through agency. We rely on this theory to hypothesize unique effects of a
motor carrier’s adoption of ISO 9000 on consistency and continuous improvement in safety
performance.

86

3.3.1

Structure: How ISO 9000 Promotes Consistency in Carrier Safety

Structure refers to the abstract idea of a routine, viewing a routine as a static pattern
of action, a habit, a heuristic or a standard operating procedure (Feldman and Pentland,
2003). In structure, routines serve as a form of organizational memory that direct recurring
activities, thereby promoting efficiency and reducing variation in organizational response
to stimuli (Huber, 1991; Knott, 2001). When viewed in this frame, routines are considered
slow to change and do so only when prompted by some exogenous event, such as adoption of
new technology, experience of failure or attainment of a milestone (Feldman, 2000; Gersick
and Hackman, 1990). This feature of stability, along with the efficiency and legitimacy that
stable routines provide, is prized by organizations (Feldman and Pentland, 2003).
Viewing ISO 9000 from this structural perspective, a motor carrier’s adoption of the
ISO 9000 series metaroutine should lead to increased stability in operational routines and
consistency in organizational outcomes. The concepts of stability and consistency are
interrelated. In the safety and risk literature, stability is a condition signaled by the return of
a system to some original state after perturbation (Wildavsky, 1988). This is to say, a stable
system is mean-reverting (trend-stationary) in the presence of shocks. Consistency describes
the pattern of variability (i.e., of an outcome) when the system is under perturbation (Su and
Linderman, 2016). Accordingly, achieving a high level of consistency requires stability, and
systems that are more stable should also tend to produce outcomes that are more consistent.
Routines promote consistency by enabling an efficient and predictable organizational response
to stimuli (Nelson and Winter, 1982; Knott, 2001). Additionally, metaroutines promote
consistency by regulating changes in existing operational routines and by turning nonroutine tasks into routine tasks through the development of new operational routines (Lewin
et al., 2011). Indeed, ISO contends that the 9000 series directly promotes consistency
by requiring organizations to implement standardized and replicable sets of routines and
procedures (Naveh and Marcus, 2005).
Motor carriers are regularly exposed to perturbations within the operational and regulatory environments in which they operate. Mello and Hunt (2009) describe the business

87

environment for motor carriers as “highly competitive, uncertain and demanding” (p. 21).
Carriers must contend with turbulence in highway conditions and congestion, in the demand
market (Dobie, 2005) and within factor markets (Torry and Murray, 2015). A carrier’s
implementation of routines which formalize organization rules, procedures and processes has
been linked to improvements in carrier outcomes. Saldanha et al. (2013) link carrier use of
formalized routines that detail rules, behavior and standards for drivers with operational
and market performance. Miller et al. (2017c) find that carriers who routinize driver-related
activities mitigate the effects of high driver turnover on unsafe driving behavior. ISO 9000
should promote the standardization and formalization of processes associated with the
production of motor carriage by regulating changes in existing routines and inventing new
routines. As the carrier’s production system becomes more stable to perturbations in the
environment, safety outcomes should vary less, improving consistency.
H1 : ISO 9000 certified motor carriers exhibit greater consistency in safety performance
than non-certified carriers.

3.3.2

Agency: How ISO 9000 Promotes Continuous Improvement in Carrier
Safety

Agency recognizes that action in the performance of routines is situated and that human
performance of a routine in practice may diverge from the performance of that routine in
principal (Feldman and Pentland, 2003). For instance, organizational actors may need to
adjust their performance of a routine to account for unique or changing contexts. There
are also practical constraints on the degree of completeness with which organizational
routines, even if formalized in writing, can routinize human action while anticipating and
addressing all possible contingencies. Other variations in routine performance may simply
be unintentional or may be a function of differences between actors in their understanding
and subjective interpretations of an organizational routine. Routine dynamics considers
organizational actors to be “knowledgeable and often reflective” (p. 506) in their performance
of routines (Feldman et al., 2016). Accordingly, actors and their organizations learn from
variations introduced in routine performance—purposeful or not—and choose to retain those

88

variations that offer added utility to the organization, appeal to its actors or improve fit with
other organizational routines (Feldman and Pentland, 2003). It is because of this sequence
of variation, reflection and selection that routines become dynamic, capable of endogenous
change. As routines change endogenously so may organizational outcomes, resulting in
continuous change (Feldman, 2000).
Routine dynamics suggests that the ISO 9000 series metaroutine can be a source for
continuous change in motor carrier safety. In production contexts, continuous (or constant)
positive change is referred to as continuous improvement (Anand et al., 2009). Zangwill
and Kantor (1998) view the process of organizational learning that underlies continuous
improvement as a cyclical pattern of trial and error, where managers retrospectively assess
outcomes from trials of variations in strategy. The literature on safety and risk has also
advocated for a trial and error approach. For instance, Wildavsky (1988) likens the process
of attaining increased safety to trial-and-error entrepreneurial activity, arguing that safety
is a condition to be discovered and continuously improved. The ISO 9000 series directly
encourages continuous improvement (Terziovski et al., 2003), thereby promoting the evolution
of operational routines under its management. While the motor carrier literature has
examined longitudinal change in safety (see Table 3.1), it has not documented the role of
metaroutines in promoting continuous safety improvement. Routine dynamics suggests that
this relationship should exist. As a dynamic routine, ISO 9000 should encourage change as
carriers select and retain variations associated with performance of the ISO 9000 metaroutine
and underlying routines.
H2 : ISO 9000 certified motor carriers exhibit greater improvement in safety performance
than non-certified carriers.

3.4
3.4.1

Methodology
Data and Sample

We test our research hypotheses using a combination of archival and secondary data on
large, for-hire motor carriers in the combination (articulated vehicle) segment. By focusing

89

our analysis on a relatively homogeneous subsection of the carrier population we are able to
reduce our exposure to uncontrolled variance on safety outcomes that might be attributed
to factors such as latent differences in competitive pressures between for-hire and private
fleets (Miller et al., 2017b) and differences in carrier operational characteristics between
the combination and straight-truck (non-articulating) segments. Our development of a
sample is discussed next. Monthly observations of carrier safety performance were obtained
from the FMCSA’s Safety Measurement System database on the following BASICs: vehicle
maintenance, hours-of-service compliance and unsafe driving (for a review, see Cantor et al.,
2017). Information on carrier characteristics were obtained from the FMCSA’s Motor Carrier
Census and from the results of an annual survey data collected on large U.S. motor carriers
by Inbound Logistics, a leading practitioner-oriented publication for the logistics industry.
Motor carriers were identified by their presence in Inbound Logistics’s annual list of
leading carriers (“Top 100 Truckers”, 2013, 2014, 2015, 2016). These lists publish carrierreported information regarding company demographics and operational characteristics, such
as each carrier’s operating area and the types of services provided. We utilize these lists for
our research since they contain carrier-reported information on ISO certification status and
since they are comprised of only established carriers in the industry, who have a history of
safety performance. Using the compiled information, we identified for-hire carriers in the
combination segment and linked these carriers to the Safety Measurement System database
and carrier census data using a DOT number or, for parent firms, DOT numbers. Since
some initial, transient effects from ISO adoption might be associated with the maturing of
an organization’s quality culture (Terziovski et al., 2003) or dwindle from a lack of long-term
organizational commitment (Sroufe and Curkovic, 2008), we examined only carriers whose
ISO certification status (certified or not certified) remained stable for all four years and,
thus, were also present in each of the carrier lists.
Our data collection resulted in a sample of 78 carriers, of which 22 were ISO 9000 certified.
Consistent with prior research into outcomes of ISO 9000 certification (e.g., Corbett et al.,
2005; Naveh and Marcus, 2007; Gray et al., 2015), we use a matching technique to create an

90

“artificial control group” (Wangenheim and Bayón, 2007, p. 39). To do so, we paired certified
carriers to non-certified carriers by propensity score. The use of propensity score matching
is an appropriate method to remove bias in observational studies and quasi-experiments
where confounds may be introduced through non-randomized assignment or self-selection
on treatment conditions (Rosenbaum and Rubin, 1983). Propensity scores for ISO 9000
certification were obtained using logistic regression. A logistic regression of ISO certification
status on a series of covariates (carrier age, size, segment, labor profile and initial safety
performance) fit significantly better to the data than a null model (χ2∆ (9) = 28.30, p < .01)
and correctly predicted certification status for 80.77 percent of carriers (50.00 percent of
ISO-certified carriers). We used nearest neighbor caliper matching (Althauser and Rubin,
1970) to match with replacement (Dehejia and Wahba, 1999) ISO 9000 certified carriers
with non-certified carriers by propensity score. Calipers were specified at a tolerance of .20
of the standard deviation of the propensity score, as recommended by Austin (2011). Of
the 22 ISO-certified carriers in our original sample, 19 successfully paired with a similar
non-certified carrier, resulting in a matched sample of 38 carriers.

3.4.2

Variable Operationalization

This subsection provides an overview of the operationalization of variables used in the study.
Descriptive statistics for all variables are provided within Table 3.2.

3.4.2.1

Dependent Variables

Inconsistency.

Consistency is measured as variability around a trend (Stephenson and

Willett, 1969), and a greater degree of variability reflects greater inconsistency. We operationalize inconsistency as the magnitude of fluctuation in a carrier’s safety performance,
measured as the scaled standard deviation of first derivative estimates of a given BASIC
with respect to time. Researchers commonly operationalize measures of variability within
a unit (e.g., consistency, volatility) using the standard deviation of raw scores (Hoffman,
2007; Deboeck et al., 2009; Matta et al., 2017). However, for a set of ordered observations

91

Table 3.2: Descriptive statistics.
Pearson Product-Moment Correlation
Variable

Mean S.D.

(1) VM Performance†
(2) HOS Performance

†

(3) UD Performance†

(1)

3.12

.93

.31

.33

.46∗

1.30

.65

.28

(4) VM Change

−.47

.37 −.57

(5) HOS Change

−.12

.20 −.11

−.07

(6) UD Change

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

.22
∗

.35 −.04

−.26

−.10

−.90∗

−.09
−.63

.22

.04
∗

−.07

−.31

∗

−.07

.03

−.45∗

.13

.10

−.20

−.76∗

.19

.81∗

.18

−.21

∗

−.70

.06

.74∗

−.09

(7) VM Inconsistency

.04

.03

.15

.51

(8) HOS Inconsistency

.01

.01

.28

.86∗
∗

.93∗

(9) UD Inconsistency

.05

.07

.28

.82

(10) ISO Certification

.50

.51

.23

−.02

.04

−.18

.09

.49

.25

<.01

−.29

−.19

.22

.26

.25

.21

.14

.20

−.12

.29

−.16

−.89∗

−.65∗

−.56∗

.05

−.28

−.13

∗

∗

∗

−.06

−.16

(11) Union Labor

.53

(12) Inspections‡

7.56

(13) Carrier Age

‡

3.53

−.34∗

1.18 −.01
.36 −.38

(12)

∗

−.48

∗

−.11

.13

.45

∗

Notes: n = 38;∗ p < .05 (two-tailed); † initial (2012) safety performance; ‡ log-transformed.

92

−.04

−.56

−.04

−.51

−.05

−.53

.10
.46∗

containing fluctuation in the presence of a linear time trend, the standard deviation of raw
scores (i.e., zeroth derivatives) is a positively-biased estimate of the true standard deviation
of fluctuations. We adopt, instead, to measure the inconsistency as the average deviation in
the rate of change, as proposed by Deboeck et al. (2009), using first differences of the raw
scores. We then divide this measure by the square root of two since the variance of the ith
difference of a set of independent, random observations and the variance of raw values of
those observations differ by a scaling factor equal to the ith central binomial coefficient (a
value of two for the first difference).
Figure 3.1 illustrates the potential for measurement bias when the standard deviation
of raw scores is taken in the presence of linear trend. Data used within the figure came
from observations of unsafe driving measures on two carriers within our original sample.
Average unsafe driving performance differed little between the two carriers during this period,
with a mean of 1.91 for Evans Delivery Company (leftmost plot) and a mean of 1.89 for
C.R. England (rightmost plot). In comparing the two plots, longitudinal performance for
Evans Delivery Company is characterized by a relatively small amount of upward trend
but a large degree of variability. In contrast, longitudinal performance for C.R. England is
characterized by a large amount of downward trend but only a small-to-moderate degree of
variability. A researcher comparing the two carriers using the standard deviations of the
raw values might mistakenly conclude that Evans Delivery Company was twice as consistent
on unsafe driving as C.R. England was during this period. However, upon examining the
scaled standard deviations of first derivative estimates, the exact opposite, and seemingly
more correct conclusion would be made that C.R. England was the more consistent carrier.
Change. We operationalize change in safety performance using the observed linear change
in a carrier’s BASIC score throughout the four year period. This is measured as the sum
of first differences of observations on the BASIC score. When examining BASIC scores, a
greater level of safety is indicated by a lower BASIC score. Accordingly, positive change
reflects a decline in carrier safety while negative change reflects a safety improvement.

93

Evans Delivery Co. (38111)

C.R. England (28406)

2.50

2.50

2.25

2.25

2.00

2.00

1.75

1.75

1.50

1.50
2013
2014
2015
2016
2017
Std. dev. of raw scores:
.14
Scaled std. dev. of first derivatives: .06

2013
2014
2015
2016
2017
Std. dev. of raw scores:
.31
Scaled std. dev. of first derivatives: .03

Figure 3.1: Comparison of alternative consistency measures using unsafe
driving data for two carriers.
3.4.2.2

Focal Predictor

Carrier observations on our focal predictor, ISO 9000 Certification, were obtained from
survey data published by Inbound Logistics. Half of the 38 motor carriers in our sample
held ISO 9000 certification after matching. In the matched sample, ISO 9000 is statistically
unrelated to the series of covariates utilized for matching (χ2∆ (11) = 10.77, p = .46). While
this suggests a partial mitigation of threats due to confounds by covariates such as carrier
size, age and initial level of safety performance, we retain the covariates used for matching
to improve efficiency in estimation and to reduce confounds between contextual variables
and safety outcomes.

3.4.2.3

Covariates

Initial Performance. Intuitively, a carrier’s level of performance should relate to their level
of consistency and likelihood of committing resources toward change. To avoid possible
confounds, we control for carrier performance on a BASIC using the initial performance
of that carrier on the BASIC. We reduce our potential exposure to measurement error by

94

operationalizing initial performance using the carrier’s annual performance in 2012, the year
prior to the start of our sample. For any of the three BASICs used in this research (VM,
HOS and UD), higher scores reflect poorer safety performance.
Labor Unionization. We include a dummy variable in our model to account for carrier use
of unionized employees. The literature is mixed with regard to the effects of unionization on
carrier safety outcomes. Miller et al. (2017c) do not find unionization to be a significant
predictor of carrier safety performance on vehicle maintenance, hours-of-service compliance
or unsafe driving. In contrast, Corsi et al. (2012) find that carriers utilizing unionized
labor perform better on FMCSA safety metrics and have fewer vehicle accidents than their
non-union counterparts. While the true nature of the relationship between carrier use of
unionized labor and safety outcomes remains unclear, we include the covariate in our model
to guard against potential confounding effects.
Carrier Size/Inspections. Cantor et al. (2016) theorize that large carriers may hold a safety
advantage over smaller carriers due to superior access to human, technological and financial
resources. Within the recent motor carrier safety literature, size is operationalized using one
of two forms: as the number of power units (Corsi et al., 2012; Cantor et al., 2016, 2017)
or as the natural logarithm of the number of power units (Britto et al., 2010; Miller and
Saldanha, 2016; Miller, 2017; Miller et al., 2017a). A significant relationship between size
and safety has been found in several studies (e.g., Cantor et al., 2016, 2017), but results
were mixed or inconclusive in others (e.g., Britto et al., 2010; Miller et al., 2017b,c).
A carrier’s size and the number of times they are inspected over a period of time are
naturally correlated. We log-transform both carrier size and number of inspections in our
sample (for skewness; Britto et al., 2010) and find a high correlation of .88 (p < .01) between
the two. When examined annually, each of the two measures also exhibit very high stability,
both having an intra-class correlation of .98. To avoid multicollinearity we select one of
the two measures, number of inspections, for inclusion as a covariate in our model. Most
importantly, data from the FMCSA on carrier size are based on carrier filings that do
not occur on a set schedule (Miller et al., 2017a); carrier size updates do not correspond

95

directly with FMCSA updates regarding carrier safety performance. In contrast, carrier
inspection data is updated in the Safety Measurement System database monthly along with
updates to carrier safety performance. While endogeneity concerns from measurement bias
on carrier size are minimal (i.e., given stability of carrier size over time), we select number
of inspections out of an abundance of caution.
Carrier Age. We include carrier age in our model as a covariate since older carriers may
have had more time to mature their organizational routines. An estimate of carrier age
was obtained from the FMCSA’s motor carrier census, using the carrier’s date of entry into
the Motor Carrier Management Information System database. Carrier age in years was
log-transformed for skewness.
Carrier Segment.

Consistent with Cantor et al. (2017) and others, we include dummy

variables in the regression equation to control for potential differences due to carrier segment.
We identify the carriers in our sample by their primary segment: less-than-truckload (16),
truckload (11), specialized/flatbed (7) and bulk (4). We use less-than-truckload as the base
category.

3.4.3
3.4.3.1

Model Specification
Contextual Factors

The literature on ISO 9000 wide recognizes that potential benefits from ISO 9000 adoption
differ based on contextual factors. Lo et al. (2013) show that poor performing firms tend
to benefit most from ISO 9000 adoption. Their findings are consistent with the ISO 9000
literature: firms with below-average quality management stand to benefit the most from
implementing a quality management program. Accordingly, we allow for the effects of ISO
9000 certification on motor carrier safety outcomes to differ based on the carrier’s initial level
of safety performance. Given the findings of Lo et al. (2013), we would expect a negative
coefficient to be estimated on the interaction term, such that higher initial BASIC scores
(greater unsafety) amplifies a negative effect of ISO certification on inconsistency and change

96

in safety performance.

3.4.3.2

Regression Equations

The model of safety consistency is defined within Equation 3.1 and the model of safety
change is defined within Equation 3.2. In each of these these models, observations on set
M = {VM, HOS, UD} of correlated response variables, given by Y M , are made on carrier i.
In Equation 3.1, the response of inconsistency is given by Ẏ . In Equation 3.2 the response
of change is given by Ÿ . We control for inconsistency in Equation 3.2 since variability
associated with trial-and-error should manifest in safety outcomes and, given Equation 3.1,
may otherwise confound the relationship between ISO 9000 and change if inconsistency is
also associated with change.
ẎiM = β0M +

3
X

βkM × SEGMENTi + β4M × UNION_LABORi

(3.1)

k=1

+ β5M × ln(CARRIER_AGEi ) + β6M × ln(INSPECTIONSi )
+ β7M × ISOi + β8M × INITIAL_PERFORMANCEM
i
+ β9M × ISOi × INITIAL_PERFORMANCEM
i

ŸiM = β0M +

3
X

βkM × SEGMENTi + β4M × UNION_LABORi

(3.2)

k=1

+ β5M × ln(CARRIER_AGEi ) + β6M × ln(INSPECTIONSi )
+ β7M × ISOi + β8M × INITIAL_PERFORMANCEM
i
+ β9M × ISOi × INITIAL_PERFORMANCEM
i
M
+ β10
× INCONSISTENCYM
i

3.5

Results
Estimation of Equations 3.1 and 3.2 was performed using seemingly unrelated regres-

sion (Zellner, 1962). We selected seemingly unrelated regression because the technique
can improve estimation efficiency (beyond that of ordinary least squares) by accounting

97

for the cross-equation residual covariance structure. We test for the appropriateness of
this technique in our setting using Breusch and Pagan’s (1980) Lagrange multiplier test.
We are able to reject a null hypothesis of diagonality of the residual correlation matrix
(χ2 (15) = 55.11, p < .01), and conclude that the equations are more efficiently estimated
as a system. The results of estimation are listed in Table 3.3. Overall fit of the system of
equations was measured using the goodness-of-fit measure proposed by McElroy (1977),
which suggests satisfactory fit (Rz2 = .93). The effects of ISO certification were jointly
significant in the system (χ2∆ (12) = 58.64, p < .01). Traditional R2 values are presented
within Table 3.3 for each of the six equations. For inconsistency, these values ranged from
.55 to .94 and, for change, they ranged from .44 to .92.

3.5.1

Hypothesis Testing

Hypothesis 1 suggests that ISO 9000 certified motor carriers will exhibit greater consistency in
safety performance than non-certified carriers. We previously discussed that our review of the
ISO 9000 literature led us to include a contextual factor, initial performance, as a moderator,
since poor performing organizations appear to benefit most from ISO 9000 certification.
This interaction term is significant in the VM regression (β = −.024, t = −4.792, p <
.001) and in the HOS regression (β = −.007, t = −3.070, p < .005), but not in the UD
regression (β = .031, t = 1.379, p = .179). For carriers with a poor initial VM performance
(one standard deviation above the mean on the VM BASIC in 2012), ISO certification
is associated with a decrease in VM inconsistency (β = −.031, t = −4.344, p < .001).
However, for carriers with strong initial VM performance (one standard deviation below
the mean on the VM BASIC in 2012), ISO certification is associated with an increase in
VM inconsistency (β = .014, t = 2.382, p = .024). Similarly, for carriers with strong initial
HOS performance, ISO certification is associated with a decrease in HOS inconsistency
(β = −.003, −2.640, p = .013). However, an effect of ISO certification on HOS inconsistency
is not present for carriers with strong initial HOS performance (β = .002, t = 1.477, p = .151).
ISO certification was not significantly associated with UD inconsistency; after removing the

98

Table 3.3: Regression results for inconsistency and change in motor carrier safety performance.
Inconsistency
Vehicle
Maintenance

Intercept

Vehicle
Maintenance

Hours of
Service

Unsafe
Driving

S.E.

Est.

S.E.

Est.

S.E.

Est.

S.E.

Est.

S.E.

Est.

S.E.

.194∗∗

.028

.026∗∗

.006

.534∗∗

.109

.739

1.170

−.420∗

.182

.907

.621

−.011
−.019

∗

SEGMENT:SPEC/FB

−.017

∗

UNION_LABOR

−.004

ln(CARRIER_AGE)

−.009

SEGMENT:BULK

Unsafe
Driving

Hours of
Service

Est.
†

SEGMENT:LTL

Change

.006
.009

−.001

.001

−.002

.007

−.004

.006

.001

.007

.001

.002
∗

−.078

.027
†

.039

−.122
−.290

.155
.227

−.002

.031

−.336

.122

†

.178

.049

−.358

†

.035

.001

−.040

.031

.140

.191

.082

.041

.159

.137

.001

−.016

.025

.035

.146

.099∗

.029

.202†

.107

.001

−.049

.030

.014

.185

.058

.038

−.107

.132

.011

−.064

.111

.024

.018

.006

.052

−.024

∗∗

ISO

.068

∗∗

.016

.002

.001

−.053

.035

−.025

.563

−.077

.031

−.012

.186

INITIAL_PERFORMANCE

.026∗∗

.005

.024∗∗

.001

.003

.018

−.242

.179

−.865∗∗

.109

−.485∗

.101

.005

∗∗

.190

∗∗

.080

−.009

.130

4.059

.789

.753

ln(INSPECTIONS)

ISO × INITIAL_PERFORM.

−.024

∗∗

.003

−.003

−.007

∗∗

−.026

∗

.001

.002

−.036

∗∗

.032

2

Notes: † p < .10, ∗ p < .05,

.008
.409

INCONSISTENCY
R

.023

.860
∗∗

.941

.546

p < .01 (two-tailed).

99

.442

3.808

.371
9.625

∗

∗

.921

.664

interaction term from the regression equation, estimated coefficients remained non-significant
for ISO certification (β = −.012, t = −.650, p = .521) and for initial UD performance
(β = .020, t = 1.556, p = .131).
Hypothesis 2 suggests that ISO 9000 certified motor carriers will exhibit greater improvement in safety performance than non-certified carriers. As before, initial performance was
included as a moderating contextual factor. First, we examined interaction terms between
ISO certification and initial performance to determine if moderation was present. These
terms are non-significant in the VM regression (β = .008, t = .041, p = .968) and in the UD
regression (β = −.009, t = −.072, p = .942). After removal of the interaction term the effect
of ISO certification remains non-significant in the VM regression (β = .02, t = .784, p = .440)
and in the UD regression (β = −.026, t = −.333, p = .742). In the HOS regression,
the interaction term is significant but the sign is opposite of our a priori expectations
(β = .37, t = 4.646, p < .01). Next, we examined the relationship between inconsistency and
change. While secondary to our hypothesized direct effect, inconsistency provides a possible
mediated pathway from ISO 9000 certification to change. We found non-significant effects of
inconsistency in the VM regression (β = .409, t = .107, p = .915) and in the UD regression
(β = .789, t = 1.048, p = .304). Again, the coefficient in the HOS regression is significant
but the sign is opposite of our a priori expectations (β = 9.625, t = 2.372, p = .025). We
employ an auxiliary regression test of endogeneity (Davidson and MacKinnon, 1993) to
test for possible bias on the estimate of inconsistency, since bias may be present due to
simultaneity. The coefficients associated with the residual terms from the auxiliary regressions are non-significant within the VM regression (β = −2.963, t = −.365, p = .718), HOS
regression (β = 8.309, t = 1.01, p = .321) and UD regression (β = 0.050, t = .030, p = .976).
Accordingly, we are unable to conclude that bias exists on estimates of inconsistency due to
endogeneity.

100

Table 3.4: Bootstrapped intra-class correlations of annualized safety performance measures.
Performance
Level

Performance
Consistency

BASIC

ICC

95% C.I.

Vehicle Maintenance (VM)

.91 [ .77 , .96 ]

Hours of Service (HOS)
Unsafe Driving (UD)

ICC

95% C.I.

Performance
Change
ICC

95% C.I.

.80 [ .64 , .89 ]

.01

[ .00 , .05 ]

.88 [ .73 , .95 ]

.73 [ .58 , .84 ]

.18

[ .06 , .28 ]

.89 [ .78 , .95 ]

.58 [ .35 , .81 ]

.01

[ .00 , .06 ]

Notes: R = 2, 500.

3.5.2

Post-Hoc Analysis

The results of our analysis lead us to revisit several assumptions central to our theorizing.
First, we made an explicit case that differences exist between carriers in the consistency
of their safety performance and that these differences persist over time. Similarly, our
theorizing of safety change assumed that carriers differ in the amount that they change their
safety performance that these differences also persist over time (i.e., for some carriers, safety
performance undergoes continuous improvement). Lastly, consistent with prior findings
in the literature, we assumed that persistent between-carrier differences exist in the level
of safety performance within the carrier population. We test our assumptions regarding
persistence of between-subject differences by developing bootstrap confidence intervals for
intra-class correlations (ICCs). We follow the procedure given by Ukoumunne et al. (2003).
For each ICC to be estimated, 2, 500 bootstrap data sets were created by random sampling
at the carrier level, with replacement. We estimate ICCs for each dataset within the mixed
effect modeling framework and use the resulting empirical distributions of bootstrapped
estimates to construct percentile confidence intervals. The results of this procedure are
presented in Table 3.4. Higher values of ICC imply that more of the total variation in
the outcome of interest resides between carriers. The data supports our assumption of
persistent differences between carriers in level and consistency, but does not widely support
this assumption regarding change.

101

3.6

Discussion
Ours is the first study that we are aware of to examine longitudinal consistency in a motor

carrier’s safety performance. We do so by conceptualizing motor carrier behavior as a function
of organizational routines and rely on a theory of dynamic routines (Feldman and Pentland,
2003) to hypothesize inter-relationships between a metaroutine (ISO 9000), consistency in
safety performance and change in safety performance. We tested our hypotheses using a
matched sample of certified and non-certified U.S. carriers with carrier-level safety data
from the FMCSA on the VM, HOS and UD BASICs. The results of these tests are discussed
next.
Our first hypothesis linked ISO 9000 certification to consistency in motor carrier safety
performance, suggesting that ISO certification would promote greater consistency. ISO
9000 series quality assurance standards specify basic requirements for organizations to
“consistently provide products and services that enhance customer satisfaction and meet
applicable statutory and regulatory requirements” (ISO, 2016, p. 2). Thus, improved
consistency in organizational outcomes should be a principal benefit from adoption of
ISO 9000 series standards. We tested this proposition directly, examining differences in
consistency of motor carrier safety performance between a sample of carriers who are ISO 9000
certified and a matched sample of carriers who are not. We found support for the hypothesis
for the VM and HOS BASICs, but not for UD. In our sample, ISO 9000 certification was a
time-invariant carrier trait and, accordingly, would best explain between-carrier differences
in BASICs. When the data were disaggregated in our post-hoc analysis to a lower level
(annual safety performance), we found UD consistency to differ more within carriers than
did VM consistency or HOS consistency. Thus, UD consistency appears to be a less stable
phenomenon over time than its VM and HOS counterparts, and would be less predictable
by a stable carrier trait such as ISO 9000 certification.
Arguably, carriers may have more control over the maintenance of vehicles and enforcing
compliance with hours-of-service regulations (i.e., via logbooks) than on-the-road behavior of
drivers. This provides one potential explanation for the lower ICC found on UD and the lack

102

of significant findings regarding its relationship with ISO 9000. In contrast, we found a link
between ISO 9000 certification and VM consistency, however the strength of the relationship
depended heavily on a carrier’s initial VM performance. Unsafe carriers appeared to benefit
the most, with ISO 9000 certification leading to an increase in VM consistency of nearly one
standard deviation. However, for safe carriers, ISO 9000 certification was associated with a
small but significant decline in VM consistency. Lastly, for HOS, we found that ISO 9000
certification of unsafe carriers was associated with a one standard deviation improvement in
HOS consistency. However, no effect was present for safe carriers. Taken together, these
findings suggest that unsafe carriers stand to benefit the most from ISO 9000 certification—at
least when it comes to maintaining a consistent level of safety performance.
Our second hypothesis suggested that ISO certification would promote change in a
carrier’s safety performance. ISO 9000 certification requires organizations to enact procedures
for the continuous improvement of organizational processes, and has been linked to improved
occupational safety outcomes (Levine and Toffel, 2010). However, this hypothesis was not
supported by the data. Upon examining the results of our post-hoc analysis, this lack
of support is not surprising: change in motor carrier safety performance is dominated by
within-subject variation for all three BASICs (Table 3.4). Thus, our sample offers very little
support for the notion that carriers differ in the degree that they continuously improve their
safety performance. On average, our sample of carriers improved across all three BASICs
during our period of analysis (Table 3.2). However, for VM and UD, we are unable to
conclude that any stable between-carriers differences exist in safety change over time. For
HOS these between-carrier differences appear to be minimal. While incentives to improve
may be greatest for unsafe carriers—our analysis suggests that carriers who were initially
unsafe were the most likely to improve—carriers may experience a rapid decline in marginal
returns from the allocation of additional resources toward improving safety. Our lack of
evidence linking ISO 9000 to safety improvement is not unprecedented. Despite the enormity
of the body of research into ISO 9000, findings have been inconclusive regarding its on-going
benefits to level firm outcomes (Singh et al., 2011). We conclude from our assessment of the

103

findings regarding our second hypothesis that changes in motor carrier safety appear to be
transient adjustments or even movements “with the tide,” but our data offer little evidence
to suggest significant differences between carriers in their ability to continuously improve
safety performance.
A primary contribution of our research is the introduction of the concept of performance
consistency to the motor carrier safety literature. To date, this body of literature has largely
focused on modeling levels of safety and changes in safety at the carrier and driver levels.
The concept of consistency in safety performance has gone unstudied despite its sizeable
implications on predictability of a carrier’s safety behavior. We show that carriers differ
in the consistency of their safety performance and that these differences persist over time.
We also offer a factor that contributes to safety consistency, ISO 9000. Su and Linderman
(2016) suggest that the organizational capabilities needed to achieve consistent performance
differ from those needed to achieve a high level of performance. We find that ISO 9000
certification contributes to carrier consistency in the VM and HOS BASICs, conditional
on a carrier’s level of performance, but does not widely influence the presence of level
changes in BASICs. Nor are we able to link ISO 9000 certification to consistency in UD,
which might be influenced by other managerial initiatives such as carrier adoption of in-cab
monitoring technologies. Our findings hold several implications for carrier management and
policymakers. We discuss these implications next.

3.6.1

Managerial and Policy Implications

Our results imply that, for carrier management, ISO 9000 may be an effective mechanism to
improve the consistency of safety performance. We found that these effects were strongest
for unsafe carriers. Inconsistency is related to predictability. Inconsistent carriers may
be more likely to experience extreme safety events such as a large number of accidents
or violations at a given measurement occasion. This holds certain financial implications
for carriers. First, improvements in predictability of a carrier’s safety performance may
reduce their level of assessed risk by insurers. Competitive and economic pressures have

104

increasingly forced insurers to rely on risk assessments when establishing insurance rates
and actuaries now commonly consider a motor carrier’s safety record (among other factors)
when assigning premiums (Hymel et al., 2012). Predictability in safety performance may
also improve a carrier’s ability to attract advantageous financing terms and reduce their own
financial exposure. Factors attributed to motor carrier safety performance are important to
investors, investment brokers and financial analysts who hold, trade or assess securities of
public carriers. Not only might poor safety performance impact a carrier’s reputation and
future business, large carriers routinely risk earnings by self-insuring a substantial portion of
their potential liability (e.g., Werner Enterprises, Inc., 2016).
For shippers, selection of an unsafe carrier can have severe consequences including disrupting the efficient flow of goods through the supply chain. Because of this, shippers and brokers
increasingly consider safety performance within their carrier selection decisions (Cantor
et al., 2013). While we were unable to link ISO 9000 certification to longitudinal change in
safety performance, our evidence suggests that ISO 9000 certification serves as a credible
public signal of quality control in a carrier’s safety performance. Prior research has shown
customer demands to be a leading cause for organizations to pursue ISO 9000 certification.
For shippers entering into high-volume, long-term contracts with carriers, demanding ISO
9000 certification may be viable mechanism to improve predictability and reduce variability
in carrier behavior. This could be especially important to shippers who utilize just-in-time
manufacturing and are especially sensitive to disruptions in the supply chain. Other shippers
holding large inventories may benefit from a more predictable supply chain by reducing
safety stock, since safety stock levels are tied to variability in the fulfillment of inventory
reorders.
The most important implications of our research are for policy makers and regulators,
such as the FMCSA. Motor carrier safety policy is largely based on the assumption that
carriers differ in their level of safety and that these differences are attributable to carrier
management practices. Our findings support this, but also show that motor carriers differ
substantially in the consistency of their safety performance over time. We examine safety data

105

from large carriers over four years and find these between-carrier differences in level of safety
consistency are persistent. Most importantly, we show that carrier management practices
can influence consistency. In our analysis, unsafe carriers exhibited significantly greater
consistency in their VM and HOS safety performance when they were ISO 9000 certified.
Other carrier management practices, such as the adoption of monitoring technologies, may
also lead to consistency improvements for UD.
Unfortunately, current FMCSA policy under the CSA program is silent regarding consistency of carrier safety performance. Yet, the carriers posing the greatest threat to public
safety may be those who exhibit a combination of low average safety performance but
high variability in performance. Not only are these carriers generally unsafe, but their
safety behavior at any time is unpredictable. The FMCSA has been criticized for its inability to consistently identify high-risk carriers under its current methodology (Government
Accountability Office, 2014). Indeed, the CSA methodology has led the FMCSA to disproportionately target their intervention resources toward small carriers with few inspections
and vehicles (Government Accountability Office, 2015). If low-consistency carriers are those
that lack the sort of internal controls needed to self-regulate safety behavior then FMCSA
methodologies for determining risk and interventions should account for carrier inconsistency
in safety performance.

3.6.2

Limitations and Future Research

Our research is limited in that we examined long-term effects of ISO 9000 certification on
motor carrier safety, but did not explore any effects that may occur prior to certification
or short-term changes to carrier safety outcomes which may occur just after certification.
Rather, we utilized a carrier’s initial level of performance on FMCSA BASICs in developing
a matched sample, thus avoiding certain confounds that might be attributable to carrier
performance at the start of our analysis. While our assessment of long-term effects does
not impact our conclusions regarding carrier consistency, it may impact certain conclusions
regarding carrier changes in safety performance that are associated with ISO 9000. Future

106

research might revisit the findings of Naveh and Marcus (2005) and determine when ISO
9000-related effects on level of carrier safety performance occur within the ISO 9000 adoption
and certification process and if these effects are stable over time.
Further, our research examined directionality of carrier change in motor carrier safety
performance and assumed that ISO 9000 certification and variability in safety outcomes
should exclusively result in positive change (safety improvements). Routine dynamics
theorizes that routines evolve when useful variations in routine performance are adopted, or
selected for retention, by managers and other organizational agents (Feldman and Pentland,
2003). However, organizational routines may constitute “competing scripts” (Aroles and
McLean, 2016, p. 542) when incentives and goals differ across levels of organizational hierarchy
or between business units. Additionally, with regard to trial-and-error induced improvements
in safety outcomes, it might be the case that some ideal level of variability (trials) exists.
Instances of too little or too much variability might uniquely constrain a carrier’s ability to
enact beneficial change, due to information underload in the former case and information
overload in the latter (see O’Reilly, 1980, for a discussion). In extreme instances, a high
level of variability may lead to an excess of “noise” in the informational environment and
cloud managerial judgment regarding the utility of any individual routine variation. Since
agents select on routine variations retrospectively, extreme levels of variability may lead to
the selection of “lemon” variations—those that inadvertently result in non-beneficial changes
to routines and organizational outcomes. Given this, we believe that future research into nonlinearities in the relationship between variability in routines and the nature of ensuing change
to those routines may contribute significantly to the understanding of routine dynamics. Such
research might also reveal the presence of a more nuanced relationship between variability
and change in motor carrier safety performance than the linear relationship we theorized.
Lastly, future research should explore alternate causal mechanisms. For instance, our
research advances a perspective of organizational routines and their underlying variability
as a source of change. An alternative theory might suggest the presence of a simultaneous
effect of change on variability and examine these relationships as a system of equations (i.e.,

107

using three-stage least squares estimation). We tested for simultaneity in our analysis but
were unable to conclude that it was present. However, if simultaneous, positive effects exist
between intra-carrier variability and absolute intra-carrier change—a downward spiral—then
it might be critical to uncover steps that carriers could take to identify and temper the
variability-change pattern.

108

REFERENCES

109

REFERENCES
Adler, P. S., Goldoftas, B., and Levine, D. I. (1999). Flexibility versus efficiency? A
case study of model changeovers in the Toyota Production System. Organization Science,
10(1):43–68.
Althauser, R. P. and Rubin, D. (1970). The computerized construction of a matched sample.
American Journal of Sociology, 76(2):325–346.
Anand, G., Ward, P. T., Tatikonda, M. V., and Schilling, D. A. (2009). Dynamic capabilities
through continuous improvement infrastructure. Journal of Operations Management,
27(6):444–461.
Anderson, S. W., Daly, J. D., and Johnson, M. F. (1999). Why firms seek ISO 9000
certification: Regulatory compliance or competitive advantage? Production and Operations
Management, 8(1):28–43.
Aroles, J. and McLean, C. (2016). Rethinking stability and change in the study of organizational routines: Difference and repetition in a newspaper-printing factory. Organization
Science, 27(3):535–550.
Austin, P. C. (2011). An introduction to propensity score methods for reducing the effects
of confounding in observational studies. Multivariate Behavioral Research, 46(3):399–424.
Bardi, E. J., Bagchi, P. K., and Raghunathan, T. (1989). Motor carrier selection in a
deregulated environment. Transportation Journal, 29(1):4–11.
Benner, M. J. and Veloso, F. M. (2008). ISO 9000 practices and financial performance: A
technology coherence perspective. Journal of Operations Management, 26(5):611–629.
Braver, E. R., Preusser, C. W., Preusser, D. F., Baum, H. M., Beilock, R., and Ulmer,
R. (1992). Long hours and fatigue: A survey of tractor-trailer drivers. Journal of Public
Health Policy, 13(3):341–366.
Breusch, T. S. and Pagan, A. R. (1980). The Lagrange multiplier test and its applications
to model specification in econometrics. The Review of Economic Studies, 47(1):239–253.
Britto, R. A., Corsi, T. M., and Grimm, C. M. (2010). The relationship between motor
carrier financial performance and safety performance. Transportation Journal, 49(4):42–51.
Cantor, D. E., Celebi, H., Corsi, T. M., and Grimm, C. M. (2013). Do owner-operators
pose a safety risk on the nation’s highways? Transportation Research Part E: Logistics
and Transportation Review, 59:34–47.
Cantor, D. E., Corsi, T. M., and Grimm, C. M. (2006). Safety technology adoption patterns
in the U.S. motor carrier industry. Transportation Journal, 45(3):20–45.
Cantor, D. E., Corsi, T. M., and Grimm, C. M. (2008). Determinants of motor carrier
safety technology adoption. Transportation Research Part E: Logistics and Transportation
Review, 44(5):932–947.

110

Cantor, D. E., Corsi, T. M., and Grimm, C. M. (2009). Do electronic logbooks contribute
to motor carrier safety performance? Journal of Business Logistics, 30(1):203–222.
Cantor, D. E., Corsi, T. M., and Grimm, C. M. (2017). The impact of new entrants and
the new entrant program on motor carrier safety performance. Transportation Research
Part E: Logistics and Transportation Review, 97:217–227.
Cantor, D. E., Corsi, T. M., Grimm, C. M., and Özpolat, K. (2010). A driver focused truck
crash prediction model. Transportation Research Part E: Logistics and Transportation
Review, 46(5):683–692.
Cantor, D. E., Corsi, T. M., Grimm, C. M., and Singh, P. (2016). Technology, firm size, and
safety: Theory and empirical evidence from the U.S. motor-carrier industry. Transportation
Journal, 55(2):149–167.
Chen, G. X. (2008). Impact of federal compliance reviews of trucking companies in reducing
highway truck crashes. Accident Analysis & Prevention, 40(1):238–245.
Corbett, C. J., Montes-Sancho, M. J., and Kirsch, D. A. (2005). The financial impact of
ISO 9000 certification in the United States: An empirical analysis. Management Science,
51(7):1046–1059.
Corsi, T. M., Grimm, C. M., Cantor, D. E., and Sienicki, D. (2012). Safety performance
differences between unionized and non-union motor carriers. Transportation Research Part
E: Logistics and Transportation Review, 48(4):807–816.
Davidson, R. and MacKinnon, J. G. (1993). Estimation and inference in econometrics.
Oxford University Press, New York, NY.
Deboeck, P. R., Montpetit, M. A., Bergeman, C., and Boker, S. M. (2009). Using derivative
estimates to describe intraindividual variability at multiple time scales. Psychological
Methods, 14(4):367.
Dehejia, R. H. and Wahba, S. (1999). Causal effects in nonexperimental studies: Reevaluating the evaluation of training programs. Journal of the American Statistical Association,
94(448):1053–1062.
Dobie, K. (2005). The core shipper concept: A proactive strategy for motor freight carriers.
Transportation Journal, 44(2):37–53.
Feldman, M. S. (2000). Organizational routines as a source of continuous change. Organization Science, 11(6):611–629.
Feldman, M. S. and Pentland, B. T. (2003). Reconceptualizing organizational routines as a
source of flexibility and change. Administrative Science Quarterly, 48(1):94–118.
Feldman, M. S., Pentland, B. T., D’Adderio, L., and Lazaric, N. (2016). Beyond routines
as things: Introduction to the special issue on Routine Dynamics. Organization Science,
27(3):505–513.

111

FMCSA (2017). Safety Measurement System Methodology: Behavior Analysis and Safety
Improvement Category Prioritization Status. Technical report, Federal Motor Carrier Safety
Administration (FMCSA).
Gersick, C. J. and Hackman, J. R. (1990). Habitual routines in task-performing groups.
Organizational Behavior and Human Decision Processes, 47(1):65–97.
Government Accountability Office (2014). Federal Motor Carrier Safety: Modifying the
Compliance, Safety, Accountability Program Would Improve the Ability to Identify High
Risk Carriers. Technical Report GAO-14-114, United States Government Accountability
Office, Washington, DC.
Government Accountability Office (2015). Federal Motor Carrier Safety: Improvements
to Data-Driven Oversight Could Better Target High Risk Carriers. Technical Report
GAO-15-443T, United States Government Accountability Office, Washington, DC.
Gray, J. V., Anand, G., and Roth, A. V. (2015). The influence of ISO 9000 certification on
process compliance. Production and Operations Management, 24(3):369–382.
Häkkänen, H. and Summala, H. (2001). Fatal traffic accidents among trailer truck drivers
and accident causes as viewed by other truck drivers. Accident Analysis & Prevention,
33(2):187–196.
Hickman, J. S. and Hanowski, R. J. (2011). Use of a video monitoring approach to reduce
at-risk driving behaviors in commercial vehicle operations. Transportation Research Part
F: Traffic Psychology and Behaviour, 14(3):189–198.
Hoffman, L. (2007). Multilevel models for examining individual differences in within-person
variation and covariation over time. Multivariate Behavioral Research, 42(4):609–629.
Huber, G. P. (1991). Organizational learning: The contributing processes and the literatures.
Organization Science, 2(1):88–115.
Hymel, K., Lee, D. B., Pearlman, J., Pritchard, R., and Rainville, L. (2012). Financial Responsibility Requirements for Commercial Motor Vehicles. Technical Report FMCSA-RRA12-045, John A. Volpe Transportation Systems Center, U.S. Department of Transportation.
ISO (2016). Selection and use of the ISO 9000 family of standards. International Organization for Standardization, Geneva, Switzerland.
Knott, A. M. (2001). The dynamic value of hierarchy. Management Science, 47(3):430–448.
Levine, D. I. and Toffel, M. W. (2010). Quality management and job quality: How the
ISO 9001 standard for quality management systems affects employees and employers.
Management Science, 56(6):978–996.
Lewin, A. Y., Massini, S., and Peeters, C. (2011). Microfoundations of internal and external
absorptive capacity routines. Organization Science, 22(1):81–98.

112

Lo, C. K., Wiengarten, F., Humphreys, P., Yeung, A. C., and Cheng, T. (2013). The
impact of contextual factors on the efficacy of ISO 9000 adoption. Journal of Operations
Management, 31(5):229–235.
Martínez-Costa, M., Choi, T. Y., Martínez, J. A., and Martínez-Lorente, A. R. (2009).
ISO 9000/1994, ISO 9001/2000 and TQM: The performance debate revisited. Journal of
Operations Management, 27(6):495–511.
Matta, F. K., Scott, B. A., Colquitt, J. A., Koopman, J., and Passantino, L. G. (2017). Is
consistently unfair better than sporadically fair? An investigation of justice variability and
stress. Academy of Management Journal, 60(2):743–770.
2
McElroy, M. B. (1977). Goodness of fit for seemingly unrelated regressions: Glahn’s ry,x
2
and Hooper’s r . Journal of Econometrics, 6(3):381–387.

Mejza, M. C., Barnard, R. E., Corsi, T. M., and Keane, T. (2003). Driver management
practices of motor carriers with high compliance and safety performance. Transportation
Journal, 42(4):16–29.
Mello, J. E. and Hunt, C. S. (2009). Developing a theoretical framework for research into
driver control practices in the trucking industry. Transportation Journal, 48(4):20–39.
Miller, J. (2017). Discrete time hazard modeling of large motor carriers’ longitudinal safety
performance. Transportation Journal, 56(2):107–139.
Miller, J., Fugate, B., and Golicic, S. (2017a). How organizations respond to information disclosure: Testing alternative longitudinal performance trajectories. Academy of
Management Journal, 60(3):1016–1042.
Miller, J. W., Golicic, S. L., and Fugate, B. S. (2017b). Developing and testing a dynamic
theory of motor carrier safety. Journal of Business Logistics, 28(2):96–114.
Miller, J. W. and Saldanha, J. P. (2016). A new look at the longitudinal relationship
between motor carrier financial performance and safety. Journal of Business Logistics,
37(3):284–306.
Miller, J. W., Saldanha, J. P., Rungtusanatham, M., and Knemeyer, M. (in press, 2017c).
How does driver turnover affect motor carrier safety performance and what can managers
do about it? Journal of Business Logistics.
Monaco, K. and Redmon, B. (2012). Does contracting with owner operators lead to worse
safety outcomes for U.S. motor carriers? Evidence from the Motor Carrier Management
Information System. Accident Analysis & Prevention, 45:654–659.
Morash, E. A. and Clinton, S. R. (1997). The role of transportation capabilities in
international supply chain management. Transportation Journal, 36(3):5–17.
Morrow, P. C. and Crum, M. R. (2004). Antecedents of fatigue, close calls, and crashes
among commercial motor-vehicle drivers. Journal of Safety Research, 35(1):59–69.

113

Moses, L. N. and Savage, I. (1992). The effectiveness of motor carrier safety audits. Accident
Analysis & Prevention, 24(5):479–496.
Naveh, E. and Marcus, A. (2005). Achieving competitive advantage through implementing
a replicable management standard: Installing and using ISO 9000. Journal of Operations
Management, 24(1):1–26.
Naveh, E. and Marcus, A. (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.
Nelson, R. R. and Winter, S. G. (1982). An evolutionary theory of economic change.
Harvard Business School Press, Cambridge, MA.
O’Reilly, C. A. (1980). Individuals and information overload in organizations: Is more
necessarily better? Academy of Management Journal, 23(4):684–696.
Pentland, B. T., Feldman, M. S., Becker, M. C., and Liu, P. (2012). Dynamics of
organizational routines: A generative model. Journal of Management Studies, 49(8):1484–
1508.
Premeaux, S. R. (2002). Motor carrier selection criteria: perceptual differences between
shippers and motor carriers. Transportation Journal, 42(2):28–38.
Robinson, J. L., Thomas, R. W., and Manrodt, K. B. (2013). Food for thought in the
transportation carrier-selection decision. Transportation Journal, 52(2):277–296.
Rosenbaum, P. R. and Rubin, D. B. (1983). The central role of the propensity score in
observational studies for causal effects. Biometrika, 70(1):41–55.
Saldanha, J. P., Shane Hunt, C., and Mello, J. E. (2013). Driver management that drives
carrier performance. Journal of Business Logistics, 34(1):15–32.
Saltzman, G. M. and Belzer, M. H. (2002). The case for strengthened motor carrier hours
of service regulations. Transportation Journal, 41(4):51–71.
Schroeder, R. G., Linderman, K., Liedtke, C., and Choo, A. S. (2008). Six Sigma: Definition
and underlying theory. Journal of Operations Management, 26(4):536–554.
Singh, P. J., Power, D., and Chuong, S. C. (2011). A resource dependence theory perspective
of ISO 9000 in managing organizational environment. Journal of Operations Management,
29(1):49–64.
Sroufe, R. and Curkovic, S. (2008). An examination of ISO 9000:2000 and supply chain
quality assurance. Journal of Operations Management, 26(4):503–520.
Stephenson, P. R. and Willett, R. P. (1969). Consistency: The carrier’s ace in the hole.
Transportation Journal, 8(3):28–33.
Su, H.-C., Dhanorkar, S., and Linderman, K. (2015). A competitive advantage from the
implementation timing of ISO management standards. Journal of Operations Management,
37:31–44.

114

Su, H.-C. and Linderman, K. (2016). An empirical investigation in sustaining high-quality
performance. Decision Sciences, 47(5):787–819.
Swartz, S. M., Douglas, M. A., Roberts, M. D., and Overstreet, R. E. (2017). Leavin’ on
my mind: Influence of safety climate on truck drivers’ job attitudes and intentions to leave.
Transportation Journal, 56(2):184–209.
Terlaak, A. and King, A. A. (2006). The effect of certification with the ISO 9000 Quality
Management Standard: A signaling approach. Journal of Economic Behavior & Organization, 60(4):579–602.
Terziovski, M., Power, D., and Sohal, A. S. (2003). The longitudinal effects of the ISO 9000
certification process on business performance. European Journal of Operational Research,
146(3):580–595.
“Top 100 Truckers” (2013). Inbound Logistics, 33(9):56–65.
“Top 100 Truckers” (2014). Inbound Logistics, 34(9):58–67.
“Top 100 Truckers” (2015). Inbound Logistics, 35(9):56–65.
“Top 100 Truckers” (2016). Inbound Logistics, 36(9):64–73.
Torry, IV, W. F. and Murray, D. (2015). An Analysis of the Operational Costs of Trucking:
2015 Update. Technical report, American Transportation Research Institute, Arlington,
VA.
Ukoumunne, O. C., Davison, A. C., Gulliford, M. C., and Chinn, S. (2003). Non-parametric
bootstrap confidence intervals for the intraclass correlation coefficient. Statistics in Medicine,
22(24):3805–3821.
Veltri, A., Pagell, M., Johnston, D., Tompa, E., Robson, L., Amick III, B. C., Hogg-Johnson,
S., and Macdonald, S. (2013). Understanding safety in the context of business operations:
An exploratory study using case studies. Safety Science, 55:119–134.
Vinodkumar, M. N. and Bhasi, M. (2011). A study on the impact of management system
certification on safety management. Safety Science, 49(3):498–507.
Voss, M. D., Page Jr, T. J., Keller, S. B., and Ozment, J. (2006). Determining important
carrier attributes: A fresh perspective using the theory of reasoned action. Transportation
Journal, 45(3):7–19.
Wangenheim, F. v. and Bayón, T. (2007). Behavioral consequences of overbooking service
capacity. Journal of Marketing, 71(4):36–47.
Werner Enterprises, Inc. (2016).
Form 10-K 2016.
Retrieved from
http://investor.werner.com/financial-information/sec-filings/ [Accessed 17-July-2017].
Wildavsky, A. B. (1988). Searching for safety. Transaction Books, New Brunswick, CA.
Zangwill, W. I. and Kantor, P. B. (1998). Toward a theory of continuous improvement and
the learning curve. Management Science, 44(7):910–920.

115

Zellner, A. (1962). An efficient method of estimating seemingly unrelated regressions and
tests for aggregation bias. Journal of the American Statistical Association, 57(298):348–368.

116