Int. J.

Production Economics 137 (2012) 29–35

Contents lists available at SciVerse ScienceDirect

Int. J. Production Economics

journal homepage: www.elsevier.com/locate/ijpe

Multiple-method analysis of logistics costs

Janne Engblom a, Tomi Solakivi b, Juuso Töyli b,n, Lauri Ojala b
a
Department of Accounting and Finance, Turku School of Economics at the University of Turku, 20014 Turun Yliopisto, Finland
b
Department of Marketing and International Business, Turku School of Economics at the University of Turku, 20014 Turun Yliopisto, Finland

articleinfo abstract

Article history: Logistics costs comprise a significant and relevant proportion of business costs, often exceeding 10 per
Received 24 February 2011 cent of company turnover. This article examines the differences and interdependencies in the self -
Accepted 12 January 2012 reported logistics costs of manufacturing and trading companies operating in Finland. Total logistics
Available online 25 January 2012
costs are taken to consist of six individual components: transport, warehousing, inventory carrying,
Keywords: logistics administration, transport packaging, and indirect costs of logistics. The analysed panel data
Logistics costs covers 241 companies identified from two surveys for the years 2005 and 2008. Logistics costs were
Panel study explored through multiple methods including descriptive analysis, generalised linear mixed models
Mixed model (GLMM), and principal component analysis. The distributions of logistics costs measured as percentages
Random effect
of turnover were skewed and best described by the beta distribution. Time, the number of employees,
turnover, industry, and level of internationalisation were shown to be statistically significant
explanatory variables of logistics costs. Logistics costs tended to be lower in larger companies, although
diseconomies of scale eventually prevail. The analysis also covers changes in costs between 2005 and
2008. In general, the results indicate the need for caution in interpreting changes in logistics costs, and
for simultaneously controlling the effects of background variables.
& 2012 Elsevier B.V. All rights reserved.

1. Introduction and LaLonde (1976) separate warehousing and inventory carrying

Logistics costs comprise a significant and relevant proportion
of business costs: depending on the method applied and the
industry in question, their share of company turnover in devel-
oped economies tends to be at least 10 per cent. However,
definitions of logistics costs are many, and vary considerably.
For the purposes of this article total logistics costs are measured
as a percentage of turnover and comprise six individual compo-
nents: transport, warehousing, inventory carrying, (logistics)
administration, (transport) packaging, and indirect costs of logis-
tics. This classification resembles that used by Naula et al. (2006)
and Töyli et al. (2008).
Heskett et al. (1973) identify transportation, warehousing,
inventory carrying and administration as components of logistics
costs, a classification that has been widely used. Gunasekaran
et al. (2001) include the opportunity cost of capital and storage,
risk costs, and the possible costs of lost sales under inventory
costs, thus combining the elements of inventory carrying and
warehousing. Beamon (1999), on the other hand, distinguishes
between operating costs and inventory costs, whereas Lambert
n
Corresponding author. Tel.: þ358 2 333 9248; fax: þ 358 2 481 4280.
E-mail addresses: janne.engblom@tse.fi (J. Engblom),
tomi.solakivi@tse.fi (T. Solakivi), juuso.toyli@gmail.com (J. Tö yli),
lauri.ojala@tse.fi (L. Ojala).
costs, but include some of the components of warehousing costs,
such as the inventory service costs and storage space costs in
inventory carrying. For more discussion about the definition of
inventory-related costs, see also Stewart (1995a, b), Lee and
Billington (1992), and Levy (1997). In sum, these four cost
components seem to be in general use on the one hand, but on
the other hand authors tend to define the limits of the categories
in unique, inconsistent ways.
The literature identifies a wide range of logistics cost compo-
nents in addition to transport, warehousing, inventory carrying
and logistics administration. For example, Zeng and Rossetti
(2003) add customs, risk and damage as well as handling and
packaging to the list, and Ojala et al. (2007) include ‘‘other
logistics costs’’ to reflect the fact that logistics costs can rarely
be divided and measured accurately. Klaus and Kille (2007) and
Klaus et al. (2010) single out order-entry costs from other
administration costs as a separate component.
The ambiguities are not limited to defining and understanding
logistics cost components. They are also rife in the measurement
approaches and the scope of analysis, which may vary from a
single function or project within a firm to the entire company, a
fraction of a supply chain, or even a demand–supply network.
There appear to be three main measurement approaches. The
first is statistics-based and uses various types of national accounts
or industrial statistics to create an estimate for logistics costs

0925-5273/$ - see front matter & 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2012.01.007
30 J. Engblom et al. / Int. J. Production Economics 137 (2012) 29–35
as a share of GDP. Examples of its implementation include Elger
et al, (2008), Wilson (2009) and Havenga (2010). Bowersox et al.
(2003) and Rodrigues et al. (2005), applied econometric model-
ling based on this kind of data in order to estimate the GDP share
of logistics costs worldwide. The so-called ‘‘Top 100’’ reports on
European logistics markets compiled by Klaus et al. (2010)
estimate the size of logistics expenditure, and ultimately logistics
costs, in Europe based on statistical data on transport volumes,
employment in logistics, and demand for logistics services.
Second, van Damme and van der Zorn (1999) and Baykasoglu
and Kaplanoglu (2008), for example, take an accounting and
activity-based-costing approach. In a similar vein, Pirttilä and
Huiskonen (1996) apply so-called mission costing. This technique
also comprises the use of specific logistics-performance indicators
in various resources-planning software products for enterprises,
such as the SCOR model developed by the Supply Chain Council.
These have enabled accounting-based comparisons across indus-
tries and countries, especially among larger firms using elaborate
ERP systems (see also Pohlen et al., 2009).
The third main measurement approach is survey-based, and
logistics costs are assessed from self-reported company data. The
surveys are usually questionnaire-based and result in estimates of
logistics costs as a percentage of sales. The European Logistics
Association and Kearney (2004, 2009) has used this approach, as
have Naula et al. (2006), and Ojala et al. (2007).
This article examines the differences and the interdependen-
cies in self-reported logistics costs of manufacturing and trading
companies operating in Finland. In order to assess temporal
differences, panel data on 241 companies was collected through
two surveys in the years 2005 and 2008. This survey-based
approach was the only feasible measurement alternative that
would yield sufficiently detailed information about the logistics
cost components in a broad range of companies.
Traditional assessments of logistics costs are based mainly on
calculations and reports of their mean values among the target
population or sample, related to a mixed set of background
variables. The limitations of this method are acknowledged here
(for a more detailed discussion about these limitations see, for
example, Dodd et al., 2006), and a different method based on
generalised linear mixed models (GLMMs) for panel data is
applied. The differences of cost estimates through descriptive
analysis and GLMM model based means are discussed. GLMMs
take into account multivariate, longitudinal, and non-normality
aspects of data, for example, and differ from descriptive analysis
in which dependent variables are analysed alone in one- or two-
dimensional space. GLMMs also allow the simultaneous analysis
of statistical dependencies between logistics costs and several
explanatory variables. Descriptive analysis may give misleading
results because differences between the direct and indirect
relationships among the variables remain unrecognised.
Generalised Linear Models (GLMs) represent a class of fixed-
effects regression models for several types of dependent variables
(i.e., continuous, dichotomous, counts; McCullagh and Nelder,
1989). In GLMs all independent variables (effects) are assumed
fixed which means that there are observations from every level of
that effect. If there is only a sample of levels in the data, the effect
is called random (McCulloch et al., 2001). For the purpose of this
article, logistics cost components are treated as continuous
variables with two observations nested within randomly selected
companies. This necessitates the use of random effects in the
model. GLMs, which assume that all observations are indepen-
dent of each other, are not appropriate for the analysis of these
correlated data structures, longitudinal panel data in particular
(see de Leeuw and Meijer, 2008). Analysis of this kind of multi-
level data requires the addition of random cluster and/or subject
effects into the regression model in order to account for the
correlation. The resulting model, with both fixed and random
effects, is called a linear mixed model (LMM). Given that the
preliminary descriptive analysis indicated non-normality of the
logistics cost components, and the beta distribution turned out to
be more suitable than gamma distribution, the beta distribution
was used in the reported analysis. For this reason, the word
‘generalised’ was added to the method description (GLMM).
2. Research design
2.1. Data and construct operationalisation
The panel data analysed in this article was a sub-sample of all
241 manufacturing and trading companies answering the identical
questions about logistics cost components both in 2006 (data for
2005) (Naula et al., 2006) and 2009 (data for 2008) (Solakivi et al.,
2009), which covered 13.8 and 13.6 per cent, respectively, of all the
responses from both industries (variable INDUST in the statistical
analysis). The logistics cost component data was measured at one-
percentage-point intervals (variable COST in the statistical analysis),
and the respondents were asked to assess the different logistics cost
components as a percentage of the turnover, which according to
Stewart (1995a, b) is a robust base for analysis. The respondents
were free to choose anything between zero and 100 per cent. Given
that the option ‘‘no response’’ was included and was the default
value, zero values were conscious responses implying that the true
cost was somewhere between zero and half a percentage point.
Determination of the company size was based on self-reported
company turnover (TURNOV) and the number of employees
(EMP). The European Commission defines the following limits:
large companies have a turnover of more than 50 million euro,
medium-sized companies between 10 and 50 million euro, small
companies between 2 and 10 million, and micro companies a
maximum of two million euro. A similar classification was used
for categorising the number of employees: 1–9, 10–49, 50–249
and at least 250 employees, respectively.
The companies were also classified in terms of their level of
internationalisation (INT) as domestic, export, and international
based on the percentage share of sales and production from
outside domestic markets. A domestic company was defined as
one generating less than 10 per cent of its sales outside domestic
markets, an export company as one generating over 10 per cent of
its sales from outside, and an international company as also
having production facilities abroad.
2.2. The GLMM model selection
Several generalised linear mixed models were used for the
different logistics cost components. A general matrix form of a
linear mixed model is defined as
Y ¼ l þ e ¼ Xb þ Zb þ e ð1Þ
where Y is a matrix for dependent variables, l the mean vector
and X and b are for fixed effects, similar to regression analysis, for
example. Linear dependency is assumed through a link function,
which in this case is identity, in other words the value of the
dependent variable itself is explained. Z is a known matrix, the
design, or incidence matrix for the random effects, and b is a
vector that contains the coefficients of all the random variables.
Matrix e represents the residuals of the model. Moreover, if
matrices b and e are normally distributed
! .. Σ. ΣΣ
b 0 D0
~N , ð2Þ
e 0 0 R
where D and R are covariance matrices of b and e. This also
 J. Engblom et al. / Int. J. Production Economics 137 (2012) 29–35 31

implies that b and e are independent (Duchateau and Janssen, P

3
COSTij aþ bi þ bk3TURNOVk þb4INDUST
e
1997). Coefficients in b are subject-specific (Molenberghs and ¼ k¼1
þeij ð8Þ
Verbeke 2005). Subjects may contain groups of measurements TURNOVij
from the same observation, or groups of observations. Group P3

means can be estimated using parameter estimates for categorical COSTij a þ bi þb1 YEAR þ bk2EMPk
¼e k¼1 þeij ð9Þ
independent variables or interaction effects. A certain covariance TURNOVij
structure is also assumed for D and R, which contain covariances
of random effects and residuals of the model. COSTij
¼ eaþ bi þb4INDUST þe ij
ð10Þ
In many cases linear dependency is assumed through link TURNOVij
functions (g) other than identity. Here again,
P
3 P
2

Y ¼ lþe ð3Þ COSTij a þ bi þ bk3TURNOVk þb 4INDUST þ bk5INTk

where
gðlÞ ¼ Xb þ Zb
When Y is assumed beta-distributed, logarithmic link function
g(l)¼ ln(l) is usually used, as in this case. If (3) and (4) are
combined with the logarithmic link function the GLMM becomes
Y ¼ g —1 ðXb þZbÞ þ e ¼ eXb þ Zb þ e
Linear mixed models with only statistically significant inde-
pendent variables (p o0.05) were used in the analyses of the
logistics cost components. The logistics cost component means
were therefore estimated based on statistically significant differ-
ences between the YEAR, EMP, TURNOV, INDUST, and INT groups.
Non-significant effects were removed from the starting model
one by one according to the largest p-value. In the starting model,
every independent variable (EMP, TURNOV, INDUST, and INT) was
assumed also to have an interaction with time (YEAR). Only the
models with significant effects are depicted in formulas (6)–(11),
and these were eventually used as final for each logistic cost
component. Beta distribution was used for every cost component
and for total logistics costs. There are several reports of the use of
gamma distribution with a logarithmic link function in cost
analysis (see e.g. Kilian, 2002; Thomson, 2005; and Dodd et al.,
2006). Because the available data was collected in the form of
relative shares (e.g., turnover percentage) of logistics costs rather
than absolute costs, however, the beta distribution was consid-
ered more suitable.
In order to compare the potential distribution candidates
(gamma, beta, and normal distribution), a small number was
added to all the responses, and all candidates were maximum-
likelihood fitted to the adjusted data. It turned out that the beta
distribution was always the best model in accordance with the
Schwarz (1978) criteria. However, the beta and gamma distribu-
tions were consistently close and, when plotted, virtually visually
indistinguishable. In contrast, the normal distribution was always
clearly the inferior model.
Model-based estimates of logistics cost components are dis-
cussed in Section 3.2. Because the number of employees, turn-
over, level of internationalisation, industry, and time are all
categorical variables, a certain number of binary variables were
used in the models (k-1, where k is the number of categories of an
independent variable). The final GLMMs for logistics cost compo-
nents and for total logistics costs thus became (for the subject, i.e.,
company i at time point j) for transport (6), warehousing (7),
inventory carrying (8), administration (9), packaging (10), and
total logistics costs (11):
P3 P2
COSTij a þ bi þ bk3TURNOVk þ bk5INTk
¼e k ¼1 k¼1
þeij
TURNOVij
P
3 Logistics cost components and total logistics costs seemed to
COSTij a þ bi þ bk2EMPk þb 4INDUST
¼e k ¼1 þeij
TURNOVij
¼e k¼1 k¼1 þeij ð11Þ
TURNOVij
The company intercepts were assumed to be random in the
ð4Þ models, and therefore the coefficients bi were included, meaning
that every company is assumed to have its own baseline of
logistics cost components, and that two measurements are con-
ducted for each company. For example, model (6) estimates the
mean transport costs for companies of different size and with
ð5Þ different levels of internationalisation. Correspondingly, model
(10) gives the estimates of packaging costs for two industries
only. No significant independent variables were found for the
indirect costs of logistics.
3. Results
SAS 9.2 was used for the data analysis, and this section reports
the study findings. The focus in the first sub-section is on the
descriptive results at a given point of time, and in the second on
the model-based results. The differences are discussed after the
analysis. The findings related to the variances and linear correla-
tions are reported in the final sub-section.
3.1. Descriptive means
In the 2008 data the one-dimensional distribution of the
numbers of employees (EMP¼ 1–9, 10–49, 50–249, 250–) was
quite right-skewed (42.7%, 22.5%, 15.6%, and 19.2%, respectively),
as was the one-dimensional distribution of turnover (TURNOV¼ h
0–2 M, h2.1–10 M, h10.1–50 M, h50.1 M– : 38.1%, 24.3%, 11.9%,
and 25.7%, respectively). Over half of the companies (57.8%) were
in manufacturing and construction, and 42.2 per cent in trading
(INDUST¼ Manufacturing and construction, Trading). The fourth
candidate as an independent variable, degree of internationalisa-
tion (INT¼ domestic, export, international), was also skewed to
the right in terms of distribution (64.2%, 13.3%, and 22.5%,
respectively).
Table 1 presents the total means and the means by indepen-
dent variable of the cost components and the total logistics costs
for each survey. The first row in each cell refers to 2005 and the
second to 2008. Most of the 2008 values are quite close to those of
2005, although there is a mild rising tendency in some cases. This
finding is interesting in that total logistics costs have been
decreasing over a longer period. European Logistics Association
and Kearney (2009), Havenga et al. (2010), Wilson (2009), and
Choi and Lee (2009), for example, make similar observations.
Of the cost components, the biggest change is the reduction in
administration costs from 2.2 per cent of turnover to 1.5 per cent.
This is surprising in that one would expect administration costs to
be the slowest component to adapt. Transport costs fell from 5.3 to
ð6Þ
5.1 per cent of turnover, whereas warehousing, inventory carrying,
packaging, and indirect costs increased.
ð7Þ differ considerably with regard to the background variables:
Logistics cost components tend to be lower in larger companies
32 J. Engblom et al. / Int. J. Production Economics 137 (2012) 29–35

measured by numbers of employees and turnover, but there is
also some indication that after a certain size diseconomies of scale
may prevail.
The total logistics costs of trading companies decreased
between 2005 and 2008, whereas the tendency was the opposite
for manufacturing companies. This trend could have been attri-
butable to changes in turnover rather than changes in logistics
costs, in that costs were measured as percentages of turnover and
the financial crisis affected the turnover of manufacturing com-
panies more than that of trading companies. The level of private
consumption affecting the trading industry is more persistent
than the demand that the manufacturing industry usually faces.
The results of Honkapohja and Koskela (1999) support that the
macroeconomic indicators of private consumption are less
affected by the economic crisis than the indicators of investments.
3.2. Model-based means
The models presented in formulas (6)–(11) were used to
estimate the means. Fig. 1 summarises the statistically significant
dependencies. Every candidate for an independent variable was
used in models at least once, but no significant independent
variables were found for the indirect costs of logistics.
Table 2 gives the estimates of logistics costs, are referring to
the models first presented in Section 2.2 using the formula
numbering (6)–(11). Turnover (TURNOV) and the level of inter-
nationalisation (INT) were statistically significant independent
variables in terms of transport costs. Company size measured in
turnover had a negative relationship with transport costs, mean-
ing that the larger companies had lower costs than the smaller
ones. In practice the second largest category had the lowest
transport costs. This finding could be attributable to the econo-
mies of scale that often apply to transport, but may also imply
that diseconomies of scale may eventually prevail. In general,
transport costs are considered to be related to the size of the
shipment and the distance transported. According to Bowersox
et al. (1986) and Coyle et al. (2003), the increase in transport costs
is not in direct proportion to distance (or shipment size) due to
economies of scale on the part of the carrier. Tyworth and Zeng
(1998) mention the size of the operation, and describe transport
cost as a non-linear function of the transported amount Q. The
level of internationalisation was also statistically significant in
relation to transport costs in the data analysed here. In particular,
export companies seemed to have higher transport costs than
domestic and international firms. This seems logical in that export
companies (i.e., companies generating over 10 per cent of sales
outside the domestic market) tend to have long and geographi-
cally dispersed delivery chains, but at the same time are unable to
enjoy the scale and the location-specific advantages of interna-
tional companies.
Company size, measured by the number of employees, was
found to be a significant independent variable in terms of ware-
housing costs. Larger companies appeared to have lower ware-
housing costs than smaller companies, and companies in the
highest category in terms of number of employees reported the
lowest cost levels. Interestingly, the estimated mean of ware-
housing costs was lower in manufacturing and construction than
in trading (2.6 and 2.9 per cent, respectively).
The findings with regard to inventory carrying costs are more
straightforward in that there was a clear decreasing trend
compared to size measured in terms of turnover: The larger
companies seemed to have lower costs than the smaller ones.
Trading companies turned out to have higher inventory carrying
costs than manufacturing companies (5.5 and 3.5 per cent,
respectively).
Two statistically significant independent variables, time and
numbers of employees, were used in estimating administration
costs. The findings are quite similar to the descriptive results
presented in Table 1: administration costs decreased from 1.7 per
cent in 2005 to 1.2 per cent in 2008. There was a negative trend
with regard to company size: the percentages ranged from 1.9 in
micro companies to 1.1 in large companies. Estimates of packa-
ging costs were higher in manufacturing and construction than in
trading (1.2 and 0.7 per cent, respectively).
Finally, total logistics costs were estimated in relation to com-
pany turnover, industry and the level of internationalisation. As in
most other cases, there was a U-shaped pattern. Interestingly, and as
in the descriptive analyses, manufacturing companies would seem
to have lower total logistics costs than trading companies. Export
companies have higher total logistics costs than companies operat-
ing either internationally or only domestically.
Changes in logistics costs over time (2005 or 2008) were
statistically significant only in the case of administration costs.
However, time was added to all the models in order to compare
the model-based with the descriptive results. The cost estimates
were now: (2005 vs. 2008) 5.7 and 5.4 for transport; 2.6 and 2.9

Table 1
Descriptive means of logistics costs to turnover (2005 and 2008, %).

Cost Year Total Number of employees Turnover 0h-2 M, h2.1–10 M, Industry manufacturing Level of
mean 1–9, 10–49, 50–249, 250– h10.1–50 M, h50.1 M– and construction, trading internationalisation
Domestic, export,
international

Transport 2005 5.3 6.1,5.5,3.5,4.8 5.8,6.0,3.8,4.7 5.3,5.3 5.0,7.2,5.0

2008 5.1 4.5,5.7,5.0,5.9 4.7,5.1,4.4,6.3 5.8,4.3 4.2,7.7,6.1

Warehousing 2005 3.0 3.5,2.8,3.2,1.9 3.5,3.0,2.7,2.3 2.9,3.3 3.1,3.2,2.9

2008 3.4 3.8,2.9,3.6,2.9 4.0,2.8,3.3,3.3 3.0,4.0 3.5,2.9,3.5

Inventory carrying 2005 5.7 8.4,3.8,3.8,2.9 8.3,3.8,3.8,2.6 3.5,8.7 6.8,3.8,3.6

2008 5.8 7.1,4.3,5.9,4.3 7.3,4.4,4.6,5.2 5.2,6.5 6.1,5.2,5.1

Administration 2005 2.2 2.8,1.5,2.2,1.2 2.7,2.0,1.3,1.9 1.8,2.6 2.3,1.9,2.0

2008 1.5 1.7,1.6,1.3,1.1 1.6,1.6,1.3,1.3 1.5,1.5 1.4,2.3,1.3

Packaging 2005 1.2 1.5,1.2,1.0,0.8 1.6,1.1,0.6,0.8 1.3,1.0 1.2,1.8,0.9

2008 1.4 1.4,1.5,0.8,1.6 1.5,1.2,1.2,1.4 1.7,0.9 1.3,1.8,1.2

Indirect 2005 1.1 1.4,0.7,0.7,0.9 1.2,1.3,0.6,0.8 1.0.1.2 1.2,0.7,0.9

2008 1.3 1.3,1.1,1.3,1.5 1.3,1.0,1.2,1.6 1.3,1.5 1.1,1.7,1.4

Total 2005 18.5 23.7,15.6,14.4,12.5 23.3,17.2,12.8,13.1 15.8,22.2 19.6,18.9,15.2

2008 18.6 20.1,17.2,18.3,16.8 20.6,16.5,15.3,19.0 18.2,19.2 18.0,22.4,17.8
 J. Engblom et al. / Int. J. Production Economics 137 (2012) 29–35 33

Logistics cost components

Industry
Transport
(INDUST)
Warehousing

Point of Inventory carrying

Turnover
time
Administration
(TURNOV)
(2005 or
2008) Packaging
Number of
Indirect employees

(EMP)
Total logistics costs

Level of
Internationalisa
tion

Fig. 1. Statistically significant dependencies between the logistics cost components and the independent variables.

Table 2
Estimates of the mean logistics costs related to turnover by cost components (%).

Cost Model Year Number of employees 1–9, Turnover h0–2 M, h2.1–10 M, Industry manufacturing and Level of internationalisation
2005, 10–49, 50–249, 250– h10.1–50 M, h50.1 M– construction, trading Domestic, export, international
2008

Transport 6 6.3,6.3,4.0,5.9 4.2,7.4,5.4

Warehousing 7 3.1,2.8,2.9,2.3 2.6,2.9
Inventory 8 6.6,4.0,4.0,3.6 3.5,5.5
carrying
Administration 9 1.7,1.2 1.9,1.4,1.6,1.1
Packaging 10 1.2,0.7
Total 11 24.2,18.3,14.7,17.2 16.0,21.1 16.1,23.1,16.6

for warehousing; 4.3 and 4.5 for inventory carrying; 1.7 and 1.2 for
administration; 0.9 and 1.0 for packaging; 0.9 and 1.0 for indirect
costs; and 18.3 and 18.5 for total costs of logistics. The changes in
the individual cost components over time appeared to be some-
what inconsistent, but small in absolute and relative terms except
for administration (decreased by 29 per cent, i.e., 0.5 percentage
points), which was also the only statistically significant change. In
contrast to the descriptive means, there was also a small increase
in total costs, which is in line with the long-term decreasing trend.
Further comparisons between the descriptive and the model-
based means can be made. There seemed to be a particularly large
difference in inventory carrying costs, with estimates of 5.7 and
5.8 per cent for the descriptive means and 4.3 and 4.5 per cent for
the model-based means Furthermore, the changes in estimates
between the two years seemed to vary: five of the seven
descriptive results (warehousing, inventory carrying, packaging,
indirect, and total logistics costs) indicate a rising trend, whereas
also five (warehousing, inventory carrying, packaging, indirect,
and total logistics costs) of the model-based means did so. This
would imply differences in both the level and the development of
cost estimates.
3.3. Linear correlations and variances
The linear correlations of logistics cost components were analysed
by means of principal component analysis, which means finding
orthogonal (i.e., independent) linear combinations of variables that
maximise the R-Square of the component. The MINEIGEN criteria
were adopted, according to which the R-square of a component is,
on average, larger than a single variable in the analysis. One principal
component was found in the 2005 data with an R-square of 39.8 per
cent, and two in the 2008 data with a cumulative R-square of 61.6
per cent. The R-squares derived from principal component analysis
could be interpreted as a proportion of the uncertainty modelled.
Given that the two samples were dependent and the distribution
non-normal, direct testing of the change in the covariance matrixes
was not feasible. Therefore, Box’s test was applied to the logarithmic
non-zero values in order to test the equality of the matrixes. The test
rejected the null hypotheses of equal covariance matrixes logistics
cost components in 2005 and 2008.
Transport costs formed an independent component in the 2008
data, and the other five components were correlated. The correlation
between transport and the other cost components was always less
than 0.33 in the 2005 data, and less than 0.25 in 2008. One principal
component with a low R-square in 2005 and the existence of two
principal components in 2008 could be related to the mutual trade-
offs between the various components. For more information on
trade-offs between transport, inventory and production costs see, for
example, Swenseth and Godfrey (2002); Blumenfeld et al. (1985)
and Jammernegg and Reiner (2007).
On account of the financial crisis of 2008 transport costs were
generally affected by the rapid decline in fuel prices and freight
rates, in addition to company-specific decisions concerning trans-
port vs. inventory types of trade-offs. For example, Havenga (2010)
found that a significant part, some 30 per cent, of transport costs
were susceptible to these kinds of external factors. The majority of
transport in Finland is outsourced to outside service providers (see
Solakivi et al., 2011), which in this competitive market with its
34 J. Engblom et al. / Int. J. Production Economics 137 (2012) 29–35
severe overcapacity are forced to take decreasing fuel prices into
account in their day-to-day pricing decisions.
Standard deviations were estimated in order to quantify the
heterogeneity of the companies and, on the other hand, the
uncertainty associated with modelling costs. The standard devia-
tions were smaller in 2005 than in 2008 for warehousing (from
3.5 to 3.9) and indirect (from 1.8 to 2.0) costs, which implies
higher uncertainty related to these two cost components, and
lower for transport (5.4 to 4.3), inventory carrying (9.7 to 6.7),
administration (3.1 to 1.7) and packaging (2.6 to 1.5) logistics cost
components. The standard deviations of total logistics costs also
declined (16.5 to 13.4).
4. Discussion and conclusions
This article examined the differences and the interdependen-
cies in the logistics cost components and in the total logistics
costs. The analyses covered the associations between time, turn-
over, numbers of employees, industry, and the degree of inter-
nationalisation and both the various components and the total
costs. Related to the company size, measured as the number of
employees or annual turnover, transport costs, warehousing
costs, and total logistics costs showed a U-shaped pattern,
whereas inventory carrying costs decreased and administration
costs were unstable, but the trend was downward. This result
could be partly attributable to economies of scale but also suggest
that after certain size the diseconomies of scale may prevail.
Total logistics costs, and warehousing costs, inventory carrying
and packaging costs turned out to be statistically significantly
higher for trading than for manufacturing companies. This result
is contradictory to those reported in most previous studies both
internationally and in Finland, most of which refer to lower costs
in trading companies. The various cost components were also
estimated to be higher among export companies than in compa-
nies operating either internationally, with production facilities
abroad, or entirely in the domestic market. It may be that export
companies face the challenges of international supply chains
without being able to exploit the scale and location-specific
advantages often associated with international firms.
The analysis also covered changes in costs between 2005 and
2008. The largest relative and only statistically significant change
was in administration costs, other changes being statistically
insignificant. This result is surprising, given that administration
costs could be considered the most inelastic of the components
under study. These findings also contribute to the discussion in
most of the studies reporting annual or biannual changes in costs.
These studies tend to report absolute changes without sufficient
testing of their statistical significance, and without simulta-
neously controlling for other background variables such as size
and level of internationalisation. The findings reported here imply
that changes in time should be interpreted cautiously, and there
should be proper testing of whether or not they are statistically
and practically significant.
The GLMM-based approach and the use of beta distribution
resulted in different estimated levels of logistics cost components
than the descriptive mean values based on the data. The reason
for this difference is the asymmetric nature of costs, which makes
beta distribution a better modelling tool than the normally
distributed measures. Moreover, multivariate modelling provides
simultaneous analysis of logistics costs with several independent
variables and their different interdependencies, thereby control-
ling for the effects and interactions of the independent variables,
which in the case reported here were sometimes large. Descrip-
tive analysis generally does not take into account the multi-
dimensional nature of business. In estimates of logistics costs in
two industries, for example, it ignores the fact that the companies
concerned also differ in size and level of internationalisation. The
implication is that traditional, one-dimensional methods should
be used cautiously.
The correlation structure of logistics cost changed between
2005 and 2008. All the components correlated in 2005, for
example, whereas in 2008 transport costs formed an independent
group. One reason for this may be that the strongest factors
affecting transport costs (the development of fuel prices and
freight rates) in 2008 came from outside the company, and were
not attributable to strategic decisions and their effects on the cost
structure.
Except in the case of warehousing and indirect costs, the
companies appear to have become more homogeneous in terms
of logistics costs, or then the logistics costs could be determined
more by external factors that are not under the control of an
individual company. Evidence of this is in the declining standard
deviations of total costs and transport, inventory carrying, admin-
istration, and packaging costs.
The general implication is that a lot of work remains to be
done in order to enhance understanding of this important yet
elusive facet of logistics – namely the logistics costs. As the
analysis on this article is based on only two points of time, the
findings and conclusions would benefit from validation based on
additional points of time.
References
Baykasoglu, A., Kaplanoglu, V., 2008. Application of activity-based costing to a land
transportation company. International Journal of Production Economics 116
(2), 308–324.
Beamon, B.M., 1999. Measuring supply chain performance. International Journal of
Operations & Production Management 19 (3), 275–292.
Blumenfeld, D.E., Burns, L.D., Diltz, D.J., Daganzo, C.F., 1985. Analyzing trade-offs
between transportation, inventory and production costs on freight networks.
Transportation Research, Part B 19 (5), 361–380.
Bowersox, Calantone D.R., Rodrigues, A., 2003. Estimation of global logistics
expenditures using neural networks. Journal of Business Logistics 24 (2), 1–16.
Bowersox, D.J., Closs, D.J., Helferich, O.K., 1986. Logistical Management, 3rd ed.
MacMillan Publishing Company, New York NY.
Choi, C.-Y., Lee, J.-Y., 2009. Strategic issues for Korean distribution companies’
penetration of the Chinese market. Journal of International Logistics and Trade
7 (2), 83–98.
Coyle, J.J., Bardi, E.J., Langley, C.J., 2003. The Management of Business Logistics:
A Supply Chain Perspective, 7th edn. South-Western, Mason, OH.
de Leeuw, J., Meijer, E., 2008. Introduction to Multilevel Analysis. Springer, New
York.
Dodd, S., Bassi, A., Bodger, K., Williamson, P., 2006. A comparison of multivariable
regression models to analyse cost data. Journal of Evaluation in Clinical
Practice 12 (1), 76–86.
Duchateau, L., Janssen, P., 1997. An example-based tour in linear mixed models. In:
Verbeke, G., Molenberghs., G. (Eds.), in: Linear Mixed Models in Practise.
Springer-Verlag, New York.
Elger, T., Lundqvist, K.-J., and Olander, L.-O., 2008. Svensk Makrologistik, Sam-
mansä ttning och Kostnadsutveckling 1997–2005. Vinnova Rapport VR
2008:13.
European Logistics Association and A.T. Kearney., 2009. Supply Chain Excellence in
der Globalen Wirtschaftskrise.
European Logistics Association and A.T. Kearney., 2004. Differentiation for Perfor-
mance Excellence in Logistics 2004.
Gunasekaran, A., Patel, C., Tirtiroglu, C., 2001. Performance measurement and
metrics in an supply chain environment. International Journal of Operations
and Production Management 21 (1/2), 71–87.
Havenga, J.H., Van Eeden, J., Simpson, Z., 2010. The state of logistics in South
Africa: sustainable improvements or continued exposure to risks, in: The 6th
Annual State of Logistics Survey for South Africa, King. D., Ed .
Havenga, J., 2010. Logistics costs in South Africa – the case for macroeconomic
measurement. South African Journal of Economics 78 (4), 460–478.
Heskett, J.L., Glaskowsky, N.A., Ivie, R.M., 1973. Business Logistics - Physical Distribu-
tion and Materials Management, 2nd edn. Ronald Press Co, New York. 1973.
Honkapohja, S., Koskela, E., 1999. The economic crisis of the 1990s in Finland.
Economic Policy October 1999, 401–436.
Jammernegg, W., Reiner, G., 2007. Performance improvement of supply chain
processes by coordinated inventory and capacity management. International
Journal of Production Economics 108 (1), 183–190.
Kilian, R., Matschinger, H., Lö effler, W., Roick, C., Angermeyer, M.C., 2002.
A comparison of methods to handle skew distributed cost variables in the
 J. Engblom et al. / Int. J. Production Economics 137 (2012) 29–35
analysis of the resource consumption in schizophrenia treatment. Journal of
Mental Health Policy and Economics, 21–31.
Klaus, P., Kille C., 2007. TOP 100 in European Transport and Logistics Services 2nd
edn. Bobingen, Germany.
Klaus, P., Hartmann, E., Kille C., 2010. Die TOP 100 der Logistik. Marktgrö ssen,
Marktsegmente und Marktfuehrer in der Logistikdienstleistungswirtschaft,
Bobingen, Germany.
Lambert, D., LaLonde, B.J., 1976. Inventory carrying costs. Management Accounting
58 (2), 31–35.
Lee, H.L., Billington, C., 1992. Managing supply chain inventory: pitfalls and
opportunities. Sloan Management Review, 65–73.
Levy, D.L., 1997. Lean production in an international supply chain. Sloan Manage-
ment Review, 94–102.
McCullagh, P., Nelder., J.A., 1989. Generalized Linear Models, 2nd edn.
Chapman&Hall/CRC. 1989.
McCulloch, C.E., Searle, S.R., Neuhaus, J.M., 2001. Generalized, Linear and Mixed
Models. Wiley. 2001.
Molenberghs, G., Verbeke, G., 2005. Models for Discrete Longitudinal Data.
Springer. 2005.
Naula, T., Ojala, L., Solakivi, T., 2006. Finland State of Logistics 2006. Ministry of
Transport and Communications Finland.
Ojala, L., Solakivi T., Hä linen H.-M., Lorentz H., Hoffmann T.M. (2007). State of
Logistics in the Baltic Sea Region 2007. Tampere: LogOn Baltic Master Reports,
2007.
Pirttilä , T., Huiskonen, J., 1996. A framework for cost-service analysis In differ-
entiation of logistics services. International Journal of Production Economics
45 (1–3), 131–137.
Pohlen, T., Klammer, T., Cokins, G., 2009. The Handbook of Supply Chain Costing.
CSCMP,Lombard, IL, USA.
Rodrigues, A.M., Bowersox, D.J., Calantone, R.J., 2005. Estimation of global and
national logistics expenditures: 2002 data update. Journal of Business Logistics
26 (2), 1–16.
35
Schwarz, G., 1978. Estimating the Dimension of a Model. Annals of Statistics 6,
461–464.
Solakivi, T., Ojala, L., Töyli, J., Hä linen, H.-M., Lorentz, H., Rantasila, K., Naula, T.,
2009. Finland State of Logistics 2009. Ministry of Transport and Communica-
tions Finland. 2009.
Solakivi, T., Töyli, J., Engblom, J., Ojala, L., 2011. Logistics outsourcing and company
performance of SMEs: evidence from 223 firms operating in Finland. Strategic
Outsourcing An International Journal 4 (2), 131–151.
Stewart, G., 1995a. Supply chain performance benchmarking study reveals keys to
supply chain excellence. Logistics Information Management 8 (2), 38–44.
Stewart, G., 1995b. Supply chain performance benchmarking study. Sloan Manage-
ment Review, 38–44.
Swenseth, S.R., Godfrey, M.R., 2002. Incorporating transportation costs into
inventory replenishment decisions. International Journal of Production Eco-
nomics 77 (2), 113–130.
Thomson, S.G., 2005. How sensitive are cost-effectiveness analyses to choice of
parametric distributions? Medical Decision Making, July/August 2005 25, 416–
423.
Tyworth, J.E., Zeng, A.Z., 1998. Estimating the effects of carrier transit-time
performance on logistics cost and service. Transportation Research Part A:
Policy and Practice 32 (2), 89–97.
Töyli, J., Häkkinen, L., Ojala, L., Naula, T., 2008. Logistics and financial performance:
an analysis of 424 Finnish small and medium-sized enterprises. International
Journal of Physical Distribution & Logistics Management 38 (1), 57–80.
van Damme, D.A., van der Zorn, F.L.A., 1999. Activity based costing and decision
support. International Journal of Logistics Management 10 (1), 71–82.
Wilson, R., 2009. Riding out the recession, CSCMP’s 20th Annual State of Logistics
Report. Council of Supply Chain Management Professionals, 2009.
Zeng, A., Rossetti, C., 2003. Developing a framework for evaluating the logistics
costs in global sourcing processes. International Journal of Physical Distribu-
tion & Logistics Management 33 (9), 785–803.