Decision Support Systems 90 (2016) 12–22

Contents lists available at ScienceDirect

Decision Support Systems

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

The complementarity and substitutability relationships between

information technology and benefits and duration of unemployment
Winston T. Lin a,⁎, Ta-Wei (Daniel) Kao b,1, Chia-Ching Chou a,1, Raj Sharman a,1, N.C. Simpson a,1
a
The State University of New York at Buffalo, School of Management, 326 Jacobs Management Center, Buffalo, New York 14260, USA
b
Department of Management Studies, College of Business, University of Michigan - Dearborn, 19000 Hubbard Drive, Dearborn, MI 48126, United States

a r t i c l e i n f o a b s t r a c t

Article history: This paper aims to analyze the complementarity and substitutability relationships between information
Received 11 August 2015 technology (IT) and unemployment benefits (B) and duration of unemployment (D), two important policy
Received in revised form 1 May 2016 and macroeconomic variables, by using the partial adjustment (PA) valuation (PAV) approaches where the
Accepted 21 June 2016
speeds of PA are fixed and time-varying. The proposed eight research models derived from the PAV approaches
Available online 26 June 2016
based on the theory of PA are fitted into a country-level panel data set covering the period from 1993 to 2008 for a
Keywords:
sample of 12 countries and are estimated by the seemingly unrelated regressions (SUR) and nonlinear SUR
Partial adjustment theory (NLSUR) methods. The findings include: (i) the speeds of PA are dynamic and variable rather than constant, caus-
Constant or time-varying speeds ing the fluctuations of the performance index over time; (ii) among the three production factors used, the tradi-
of partial adjustment tional (non-IT) capital (Kjt) is found to exert the greatest impact upon the desired (true) output and,
The Box–Tidwell transformation consequently, upon the performance index; (iii) B or D or joint B and D has the power to shift the speeds of PA
Unemployment benefits in the countries considered; (iv) the impacts of B, D, and the combination of B and D upon the actual (observed)
Duration of unemployment output and, hence, upon the business value of IT vary from country to country; and (v) all in all, the empirical
Nonlinear seemingly unrelated regressions
evidence strongly suggests that the government policy of B, D, or B and D combined is good for some advanced
developed countries (e.g., US and CN) but may be ineffective for some newly developed nations (e.g., Greece and
Norway) in the presence of IT, and that the policy of joint B and D is good for all four advanced developed coun-
tries and five (Australia, Denmark, Norway, Portugal, and Switzerland) of the eight newly developed economies
in the absence of IT.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction
In the past decades, information technology (IT) has become one of
the most prominent factors that impacts the firm's or country's produc-
tivity and performance. The term IT in its modern sense first appeared in
an article published in the Harvard Business Review in 1958, in which the
authors commented that, “the new technology does not yet have a
single established name. We shall call it information technology.”
Abundant research on the business value of IT has emerged since
then, both at the firm and country levels. Numerous firm level studies
[4,6,11,17,18,27, and others] claim that IT impacts the firm's perfor-
mance positively and the IT productivity paradox has disappeared,
where the productivity paradox is defined as the apparent lack of
productivity improvements that are expected to occur as a result of IT
investment; in other words, it means the lack of a substantial and statis-
tically significant contribution of IT investment to firm output at the
⁎ Corresponding author. Tel.: +1 716 645 3257; fax: +1 716 645 5078.
E-mail addresses: mgtfewtl@buffalo.edu (W.T. Lin), taweikao@umich.edu
(T.-W.(D.) Kao), chiachin@buffalo.edu (C.-C. Chou), rsharman@buffalo.edu (R. Sharman),
nsimpson@buffalo.edu (N.C. Simpson).
1
Tel.: +1 716 645 3257; fax: +1 716 645 5078.
firm level or to aggregate output at the country level [cf. 6,28]. Especial-
ly, [28] provides a conflicting opinion suggesting that the productivity
paradox remains as an unpleasant phenomenon.
At the country level, the debates over the productivity paradox
are even more severe. Some researches [e.g., 12] argue that the pro-
ductivity paradox exists only in developing countries, while some
others [7,23,25,26] demonstrate that the productivity paradox exists in
both developing and developed countries. These conclusions provide im-
portant managerial implications for investors, governments and decision-
makers as IT investment is expected to increase over time [6,18,23,25,
27–30].
During an economic recovery and an expansion, there is a miscon-
ception that benefits and duration of unemployment are harmful to
the country's economic recovery and growth, thereby are not effective
national policies; and they are actually the hurdles of economic
recovery and growth. During an economic downturn and a recession
(e.g. the most recent one taking place in 2008–2009), the misconcep-
tion gets even more popular. Decision makers, managers, and ordinary
people as well are concerned with the increasing unemployment rate,
declining housing prices, scary stock prices, and paltry yields on bank
accounts, etc. The media constantly reports how many Americans
are looking for work and that payrolls, stock prices, and GDP growth

http://dx.doi.org/10.1016/j.dss.2016.06.015
0167-9236/© 2016 Elsevier B.V. All rights reserved.
 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22
point to another contraction, etc. The Congress has passed the bills of
extending jobless benefits. No other U.S. post-war recession than the
most recent one has been so prolonged and severe. Indeed, every
recession causes a lot of pains, job losses, and the financial crisis.
Mired in the stress of recession, people get desperate and overly sensi-
tive to lengthened benefits and duration of unemployment.
Nevertheless, no one seems to question whether unemployment
benefits (denoted by B hereafter) and/or the duration of unemployment
(denoted by D hereafter) may have the potential positive or adverse im-
pacts upon the country's GDP and, hence, the value of IT. Such an impor-
tant economic issue should be taken seriously as a topic of research
inquiry. The present study aims to bridge the gap by pursuing this
important research topic.
In substance, this study concentrates on the complementarity and
substitutability relationships between two well-familiar country
characteristics (B and D) and IT within the framework of the partial ad-
justment methodologies. Here, a complementarity (a substitutability)
relationship between any two variables (say, X1 and X2) is defined as
a phenomenon in which the presence of X2 strengthens (weakens)
the country's performance or the value of X1, measured by a
performance metric (e.g., APRj in Eq. (9) below) and, consequently, X1
and X2 are complementary (substitutable), and vice versa.
More specifically, the objective of this study is to analyze the rela-
tionships among IT, B, and D, and the effects of B, D, or B and D on the
country's performance as measured by the performance index (ratio)
when IT is present. In other words, the objective of this research is
to answer three research questions, namely, (i) Does IT investment
improve or worsen the country's performance as measured by the
average performance ratio called APR in the presence or absence of B
or D or B and D together? (ii) Does B, D or joint B and D have a positive
or adverse (a negative) impact upon the country's performance as
measured by APR in the presence of IT? (iii) Does B, D, or B and D com-
bined strengthen or weaken the country's performance as measured
by APR in the absence of IT? These research questions are critically
important to an economy regardless of whether the economy is in a
depression or an expansion era, amid the fact that IT value and impact
research at the firm level is abundant but, in contrast, knowledge accu-
mulation of IT value and impact at the country level is poor [7,23,29].
To achieve the objective, we apply the partial adjustment valuation
(PAV) approaches with constant and time-varying speeds of partial
adjustment.
The rest of the paper is organized as follows. Section 2 conducts a lit-
erature review; Section 3 describes the theory of partial adjustment
(PA), the PAV approaches, the performance measure, the research
models, and the estimation methods. Section 4 discusses the variables
and the set of country-level panel data used, while Section 5 presents
the various specification and validity tests, the empirical results needed
to answer the three research questions, a detailed discussion of the re-
sults to answer the research questions, implications for policy makers
and for academic researchers, and the links of the study to DSS.
Section 6 concludes the paper with some remarks and managerial
decision-making benefits of this study.
2. Literature review
As mentioned above, numerous researches have investigated the
value of IT and held conflicting opinions on the so-called productivity
paradox phenomenon. The studies can be classified into two major
groups: one group focuses on IT and productivity [6,12,13,30]; and the
other on IT and productive (technical) efficiency [7,20,23–25,27].
These researchers undertake either a firm- or a country-level analysis;
and conflicting conclusions have been reached towards the existence
of the productivity paradox, as we discussed in the preceding Section.
Meanwhile, previous studies apply different methodologies. Some
[6,12,18] of them have applied the traditional regression-analysis ap-
proach; many others have used the data envelopment analysis (DEA)
13
approach; and still other studies [7,17,23,27,28] are based on the
time-varying stochastic production frontier (SPF) approach. Moreover,
Lin at el. [25] and Lin and Kao [26] proposed the one-equation and the
two-equation PAV approach with constant and time-varying speeds of
partial adjustment, respectively, to evaluating and measuring the
value of IT. Though different approaches have been employed, there
are only a few studies [e.g., 7] that concern the factors that influence
the business value of IT.
In a narrow sense, IT value can be measured by technical efficiency
(TE) associated with the SPF approaches or by the performance index
(PI) associated with the PAV approaches. In a broad sense, IT value
can be expanded to include the TE (or PI) change index that is the
ratio of the productivity growth index (or total factor productivity
index) to the technological progress index [24]. Moreover, according
to Steil et al. [39] and Miozzo and Walsh [32], IT induces technological
change and innovation which in turn stimulate economic growth, glob-
alization, and international competitiveness. Having reviewed some se-
lected firm-level studies of IT value published in the 1990–2002 period
and selected industry- and country- level studies of IT returns during
the 1987–2001 period, Dedrick et al. [14] have found the relationship
between IT and productivity and, hence, economic performance, and
concluded that the productivity paradox has been effectively refuted.
In contrast, however, the firm- and country- level studies of [28]
(which used the same set of data as used in [e.g., 6,18,27]), [7,23,28,
etc.] have concluded that the productivity paradox remains in large
U.S. firms and in a country regardless of whether it is a developed or
developing country.
The aim of this study is to investigate the impacts of the country's
two essential macroeconomic unemployment factors, namely, unem-
ployment benefits (B) and the duration of unemployment (D), upon
the business value of IT, and vice versa.
To fulfil the goal, we apply the two-equation PAV approach with
time-varying speeds of adjustment [26] and, simultaneously, compare
to the one-equation PAV approach with constant speeds of adjustment
[25]. The two-equation PAV approach is featured by several merits. First,
like the one-equation PAV approach, the two-equation PAV approach
has a solid theoretical foundation since it is based on the theory of par-
tial adjustment.
Second, the two-equation PAV approach is appropriate for the
present study investigating the influence of B and D upon the business
value of IT. We base this point on the scientific literature and on con-
vincing explanations. (i) On the basis of the scientific literature, the
two-equation PAV approach is featured by several merits as compared
with other methods such as DEA and SPF: (a) DEA is mathematical
programming-oriented, lacks theoretical foundation, and is determinis-
tic and nonparametric; thus, unlike DEA, the PAV approach is stochastic
and parametric as SPF. (b) The two-equation PAV approach with
dynamic and variable speeds of adjustment is particularly suitable for
research topics engaged in PA towards a desired (maximum) target
(e.g., PA towards maximum output in this research, PA towards target
capital structure, etc.), but both DEA and SPF are not. (ii) On the basis
of convincing explanations, the PAV approach with dynamic and
variable speeds of adjustment is constituted by two equations (see
Eqs. (3) and (4) below) to evaluate and measure the performance of
the country and, hence, the value of IT; and the performance is influ-
enced by the desired (maximum) output through the first equation
where IT capital appears, and is also impacted by the changes in the
observed (actual) output via the second equation where B, or D, or
both B and D appear.
Third, the PAV approaches (one-equation and two-equation) and
their accompanied (built-in) performance measures are easy to apply
since, unlike the SPF approach, the PAV approach with either constant
or time-varying speeds of partial adjustment does not involve a half-
normally distributed random inefficiency variable. Simply because the
SPF involves a half-normally distributed random inefficiency variable,
its two-stage estimation procedure is complicated and may fail if
14 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22
the so-called exception condition (wrong skewness) prevails [28];
in contrast, since the PAV approaches do not involve a half-normally
distributed random inefficiency variable, they are immune from the es-
timation difficulty associated with SPF and their estimation processes
are greatly simplified. Moreover, the PAV approaches offer an additional
performance measure, ASAj (see Eq. (5) below).
Unemployment, job search theory, and their companions (notably, B
and D) have drawn much attention of wisdom for a long time. In partic-
ular, job search theory has been studied for decades [e.g. 14,33–35,37].
Nonetheless, both job search and D have become critical issues and
been hotly debated before the start of the global financial crisis in
2008 [1,2,31,33–35,37,38, among others]. With the growing pain of
the unemployment rate, the U.S. Congress and government have ex-
tended unemployment benefits several times, but the action seems to
fail to encourage jobless people to actively seek jobs and the unemploy-
ment rate did not decrease fast enough as expected [10]. This evidence
corresponds to the study of [31] suggesting that the higher unemploy-
ment insurance has a negative effect on the probability of leaving
unemployment.
Since unemployment situations and unemployment compensation
systems differ from country to country, unemployment has been inves-
tigated from different perspectives, including the impacts of unemploy-
ment and related compensation or insurance on D [2,19, et al.], the
effects of B and the business cycle on D [2], the influence of extended
B on D [22], and the impacts of B and D upon forecasting stock returns.
However, virtually all of previous research focuses only on either B
or D. The impacts of B and D upon the business value of IT, and vice
versa, have been overlooked and, consequently, missed from the litera-
ture. This research attempts to close the gap.
3. Theory, research methodology, performance measures, research
models and estimation methods
3.1. The theory of partial adjustment (PA) and the partial adjustment
valuation (PAV) approach with constant speeds of adjustment [25]
The theory of partial adjustment was originally introduced by
Nerlove [36]. It was subsequently applied and extended by numerous
authors [25,26, and others]. The theory states that the observed output
of a decision making unit j (DMUj) at time t is assumed to be equal to a
weighted average of the desired output at current time t and the actual
output at previous time t−1, with weights of δj (the constant speed of
partial adjustment) and 1- δj, respectively. In symbol, the theory is de-
scribed by


Y jt ¼ δ j Y jt
þ 1−δ j Y j;t−1 or; equivalently;
 
Y jt −Y j;t−1 ¼ δ j Y jt −Y j;t−1 ; 0 b δ j b 1; j ¼ 1; 2; …; m and t ¼ 1; 2; …; n
ð1Þ
where Yjt is the actual (observed) output of DMUj (country j in this
research) at time t, Yj , t − 1 is the actual (observed) output of DMUj at
time t− 1, Yjt⁎ is the desired (ideal, maximum) output of DMUj at time
t, and δj is the constant speed of partial adjustment of country j. Accord-
ingly, the speed of adjustment represents the ratio of the change in the
actual output to the change in the desired output in any time period
and, therefore, it must lie between zero and one. In one extreme case,
if the speed of adjustment is equal to zero, then the actual output at
time t equals the actual output at t−1. In another extreme case, if the
speed of adjustment is equal to one, then the actual output is equal to
the desired output; in other words, there is no gap between the actual
output and the desired output and it is said to be fully adjusted. But
these two extreme cases are rare in reality, due primarily to the techno-
logical conditions. Consequently, the value of the speed of adjustment
cannot be negative or greater than 1; hence, “the theory of partial
adjustment” is named.
Eq. (1) can be generalized by adding the traditional error ujt to the
right-hand side, so we have


Y jt −Y j;t−1 ¼ δ j f X jt ; β j −δ j Y j;t−1 þ ujt ; j ¼ 1; …; m and t
¼ 1; …; n ð2Þ
where the desired (true or maximum) output is not directly observable
but can be quantified in practice by a production function, as suggested
by Lin et al. [25], which is consistent with the practice in the SPF
approach; the production function is denoted by f(Xjt; βj) in which
Xjt is a vector of production factors including traditional capital Kjt, tradi-
tional labor Ljt, and IT capital Ijt, thereby, Xjt = (Kjt,Ljt) when Ijt is absent
and Xjt = (Kjt,Ljt, Ijt) when Ijt is present; βj is a vector of unknown coeffi-
cients; and ujt is the traditional uncontrollable random error assumed to
be distributed according to N(0, σu2).
There are several well-known production functions, including the
Cobbs–Douglas (CD) production function and the Box–Cox (BC) and
Box–Tidwell (BT) transformations [3]. Mathematically, as the two trans-
formation parameters involved in BT tend to be equal, BT becomes iden-
tical with BC; as both the two transformation parameters of BT tend to
be zero, the limit of BT is CD; and as the only transformation parameter
of BC tends to be zero, BC collapses to CD [also see 3,23]. Because of the
mathematical reasons just provided and the empirical justifications to
be explained in Subsection 3.5 below, we employ the BT function to
quantify the true (desired, expected, or maximum) output in our
research models.
Eq. (2) combined with the built-in performance measures as stated
below has been referred to as the one-equation PAV approach with con-
stant speeds of adjustment to evaluating and measuring the business
value of IT.
3.2. The PAV approach with time-varying speeds of adjustment [26]
Both the original theory of PA and the one-equation PAV approach
assume constant speeds of PA over time, implying that the speed of
PA is not affected by any unobserved or observed macroeconomic and
microeconomic factors. To achieve the goal of the study requires us to
consider unemployment benefits (Bjt) and the duration of unemploy-
ment (Djt) (i.e., two important macroeconomic policy factors or country
characteristics) using the two-equation PAV approach with time-
varying speeds of PA over time [26]. That is, we apply the generalized
(two-equation) PAV approach given as follows2:


Y jt −Y j;t−1 ¼ δjt f X jt ; β j −δjt Y j;t−1 þ ujt ð3Þ


δjt ¼ g Z jt ; π j ; 0 b δjt b 1; j ¼ 1; …; m and t ¼ 1; …; n ð4Þ
where Zjt = (Bjt, Djt), and πj represents a vector of unknown coefficients.
The estimated average speed of adjustment of country j can be obtained
by
X
ASA j ¼ δ̂ jt =n; j ¼ 1; …; m ð5Þ
t
where δ̂ jt is the estimate of δjt.
The relationships of Bjt, Djt, and IT are analyzed by treating Bjt and Djt
as country characteristic factors that affect the speeds of adjustment in
four cases, each case with two or three factors: Case 1: without Bjt and
Djt, which is the case of constant speeds of adjustment to be used and
served as the base of comparison with other three cases; Case 2: Bjt is
the only source of the dynamic and variable speeds of adjustment;
Case 3: Djt is the only source; and Case 4: both Bjt and Djt enter Eq. (4).
2
There are fundamental differences between the speed of PA associated with the PAV
approaches and TE associated with the time-varying SPF approaches.
 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22 15

3.3. The performance measures Model 1. Case 1, two-factor, constant speeds (δjt = π0j, for all t) without Bjt
and Djt
As proposed by Lin at el. [25], the performance metric can be repre-
!
sented by the product of the speed of adjustment (δjt = δjif constant) Y λjt −1 Y λj;t−1 −1 K θjt −1
and the desired output, that is, PMjt = δjtf(Xjt; βj), which has the same − ¼ π0 j β0 j þ π0 j β1 j
λ λ θ
physical units as the desired output, actual output, and production ! !
Lθjt −1 Y λj;t−1 −1
factors denoted by Xjt. This performance measure is affected by (i) the þ π0 j β2 j −π0 j þ ujt ð10Þ
estimate of the speed of adjustment δjt, (ii) the functional form of θ λ
f(Xjt; βj), (iii) the production factors Xjt in the production function,
(iv) the components of Zjt, (v) the functional form of g(Zjt; πj), and Model 2. Case 1, three-factor, constant speeds (δjt = π0j, for all t) without
(vi) the estimates of unknown coefficients πj and βj. Any change of the Bjt and Djt
six factors would change the value of the performance measure. The ! !
value of the performance measure is non-negative and tends to be Y λjt −1 Y λj;t−1 −1 K θjt −1 Lθjt −1
− ¼ π0 j β0 j þ π j0 β1 j þ π0 j β2 j
zero when the speed of adjustment (fixed or dynamic) approaches to λ λ θ θ
! !
zero. Iθjt −1 Y λj;t−1 −1
The estimated value (PVjt) of PMjt is represented by þ π0 j β3 j −π0 j þ ujt ð11Þ
θ λ


PV jt ¼ δ̂ jt f X jt ; β̂ j Þ ð6Þ
Model 3. Case 2, two-factor, dynamic and variable speeds with Bjt and
δjt = π0j + π1j ln Bjt
! !
where δ̂ jt and β̂ j are the estimates of δjt and βj, respectively. Y λjt −1 Y λj;t−1 −1 K θjt −1 Lθjt −1
Furthermore, in order to compare the performance of DMUs, the − ¼ π0 j β0 j þ π0 j β1 j þ π0 j β2 j
λ λ θ θ
average performance value of j (APVj) can be calculated by !
K θjt −1
þ π1 j β 0 j lnBjt þ π 1 j β 1 j lnBjt
θ
X
n ! !
APV j ¼ PV jt =n; j ¼ 1; 2; …; m ð7Þ Lθjt −1
 Y λj;t−1 −1
þ π1 j β 2 j lnBjt − π0 j þ π 1 j lnBjt þ ujt
t¼1 θ λ
ð12Þ
Both PVjt and APVj are monetary gain as PMjt. Now, PVjt can be
transferred into a performance ratio (PR, or index) by dividing the Model 4. Case 2, three–factor, dynamic and variable speeds with Bjt and
corresponding actual output denoted by YΔ
jt , that is: δjt = π0j + π1j ln Bjt
! !
Y λjt −1 Y λj;t−1 −1 K θjt −1 Lθjt −1
PRjt ¼ PV jt =Y Δjt ; ð8Þ − ¼ π0 j β0 j þ π0 j β1 j þ π 0 j β2 j
λ λ θ θ
! !
I θjt −1 K θjt −1
where YΔ γ⁎ ⁎ λ⁎ ⁎ þ π0 j β3 j þ π 1 j β0 j lnBjt þ π 1 j β1 j lnBjt
jt = lnYjt for CD, =(Yjt − 1)/γ for BC, and =(Yjt − 1)/λ for BT θ θ
[25]. θ
! θ
!
Ljt −1 I jt −1
Then the average performance index (ratio) is given by þ π1 j β 2 j lnBjt þ π1 j β3 j lnBjt
θ θ
!
X
n
 Y λj;t−1 −1
− π0 j þ π 1 j lnBjt þ ujt ð13Þ
APR j ¼ PRjt =n; j ¼ 1; 2; …; m ð9Þ λ
t¼1

Model 5. Case 3, two-factor, dynamic and variable speeds with Djt and
It is clear that PRjt and APRj lies between zero and one. In particular, δjt = π0j + π2j ln Djt
PRjt and APRj facilitate a convenient comparison with the performance ! !
Y λjt −1 Y λj;t−1 −1 K θjt −1 Lθjt −1
measures from other approaches (e.g., DEA and SPF). − ¼ π0 j β0 j þ π0 j β1 j þ π0 j β2 j
λ λ θ θ
θ
!
3.4. Research models K jt −1
þ π2 j β0 j lnDjt þ π 2 j β1 j lnDjt
θ
! !
As mentioned above, the desired output is represented by the pro- Lθjt −1
 Y λj;t−1 −1
þ π 2 j β2 j ln Djt − π0 j þ π2 j lnDjt þ ujt
duction function f(Xjt; βj)which is specified to be the BT transformation θ λ
production function, yielding a total of eight (8) research models. Under ð14Þ
the BT specification, the desired output in Eq. (3) is f ðX jt ; β j Þ ¼ β j0 þ
 K θ −1   Lθ −1   Iθ −1  Model 6. Case 3, three-factor, dynamic and variable speeds with Djt and
β j1 jtθ þ β j2 jt θ þ β j3 jt θ for three factors and f ðX jt ; β j Þ ¼ δjt = π0j + π2j ln Djt
 K θ −1   Lθ −1 
β j0 þ β j1 jtθ þ β j2 jt θ for two factors. The eight research ! ! !
Y λjt −1 Y λj;t−1 −1 K θjt −1 Lθjt −1 I θjt −1
models are grouped into four cases, as stated above. Case 1 consists − ¼ π0 j β0 j þ π0 j β1 j þ π0 j β2 j þ π0 j β3 j
λ λ θ θ θ
of Models 1 and 2 which are based on the PAV approach with constant ! !
K θjt −1 Lθjt −1
speeds of adjustment [25] due to the fact that both Bjt and Djt are ab- þ π2 j β0 j lnDjt þ π 2 j β1 j lnDjt þ π2 j β2 j lnDjt
θ θ
sent, while Models 3–8 in Cases 2, 3, and 4 are based on the PAV ap- ! !
I θjt −1
 Y λj;t−1 −1
proach with dynamic (and variable) speeds of adjustment [26]. The þπ 2 j β 3 j lnDjt − π 0 j þ π 2 j lnDjt þ ujt
θ λ
eight models in four cases described as follows are needed to achieve
our research objective. ð15Þ
16 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22

Model 7. Case 4, two-factor, dynamic and variable speeds with Bjt and Djt 0.60 to 0.83.). Thus, Models 1 and 2 with constant speeds of PA and
and δjt = π0j + π1j ln Bjt + π2j lnDjt Models 3–8 with dynamic and variable speeds of PA were, respectively,
estimated by SUR and NLSUR.
! !
Y λjt −1 Y λj;t−1 −1 K θjt −1 Lθjt −1
− ¼ π 0 j β0 j þ π0 j β1 j þ π0 j β2 j þ π 1 j β 0 j lnBjt
λ λ θ θ 4. Data and sources
! !
K θjt −1 Lθjt −1
þ π 1 j β1 j lnBjt þ π1 j β2 j lnBjt þ π 2 j β0 j lnDjt 4.1. Selection of countries
θ θ
! ! ð16Þ
K θjt −1 Lθjt −1
þ π2 j β1 j lnDjt þ π 2 j β2 j lnDjt This study uses a sample of twelve countries for the period from
θ θ
!
1993 to 2008, yielding a panel data set consisting of 192 observations.
λ

 Y j;t−1 −1 The twelve countries are: Canada (CN), France (FR), the United
− π 0 j þ π 1 j lnBjt þ π 2 j lnDjt þ ujt
λ Kingdom (UK), the United States (US), Australia (AS), Belgium (BG),
Denmark (DM), Greece (GR), Norway (NW), Portugal (PT), Spain (SP),
and Switzerland (SZ). One major reason why we chose these countries
Model 8. Case 4, three-factor, dynamic and variable speeds with Bjt and Djt
is the availability of the data on Bjt (the benefits) and Djt (the duration).
and δjt = π0j + π1j ln Bjt + π2j lnDjt
Another reason is the stage of their economic developments. The chosen
! ! ! countries have the data on Bjt and Djt available with the economic devel-
Y λjt −1 Y λj;t−1 −1 K θjt −1 Lθjt −1 Iθjt −1
− ¼ π0 j β0 j þ π0 j β 1 j þ π0 j β2 j þ π 0 j β3 j opment either in the advanced developed or in the newly developed
λ λ θ θ θ
! ! stage. These countries are divided into two groups according to their in-
K θjt −1 Lθjt −1 come criteria. The first four countries (CN, FR, UK, and US) are classified
þ π1 j β0 j lnBjt þ π 1 j β 1 j ln Bjt þ π1 j β2 j lnBjt
θ θ
as advanced developed countries and the rest of eight countries as
! !
I θjt −1 K θjt −1 newly developed countries. Such classification offers a possible compar-
þ π1 j β 3 j lnBjt þ π2 j β 0 j lnDjt þ π2 j β1 j lnDjt
θ θ ison across countries according to the stages of the economic develop-
! ! ment level.
Lθjt −1 I θjt −1
þ π 2 j β 2 j lnDjt þ π 2 j β 3 j lnDjt The time period that spanned 16 years from 1993 to 2008 was se-
θ θ
! lected simply because the data on B and D were available for the 12

 Y λj;t−1 −1
− π0 j þ π 1 j lnBjt þ π2 j lnDjt þ ujt ð17Þ sample countries when we undertook our empirical work at the begin-
λ
ning of 2012. Therefore, the sample was not chosen to circumvent the
Great Recession of 2008.
The optimal values λ⁎ and θ⁎ of the transformation parameters
λ and θ involved in the above-stated eight models are obtained via 4.2. Variables and data

nm SSE X
 X
 According to the research models, we need these variables to conduct
L max ðλ; θÞ ¼ − ln þ ðλ−1Þ ln Y jt þ ðθ−1Þ ln Y jt
2 nm j;t j;t
empirical work: (i) capital; (ii) labor; (iii) IT investment on computer
software, hardware, and other office equipment; (iv) IS (information
ð18Þ
systems) staffing costs; (v) unemployment benefits; and (vi) the dura-
tion of unemployment. Then, the production factors are (i) ordinary cap-
[23,25–27]. Then, Models 1 to 8 are fitted into the data transformed
ital (Kjt) = capital minus IT capital; (ii) ordinary labor(Ljt) = labor minus
by λ⁎ and θ⁎.
IS staffing costs; and (iii) IT capital (Ijt) = IT investment +3 * IS staffing
costs, where 3 is usually called the multiplier applied to each and every
3.5. Justifications for the deployment of the Box–Tidwell transformation
year of the time period to construct the Ijt series [6,23,25–28]. The macro-
function
economic factors are (i) unemployment benefits (Bjt) which are annual
unemployment compensation costs and (ii) the duration of unemploy-
The mathematical justifications given in Subsection 3.1 above for
ment (Djt) which is the average duration of unemployment measured
deploying BT to quantify the unobserved desired output have been con-
in months. The dependent variable (Yjt) is represented by the value-
firmed empirically. The estimates, λ⁎ and θ⁎, of the two transformation pa-
added created by the input factors and, therefore, is measured by the
rameters contained in the BT function, obtained by Eq. (18), are (0.0935,
gross domestic product (GDP) of a country. All the data used have been
0.0772) and (0.0448, 0.0280) for the two-factor and three-factor models,
transformed into 2000 million U.S. dollars.
respectively. Because the estimates are neither identical nor 0, the BT
transformation production function is certainly more appropriate than
both CD and BC for this research. In addition, BT is known to be the gener- 4.3. Data sources
alization of both BC and CD. Thus, specifying the true output as BT is con-
sistent with the mathematical and empirical justifications. The data needed are collected from several sources. The data on IS
staffing costs and IT investment were secured from Digital Planet
3.6. Estimation methods 2002–2010. The remainder was retrieved from The OECD Statistics,
The United Nations Common Database, and Eurostat Database (http://
While Models 1 and 2 with constant speeds of PA are just-identified, epp.eurostat.ec.europa.eu).
Models 3–8 are obviously over-identified. According to Zellner [41] and
Kmenta [21], when the disturbances in different equations for different 5. Specification and validity tests,3 results, and discussion
countries are highly correlated, the standard seemingly unrelated
regression (SUR, Zellner [41]) and the nonlinear seemingly unrelated 5.1. The Chow test for structural breaks
regression (NLSUR) estimators are more efficient than the ordinary
least-squares (OLS) and the nonlinear least squares (NLS) or the nonlin- To determine whether there exist structural breaks in the internet
ear maximum-likelihood (NML) method. Here, we are exactly facing bubble period (1999, 2000, and 2001) and 2008, we perform the
the situation in which the disturbances of the equations in each model
are highly correlated (with many correlations lying in the range from 3
Details of the tests are available upon request.
 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22
Chow [8] test. We find no significant structural breaks that exist in our
dataset. Consequently, we are not surprised that a dummy included
for 2008 is insignificant.
5.2. Specification tests: the White and Breusch–Pagan tests
Here, we undertake the White [40] test and the Lagrange multiplier
(LM) test proposed by Breusch and Pagan [5] to see if the random errors
in our research models are homoscedastic (H0) or heteroskedastic (H1).
The test results clearly indicate that the random errors in each country's
equation are homoscedastic for all eight models.
5.3. The validity and reliability tests to evaluate model parameter
magnitude differences: the likelihood ratio (LR), Wald, and LM tests
In this subsection, we evaluate the validity and reliability of the re-
search models by conducting the LR, Wald, and LM tests [9,15,16,22,23].
We are interested in testing the hypothesis composed of the null hy-
pothesis (H0): model parameter magnitude differences do not exist
versus the alternative hypothesis (H1): model parameter magnitude
differences do exist. Stated alternatively, we would like to determine
if the differences in the parameters associated with any pair of models
differ significantly from zero (H1) or not (H0). The test results strongly
suggest that H0 should be rejected.
5.4. One more validity and reliability test: the nonparametric Wilcoxon
signed-ranks test
In this subsection, we perform one more validity and reliability test
called the Wilcoxon signed-ranks test [23] to determine whether the
differences in the performance values (PVjt's from Eq. (6)) across coun-
tries and through time, related to the individual APVj 's (Eq. (7)) for any
two different models differ significantly from zero (H1) or not (H0). The
test results lead us to conclude that H1 is not rejected at the 1% or 5%
level. We reach the same conclusion based on PRjt's (Eq. (8)) and
APRj ' s (Eq. (9)).
The conclusions of the various tests suggest that the global
recessions in some years, especially 2007 and 2008, involved in the
sample period did not affect the observed patterns of the results.
5.5. The empirical results needed to answer the three key research questions
Based on the SUR and NLSUR estimates,4 we are able to set up
Tables 1 to 3 needed to answer the three research questions. Specifical-
ly, Table 1 provides the answer to the first research question: Does IT
improve (I) or worsen (W) the country's performance as measured by
APRj in the presence or absence of B, D, or joint B and D? Using Model
2 as the basis, Table 2 is designed to answer the second research ques-
tion: Does B, D, or joint B and D have a positive (P) or an adverse
(A) impact on the country's performance or IT value as measured by
APRj in the presence of IT? Then, using Model 1 as the basis, Table 3 is
presented to answer the third research question: Does B, D, or joint B
and D improve (I, impact favorably) or worsen (W, impact unfavorably)
the country's performance as measured by APRj in the absence of IT? In
short, Tables 1, 2, and 3 employ the values of the APRj to demonstrate
the numerical impacts of B, D, or joint B and D upon the country's per-
formance as measured by APRj in the presence or absence of IT, as well
as to show the numerical impacts of IT upon the country's performance
in the presence or absence of Bjt and Djt, or joint Bjt and Djt.
4
Because a maximum page number is imposed on the length of the paper, the vast
amount of the SUR and NLSUR estimates is available upon request.
17
5.6. Discussions in detail: the answers to the three research questions
This section focuses on answering the three research questions. In
the first place, Table 1 demonstrates the contribution of IT investment
in the country as measured by APRj in the absence or presence of Bjt,
Djt, or both Bjt and Djt. In other words, Table 1 serves to answer the
first research question: When comparing the production process in
which IT capital (Ijt) is present along with non-IT capital (Kjt) and tradi-
tional labor (Ljt) with one where Ijt is absent, does IT improve (I) or
worsen (W) the country's APRj without or under the impacts of Bjt or
Djt or joint Bjt and Djt? There are four situations characterized by the
four cases. Here, we summarize the main points based on the results
given in Table 1 according to the four cases.
(i) Case 1 in the absence of B and D: Model 2 vs. Model 1 indicating
whether IT does improve (I) or worsen (W) the country's APRj in
the absence of Bjt and Djt. In Table 1, after taking into account the
effect of IT on the country's APRj, the average APRj of the whole
sample group (AVGW) has increased 12.56%, meaning that on
average IT investment did contribute to the country's APRj. Inter-
estingly, when comparing other two country groups, we find that
the AVG of advanced developed countries has increased 31.95%,
while the AVG of newly developed economies has increased
just 5.78% which is much smaller than the AVG of the advanced
developed countries and is also below the average APR of the
whole sample. In other words, collectively, IT investment has
exerted a larger positive effect on the APR of advanced developed
countries than that of newly developed countries, suggesting
that collectively, the advanced developed country group is
more productively efficient than the newly developed country
group in the absence of B and D.
Individually, the tale differs while looking at the effect of IT in-
vestment on the APR of individual advanced developed and
newly developed countries. We find that all the countries in the
group of advanced developed countries composed of CN, FR,
UK, and US have enjoyed improvements in APRj (17.83, 40.96,
55.28, and 16.01, respectively). In contrast, among the newly de-
veloped nations, IT has caused APRj to worsen in AS, GR, and SP
(− 1.59, −20.81, and −1.15, respectively), suggesting that AS,
GR, and SP in the group of newly developed countries have suf-
fered from the productivity paradox in the absence of Bjt and Djt.
(ii) Case 2 in the presence of B only: Model 4 vs. Model 3 suggesting
whether IT does improve (I) or worsen (W) the country's APRj in
the presence of B only. In comparing with Case 1, B improves the
AVG of APRj's for the group of advanced developed countries
(from 31.95% to 43.93%), and also improves the AVG of APRj's
for the group of newly developed countries (from 5.78% to
11.38%), implying that collectively, Bjt and IT are complementary
for both the advanced and the newly developed groups. They are
also complementary for the two groups combined because the
AVGW also improves from 12.56% in the absence of B to 20.14%
in the presence of Bjt. Individually, in the presence of Bjt, IT still
has shown a positive impact on the APRj of the four advanced de-
veloped countries, CN, FR, UK, and US (26.60, 29.65, 71.40, and
73.93, respectively); and in the presence of Bjt, the impact of IT
on the APRj of GR (−18.91%) remains negative. Thus, Bjt and IT
in GR is substitutable; but, in other eleven countries, Bjt and IT
are complementary, implying that in the presence of Bjt, the pro-
ductivity paradox remains in GR but disappears in AS and SP
along with other nine countries (i.e., CN, FR, UK, US, BG, DM,
NW, PT, and SZ).
(iii) Case 3 in the presence of D only: Model 6 vs. Model 5 indicating
whether IT does improve (I) or worsen (W) the country's APRj in
the presence of Djt only. The situation in this case is quite differ-
ent from Case 2. In the presence of Djt, IT contributes to the
AVGW slightly higher than Case 1 but much lower than Case 2
18 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22

Table 1
When comparing the production process in which Ijt is present along with Kjt and Ljt with one where Ijt is absent, does IT improve (I) or worsen (W) the country's APRj under the impacts of
Bjt or Djt or joint Bjt and Djt?

Country Case 1: Model 2 vs. Model 1 Case 2: Model 4 vs. Model 3, Case 3: Model 6 vs. Model 5, Case 4: Model 8 vs. Model 7,
without B and D with B with D with B and D

I or W % RK I or W % RK I or W % RK I or W % RK

CN I 17.83 6 I 26.60 5 I 0.37 9 W −1.95 9

FR I 40.96 3 I 29.65 4 I 41.82 2 I 10.23 5
UK I 55.28 1 I 71.40 2 I 25.79 5 I 27.11 3
US I 16.01 7 I 73.93 1 I 13.65 6 W −5.90 11
AVG of advanced developed countries I 31.95 I 43.93 I 18.55 I 3.84
AS W −1.59 11 I 1.68 11 I 1.36 8 I 4.72 6
BG I 41.20 2 I 64.31 3 I 127.97 1 I 236.10 1
DM I 14.10 8 I 21.66 7 I 10.53 7 I 11.50 4
GR W −20.81 12 W −18.91 12 W −24.96 12 W −9.47 12
NW I 28.68 4 I 20.82 8 I 38.52 3 I 4.32 7
PT I 23.77 5 I 25.23 6 I 32.14 4 I 42.41 2
SP W −1.15 10 I 7.60 10 W −1.66 10 I 1.63 8
SZ I 5.25 9 I 9.68 9 W −2.16 11 W −5.08 10
AVG of newly developed countries I 5.78 I 11.38 I 10.76 I 16.79
AVGW I 12.56 I 20.14 I 13.18 I 12.11

(13.18 vs. 12.56 and 20.14); and IT in Case 3 contributes to the
AVG of the advanced developed countries less than in Case 1
and Case 2 (18.55 vs. 31.95 and 43.93). In contrast, IT in Case 3
contributes to the AVG of the newly developed economies great-
er than in Case 1 but slightly less than in Case 2 (10.76 vs. 5.78
and 11.38). In other words, collectively, Djt and IT are comple-
mentary for both the advanced and newly developed groups, so
are for the whole sample. Now, we turn to the effect of IT invest-
ment in the presence of Djt on the individual country's APRj. It ap-
pears that D and IT are substitutable for GR, SP, and SZ, and are
complementary for other nine countries. Thus, in the presence
of D only, the productivity paradox does exist in GR, SP, and SZ.
(iv) Case 4 in the joint presence of B and D: Model 8 vs. Model 7 indi-
cating whether IT does improve (I) or worsen (W) the country's
APRj in the joint presence of Bjt and Djt. Collectively, the AVGW
(12.11) and the AVG of the advanced developed group (3.84)
are the lowest among the four cases; but the AVG of the newly
developed group (16.79) is the highest among the four cases.
This evidence suggests that overall, B and D jointly have exerted
adverse impacts on the value of IT for both the advanced devel-
oped group and the whole sample of countries, but have given fa-
vorable impacts on the value of IT for the newly developed group,
as measured by the APRj, even though the relationship between
IT and joint Bjt and Djt is still complementary for the advanced de-
veloped, newly developed, and whole groups. Furthermore,
comparing Case 4 with Case 1 reveals that in the joint presence
of Bjt and Djt, IT has greatly improved the AVG of the newly devel-
oped group (16.79 vs. 5.78) but has merely slightly worsened the
AVGW of the whole sample of countries (12.11 vs. 12.56) and the
AVG of the advanced developed group (3.84 compared to 31.95).
Individually, Bjt and Djt combined and IT are substitutable for CN, US,
GR, and SZ, and are complementary for other eight countries. Thus, the

Table 2 Table 3
When considering only the production process in which Ijt is used as an input along with When Ijt is not used in (i.e., Ijtis absent from) the production process (i.e., the production
Kjt and Ljt, does Bjt or Djt or joint Bjt and Djt have a positive (P) or an adverse (A) impact on function involves Kjtand Ljtonly), does Bjt or Djt or jointBjt and Djtimprove (I) or worsen
the country's performance or IT value as measured by APRj? (W) the country's performance as measured by APRj ?

Country Model 4 vs. Model 6 vs. Model 8 vs. Country Model 3 vs. Model 5 vs. Model 7 vs.
Model 2 Model 2 Model 2 Model 1 Model 1 Model 1

P or A % RK P or A % RK P or A % RK I or W % RK I or W % RK I or W % RK

CN P 13.98 4 P 42.06 1 P 55.41 3 CN I 6.08 3 I 66.78 1 I 86.75 2

FR P 16.21 2 P 5.84 6 A −4.15 7 FR I 26.36 2 I 5.20 7 I 22.58 3
UK P 15.85 3 A −2.61 9 A −12.45 10 UK I 4.96 4 I 20.23 3 I 6.96 6
US P 10.40 6 P 27.40 3 P 61.62 1 US W −26.36 12 I 30.04 2 I 99.25 1
AVG of P 14.44 P 16.76 P 21.13 AVG of I 4.91 I 29.95 I 53.91
advanced advanced
developed developed
countries countries
AS P 4.23 10 P 3.32 8 P 8.44 5 AS I 0.87 8 I 0.32 8 I 1.90 9
BG P 11.83 5 P 37.25 2 P 55.46 2 BG W −3.90 9 W −15.00 12 W −34.69 11
DM P 8.09 8 P 8.98 5 P 6.28 6 DM I 1.37 6 I 12.49 6 I 8.76 5
GR A −13.84 11 A −9.65 12 A −29.75 12 GR W −15.86 11 W −4.65 9 W −38.54 12
NW A −20.45 12 A −4.68 10 A −13.61 11 NW W −15.28 10 W −11.45 11 I 6.57 7
PT P 33.89 1 P 24.86 4 P 32.98 4 PT I 32.33 1 I 16.95 4 I 15.57 4
SP P 9.91 7 A −6.47 11 A −4.20 8 SP I 0.98 7 W −5.98 10 W −6.82 10
SZ P 5.76 9 P 5.13 7 A −4.80 9 SZ I 1.48 5 I 13.08 5 I 5.55 8
AVG of newly P 4.93 P 5.45 P 4.89 AVG of newly W −0.34 I 0.71 W −5.00
developed developed
countries countries
AVGW P 7.82 P 8.88 P 9.82 AVGW I 1.02 I 8.28 I 10.26
 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22
productivity paradox has appeared in CN, US, GR, and SZ during the time
period studied due to the joint presence of Bjt and Djt.
In the second place, Table 2 intends to answer the second research
question: When considering only the production process in which Ijt is
used as an input along with Kjt and Ljt, does Bjt or Djt or joint Bjt and Djt
have a positive (P or +) or an adverse (A or −) impact on the country's
performance or IT value as measured by APRj? We can pursue the an-
swer to the question by engaging in three aspects, using Model 2 as
the basis.
(i) Model 4 vs. Model 2 denotes the impact of Bjt upon the country's
performance or the value of IT as measured by APRj: Table 2
shows that Bjt has improved the country's performance or IT
value as measured by APRj for the advanced developed countries
group (14.44%), for the newly developed countries group
(4.93%), and for the whole group (7.82%). In other words, Bjt
does enhance the IT contribution by increasing the average APR
by 14.44% for the advanced developed subsample, by 4.93% for
the newly developed subsample, and by 7.82% for the sample
as a whole.
Consider the situation in individual countries. NW is the country
which suffers most from a decrease of APR(−20.45), followed by
GR (− 13.84), because of the payment of unemployment bene-
fits. In contrast, the value of IT in other ten countries was favor-
ably affected by Bjt in varying degrees, suggesting that the
effects of unemployment benefits on IT investment differ from
country to country. In other words, the effects of Bjt on IT value
are mixed (positive for a majority of ten countries and negative
for only two countries). Consequently, Bjt and IT have interacted
to create the phenomenon of complementarity in CN, FR, UK, US,
AS, BG, DM, SP, SZ, and PT and the phenomenon of substitutabil-
ity in GR and NW.
(ii) Model 6 vs. Model 2 denotes the impact of Djt upon the value of IT
as measured by APRj: Again, from Table 2, we can observe the im-
pact of Djt (unemployment duration) upon the average APR rates.
Interestingly, we find that like the previous case of Model 4 vs.
Model 2, Djt has increased the average APRs of the advanced de-
veloped group, the newly developed group, and the whole sam-
ple by 16.76%, 5.45%, and 8.88%, respectively.
In the advanced countries group, UK is the only country which
has suffered from the decrease of its APR(− 2.61); but in the
newly developed countries group, GR, SP, and NW have experi-
enced a decrease in their APRs (− 9.65, − 6.47, and − 4.68, re-
spectively), caused by the effect of D on the business value of
IT. Again, this evidence implies that the effect of the duration of
unemployment upon IT investment differs from country to coun-
try with mixed contribution to the business value of IT; and the
interaction between Djt and IT has created the substitutability re-
lationship in UK, GR, SP, and NW and the complementarity phe-
nomenon in other countries in the sample.
(iii) Model 8 vs. Model 2 denotes the influence of the joint presence of
Bjt and Djt upon the value of IT as measured by APRj: Considering
Bjt and Djt jointly, we again discover the varying effects in differ-
ent countries groups. Specifically, the impacts of the joint pres-
ence of Bjt and Djt upon IT value in the advanced group, the
newly developed group, and the whole sample are all positive
as measured by the average APR (21.13%, 4.89%, and 9.82%, re-
spectively), meaning that the combination impacts of Bjt and Djt
on the average APR rates for the three countries groups are all
positive, strongly suggesting that overall, the joint practice of Bjt
and Djt is an effective national policy as far as the country's eco-
nomic (GDP) growth and IT value are concerned.
Individually, two (CN and US) of the countries in the advanced de-
veloped group and four (AS, BG, DM, and PT) of the newly developed
19
countries have enjoyed positive influences of the joint presence of Bjt
and Djt upon their APRs. However, the rest of the advanced and newly
developed countries have experienced a decline in their APRjs because
of the installments of Bjt and Djt. This empirical evidence suggests that
Bjt and Djt together would cause varying impacts on IT investment
from country to country. To state alternatively, the interaction between
IT and jointBjt and Djt has created the phenomenon of substitutability in
six countries (FR, UK, GR, NW, SP, and SZ) and the relationship of com-
plementarity in the reminder of six nations. US and GR are the biggest
winner and loser, respectively.
In the third and final place, Table 3 is established to answer the
third research question, namely: When IT capital (Ijt) is not used in
(i.e., Ijt is absent from) the production process, that is, the production
function involves Kjt and Ljt only, does Bjt or Djt or joint Bjt and Djt im-
prove (I) or worsen (W) the country's performance as measured by
APRj? Once more, we can discuss three aspects by treating Model 1
as the basis in order to search for the answer to the third research
question.
(i) Model 3 vs. Model 1 measures the impact of Bjt upon the
country's performance as measured by APRj in the absence of
IT: We can see from Table 3 that Bjt exerts varying degrees of im-
pacts on the average APR for the whole sample (1.02%), for the
group of advanced developed countries (4.91%), and for the
group of newly developed countries (− 0.34%), in the absence
of IT. Thus, Bjt has worsened the performance of the newly devel-
oped group, but has improved the performance of the advanced
developed group. Overall, Bjt, ordinary capital, and ordinary labor
interact to create the phenomenon of complementarity (substi-
tutability) in the advanced (newly) developed group.
For individual advanced developed countries, US is the only
country which suffers from the decrease of APR (− 26.36%),
while among the newly developed countries, the APRs of BG,
GR, and NW have declined (− 3.90, − 15.86, and − 15.28, re-
spectively) because of the payment of B. This empirical evidence
suggests that a significant majority of countries (eight of twelve)
have benefited with the payments of B in the absence of IT.
(ii) Model 5 vs. Model 1 reveals the impact of Djt upon the country's
performance as measured by APRj in the absence of IT: Again,
from Table 3, we can observe the mixed effects of unemployment
duration on the average APRs of different groups. The perfor-
mance of the group of advanced developed countries has been
improved by the duration of unemployment for a significant
percentage (29.95%); but the performance of the group of
newly developed countries has been enhanced by the duration
of unemployment for a negligible performance (0.71%).
All individual countries in the advanced developed countries
group have benefited from the policy of the duration of
unemployment (D), as measured by the increases of their APRjs
(as high as 66.78% for CN, 30.04% for US, and 20.23% for UK). In
contrast, among the newly developed countries studied, the per-
formance of BG, GR, NW, and SP has got worsened because of the
practice of D (ranging from −15.00% for BG to −4.65% for GR).
Accordingly, the results suggest that the effect of D on the
country's performance in the absence of IT varies from country
to country and that D is likely a good national policy for all
advanced developed countries in the absence of IT.
(iii) Model 7 vs. Model 1 discloses the impact of joint Bjt and Djt
upon the country's performance as measured by APRj in the ab-
sence of IT: The results presented in Table 3 indicate that the
policy practice of joint Bjt and Djt has positively impacted the
average APRsof the whole sample (10.26%) and the advanced
developed countries group (53.91%); but it has negatively
affected the average APR of the newly developed group
(− 5.00%). It is worth mentioning that comparing with Bjt or
Djt, the policy of joint Bjt and Djt is the most powerful policy
20 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22

Table 4
An outline of some other interesting conclusions drawn from the empirical estimates.

Issue Finding

Explanatory power (R2)
Statistical significance
Variation of PA speeds
The PP caused by B or D
Inputs contributions to desired outputs
Are B, D, or B and D the cause of the PP?
Which policy, B or D or joint B and D, is most effective
based on its influence on APRj?
The dynamic patterns of the PA speeds under the
impacts of B, D, or joint B and D
The contribution of B, D, or B and D to PA speeds
In the 2-factor models, R2s range from 0.112 (NW) to 0.956 (AS), whereas in the 3-factor models, R2s range from 0.28
(NW) to 0.959 (AS)
Virtually all the speed and coefficient estimates are significant at the 1% or 5% level
Most of the speeds of PA (δjt′s) in the 3-factor models are higher than their counterparts of the 2-factor models,
implying that IT has the power to accelerate the PA speed in all the sample countries
In some countries (e.g., US, AS, and SP), B or D causes the PP but joint B and D makes the PP disappear
In the 2-factor models, K has greater impact than L on the desired output, while in the 3-factor models, though the
influence of K is diluted by IT, K remains to have the greatest impact upon the desired output and performance indexa
Judging from ASAj and APRj, the impacts of B, D, and Joint B and D on IT value vary from country to country, with
positive impacts for most countries and negative effects for only a few countries in the sample
Judging from APRj, the joint B and D is the most effective for advanced ad newly developed countries in the presence
of IT; and the joint B and D remains the most effective for newly developed nations in the absence of IT
In CN and BG, an increasing but fluctuating pattern over time prevails; in US and PT, the fluctuating patterns from
different models converge at the year of 2000 and diverge after 2000; in UK and SP, the speeds fluctuate over time in
the absence of IT, but its pattern is stabilized in the presence of IT; finally, the presence of IT accelerates fluctuations of
PA speeds in at least two-third of the countries studied
The joint presence of B and D with IT contributes to PA speeds more than B or D does in a significant majority of
countries

Notes: PA stands for partial adjustment, B for unemployment benefits, D for the duration of unemployment, and PP for the IT production paradox.
a
B and D are a labor problem of the country's labor market. The problem signals that there are displaced, unemployed, and laid-off employees who are discouraged workers. Conse-
quently, the problem in L makes, via the interaction between the two equations of the PAV approach, K outperform L and I; thereby K plays the role as the greatest contributor of the output
and performance index. The conclusion under B and D is obviously contrary to the finding of [27, among others] that L exerts the greatest impact upon the output in the absence of B and D.

instrument to improve the performance of advanced devel-
oped countries in the absence of IT.
Individually, the policy of joint Bjt and Djt appears to be most power-
ful for US as reflected by the largest increase of 99.25% in its APR, follow-
ed by CN (86.75%), FR (22.58%), and UK (6.96%). It is also an effective
policy for PT, DM, and SZ. However, the policy seems to be ineffective
for several newly developed countries (e.g., GR, BG, and SP) which
have suffered from the implementation of the joint Bjt and Djt policy.
After all, the joint Bjt and Djt policy may not be a bad one.
To summarize, in the absence of IT, the impacts of B, D, and joint B
and D upon the country's performance as measured by APRj are
mixed; and the public policies of D and joint B and D are likely to be
more appealing for advanced developed countries than for newly devel-
oped economies. Prudent exercises of these policies are required.
In addition, Table 4 is produced to outline some other interesting
conclusions drawn from the vast amount of the SUR and NLSUR esti-
mates. The table is self-explanatory.
5.7. More on the dynamic patterns of the speeds of partial adjustment
Fig. 1 shows explicitly the dynamic patterns of the speed of PA
with the impacts of B, D, or joint B and D, i.e., based on the eight re-
search models, for each country.5 We can group them with similar
patterns. For example, CN and BG exhibit a pattern of increasing
fluctuations over time based on the eight models, meaning that at
the very beginning when B, D or joint B and D was implemented,
the speeds of PA did not change; but they started fluctuating more
and more after the year of 2000; in other words, the government
policy of B, D, or joint B and D has played a greater role of shifting
the speeds of PA after 2000.
Another pattern is found in US and PT where the fluctuating patterns
of different models converge at the year of 2000, and then diverge after
2000, caused by B, D, or joint B and D. A third pattern is discovered for
UK and SP where the speeds of PA fluctuate over time in the absence
of IT, but the pattern is stabilized in the presence of IT. Moreover, for
FR, and SP, a very stable pattern over time is found. Most countries
have a flexion point around 2000 but reveal quite varied patterns.
5
Fig. 1 is incomplete to save space.
Furthermore, Fig. 1 clearly indicates that in most countries, the
speeds of PA from the three-factor models are more nonlinear than
their counterparts from the two-factor models, regardless of whether
B, D, or joint B and D is implemented, with AS, GR, SP, and SZ being
the exceptions. The implication is that the presence of IT accelerates
the speeds of PA in at least two-third of the countries considered.
5.8. Implications
Implications for policy makers and academic researchers differ.
These include, but are not limited to, the following.
(i) Implications for policy makers:
(a) The PA theory, PAV approaches, research models, built-in perfor-
mance index, estimation methods, panel dataset, and empirical
evidence represent a system that provides enhanced decision support.
(b) The results imply that B, D, and joint B and D can be three effective
national policies; as such, prudent choice among them is of crucial im-
portance in government decision-making processes. (c) The findings
that B, D, and B and D combined have the power to slow or accelerate
the speeds of PA and, hence, affect the performance index points to
the need to exercise these national policies effectively and carefully in
order for them to exert positive impact upon the output and perfor-
mance index. (d) The results also imply that the choice among the
three national policies differ in the presence and absence of IT; there-
fore, the IT investment becomes a decision-making component needed
to make B, D, or B and D combined effective. (e) In the presence of B, D,
or joint B and D, the allocation among Kjt, Ljt, and Ijt, is relevant in view of
the finding that Kjt exerts the greatest impact upon the output and
performance index.
(ii) Implications for academic researchers:
(a) The importance of choosing an appropriate production frontier is
implied by the two-equation PAV approach, e.g., between BT if the sub-
stitution and complement relations of inputs are neglected (in this
study) and CES (constant elasticity of substitution [7]) if the relations
among the three inputs can't be ignored (an extension to this study);
then in the latter situation, an in-depth analysis of the complementarity
and substitutability relationships created by IT and unemployment
factors becomes even more complicated but is more interesting. (b) A
 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22
Fig. 1. Patterns of the constant and dynamic speeds of adjustment.
second implication for researchers is that the present study can be
moved a step forward by analyzing the relationships among perfor-
mance index change, technological progress, and productivity growth
(measured by the total factor productivity) in the presence or absence
of B, D, or joint B and D. (c) Broad firm-level applications of the PAV
approach in which the speeds of PA are dynamic are possible [35]. For
example, the chief executive officer (CEO) pay can be investigated by
using the decision support system as applied in this research. In other
words, the PAV approaches (with constant and dynamic PA speeds) can
be applied to bridge a gap created by such issues as: what are the PA
speeds of actual (observed) towards desired (maximum) compensation?
Whether or not the PA speeds are constant or dynamic over time? And
whether or not the firm's performance is positively affected by the PA
speeds and desired compensation so that the actual compensation
is justified. Of course, in applying the two-equation PAV approach, B
and D will be replaced by other variables appropriate for the CEO pay
problem. These are ongoing research projects of the first author with an
advanced Ph.D. candidate.
5.9. The links of this study to DSS
As stated above, the sound theory, novel methodologies with built-
in performance metrics and research models, estimation methods,
panel dataset, and empirical findings constitute an innovative decision
support system that has a wide range of practical applications [25,26]
with the firm, industry, or country as the level of study. It is capable of
enhancing decision making and providing decision support for policy
decision makers at the country level and managerial decision makers
at the firm and industry levels as well. As such, the links of this research
to DSS are strong, rendering this research work undoubtedly suitable for
the general DSS readership.
6. Concluding remarks and managerial decision-making benefits of
this research
This country-level study investigated whether the (inter)relation-
ships among information technology (IT), unemployment benefits (B),
and unemployment duration (D) are complementary or substitutable,
as judged by the speed of partial adjustment (PA) and the performance
ratio (index) which is the desired (maximum) output multiplied by the
speed of PA.
21
The methodology used is known as the PA valuation (PAV) ap-
proaches with constant and time-varying speeds of partial adjustment.
The PAV approaches are noted for having strong theoretical foundation,
ease of applications, and built-in performance measures. They are
especially appropriate for this research.
The objective of the study was transformed into three major
research questions and the answers to the questions were pursued.
The empirical evidence related to the answers to the research questions
reveals that the (inter)relationships among IT, B, and D are a complex
issue and differ from country to country and from different levels of
economic development.
The complex (inter)relationships imply that prudent exercises of na-
tional policies as to the implements of IT, B, and D are critical to
promoting recovery and boosting economic growth. The evidence also
implies that the policies related to IT, B, and D are more effective for
the advanced than for the newly developed countries and that the
practice of joint B and D is more powerful than B or D standing alone
in influencing the business value of IT and curing the disease of the pro-
ductivity paradox and, hence, the country's performance. Thus, there
are important implications for policy makers and academic researchers;
and the links of this work to DSS are strong.
It is noted that as a byproduct, the data do not support the notion
that the productivity paradox exists only in developing countries and
reject the widely shared view that the productivity paradox has
disappeared since the early 1990s. The byproduct is consistent with
the conclusions reached by [7,23].
Finally, we point out that, though this research takes the country as
the level of study, there are managerial decision-making benefits of this
research for managers and firms. The following are just some of them.
First, this research offers a comprehensive application of the above-
stated decision support system (Subsection 5.9). The system consists
of a series of instruments of enhanced decision making and is well appli-
cable for firm-level studies. The third extension proposed to investigate
the firm's CEO pay (Section 5.8) is a witness.
Second, it fallows that the outcomes of this work yield in the way of
aid to managerial decision making. For example, in the economic envi-
ronment with B and D, one outcome indicates that ordinary capital or
non-IT capital (Kjt) is the greatest contributor of the output and
performance index, followed by ordinary labor (Ljt) and IT capital (Ijt).
We believe this outcome equally applies to the firm under the same eco-
nomic environment. Thus, the outcome is an aid to managerial decision
22 W.T. Lin et al. / Decision Support Systems 90 (2016) 12–22
making of the resources allocation among Kjt, Ljt, and Ijt. Take another
example. In the economy characterized by B and D, IT and B and D com-
bined were found to be substitutable (see US and CN with Case 4 in
Table 1), meaning the existence of the IT productivity paradox. Under
this situation, IT investment may be considered crucial for the firm's
survival, competitiveness and growth, the decision support system
used in the research becomes an aid to the managerial decision making
of IT investment.
Acknowledgements
For constructive and useful comments and suggestions that consid-
erably improved the paper, the authors are greatly grateful to the
Editor-in-Chief, Professor James R. Marsden, and the anonymous
referees of this journal. The first author thanks the Research Committee
of the School of Management at The University at Buffalo, The State
University of New York, for financial support. Y.B. Zhao and Ta-Wei Kao
served in the capacity of research assistants for two years and
contributed to data collection and empirical estimation. The authors
bear the sole responsibility for any errors that remain.
References
[1] J.M. Barron, Search in the labor market and the duration of unemployment: some
empirical evidence, The American Economic Review 65 (1975) 934–942.
[2] O. Bover, M. Arellano, S. Bentolila, Unemployment duration, benefit and the business
cycle, The Economic Journal 112 (2002) 223–265.
[3] G.E.P. Box, P.W. Tidwell, Transformation of the independent variables, Technometrics
4 (1962) 531–550.
[4] T.F. Bresnahan, Measuring the spillovers from technical advance: mainframe com-
puters in financial services, The American Economic Review 76 (1986) 742–755.
[5] T.S. Breusch, A.R. Pagan, The Lagrange multiplier test and its applications to model
specification in econometrics, The Review of Economic Studies 47 (1980) 239–253.
[6] E. Brynjolfsson, L. Hitt, Paradox lost? Firm-level evidence on the returns to
information systems spending, Management Science 42 (1996) 541–558.
[7] Y.H. Chen, W.T. Lin, Analyzing the relationships between information technology,
inputs substitution and national characteristics based on CES stochastic frontier pro-
duction models, International Journal of Production Economics 120 (2009)
552–569.
[8] G.C. Chow, Tests of equality between sets of coefficients in two linear regressions,
Econometrica 28 (1960) 591–605.
[9] T. Coelli, D.S.P. Rao, C.J. O'Donnell, G.E. Battese, An Introduction to Efficiency and
Productivity Analysis, second ed. Springer Science & Business Media, New York,
2005.
[10] P. Coy, A.R. Hunt, Jobless benefits: corrupting or compassionate? Bloomberg
Businessweek Dec 13–192010.
[11] J. Dedrick, V. Gurbaxani, K.L. Kraemer, Information technology and economic perfor-
mance: a critical review of the empirical evidence, ACM Computing Surveys 35
(2003) 1–28.
[12] S. Dewan, K.L. Kraemer, Information technology and productivity: evidence from
country-level data, Management Science 46 (2000) 548–562.
[13] S. Dewan, C.K. Min, The substitution of information technology for other factors of
production: a firm level analysis, Management Science 43 (1997) 1660–1675.
[14] P.A. Diamond, Unemployment, vacancies, wages, American Economic Review 101
(2011) 1045–1072.
[15] L.G. Godfrey, Misspecification Tests in Econometrics: the Lagrange Multiplier
Principle and Other Approaches, Cambridge University Press, New York, NY, 1988.
[16] W.H. Greene, Econometric Analysis, seventh ed. Pearson Education, Inc., Upper
Saddle River, NJ, 2012.
[17] S. Grosskopf, The Measurement of Productive Efficiency: Techniques and Applica-
tions, Oxford University Press, New York, NY, 1993 (Chapter 4).
[18] L.M. Hitt, E. Brynjolfsson, Productivity, business profitability, and consumer surplus:
three different measures of information technology value, MIS Quarterly 20 (1996)
121–142.
[19] J. Hunt, The effect of unemployment compensation on unemployment duration in
Germany, Journal of Labor Economics 13 (1995) 88–120.
[20] H. Kaynak, J.A. Pagan, Just-in-time purchasing and technical efficiency in the US
manufacturing sector, International Journal of Production Research 41 (2003) 1–14.
[21] J. Kmenta, Elements of Econometrics, second ed. The University of Michigan Press,
Ann Arbor, MI, 1997.
[22] R. Lalive, How do extended benefits affect unemployment duration? A regression
discontinuity approach, Journal of Econometrics 142 (2008) 785–806.
[23] W.T. Lin, The business value of information technology as measured by technical
efficiency: evidence from country-level data, Decision Support Systems 46 (2009)
865–874.
[24] W.T. Lin, Assessing the Impacts of the Integration of the ICT Investments of Taiwan
and China upon Economic Growth in Taiwan, in: P.C.Y. Chou (Ed.), Economic Inte-
gration across the Taiwan Strait: Global Perspectives, Edward Elgar Publishing,
Northampton, MA 2013, pp. 56–80.
[25] W.T. Lin, C.H. Chuang, J.H. Choi, A partial adjustment approach to evaluating and
measuring the business value of information technology, International Journal of
Production Economics 127 (2010) 158–172.
[26] W.T. Lin, T.W. Kao, The partial adjustment valuation approach with dynamic and
variable speeds of adjustment to evaluating and measuring the business value of in-
formation technology, European Journal of Operational Research 238 (2014)
208–220.
[27] W.T. Lin, B.B.M. Shao, Relative sizes of information technology investments and pro-
ductive efficiency: their linkage and empirical evidence, Journal of the Association
for Information Systems 1 (2000) 1–35.
[28] W.T. Lin, B.B.M. Shao, The business value of information technology and inputs sub-
stitution: the productivity paradox revisited, Decision Support Systems 42 (2006)
493–507.
[29] N. Melville, K. Kraemer, V. Gurbaxani, Review: information technology and organi-
zational performance: an integrated model of IT business value, MIS Quarterly 28
(2004) 283–322.
[30] N. Melville, V. Gurbaxani, K. Kraemer, The productivity impact of information
technology across competitive regimes: the role of industry concentration and
dynamism, Decision Support Systems 43 (2007) 229–242.
[31] B.D. Meyer, Unemployment insurance and unemployment spells, Econometrica 58
(1990) 757–782.
[32] M. Miozzo, V. Walsh, International Competitiveness and Technological Change,
Oxford University Press, Oxford, 2006.
[33] D.T. Mortensen, Job search, the duration of unemployment, and the Phillips curve,
The American Economic Review 60 (1970) 847–862.
[34] D.T. Mortensen, Markets with search friction and the DMP model, American
Economic Review 101 (2011) 1073–1091.
[35] D.T. Mortensen, C.A. Pissarides, Job creation and job destruction in the theory of un-
employment, The Review of Economic Studies 61 (1994) 397–415.
[36] M. Nerlove, Distributed Lags and Demand Analysis for Agricultural and Other Com-
modities, Agricultural Handbook, vol. 141, US Department of Agriculture,
Washington, D.C., 1958
[37] C.A. Pissarides, Equilibrium in the labor market with search frictions, American
Economic Review 101 (2011) 1092–1105.
[38] S.C. Salop, Systematic job search and unemployment, The Review of Economic
Studies 40 (1973) 191–201.
[39] B. Steil, D.G. Victor, R.R. Nelson, Technological Innovation and Economic Perfor-
mance, Princeton University Press, Princeton, 2002.
[40] H. White, A heteroskedasticity-consistent covariance matrix estimator and a direct
test for heteroskedasticity, Econometrica 48 (1980) 817–838.
[41] A. Zellner, An efficient method of estimating seemingly unrelated regressions and
tests for aggregation bias, Journal of the American Statistical Association 57
(1962) 348–368.
Winston T. Lin received his Ph.D. degree from Northwestern University, and is a Professor
in the School of Management at The University at Buffalo, The State University of New
York. He has published nearly 100 journal articles, presented papers at national and inter-
national conferences, and received best paper awards and research grants. He also won a
Sustained Achievement and Exceptional Scholar Award from The University at Buffalo,
The State University of New York.
Ta-Wei (Daniel) Kao is a Ph.D. Candidate in Operations Management and Strategy. He has
published articles in the European Journal of Operational Research and Computers in Human
Behavior. He will be an Assistant Professor at the University of Michigan at Dearborn
starting with the Fall 2016 semester.
Chia-Ching Chou is an Adjunct Professor in the Department of Management at Canisius
College. He is currently a Ph.D. Candidate in Operations Management and Strategy at
The University at Buffalo, The State University of New York and a member of Alpha Iota
Delta. His research interests include production and operations management, information
technology and e-commerce, sustainability, and environmental management. His studies
have been presented in several major and international conferences such as Decision
Sciences Institute (DSI), Production and Operations Management Society (POMS), and In-
ternational Conference on Engineering, Technology, and Applied Science (ICETA). Also, he
has been awarded several research grants and awards in the U.S. and Taiwan, and praised
by his students with great teaching evaluations.
Raj Sharman is an Associate Professor in the School of Management at The University at
Buffalo, The State University of New York, and earned his Ph.D. degree from Louisiana State
University. He has published more than 45 co-authored papers in DSS, MISQ, etc.
Natalie C. Simpson is an Associate Professor in the School of Management at The
University at Buffalo, The State University of New York, and earned her Ph.D. degree from
the University of Florida. She has published papers in International Journal of Production
Research, Journal of Operations Management, etc.