Professional Documents
Culture Documents
Knowledge Hiding
Knowledge Hiding
Research Note
A R T I C LE I N FO A B S T R A C T
Keywords: Drawing on psychological ownership and social exchange theories, this study suggests theoretical arguments and
Knowledge hiding behavior empirical evidence for understanding employee reactions to distributive, procedural, and interactional (in)jus-
Service employees tice — three crucial bases of employees’ feelings of social self-worth. Utilizing field data and artificial in-
Artificial intelligence telligence technique, this paper reveals that distributive, procedural, and interactional (in)justice contribute to
Organizational injustice
higher levels of knowledge hiding behavior among employees and that this impact is non-linear (asymmetric).
By reuniting the discourses of organizational justice and knowledge management, this study indicates that
feelings of psychological ownership of knowledge and the degree of social interaction are mechanisms that work
with organizational (in)justice to influence knowledge hiding behavior. The current research may inform con-
temporary theories of business research and provide normative guidance for managers.
⁎
Corresponding author.
E-mail addresses: mohammed.abubakar@antalya.edu.tr (A.M. Abubakar), elaheh.behravesh@emu.edu.tr (E. Behravesh),
hamed.aqdam@cyprus.bau.edu.tr (H. Rezapouraghdam), baha.yildiz@cbu.edu.tr (S.B. Yildiz).
https://doi.org/10.1016/j.ijinfomgt.2019.02.006
Received 24 October 2018; Received in revised form 23 February 2019; Accepted 23 February 2019
Available online 22 March 2019
0268-4012/ © 2019 Elsevier Ltd. All rights reserved.
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
(Wang, Noe, & Wang, 2014; Zhao & Xia, 2017), and individuals’ per- artificial intelligence technique.
ceptions of distrust, reciprocity, and unfairness (Connelly et al., 2012).
Notably, employees withhold knowledge from those they distrust 2. Theory and hypotheses
(Connelly, 2012), and willingness to contribute to teamwork decreases
as mistrust increases (Huo, Cai, Luo, Men, & Jia, 2016). 2.1. Social exchange and psychological ownership of knowledge
A vast amount of organizational knowledge is controlled at the in-
dividual level. Because organizations must ensure that employees are Knowledge ownership is a disputed zone for employers and em-
willing to share knowledge, it is crucial for managers and policymakers ployees in knowledge-intensive industries. According to Pierce,
to seek strategies for mitigating and controlling employees’ knowledge Kostova, and Dirks (2001), psychological ownership is a state in which
hiding behavior, especially in knowledge-intensive industries such as individuals feel as though the target of ownership or some piece of that
banking and finance, information technology, and healthcare (Kumar target is theirs. Ownership can be conceptualized as both an objective
and Varkkey, 2018). Increased understanding of knowledge hiding and a psychological state (Pierce & Jussila, 2011). Ownership di-
behavior will help organizations develop appropriate HRM practices minishes extra-role behavior by motivating counterproductive beha-
(Xiao & Cooke, 2018). Previous research has revealed that antecedents viors among incumbent employees (Pierce, Kostova, & Dirks, 2003). In
of knowledge hiding include knowledge-related factors (Černe et al., the context of knowledge, the psychological ownership of knowledge
2017), interpersonal factors (e.g., interpersonal distrust (Černe et al., (POK) among incumbent workers, a feeling of knowledge ownership
2014)), situational factors (e.g., knowledge sharing climate (Connelly (i.e., knowledge, skills, and know-how) and its actual possession (Han,
et al., 2012)), and personality traits (Demirkasimoglu, 2015). Chiang, & Chang, 2010; Pierce, Jussila, & Li, 2017), has the tendency to
However, a consideration of how organizational factors (i.e., orga- predict greater knowledge sharing or hiding. Ownership-driven
nizational (in)justice) can affect knowledge hiding provides us with a knowledge hiding occurs through either (1) the overvaluing of knowl-
more complete understanding of why and when this behavior occurs in edge or (2) an anticipated loss of control (Von der Trenck, 2015). Given
organizations. Given this untapped research area, the primary motiva- this premise, this study theorizes that top management's failure to en-
tion for this study comes from the need to explore the underlying or- sure justice across an organization may backfire in the form of knowl-
ganizational (in)justice mechanisms associated with knowledge hiding. edge-hiding behavior.
To advance theory, research, and practice on how managers can miti- According to Blau (1964), social exchange theory (SET) is an at-
gate the effects of knowledge hiding, it is critical to know understand tempt to explain the dynamics involved in the relationships and
the nature or sources of knowledge hiding based on organizational transactions between two or more parties. Any intervention into the
factors such as (in)justice. In the interim, multiple perspectives have reciprocal process of exchanging resources between organizations and
been used independently to suggest the use of social exchange theory their members can affect employees’ attitudes and behaviors (Cro-
(Parker, Gerbasi, & Cullen-Lester, 2018) and psychological ownership panzano et al., 2016). Thus, positive treatment on the part of employing
theory (Peng, 2013; Von der Trenck, 2015) to explore knowledge organizations in the form of fairness/justice would be reciprocated by
hiding, positive employee behaviors, such as knowledge sharing (Cropanzano,
However, to the best of our knowledge no study has consider the Anthony, Daniels, & Hall, 2017), and vice versa. Employees will tend to
ability of the theories to work together in explaining knowledge hiding. withhold discretionary positive behaviors (i.e., knowledge exchange)
Thus, the second aim and contribution of the current investigation is to when their organizations or colleagues do not treat them fairly (Khalid,
address the possibility by which social exchange theory (SET) and Bashir, Khan, & Abbas, 2018). Drawing on POK and SET, this study
psychological ownership of knowledge (POK) theory interact together offers theoretical arguments and empirical evidence in an attempt to
to predict knowledge hiding. This possibility seems to have been ne- understand employees’ reactions to distributive, procedural, and in-
glected so far. Recently a study by Pirkkalainen, Pawlowski, Bick, and teractional (in)justice, which are parts of a broader construct, that of
Tannhäuser (2018) highlighted the scarcity of investigations linking perceived justice (Alexander & Ruderman, 1987; Bies & Shapiro, 1987).
POK with the intention to share knowledge. This article responds to this The background regarding the assumptions and the development of the
call by linking (in)justice and knowledge hiding, by theorizing that hypotheses will be discussed in detail in the following section.
knowledge hiding simply makes employees feel better because it serves
as a mechanism to punish unfair operant in a social setting (Tepper, 2.2. Organizational (In)Justice
Mitchell, & Almeda, 2011). This paper draws upon SET and POK to
explore the relationship between organizational (in)justice and Organizational justice is defined as a group's or individual's per-
knowledge hiding behavior. ception of the fairness of the treatment received from an organization
Third, the focal point of the current study is justice as a key pre- and the behavioral reaction in response to that perception (James,
dictor of knowledge hiding, more specifically, this paper examines how 1993). Perceived organizational justice represents a cognitive appraisal
the three dimensions of organizational (in)justice explains knowledge concept, which is subject to positive, negative, or non-responsive re-
hiding behavior in Turkish banks. Fourth, artificial intelligence tech- actions on the part of the appraiser. Organizational justice is oper-
niques such as artificial neural networks (ANN) have been used for data ationalized using three dimensions. Namely, distributive justice refers to
analysis in various fields e.g., economics and finance – for stock pre- the fairness of the outcomes employees receive. It involves correlating
diction (Göçken, Özçalıcı, Boru, & Dosdoğru, 2016), medicine – disease organizational reactions with specific evaluations and a resulting in-
prediction (Li, Chen, Sun, & Tao, 2015; Weng, Huang, & Han, 2016), crease in pay, e.g., performance appraisal. Procedural justice refers to the
cloud computing adoption (Priyadarshinee, Raut, Jha, & Gardas, 2017), fairness of the process used to make decisions and determine which
mobile/social commerce (Hew, Leong, Tan, Ooi, & Lee, 2017; Liébana- outcomes employees receive. Lastly, interactional justice involves the
Cabanillas, Marinković, & Kalinić, 2017; Liébana-Cabanillas, interpersonal and informational interactions among employees (Bies,
Marinkovic, de Luna, & Kalinic, 2018), workplace withdrawal 1987; Greenberg, 1990; Thibault and Walker, 1975; Konovsky, 2000).
(Abubakar, Namin, Harazneh, Arasli, & Tunç, 2017), decision-making Contemporary organizations expect their employees to share their
(Duan, Edwards, & Dwivedi, 2019) and supervisor turnover (Quinn, knowledge with peers (Cabrera & Cabrera, 2002; Gagné, 2009). Scho-
Rycraft, & Schoech, 2002). Despite its robustness, the knowledge hiding lars believe that the obstacles complicating the knowledge transfer
research stream is devoid of artificial intelligence data analysis tech- process in organizations can be attributed to the characteristics of the
nique. The present research contributes to the literature methodologi- organizational systems at work (Michailova & Husted, 2003; Wolfe &
cally by providing a stringent examination of the causal relationship Loraas, 2008). Some scholars have cited distrust, HRM practices, per-
between organizational (in)justice and knowledge hiding behavior with ceived support, and organizational identification as factors hindering
46
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
knowledge transfer (Connelly & Zweig, 2015; Koriat & Gelbard, 2014; 2.4. Procedural (in)justice and knowledge-hiding behavior
Wang et al., 2018a; Wang et al., 2018b). Despite this, empirical evi-
dence is lacking in the literature. Knowledge hiding is not simply the In the literature of organizational justice, procedural justice refers to
absence of knowledge sharing. Rather, it is an intentional attempt by an the fairness of the process, formal policies, or procedures via which
individual to conceal or withhold requested knowledge from another decisions are made (Konovsky, 2000) and outcomes are allocated.
person (Connelly et al., 2012; Pan, Zhang, Teo, & Lim, 2018). For ex- Procedural fairness concentrates on the relationship between em-
ample, a situation in which an employee provides false information or ployees and their organization (Cropanzano, Prehar, & Chen, 2002)
pretends not to have the requested information would constitute and, accordingly, tends to be a good predictor of reactions to an entire
knowledge hiding. organization (Folger & Konovsky, 1989) and human resource systems
(Sapienza & Korsgaard, 1996). Organizations are forums for transac-
tions; individuals form perceptions about the fairness of the transac-
2.3. Distributive (in)justice and knowledge-hiding behavior tions with their exchange partners, such as their company. A formal
procedure is established by organizations, and procedural justice is
Distributive (in)justice refers to the fairness of the outcomes re- associated with an organization as whole. Therefore, a social exchange
ceived (e.g., pay, promotion, and compensation), and it is more strongly perspective can help in establishing that procedural justice should be
related to reactions to specific outcomes than to reactions to an entire associated with responses to an organization as a whole.
organization (Folger & Konovsky, 1989; Sweeney & McFarlin, 1993). As stated earlier, POK theory indicates that the situational and in-
According to Connelly and Zweig (2015), knowledge hiding is triplex in dividual factors that foster psychological ownership are context driven
nature, being comprised of (1) evasive hiding, i.e., “where an individual (Pierce et al., 2003). The strength of employees’ claims regarding the
gives incorrect information or a confounding promise of a complete possession of knowledge may vary in the workplace based on the justice
answer in the future;” (2) playing dumb, i.e., “where an individual or injustice of an organization's procedures. In other words, procedural
pretends not to have the requested information;” and (3) rationalized justice refers to the perceived fairness of the allocation process that
hiding, i.e., “where an individual provides a justification, saying that leads to outcomes (e.g., promotion), and it can enhance perceived
he/she is not allowed to share information or blaming another party.” fairness, even if the outcomes themselves cannot be influenced (Thibaut
Existing scholarly works have consistently shown that trust for an or- & Walker, 1975); it may even affect the knowledge-hiding behaviors of
ganization facilitates organizational identity, organizational commit- employees. Specifically, employees may engage in knowledge-hiding
ment, and participative and knowledge-sharing behaviors (Cox, behavior as a strategy with which to cope with the unfairness they
Zagelmeyer, & Marchington, 2006; Han et al., 2010; Meyer, Allen, & encounter in their organization's procedures.
Allen, 1997). According to SET, the feeling of being treated fairly by an organi-
Interestingly, POK is associated with the strength of workers’ claims zation will make subordinates more occupied with their work because
about the knowledge they possess. Specifically, distributive in(justice) reasonable authoritative systems can improve the level of trust and
impairs organizational social relationships and balance. Under such certainty within an organization. This encourages positive attitudes and
circumstances, individuals will be less likely to share knowledge with behaviors in the workplace (He, Zhu, & Zheng, 2014). As suggested by
their peers to avoid losing control because this knowledge was gained Konovsky and Pugh (1994), the use of fair procedures exhibits an or-
and is controlled by them (Pierce et al., 2003). This paper advances the ganization's respect for the dignity and rights of its employees. Orga-
theory that employees may engage in knowledge-hiding behaviors to nizational procedural justice can increase employees’ knowledge-
restore balance after experiencing distributive injustice. The primary sharing behavior because employees’ discretionary behaviors, such as
drivers of such behavior seem to be knowledge ownership and a lack of sharing knowledge according to the rules of reciprocation, will be en-
trust in the employing organization. This conception centers on the hanced when employees are treated genuinely by an organization
value of knowledge, ideas, and know-how for the organizations; thus, (Schepers & Van den Berg, 2007; Wang & Noe, 2010). According to the
knowledge-hiding behaviors impair both peers’ productiveness and above discussion, it is reasonable to assume that the fairness of an or-
organizational productivity. ganization's procedures and process in terms of allocating outcomes
Drawing on the tenets of SET, dyadic interactions within organiza- leads to reciprocal employee organizational behaviors, such as the
tions, i.e., employer-employee and employee-employee, are generally avoidance of knowledge-hiding behavior. Thus, the following hypoth-
governed by an implicit and unspoken social exchange (Blau, 1964). esis is proposed:
Thus, organizations are setting for transactions (Cropanzano, Howes,
H2 (:). Procedural (in)justice has an impact on employee knowledge-
Grandey, & Toth, 1997) in which individuals recognize at least two
hiding behavior.
imperative exchange partners, the organization and peers. Unfair work
outcomes resulting from distributive injustice trigger the development
of a mindset of distrust. Just as trust and distrust are opposites, so are 2.5. Interactional (in)justice and knowledge-hiding behavior
the positive and negative reciprocity present in organizational settings.
Unfair treatment is associated with unfavorable work attitudes and Interactional (in)justice is the third dimension of justice that is
behaviors (Konovsky, 2000). Employees often engage in reciprocating perceived by employees. Bies and Moag (1986) advanced the justice
counterproductive behaviors in their work settings because this enables literature by introducing the concept of “Interactional Justice,” which
them to penalize unfair partners in a social setting and makes them feel includes the quality of the interpersonal interactions between in-
better (Kahneman, Knetsch, & Thaler, 1986). This suggests the essential dividuals. It shapes employees’ reactions to their immediate work en-
need to restore justice when rules and norms are violated (Lerner, vironment, their supervisors, and their peers (Masterson, Lewis,
1980), for instance, when workers hide their knowledge. Based on the Goldman, & Taylor, 2000). Interactional justice is most likely to occur
norm of reciprocity and related outcomes, distributive injustice can when superiors or group members treat employees with interpersonal
impair reciprocal norms, thus encouraging knowledge-hiding behavior. dignity, provide justifications or explanations for all actions, discuss
Therefore, we propose the following hypothesis: implications, and engage in honest transactions with employees
(Cropanzano et al., 2002). This explains why Connelly et al. (2012)
H1 (:). Distributive (in)justice has an impact on employee knowledge-
stress the importance of social relationships at work. Surprisingly, the
hiding behavior.
nature of the association between interactional (in)justice and knowl-
edge-hiding behavior is still unknown.
One encouraging aspect of the present study that extends beyond
47
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
the previous research is that the relationship between interactional (in) assured to abate the potential effects of common method bias. In ad-
justice and knowledge-hiding behavior is explored using POK theory dition, a self-reported design was adopted in this study because per-
and SET. Knowledge is acquired and controlled by individuals. ceived organizational (in)justice and knowledge hiding behavior are
Individuals tend to treat knowledge as their personal property (Peng, private and sensitive issues, which are best known by the employee
2013). According to POK theory, individuals can create feelings of alone (Chan, 2009; Conway & Lance, 2010).
ownership over anything they have acquired by spending time and
energy or have within their control. The fairness of interpersonal 3.2. Instruments
treatment (Martínez-Tur, Peiró, Ramos, & Moliner, 2006) may de-
termine whether employees tend to hide knowledge from their orga- Knowledge Hiding Behavior has 3 dimensions namely: evasive
nizations and/or peers. Thus, if employees perceive interactional (in) hiding, rationalized hiding and playing dumb, each of these dimensions
justice, they will be reluctant to share their property with others. were measured with 4 items each adopted from (Connelly et al., 2012)
Drawing from SET, the interpersonal mechanisms involved in work. Employees were asked to respond using a 7-point scale (1
dyadic interaction and dyadic social exchange are an imperative part of strongly disagree to 7 strongly agree) the value that reflects their be-
the current study and have been considered a prominent factor in haviors in a specific situation and interaction with co-workers. Items
knowledge hiding. According to SET, social interaction and treatment are presented in Table 1.
are the lifelines of quality workplace relationships, which may spill into Organizational (In)Justice encompasses three dimensions: dis-
other important organizational domains. Thus, employee perceptions of tributive, procedural, and interactive justice. Distributive justice (DJ)
interpersonal treatment and interactions are important because they was measured with five items, procedural justice (PJ) with six items
shape several employee and organizational outcomes. For instance, and interactive justice (IJ) was measured with nine items adopted from
employees may engage in hiding or withholding knowledge from those (Niehoff & Moorman, 1993). These items have been validated in Turkey
with whom they interact, e.g., supervisors and/peers. In the current by (Gürbüz & Mert, 2009). Participants were asked to rate (in) justice
study, we propose that this reciprocal justice loop applies to situations scale based on to their perception using a 7-response point that spans
in which an employee requires information and knowledge. Individuals from 1 = strongly disagree to 7 = strongly agree. Items are presented
who have feelings of injustice and distrust are more likely to reciprocate in Table 1.
and hide knowledge in return. Based on the above reasoning, the fol-
lowing hypothesis was proposed: 3.3. Demographic variables
H3 (:). Interactional (in)justice has an impact on employee knowledge-
hiding behavior. Of the 152 responses, 59.2% are male and the rest are female em-
ployees. Age wise, 36.8% are between 31 and 40, 34.9% are between 21
The research model and hypotheses are presented in Fig. 1. and 30, 15.1% are between 41 and 50, 9.2% are above 50 and the rest
less than 20 years old. In terms of marital, 64.5% are either married or
3. Methods divorced and the rest are single. Approximately, 30.3% are earning less
than 3,000TL monthly; 27% are earning between 3000 and 4,999TL
3.1. Data collection and sampling procedure monthly; 22.4% are earning between 5000 and 6999 TL; 9.9% are
earning between 7000 and 8999 TL; and the rest more than 9000 TL.
One of the cutting edge human technological and scientific inven- About 53.3% of the respondents have bachelor's degrees, 33.6% have
tions was the Internet technology, with various contribution to man higher degrees, 4.6% are high school graduates and the rest have some
ranging from economic, social, technology to cultural facets. The fur- college degrees. Approximately, 30.9% of the respondents have more
ious competition on price and service to attract the consumers has in- than 9 years organizational tenure; 27% have between 1 and 4 years;
tensified the nature of competition in the sector. The competitive, 23.7% have between 5 and 8 years; and the rest less than a year.
labor-intensive and information-intensive nature of the banking in-
dustry makes it suitable to investigate knowledge hiding behavior. The 3.4. Artificial intelligence analysis
research instruments were developed in English and back-translated to
Turkish by two bilingual Academics. Utilizing simple random sampling Partial least squares (PLS) regression is an addendum to the general
technique, 152 valid responses were retrieved from bank employees in stepwise, generalized and multiple linear regression modeling. These
the Aegean region of Turkey. Confidentiality of the respondents were modeling techniques operate based on the effect of predictor variables
48
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
Table 1
Scale items.
Cronbach alpha-α Mean
(i.e., X's) on response variable (i.e., Y), with the following equation: Liébana-Cabanillas et al., 2018; Priyadarshinee et al., 2017; Raut,
Priyadarshinee, Gardas, & Jha, 2018), because it outperformed tradi-
Y = b0 + b1 X1 + b2 X2 + …+bp Xp
tional methods such as regression, linear modeling, CB-SEM and PLS-
SEM.
These modeling techniques were designed to meet convergent and
This is primarily due to its ability to detect both linear and non-
divergent validity as well as canonical correlation assumptions. These
linear relationships (Abubakar, 2018) couple with (1) high predictive
assumptions and methods “impose restrictions such that (1) factors
accuracy and validity predictions, (2) fast learning and accurate pre-
underlying the Y and X variables are extracted from the Y’Y and X’X
dictions – ANN can acquire new learning and store this memory (3)
matrices, respectively, and never from cross-product matrices involving
exerts reliability dominance over regression, CB-SEM and PLS-SEM, (4)
both the Y and X variables, and (2) the number of prediction functions
ANN is less restrictive in nature – no mandate for factor loadings,
can never exceed the minimum of the number of Y variables and X
normality assumptions, linearity, homoscedasticity, lack of power,
variables” (Hill & Lewicki, 2007, pg. 234). PLS is among the least re-
sample error and sample size (Abubakar et al., 2017; Abubakar,
strictive multivariate modeling technique. According to StatSoft (2013),
Karadal, Bayighomog, & Merdan, 2018; de la Paz-Marín, Campoy-
“in PLS regression, prediction functions are represented by factors ex-
Muñoz, & Hervás-Martínez, 2012), (5) fault tolerance – ANN can ac-
tracted from the Y’XX’Y matrix. The number of such prediction func-
commodate samples with great individual differences (Li et al., 2015),
tions that can be extracted typically will exceed the maximum of the
and (6) ANN has good generalization capabilities – robustness against
number of Y and X variables”.
noisy or missing data (Göçken et al., 2016).). These attractive features
A vast majority of research in the field of knowledge management
of ANN contribute to its versatility, more specifically, this implies that it
and information systems rely on methods such as regression, covar-
is largely insulated from statistical flaws and myths applicable to tra-
iance-based (CB-SEM) and partial least squares (PLS-SEM) structural
ditional methods. Scholars (i.e., Xiao & Cooke, 2018) have urged re-
equation modeling to test for causal-relationships (e.g., Černe et al.,
searchers to utilize variety of methods diagnosing or predicting
2014; Černe et al., 2017; Khalid et al., 2018; Wang et al., 2018a; Wang
knowledge hiding behavior citing that doing so may reduce the nuances
et al., 2018b; Zhao & Xia, 2017). Despite the widely acceptance of these
in knowledge hiding researches.
methods, they have the possibilities of over-simplifying the complex-
This paper responds to this call by employing ANN because it out-
ities in decision-making processes (Haykin, 1999; Abubakar et al.,
smarts other techniques such as regression, PLS-SEM and CB-SEM,
2017). As an artificial intelligence tool, ANN has recently gained mo-
while performing estimations with high level of accuracy (Abubakar
mentum from information systems and knowledge management re-
et al., 2017). The capability of ANN to model complex interactions with
searchers (i.e., Duan et al., 2019; Liébana-Cabanillas et al., 2017;
49
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
flexible non-linear response values gave the method superior predictive illustrated diagrammatically in Fig. 2. The result shows that component
power in comparison to traditional methods (Liébana-Cabanillas et al., 1, 2, and 3 explained 75.72%, 19.32%, and 4.96% of the variance in
2017; Hew et al., 2017; Olden & Jackson, 2002; Taneja & Arora, 2019). knowledge hiding behavior. To visualize the explained variance the
Apart from the fact that ANN can detect non-linear functional re- following functions were utilized (see Fig. 3). The test data model
lationships that are hidden in case data, ANN's are able to apply these produces an RMSEP of 1.25. The algorithm (codes) used are provided in
relationships to new data (de la Paz-Marín et al., 2012). Knowledge Appendix 1.
hiding and organizational (in)justice are complex issues, as such, ANN ANN modeling was conducted using a Multi-Layer Perceptron with
seems suitable to test these variables. 3:2 hidden nodes and Resilient Backpropagation with Weight
More subtly, ANNs are information processing systems that com- Backtracking algorithm. Logistic function was used as the activation
prise a certain number of information processing units (also “cells,” function for both hidden and output layer, and Sum Squared Errors
“neurons”) that incorporate mathematical functions and are connected (SSE) was used as differentiable error function. The algorithm (codes)
by directed weighted links (Strohmeier & Piazza, 2015). ANNs con- used is provided in Appendix 1. Training data used was 75% and 25%
stitute a category of knowledge discovery that is able to solve clus- for testing, the synaptic weights of the input nodes on the hidden and
tering, classification, estimation and prediction tasks (Haykin, 1999). output nodes are shown in Fig. 4. The objective of this algorithm is to
The information processing units are usually organized in layers, with minimize error until the ANN learns through the learning or training
an input layer to provide the input data, one or more hidden layers to process. During the training process, random synaptic weights were
process the data through the neural network and an output layer to assigned to the connections and the aim is to adjust them to obtain
provide the result (Abubakar, 2018). minimal error. The training process needed 372 steps until all absolute
ANN can forecast both symmetric and asymmetric relationships partial derivatives of the error function were smaller than 0.010. ANN
with nearly 100% accuracy, and without a mandate for multivariate are difficult to work with, for example, visually they suffer from a
assumptions e.g., homoscedasticity, normality, and linearity problem of serious clutter, making the interaction weights mostly ille-
(Abubakar, 2018). Additionally, this strength of ANN as an AI method gible (Beck, 2013).
may help overcome the problem of common method bias in knowledge In addition, the plot.nn function we used has some undesirable
management studies, since research models can be simulated and in behavior (Beck, 2015) in terms of interpretation. ANN experts (i.e.,
each case a test for accuracy in prediction can be conducted. It is on this Alice, 2015; Günther & Fritsch, 2010) suggested that distribution of the
premises that this study combines PLS package (i.e., plsr) and ANN (i.e., generalized weights is easy and useful in interpreting the nature of the
neuralnet) in R to predict knowledge hiding behavior among bank predictor variable effect on the response variable. This is because the
employees. PLS regression is used for factor structuring and to under- weight parameters are the knowledge carriers that keep the learning
stand the explained variance of predictor variables on a response results. The generalized weights were evaluated based on closeness to
variable; and ANN is used in modeling and testing the accuracy of the or below 0, all the generalized weights for the predictor variables were
research model. 75:25% ratio of the collated data is assigned for mostly above 0. The distributions of the generalized weights illustrated
training and testing following the procedures adopted by prior studies in Fig. 5 shows that predictor variables exert a non-linear effect on the
(i.e., Abubakar et al., 2017; Liébana-Cabanillas et al., 2017; Liébana- response variable. More specifically, the impact of organizational in
Cabanillas et al., 2018; Raut et al., 2018). (justice) on knowledge hiding behavior is non-linear (asymmetric).
The author utilized two cross-validation estimates, CV is the or- Table 4 illustrate that low and moderate distributive justice are
dinary CV estimate, and adjCV is the bias-corrected estimate. The root more likely to trigger knowledge hiding behavior when compared with
mean squared errors of prediction (RMSEP) were estimated during the higher distributive justice. The lower the procedural justice, the higher
cross-validation process for the PLSR models. The probable prediction the tendency of employees to exhibit knowledge hiding behavior; and
errors for the new samples were estimated using cross-validation. The moderate procedural justice is likely to trigger mediocre knowledge
cross-validation was repeated ten times by which data from each of the hiding behavior as opposed to higher procedural justice. Low inter-
subsample group used exactly once for validating the model. Average of personal justice is less likely to knowledge hiding behavior, the results
ten iterations was reported for the model performance. In Table 2, a 3- denote that low and high interpersonal justice are likely to cause slight
components leave-one-out (LOO) cross-validated prediction analytics is effect, more specifically, some level of employee's knowledge hiding
illustrated. Root Mean Square Error of Prediction (RMSEP) value 1.37 behavior might be exhibited. However, moderate level of interpersonal
suggest that the 3-components model is reliable as it does not differ justice is more likely to trigger knowledge hiding behavior than low and
significantly from 1 or 2-components model. RMSEP values are high. Cross validation is another important step for building predictive
models. To avoid over-fitting, a 10-fold cross-validation modeling with
Table 2 a ratio of 75:25 data for training and testing of predictions was con-
PLS coefficients. ducted. Mean Square of Error (MSE) from ten networks was used to
(Intercept) 1 comps 2 comps 3 comps
examine the accuracy of the model. Table 3 shows that MSE stretches
from 0.026 to 0.046 for training and 0.009 to 0.030 for testing. Based
Data: X dimension : 114-3 on this outcome, we concluded that the model is reliable in predicting
Y dimension : 114-1 output variable and that the present predictions are reliable since the
Fit method : kernelpls
VALIDATION: RMSEP
MSE are closer to .000. Thus, all hypotheses [1, 2 and 3] received
Cross-validated empirical support.
using 10
RANDOM 4. Discussion
segments
CV 1.310 1.335 1.375 1.377
adjCV 1.310 1.333 1.370 1.373 Knowledge management within organizations is challenging and
TRAINING: % variance imperative in today's knowledge-based economy. Understanding the
explained dynamics of knowledge exchange within organizations can help man-
X - 75.720 95.036 100.000 agers and policy makers to design systems and apply processes facil-
Knowledge hiding - 1.035 1.046 1.046
behavior
itating knowledge sharing. Organizations expect their employees to
share their knowledge with their colleagues (Cabrera & Cabrera, 2002;
Note: Fit method = kernelpls; RMSEP = RMSEP stands for Root Mean Square Gagné, 2009; Hwang, Lin, & Shin, 2018). However, the “intellectual
Error of Prediction. assets” of employees do not belong to organizations, so these
50
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
organizations cannot force workers to transfer their knowledge to other predict knowledge hiding, but the antecedents of knowledge hiding in
organization members (Kelloway & Barling, 2000). Knowledge hiding, non-Western countries have not been widely investigated. Thus, we
although it is not always harmful, does hinder effective knowledge extend the literature by identifying a potential antecedent and, in doing
transfer within an organization (Bogilovic, C, & Skerlavaj, 2017; Černe so, interrogating concepts developed in the Western World.
et al., 2014; Connelly et al., 2012). In addition, affect climate is closely intertwined with organizational
Recently, the issue of “employees’ knowledge withholding” and its (in)justice. Thus, exploring this link will aid organizations in designing
motivations and consequences has attracted a great deal of attention and implementing knowledge-management practices to foster knowl-
(Černe et al., 2014; Connelly et al., 2012; Connelly & Zweig, 2015; De edge sharing and decrease hiding behavior. The current study extends
Geofroy and Evans, 2017; Huo et al., 2016; Kumar and Varkkey, 2018; the literature on knowledge-hiding behavior empirically by examining
Serenko & Bontis, 2016). Still, more work is needed. An increased un- the role of organizational in(justice) in predicting banking employees’
derstanding of knowledge hiding will benefit knowledge management knowledge-hiding behaviors. This study highlights the contextual ap-
and human resource management practices in various ways. For in- plication of artificial intelligence (AI) techniques in an organizational
stance, affect climate has been shown to be a determinant of personal setting to help predict organizational behavior. Using primary data, a
mood and is defined as “employees’ shared perception of organizational normative conceptual model that seeks to determine whether organi-
aspects such as policies, practices and routines, as well as the behaviors zational (in)justice can instigate employee knowledge-hiding behavior
that are expected, supported or rewarded regarding their affective ex- is used. Several evaluations and prototypes reveal the application of AI
pressions or experiences” (Parke & Seo, 2017, 335). Xiao and Cooke to be a knowledge-discovery technique in human resource management
(2018) argue that this could be linked to knowledge hiding among (HRM) research (e.g., Fan, Fan, Chan, & Chang, 2012; Lawler & Elliot,
employees. Previous research has claimed to explore the factors that 1996; Quinn et al., 2002; Strohmeier & Piazza, 2015).
51
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
Xiao and Cooke's (2018) meta-analysis, drawn from published work Table 3
on knowledge hiding, shows that only a few studies have been con- Neural network - MSEs.
ducted on the subject of knowledge hiding. Of the 52 papers sampled, Neural network Training Testing
17 articles (33%) used qualitative methods, two (4%) used mixed
methods, and the remainder (63%) used quantitative methods. More- 1 0.046 0.014
over, most of these quantitative studies adopted a linear approach, e.g., 2 0.026 0.030
3 0.038 0.022
structural equation modeling, regression, and/or ANOVA. Xiao and 4 0.041 0.009
Cooke (2018) urged future research to adopt a variety of methods to 5 0.033 0.012
cater to investigations with different themes and levels of analysis re- 6 0.042 0.012
garding knowledge hiding. The deployment of (artificial neural net- 7 0.039 0.016
8 0.029 0.013
works) in knowledge management studies for behavior prediction is
9 0.038 0.019
scarcely discussed in the literature. For instance, Zhao and Xia (2017) 10 0.039 0.030
argue that the relationship between ostracism and knowledge hiding is Mean MSE 0.037 0.018
curvilinear. In response to the above-mentioned call, this study uses
ANNs to explore whether the effect of organizational (in)justice on Notes: Input nodes – distributive, procedural and interactional (in)justice;
knowledge hiding behavior is non-linear. The findings are imperative output node – knowledge hiding behavior.
because the central paradox in knowledge management studies is the
issue of linearity. Existing studies most commonly relate variables
52
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
Table 4 Dávila, & Kaplan, 2000). Because of its effect on employees’ perceptions
Non-linear effect of organizational in(justice) on knowledge hiding behavior. of fairness, which may subsequently force them to take refuge in
Exogenous variables Low Moderate High knowledge hiding, the socialization process can be effective in pro-
moting social interactions among employees (Chatman, 1989). Thirdly,
Distributive justice High Effect High Effect Slight effect in terms of organizational culture, adopting a teamwork-based ap-
Procedural justice High Effect Moderate Effect Slight effect
proach can promote interaction and bonding among employees. This
Interpersonal justice Slight Effect High Effect Slight Effect
may alleviate knowledge hiding and thus motivate the sharing of
knowledge.
linearly, ignoring the simple fact that human behavior is inconsistent Subsequently, POK theory accepts employees’ claims on the
and full of uncertainty. The present study contributes to the existing knowledge and skills they possess. Organizational (in)justice impedes
literature by showing that this relationship is non-linear. Our findings organizational harmony and social balance. Thus, individuals will be
provide empirical support for Serenko and Bontis (2016) and Kumar, less likely to share knowledge with peers to avoid losing hard-earned
Jha, and Varkkey (2018), who argued that organizational culture and resources. For instance, a performance-oriented climate was found to be
related factors (i.e., the norm of sharing knowledge, the presence of positively associated with knowledge hiding (Anaza & Nowlin, 2017).
social trust, fairness, job insecurity, and justice) are strong determinants This implies that programs designed to promote work performance and
of knowledge-hiding behavior. knowledge sharing do not necessarily reduce knowledge hiding beha-
vior. The findings of the current study have important implications
regarding business practices that can be useful from a managerial
4.1. Implications for theory perspective. Firstly, management should ensure that there is a climate in
which employees perceive that the formal processes involved in deci-
The contributions of the current study are three-fold. Firstly, in sion-making, the allocation of rewards, and the quality of the interac-
comparison to knowledge-sharing behavior, the knowledge-hiding tion and information exchange are fair. Secondly, organizations can
construct is distinct and relatively new. Scholars (Issac & Baral, 2018; develop “information systems” to capture certain types of knowledge
Lanke, 2018) have given knowledge hiding in organizations, especially and keep them in data warehouses to ensure equal and fair access for all
its antecedents and consequences, increased attention. In congruence to employees. Thirdly, managers and practitioners can increase employees’
this, the current study contributes to the body of existing knowledge by perceptions of the trustworthiness of their colleagues by granting in-
observing how each component of organizational (in)justice may affect centives for sharing knowledge, providing managerial support, and
knowledge hiding in organizations. Secondly, previous studies suggest enhancing the level of personal contact with colleagues to improve
that employees with perceptions of fairness in regard to the quality of organizations’ knowledge-sharing climates. Fourthly, managers must
workplace interpersonal relationships, information flow, and the allo- know that (in)injustice results in knowledge hiding, which results in a
cation and distribution of rewards are less likely to exhibit unwanted reciprocal distrust loop and leads to further knowledge hiding. Thus, it
behaviors (e.g., Cohen-Charash & Spector, 2001; Hemdi & Nasurdin, is fair to say that (in)injustice augments counterproductive work be-
2007). The significant contribution of the study is the association of two haviors. For instance, knowledge hiding can inhibit the flow of
complementary theoretical perspectives, POK theory and SET, which knowledge due to retaliation, which can start at the unit level before
highlight the positive and negative consequences of knowledge hiding, expanding to an organization-wide level and even spilling out into the
respectively. The study applied and connect these unlinked theories to community. A community, society, or organization characterized by a
uncover how (in)justice can trigger knowledge-hiding behavior. high level of knowledge hiding will be less likely to be innovative.
Thirdly, the AI findings of the current study indicate the influence of (in)
justice on knowledge hiding, thereby making significant theoretical 5. Conclusion
contributions on the methodological front. In summary, this paper ar-
gued that knowledge-hiding prediction using ANNs allows for proactive Knowledge hiding is inherently risky and can potentially harm or-
knowledge-based HRM. For instance, ANNs have the capacity to predict ganizational objectives. Building on SET and POK theory, this paper
and forecast. Predicting and proactively managing the prevalence of explained how and when organizational (in)justice may lead to
knowledge-hiding behavior can avoid or at least alleviate the severe knowledge-hiding behavior. The results not just advance the knowl-
downsides of dysfunctional behavior, such as declines in performance, edge-hiding and organizational (in)justice literature but also demon-
increased turnover, and declines in the number of innovations and or- strate the benefits of using an advanced methodological approach e.g.,
ganizational productivity. Additionally, the costs of recruiting, in- ANNs, in studying sensitive issues. Specifically, the results revealed that
troducing, training, and mentoring new employees can be alleviated, distributive, procedural, and interpersonal (in)justice have a non-linear
which is especially important given intense competition and the current (asymmetric) impact on knowledge-hiding behavior. These results also
global economic outlook. suggest that future research and practice should account for the po-
tential detrimental effects of organizational (in)justice on changes in
4.2. Implications for practice knowledge-hiding behavior.
SET suggests that knowledge hiding is dependent on the frequency 5.1. Limitations and future research courses
and quality of social interactions among entities, including the orga-
nization and its members. Firstly, the findings indicate that not only the This study's limitations provide several opportunities for future re-
distributive and procedural fairness of organizations but also the search. Although the sample size for this study was mediocre given the
quality of interpersonal interactions among supervisors and sub- required analysis, the limited number of participants restricted the
ordinates are quite effective in mitigating workplace misbehavior. All generalizability of the current findings. Most empirical and self-report
three dimensions of (in)justice help in avoiding the reciprocal distrust data are subject to common method bias (CMB). However, the use of an
loop. Human resource managers and supervisors have the power to ANN may have made a profound improvement in terms of abating the
alleviate the negative effects of knowledge hiding in the workplace by potential effects of CMB due to its complex nature and ability to identify
creating a favorable justice climate, as well as a friendly and knowl- linear and non-linear relationships. The use of activation functions and
edgeable work environment. Secondly, the current study suggests that bias functions in training the neural networks and the testing phase also
organizations should have a design process and performance evalua- served as a validation measure. For instance, Weng et al. (2016) argue
tion/reward system that consider both process and outcomes (Simons, that an ANN can significantly improve the generalization ability of
53
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
learning systems via training; the cross-validation and test data pro- information hiding at the team or group level. For instance, in-
vided by an ANN can help optimize the network parameters and ar- vestigating knowledge hiding through the lens of the social network
chitecture (e.g., by reducing residual generalization errors), thereby perspective may be fruitful because networks are components of orga-
allowing for near 100% accuracy and successful application in real-life nizations. Thus, the strength or weakness of network ties has a tendency
decision making (Weng et al., 2016). Additionally, the replication of to strengthen or dampen knowledge-hiding behavior. Stronger ties,
this model in other service industries, such as tourism, aviation, and which denote closeness and/or friendship, may dampen the level of
retailing, as well as using a larger sample size, a time lag design, and hiding (Granovetter, 1977). The literature frequently considers
multiple data sources, is recommended. knowledge hiding from the perspective of the knowledge hider. It will
The obstacles facing the knowledge transfer process in organizations be imperative to also study the perceptions of knowledge seekers, i.e., a
can be attributed to the characteristics of the associated organizational knowledge seeker's perception that a colleague has deliberately with-
systems and the personalities of their members (Michailova & Husted, held knowledge from him or her.
2003; Wolfe & Loraas, 2008). This study only observed the organiza-
tional factors linked to employees’ knowledge-hiding behavior. There-
fore, it would be useful for future research to extend this study and Acknowledgements
consider personal factors because personality affects people's cognitive,
affective, and behavioral outcomes, such as employees’ deviant beha- This research did not receive any specific grant from funding
vior at work (Mount, Barrick, Scullen, & Rounds, 2005). In addition to agencies in the public, commercial, or not-for-profit sectors. The au-
personal factors, future studies can investigate the process of thors have no conflict of interest.
Appendix A. Appendix 1
54
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
nn < - neuralnet (KHB ∼ DJ + PJ + IJ, data = train.cv, hidden = c(3,2), err.fct = “sse”)
columns < - c(“DJ”, “PJ”, “IJ”)
covariate < - subset(train.cv, select = columns)
pr.nn < - compute(nn, covariate, rep = 1)
pr.nn < - pr.nn$net.result*(max(train.cv$KHB)-min(train.cv$KHB)) + min(train.cv$KHB)
train.cv.r < - (train.cv$KHB)*(max(train.cv$KHB)-min(train.cv$KHB)) + min(train.cv$KHB)
cv.error[i] < - sum((train.cv.r - pr.nn)^2)/nrow(train.cv)
print(paste(cv.error[i]))
pbar$step()
}
mean(cv.error)
############################FOR TESTİNG##########################
set.seed(450)
cv.error < - NULL
k < - 10
library(plyr)
pbar < - create_progress_bar(’text’)
pbar$init(k)
for(i in 1:k)
{
index < - sample(1:nrow(KHBDAT),round(0.75*nrow(KHBDAT)))
train.cv < - scaled[index,]
test.cv < - scaled[-index,]
library(neuralnet)
nn < - neuralnet (KHB ∼ DJ + PJ + IJ, data = test.cv, hidden = c(3,2), err.fct = “sse”)
columns < - c(“DJ”, “PJ”, “IJ”)
covariate < - subset(test.cv, select = columns)
pr.nn < - compute(nn, covariate, rep = 1)
pr.nn < - pr.nn$net.result*(max(test.cv$KHB)-min(test.cv$KHB)) + min(test.cv$KHB)
test.cv.r < - (test.cv$KHB)*(max(test.cv$KHB)-min(test.cv$KHB)) + min(test.cv$KHB)
cv.error[i] < - sum((test.cv.r - pr.nn)^2)/nrow(test.cv)
print(paste(cv.error[i]))
pbar$step()
}
mean(cv.error)
55
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
56
A.M. Abubakar, et al. International Journal of Information Management 49 (2019) 45–57
integration in wikis: The moderating role of knowledge equivocality. International Xiao, M., & Cooke, F. L. (2018). Why and when knowledge hiding in the workplace is
Journal of Information Management, 43, 64–75. harmful: A review of the literature and directions for future research in the Chinese
Weng, C. H., Huang, T. C. K., & Han, R. P. (2016). Disease prediction with different types context. Asia Pacific Journal of Human Resources (In press).
of neural network classifiers. Telematics and Informatics, 33(2), 277–292. Zhao, H., Xia, Q., He, P., Sheard, G., & Wan, P. (2016). Workplace ostracism and
Witherspoon, C. L., Bergner, J., Cockrell, C., & Stone, D. N. (2013). Antecedents of or- knowledge hiding in service organizations. International Journal of Hospitality
ganizational knowledge sharing: A meta-analysis and critique. Journal of Knowledge Management, 59, 84–94.
Management, 17(2), 250–277. Zhao, H., & Xia, Q. (2017). An examination of the curvilinear relationship between
Wolfe, C., & Loraas, T. (2008). Knowledge sharing: The effects of incentives, environment, workplace ostracism and knowledge hoarding. Management Decision, 55(2), 331–346.
and person. Journal of Information Systems, 22(2), 53–76.
57