Pilot study was conducted before collecting the whole sample for the main study. The purpose of the pilot study is to validate the quality of selected measurement scales. The sample size for the pilot study is n = 50. Results of reliability, validity, dropped items (if any) and relationships among variables will be presented.
KMO and Bartlett’s Test of Sphericity
The main purpose of KMO and Bartlett’s Test of Sphericity is to assess whether correlations exist among variables (or among indicators) so that a following factor analysis is applicable. KMO values range from 0 to 1. Values close to 1 indicate small partial correlation coefficients while values close to 0 imply that factor analysis is not a good idea. KMO value should be higher than 0.5 to be acceptable, higher than 0.6 to be mediocre fit, higher than 0.7 to be middling fit (Kaiser & Rice, 1974). For factor analysis to work, some correlations among variables are required. Therefore, a significant Bartlett’s test of sphericity is needed. Small p-values (p < .05) from Bartlett’s test of sphericity indicate appropriateness to run factor analysis.
Table 4.1 summarizes the results from KMO and Bartlett’s Test of Sphericity of the pilot study.
All the KMO values are higher than 0.6, and all constructs’ p-values are smaller than .001, which is satisfactory.
Table 4.1.
KMO and Bartlett’s Test of Sphericity of Pilot Study
Constructs Number of Items KMO Variance Explained Bartlett’s Test of Sphericity
KM_H 4 .658 62.09% 82.280***
KM_S 3 .703 75.25% 58.390***
I_PD 3 .719 74.67% 53.820***
I_PC 3 .760 86.02% 103.012***
BP_C 3 .669 64.22% 28.513***
BP_F 3 .725 89.18% 137.701***
Note. (n = 50) *** p < 0.001
Factor Loadings
Factor analysis was conducted using AMOS 23.0 to achieve factor loadings. Basically, factor loadings represent the relationship between a latent variable and its underlying factors. If a factor loading is below 0.4, the item should not be retained because it doesn’t well represent the latent variable. Accordingly, item km_h4 with factor loading of .336 may be dropped from further analysis. However, since the value is close to 0.4, the item may undergo factor analysis again in the main study. If its factor loading is still below 0.4 in the main study, the item will be dropped.
There are two criteria to assess construct reliability: Cronbach’s alpha or Composite Reliability (CR). Cronbach’s alpha or CR value of 0.7 or greater is required to ensure internal consistency of measurement scales (Hair Jr, Hult, Ringle, & Sarstedt, 2016). Table 4.3 shows Cronbach’s alpha values of every construct before dropping items. All the values are satisfactory (higher than 0.7).
Table 4.3.
Reliability and Validity Results of the Pilot Study
Construct Number of items
Cronbach’s alpha
Composite Reliability
Average Variance Extracted
KM_H 4 0.78 0.86 0.62
KM_S 3 0.83 0.90 0.75
I_PD 3 0.83 0.89 0.74
I_PC 3 0.91 0.94 0.86
BP_C 3 0.72 0.84 0.63
BP_F 3 0.93 0.96 0.89
Validity Test
Construct validity is tested on convergent validity and discriminant validity. Convergent validity is confirmed when CR is of 0.7 or higher, and Average Variance Extracted (AVE) is of 0.5 or greater (Hair Jr et al., 2016). Table 4.3 summarizes every CR (ranging from 0.86 to 0.96) and AVE (ranging from 0.62 to 0.89), which all satisfy the cut-off criteria. Hence, convergent validity is established. Discriminant validity can be examined by Fornell and Larcker criterion which compares the square-root of AVE (i.e. √𝐴𝑉𝐸 ) with correlations among latent constructs.
A latent construct’s square-root of AVE should be higher than its correlations with other constructs to ensure discriminant validity. Table 4.4 shows that the square-root of AVE of each construct (in the diagonal) is higher than its correlations with other constructs (in the corresponding rows and columns), thus, confirm discriminant validity of measurement scales used.
Table 4.4.
The Square-root of AVE (in bold) and Correlations among Constructs of the Pilot Study
KM_H KM_S I_PD I_PC BP_C BP_F
KM_H .787
KM_S .694 .867
I_PD .461 .401 .862
I_PC .529 .588 .685 .927
BP_C .275 .343 .327 .587 .799
BP_F .208 .210 .091 .329 .701 .944
Hypothesis Testing of Pilot Study
Results were obtained from an examination of the structural relation model. SmartPLS 3.0 was used in the pilot study due to the software’s ability to handle small samples effectively.
Beta coefficients and their t-values are presented in Table 4.5 below. Besides p-values, t-values are also intended for determining the significance of path coefficients. T-values for two-tailed test higher than 1.65, 1.96 and 2.58 represent weak, moderate and strong relationships. The results suggest that KM does not have a significant effect on BP (β = .046, t < 1.65); KM has a positive, significant effect on Innovation (β = .606, t = 6.669); and Innovation has a positive, significant effect on BP (β = .497, t = 2.599). The fact that KM does not have a significant effect on BP is potentially because Innovation has fully mediated KM’s effects on BP, therefore, weakening the significance of the path coefficient from KM to BP (see Figure 4.1 for graphic illustration). The structural model explained 36.7% variance of Innovation (R2 = 0.367) and 27.6% of Business performance (R2 = 0.276).
Table 4.5.
Hypothesis Testing Results of the Pilot Study
Hypothesis β-path t-value Direction Results
KM BP H1 .046 0.198 + Rejected
KM Innovation H2 .606*** 6.669 + Accepted
Innovation BP H3a, H3b .497*** 2.599 + Accepted
Note. *** p < .001
Main Study Descriptive Statistics
The study was conducted within IT industry. A total of 230 responses were collected. Out of them, 219 were valid, representing 95% valid response rate. Characteristics of the respondents are listed in Table 4.6, which are divided into gender and marital status. In terms of gender, there is a higher percentage of male respondents (68.9%), female respondents account for 31.1%. In terms of marital status, there is a higher percentage of employees who are single (65.8%), married employees account for 33.8%, 0.5% of respondents selects ‘Other’ for Marital status.
Table 4.6.
Sample Characteristics
Variable Entries Percentage
Gender Male 151 68.9%
Female 68 31.1%
Marital status Single 144 65.8%
Married 74 33.8%
Others 1 0.50%
Note. n = 219
Items’ means and standard deviations are shown in Table 4.7 Table 4.11 contains correlations among variables, the highest correlation is between I_PC and I_PD (.782), none of the correlations exceeds 0.8 threshold (Kennedy, 1985), therefore, multi-collinearity is not present.
Table 4.7.
KMO and Bartlett’s test of sphericity. Table 4.8 demonstrates KMO and Bartlett's test of sphericity values for every construct. All the values reach the cut-off criteria (KMO > 0.6, Bartlett's test of sphericity p-values < .001), thus, a following CFA can be conducted.
Table 4.8. loading is below 0.4, it means the indicator does not well represent its parent construct, therefore, should be withdrawn from further analysis. Table 4.9 shows factor loadings of all constructs.
Different from the pilot study, in the main study, item km_h4 has factor loading of higher than 0.4, hence, the item was retained.
Construct Reliability. As shown in table 4.10, the lowest Cronbach’s alpha value is 0.78, and the lowest Composite Reliability value is 0.87, thus, satisfying the cut-off criterion of 0.7.
Table 4.10. range from 0.63 to 0.93 (Table 4.10), which exceed the cut-off criteria of 0.7 and 0.5 respectively, which ensures convergent validity. Table 4.11 shows that the square-root of AVE of each construct (in the diagonal) is higher than its correlations with other constructs (in the corresponding rows and columns), thus, confirm discriminant validity.
Table 4.11.
The Square-root of AVE (in bold) and Correlations among Constructs
KM_H KM_S I_PD I_PC BP_C BP_F
After the measurement scales’ reliability and validity are confirmed, the next step is using fit indices to evaluate the research model. The fit indices will help determine how adequately the model explains the data. The model fit statistics are summarized in Figure 4.1. Fit indices show that the model was adequate and reasonably fitted. Although the large value of chi-square (χ2 =
342.688) suggests an inadequate fit, conclusion can’t be made without taking other fit indices into consideration. Moreover, researchers don’t favor chi-square because it is an absolute fit index that is affected by sample size and model size (Newsom, 2015). Chi-square values tend to be large as sample size is large (over 200) or model size expands (Newsom, 2015). Other fit indices such as Bentler’s Comparative Fit Index (CFI), Standardized Root Mean Square Residual (SRMR), Root Mean Square Error of Approximation (RMSEA) provide more details about the model and should be reported along with chi-square. According to Hu and Bentler’s (1999), Type I and Type II error are best minimized when using the combination of CFI and SRMR with values greater than .95 and smaller than .08 (or .09) respectively. The model’s CFI is .951, and SRMR is .053, thus, satisfying the cut-off criteria. A RMSEA cut-off value close to .07 is also recommended for a good fit of model. The model’s RMSEA is .065, indicating a good fit of the model. Summarized in Figure 4.1, the model’s fit indices (CFI = .951, SRMR = .053, RMSEA
= .065) satisfy all the mentioned cut-off values, indicating that the model fits the data relatively well.
Hypothesis Testing
Beta coefficients and their t-values are presented in Table 4.12 and depicted in Figure 4.1.
The structural model explained 45.2% variance of Innovation (R2 = 0.452) and 42.2% of Business performance (R2 = 0.422). KM strategies has a significant and positive direct effect on Business performance (β = .411, t = 3.030). Also, the indirect effect of KM strategies to Business performance is at β = .193, the total effect is at β = .604. Thus, Hypothesis 1 is accepted. KM strategies has a significant and positive direct effect on Innovation (β = .648, t = 4.900). Thus, Hypothesis 2 is accepted. Innovation has a significant and direct effect on Business performance (β = .297, t = 2.712). Thus, the Hypothesis 3a is accepted. For Hypothesis 3b, the procedure adopted to test Innovation’s mediating effects was from Nitzl, Roldan, and Cepeda (2016).
Innovation’s mediating effect is tested as follows:
Step 1. An indirect effect of KM on BP was established (β = .193, reported by AMOS 23.0). The value is also equal to KM’s direct effects on Innovation (β = .648) multiplied by Innovation’s direct effects on BP (β = .297).
0.193 = 0.648 x 0.297
Step 2. The indirect effect of KM on BP is tested for significance. Bootstrap Maximum Likelihood technique was employed to calculate two tailed significant of the indirect effect. The
indirect effect is statistically significant at p < .05 (β = .193). Thus, the mediating effects of Innovation is confirmed. A mediating effect is always present when the indirect effect is statistically significant (Nitzl et al., 2016).
Step 3. Type of mediation is determined by the effects’ portion mediated. Mediation can be classified into full mediation and partial mediation. VAF (variance account for) ratio can be calculated to find out what type of mediation is taking place. If the ratio is under 20 percent, (31.95% of the effects was mediated), thus, Hypothesis 3b is accepted.
Table 4.12.
Hypothesis Testing Results
Hypothesis β-path t-value Direction Results
KM BP H1 .411** 3.030 + Accepted
Figure 4.1. Research Model.
Background variables. Although demographic features are not the main concerns of the study, the background variable Gender was examined for additional info on the data. The results are shown in Figure 4.1 and Table 4.12. Gender has no significant effect on Innovation (β =-.084, t = -1.401). Gender has no significant effect on BP (β = -.066, t = -1.087), suggesting that both male and female employees have the same tendency when evaluating their companies’ levels of innovation and how well the companies’ businesses are doing.
Findings Discussion
The primary aim of this study is to examine the effects of KM strategies on Innovation and Business Performance and whether Innovation mediates the effects of KM strategies to Business Performance. Several aspects of the study findings can be discussed.
First, the strongest direct relationship in the model is between KM and Innovation as their path coefficient has the highest absolute value and are the most statistically significant (β = .648, t = 4.900). This result implies that KM is a crucial driver behind Innovation’s growth. The higher the level of KM strategies in a company is, the higher the levels of novelty and quality in products and processes are. What it means to companies is that KM strategies should be built to
Innovation
facilitate the creating, transferring, and exchanging knowledge among their employees as well as to make knowledge easily accessible by codifying and storing knowledge in forms of database, documents, manuals, etc. One example of these strategies is the use of online platform such as social media sites, cloud services, instant message apps for collaborative projects and streaming communication. In online environment, people communicate more, share more, create more contents, store more information. And not only information could be quickly accessed, business contacts are reached within seconds of need, clients are met virtually, employees can take online courses, online training programs anytime, anywhere. All of those conveniences not only help shorten time needed to search for knowledge but also enlarge the scope of knowledge which is the main input for innovation.
Second, Innovation as a mediator mediated 31.95% of the total effects of KM on Business Performance, which means around one-third of KM strategies’ effects on business performance is manifested through the rise in companies’ innovation. KM strategies enhance knowledge resources and make them available to be used for creating better, newer products to launch in the market, along with smoother, quicker internal processes to streamline production.
Innovation adds values to a company’ offerings and differentiates its business in the saturated market. Consequently, companies stay ahead of their competitors, and secure their own market share by attracting customers who take interest in their unique products. As a company’s market share grows, so does their profit. Satisfied customers and successful financial results are what largely made up a superior business performance for every organization. Besides the portion of effects that was mediated, two-third of KM strategies’ effects on Business Performance didn’t go through Innovation. This suggests the existence of other mechanisms through which KM strategies exert their impacts on business performance. For example, KM strategies might help nurture employees in a company by equipping them with knowledge on how to better carry their tasks which doesn’t necessarily lead to any innovation but may result in less mistakes in job and even higher productivity. The system-oriented KM strategy which revolve around systemizing knowledge could provide timely answers for urgent matters as database and technology are applied for quick searching of information. This application is particularly beneficial in areas such as answering customers, managing inventory, and dealing with business emergencies such as tracking lost shipments, repairing damaged equipment, accessing back-up plans, etc. With so
many aspects of business that are potentially influenced by KM strategies explained the great amount of total effects KM strategies have on business performance (β = .604, p < .01).
Finally, the results bear especially important meaning for IT industry in comparison with other industries because of IT industry’s fast-paced nature that shortens the life cycles of products, thus, requires constant developments of new products to be ahead in the market.