160 H9: Technological Competence has a positive effect on the MDM adoption by local government 4.3.10 Government Policy Policy refers to the existence of fundamental policies or standards to adopt or implement an IT innovation in organization (Awa & Ojiabo, 2016; Kuan & Chau, 2001; Lian et al., 2014; Pan & Jang, 2008). Local government organisations adopt IT innovation if there is a policy at central level for the implementation. Government Policy evaluates the existence of government fundamental policies or standards for the MDM adoption or implementation by the organization. For example, policy would influence local government in adopting Cloud Computing (Ali et al., 2016), eGovernment (Jans et al., 2016), Web Accessibility Standard (Velleman et al., 2015), Social Media (Rubin et al., 2014; Sharif et al., 2015) Smart Grid Technologies (Dedrick et al., 2014), and e-Government (Kamal, Hackney, & Sarwar, 2013). Haug et al. (2013) stated that MDM implementation would be hampered without efficient procedures and written data quality policies in the organization. Thus, we proposed that: H10: Government Policy has a positive effect on the MDM adoption by local government 4.3.11 Citizen Demand Local government organisations deal with a citizen on land and housing taxation, business licensing and rules enforcement in the certain coverage area. Citizen demand towards the IT innovation regarding these businesses would influence the IT adoption, such as in the domain of e-Services (Li & Feeney, 2014; Wang & Feeney, 2016) Web Accessibility Standard (Velleman et al., 2015), Social Media (Sharif et al., 2015), Smart Grid Technologies (Dedrick et al., 2014), and e-Government (Kamal, Hackney, & Sarwar, 2013). According to Otto, Hüner and Österle 2012, in a
161 telecommunications context, the rising of orders, complaints, contract, service quality, and billing data around individual customer information, require significant changes on the master data lifecycles as similarly as MDM implementation is requires. Thus, we hypothesized that: H11: Citizen Demand has a positive effect on the MDM adoption by local government 4.3.12 Citizen Population Density and Citizen Demand This research posited that citizen population density of each local government would affect citizen demand towards the MDM adoption. The number of citizen population served by each local government in Malaysia is retrieved from the electronic data bank of the Department of Statistics, Malaysia (DOSM, 2010). McCullough et al. (2015) and Rubin et al. (2014) proposed that citizen population density is measured by three levels; low, medium, and high, depending on the number of the citizen served by each local government. Based on the expert suggestion (Section 4.1.1, page 141), this research has classified a citizen population density of Malaysia local government into ‘low’ if less than 100,000 people, ‘medium’ if 100,000 to 3000,000 people, and ‘high’ if more than 3000,000 people served by the local government. Moderator variable is typically introduced when there is an unexpectedly weak or inconsistent relation between a predictor and a criterion variable. In this research, citizen population density is suggested as a moderator variable on the relationship between citizen demand and MDM adoption. This is due to the inconsistent relationship between citizen demand and IT adoption in the existing literature. On one hand, citizen demand has a significant effect to the adoption of e-Services (Li & Feeney, 2014; Wang & Feeney, 2016) Web Accessibility Standard (Velleman et al., 2015), Social Media (Sharif et al., 2015), Smart Grid Technologies (Dedrick et al., 2014), and e-Government (Kamal, Hackney, & Sarwar, 2013). However, on the other hand, citizen demand has insignificant effect towards the adoption of digital government (McNeal, Tolbert, Mossberger, & Dotterweich, 2003), and e-democracy
162 (Lee, Chang, & Berry, 2011). Hence, due to the inconsistency effect of the citizen demand, this research proposed citizen population density as a moderator that influence the relationship between citizen demand and MDM adoption. The higher the citizen population density of local government, the higher citizen demand on the MDM adoption for integrated government online service enablement. Citizen population density had influenced the adoption of Electronic Health Record (McCullough et al., 2015) and Social Media (Rubin et al., 2014). Thus, we proposed that: H12: The positive relationship between citizen demand and MDM adoption by Malaysia local government will be stronger when citizen population density is high 4.4 Operational Definition of the Measurement Terms It is common that definition of terms and key concepts vary form study to another depending on the research domain and context. Hence, it is important to clearly define the operational definitions of the terms and key concepts used in this research. The operational definitions for measurement terms in this research were adapted from IT adoption studies as presented in Table 4.3. Table 4.3 Operational definitions of the measurement terms Measurement Term Original Definition Operational definition in this research Technological Dimension The characteristic of the innovation relevant to the organization (Baker, 2012; Tornatzky & Fleischer, 1990). The characteristic of the MDM innovation relevant to the organization Organizational Dimension The resources characteristic (Baker, 2012; Tornatzky & Fleischer, 1990) and linking structure of the personnel of the organization (Krishnan et al., 2017; Smallwood, 2014) The resources characteristic and linking structure of the personnel of the organization which related to the MDM innovation Environmental Dimension The arena in which the organization conducts its business (Baker, 2012; Tornatzky & Fleischer, 1990) The arena in which the organization conducts its business
163 Measurement Term Original Definition Operational definition in this research Relative advantage The degree in which IT innovation could provide benefits to the organization in terms of reduced turn-around time, better customer service, reduce costs and timely information availability for decision making (Premkumar & Roberts, 1999) The degree in which MDM innovation could improve service delivery, provide better communication, reduce data management cost, providing timely decision-making, and reduce data quality issue. Complexity The degree of organization’s difficulty associated with understanding and learning to use an IT innovation (Premkumar & Roberts, 1999) The degree of organization’s difficulty to understand and implement the MDM innovation Quality of master data The degree of completeness, uniqueness, timeliness, validity, accuracy, and consistency of data (DAMA UK Working Group, 2013) The degree of completeness, uniqueness, timeliness, validity, accuracy, and consistency of master data at the local government organization Data security The degree in which inter-organizational information systems could preserve data confidentiality, integrity and availability (Soliman & Janz, 2004) The degree in which MDM innovation could preserve data confidentiality, integrity and availability Data governance The strategy and operating model used by an SME for internal activities and website (W. Hung et al., 2014) The strategy of the organization in terms of defining operation procedures, roles and responsibilities in steering the MDM innovation Top management support The degree of top management of the organization to create a supportive environment and providing adequate resources for the adoption of IT innovation (Premkumar & Roberts, 1999) The degree of top management to create a supportive environment and providing adequate financial and human resources for the adoption of MDM innovation Technological competence The degree of organization’s capability which includes IT infrastructure and human resources availability in terms of expertise, skills sufficient number of personnel to adopt and implement an IT innovation (Lin, 2006; Wang & Wang, 2016) The degree of organization’s capability which includes IT infrastructure and human resources availability in terms of expertise, skills and a sufficient number of personnel to adopt and implement the MDM innovation.
164 Measurement Term Original Definition Operational definition in this research Government policy The existence of fundamental policies or standards to adopt or implement an IT innovation in an organization (Awa & Ojiabo, 2016; Kuan & Chau, 2001; Lian et al., 2014; Pan & Jang, 2008) The existence of government fundamental policies or standards for the MDM adoption or implementation by the organization Citizen demand The extend of customer or nongovernmental stakeholder demand towards an IT innovation (Wang & Feeney, 2016) The extend of citizen demand towards the MDM innovation Citizen population density The size of the citizen population served by the organization (McCullough et al., 2015; Rubin et al., 2014) The size of citizen population number served by local government in Malaysia (Low: Less than 100,000 people, Medium: Between 100,000 and 300,000 people, High: More than 300,000 people) MDM adoption by local government The decision to adopt or implement an IT innovation (Rogers, 1995) The willingness of Malaysia local government to participate in sharing their master data to the MDM innovations 4.5 Chapter Summary This chapter discussed the development of the conceptual model. The discussion includes experts’ verifications, conceptual model, research hypotheses, and operational definition development. The conceptual model was developed based on the TOE framework by Tornatzky and Fleischer (1990), DOI theory by Rogers (1995) and Fit-Viability model by Tjan (2001). Twelve research hypotheses were constructed, and the operational definition of the measurement terms was specified. Based on the conceptual model, the following Chapter 5 describes the model validation, including initial data preparation, descriptive analysis, measurement model analysis, structural model analysis and moderation effect analysis.
165 CHAPTER 5 DATA ANALYSIS AND FINDINGS 5.1 Introduction This chapter presents the empirical findings of the research. The data analysis of model validation was conducted using the statistical technique as discussed in Chapter 3. The reporting of the data analysis is based on the widely accepted PLSSEM reporting as described by Hair et al. (2016). Figure 5.1 shows the analysis steps followed in data analysis in order to validate the conceptual model. Figure 5.1 Data analysis steps First, initial data preparation was performed including the analysis of response rate, data cleaning, non-response bias test, common method of bias test, and normality test (Section 5.2, page 166). Second, the descriptive analysis of the demographics was conducted (Section 5.3, page 171). Third, the measurement model analysis was performed to validate the outer model by examining the internal consistency reliability, convergent validity, and discriminant validity (Section 5.4.1, page 174). Fourth, the structural model analysis was conducted to assess the collinearity, path coefficient, Initial Preparation Response Rate Analysis Data Cleaning Non-response Bias Test Common Method Bias Test Normality Test Descriptive Analysis Demographic Analysis Measurement Model Analysis Internal Consistency Reliability Convergent Validity Discriminant Validity Structural Model Analysis Collinearity Assessment Path Coefficient Analysis Coefficient of Determination Effect Size Blindfolding and Predictive Relevance Moderation Effect Analysis Moderation Analysis Slope Analysis
166 coefficient of determination, effect size, and blindfolding and predictive relevance (Section 5.4.2, page 183). Fifth, since this research also aims to assess the moderating effect of citizen population on the relationship between citizen demand and MDM adoption, moderation analysis was conducted to examine this effect (Section 5.4.3, page 191). Finally, the summary of hypotheses testing is presented (Section 5.4.4, page 194). 5.2 Initial Data Preparation The initial data preparation is an essential step in the application of PLS-SEM. Upon survey completion, the data collection issues that need to be checked, such as response rate, missing data, suspicious response pattern, non-response bias, common method bias, and data distribution was performed. To address these issues, this research performed five steps during this stage, namely response rate analysis, data cleaning, non-response bias test, common method variance test, and normality test. 5.2.1 Response Rate Analysis The survey was distributed to 465 head of department units from 155 local government organizations in Malaysia. Only three departments from each local government were selected, namely Information Management Department, Business Licensing and Petty Traders Department, and Town Planning Department. These departments mainly involved in managing the master data regarding business cluster in a local government. From 465 contacted respondents, a total of 227 survey responses from 176 department units were collected equalling to a response rate of 38%. Initially, there were only 169 responses through the internet-based approach and after conducting follow-ups via paper-based approach at a later stage, the number of responses increases by 58 responses. However, 38% response rate is generally accepted according to
167 Baruch and Holtom (2008), in which the argued that the average response rate in organizational research studies is 35.7%. 5.2.2 Data Cleaning The data cleaning process was performed to examine the missing data and eliminate the suspicious response pattern from the data. This process was performed by examining the total number of questions answered and how the respondents answered the question. According to Sekaran (2016), if 25% of the questions were not fully answered, the data could be dropped. Only two data have missing values, but not more than 25% of the total questions, hence the data were retained and the missing values were replaced with mean values as suggested by Pallant (2016). In addition, data should also be dropped if respondents repeat the same answer for each question (Hair et al., 2016). Out of the 227 collected responses, three data were abolished because respondents repeat the same answer for each question and interestingly, all these three data were from the late responses. Therefore, only data from 224 respondents were left for the analysis. The 224 valid responses are considered adequate sample size since it is higher than a minimum sampling size as discussed in Section 3.6.2.1. 5.2.3 Non-response Bias Test A major concern of the survey-based study is the extent in which the validity of the results may be compromised by non-response bias. It is recommended to do the analysis on the non-respondents to check whether the results would have changed if non-respondents had responded (Creswell 2014). However, it is difficult to test the non-response bias with a limited information about the non-respondents. Hence, to handle non-response issue, Miller and Smith (1983) suggested that late respondents information can be used to represent the non-respondents. Hence, the independent
168 sample t-tests were conducted to identify differences between two groups sorted according to their returned date. The analysis of response rate shows that there was 62% of non-response rate of this research making the plausibility to conduct non-response bias test. Nonresponse bias test was conducted by comparing the early responses with the late responses. From 224 valid responses, 169 were early responses and 55 were late responses. Following Ram, Corkindale, & Wu (2014), the research conducted an independent sample t-test using SPSS for early responses versus late respondents on the five demographic characteristics, such as population density of the local government, department, designation, working experience, and data management experience. The comparison also conducted for the Likert scale answers of the survey questions. The non-response bias results in Appendix L indicates that there is no significant difference (p>0.05, two-tailed test) between early and late respondents on any of the five demographics characteristics and survey responses. Hence, it can be concluded that this research is free from non-response bias, which is a potential problem in survey-based studies. 5.2.4 Common Method Bias Test According to Richardson, Simmering and Sturman (2009), common method bias or also known as common method variance (CMV) is defined as the systematic error variance that is occurred from the measurement by the same source or method. Common method bias was concerned in the research because both independent and dependent variables were collected from the same respondents (Podsakoff et al., 2003). Typically, common techniques to test common method bias include Harman’s single factor test (Podsakoff & Organ, 1986), common latent factor (Liang et al., 2007), marker variable (Lindell & Whitney, 2001), and full collinearity VIF (Kock, 2015). Harman’s single-factor test and full collinearity variance inflation factors were chosen to examine the presence of common method bias in this research. Harman’s
169 single-factor test results from SPSS showed that the maximum co-variance, explained by one factor at 27.87% variance which is less than 50%. The result indicates that the data did not have any common method bias problem (Podsakoff & Organ, 1986). On the other hand, all VIF results from the full collinearity VIF test using SmartPLS 3.0 were lower than 3.3, demonstrating that the data is free from common method bias (Kock, 2015). Appendix M shows the common method bias results for both Harman’s single-factor test and full collinearity VIF. The results from both approaches demonstrate that there is no common method bias problem occurred in this research; even though the same respondents were used to answer both independent and dependent variables. 5.2.5 Normality Test Normality test examines the distribution of the data. It is important to determine the statistical analysis to be performed whether parametric or nonparametric. There are three methods commonly used for normality test, namely the graphic methods, formal method or statistical methods. Graphic methods include a histogram, distribution plot, and Q-Q plot to present results graphically. However, there are arguments on graphic methods because they may not give accurate results, whether they are normally distributed or not (Field, 2009). Formal methods namely Shapiro-Wilk and Kolmogorov-Smirnov tests commonly have an issue of showing bias on the result especially for a large sample size. (Pallant, 2016; Tabachnick & Fidell, 2013). While, the statistical methods involve the analysis of skewness and kurtosis values and they are appropriate to test the data normality if the sample size is larger than 200 (Tabachnick & Fidell, 2013). This research performed skewness and kurtosis for the normality test because the data exceeds 200. Skewness refers to the direction of the distribution of data either positive, negative or symmetrical (Pallant, 2016). On the other hand, kurtosis shows the shape of the data distribution curve either peak or flat. Normality of data can be achieved if the value of skewness and kurtosis is 0 (Pallant, 2016). However, to estimate the normal distribution of data, the skewness and kurtosis values should be
170 within the range of -3 and +3 (Tabachnick & Fidell, 2013) or between -1 and +1 (Hair et al., 2016). Following Ramayah et al. (2017), we analysed the multivariate skewness and kurtosis using the tool available at https://webpower.psychstat.org/models/kurtosis/. Figure 5.2 shows the normality test results. Normal distribution of data is determined by the value of z obtained through the distribution of statistical value over the standard deviation and kurtosis error (Hair et al., 2016). Figure 5.2 Normality test result The normality test result shows that that the z value of Mardia’s skewness and kurtosis is not multivariate normal, because it is not between -3 and +3. Mardia’s multivariate skewness (β = 29.94, p < 0.01) and Mardia’s multivariate kurtosis (β = 139.01, p < 0.01). This indicated that the data is abnormal. Thus, we proceeded the data analysis to use SmartPLS 3.0 which is a non-parametric analysis software. This is because PLS-SEM has a linear regression feature that is less influenced by the diversion of the normal distribution of data (Gefen et al., 2000). Therefore, PLS-SEM analysis can be done even if the data is normal or abnormal.
171 5.3 Descriptive Analysis Descriptive analysis was performed against the 224 valid demographic profiles data. Demographic analysis is important to quantify the characteristic of the respondents of this research. It also important to achieve stratified sampling condition. 5.3.1 Demographic Analysis Descriptive analysis on demographic profiles examines the respondents’ demographic characteristics by calculating the frequency and percentage of each characteristic category (Pagano, 2012). Demographic characteristics involved in the analysis were citizen population, type of local government, departments, designation group, working experience in local government and working experience in data management. Table 5.1 presents the result of demographics analysis. Table 5.1 Demographic analysis results Demographic Category Frequency Percentage (%) Type of Local Governments City Council 46 20 Municipal Council 88 40 District Council 66 29 Special Council 24 11 Citizen Population Density High (more than 3000,000 people) 68 30 Medium (100,000 – 3000,000 people) 74 33 Low (less than 100,000 people) 82 37 Departments IT 122 55 Business Licensing and Petty Traders 58 25 Town Planning 44 20 Designation Group Top management 31 14 Executive 99 44
172 Demographic Category Frequency Percentage (%) Support group 94 42 Working experience in Local Governments More than 10 years 118 53 6 – 10 years 47 21 1 – 5 years 55 25 Less than 1 year 4 2 Working experience in data management More than 10 years 109 49 6 – 10 years 54 24 1 – 5 years 50 22 Less than 1 year 11 5 This research obtained 224 valid respondents. The respondents work in city councils, municipal councils, district councils, and special councils by 46 (20%), 88 (40%), 66 (29%), and 24 (11%) respectively. The majority of respondents (37%) work in local government that have low citizen population coverage (less than 100,000 people), followed by 33% in medium citizen population coverage (100,000 – 3000,000 people), and 30% in high citizen population coverage (more than 3000,000 people). In terms of the department, the respondents work in IT, Business Licensing and Petty Traders, and Town Planning departments, 122 (55%), 58 (25%), 44 (20%) respectively. The majority of respondents (44%) are executives, followed by 42% from the support group, and 14% from top management. The respondents have been working in local government for more than 10 years (53%), between six to ten years (21%), between one and five years (25%), and less than 1 year (2%). In terms of experience in data management, the majority of the respondents (49%) have more than ten years of experience, (24%) have between six and ten years of experience, (22%) have between one to five years of experience, and (5%) have less than one year of experience. Based on the results from demographic analysis, the stratified random sampling designed for this research is fulfilled (Section 3.6.2.1). The results show that the respondent numbers exceed the stratified sampling condition for each stratum
173 (citizen population density). The respondents of each stratum were 68, 74, and 82, which is above the stratified sampling condition of 19, 42, and 66 department units. 5.4 Model Validation using PLS-SEM To validate the conceptual model, PLS-SEM approach by Hair et al. (2016) was employed which involve the measurement model, structural model, and moderation effect analysis. Measurement model analysis evaluates the reliability and validity of the relationships between the measurement items and the constructs. Structural model analysis validates the relationship between the constructs in relation to hypothesis testing. While moderation effect analysis validates the moderating effect of the moderator variable on the relationship between independent variable and dependent variable. Figure 5.3 presents the overall measurement model and structural model specification of determinants affecting MDM adoption by local government by SmartPLS 3.0. Figure 5.3 Overall measurement and structural model for PLS-SEM analysis
174 5.4.1 Measurement Model Analysis The objective of the measurement model analysis was to examine the measurement items reliability and validity of each construct. The measurement model was analysed through a reflective measurement model assessment, namely internal consistency reliability, convergent validity and discriminant validity (Hair et al., 2016). To do the measurement model analysis, the moderator variable (i.e. citizen population) was excluded in the model specification because it was measured using single item. Figure 5.4 illustrates the model specification for measurement model analysis in Smart PLS. Figure 5.4 Measurement model specification in SmartPLS The result of internal consistency reliability, convergent validity, and discriminant validity are the most important metrics for measurement model in PLSSEM to evaluate the relationship between the measurement items and the construct. Table 5.2 shows the assessments of measurement model analysis and the value of threshold limit need to be achieved, which used in this research.
175 Table 5.2 Measurement model assessment and threshold limit Measurement model assessment Criterion Threshold limit Internal consistency reliability Composite Reliability Accept values between 0.7 and 0.95, eliminate construct which has value less than 0.7 or above 0.95 (Hair et al., 2016) Convergent Validity Indicator Reliability or Outer Loading Accept values equal or above 0.5, eliminate measurement item which has value less than 0.5 (Hair et al., 1998) Average Variance Extracted (AVE) Accept values above 0.5 (Fornell & Larcker, 1981; Hair et al., 2016) Discriminant validity Fornell-Lacker The square root of AVEs is larger in all cases than the off-diagonal elements in their corresponding row and column (Chin, 1998). Cross-loadings The loading of each item should be higher than all cross-loadings (Fornell and Larcker 1981). HTMT Accept the value less than 0.85 (Kline 2011) or 0.90 (Gold, Malhotra, and Segars 2001) 5.4.1.1 Internal Consistency Reliability Internal consistency reliability examines the measurement items of a construct by analysing the inter-correlations among them. Cronbach’s alpha is used to as a criterion to evaluate the internal consistency reliability, however, due to Cronbach’s alpha limitation, this research used composite reliability. The limitation of the Cronbach’s alpha is that it assumes that all measurement items have equal outer loadings on the construct. The composite reliability varies between 0 and 1, where higher values indicating higher levels of reliability. However, composite reliability value above 0.95 is not appropriate because they show that all measurement items are measuring the same phenomenon and are hence implausible valid measure of the construct. The threshold limit by Hair et al. (2016), which is to accept values between 0.7 and 0.95 was applied in this research. Table 5.3 present the composite reliability
176 for each construct which ranged from 0.886 to 0.946, indicating that all measurement items are reliable to measure the constructs. Table 5.3 Internal Consistency Reliability 5.4.1.2 Convergent Validity Convergent validity is an important assessment to ensure that each measurement item only measuring its own construct and not measuring other constructs (Urbach & Ahlemann, 2010). The convergent validity was assessed by using indicator reliability also known as outer loadings, and average variance extracted or AVE in-short (Hair et al., 2016). Good convergent validity indicates that the outer loading of each measurement item is above 0.5 (Hair et al., 1998), thus measurement items below 0.5 should be eliminated. Based on the results presented in Table 5.4, outer loadings for all measurement items exceeded 0.5, hence, none of the measurement items were deleted. AVE is the average amount of the squared loadings of the measurement items associated with the construct (Hair et al., 2016). Convergent validity is sufficient if the value of AVE is equal or higher than 0.5 (Fornell & Larcker, 1981; Hair et al., 2016) indicating that the construct explains more than half of the variance of its measurement items. Table 5.4 presents the results of AVE values which ranged from 0.622 to 0.850, Constructs Composite Reliability Constructs Composite Reliability Relative Advantage (T_RA) 0.891 Top Management Support (O_TS) 0.939 Complexity (T_CX) 0.926 Technological Competence (O_TC) 0.926 Quality of Master Data (T_DQ) 0.925 Government Policy (E_GP) 0.928 Data Security (T_DS) 0.942 Citizen Demand (E_CD) 0.886 Data Governance (O_DG) 0.903 MDM Adoption (MA) 0.946
177 and hence more than 0.5. The result from both outer loadings and AVE values in this research demonstrated that the measurement model conforms the convergent validity assessment. Table 5.4 Convergent validity results Construct Item Outer Loading AVE Relative Advantage (T_RA) RA1 0.807 0.622 RA2 0.814 RA3 0.647 RA4 0.816 RA5 0.843 Complexity (T_CX) CX1 0.802 0.763 CX2 0.822 CX3 0.937 CX4 0.917 Quality of Master Data (T_DQ) DQ1 0.818 0.675 DQ2 0.635 DQ3 0.869 DQ4 0.817 DQ5 0.890 DQ6 0.873 Data Security (T_DS) DS1 0.935 0.798 DS2 0.942 DS3 0.937 DS4 0.894 DS5 0.744 Data Governance (O_DG) DG2 0.827 0.651 DG3 0.859 DG4 0.749 DG5 0.813
178 DG6 0.781 Top Management Support (O_TS) TS1 0.910 0.794 TS2 0.890 TS3 0.881 TS4 0.882 Technological Competence (O_TC) TC1 0.788 0.648 TC2 0.871 TC3 0.828 TC4 0.891 TC5 0.857 TC6 0.824 TC7 0.511 Government Policy (E_GP) GP1 0.836 0.722 GP2 0.889 GP3 0.780 GP4 0.872 GP5 0.868 Citizen Demand (E_CD) CD1 0.836 0.661 CD2 0.864 CD3 0.749 CD4 0.798 MDM Adoption (MA) MA1 0.905 0.850 MA3 0.928 MA4 0.925 MA5 0.928 MA6 0.924
179 5.4.1.3 Discriminant Validity Discriminant Validity is important to verify that each construct is unique and observe the phenomenon which is not presented by other constructs in the model. It is to verify that the construct is truly distinct from other constructs (Hair et al., 2016). In this research, discriminant validity was examined by three assessment outcomes, namely Fornell-Lacker, cross-loadings, and the heterotrait-monotrait ratio of correlations (HTMT) (Ab Hamid, Sami, & Sidek, 2017). First, the correlations between the items of potentially overlapping constructs according to (Fornell and Larcker, 1981) were assessed. Table 5.5 shows that the square root of AVE are larger in all cases than the off-diagonal elements in their corresponding row and column, suggesting that the required discriminant validity has been achieved (Chin, 1998). Table 5.5 Fornell-Lacker analysis result E_CD E_GP MA O_DG O_TC O_TS T_CX T_DQ T_DS T_RA E_CD 0.813 E_GP -0.506 0.85 MA 0.496 -0.372 0.922 O_DG 0.181 -0.165 0.371 0.807 O_TC 0.457 -0.697 0.431 0.252 0.805 O_TS 0.185 -0.125 0.548 0.070 0.158 0.891 T_CX -0.074 0.042 -0.101 0.090 -0.004 -0.087 0.871 T_DQ 0.377 -0.298 0.506 0.483 0.355 0.078 0.147 0.822 T_DS 0.36 -0.581 0.183 -0.048 0.546 0.073 0.037 0.285 0.894 T_RA 0.226 -0.379 0.307 0.235 0.375 0.249 -0.032 0.265 0.420 0.789 Second, the cross-loadings assessment was performed. The cross-loadings condition proposes that the loading of each item should be higher than all cross-
180 loadings (Fornell and Larcker, 1981). Table 5.6 shows that all the loadings are greater than the correspondent cross-loadings, except TC7. Hence, TC7 was deleted from the model before conducting the structural analysis. Table 5.6 Cross-loading analysis result E_CD E_GP MA O_DG O_TC O_TS T_CX T_DQ T_DS T_RA CD1 0.836 -0.407 0.466 0.091 0.385 0.173 -0.133 0.308 0.179 0.106 CD2 0.864 -0.44 0.38 0.175 0.377 0.127 -0.025 0.317 0.375 0.169 CD3 0.749 -0.336 0.373 0.204 0.295 0.161 -0.062 0.344 0.283 0.252 CD4 0.798 -0.463 0.382 0.134 0.425 0.134 -0.006 0.259 0.36 0.225 GP1 -0.382 0.836 -0.201 -0.062 -0.615 -0.102 0.057 -0.184 -0.568 -0.404 GP2 -0.462 0.889 -0.431 -0.174 -0.686 -0.093 0.07 -0.341 -0.555 -0.314 GP3 -0.435 0.78 -0.325 -0.192 -0.409 -0.153 0.08 -0.188 -0.297 -0.269 GP4 -0.44 0.872 -0.286 -0.109 -0.605 -0.058 -0.016 -0.265 -0.544 -0.272 GP5 -0.397 0.868 -0.231 -0.112 -0.648 -0.129 -0.047 -0.228 -0.532 -0.406 MA1 0.451 -0.335 0.905 0.353 0.363 0.491 -0.105 0.416 0.134 0.231 MA3 0.457 -0.297 0.928 0.346 0.333 0.565 -0.09 0.476 0.116 0.279 MA4 0.454 -0.273 0.925 0.326 0.312 0.531 -0.163 0.388 0.113 0.228 MA5 0.449 -0.477 0.928 0.334 0.512 0.485 -0.031 0.538 0.296 0.34 MA6 0.478 -0.331 0.924 0.355 0.463 0.455 -0.083 0.509 0.179 0.332 DG2 0.127 -0.168 0.226 0.827 0.145 -0.045 0.161 0.374 -0.049 0.116 DG3 0.175 -0.225 0.371 0.859 0.331 0.09 0.124 0.444 -0.006 0.207 DG4 0.111 -0.077 0.157 0.749 0.062 0.047 0.103 0.437 -0.004 0.032 DG5 0.185 -0.178 0.313 0.813 0.275 0.056 -0.044 0.327 -0.002 0.211 DG6 0.112 0.003 0.331 0.781 0.105 0.088 0.055 0.385 -0.122 0.275 TC1 0.373 -0.598 0.33 0.02 0.788 0.133 0.077 0.162 0.496 0.271 TC2 0.438 -0.59 0.436 0.178 0.871 0.191 -0.062 0.26 0.491 0.279 TC3 0.351 -0.613 0.245 0.016 0.828 0.141 0.014 0.223 0.419 0.258 TC4 0.43 -0.653 0.325 0.165 0.891 0.128 0.011 0.35 0.51 0.377
181 E_CD E_GP MA O_DG O_TC O_TS T_CX T_DQ T_DS T_RA TC5 0.387 -0.534 0.325 0.177 0.857 0.109 0.001 0.289 0.485 0.286 TC6 0.36 -0.59 0.225 0.062 0.824 0.077 -0.054 0.21 0.487 0.233 TC7 0.187 -0.326 0.401 0.643 0.511 0.69 -0.004 0.426 0.17 0.344 TS1 0.183 -0.101 0.561 0.062 0.139 0.91 -0.132 0.012 0.039 0.218 TS2 0.167 -0.202 0.479 0.016 0.19 0.89 -0.021 0.105 0.106 0.221 TS3 0.183 -0.064 0.484 0.104 0.122 0.881 -0.18 0.1 0.078 0.213 TS4 0.115 -0.074 0.409 0.069 0.108 0.882 0.049 0.07 0.034 0.238 CX1 -0.035 -0.017 -0.045 0.119 0.158 -0.103 0.802 0.219 0.071 0.04 CX2 -0.012 0.054 -0.055 0.142 0.031 -0.092 0.822 0.224 0.012 -0.064 CX3 -0.102 0.04 -0.123 0.04 -0.081 -0.064 0.937 0.057 0.029 -0.07 CX4 -0.067 0.05 -0.091 0.077 0 -0.076 0.917 0.131 0.031 0.017 DQ1 0.317 -0.279 0.436 0.508 0.319 0.048 0.089 0.818 0.232 0.317 DQ2 0.148 -0.065 0.236 0.265 0.07 0.182 0.196 0.635 0.186 0.23 DQ3 0.311 -0.308 0.409 0.452 0.329 0.102 0.128 0.869 0.277 0.225 DQ4 0.287 -0.319 0.41 0.345 0.382 -0.024 0.11 0.817 0.216 0.172 DQ5 0.319 -0.223 0.42 0.375 0.215 0.048 0.113 0.89 0.231 0.149 DQ6 0.408 -0.223 0.515 0.403 0.349 0.083 0.129 0.873 0.255 0.227 DS1 0.328 -0.533 0.197 -0.05 0.526 0.107 0.032 0.253 0.935 0.43 DS2 0.278 -0.534 0.137 -0.028 0.542 0.065 0.003 0.287 0.942 0.433 DS3 0.318 -0.579 0.163 -0.075 0.551 0.069 -0.019 0.247 0.937 0.408 DS4 0.378 -0.486 0.155 -0.041 0.438 0.011 0.047 0.256 0.894 0.332 DS5 0.299 -0.454 0.151 -0.015 0.371 0.06 0.102 0.231 0.744 0.26 RA1 0.231 -0.274 0.317 0.155 0.258 0.252 0.035 0.173 0.314 0.807 RA2 0.097 -0.22 0.155 0.07 0.131 0.195 -0.06 0.134 0.301 0.814 RA3 0.071 -0.174 0.161 0.279 0.312 0.133 -0.022 0.221 0.273 0.647 RA4 0.118 -0.296 0.209 0.24 0.282 0.086 0.001 0.172 0.314 0.816 RA5 0.273 -0.453 0.285 0.195 0.438 0.258 -0.095 0.317 0.426 0.843
182 Third, the discriminant validity was also examined through the most recent analysis using HTMT as recommended by Henseler, Ringle and Sarstedt (2015). HTMT requires the calculation of a ratio of the average correlations between constructs to the geometric mean of the average correlations within items of the same constructs. The HTMT assessment validates whether HTMT ratio approaches 1.0. If exceed, the discriminant validity of the construct is interpreted as a violation. HTMT ratio was compared to a predefined threshold. If the value of the HTMT is higher than this threshold, it means that there is a lack of discriminant validity. Kline (2011) proposes a threshold of 0.85 and Gold, Malhotra and Segars (2001) suggest a value of 0.90. Table 5.7 shows all HTMT ratios were less than those threshold limits 0.85 and 0.90 for all constructs. Overall, the measurement model demonstrated the adequate convergent validity and discriminant validity. Table 5.7 HTMT analysis result E_CD E_GP MA O_DG O_TC O_TS T_CX T_DQ T_DS T_RA E_CD E_GP 0.574 MA 0.554 0.371 O_DG 0.212 0.182 0.378 O_TC 0.521 0.772 0.44 0.267 O_TS 0.208 0.142 0.58 0.1 0.168 T_CX 0.081 0.085 0.105 0.152 0.11 0.124 T_DQ 0.423 0.304 0.53 0.544 0.365 0.12 0.209 T_DS 0.418 0.636 0.19 0.072 0.594 0.077 0.071 0.311 T_RA 0.26 0.423 0.316 0.265 0.401 0.266 0.095 0.297 0.461 ▪ HTMT.85 < 0.85 ▪ HTMT.90 < 0.90 ▪ HTMT.inference < 1.00 (Significant)
183 5.4.2 Structural Model Analysis After the measurement model analysis meets the reliability and validity conditions, the structural model analysis was conducted. The structural model analysis examines the strength of the relationship between constructs, effect size and predictive relevance of the model. The structural model analysis examines the direct effect of the independent variables and dependent variable. When doing the structural model analysis, the moderator variable (i.e. citizen population) was excluded in the model specification. Figure 5.5 shows the model specification for structural model analysis in Smart PLS 3.0. Figure 5.5 Structural model specification in SmartPLS The result of collinearity, path coefficient, the coefficient of the determinant (R2 ), effect size (f2 ), predictive relevance Q2 , and effect size (q2 ) are the most important metrics for the structural model in PLS-SEM when evaluating the relationship among the constructs. Table 5.8 shows the assessments in structural model analysis and the value of threshold limit that need to be achieved, which used in this research.
184 Table 5.8 Structural model assessment and threshold limit Structural model assessment Criterion Threshold limit Collinearity Variance Inflation Factor (VIF) Accept value equal or less than 5, eliminate from the model or merged into a single construct if more than 5 (Hair et al., 2016) Path Coefficient Path coefficient (β) Accept if t-value >1.65 and p-value < 0.05, reject if tvalue <1.65 and p-value > 0.05 (Hair et al., 2016). Coefficient of determinant R 2 R 2 value around 0.75 is high, around 0.50 is moderate, around 0.25 and below is weak (Hair, Ringle, & Sarstedt, 2011; Henseler, Ringle, & Sinkovics, 2009) Effect size f 2 f 2 value around 0.35 is high, around 0.15 is moderate, around 0.02 and below is weak (Gotz, Liehr-Gobbers, & Krafft, 2010) Blindfolding and Predictive Relevance Q2 Q2 value around 0.35 is high, around 0.15 is moderate, around 0.02 and below is weak (Hair et al., 2016) Effect size q 2 q 2 value around 0.35 is high, around 0.15 is moderate, around 0.02 and below is weak (Hair et al., 2016) 5.4.2.1 Collinearity Assessment Collinearity assessment is the essential step in validating structural model before proceeding to analyse the path coefficient. Collinearity refers to the problem when independent variables in a model are linearly related to each other (Dormann et al., 2013). According to Hair et al. (2016), collinearity can reduce the predictive power of independent variables. Independent variables that have a high collinearity relationship are not reliable to explain the dependent variable in the model. Collinearity can be assessed through VIF (Hair et al., 2016). Each VIF value should be lower than 5, otherwise, the construct should be eliminated from the model or merged into a single construct (Hair et al., 2016). Table 5.9 presents the results of
185 collinearity analysis t of the model, demonstrating that collinearity was not exhibited in any of the constructs in the model since all VIF values were less than 5. Table 5.9 Collinearity analysis result Constructs VIF Constructs VIF T_RA 1.420 O_TS 1.104 T_CX 1.060 O_TC 2.260 T_DQ 1.615 E_GP 2.385 T_DS 1.972 E_CD 1.524 O_DG 1.529 5.4.2.2 Path Coefficient Analysis The path coefficient analysis is also known as hypotheses testing of the model. The path coefficient (β) estimates the relationship between exogenous and endogenous variables based on algebraic symbols (negative or positive), magnitude or standard values (-1 to +1) and significant value levels (Urbach & Ahlemann, 2010). Algebraic symbols are the sign of relationships between exogenous and endogenous constructs either positive or negative in the hypothesis. The path coefficient (β) is insignificant if the output results show an opposite symbol for the hypothesis of the study and vice versa (Gotz et al., 2010; Urbach & Ahlemann, 2010). The magnitude or standard value refers to the strength of the relationship between exogenous and endogenous constructs. The value of the path coefficient (β) which is approximately +1 shows a strong positive relationship and likewise for a value approaching -1 shows a strong negative relationship (Hair et al., 2016). The path coefficient (β) ultimately depends on the standard error obtained through the bootstrap function also known as bootstrapping. Standard Error (SE) from bootstrapping allows the computation of t-values and p-values for all path coefficients. When t-value is higher than the critical value, it is concluded that the coefficient is
186 statistically significant at a certain significant value level. The critical value used in this research is 1.65 and 0.05 significant value level (Hair et al., 2016). In other words, the hypothesis is accepted if the t-value is greater than 1.65 (one-tailed) and the output of the p-value is less than the significant value level of 0.05. Bootstrap of 5000 samples as suggested by Hair et al. (2014) to provide consistent value. Table 5.10 presents the path coefficient or hypotheses testing results. Table 5.10 Path coefficient results Hypothesis Relationship Path Coefficient (β) SE t-value p-value Decision H1 T_RA -> MA 0.018 0.049 0.370 0.356 Rejected H2 T_CX -> MA -0.096 0.051 1.905 0.029 Accepted H3 T_DQ -> MA 0.323 0.055 5.86 0 Accepted H4 T_DS -> MA -0.125 0.085 1.475 0.07 Rejected H5 O_DG-> MA 0.117 0.053 2.214 0.014 Accepted H6 O_TS -> MA 0.444 0.048 9.225 0 Accepted H7 O_TS –> O_DG 0.069 0.07 0.992 0.161 Rejected H8 O_TS -> O_TC 0.158 0.079 2.006 0.023 Accepted H9 O_TC-> MA 0.116 0.068 1.7 0.045 Accepted H10 E_GP-> MA -0.074 0.058 1.287 0.099 Rejected H11 E_CD -> MA 0.215 0.059 3.667 0 Accepted The results revealed that among the technological determinants, complexity and quality of master data have significant relationship with MDM adoption by local government with β = -0.096, p < 0.05 and β = 0.323, p < 0.01 respectively. With regard to the organizational determinants, data governance, top management support and technological competence have positive relationships with MDM adoption by local
187 government with β = 0.117, p < 0.05, β = 0.444, p < 0.01, and β = 0.116, p < 0.05 respectively. in terms of organizational determinants, only citizen demand shows a positive relationship towards MDM adoption by local government with β = 0.215, p < 0.01. Furthermore, the relationship between top management support and technological competence is significant with β = 0.158, p < 0.05. Thus, H2, H3, H5, H6, H7, H9 and H11 are accepted. Amazingly, relative advantage, data security, and government policy have no significant relationships with MDM adoption by local government with β = 0.018, p > 0.05, β = -0.125, p > 0.05, β = -0.074, p > 0.05 respectively. In addition, the relationship between top management support and data governance is insignificant with β = 0.069, p > 0.05. Thus, H1, H4, H8 and H10 are rejected. Figure 5.6 illustrates the path coefficient analysis in SmartPLS 3.0 with t-value and p-value on each relationship between exogenous and endogenous variable. Figure 5.6 Path coefficient result
188 5.4.2.3 Coefficient of Determination (R2 Value) The utmost important measure to validate the structural model is through the coefficient of determination (R2 value). R2 value explains the model’s predictive power by presenting the combination of all exogenous variable effect on the endogenous variables (Gotz et al., 2010). The R2 value is usually between 0 and 1. The higher R2 value shows that the accuracy of the model's prediction is high, while the lower R2 indicates that the accuracy of model’s predictions is low. In this research R 2 values around 0.75 is considered high, around 0.50 is considered moderate, around 0.25 and below is considered weak to determine the accuracy of model prediction (Hair et al., 2011; Henseler et al., 2009). The R2 value of the structural model in this research is 0.625 suggesting high model predictions with 62.5% of the variance in MDM adoption by local government and can be explained by the TOE determinants. 5.4.2.4 Effect size f 2 In addition to the coefficient assessment of determination (R2 value), the measure of effect size f 2 is progressively encouraged by journal editors and reviewers. Effect size f 2 is the calculation of the change in the R2 value when a specified exogenous variable is eliminated from the model. Effect size f 2 can be used to evaluate whether the eliminated variable has a substantial impact on the endogenous variable. This research applies f 2 values at around 0.35 (high), around 0.15 (moderate), around 0.02 and below (weak) to determine the effect size f 2 of exogenous towards endogenous variable (Gotz et al., 2010). Using SmartPLS, Table 5.11 presents the results of effect size f 2 of the exogenous variable to the endogenous variable. The effect size f2 result shows that the effect size of exogenous variable TS on the endogenous variable MA is high, the effect size of exogenous variable DQ on the endogenous variable MA is moderate, and the effect size of exogenous variable CX, DS, TC, and CD on the endogenous variable MA are weak.
189 Table 5.11 Effect Size f2 result Construct Effect Size f2 Effect size Construct Effect Size f2 Effect size T_RA 0 None O_TS 0.475 High T_CX 0.022 Weak O_TC 0.026 Weak T_DQ 0.169 Moderate E_GP 0.004 None T_DS 0.023 Weak E_CD 0.081 Weak O_DG 0.017 None 5.4.2.5 Blindfolding and Predictive Relevance (Q2 ) Further analysis in determining the accuracy of model’s prediction was performed using Stone-Geisser’s predictive relevance (Q2 ) assessment. Predictive relevance (Q2 ) values are estimated through blindfolding procedures to find out how the path model can predict the true phenomenon. Blindfolding technique reuses sample data by not including some data for some indicators in endogenous constructs (Hair et al., 2016). In structural models, if the predictive relevance (Q2 ) values are greater than zero for endogenous constructs means that the forecasting of a structural model of the constructs is relevant. If the relevant predictive relevance (Q2 ) is smaller than zero it means that the lack of predictive relevance. Predictive relevance value is considered small, medium, or high with a value of 0.02, 0.15 or 0.35 respectively. Using SmartPLS 3.0 algorithm calculated, based on cross-validated redundancy approach as recommended by Hair et al. (2016), the evaluation of predictive relevance (Q2) of this research presents a sufficient high predictive relevance for the endogenous variable MA with a value of 0.459.
190 5.4.2.6 Effect size (q2 ) In addition to effect size (f2 ), the effect size value (q2 ) is used to determine the effects of predictive relevance (Q2 ) when an exogenous construct is removed from the model. The effect size (q2 ) is required to measure the effects of relevant changes in the prediction of exogenous constructs on the endogenous construct. The effect size (q2 ) is estimated from the values of Q2 included and Q2 excluded. The Q2 included results from the previous blindfolding estimation are available from Q2 . The Q2 excluded value is obtained from a model re-estimation after deleting a specific predecessor of that endogenous latent variable. The following formula is used to calculate the q2 . The effect size (q2 ) value is considered low, medium, or high with a value of 0.02, 0.15 or 0.35 respectively. Table 5.12 shows that the effect size (q2 ) of all exogenous constructs on the endogenous construct (MA) is relatively small, except for exogenous construct TA which is high. Table 5.12 Effect size (q2 ) result Construct Q2 included Q2 excluded q 2 T_RA 0.459 0.465 0 T_CX 0.459 0.457 0.013 T_DQ 0.459 0.421 0.089 T_DS 0.459 0.46 0.011 O_DG 0.459 0.459 0.013 O_TS 0.459 0.326 0.244 O_TC 0.459 0.459 0.007 E_GP 0.459 0.464 0.004 E_CD 0.459 0.442 0.041
191 5.4.3 Moderation Effect Analysis According to Holmbeck (1997), a moderator variable is the one that affects the relationship between two variables. In another word, the impact of independent variable on dependent variable varies according to the level of the moderator. In this research, citizen population density is hypothesized (H12) as a moderator variable that affects the relationship between citizen demand and MDM adoption by local government, as shown in Figure 5.7. The higher is the citizen population density, the higher is the relationship between citizen demand and MDM adoption by local government. Figure 5.7 Moderating effect of citizen population density To perform the moderation analysis, the model specification from the structural model analysis was extended by including citizen population as a moderator variable. Then, interaction term using two-stage approach was introduced. The two-stage approach was chosen primary to reveal the significance of a moderating effect (Hair et al., 2016). Figure 5.8 presents the model specification with the interaction term for moderating effect analysis in SmartPLS.
192 Figure 5.8 Model specification for moderating effect analysis in SmartPLS The moderating effect of citizen population (CP) on the relationship between citizen demand (E_CD) and MDM adoption by local government (MA) was performed using SmartPLS 3.0 as shown in Figure 5.9. The result of the bootstrapping procedure with 5,000 bootstrap samples using one-tailed testing is presented. The result shows a p-value of 0.010 for the path interaction term CD*MA. Using 0.05 significant value level (Hair et al., 2016), thus, it can be concluded that the moderation effect is significant. In addition, the significance of the moderation effect was to analyse the size of the moderating effect. The moderation effect analysis in this research shows that the interaction term has a positive effect on MA at 0.100, while the simple effect of E_CD to MA is 0.192. These results propose that the relationship between E_CD and MA is at 0.192 for a moderate level of CP. For higher levels of CP (e.g. CP is increased by one standard deviation unit), the relationship between CD and MA increases by the size of the interaction term (i.e., 0.192 + 0.100 = 0.292). In contrast, for lower levels
193 of CP (e.g. CP is decreased by one standard deviation point), the relationship between CD and MA becomes lower (0.192 - 0.100 = 0.092). Figure 5.9 Moderation effect analysis result Next, slope analysis was used to better comprehend the results of the moderator analysis. The three lines shown in Figure 5.10 represent the relationship between E_CD (x-axis) and MA (y-axis). The middle line represents the relationship for an average level of the moderator variable CP. The other two lines represent the relationship between E_CD and MA for higher (i.e., mean value of CP plus one standard deviation unit) and lower (i.e., mean value of CP minus one standard deviation unit) levels of the moderator variable CP. As it can be seen, the relationship between E_CD and MA is positive for all three lines as indicated by their positive slope. Hence, higher levels of citizen population density are parallel with higher levels of citizen demand. The moderating effect’s slope was analysed in greater detail. The upper line, which represents a high level of the moderator CP, has a flatter slope while
194 the lower line, which represents a low level of the moderator CP, has a steeper slope. According to thumb condition, the slope of the high level of the moderator CP is the simple effect (i.e. 0.192) plus the interaction effect (0.100), while the slope of the low level of the moderator CP is the simple effect (i.e., 0.192) minus the interaction effect (0.100). Hence, the simple slope plot supports our previous discussion of the positive interaction term. Figure 5.10 Simple slope analysis of moderation effect in SmartPLS The final assessment of the moderation effect analysis includes the identification of moderator’s effect size f 2 . Using SmartPLS 3.0 algorithm, it shows that the interaction term’s f 2 effect size of CP*E_CD has a value of 0.027. According to (Gotz et al., 2010), this value indicates a weak effect. Overall, this result demonstrates that CP is significantly and positively affecting the relationship between CD and MA, indicating that hypothesis H12 is accepted, the positive relationship between citizen demand and MDM adoption by Malaysia local government will be stronger when citizen population density is high. 5.4.4 Summary of Hypothesis Testing In total twelve hypothesized relationships were tested in this research. Eight hypotheses (H2, H3, H5, H6, H8, H9, H11, and H12) were accepted while four hypotheses (H1, H4, H7, H10) were rejected. Table 5.13 summarises the hypotheses
195 testing results. The implications of these results are discussed further in the next chapter. Table 5.13 Hypotheses Testing Result Hypotheses Result Technological H1: Relative Advantage has a positive effect on the MDM adoption by local government Not supported H2: Complexity has a negative effect on the MDM adoption by local government Supported H3: Quality of master data has a positive effect on the MDM adoption by local government Supported H4: Data security has a positive effect on the MDM adoption by local government Not supported Organizational H5: Data governance has a positive effect on the MDM adoption by local government Supported H6: Top Management Support has a positive effect on the MDM adoption by local government Supported H7: Top Management Support has a positive effect on the Data Governance Not supported H8: Top Management Support has a positive effect on the Technological Competence Supported H9: Technological Competence has a positive effect on the MDM adoption by local government Supported Environmental H10: Government Policy has a positive effect on the MDM adoption by local government Not supported H11: Citizen Demand has a positive effect on the MDM adoption by local government Supported Moderating effect of citizen population density H12: The positive relationship between citizen demand and MDM adoption by Malaysia local government will be stronger when citizen population density is high Supported
196 5.5 Chapter Summary This chapter discussed the overall data analysis using PLS-SEM statistical technique. It began with initial data preparation followed by descriptive analysis of the demographic profile of the respondents. Then, the crucial assessments in PLS-SEM which are measurement model analysis and structural model analysis were discussed. In addition, with regards to the proposal of moderation effect of citizen population density on the relationship between citizen demand and MDM adoption by local government, moderation effect analysis was presented. Overall, eight hypotheses (H2, H3, H5, H6, H8, H9, H11, and H12) were accepted in this research, whereas four hypotheses (H1, H4, H7, H10) were rejected. In the next Chapter 6, the data analysis results were interpreted based on the hypotheses in the context of Malaysia local government.
197 CHAPTER 6 RESULTS AND DISCUSSION 6.1 Introduction This chapter discusses the model evaluation according to data analysis results from the previous chapter. Based on the survey that was conducted, the research identifies the determinants that influence the decision of the Malaysia local government organizations to participate as data providers to the MDM. This chapter discuss data analysis based on the twelve research hypotheses and the TOE dimension of technological (Section 6.2, page 197), organizational (Section 6.3, page 200), and environmental (Section 6.4, page 203). Then, the final model of the research is discussed in terms of the evolutionary stages it underwent throughout the research (Section 6.6, page 208). And finally, to evaluate the key findings of the final model to the real-phenomenon of practical world, this chapter discusses the development of guidelines and strategy of MDM adoption for the Malaysian public sector (Section 6.7, page 210). 6.2 Technological Dimension H1: Relative Advantage has a positive effect on the MDM adoption by local government Unexpectedly, relative advantage does not associate with MDM adoption by local government. This is indicated by the value of Path Coefficient (β = 0.018) and probability (p > 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). This finding is inconsistent with past studies on adoption of e-government (Kamal, Hackney, & Sarwar, 2013), enterprise application integration (Kamal, Hackney, & Ali, 2013), and social media (Sharif et al., 2015) where this factor was a
198 critical determinant affecting IT adoption in the local government context. In this research, local government organizations are interested to adopt MDM even if they do not perceive benefits from it. One possible reason which relates to the other finding of this research: that citizen demand has a significant effect on MDM adoption by local government in Malaysia. Regardless of the relative advantage of MDM, local government entities tend to adopt MDM because there is a demand from the citizens for the integrated services across public sector organizations. This relates to the major role of local government, i.e. providing public goods and services to the local citizens (Ibrahim & Karim, 2004). This finding also reveals the reason the lower MDM adoption rates in local government than expected. In this regard, Rogers (1995) stated that when the perceived benefits of the technology are high, then the innovation’s adoption rates increase. H2: Complexity has a negative effect on the MDM adoption by local government Complexity is revealed to have a significant negative effect on MDM adoption by local government. This is consistent with previous research findings on the adoption of e-services and municipal government (Lagrandeur & Moreau, 2014). This is indicated by the value of Path Coefficient (β = -0.096) and probability (p < 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). If it is complex to learn and use, MDM adoption by local government tends to be lower. As MDM has the inherent complexity of incorporating services from one platform to another (Alharbi, 2016), it requires many skilful professionals to work together. Introducing MDM to the organization is a complex process and requires numerous steps and viewpoints (Loshin, 2009). The typical practices involved data profiling, data cleansing, and data integration seem. The complexity of activities has influences the decisions of local government organizations to adopt MDM. In the Malaysia local government context, there are different educational levels among members of city councils, municipal councils, and district councils’ IT leaders. Most IT personnel managers in the city and municipal councils are bachelor or master degrees holders, while in district councils the managers have only reached diploma level. The learning
199 curve of MDM technology is higher when the educational level is lower. In total, over half of Malaysia local government entities are district councils. H3: Quality of master data has a positive effect on the MDM adoption by local government The quality of master data is found to have a positive effect on MDM adoption by local government. This is indicated by the value of Path Coefficient (β = 0.323) and probability (p < 0.01) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). This result is similar to the other findings on MDM literature, i.e. (Silvola et al., 2011). Hence, it is suggested that the higher the quality of master data in the organization, the greater the likelihood of MDM adoption. The adoption of MDM means that the organizations will provide their master data to the MDM centralized platform. These centralized data can be referred to as the authorized data to other services. If the master data on the sources are incomplete, redundant, invalid, obsolete, inaccurate, and inconsistent, the organization may face difficulties during the data cleansing process before they send the data to the MDM. Any discrepancy of the data may affect their reputation as organizations that own the data. Hence, the quality of master data on the source plays an important role in adopting MDM among local government in Malaysia. H4: Data security has a positive effect on the MDM adoption by local government Remarkably, data security is found to have no association with MDM adoption. This is indicated by the value of Path Coefficient (β = -0.125) and probability (p > 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). This result is inconsistent with previous studies in which data security has found to have a positive relationship with IT adoption in local government (Ali et al., 2016; Kamal, Hackney, & Ali, 2013; Lagrandeur & Moreau, 2014; Rubin et al., 2014). The adoption of MDM by local government involves data sharing between local government departments as data owner or data provider and the MDM initiator as data steward or
200 data custodian (Dreibelbis et al., 2008). One possible reason is that data security management is one of the essential embedded elements in the MDM services and is responsible for handling the security of the transaction channel and for protecting data from unauthorized changes and access. In addition, in Malaysian context, most of the MDM initiators are central agencies such as ministries and state government agencies, which normally have sufficient security infrastructure and appropriate security procedures for information managing. Furthermore, the Malaysian public sector also has established an initiative known as Malaysian Government Public Key Infrastructure to ensure that all electronic transactions within e-government applications go through identity verification which retains the privacy and integrity of data through highly secure data encryption (MAMPU, 2018). Therefore, the security concern is less critical in the adoption of MDM by local government since the data is shared between government entities and the security infrastructure to support the implementation is ready and reliable. 6.3 Organizational Dimension H5: Data governance has a positive effect on the MDM adoption by local government As shown in the previous chapter, the relationship between data governance and MDM adoption by local government is found to be significant. This is indicated by the value of Path Coefficient (β = 0.117) and probability (p < 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). This empirical result is consistent with findings from other research which established the significant role of governance in IT adoption (Ali et al., 2016; McCullough et al., 2015; Velleman et al., 2015; Wang & Feeney, 2016). Local government in Malaysia is generally under the exclusive purview of the state government, except in the federal territories. Nevertheless, Malaysia local government organizations are also under the management of central government, i.e. the Ministry of Urban Wellbeing, Housing, and Local Government in terms of policy coordination, technical advice, and federal budget distribution (KPKT, 2017a). Due to the complex of organizational structure,
201 data governance in MDM influences MDM adoption by local government in Malaysia with the clear identification of stakeholders, roles and responsibilities, and continuous commitment across many parties involved in MDM implementation. Most of the local government departments are reluctant to adopt MDM when there are unclear responsibilities for decision-making, procedures, and measurement of the impact assessment. In addition, continuous commitment from all stakeholders is a critical factor influencing the decisions of local government to adopt MDM. H6: Top Management Support has a positive effect on the MDM adoption by local government The findings of this research show that top management support is the most important determinant for MDM adoption in a local government context. This is indicated by the value of Path Coefficient (β = 0.444) and probability (p < 0.01) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). This is consistent with recent literature on IT adoption in the local government context in the adoption of social media (Rubin et al., 2014; Seigler, 2017), web accessibility standards (Velleman et al., 2015), smart grid technologies (Dedrick et al., 2014), enterprise application integration (Kamal, Hackney, & Ali, 2013), and e-government (Kamal, Hackney, & Sarwar, 2013). A possible reason for this scenario is that MDM implementation comprises multiple organizations cooperation, from data provider organizations to data consumer organizations. Hence, top management support is required to ensure inter-organizational commitment and encourage the personnel in the organization to adopt MDM. In addition, as local government entities in Malaysia are generally under the exclusive purview of the state government and Ministry of Urban Wellbeing, Housing and Local Government, top management of the local government will tend to adopt MDM, especially if it is initiated by the state government or by the ministry.
202 H7: Top Management Support has a positive effect on the Data Governance Top management support is revealed to have no significant relationship with the adoption of MDM by local government. This is indicated by the value of Path Coefficient (β = 0.069) and probability (p > 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). Even though strong top management support is supposed to be the first key to an effective data governance (Smallwood, 2014), it appears to be non-significant in the context of MDM implementation in Malaysia. This is because the MDM data governance already specified the roles of Data Governance Oversight Board (DGOB) committee which is responsible to guide and oversee the adoption and implementation of MDM. The representatives of the DGOB are selected from the MDM participats in government organizations. According to (Loshin, 2013), the main responsibilities of this steering committee is to provide a strategic direction for MDM data governance, approve data governance policies and procedure, and monitor the compliance of data governance policies. Hence, with this committee, the dependency of the MDM adoption on the top management of the local government organization is less. H8: Top Management Support has a positive effect on the Technological Competence As mentioned earlier, top management support is found to have positive effect on technological competence in the organization. This is indicated by the value of Path Coefficient (β = 0.158) and probability (p < 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). Top management support includes the extent to which senior managers understand the perceive the benefits of the technology, have clear vision about the application of technology in the organization, and allocate sufficient funding and other resources to technology implementation (Dong et al., 2009). On the other hand, technological competence refers to the ICT infrastructure, knowledge, skills, experience and a sufficient number of personnel to implement technology (Wang & Wang, 2016). Top management support will increase the technological competence of personnel through the budget allocation for technical
203 training and change management programs (Rezvani et al., 2017). In addition, the clear vision of MDM among senior managers leads to the prioritizing allocation of sufficient resources for MDM implementation. H9: Technological Competence has a positive effect on the MDM adoption by local government The findings of this research also confirm that technological competence has a positive association with MDM adoption in a local government context. This is indicated by the value of Path Coefficient (β = 0.116) and probability (p < 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). The finding is aligned with previous studies on IT adoption in the local government context, in particular social media (Seigler, 2017), cloud computing (Ali et al., 2016), e-services (Lagrandeur & Moreau, 2014; Wang & Feeney, 2016), web accessibility standards (Velleman et al., 2015), electronic health records (McCullough et al., 2015), smart grid technologies (Dedrick et al., 2014), enterprise application integration (Kamal, Hackney, & Ali, 2013), and e-government (Kamal, Hackney, & Sarwar, 2013). One of the reasons is that the main focus of MDM is to provide high-quality master data to the consumers. In dealing with their master data, local government depends on the knowledge and skills of internal personnel instead of outside vendors. This is due to the sensitivity of data and security reasons. In addition, technical activities such as profiling and cleansing master data require skilled and experienced personnel from the local government departments. 6.4 Environmental Dimension H10: Government Policy has a positive effect on the MDM adoption by local government The relationship between government policy and MDM adoption is found to be non-significant. This is indicated by the value of Path Coefficient (β = -0.074) and
204 probability (p > 0.05) derived from the Path Coefficient analysis (Section 5.4.2.2, page 185). This means that the government policy on MDM adoption at a central level does not influence local government to adopt MDM. This result is inconsistent with findings from previous studies on IT adoption in the local government context, in particular in cloud computing (Ali et al., 2016), e-government (Jans et al., 2016), web accessibility standards (Velleman et al., 2015), social media (Rubin et al., 2014; Sharif et al., 2015), smart grid technologies (Dedrick et al., 2014), and e-government (Kamal, Hackney, & Sarwar, 2013). However, this finding seems to be similar to a study of Government-to-Government (G2G) information sharing in China by (Fan, Zhang, & Yen, 2014) which stated that laws and policies have not been found to have a significant effect on G2G information sharing among government agencies. In this research, government policy on MDM adoption refers to the policy at the central level, which is commonly initiated by the federal government to encourage different types of government organizations (e.g. federal government, local government, state government, and government-linked companies) to provide their data to MDM initiatives. One possible explanation for this non-significant relationship may be due to the organizational structure of the Malaysia local government. Since local government organizations in Malaysia are under the exclusive purview of the state government rather than the federal government (KPKT, 2017a), the local government can choose whether to adhere to the central policies or not. In addition, although there is a policy to support MDM adoption, with limited IT resources in each local government in Malaysia, the local government organizations are reluctant to adopt MDM. This is due to MDM adoption requires highly skilful IT professionals to conduct data profiling, data cleansing, data enrichment, and data analysis (Scheidl, 2011). H11: Citizen Demand has a positive effect on the MDM adoption by local government The findings also revealed that citizen demand is important for MDM adoption by local government. This is indicated by the value of Path Coefficient (β = 0.215) and probability (p < 0.01) derived from the Path Coefficient analysis (Section 5.4.2.2, page
205 185). This finding is in line with findings from previous studies on IT adoption such as e-services (Li & Feeney, 2014; Wang & Feeney, 2016) web accessibility standards (Velleman et al., 2015), social media (Sharif et al., 2015), smart grid technologies (Dedrick et al., 2014), and e-government (Kamal, Hackney, & Sarwar, 2013). With regards to the non-significant relationship between government policy and MDM adoption in Malaysia, citizen demand might be unlikely to influence government policies since civic participation (e.g. voicing opinions, joining together to address public issues, collaborating with the government on decision-making processes to form policies) is still at the early stage. The open government index in 2015 shows that Malaysia has ranked number #90 with a score of only 37% in civic participation in government policies, which is lower than in other Asian developing countries such as Indonesia, India, Philippines, Singapore, Thailand, and Cambodia: 68%, 65%, 62%, 55%, 52%, and 42% respectively (WJP, 2015). Local government organizations tend to adopt MDM if there is high demand from citizens although they can decide whether to adhere to the MDM central policies or not. Citizens are demanding online and seamless services across government organizations, thus, failing to provide may lead to frustration. As government entities that closely deal with citizens, meeting citizen demand is a priority. Local government in Malaysia may participate in the MDM initiative by providing their master data to the MDM repository to meet the citizen demand for data collaboration and seamless services. 6.5 Moderation Effect of Citizen Population Density H12: The positive relationship between citizen demand and MDM adoption by Malaysia local government will be stronger when citizen population density is higher In this research, citizen population density is hypothesized as a moderator variable to the relationship between citizen demand and MDM adoption by local government (H12). This research indicated a significant positive moderation effect of citizen population density on the relationship between citizen demand and MDM adoption by local government. This is shown by the simple slope analysis of the citizen
206 population density with the citizen demand as illustrated in Figure 5.10, page 183. As posited, the finding revealed that there is a positive relationship between citizen demand and MDM adoption by Malaysia local government, where the relationship is stronger when citizen population density is high. In the local government context, citizen demand is revealed to have a significant effect to the adoption of e-Services (Li & Feeney, 2014; Wang & Feeney, 2016) Web Accessibility Standard (Velleman et al., 2015), Social Media (Sharif et al., 2015), Smart Grid Technologies (Dedrick et al., 2014), and e-Government (Kamal, Hackney, & Sarwar, 2013). However, on the other hand, citizen demand is also revealed to have a non-significant effect towards the adoption of digital government (McNeal et al., 2003), and e-democracy (Lee et al., 2011). This inconsistency effect is explained in this research with the discovery of the moderation effect of citizen population density on the relationship between citizen demand and the MDM adoption by local government. Local government organizations with low citizen population density are less likely to adopt MDM as compared to the local government organizations with high citizen population density. The heat map plot of citizen population density of each local government as shown in Figure 6.1 indicates that in Malaysian context, the citizen population density of local government represents the location of the local government organization whether it is in the rural or urban areas. Figure 6.1 Heat map of citizen population density of Malaysia local government organizations
207 The relationship between citizen population density and the location of the local government organization in Malaysia may shed light on why the citizen demand on MDM in urban areas is higher than in the rural areas. According to Pateman (2010), citizen demographics in urban areas are different as compared to the rural areas in terms of educational qualifications and occupation type, among others. Urban areas have higher proportions of citizens with tertiary qualifications as compared to the rural areas. The pattern of education qualification in the urban and rural areas in Malaysia could explain why citizen demand towards MDM is higher in urban areas which have a high number of citizen populations. According to (Rehman, Esichaikul, & Kamal, 2012), citizens with higher education level typically have a high demand for IT innovations. With their knowledge and exposure to the latest trend of government service around the world, they know their rights as citizens to demand the highest level of service delivered by the government. Their expectation for the seamless services across government organizations is very high and which can be achieved by MDM implementation. On the other hand, with respect to the occupation type, urban areas show more citizens working as employed workers and professionals whereas in rural areas, working from home or self-employed are far more common (Pateman, 2010). Employed workers and professionals in urban areas are typically tight with the office hours which normally eight hours per day. Integrated online services through MDM implementation by the government is demanded by the citizens because it will ease their interaction with government services without interrupting their time commitment (Mansor, 2010). This scenario is different in rural areas where self-employed workers could have flexible time to use traditional medium of service delivery such as government counters in order to get the required services (Mansor, 2010). Hence, the demand for integrated online services through MDM implementation is less in such areas.
208 6.6 Final Model This research aims to develop a new model of determinants that influence the MDM adoption by Malaysia local government organizations. Based on the literature reviews, expert verifications, and survey data collection, the research examined the determinants influencing local government in Malaysia to participate as MDM data providers by sharing their master data with the MDM platform. From the beginning of the research, the final model of this research was undergoing a series of evolutionary stages. It started with the initial conceptual model constructed based on the related works of previous literature. The initial conceptual model outlined eleven hypotheses to be tested (Figure 2.23, page 89). Then, the initial conceptual model was revised to twelve hypotheses to be tested based on the expert verifications (Figure 4.5, page 153). After that, the survey instrument was developed through content analysis and pilot test. Then, data were collected from local government organizations in Malaysia and 224 responses were analysed to validate the conceptual model and seven hypotheses were accepted to be in the final model (Figure 6.2, page 210). Overall, Table 6.1 shows the research model evolution from the initial conceptual model, the conceptual model, and the final model. Table 6.1 Research model evolution Hypotheses Initial Conceptual Model Conceptual Model Final Model Technological Relative advantage -> MDM adoption by local government ✓ ✓ X Complexity -> MDM adoption by local government ✓ ✓ ✓ Cost-> MDM adoption by local government ✓ X - Quality of Master Data -> MDM adoption by local government - ✓ ✓ Data Security -> MDM adoption by local government ✓ ✓ X
209 Hypotheses Initial Conceptual Model Conceptual Model Final Model Organizational Data Governance -> MDM adoption by local government ✓ ✓ ✓ Top Management Support -> MDM adoption by local government ✓ ✓ ✓ Top Management Support -> Data Governance ✓ ✓ X Top Management Support -> Technological Competence ✓ ✓ ✓ Technological Competence -> MDM adoption by local government ✓ ✓ ✓ Environmental Government Policy -> MDM adoption by local government ✓ ✓ X Citizen Demand -> MDM adoption by local government ✓ ✓ ✓ Citizen Population Density * Citizen Demand - ✓ ✓ Total (significant) 11 12 8 ✓: Significant, X: Not Significant, -: Not Applicable The research revealed that MDM adoption by local government was found to be contingent on the determinants of technological (complexity, quality of master data), organizational (data governance, top management support, technological competence), and environmental (citizen demand). It is also revealed that top management supports as an influential factor to the technological competence in the organization. In addition, the research also discovered the moderation effect of citizen population density on the relationship between citizen demand and MDM adoption by local government. The following Figure 6.2 demonstrates the final model of determinants affecting MDM adoption by Malaysia local government.