This study undertaken by means of quantitative research methods developed a multilevel model of employee innovative work behaviour. Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) were performed to confirm the factorial validity and construct reliability of the constructs. Hierarchical linear modeling (HLM) was used to conduct multilevel analyses for verifying the hypotheses. This chapter includes six sections, namely, research framework, research procedures, data collection, measures, data analysis methods, and data analysis procedures.
Research Framework
Figure 3.1 shows the hypothesized multilevel model of innovative work behaviour as research framework of this study.
Employee level Firm level
Figure 3.1. Hypothesized Multilevel Model of Innovative Work Behaviour.
Organisational Climate
As shown in Table 3.1, Hypotheses 1, 2, and 3 (H1, H2, and H3) were proposed to examine that whether the employee-level factors of intrinsic motivation (IM), extrinsic motivation (EM), and accessed social capital (ASC) can influence innovative work behaviour (IWB). Hypotheses 4, 5, and 6 (H4, H5, and H6) were proposed to examine that whether the firm-level factor of organisational climate for innovation (OCI) exerts a contextual effect on IWB, and that whether OCI exerts cross-level moderating effects on the hypothesized relationships of IM－IWB and EM－IWB.
Table 3.1.
The Hypotheses
Hypotheses Content
H1 Intrinsic motivation is positively related to innovative work behaviour (Employees with a higher orientation to intrinsic motivation are more likely to exhibit stronger propensities of innovative work behaviour.).
H2 Extrinsic motivation is positively related to innovative work behaviour (Employees with a higher orientation to extrinsic motivation are more likely to exhibit stronger propensities of innovative work behaviour.).
H3 Accessed social capital is positively related to innovative work behaviour (Employees with more diverse acquaintances are more likely to exhibit stronger propensities of innovative work behaviour.).
H4 Organisational climate for innovation is positively related to innovative work behaviour (On average, employees are more likely to exhibit stronger propensities of innovative work behaviour when they perceive a higher extent of supportive climates for innovation.).
(continued)
Table 3.1. (continued)
Hypotheses Content
H5 Organisational climate for innovation positively moderates the relationship between intrinsic motivation and innovative work behaviour (On average, when employees perceive a higher degree of supportive climates for innovation, the extent to which employees’ intrinsic motivational orientations contribute to their propensities of innovative work behaviour is more likely stronger.).
H6 Organisational climate for innovation positively moderates the relationship between extrinsic motivation and innovative work behaviour (On average, when employees perceive a higher degree of supportive climates for innovation, the extent to which employees’ extrinsic motivational orientations contribute to their propensities of innovative work behaviour is more likely stronger.).
Research Procedures
As shown in Figure 3.2, the research procedures comprising thirteen steps divided into 3 major stages are illustrated as follows.
Preparation
The researcher first performed an extensive review of innovation literature focusing on individual’s innovation of organisation to acquire in-deep understanding on the topic of employee innovative work behaviour. The knowledge known of the topic was systematically sorted out to identify research gap and to clarify problematic consciousness. After reviewing the literature, the hypothesized relationships between the variables were developed, followed by the construction of research framework. The strategies and methods of data collection and data analysis were evaluated and selected based on the needs of responding the research questions and purpose.
Implementation
This stage comprises two major procedures of data collection and data analyses. The collection of data was performed twice. The returned questionnaires that featured a clear response orientation and/or an excessive number of omitted responses were eliminated. The pilot study data first collected was used to conduct EFA for examining the appropriateness of the factor loadings for questionnaire items. The modified scales created based on the EFA results were used to collect the formal study data that further employed to conduct CFA and to test the hypotheses.
Completion
After the data was analyzed, the analysis results were discussed and concluded. The theoretical and managerial implications were discussed based on the research findings. The limitations of the study and future research directions were elucidated.
Figure 3.2. Research Procedures.
Identification of Problematic Consciousness
Establishment of Hypothetical Framework
Research Design
Data Collection Proposal Meeting Research Motivation
Reviews of Literature
Data Analyses Reviews of Response Set
Data Coding
Conclusions and Implications
Final Defense
End of Thesis Project Preparation
Implementation
Completion
Results and Discussion
Measures
The corresponding sources of the information for specific measures are described below.
The texts of items for all scales are provided in Appendix A.
Innovative Work Behaviour
This study measured employee’s innovative work behavioural propensity with a 6-item modified IWB scale adapted from Chen’s (2006) Innovative Behaviour Scale. This 6-item single-dimensional measure captures various behaviours regarding four major types of innovation activities, namely, the idea exploration, idea generation, idea championing, and idea realization. Employees were instructed to select the response that most reflected their personal experiences or behaviours related to innovation on a 5-point Likert-type scale with scale anchors ranging from “never” (1) to “always” (5). High scores indicated strong IWB propensity. The intraclass measure, ICC (1), of .71 provided the preliminary information of the appropriateness for conducting multilevel analyses.
Accessed Social Capital
This study measured employee’s quantity and diversity of acquaintances as capability of social relations with Hwang’s (2011) measure of position generators. This measure listed 22 occupations with its corresponding scores of socioeconomic status that were based on Ganzeboom & Treiman’s (1996) Standard International Occupational Scale (SIOPS) (Hwang, 2011). Employees were instructed to tick occupations with the question that do you have any acquaintance that has the following occupations? High scores indicated strong capability of social relations. In addition, the correlation analysis results indicated that of the three indicators suggested by Lin (2001), upper reachability was found to be correlated highly with heterogeneity (r = .71, p <.01) and extensity (r = .75, p <.01) where heterogeneity was associated with extensity (r = .59, p <.01). Given the possibility of multicollinearity, ASC is constructed using the indicators of heterogeneity and extensity in this study.
Work Motivation
This study measured employee’s work motivation as intrinsic and extrinsic motivational orientation with a 13-item modified scale (i.e., 7-item IM subscale and 6-item EM subscale) adapted from Chiou’s (2000) 26-item four-dimensional measure which was derived from Amabile et al.’s (1994). Employees were instructed to rate how accurately each item described them as they generally were on a 5-point Likert-type scale with scale anchors ranging from “strongly agree” (1) to “strongly disagree” (5). High IM scores indicated strong intrinsic motivational orientation.
Organisational Climate for Innovation
This study measured organisational climate for innovation with an 18-item shortened version (Chiou, 2011) of Chiou et al.’s (2009) 35-item Creativity Organisational Climate Inventory. This measure consists of 6 sub-scales, namely, Value (3 items), Jobstyle (3 items), Teamwork (3 items), Leadership (3 items), Learning (3 items) and Environment (3 items).
Employees were instructed to rate, on the basis of their personal observation, the extent to which their companies support innovation regarding the described conditions on a 5-point Likert-type scale with scale anchors ranging from “strongly agree” (1) to “strongly disagree”
(5). High scores indicated strong innovation climate. The interrater and intraclass measures (median rWG .99 and ICC2 = .95) justified aggregation across raters (Bliese, Halverson, &
Schriesheim, 2002; LeBreton & Senter, 2008).
Data Collection Sampling
The data collected from Taiwanese working adults in business firms across various industries was based on the method of convenience sampling. The present study adopts the sampling strategy because the researcher anticipated that the theoretical and empirical linkages between behavioural propensity, social relations, motivational orientation, and perceived climate would remain robust, although in the case of not considering the factor of
industry. Numerous studies regarding innovative work behaviour (IWB) conducted in a context of various industries (Chen, 2006; Dorenbosch et al., 2005; Kleysen & Street, 2001;
Krause, 2004; Scott & Bruce, 1994; Tsai, 2008; Tsai & Kao, 2004) showed that employee’s IWB propensity was significantly related to certain individual’s characteristics (e.g., motivation) and social environment (e.g., the climate for innovation), though they had different conclusions. Therefore, the theoretical pattern of the relationships between employee’s personal characteristics, psychological conditions of social environment and behavioural propensity may empirically exist in various industrial departments. In addition, from the perspectives of social psychology and interactionism, individual’s behaviour was viewed as the function of people’s characteristics and social environment, which has been a general consensus in the field of behavioural research, therefore the attempts of establishing the relationships between the variables studied should be theoretically acceptable in the heterogeneous contextual background of industry.
Procedures
The data used to conduct pilot and formal study were collected using online surveys.
Ninety prizes of gift voucher worth approximately $10 in value per each were provided to enhance response rate for the collection of formal study data. The researcher’s acquaintances as organisational contacts sent the inquiry letters by means of emails to their colleagues for exploring their willingness of participating in this survey. Participants who replied the emails with consent to participate in the survey received a link of web page created by the researcher and distributed through the participant’s corresponding organisational contacts.
The web page first shows an informed consent form ensuring the confidentiality of the information, followed by the instructions of how to participate in the survey. During the next phase, participants were asked to complete the questionnaire items for assessing their accessed social capital (ASC), followed by the items for IWB, intrinsic motivation (IM), extrinsic motivation (EM), organisational climate for innovation (OCI), and demographic
characteristics. Participant’s each set of response answer was automatically directed and saved into its respective database by the online survey system.
Each participant’s database was expected to have only one set of the answer; otherwise, the database was manually deleted by the researcher. The links of online survey questionnaire were coded according to employee’s organisation; hence the returned online questionnaires can be differentiated and retrieved according to the catalog of organisation. An organisational data with the returned questionnaires less than 2 apparently distinct departments was excluded to ensure the representativeness of within organisational consensus on OCI. The researcher is the only one person allowed to access into the databases.
The pilot study data collected from mid-June through mid-August 2013 and the formal study data collected from mid-September through mid-December 2013 were independent from each other. There were no missing data.
Participants
The pilot study participants were 346 employees from 13 companies across 6 industries including high-tech, banking, retail, software, traditional manufacturing, and information.
43% percent of the participants were women and 81.3% of the participants possessed a bachelor’s or master’s degree. The formal study participants were 922 employees from 36 companies ranging in size from 81 to 1,500. A total of 1129 questionnaires were distributed, of which 978 were returned and 56 that featured a clear response set and/or an excessive number of omitted responses were eliminated, achieving a valid return rate of 81.67%. Of these valid questionnaires, 581 were completed by female employees (63.01%), and 716 were completed by the employees with a bachelor’s or master’s degree (77.66%). 34.6% of participants were from high-tech industry, and others were from general service (17.2%), financial service (12.6%), traditional manufacturing (13.1%), banking (9.1%), information (7.4%), retail (3.1%), and software (2.9%).
Common Method Variance
Common method variance (CMV) occurs in situations where two or more variables are assessed with the same method, which may inflate or deflate the observed relationships among constructs (Cote & Buckley, 1988; Spector & Brannick, 2010), leading to both Type I and Type II errors. This study might fall a victim to the potential threats of CMV because all the variables were measured by self-report assessment format. This study adopted six prior remedies to alleviate the negative effects of CMV (Peng, Kao, & Lin, 2006; Podsakoff et al., 2003; Tourangeau, Rips, & Rasinski, 2000), namely:
1. The question items for measuring intrinsic and extrinsic motivation were intermixed with one another in random order.
2. The variable of OCI was constructed by aggregating individual employee scores to a firm-level construct (i.e., mean OCI), as is also a necessary step in the conduction of hypothesis tests in this study.
3. The order of the variables presented in the questionnaire was arranged in such a way that ASC (independent variable) was shown first, followed by IWB (dependent variable), IM and EM (independent variables), and OCI (independent variable as a moderator).
4. The names of the variables were not revealed in the questionnaire so as to reduce psychological interference on the part of participants.
5. Surveys were allowed to be done anonymously.
A Harman single-factor test was performed to examine the severity of CMV. The results of unrotated factor analysis of the items included in all the questionnaire subscales showed 15 distinct extracted factors with variance ranging between 1.24 and 26.3 accounted for 36.12%
of the total cumulative variance explained. Because the first principal component (with factor loadings ranging between .41 and .76) were not the general factor for all independent and dependent variables, the CMV in this study was not severe.
Data Analysis Methods
The study was undertaken by means of quantitative research methods. A questionnaire survey was conducted to collect quantitative data, validate measures, and test the research hypotheses. The corresponding statistical methods and techniques used to perform data analyses are described as follows.
Descriptive statistics and correlation analysis
The descriptive statistics performed using SPSS version 21 for Windows was used to capture and describe the demographic characteristics of the sample. In addition, correlation analyses were used to investigate the direction and strength among the variables.
Confirmatory factor analysis
The method of structural equation modeling (SEM) was used to perform confirmatory factor analyses (CFA) for testing the factorial validity and construct reliability of the constructs by examine the relations among observed variables (i.e., questionnaire items) and latent variables (i.e., the factors of the constructs). The CFA conducted using LISREL 8.51 statistical software (Jöreskog & Sörbom, 2001) was based on the covariance structure analyses, where one item per construct was fixed to 1.0, and a simple structure was maintained. The factorial validity was confirmed by showing that the fit indexes fell within an acceptable range (e.g., normed chi-square < 3, GFI (goodness-of-fit index) > .90, NFI (normed fit index) > .90, NNFI (non-normed fit index) > .90, RMSEA (root mean square error of approximation) < .05, and SRMR (standardized root mean square residual) < .08) (Bentler, 1980; Browne & Cudeck, 1993; Carmines & McIver, 1981; Hu & Bentler, 1999).
The tests of construct reliability were performed to examine the extent to which latent variables can explain the proportional variance of its corresponding observed variables (Fornell & Larcker, 1981). The values of composite reliability (CR or c)1 were computed
1 𝐶𝑅 = 𝜌𝑐 = [(∑ 𝜆(∑ 𝜆𝑖)2
𝑖)2+∑ Θ𝑖𝑖)] where: (∑ 𝜆𝑖)2 is the squared sum of standardized factor loadings,
and used to confirm construct reliability (Raine-Eudy, 2000). The tests of convergent ability were performed by computing average variance extracted (AVE or v)2, which was aimed to explain the extent to which latent variables can be effectively estimated by observed variables (Hair, Black, Babin, Anderson, & Tatham, 2006). The Cronbach’s alphas of the measures were also reported.
Multilevel analysis
This study used hierarchical linear modeling (HLM) to conduct multilevel analyses for hypothesis verification. Statistical software of HLM version 6.08 was used to perform the analyses (Raudenbush, Bryk, & Congdon, 2009). HLM provides theoretical, conceptual, and statistical mechanism for investigating organisational phenomena (Hofmann, 1997). HLM rationale and related issues comprising intraclass correlation, aggregation, estimating effects, explanatory power, and model fit are illustrated in the following paragraphs.
Hierarchical linear modeling.
The statistical methodology and techniques of HLM was developed primarily based on the needs of analyzing hierarchical data structures (Raudenbush & Bryk, 2002). From a theoretical perspective, because all organisational phenomena are hierarchically embedded in a higher-level context (Kozlowski & Klein, 2000), it is advisable that the researchers take a meso perspective to investigate and discuss multilevel phenomena (House, Rousseau, &
Thomas-Hunt, 1995). In addition, from the perspective of statistical methodology, because the phenomena are multilevel, the data is clustered; it has been noted that the statistical techniques of ordinary least square (OLS), such as regression analysis and/or analysis of variance (ANOVA), may not be relatively appropriate methods for analyzing clustered data
∑ Θ𝑖𝑖 .is the sum of indicator measurement errors.
2 𝐴𝑉𝐸 = 𝜌𝑣 = (∑ 𝜆∑ 𝜆𝑖 2
𝑖 2+∑ Θ𝑖𝑖) where: ∑𝜆 2𝑖 .is the sum of squared standardized factor loadings.
due to the assumption of the methods that the observations are independent3 (Luke, 2004).
The statistical decision may be too liberal (Type I errors) or too conservative (Type II errors) if the aforesaid theoretical and methodological concerns were ignored (Bliese & Hanges, 2004). The main idea to describe HLM was that it partitioned the variance of individual-level outcomes into level 1 (e.g., employee-level) and level 2 (e.g., team-level) components, and then regressed the level 1 variance component onto individual-level predictors, and the level 2 variance component onto organisation-level predictors. Because HLM was allowed to integrate individual- and higher-level data into a multilevel model to be analyzed, HLM can better serve to observe and investigate organisational multilevel phenomena. In addition, HLM introduced macro-level (i.e., higher-level) error terms in regression models to capture the intraclass correlations of nested relationships by estimating the variance of macro-level error terms, therefore the statistical results of tests would be more realistic when drawing conclusions related to the influences of phenomena at different levels.
Intraclass correlation.
Intraclass correlation describes the relative degree of within-group and between-group variance. Intraclass correlation coefficient, ICC (1)4, is often used to capture the similarity and/ or non-independency of observations by examining the extent to which the total variance of outcome variables can be explained by the between-group variance (Blises & Hanges, 2004). In other words, the measurement of ICC (1) aims to investigate the extent to which the
3 The observations are assumed to be independent or non-interdependent in situations where the analyses are conducted using ordinary least square (OLS) techniques. The assumption would be violated in various degrees when the data is clustered. In an organisational context, because employees are nested in departments that are nested in the company, the survey responses collected from the employees from a same organisation may be dependent due to the inherently hierarchical nature of organisations. For example, employee A’s and B’s testing scores of IWB propensity may be correlated if they perceive the identical innovation climate (i.e., the same higher-level contextual factor) which influences their IWB propensities.
4 𝐼𝐶𝐶(1) = 𝜌 = 𝜏𝜏00
00+𝜎2 where: 𝜏00 is the between-group variance, 𝜎2 ..is the within-group variance.
individuals/members are dependent within a particular group or a context (Hox, 2002). It has been noted that the hypothesis verification should be undertaken using multilevel analysis strategies and methods in situations where the value of ICC (1) is high which implies that the assumption of independent observations may be violated in a large degree (Hoffman, 1997;
Luke, 2004). According to Cohen (1988), the effect of group membership should not be ignored when the value of ICC (1) ranges between .059 and .138.
Aggregation and contextual variable.
Aggregation is related to the conceptualisation and measurement of macro-level variable, namely, contextual variable (e.g., innovation climate of organisation). This kind of variable reflects particular characteristics of organisational social environment (e.g., the behavioural patterns of innovation) which can produce corresponding atmosphere that may exert contextual influences on organisational members. In other word, therefore, the members of an organisation may experience homogeneous shared psychological perception/psychological climate (James & Jones, 1974). From the perspective of multilevel analysis, contextual influences can be understood as net effect of a contextual variable after controlling the same variable at the level of individual (Pedhazur, 1997). In this study, although the construct of OCI was measured based on employees’ psychological perceptions of organisational settings for innovation, this construct is centered on its inherent meaning of organisational. Whether the contextual variable of OCI can be constructed by aggregating individual employee scores to a firm-level construct remains to be confirmed. An assessment of within-group agreement thus was performed to confirm that whether the data justified the aggregation of firm-level
𝑤 is the within-group agreement coefficient for raters’ mean scores based on J items,
Wolf, 1984), was calculated to examine the appropriateness of conducting the aggregation.
The intraclass correlation coefficient, ICC (2)6, a measure for testing the stability of group means, was used to assess the reliability of company means of OCI (Kozlowski & Klein, 2000). Appendix B provides the SPSS scripts for computing rWG.
Estimation of fixed and random effects.
The situation in which an employee-level variable can significantly predict IWB was determined by that the t tests results for the firm-level parameters (i.e., γ̂s) were statistically significant. In addition, the chi-square tests were used to investigate the significance of within- and between-company variance by examining the estimates of σ2 and τ̂. A significant
The situation in which an employee-level variable can significantly predict IWB was determined by that the t tests results for the firm-level parameters (i.e., γ̂s) were statistically significant. In addition, the chi-square tests were used to investigate the significance of within- and between-company variance by examining the estimates of σ2 and τ̂. A significant