This chapter is divided into four sections. Section one presents the information regarding the validity and reliability of the measures. Section two reports the HLM analysis results of hypothesis verification. Section three summarizes the results. The final section conducts the discussion of the results.
Psychometric Characteristics of the Measures
This study conducted exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) to examine and confirm the appropriateness of the factor loadings for questionnaire items, the factorial validity, the construct reliability, and the convergent reliability of the measures. The EFA performed using Varimax procedure was based on principal-components factor analyses. The extractions were followed by an oblique rotation to determine the extent to which the factors were orthogonal. The CFA was conducted based on covariance structure analyses, where the first observed variable of a latent variable was fixed to 1.00, and a simple structure was maintained. The factorial validity was confirmed by showing that the goodness-of-fit indexes fell within an acceptable range. The construct reliability was confirmed by demonstrating acceptable values of composite reliability (CR), c. The convergent reliability was presented by showing average variance extracted (AVE), v. Analysis results were illustrated as follows.
Innovative Work behaviour
This variable aims to assess employee’s IWB propensity. The pilot study data was collected using Chen’s (2006) 9-item measure. The analysis results of sampling adequacy for the IWB pre-test data (n = 346) showed that the Kaiser-Meyer-Olkin (KMO) measure was .87 (χ2(36)= 3044.25, p < .001), indicating the suitability of the data for exploratory investigations.
An initial EFA yielded two factors with eigenvalues greater than 1. The total variance of
the two-factor solution accounted for 52.38%, where the first factor extracted accounted for 43.92%. A review of factor loadings for the 9 items ranging from .23 to .87 indicated 3 items with poor factor loadings, namely, item number 3 (.23), 4 (.37), and 8 (.42). A further EFA was conducted without the 3 items yielded a single factor solution with factor loadings ranging between .58 and .88. A modified 6-item IWB scale with total variance accounting for 49.15% therefore was created to collect formal study data.
Table 4.1 presents the CFA results based on formal study data (n = 922) for the IWB scale. Standardized factor loadings, R2 values, CR, and AVE as total variance explained were reported. The first-order oblique IWB measurement model without offending estimate exhibited significant factor loadings for the 6 items ranging between .50 and .87 (p <.001), with no significant standard error or negative error variance. Although the reliability of individual item should be further enhanced, such as item number 1 (R2 = .25), in overall, the quality of the items was acceptable (R2 >.5) (Tabachnica & Fidell, 2006).
In addition, the CFA results showed strong fit indexes of the model (χ2(6) = 21.22, p
< .001; χ2/df = 3.54; NFI = .99, NNFI = .98, GFI = .99, RMSEA = .056, and SRMR = .021).
As shown in Table 4.2, the CR of the scale was satisfactory (c = .85) (Bagozzi & Yi, 1988).
The convergent reliability of the construct was also acceptable (v = .50) (Hair et al., 2006).
The Cronbach’s coefficient alpha for the scale was .851.
Table 4.1.
Confirmatory Factor Analysis Results for Innovative Work Behaviour Scale
Factor Item no.a λ (SEm) t value R2 c v
IWBb
1 .50 (.03) 14.58*** .25
.85 .50
2 .87 (.03) 27.87*** .77
5 .69 (.03) 21.55*** .48
6 .62 (.03) 17.82*** .38
7 .76 (.03) 24.13*** .57
9 .75 (.03) 22.46*** .56
Note. a = Item numbers were based on the original scale; b = innovative work behaviour; SEm
= standard error of measurement; λ = standardized factor loadings; c = composite reliability;
v = average variance extracted; n = 922.
*** p < .001.
Work Motivation
This variable aims to assess employee’s intrinsic and extrinsic motivational orientation.
The pilot study data was collected using Chiou’s (2000) 26-item four-dimensional measure.
The EFA and CFA analysis results for IM and EM scales were illustrated in the following paragraph.
Intrinsic Motivation.
The analysis results of pre-test data showed a KMO value of .79 for the IM scale (χ2(78) = 3069.90, p < .001). Three crucial factors were extracted from 13 IM items with eigenvalues greater than 1. The cumulative variance explained by the three factors was 42.32%, where the first factor accounted for 27.07% and the second factor accounted for 10.04%. After each questionnaire item was considered and 6 items that possess low factor loadings were eliminated (i.e., the item number 5 (.23), 20 (.34), 23 (.26), 30 (.34), 8 (.39), and 9 (.19)), 7
items with factor loadings ranging between .47 and .81 were retained. The total variance explained of 7 items accounted for 38.95%.
As shown in Table 4.2, the CFA results for IM scale indicated that all parameter estimates of factor loadings ranging between .45 and .81 were significant (p <.001). The chi-square test result of the t-value for the first-order oblique IM measurement model was significant (χ2(9) = 24.05, p < .001). The normed chi-square value (χ2/df = 2.67), goodness-of-fit indices, and alternative indices (NFI = .99, NNFI = .97, GFI = .99, RMSEA
= .045, and SRMR = .025) all indicated that the model was acceptable. The CR and AVE were .82 and .41. The Cronbach’s coefficient alpha of the scale was .806, indicating acceptable internal consistency.
Table 4.2.
Confirmatory Factor Analysis Results for Intrinsic Motivation Scale
Factor Item no.a λ (SEm) t value R2 c v
IMb
7 .52 (.02) 15.37*** .27
.82 .41
11 .47 (.03) 13.62*** .22
17 .52 (.03) 13.21*** .17
27 .45 (.02) 12.36*** .20
3 .81 (.03) 24.43*** .65
13 .77 (.03) 22.56*** .59
26 .81 (.02) 25.62*** .64
Note. a = Item numbers were based on the original scale; b = intrinsic motivation; c = extrinsic motivation; SEm = standard error of measurement; λ = standardized factor loadings; c = composite reliability; v = average variance extracted; n = 922.
*** p < .001.
Extrinsic Motivation.
The analysis results first indicated the adequacy for conducting EFA (KMO = .71; χ2(78)
= 2092.49, p < .001). Four factors with eigenvalues greater than 1 were extracted from 13 EM items. The total variance explained of the four-factor solution accounted for 40.18%. A further EFA yielded a 6-item single-factor solution with factor loadings ranging between .51 and .67, after the 7 items that possess poor factor loadings were eliminated (i.e., item number 1 (.25), 2 (.27), 6 (.28), 12 (.34), 16 (.31), 19 (.26), and 22 (.24)). The cumulative variance coefficient alpha for the scale was .746, indicating acceptable internal consistency.
Table 4.3.
Confirmatory Factor Analysis Results for Extrinsic Motivation Scale
Factor Item no.a λ (SEm) t value R2 c v reliability; v = average variance extracted.
*** p < .001.
Organisational Climate for Innovation
This firm-level variable aims to assess the extent to which employees’ companies support innovation, which is based on employees’ psychologically meaningful perceptions of organisational settings for innovation. As shown in Table 4.4, the results of high-order confirmatory factor analysis (HCFA) for OCI scale indicated that all parameter estimates of factor loadings ranging between .61 and .92 were significant (p < .001). The chi-square test result of the t-value for the model was significant (χ2(120) = 649.64, p < .001). The goodness-of-fit indices and alternative indices (GFI = .92, NFI = .92, NNFI = .91, RMSEA
= .076, and SRMR = .053) indicated that the model was acceptable.
The CR values ranged between .64 and .83, exceeding .60, implying that the 6 subscales were considered reliable. The AVE for 6 first-order latent variables exceeded .50, indicating that the 6 subscales possessed adequate convergent reliability. The Cronbach’s αs for the 6 subscales were .85 (Value), .74 (Jobstyle), .85 (Teamwork), .88 (Leadership), .84 (Learning), and .85 (Environment), and that for an overall OCI scale was .93, indicating satisfactory internal consistency.
In addition, as shown in Table 4.5, the HCFA results indicated that the second-order OCI measurement model showed stronger goodness-of-fit indexes than the first-order oblique OCI model (χ2(120) = 1139.73, p < .001; GFI = .87, NFI = .87, NNFI = .85, RMSEA = .097, and SRMR = .055).
Table 4.4.
Confirmatory Factor Analysis Results for Organisational Climate for Innovation Scale
Factor Item no.a λ (SEm) t value R2 c v reported because the path coefficients for the first observed variables of latent variables were set to 1.00; SEm = standard error of measurement; λ = standardized factor loadings; c = composite reliability; v = average variance extracted;
n = 922.
*** p < .001.
Table 4.5.
The Goodness-of-fit Comparison between First-order and Second-order Measurement Models of Organisational Climate for Innovation Scale
GFIsa χ2(df) χ2/df GFI NFI NNFI RMSEA SRMR
First-order oblique model
1139.73 (120) 9.50 .87 .87 .85 .097 .055
Second-order model
649.64 (120) 5.41 .92 .92 .91 .076 .053
Note. a GFIs = goodness-of-fit indexes.
As shown in Table 4.6, the correlation coefficients for the six factors in the OCI scale ranged between .36 and .72 (p < .001) and did not differ significantly, indicating that the correlation coefficients may be influenced by an identical second-order factor. In addition, the factor loadings for the second-order factor and six OCI factors were high and ranged between .71 and .85 (p < .001), indicating that the six factors were closely related.
Table 4.6.
The Second-order Factor Loadings and the Correlation between the First-order Factors
Factora
Organisational Climate for
Innovation
Correlation coefficients of the first-order factors
1 2 3 4 5
1. VL ..76b
2. JS .77 .59*** c
3. TW .78 .56*** .50***
4. LD .85 .58*** .69*** .68***
5. LN .71 .72*** .36*** .55*** .55***
6. EN .83 .49*** .64*** .59*** .64*** .60***
Note. a VL = value, JS = Jobstyle, TW = teamwork, LD = leadership, LN = learning, EN = environment; b = factor loadings; c = correlation coefficients.
n = 922.
*** p < .001.
Two-tailed tests.
Finally, the appropriateness of aggregating individual employee scores to a firm-level construct of OCI was confirmed by the median rWG values of .99 (SD = .004) which was calculated using a uniform null distribution (James, Demaree, & Wolf, 1984), indicating very strong within-company agreement (LeBreton & Senter, 2008) on the perceptions of OCI. The ICC (2) value of .95 suggested that, on average, employees possessed high consensus on OCI of their companies (Bliese, Halverson, & Schriesheim, 2002). According to the results of interrater reliability and intraclass correlation tests, the combination of individual employee’s responses into a single measure of firm-level OCI therefor was supported.
Hypothesis Tests
This section presents the information of descriptive statistics and the analysis results of bivariate correlation, followed by the illustration of model establishment and the results of hypothesis verification. Further details are provided in the following paragraph.
Descriptive Statistics and Correlation Analyses
Table 4.7 presents the means, standard deviations, correlation coefficients for the variables, and the reliabilities of the measures used in this study. The bivariate correlation matrix shows that all independent variables (i.e., IM, EM, ASC, and OCI) were significantly and positively correlated with IWB.
Of all the employee-level variables, IM was more strongly related to IWB (r = .51, p
< .01), followed by ASC (r = .33, p < .01) and EM (r = .22, p < .01). Employee’s perceived OCI was also positively related to IWB (r = .27, p < .01). These findings provided initial evidence that supports the hypotheses 2, 3, and 4.
In addition, although the results showed that ASC was positively correlated with its three dimensions, namely, upper reachability (r = .75, p < .01), heterogeneity (r = .79, p
< .01), and extensity (r = .73, p < .01), indicating that the three dimensions may reflect the construct of ASC, 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, the score of ASC was computed only using the indicators of heterogeneity and extensity in the study.
Table 4.7.
Descriptive Statistics and Bivariate Correlations
Measuresa Min Max M SD 1 2 3 4 5 6 7 8 9
1. Tenure 0.8 21 10.12 7.50
2. Education 1 5 3.07 .76 -.38**
3. IWB 1.83 5 3.45 .63 -.03 -.13**
4. IM 2.43 4.16 3.66 .51 -.03 -.11** .51**
5. EM 1.83 4.31 3.67 .51 -.09** -.01 .22** .15**
6. ASC 2.91 1.83 0 1 -.01 -.05 .33** .13** -.06
7. UR 52 85 77.43 6.78 -.17** -.24** .26** .20** -.03 .75**
8. HG 1 17 8.65 3.33 -.05 -.07 .33** .16** -.03 .79** .71**
9. ES 0 69 53.56 12.22 -.00 -.03 .22** .11** -.06 .73** .75** .59**
10. OCI 2.09 4.97 3.38 .57 -.14** -.01 .27** .40** -.27** .14** .12** .19** .13**
Note. a IWB = innovative work behaviour; IM = intrinsic motivation; EM = extrinsic motivation; ASC = accessed social capital; UR = upper reachability; HG = heterogeneity; ES = extensity; OCI = organisational climate for innovation.
n = 922.
** p < .01.
Two-tailed tests.
Random Effects of Innovative Work Behaviour
Table 4.8 presents the null model established to investigate the between-company variance in the mean IWB. The employee-level model of Equation 4.1 predicted the mean IWB in each company with just one parameter of the intercept, namely, β0j. The firm-level model of Equation 4.2 shows that each company’s mean IWB, β0j, is represented as a function of the grand-mean IWB in the companies, γ00, plus a random error of u0j. The mixed model of Equation 4.3 was yielded by substituting Equation 4.2 into Equation 4.1. The one-way ANOVA with random effects was used to capture the proportion of variance in IWB resided between companies by examining the firm-level residual variance of the intercept (i.e., τ00) and the intraclass correlation coefficient (i.e., ICC(1)).
Table 4.8.
The Null Model for Testing the Between-company Variance in Innovative Work Behaviour Employee-level Model:
IWBij = β0j+ rij [4.1]
where:
rij is the random effect associated with employeeinested in companyj, Variance (rij) = σ2 = within-company variance in IWB.
Firm-level Model:
β0j = γ00 + u0j [4.2]
where:
γ00 is the grand-mean IWB across the companies, u0j is the random effect associated with companyj, Variance (u0j) = τ00 = between-company variance in IWB.
Mixed Model:
IWBij = γ00 + u0j + rij [4.3]
As shown in Table 4.9, HLM analysis results for the null model indicated that the coefficient of grand-mean IWB with fixed effect, γ̂00, was 3.46 (p < .001). The results of
variance components analyses showed a τ̂00 of .12 (χ2(35) = 391.82, p < .001) and a σ̂ 2 of .29, which further yielded an ICC (1) of .29, indicating that about 29% of the variance in IWB was between companies and about 71 % was within companies. These results indicated that significant variance does exist among companies in the mean IWB propensity.
In addition, according to Cohen (1988), the influence of group membership should be taken into account in a situation where the value of ICC (1) ranges between .059 and .138, which implies that a certain proportion of total variance in the outcome variable can be explained by the groups or by the differences between the groups. Considering the multilevel data structure studied and the ICC (1) value of .29, the precondition of conducting multilevel analyses to verify the hypotheses therefore was supported.
Table 4.9.
The Test Results for the Between-company Variance in Innovative Work Behavioura The Null Model (M0)
Variable
Fixed effects Random effects
Estimateb s.e. t Estimate χ2 σ̂
Intercept γ̂00 3.46*** .06 57.82 τ̂00 .12*** 391.82 .29
Model deviancec 1558.91
Note. a n = 36 at the firm level; b γ̂00 with robust standard errors; c Deviance is defined as -2 × the log-likelihood of a maximum-likelihood, which is a measure of model fit.
*** p < .001.
Two-tailed tests.
Main Effects of Intrinsic Motivation, Extrinsic Motivation, and Accessed Social Capital on Innovative Work Behaviour (Hypotheses 1, 2, and 3)
Table 4.10 presents the random coefficient regression model established to test the Hypotheses 1, 2, and 3. The employee-level model of Equation 4.4 indicated that the IWB was regressed on IM, EM, and ASC after controlling for education and current tenure. Each company’s distribution of IWB was characterized by six parameters: the employee-level intercept of IWB, β0j, and the slopes, β1j-5j, for the relationships between the employee-level predictors (i.e., education, current tenure, IM, EM, and ASC) and IWB. The firm-level model comprising Equations 4.5 to 4.10 indicated that the intercept and the slopes were estimated by the parameters of γ00-50 and u0j-5j. The mixed model of Equation 4.11 was yielded by substituting Equation 4.5 to 4.10 into Equation 4.4.
Table 4.10.
The Random Coefficient Regression Model for Testing the Main Effects of Intrinsic Motivation, Extrinsic Motivation, and Accessed Social Capital on Innovative Work Behaviour
Employee-level Model:
IWBij = β0j + β1j*(Eduij) +β2j*(Cur_tenij) + β3j*(IMij) + β4j*(EMij) + β5j*(ASCij) + rij [4.4]
where:
rij is the random effect associated with employeei nested in companyj. Firm-level Model:
β0j = γ00 + u0j [4.5]
β1j = γ10 + u1j [4.6]
β2j = γ20 + u2j [4.7]
β3j = γ30 + u3j [4.8]
β4j = γ40 + u4j [4.9]
β5j = γ50 + u5j [4.10]
where:
γ00
γ10-50
u0j
u1j-5j
is the mean of the IWB intercepts across companies,
are the mean predictor－IWB regression slopes across companies, is the random effect to the intercept of IWB associated with companyj, are the random effects to the slopes associated with companyj.
Mixed Model:
IWBij = γ00 + γ10*Eduij + γ20*Cur_tenij + γ30*IMij + γ40*EMij + γ50*ASCij + u0j + u1j*Eduij + u2j*Cur_tenij + u3j*IMij + u4j*EMij + u5j*ASCij + rij [4.11]
Table 4.11 presents the analysis results for the random coefficient regression model. Of the two control variables, only education was significantly but negatively related to IWB (γ̂10
= -.07, p < .05). In addition, the results showed that IM was more significantly related to IWB (γ̂30 = .41, p < .001), followed by ASC (γ̂50 = .25, p < .01) and EM (γ̂40 = .31, p < .05).
Therefore, the findings supported Hypotheses 1, 2, and 3. The pseudo R2within-company for the model was .91, indicating that, in overall, adding IM, EM, ASC, and the two control variables as employee-level predictors of IWB reduced the within-company variance by 91%. In other words, the predictors account for about 91% of the employee-level variance in the IWB.
Furthermore, the chi-square test for the firm-level residual variance of the intercept, β0j,
was significant (τ̂00 = .19, p < .001), indicating that the significant difference among companies in their IWB not only was influenced by the employee-level predictors but also probably influenced by certain firm-level factors, whereby the study further investigated that whether OCI can exerts a cross-level effect on IWB (i.e., Hypothesis 4).
Besides, the results of components variance analyses indicated that the chi-square tests for the firm-level residual variance of the IM and EM slopes (i.e., β3j andβ4j) were significant, with the τ̂ values of .37 (p < .001) and .53 (p < .001), implying that the significant difference of IM and EM slopes may be influenced by firm-level predictors, whereby the study further investigated that whether OCI can exerts cross-level moderating effects on the slopes (i.e., Hypotheses 5 and 6).
Finally, the analysis result of a parallel model comparison indicated that a chi-square test of the change in the deviance statistic from null model (M0) to random coefficient regression model (M1) confirmed that the inclusion of employee-level predictors significantly improved the overall model fit (Δ-2LL = 1548.31, Δdf = 20, p < .001).
Table 4.11.
The Test Results for the Main Effects of Intrinsic Motivation, Extrinsic Motivation, and Accessed Social Capital on Innovative Work Behavioura
The Random Coefficient Regression Model (M1)
Variable
Fixed effects Random effects
Estimateb s.e. t Estimate χ2 σ̂
Intercept γ̂00 3.44*** .08 40.74 τ̂00 .185*** 423.74
.025 Education γ̂10 -.07* .02 -2.60 τ̂11 .018*** 292.46
Tenure γ̂20 -.03* .07 -0.35 τ̂22 .164*** 366.71
Intrinsic motivation γ̂30 -.41*** .11 3.68 τ̂33 .373*** 240.87 Extrinsic motivation γ̂40 -.31* .13 2.38 τ̂44 .528*** 118.32 Accessed social capital γ̂50 -.25** .08 2.96 τ̂55 .194*** 307.04 pseudo R2within-companyc
.91
Model devianced 10.61
Note. a n = 922 at the employee level, n = 36 at the firm level; b γ̂s, with robust standard errors;
c R2 was calculated based on the proportional reduction of within-company variance in the IWB; d Deviance is defined as -2 × the log-likelihood of a maximum-likelihood, which is a measure of model fit; All the predictors were group-mean-centered in theses analyses.
* p < .05.
** p < .01.
*** p < .001.
Two-tailed tests.
Cross-level Effects of Organisational Climate for Innovation on Innovative Work Behaviour (Hypothesis 4)
This hypothesis seeking to understand why some companies have higher means of IWB propensity than others was aimed to investigate that whether the firm-level predictor of OCI has positive relation with the outcome variable of IWB, specifically, whether the mean IWB of each company can be predicted by OCI after controlling for the employee-level predictors.
Table 4.12 presents the mean-as-outcome model incorporating with the main effects of employee-level predictors for testing Hypotheses 5. The employee-level model remains the same as in Equation 4.4, whereas the firm-level predictor of OCI was now added into Equation 4.5 to formulate Equation 4.11, indicating each company’s mean IWB now to be predicted by the mean OCI of the company. The mixed model of Equation 4.12 was yielded by substituting Equation 4.6 to 4.11into Equation 4.4.
Table 4.12.
The Mean-as-Outcome Model Incorporating with the Main Effects of Employee-level Predictors for Testing the Cross-level Effect of Organisational Climate for Innovation on Innovative Work Behavior
is the mean of the IWB intercepts across companies, is the fixed effect of the mean OCI on β0j,
are the mean predictor－IWB regression slopes across companies, is the conditional residual β0j － γ00 － γ01*(OCIj),
are the random effects to the slopes associated with companyj. Mixed Model:
IWBij = γ00 + γ01*(OCIj) + γ10*Eduij + γ20*Cur_tenij + γ30*IMij + γ40*EMij + γ50*ASCij + u0j + u1j*Eduij + u2j*Cur_tenij + u3j*IMij + u4j*EMij + u5j*ASCij + rij [4.12]
As shown in Table 4.13, the HLM results for the model (M2) indicated that OCI was significantly and positively related to IWB (γ̂01= .47, p < .001), supporting Hypothesis 4. In addition, all employee-level predictors, except for tenure, were significantly related to IWB, with γ̂s ranging between -.06 and .41. The conditional between-company variance (τ̂00) in the intercept of IWB, β0j, was reduced .002 from .185 (M1) to .183 (M2). The pseudo R2between-company was .011, indicating that adding OCI as the firm-level predictor of mean IWB reduced the between-company variance by 1.1%. In other words, OCI accounts for about 1%
of the firm-level variance in the IWB. The chi-square test result for the firm-level residual variance in the intercept of IWB remains significant (τ̂00 = .18, p < .001), indicating that there may have certain firm-level factors to explain the significant difference among companies in their IWB after adding the firm-level predictor of OCI. Finally, a chi-square test of the change in the deviance statistic from random coefficient regression model (M1) to mean-as-outcome model (M2) confirmed that including OCI as firm-level predictor significantly improved the overall model fit (Δ-2LL = 23.26, Δdf = 1, p < .001).
Table 4.13.
Cross-level Moderating Effects of Organisational Climate for Innovation on the Relationships between Intrinsic Motivation, Extrinsic Motivation, and Innovative Work Behaviour (Hypothesis 5 and 6)
The Hypotheses aim to seek that whether in some companies the correlations between IM and IWB, and EM and IWB are stronger than in others due to OCI, specifically, does OCI exerts cross-level moderating effects on such relationships. Table 4.14 presents the full model created to test the Hypotheses 5 and 6, incorporating the main effects of employee-level predictors and the cross-level effect of OCI on IWB. The employee-level model remains the same as in Equation 4.4, whereas the OCI as the firm-level predictor for the slopes of IM-IWB and EM-IWB was now added into Equation 4.8 and 4.9 to form Equation 4.13 and 4.14, indicating that the slopes were now predicted by OCI. The mixed model of Equation 4.15 was yielded by substituting all firm-level equations into Equation 4.4.
Table 4.14.
The Full Model for Testing the Cross-level Moderating of Organisational Climate for
The Full Model for Testing the Cross-level Moderating of Organisational Climate for