The formal questionnaire that was applied in this study covered innovation speed, leader encouragement of creativity, organizational ambidexterity, creative process engagement, and control variables (gender, tenure, firm ownership, firm age, firm size). Most of the items adopted in this research were selected from prior studies, though the researchers modified several to reflect the specific context of China. All scales in this study followed a five-point Likert format. Evaluation of the scales showed that all Cronbach’s alpha values were above 0.70. To detect common method bias due to self-reports, Harman’s one-factor test was used. The results showed that none of the factors significantly dominated the variance, so common method variance is not an issue [45
Innovation speed (IS).
This was conceptualized by a three-item scale borrowed from previous literature: (1) time effectiveness, (2) time efficiency, and (3) time relative to competitors in the industry [47
]. We copied these three items and, to be consistent, amended them for our other measure of speed as well as the corresponding measure of technical time efficiency, which enabled us to compare more than incremental innovation [48
]. The response options ranged from 1, “strongly disagree,” to 5, “strongly agree.” The Cronbach’s alpha reliability value of this scale was 0.79.
Leader encouragement of creativity (LEC).
A six-item scale developed by Zhang and Bartol [7
], drawing from Scott and Bruce [50
], was used to measure leader encouragement of creativity. Response options ranged from 1, “strongly disagree,” to 5, “strongly agree.” The reliability of this scale was 0.86.
Organizational ambidexterity (OA).
The researchers adopted the items developed by Gibson and Birkinshaw [20
] to measure organizational ambidexterity. According to their study, alignment and adaptability are two dimensions of organizational ambidexterity. Hence, two different subscales were used to measure alignment and adaptability, respectively.
Alignment (AL). A three-item scale was adopted to measure alignment. Response options ranged from 1, “strongly disagree,” to 5, “strongly agree.” The reliability of this scale was 0.87.
Adaptability (AD). A three-item scale was adopted to measure adaptability. Response options ranged from 1, “strongly disagree,” to 5, “strongly agree.” The reliability of this scale was 0.91.
Creative process engagement (CPE).
Problem identification, information search and encoding, and idea generation are the three factors of creative process engagement in Zhang and Bartol’s [8
Problem identification (PI). A three-item scale was used to measure problem identification. Response options ranged from 1, “strongly disagree,” to 5, “strongly agree.” The reliability of this scale was 0.81.
Information searching and encoding (ISE). In the same way, a three-item scale was used to measure information searching and encoding. Response options ranged from 1, “strongly disagree,” to 5, “strongly agree.” The reliability of this scale was 0.74.
Idea generation (IG). A five-item scale was used to measure idea generation. Response options ranged from 1, “strongly disagree,” to 5, “strongly agree.” The reliability of this scale was 0.84.
Control variables. Gender and tenure were two controls of participants in this study. Gender was coded “0” for female and “1” for male, and tenure represented the number of years that a participant had been in the company. In addition to individual-level controls, three firm characteristics were included: firm ownership, which was coded “0” for state-owned enterprises and “1” for non-state-owned enterprises, firm age, and firm size.
3.3. Participants and Procedures
With the help of local government agencies (e.g., Department of Commerce, Development and Reform Bureau), we collected data from 300 companies located in Zhejiang, Jiangsu, and Shanghai. Participants were selected from R&D-relevant divisions, and had considerable experience in innovation work. The participants were professional-level employees, such as hardware engineers and software engineers, because those jobs require substantive creative activities. In the first stage (T1), demographic information was asked for when participants reported their perception of organizational ambidexterity; 279 surveys were received (response rate 93%). In the second stage (T2), carried out about two months after T1, the 279 participants who responded were asked to report their perception of the encouragement of creativity by their supervisors for innovation activities and their level of creative process engagement; 253 responses were returned (response rate 90.7%). In the last stage (T3), two months later again, the 253 participants from T2 were required to evaluate their perceptions of innovation speed in their departments; 245 responses were received (valid response rate 96.8%). The procedure of data collection was from December 2015 to May 2016. In the valid surveys, females represented 20% while males constituted the remainder; 232 participants had worked for more than 3 years. As for firms, 14.3% had been founded for fewer than 3 years, 37.1% were founded between 3 and 5 years ago, 35.5% had been founded for between 5 and 10 years, 9% had been founded for between 10 to 20 years, and 4.1% had been founded for more than 20 years. Most of the investigated firms were small and medium-sized enterprises (SMEs), and the number of employees was fewer than 500. Regarding ownership, 31.4% were state-owned enterprises (SOEs) and 68.6% were not SOEs.
3.4. Method Choice
Structural equation modeling.
In this study, Amos 17.0 at ZJUT, China which developed by IBM, New York, United States, was used to examine the measurement model. The technique includes a set of suitable methods for both confirmatory factor analysis (CFA) and SEM through visualized modeling techniques. Compared with conventional linear regression, SEM allows comparisons among alternative theoretical models via indexes. In this study, indexes including the values of X2
goodness-of-fit test (X2
) comparative fit index (CFI), Tucker–Lewis index (TLI), incremental fit index (IFI), and root mean square error of approximation (RMSEA) are adopted to compare the goodness of fit among the proposed models to produce the best model. Regarding sample size, the 245 responses received exceeded the recommendation for a sample size of about 200 for SEM [51
]. Therefore, SEM is an appropriate tool for this study.
However, there are still some limitations of SEM. First, SEM emphasizes comparison among alternative models, rather than the effect size within each model. According to Preacher and Hayes [52
], the qualified model is a necessary, but not sufficient, condition for the effects in the model. While SEM captured effect size by path coefficients, it failed to detect indirect effect size within the model. Thus, this study adopted SPSS 23.0 at ZJUT, China which developed by IBM, New York, United States, and the PROCESS 3.0 macro developed by Andrew F. Hayes to capture the indirect effect size; that is, the indirect effect from leader encouragement of creativity to innovation speed through creative process engagement with organizational ambidexterity as moderator. Second, both tools mentioned failed to capture the variety or discrepancy of the causal conditions underpinning the same outcomes [53
]. To avoid the limitations of quantitative tools, qualitative comparative analysis (QCA) was introduced to this study to uncover the causal factors within innovation speed
Fuzzy-set qualitative comparative analysis.
QCA is a set theory-based method used for detecting sufficient but not necessary combination(s) of one or more variables leading to the expected dependent variable. The core principle of QCA is logic optimization, in which researchers could simplify combinations from cases and construct sufficient combinations with the fewest variables to outcome [49
]. Based on that, QCA enables researchers to gain various combinations linking the same outcome with a non-linear nature [56
To adopt QCA, users should convert the original data into membership data ranging from 0 to 1; membership of 0 represents the variable being existence in the given set, and membership of 1 represents the variable being present in the given set. Depending on the membership score in the truth table, QCA is divided into crisp-set QCA (czQCA) and fuzzy-set QCA (fsQCA) [58
]. Membership scores in czQCA are only binary data. However, fsQCA expands czQCA by permitting membership scores as continuous variables between 0 and 1. This improvement allows more underlying information to be retained. Additionally, although QCA is initially designed for small-N studies, in work by Fiss et al. [59
] QCA is adapted to larger-N analysis. Therefore, in conclusion, fsQCA is suitable for this study.
Noticeably, compared with SEM, there is one obvious limitation of QCA: there is no causal structure in its solutions. Because of the causal principles of set theory, the outcomes of QCA only show the sufficient configurations of the result, without inner relationships within one combination or among combinations [60
]. However, in this study, this limitation was partly reduced by combining the outcomes from SEM and fsQCA, so that the researchers could reveal more comprehensive and integral findings.