4 When Do Written Rules Persist in Routines? 82
5.4 Sample and Method 119
5.4.1 Sample Selection and Description
We collected data from one site of a large German chemical company. In line with the notion that employee creativity is crucial for firm competitiveness and survival in the chemical industry (Rammer, 2007), our sample company encour- ages employees to come up with creative ideas for improving products, services, and production processes by means of a suggestion system that has been in place for over thirty years. Representatives of the company told us that their suggestion system has been in place for this long time because the ideas submitted have a
significant positive economic impact for the company. All suggestions made by employees are evaluated by management representatives and representatives from the workers’ council (cf. Arthur and Aiman-Smith, 2001). Accordingly, the evaluation of ideas is strictly disconnected from the suggestion submitter. Each suggestion is scrutinized through a formalized decision process that is designed to produce a rational estimate of the company’s potential profit from implement- ing the suggestion. Employees receive the evaluation of their suggestions in their private company mail. If an idea is considered new and valuable and thus accept- ed for implementation, employees are rewarded by receiving a premium reflect- ing the idea’s economic benefit for the company. Suggestions made at the time of this study were complex and included proposals about how to increase the effi- ciency of operating routines (e.g., to reduce the amount of mercury needed and emitted in the production process), how to increase the flexibility of site logistic routines (e.g., a change in weighting procedures for return shipments of raw ma- terials), and how to increase the effectiveness of chemical-analysis routines (e.g., higher precision in chlorine analysis). As several employees confirmed, develop- ing such suggestions typically takes a considerable amount of time and effort.
To ensure support for our research, we had meetings with the company’s senior managers and employee representatives, presented our research design and ques- tionnaire, and offered a summary of the main study findings upon study comple- tion to the management. This procedure gave us access to the 51 teams compris- ing 299 employees working in the production and service areas of the company site under study. While production teams are responsible for running and improv- ing chemical and other operating routines, service teams are responsible for run- ning and improving the infrastructural routines. For example, service teams ar- range the logistic activities required to transport needed raw materials to the pro- duction site and chemical products to customers.
Employee participation in our study was anonymous, voluntary, and not incentiv- ized. In total, 267 employees (89.3% response rate) from 48 teams returned ques- tionnaires. We used Dawson’s (2003) selection rate to exclude groups with low group-level response rates from further analyses. Following earlier research (Richter et al., 2006), we chose a selection rate ([N-n]/Nn) of .32 as cut-off point, which suggests that the data from our sample correlated with true scores to .95 or higher. In total, six teams did not meet this cut-off point and were excluded from further analysis. The final sample thus included 42 production and service teams (218 respondents). On average, teams had eight members (SD = 3.25), and em- ployees had worked on their jobs for 14 years (SD = 10.49). Respondents were on average 44 years old (SD = 9.42). 204 (94%) employees had completed pro-
fessional schooling, and 24 (11%) master craftsmen training. 196 (90%) re- spondents were male.
In addition to the survey, we were allowed to use archival data on the suggestions made by those employees that took part in our study. Specifically, based on a matching process employing an anonymized code issued by the company, we received information on the suggestions made by employees in a two-year timespan, as well as information on whether the suggestions were rejected or ac- cepted, and the bonus payments that were granted for every suggestion that was accepted. For testing our hypotheses, we could thus rely on archival data indicat- ing prior creative failure, subsequent creative performance behavior, and the re- sulting creative outcome, in two subsequent years (year 1, year 2). The modera- tors, as well as several control variables, were based on survey data that were collected in the first half of year 2. All survey items used were translated and back-translated following the procedures described by Brislin (1980). Prior to the main study, we additionally discussed the entire survey with managers and em- ployee representatives and conducted a qualitative pre-test with employees that did not participate in the main study. This pre-test consisted of a think-aloud pro- tocol to receive structured feedback on the validity and comprehensibility of the questionnaire (Sudman et al., 2010). Following a consultation with employee representatives, the final pen-and-paper survey was distributed to all team mem- bers by the researchers. Finally, we conducted several in-depth interviews with employees from different teams in order to cross-validate the interpretation of our findings.
5.4.2 Measures
Dependent Variables. We followed recent calls for using objective indicators of
creativity (e.g., Amabile and Mueller, 2008) and relied on archival data from the firm’s suggestion system to capture our dependent variables. Consistent with pri- or work (Oldham and Cummings, 1996; Montag et al., 2012), we measured crea-
tive performance behavior by counting the suggestions an employee submitted in
year 2. Following Liao et al. (2010), we captured employee creative outcome by the sum of bonus payments employees received for accepted suggestions submit- ted in year 2.
Independent Variable and Moderators. To capture creative failure, we counted
the number of rejections each employee received for his or her submissions in year 1. Following earlier research (Nembhard and Edmondson, 2006; Detert and Burris, 2007; Tucker, Nembhard and Edmondson, 2007), we relied on a short- ened and context-adapted version of Edmonson’s (1999) scale to assess psycho-
logical safety. The items we used are: “Members of this team are able to bring up
problems and tough issues”, “No one on this team would deliberately act in a way that undermines my efforts”, “It is difficult to ask other members of this team for help (reverse-coded)”, and “If you make a mistake on this team, it is often held against you (reverse-coded)” Answers ranged from 1, “strongly disa- gree,” to 5, “strongly agree”. Items formed a single scale (Cronbach’s alpha = .72). We measured teams’ transactive memory system using Lewis’s (2003) 15- item scale ranging from 1, “not at all correct,” to 5, “completely correct.” A sam- ple item is, “I know which team members have expertise in specific areas.” Con- sistent with prior research (Lewis, 2004; Gino et al., 2010) the items formed a single scale (Cronbach’s alpha = .78).
As psychological safety and transactive memory are both considered team-level constructs, we assessed levels of inter-rater agreement within teams, as well as significant variance between teams (Bliese, 2000). Inter-rater agreement analyses by means of median rwg(j) tests across teams (James, Demaree and Wolf, 1984;
LeBreton and Senter, 2008) revealed adequate within-team agreement for psy- chological safety (.82) and transactive memory system (.95). One-way ANOVAs further revealed significant between-team variance for psychological safety (F = 2.68, p = .00) and transactive memory system (F = 2.98, p = .000). This result was confirmed by intra-class correlation coefficients for psychological safety (ICC[1] = .24, ICC[2] = .63) and transactive memory (ICC[1] = .29, ICC[2] = .66), which are acceptable for teams as small as the ones in our sample (James, 1982; Bliese, 2000; Klein et al., 2000). Overall, results suggest that the aggrega- tion of both measures to the team level is justified.
Controls. In all analyses, we controlled for several individual and team-level var-
iables that have previously been associated with employee creativity and team processes (see Zhou and Shalley, 2008, for an overview). At the individual level, we controlled for employees’ age, gender, job tenure, and risk propensity. We measured risk propensity with a six–item scale developed by Hao et al. (2005). A sample item is: “I am willing to take significant risk if the possible rewards are high enough.” Response categories ranged from 1, “strongly disagree,” to 5, “strongly agree”. Items formed a single scale (Cronbach’s alpha = .70). Addi- tionally, we controlled for employees’ prior creative performance behavior and
creative outcome by including the number of suggestions made, as well as the
number of suggestions accepted, and the bonus payments employees received for their accepted suggestions in year 1.
At the team level, we controlled for team size, team job tenure, and whether the team operates in a production or service setting. Additionally, we included a dummy variable indicating whether the team participated in a quality circle pro- ject which had been introduced in the beginning of year 2, as quality circles may shift the focus from individual-level to team-level suggestions. Finally, we con- trolled for prior experiences with the suggestion system on a team level by in- cluding the number of team members that submitted ideas and the number of team members that had ideas accepted in year 1.