*3.2. Methodology: Qualitative Methods*

We used a qualitative method to test PRAC (hypotheses H1 and H2 in Model 1; hypotheses H3 and H4 in Model 2; hypotheses H5 and H6 in Model 3). Osabutey and Jin [54] showed that traditional quantitative methods have limitations when it comes to explaining complex interactions between variables. In addition, other recent studies applied qualitative methods. For example, Oyemomi et al. [55] used fsQCA. FsQCA is a qualitative method that identifies the essential and necessary conditions of the configurations that will lead to the outcome variable or its absence [56]. Therefore, this study completely covers all pathways for PRAC.

In this paper, we used fuzzy set QCA (fsQCA) from [57] to evaluate these hypotheses. There are many applications of quantitative methods in managemen<sup>t</sup> and business [37] but few applications of qualitative methods and mixed methods.

In similar fields to management, fsQCA has been applied instead of quantitative methods (PLS, SEM and others) and also as a complementary method to quantitative methods [58,59]. Any correlational method in general assumes symmetrical relations between variables and measures the net effect of each variable on the assessed output. On the other hand, fsQCA allows discovery of the combinatorial effects of variables on the output as well as accepting that these interactions could be asymmetrical [58].

Therefore, in this paper, we applied fsQCA and found the logical implication that combining the presence/absence of input variables provides better output results. Consequently, consistency and coverage measures inform the relevance of the discovered logical implications.

#### *3.3. Sample and Measurement Assessment*

An online survey was conducted in Qualtrics® and sent to 6846 innovative SMEs in Portugal within twelve industrial sectors: manufacturing, energy supply and gas; water supply and pollution; edifice; trade and repair of vehicles; transportation and storage; catering; information and communication; accommodation; scientific activities; administrative activities; health activities; and other services. In total, 385 responses were obtained. After applying a rigorous cleaning process [60], the final sample constituted 349 firms. Therefore, the response rate was 5.1%. The sample included firms from the twelve industrial sectors existing in the population.

The questionnaire was originally written in English and was later translated into Portuguese by an expert translator for sending to firms. It was then back-translated into English. In this way, errors related to language interpretation were minimized. In addition, to carry out the questionnaire, five academics and managers who are experts in the field were contacted to show them a pilot questionnaire and subsequently launch the final questionnaire. Finally, the firms were called by telephone to inform them of the study and the questionnaires were then sent to them. It is a simple and quick survey to answer that takes an average of 20 min to answer, and the people surveyed were the CEOs in each firm. Our goal was for CEOs to respond to the questionnaire because they are the top decision-makers and know all the tools and information necessary to try to achieve economic and environmental development in the firm [61].

The characteristics of the sample are detailed below. The majority of CEOs surveyed were women (56.4%), and more than 76.5% had undergraduate or graduate degrees. Their average age was 43.6 years and more than 77.4% had more than five years of seniority in their firms. In relation to the characteristics of the firm, more than 92.4% had more than 10 years of experience and the majority (65.9%) had 50 employees or fewer. Of them, 63.6% were public limited companies and 36.4% were collective firms. On the other hand, to test the non-response bias, a trend extrapolation test was used to compare the responses that took the longest to arrive with those that arrived first. Our study considers that the responses that took the longest to arrive are those that arrived in the second phase of the study and after carrying out a reminder. These responses may be very similar to those of those firms that never responded since, had it not been for the reminder, they might

never have responded [62]. After conducting a one-way analysis of variance (ANOVA), we could see that there were no significant differences between CEOs who responded early and those who responded late in terms of company size (number of employees in the firm) and age.

Therefore, the sample was representative of the population.

## *3.4. Variables and Measurements*

HRC measures the total expenditure on human resources per year as a proportion of the firm's total invoicing [63].

All other measurements used Likert-type scales [64]. The ranges were from 7 (strongly agree) to 1 (strongly disagree). To measure OLC, we used a scale validated by Alegre and Chiva [65]. This scale has five dimensions: risk-taking, experimentation, participatory decision-making, interaction with the external environment and dialogue. To measure ITS, a validated scale by Lee and Choi [18] was used, and finally, PRAC was measured with the scale by Molina-Azorín et al. [66]. Therefore, we had the construct OLC with five dimensions and the two constructs ITS and PRAC with one dimension.

The level of education of the manager (EL) was measured taking into account what type of education they have: high school, undergraduate and graduate. The experience (EXP) of the manager was measured according to the number of years in the firm: junior, <2 years; intermediate, 2 to 5 years; senior, >5 years. Finally, the size of the firm (SIZ) was measured according to the number of employees: small: <50 employees; medium: >50 employees [67,68].

A confirmatory factor analysis (CFA) was used in AMOS® to study the validity of the measurement. All of the items showed good levels; therefore, there is a good validity in the measurement.

Table 1 summarizes the variables' descriptions and the CFA.

#### **Table 1.** Variables' descriptions and CFA.


The survey was designed to reduce common method bias (CMB) [69]. We used Harman's single factor test to evaluate the existence of CMB. No evidence of CMB existed.

#### *3.5. Fuzzy Set Qualitative Comparative Analysis (fsQCA)*

As demonstrated in this study, by using fsQCA [52], we can obtain more than one configuration or causal pathway that leads to PRAC. In this way, we can find a set of alternative causal configurations that a firm can use to arrive at PRAC. In our study, the causal pathways in fsQCA that could lead to PRAC corresponded to combinations of the following variables: HRC, OLC, ITS, EL, EXP and SIZ. Since OLC has five dimensions (OLC-E, OLC-R, OLC-I, OLC-D and OLC-P), it was measured using the fsQCA "fuzzyand" function. This function corresponds to the mathematical logical operation in Boolean algebra called "intercept." Therefore, when OLC appears in a causal configuration, it means that this condition is a five-dimensional cumulative condition. The end result will be PRAC.
