Indicator-driven data calibration of expert interviews in a configurational study

Expert interviews can provide interesting data for the use in qualitative comparative analysis (QCA) to investigate complex social phenomena. To guide the challenging task of data calibration from qualitative data sets, techniques have already been suggested for the transformation of qualitative data into fuzzy sets. The current article follows existing guidelines and extends them with a system for indicator-based data calibration of expert interviews. While the underlying data set is confidential due to its corporate setting, in this article the analysis of the data is made transparent and hence reproducible for potential follow-up studies. First, the process of data collection is described, and the final data sample is characterized. Consequently, a system for indicator-based data calibration is presented and the calibration results for the empirical sample are provided in form of the set membership of cases and truth tables. • Data collection from expert interviews is described for a configurational setting • A combined indicator-based system is used for the calibration of qualitative data


Specifications
Area; Economics and Finance More specific subject area; Business research Method name; Data calibration for fsQCA Name and reference of original method; • Ragin, C. C. [1] . Redesigning social inquiry: Fuzzy sets and beyond. Chicago/London: University of Chicago Press. • Basurto, X., & Speer, J. [2] . 'Structuring the calibration of qualitative data as sets for qualitative comparative analysis (QCA)'. Field Methods, 24 (2), 155-174. doi: 10.1177/1525822 ×11433998 • De Block, D., & Vis, B. [3] . 'Addressing the challenges related to transforming qualitative into quantitative data in qualitative comparative analysis'. Journal of mixed methods research, 13(4), 503-535. doi: 10.1177/1558689818770061 Resource availability; The analysis is based on confidential primary data from interviews with corporate innovation managers that cannot be shared. The interview data was verified and triangulated via additional quantitative and qualitative data from participatory observations at conferences and desk-based research into the corporations and their activities.

Fuzzy set qualitative comparative analysis (fsQCA)
Qualitative comparative analysis (QCA) is a well-established field of methods to investigate causal configurations, originally developed and further refined by Ragin [1 , 4 , 5] . QCA enables causal investigations into social phenomena based on set theory and qualitative or quantitative information and has found wide application across research fields at the macro, meso, and micro levels. For example, in innovation research several studies have recently applied QCA with a focus on innovation systems [6 , 7] , innovation clusters [8] , innovation performance [9][10][11] , management innovations [12] , service innovations [13 , 14] and eco-innovations [15] . An important advantage compared to conventional methods is that QCA allows for equifinality, i.e., more than one causal path to an outcome [16] . Moreover, causality is directional and one-way and hence can be asymmetric [16] . For management and business research, Misangyi et al. [17] claim that QCA enables a "neo-configurational perspective" that is particularly promising for certain types of research, including studies on expected but unobserved strategy, strategic change, and managerial decision making. Hence, the present study conducts a QCA according to Ragin [1] and the best practices for strategy and organizational research defined by Greckhamer et al. [18] to investigate unobserved strategies among firms involved in research and development (R&D) for carbon capture and utilization (CCU) technologies. While QCA can be applied to different kinds of sets, fuzzy sets allow researchers to calibrate their data between non-membership (0.0) and full membership (1.0) in ordinal or continuous scales based on substantial knowledge and qualitative assessment [1 , 5] . Due to the complexity of the task, this article presents in detail the data collection process from expert interviews and the subsequent data calibration for the original research article [19] . Since data calibration is of paramount importance for the quality of QCA, this article follows the research techniques suggested by Basurto and Speer [2] and De Block and Vis [3] and further details an indicator-driven approach for data calibration in the specific research context.

Preparation of the interview guidelines
To prepare data collection, as a first step a guideline for the interviews with innovation managers (see Table 1 ) was prepared incorporating epistemological and methodological recommendations for expert interviews from Bogner et al. [20] and the theoretical concepts of a configurational system of innovations (see Naims and Eppinger [19] ). The guideline starts out with the collection of relevant personal information (section A) and company information (section B). Section C presents a definition of CCU and facilitates a discussion to reach a common understanding or uncover potential interview How are production inputs impacted by the CCU innovation? interview Where does the CO 2 for the CCU process come from?
interview What is the price of CO 2 ?
interview Which role will transport costs of CO 2 play? interview Is there a profit margin on CO 2 ? Which one?
interview What is the effect on efficiency of the production process? interview How will the profit margin of the CCU based product change (compared to conventional technologies)? interview Which topic has been missing? What would you like to add? interview discrepancies between the interviewer and the interviewed expert. Section D collects all relevant information on the status of the CCU projects within the firm. The subsequent sections were designed to harvest the experts' knowledge of R&D resources, results, policy conditions, and their expectations for economic progress: Section E collects information on profitability and production costs, section F on revenues, section G on intangible value, section H on Investments, section I on policy and external conditions, and section K on economic progress. The list of criteria and questions were formulated based on the research targets and the theoretical literature, in particular Grupp [21] .

Expert selection and interview process
Initial candidates for the expert interviews were representatives of firms identified from the participant lists and agendas of relevant scientific or business conferences and workshops on CCU (see Table 2 ) in which the authors participated between 2014 and 2017. Upon invitation, experts from only three companies declined to participate. Thus, the final sample sufficiently represents the available expertise on CCU in European-based corporations that actively (and publicly) engaged in CCU Each interview was prepared in advance based on participatory observations at the listed conferences ( Table 2 ) combined with desk-based research into the corporations and their CCU-related activities, their intangible assets, and economic performance. When analyzing the interviews, further questions could be addressed to the interviewees. The data collected in the interviews hence could be sufficiently verified and triangulated.
In a few cases, two experts were interviewed per company. Then, certain parts of the interview were split between interviewees depending on their respective knowledge and corporate functions; for example, a marketing expert answered the marketability questions while an R&D expert answered the investment questions. Since these experts always work in project teams, in those cases the individual responses of the separated sections were treated as valid for both experts. However, in other cases, two experts from the same company completed the entire interview, for example when both worked in R&D but in different business units. The section on economic progress was completed by all experts individually in order to cover the entire qualitative spectrum of their expectations for achieving growth and transformation goals based on their personal and context-specific experience and beliefs. We conducted the in-depth semi-structured interviews in person, or by video/phone call in German or English between 06/2016 and 03/2017. The interviews lasted between one to three hours. The dialogue was recorded and consequently transcribed.

Characterization of the final sample and the data
The interviewed experts are, overall, highly experienced professionals who serve diverse departments, including R&D, technology and innovation, environment and sustainability, public affairs, and marketing (see Figure 1 ).
The interviews examined the experts' knowledge of R&D resources, results, policy conditions, and their expectations for economic progress. Since all interviewees are involved in the management and/or advancement of corporate R&D projects, their knowledge of R&D resources and results is of very high quality. Furthermore, most experts are very knowledgeable about policy conditions relevant for R&D in CCU. While those experts from public relations or environmental departments often have a more detailed knowledge of policies, even those with a more technical R&D background were able to reflect on the marketability conditions of their work in detail. Despite their subjectivity, all expectations are shaped within a profit-driven environment with explicit or implicit innovation strategies. Hence, the experts' expectations provide valuable insights on the progress potentials of such innovations. Moreover, we analyzed public company data for the experts' firms to further characterize the sample. For this, the CO 2 intensity of the firm was calculated as the ratio of CO 2 emissions (including scope 1 and scope 2) to revenues measured in tCO 2 /m US$. This measure is common for environmental, social, and governance (ESG) stock market index evaluations, e.g., MSCI Inc. [22] . Intervals were classified based on the observed distribution of the sample as low when below 300, as medium when between 300 and 800, and as high when above 800. Moreover, the R&D intensity of the firm was calculated as the ratio of R&D expenses to revenues. Intervals were classified as follows: low is below 1%, medium is between 1% and 4 %, and high is above 4%. This is in line with the classification by the European Commission [23] except that the latter defines high R&D intensity as above 5%. In contrast, Grupp [21] defines high R&D intensity as above 3.5%. Since our sample only contains one firm with an R&D intensity between 3.5% and 5%, this was categorized as high, and the corresponding boundary set to values above 4%. Data on revenues, R&D expenses, and CO 2 emissions (including scope 1 and scope 2) is sourced from annual reports for the year 2017. Only for one start-up company, financial data for 2017 were unavailable and replaced by data for 2018. For two start-ups, emissions data were unavailable but were assumed to be in the low category. The analysis of the firms reveals three groups: (i) CO 2 -intensive firms with low R&D intensity, (ii) R&D-intensive firms with low CO 2 intensity, and (iii) firms with medium CO 2 intensity and low or medium R&D intensity (see Naims and Eppinger [19] ).
The presented attributes characterize the sample. Due to its small size systematic sensitivities of the result towards certain characteristics of the expert (background, experience) or the firm (e.g., size, R&D intensity) commonly determined in quantitative research cannot be determined. However, in the phases of data calibration and analysis these characteristics can be considered in addition due to the small size of the sample. Generally, the authors are very familiar with the sample and have extensive case knowledge common to qualitative research, which is useful in the phases of data calibration as well as during the interpretation of the results.

Data calibration
Data calibration is of paramount importance for the quality of QCA. The present study followed the technique suggested by Basurto and Speer [2] to transform qualitative interview information into fuzzy sets by identifying measures, anchor points, interview coding, summarizing data through classification, and assigning and revising fuzzy set values. The interview data were coded in MaxQDA software and summarized in Microsoft Excel. Subsequently, the data were calibrated. According to Ragin [1] "fuzzy sets […] are calibrated using external criteria, which in turn must follow from and conform the researcher's conceptualization, definition, and labeling of the set in question." For those conditions measured against several indicators, fuzzy set memberships were calculated based on a system of qualitative and quantitative indicators and thresholds. The groundwork for the system of indicators is described in the original article , which describes the configurational theorizing and relevant indicators from the literature. When available, established literature thresholds were chosen. However, the thresholds were often derived from qualitative observations in the interview data.  Table 3 details how investments were assessed indirectly based on a combined logic of indicator thresholds for the absolute and relative size of investments, the status of investment and technology readiness level (TRL), to ensure that the calibrated set sufficiently reflects the observed spectrum of commitments. Especially the differentiation between diverse investments and major investments required the analysis of combined indicators and multiple observations to separate those with a particularly high commitment from those that are "mostly" committed. Table 4 details how profitability was assessed indirectly based on logical combinations of the experts' judgments about production costs and revenues.

Calibration of intangible value (IV)
Relevant categories and indicators for measuring IV were derived from the literature, in particular Lev [25] . Table 5 details how IV was assessed as a continuous fuzzy set with the mean of the indicator groups patents, product & customer value, and public perception. Within the sub-indicator groups a median is calculated to level out outlier values from the interviews. Across the indicator groups a mean is chosen to weigh all three categories equally. The thresholds were partially derived from qualitative observations in the interview data, e.g., the number of patents per year. Table 4 Detailed calibration of profitability in a four-value fuzzy set.

Indicators
Selection of cut-off points Calibration of fuzzy set membership Production cost Cut-off points based on experts' qualitative statements on whether production costs are expected to:

Increased profitability (1 -fully in)
if one of the following combinations applies: Ü Revenue increases + decreased production cost Ü Revenue increases + constant production cost Ü Constant revenues + decreased production cost Constant profitability (0.67 -mostly in) if: Ü Constant revenues & constant production costs Ambivalent profitability outlook (0.33 -mostly out) if one of the following combinations applies: Ü Revenue increases + increased production cost Ü Revenue decreases + decreased production cost Ü Ambivalent production costs + constant / increasing revenues Ü Ambivalent revenues outlook + constant / decreasing production cost Decreased profitability (0 -fully out) if one of the following combinations applies: Ü Constant revenues + increased production costs Ü Revenue decreases + increased production cost Ü Revenue decreases + constant production cost Ü Ambivalent production costs + decreasing revenues Ü Ambivalent revenues outlook + increasing production cost Revenues Cut-off points based on experts' qualitative statements on whether revenues are expected to:

Calibration of policy conditions
The calibration of policy conditions was derived directly from the interview data. The statements of the experts on policies were calibrated to the degree they support or hinder CCU implementation as shown in Table 6 . Table 7 details how, to calibrate progress, we assessed the expectations of the experts concerning growth and transformation. As suggested by De Block and Vis [3] a cluster analysis assessed the spectrum of combinations for growth and transformation. Naims and Eppinger [19] illustrates the observed clusters and their interpretation, firstly as transformation winners and opportunists who both expect to benefit from CCU, and secondly those that do not expect to benefit, including transformation underdogs, pessimists, and impact sceptics. Hence, we calibrated the outcome progress using formulae summarizing the defined clusters, as detailed in Table 7 . Consequently, transformation winners are fully in the set, whereas impact sceptics are fully out of the set.

Calibration of progress
After completing the data matrix, all calibrations and thresholds were revised to improve their quality and consistency. The assigning of thresholds and degrees of set membership were made explicit, in accordance with the recommendations by De Block and Vis [3] for calibrating qualitative information. Through testing and revising with different thresholds, the calibrations were improved to allow for robust interpretations. Moreover, selected sensitivity checks revealed that small changes in the data assessment did not significantly impact the overall results of the analysis, since the formulae combine a multitude of indicators.

Results: set membership of cases
The result of the data calibration is the set membership of all cases summarized in Table 8 . While the raw data must remain confidential as agreed with the interviewed experts before the interviews, the calibrated data is anonymized and does not allow any identification on the individual cases. The IV fuzzy set is calculated as the mean of three indicator groups as follows:    (1)

Results: truth tables
Consequently, fs/QCA software was used to identify truth tables for the presence ( Table 9 ) and the absence ( Table 10 ) of the outcome from the calibrated data. The presented truth tables are hence the concluding result of data calibration of our empirical sample. Consequently, they permit the configurational analysis and interpretation described in Naims and Eppinger [19] .

Declaration of Competing Interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.