FsQCA in entrepreneurship research: Opportunities and best practices

ABSTRACT This paper offers step-by-step guidance and best practices for using fuzzy-set qualitative comparative analysis (FsQCA) in entrepreneurship research. We analyzed publications in which FsQCA was utilized as the primary methodological choice, focusing our scope on 10 top entrepreneurship journals. Our review reveals that inconsistencies exist regarding how to use the method, what processes to follow, and how to interpret results. Specifically, there is a need for an improved understanding regarding (a) considerations for performing direct calibration, (b) performing (necessity) analysis for both the outcome of interest and its negation, and (c) assessing the sensitivity of the resulting configurations to the selected threshold and cutoff values. To illustrate best practices, we conclude by analyzing GEM data from 43 countries to demonstrate how to best use the method, what to consider, and how the findings change if best practices are not followed.


Introduction
There exist numerous contributions applying configurational thinking as a research methodology in a wide range of disciplines, including entrepreneurship.However, in our opinion, step-by-step guidelines of the application of the method combined with state-of-the-art literature are lacking, specifically in entrepreneurship research.Thus, this paper aims to provide methodological best-practice guidelines on the application of the configurational thinking approach (CTA), specifically fuzzy-set qualitative comparative analysis (FsQCA) in entrepreneurship research.By employing this relatively new method in entrepreneurship, researchers can better understand the complexity of entrepreneurial phenomena.We hope to accomplish two goals: (a) provide step-by-step instructions on how to use the FsQCA technique in entrepreneurship research and (b) highlight some of the most relevant features (e.g., data asymmetry, equifinality, and interdependency among antecedent variables) of the method in entrepreneurship study.We also aim to share our views regarding current, but inappropriate, use of the FsQCA method in entrepreneurship research, which has led to restricting the understandability of the results.We begin by providing an overview of previous and current empirical research to demonstrate why CTA might be beneficial to entrepreneurship research.
To publish in the top entrepreneurship and management journals, traditionally, entrepreneurship researchers have focused on theoretical and practical contributions (Corley & Gioia, 2011).The core theoretical and methodological objectives in many of those studies propose a conceptual model and then evaluate the model to demonstrate a linear relationship between independent variables (IVs) and a dependent variable (DV), primarily using regression or multiregression analysis.While we acknowledge the substantial knowledge and contributions made via this approach, we suggest that entrepreneurship researchers also consider placing more attention on contingency arguments by exploring interactions rather than simply linear relationships (Parente & Federo, 2019).One potential approach is configurational thinking (Qualitative Comparative Analysis: QCA), an emerging alternative to conventional methods, which uses concepts from set theory and Boolean algebra.Qualitative comparative analysis is a comparative caseoriented approach (Wickham-Crowley, 1992) combining qualitative and quantitative research methods and can be used to evaluate existing theories and theory-testing research (e.g., Misangyi et al., 2017;Schneider & Wagemann, 2012).
Over the last two decades, there has been a surge in the number of studies using CTA as scholars used CTA to further understand and explore causal complexity (Misangyi et al., 2017;Nikou et al., 2019;Santos et al., 2021).Specifically, qualitative comparative analysis and its variant fuzzy-set qualitative comparative analysis (FsQCA) are increasingly applied in entrepreneurship research as they can be used not only to combine the advantages of qualitative and quantitative methods but also as a viable method for situations wherein the sample size is more limited (e.g., Beynon et al., 2020;Douglas et al., 2020;Kraus et al., 2018;Sahin et al., 2019).Unlike the traditional statistical approach that aims to obtain individual net effects of a specific variable, this approach aims to produce theories that allude to a pattern of multiple independent variables that together are related to a dependent variable (Ragin & Fiss, 2008).In other words, the outcome may occur based on combinations of several attributes.From a set theory standpoint, CTA takes a comprehensive approach (Miller & Friesen, 1984), assuming equifinality and complex causality (cf., Fiss, 2007;Payne, 2006;Woodside, 2015), allowing for the creation of typologies based on theoretical concepts (Yoruk & Jones, 2020).
The configurational thinking approach relies on two logics, complementarity and substitution, to understand a phenomenon and the ways multiple attributes are interrelated (Misangyi & Acharya, 2014).The complementarity logic focuses on a synergetic relationship (Milgrom & Roberts, 1992) and assumes the attributes' effects are mutually enhanced (Aguilera et al., 2008); whereas, substitution logic suggests that attributes can be substituted for one another to produce an outcome (Rediker & Seth, 1995).However, CTA considers the potential of complementarity and substitution among the conditions (attributes) to produce multiple combinations of conditions (i.e., conjunction) and to yield the same effect (i.e., equifinality).The concept of equifinality helps explain complex causal relationships by systematizing the analysis of CTA (Fiss et al., 2013;Fiss, 2011).As such, this method is appropriate for research in which the focus is on the assessment of causal complexity underlying a given phenomenon, such as entrepreneurial intentions (Fiss, 2007), entrepreneurial decision-making (Stroe et al., 2018), and entrepreneurial dynamics and entrepreneurial ecosystems (Muñoz et al., 2020;Vedula & Fitza, 2019).

Key aspects of the configurational thinking approach
Configuration theories are based on an understanding of patterns and combinations of conditions and how they, as configurations, lead to outcomes.In terms of CTA, a configuration is a specific combination of causal conditions that produce the outcome of interest (Rihoux & Ragin, 2009).In other words, configuration theories see phenomena as a collection of interconnected attributes (in terms of FsQCA: conditions) that must be understood as a whole, rather than as individual attributes or separable entities, to avoid reductionism (Meyer et al., 2005).The fundamental aspects of CTA are covered in the remainder of this paper, with a focus on the core features of CTA for analyzing a phenomenon (cf., El Sawy et al., 2010;Fiss, 2007Fiss, , 2011;;Ragin, 2000Ragin, , 2008;;Rihoux & Ragin, 2009).

Holistic and systemic perspective
From a CTA standpoint, it is not individual independent variables that are related to dependent variables but rather systemic patterns and combinations of causal conditions that lead to the outcome of interest.Furthermore, unlike traditional methods, which focus on determining which causal variable has the greatest effect, configuration theory focuses on how different conditions (variables) interact to produce the desired outcome.In other words, with this theory, researchers seek to uncover causal combinations or causal recipes.

Equifinality and conjunction
The configurational thinking approach focuses on the concept of equifinality, meaning researchers can obtain multiple distinct configurations of conditions leading to outcomes of interest (Nikou et al., 2019).Therefore, different causal recipes may yield similar outcomes (Rippa et al., 2020).Configuration theory also enables researchers to account for the concept of conjunction (viz., when the outcome of interest occurs from the interdependence of multiple conditions; Schneider & Wagemann, 2012).

Hypotheses versus propositions
In conventional statistical methods examining linear relationships, hypotheses are viewed as correlational expressions to unfold pathways.In contrast, in configuration theories, instead of hypotheses, researchers suggest propositions as causal receipts to show how combinations of conditions together lead to the outcome of interest.In addition, with this approach, researchers formulate propositions to specify whether a condition should be present or absent and whether a condition is a core or peripheral condition in the causal receipt leading to the outcome of interest.As a result, asymmetric and multidimensional causality can be handled with configuration theories.Finally, based on the concept of causal asymmetry in configuration theory, the combination of conditions that leads to the presence of an outcome can be distinct from those that lead to the absence of the same outcome of interest.In other words, conditions "found to be causally related in one configuration may be unrelated or even inversely related in another" (Meyer et al., 1993(Meyer et al., , p. 1178)).
The rest of the paper is structured as follows.In the following section we provide a literature review with a descriptive analysis of articles applying FsQCA from a selection of journals in the entrepreneurship domain.In the subsequent section we discuss some of the best practices of applying FsQCA making use of a data set from the widely used Global Entrepreneurship Monitor.Finally, we present some conclusions and future recommendations.

Literature review
To assess the current state of the academic literature in entrepreneurship, we conducted a critical literature review focusing on the contributions that make use of (Fs)QCA.To present a sufficient, and a comprehensive overview, we selected 10 highly ranked academic journals in entrepreneurship research.The selection was based on the top journal rankings as presented by Google Scholar for the entrepreneurship and innovation category.As it has been pointed out in analyses of academic databases, citations in Google Scholar, Web of Science, and Scopus are very similar to each other, particularly for the business and management category with correlations of >0.90 (Martín-Martín et al., 2018).Considering this similarity and that Google Scholar journal rankings are based on the h5 citation index, we believe this to be an appropriate basis for our journal selection.Journals with wider scopes beyond entrepreneurship (e.g.Journal of Business Research, Research Policy, Journal of Intellectual Capital) were not included.We focused only on the top 10 journals for which entrepreneurship is specified as the main focus and scope.Table 1 presents the frequency of articles making use of QCA tabulated by year and journal name.The collection of articles was completed in October 2022, so the number for 2022 includes counts for the first nine months of the year.We identified 76 articles in total, a visual representation of the yearly distribution of articles is presented in Figure 1.
To understand the research problems to which FsQCA has been applied, we collected the keywords, as specified by the authors of articles, to perform frequency analysis with the aim of identifying the core themes.To offer a more reasonable ranking of the keywords, we performed basic text pre- processing to combine keywords (expressions) that refer to the same concept.As a result of the analysis, we have identified the five most frequent keywords: (a) FsQCA ( 48), (b) entrepreneurial intention ( 12), (c) entrepreneurial orientation ( 8), (d) human capital (7), and (e) firm performance (6).We observe that the authors in most of the cases included FsQCA as a key component of the article worth mentioning in the keywords.The most frequently analyzed topics, not unlike general trends in entrepreneurship research, explore the determinants of intention (frequency count of 12; 20 percent of all the papers).We note that, while the five mentioned topics occur most frequently, there are many specific topics that are analyzed at least a couple of times in the articles.
It is safe to say that the use of FsQCA is not restricted to a very small set of problems.

Best practices using FsQCA in entrepreneurship research
In this section, we aim to explain the distinct procedures of performing FsQCA analysis using a data set from the domain of entrepreneurship.In the data analysis, we focus on entrepreneurial intention as an outcome variable in the entrepreneurship literature.This is consistent with the criterion we used to select the top 10 entrepreneurship journals, as this variable has been the most frequently considered outcome of interest in contributions applying the FsQCA methodology.The data used in the analysis is retrieved from the Global Entrepreneurship Monitor (GEM1 ), which provides access to the result of surveys conducted across the globe on issues related to entrepreneurship over more than a decade.Data from GEM has been analyzed continuously in the academic literature, including an analysis of African countries using FsQCA (Decker et al., 2020).In the analysis, we focused on the publicly available data from 2020 and considered all the countries (N = 43) for which data is available (the list of countries included in the analysis is presented in Appendix A).The antecedent variables (referred to as conditions) for the model building were selected according to their relevance to the entrepreneurial intention.The conditions (e.g., perceived opportunities rate, and perceived capabilities rate) are defined according to GEM as presented in Appendix B.
As we focused on the year 2020 (the most recent available at the time of writing this article), the number of observations in the final data set is equal to the number of countries considered-that is, 43 observations in total.
The different steps and procedures of the analysis were performed using the statistical programming language R, making use of the QCA package, the tool with the most complete set of functionalities available for performing (any type of) QCA data analysis, and related visualization and tests.Before we present the main steps of the FsQCA analysis, we note that traditional statistical methods used for construct reliability and validity should be utilized.This is not related to the FsQCA methodology itself, but related analysis should be performed when appropriate.Specific tests and tools include the calculation of the Cronbach alpha measure for internal consistency, average variance extracted (AVE), and correlation.These measures are especially important when the data is based on constructs created from the combination of indicators, which is the most frequently encountered data type in the FsQCA literature.However, in this editorial, we will not discuss how to perform those tests.

Calibration
After the basic statistical analysis, the first step is to perform data calibration.This requires the data to be transformed in such a way that the resulting values range in the [0,1] interval.This reflects the notion of membership: The higher the value, the more characteristic the condition (attribute) is for the specific data point.In crisp QCA, 0 and 1 are the only values used to represent no membership (the variable is not observed in the data point) and full membership (the variable is observed in the data point).As an extension of this binary evaluation, fuzzy membership values are used in FsQCA to express the notion that some conditions are only partially present when considering a data point.A membership value of 0.5 is of particular importance in this case, as it represents maximum ambiguity.Data calibration can typically be performed according to two main approaches: direct and indirect.In the direct calibration, one needs to first choose the values of the original range of the variable that should be transformed into 0 (full nonmembership), 0.5 (intermediate value), and 1 (full membership).This is typically done using statistical measures, such as specific percentile values.In contrast, indirect methods rely on expert judgment and the researchers' domain knowledge.For example, an expert in entrepreneurship could specify values based on decades of experience, such as any value higher than 80 percent for perceived capabilities classifies a country as having full membership in the set of countries with high Perceived Capabilities.
In general, it is recommended, if possible, to use direct calibration (Pappas & Woodside, 2021), as it makes the research more reproducible and generalizable to similar situations.In our analysis, as is commonly done, we used the 5 percent, 50 percent and 95 percentpercentiles of the original variables as thresholds for full nonmembership, intermediate membership, and full membership, respectively.The specific values for each variable are presented in Table 2.There are recommendations on when to deviate from these percentile values-for example, data collected using Likert scales or data that does not conform to normality assumptions, but in most cases the fifth, 50th, and 95th percentiles are appropriate choices for the thresholds.Taking the example of Perceived Capabilities, any value above 83.92 will be transformed to 1, while any value below 41.80 will be transformed to 0. Furthermore, 60 will be assigned the transformed value 0.5.The next decision to make relates to how we perform the transformation to any intermediate values.The original and still widely used choice is to use a linear transformation: Any value between 41.80 and 60.00 will be transformed into a value between 0 and 0.5 according to the proportionality of distances from the two threshold values.For example, 50.90 will be transformed to 0.25 = 0.5 × 0 + 0.5 × 0.5, as 50.90 = 0.5 × 41.80 + 0.5 × 60.00.
A different, more statistically motivated choice is to use slightly more complex transformation functions, such as a logistic function.This allows us to account for the typical (normal) distribution of datapoints by assigning more weight to the intermediate threshold: When a variable follows a normal distribution, on average, values tend to be close to the mean rather than to the tail of the distribution.Figure 2 illustrates the outcome of the transformation as performed by the calibration function of the QCA library in R, using the identified threshold values and the choice of logistic function.

Necessity analysis
fhe next step of the analysis is to identify attributes that are required for the outcome of interest to take on a high value.This is termed as a necessary  condition in FsQCA analysis.Identifying an attribute as a necessary condition would imply a very strong relationship according to the definition; for example, identifying fear of failure as a necessary condition would mean that country can only have a high average rate of entrepreneurial intention when the rate of fear of failure is also high.Looking at it from another perspective, it would mean that if we find a country with a low rate of fear of failure, we conclude the the level of entrepreneurial intention cannot be high.As we typically analyze overly complex phenomena that can be characterized through the interrelationships of several variables, in practice it is rarely the case that we can find a necessary condition.However, it is always a crucial step to confirm the (lack of) existence with appropriate tests.The results of the necessity analysis are presented in Table 3.
To determine whether an attribute is a necessary condition and if it is what its scope is in terms of the data points explained by this relationship, consistency and coverage measures are calculated.Consistency values higher than 0.9 indicate the presence of a necessary condition as a general recommendation (Ragin, 2008).Coverage captures the importance of the relationship; the lower the relationship, the smaller the number of cases to which the identified relationship applies.As highlighted by the property of asymmetry, this and the following steps of the FsQCA analysis are performed separately for both the outcome of interest and the lack of it.As discussed earlier, we do not, in general, expect that by simply converting any configuration to its opposite we will obtain a good explanation for the opposite of the outcome of interest.
In Table 3, we present the results for identifying necessary conditions for both high and low levels of continued intention.By the same reasoning related to asymmetry, we do not assume that a condition being necessary (not necessary) implies that the opposite of the condition is not necessary (necessary).For this reason, all the calculations need to be performed for both conditions and their negatives with respect to the outcome and its negation.As we can observe in this specific case, we do not have any individual condition that could be identified as necessary for achieving high or low entrepreneurial intention in a country (see Table 3).This confirms that the phenomenon we are investigating is complex enough to warrant moresophisticated analysis.

Truth table
The next step of the analysis is to identify all the possible combinations of conditions that are present in the data and evaluate how each combination is consistent with the outcome.In other words, when assessing whether "the combination leads to the outcome" we determine whether the configuration is sufficient to lead to the outcome of interest.As the basis of performing the sufficiency analysis, identifying the configurations that are sufficient to result in the presence/absence of entrepreneurial intention, one needs to construct a truth table.This requires calculating the frequency of all the possible combinations for the presence (over 0.5) and absence (below 0.5) of the conditions.As we have five conditions (OPP, CAP, FEAR, STATUS, CAREER) and we need to consider both the presence and the negations of the conditions, there are 2^5 = 32 possible combinations.As we have data for 43 countries, we cannot expect to have all the theoretically possible configurations present, but this data should be sufficient to allow for observing the most-relevant ones.In the data, we found at least one corresponding country for 19 of the 32 possible configurations.In the most frequent combination, all five conditions (attributes) take on a high value; seven countries are characterized with this configuration.
After the truth table is constructed, for each configuration we need to assign a label indicating whether the configuration corresponds to the presence or absence of the outcome of interest.To rigorously determine this label and perform systematic calculations, consistency measures can be utilized.The value of consistency quantifies the extent to which a given configuration "agrees" (cooccurs) with the high/low values of the outcome variable.While there is a widely employed cutoff value (0.75: Ragin, 2008), as consistency is not typically associated with any corresponding statistical significance test, it has been pointed out in the literature that the optimal consistency value should be determined after carefully considering the underlying data set.In this study, as the number of data points is not significantly higher than the number of possible configurations, one would expect moderately higher sensitivity in the results.The cutoff value was set as 0.75 for both the presence and absence of entrepreneurial intention.Furthermore, we set the frequency cutoff for configurations to be included in the analysis as 1, to retain a sufficient number of cases.This means that we consider information of a configuration even if it only appears once in the data; this is common in small-N FsQCA analysis, with sample sizes of less than 50.The results of the analysis are presented in Table 4, for the high and low values of entrepreneurial intention.In the table, • and ◌ stand for the presence and absence of a condition in the configuration and present the so-called parsimonious solutions.
In FsQCA, three possible sets of solutions (complex, parsimonious, and intermediate solutions) can be calculated based on how we interpret the configurations missing in the data and our assumptions based on the available general knowledge.The complex solutions present all the possible combinations of conditions without any prior or further assumptions on how we should deal with configurations that are not present in the data or our knowledge of the domain.Because of this, one can typically obtain a large number of complex solutions that makes it difficult to interpret and extract useful and generalizable insights.Parsimonious solutions present a simplified version of the complex solutions and can be seen as the combinations of the most important conditions that cannot be left out from any solution.Indeed, as it can be proven mathematically, every single complex solution has a corresponding parsimonious solution.Without going into detail on the calculations to arrive at parsimonious solutions from complex ones, the general idea is to consider and make use of any counterfactual configurations that can contribute to logically simplified solutions.A different approach to simplifying complex solutions is to make use of the available general domain knowledge and, based on that, use only the counterfactuals (configurations not present in the data) that can be assumed to be reasonably consistent with theoretical and empirical knowledge.
An important, but often neglected, task to be performed after generating the solutions is to perform sensitivity analysis and assess the robustness of the results.In practice, one attempts to see whether changing various parameters and threshold values chosen in the analysis will change the results and what the extent of the change is, if any.Table 5 summarizes the three most important parameters that a researcher needs to determine for each individual  analysis and what the recommended values are for each step, as we discussed above.When using FsQCA, it is especially vital to test these values because, unlike statistical approaches in which similar threshold values are calculated based on precise statistical measures of the data, we are working with general guidelines and recommendations.As the analysis of the articles in the literature making use of QCA shows, authors typically rely on their domain knowledge and skill and specifically rely on the recommendations by Ragin (2000) when determining these threshold values.In most cases, the problem of threshold value selection and sensitivity analysis is typically not evaluated fully and in-depth.
Considering our study, the focus of assessing the validity of the results should be on the consistency threshold value.Regarding the necessity analysis, the highest value we obtained according to Table 3 is 0.8, which is significantly lower than the recommended 0.9, implying that indeed there is no necessary condition for either high or low levels of entrepreneurial intention.
Regarding the frequency threshold, as we deal with a small data set, increasing the cutoff value would mean losing close to 30 percent of the data (configurations that occur only once in the data set), a loss too large considering the resulting data size, number of variables, and proportion of configurations the remaining data would cover.To test the obtained solutions, we experimented with various consistency threshold values ranging from 0.70 to 0.80 to assess the sensitivity of the results with respect to the threshold selection.If the configurations vary considerably due to a slight change in threshold value, it indicates that the configurations are not stable enough.The sensitivity analysis shows that we can divide the threshold range into three intervals: • Between 0.72 and 0.75: no change, we obtain the two solutions as presented in Table 4 • Between 0.76 and 0.80: we obtain a single solution, Solution 1 from Table 4, highlighting the core importance of capabilities • Between 0.70 and 0.71: we obtain a third solution in addition to those in Table 4 (the extra rule indicates that low perceived opportunities, low fear of failure, and high good career choice are sufficient to achieve high intention) As these results show, the obtained configurations are sufficiently stable.Specifically, we have a configuration, involving the single condition perceived capabilities that is always present, independent of the threshold used, so this should be considered the main finding of this analysis.Furthermore, perceived opportunities and fear of failure play a clear role too.Please note that in this case when performing FsQCA, there are several conditions (attributes), such as STATUS, that do not appear in any of the configurations even when we change the threshold, which is an interesting, but not completely expected finding based on the recognition of this attribute in the literature.

Conclusions
In this paper, we aimed to offer twofold insight by reviewing the current entrepreneurship literature using FsQCA and discussing the features and use of FsQCA in this domain.According to the findings of a literature review of 10 major journals in the field, FsQCA is becoming a more widely used methodology in entrepreneurial research.According to the literature review, the number of articles using FsQCA has increased dramatically in recent years, as has the diversity of issues studied.The most frequently investigated topics in the reviewed articles where CTA was applied, as shown by the analysis of the extracted keywords, are entrepreneurial intention, orientation, human capital, and firm performance.These research topics are in the core of entrepreneurship research and have been investigated numerous times in the past.As such, researchers were able to generate novel insights making use of the benefits offered by FsQCA.As previously stated, when compared to traditional statistical models, FsQCA provides a comprehensive approach that allows for the construction of nonlinear models that include equifinality.We can anticipate, based on the trend depicted in Figure 1, that this popularity will continue to grow in terms of number of articles and range of research problems analyzed.
As the number of researchers using FsQCA grows, it is more important than ever to follow the preprocessing data and perform analysis in a rigorous way.In this work, we set forth the underlying steps of using FsQCA by applying it to one of the most widely used open data sets in entrepreneurship research-the Global Entrepreneurship Monitor (GEM).We have selected the variables that are used most frequently in the literature to explain entrepreneurial behavior as the outcome and made use of the most recent data from 43 countries.A step-by-step description is offered to highlight the recommended procedures to perform data calibration and necessity analysis and to determine sufficient configurations.We discussed step-by-step procedures for performing FsQCA analysis that are frequently overlooked and/or oversimplified in several articles.The major inconsistencies found in the reviewed articles were (a) considerations for performing direct calibration, (b) performing (necessity) analysis for both the outcome of interest and its negation, and (c) assessment of the sensitivity of the resulting configurations to the selected threshold and cutoff values.Furthermore, we have obtained results that can be seen as novel considering the state-of-the-art understanding of entrepreneurial intention.
Based on the articles considered in the literature review, we can make the following final observations and recommendations for prospective users of FsQCA in the entrepreneurship domain: • Calibration: the main recommended approach is direct calibration, making use of the empirical data and conceptual knowledge when available.As we found in the literature review, most of the papers already made use of direct calibration.However, it is still important to highlight the importance of making use of statistical measures (e.g., quantiles), rather than using predetermined values, such as when calibrating Likert-scale data from questionnaires (most of the papers).Furthermore, researchers can easily make use of more-complex transformation functions, such as logistic, instead of using basic linear membership functions, as appropriate tools are implemented, such as the QCA package in R.
• Necessity analysis: a crucial step in performing FsQCA to establish whether there is/are individual condition(s) without which the outcome of interest cannot take place.While it is rarely the case that we find a necessary condition, it is crucial to perform this part of the analysis as failing to correctly use a necessary condition in the subsequent sufficiency analysis can result in incorrect configuration solutions.In fact, this is one of the key problems we identified in the literature: In 20 papers (1 out of each 3 included in the analysis), the authors failed to perform (or at least to report) necessity analysis.For this reason, we encourage prospective users not to neglect this component of the analysis.• Truth table and sufficiency analysis: this is the main step in determining the relevant configurations, including some issues that require the attention of researchers.The choice of parameters (frequency cutoff and consistency threshold) can impact the results significantly.While there are some recommended choices, as presented in Table 5, researchers always need to rely on the specific features of the data (size and distribution of variables) instead of making use of the generic recommendations without further consideration.At a minimum, we recommend that researchers perform a sensitivity analysis and robustness test, by looking at how changes in the initial parameter values impact the final solution, to avoid presenting configurations that are just "random noise" in the data.
In the analyzed literature, we found fewer than 10 papers of the 60 that performed any kind of sensitivity analysis, implying that this is definitely an issue requiring much more attention from researchers.• Solutions: output from the sufficiency analysis is referred to as a solution, and FsQCA provides three different solution sets: parsimonious, intermediate, or complex.As a result, researchers must take into account a number of crucial factors.For instance, it is critical that authors make clear which of these they are reporting.This seems like an obvious statement, however, we found that 14 out of 60 papers (almost 25 percent), did not include any information on the solution type used.Second, while it can be reasonable to use any of the solution sets, recent results (Baumgartner & Thiem, 2020) indicate that, as a default choice, a parsimonious solution should be used, as the other solution types can be incorrect.Third, when applying intermediate solutions, authors need to specify what additional assumptions (general knowledge) they used as an input to construct the configurations that supplement the empirical data.Without this, intermediate solutions do not provide anything new compared to other solution types.In the literature review, there were close to 20 articles in our sample that claimed to present intermediate solutions; however, we found that only two of these specified the additional assumptions and general knowledge they considered in the analysis.
We hope that the presented insights and recommendations will further increase the popularity of FsQCA in entrepreneurship research and that subsequent studies will utilize FsQCA in a rigorous manner.
Figure 1.Year of publication vs. number of publications.

Table 2 .
Calibration of the conditions.

Table 3 .
The results of the necessity analysis.

Table 4 .
The FsQCA results leading to high/low entrepreneurial intention.

Table 5 .
The default parameter choices to be considered when using FsQCA method.