Prior empirical evidence on the relationship between innovation capabilities and firm performance remains inconsistent, leaving unresolved questions regarding the magnitude of this association and the conditions under which it varies. This study addresses this gap by conducting a cross-national meta-analysis of 54 independent samples from 19 countries (N=320,529). Grounded in Schumpeter’s hypothesis that innovation is a primary driver of economic growth, we distinguish between the Mark I. pattern (characterized by creative destruction) and the Mark II. pattern (characterized by creative accumulation). In support of Schumpeter’s hypothesis, our findings indicate that, on average, innovation capabilities exhibit a positive, small-to-moderate, and statistically significant association with firm performance. When decomposing Schumpeter’s hypothesis into its ‘narrow’ forms, the relationship is strongest for Mark I. and for studies combining both Mark I. and Mark II. patterns, whereas evidence for the Mark II. pattern alone suggests no meaningful performance effect. Additional analyses show that heterogeneity across studies is partly explained by contextual and methodological moderators, including the proximity of capability measures to performance outcomes, performance indicators, and study design characteristics. By synthesizing a broad base of empirical evidence and identifying the conditions under which innovation capabilities are associated with firm performance, this study provides a more precise and generalizable understanding of the innovation capabilities–firm performance relationship and strengthens the empirical foundations of the Schumpeterian framework.
Since the early 1940s, researchers and practitioners have revisited the Schumpeterian claim that innovation capabilities enhance, rather than obstruct, firm performance (Cevik, 2025; Guerrero & Siegel, 2024; Schumpeter, 1934). This relationship remains central to contemporary debates on competitiveness, as firms increasingly rely on their ability to sense, seize, and transform innovation opportunities in dynamic markets (Moreira et al., 2024; Teece et al., 1997). Recent studies continue to highlight the strategic importance of innovation capabilities as a drivers of firm performance (López-Cabarcos et al., 2019; Piñeiro-Chousa et al., 2025), even as the literature continues to report conflicting evidence (Mendoza-Silva, 2021). Although innovation may entail substantial costs and uncertainty (Malik et al., 2022; Subramaniam & Youndt, 2005), Schumpeterian hypotheses suggest that firm-level innovation capabilities generate competitive advantage through processes of creative destruction (i.e., Mark I) and creative accumulation (i.e., Mark II).
Despite extensive empirical work, research findings remain inconsistent. For example, patent-based innovation capabilities have been reported to have significantly negative (DeCarolis & Deeds, 1999, pp. 963, model 3. and 4.), positive (Arora et al., 2014, oo. 332, model 2.), or nonsignificant effects (Kang et al., 2017, pp. 1364, model 1., 2., and 3.) on firm performance. Although several systematic reviews exist (Iddris, 2016; Mendoza-Silva, 2021; Moreira et al., 2024), no study has yet synthesized this fragmented evidence meta-analytically, as indicated by the absence of published meta-analyses, leaving the magnitude and variability of the innovation capability–firm performance relationship unresolved. This gap is particularly significant because inconsistent empirical findings call for systematic synthesis to clarify the true magnitude of the effect of innovation capabilities on firm performance (Piñeiro-Chousa et al., 2020).
The aim of this study is to quantitatively determine the overall magnitude of the relationship between innovation capabilities and firm performance, and to explain why and when (i.e., under what conditions) this relationship varies across studies. To achieve this aim, we employ meta-analysis as the most appropriate method. Specifically, we integrate empirical findings across diverse dimensions of innovation capabilities, performance metrics, and contextual characteristics. Our study evaluates: (i) the overall effect of innovation capabilities on firm performance, including distinctions between the Mark I and Mark II patterns of Schumpeter’s hypothesis; and (ii) the extent to which differences in capability proximity, performance measurement, cultural context, theoretical framing, and study design explain variability in prior findings.
By doing so, our research addresses the following questions: What is the overall association between innovation capabilities and firm performance, and how various theoretical and methodological factors influence Schumpeter’s hypothesis? To answer this question precisely, we conducted a systematic search of studies published between 2001 and 2024, yielding 54 independent samples and 320,529 firm-level observations. We followed the latest ‘Preferred Reporting Items for Systematic Reviews and Meta-Analyses’ (PRISMA) guidelines and developed a four-step approach (Page et al., 2021). Effect sizes were extracted and aggregated using random-effects meta-analytic techniques (Hunter & Schmidt, 2000). We also performed a series of moderator analyses to examine differences across capability types, proximity levels, performance measures, and study-level attributes (Cooper et al., 2019). A summary of our results indicates that, on average, innovation capabilities are positively associated with firm performance. This relationship is particularly strong for Mark I patterns, proximal capabilities, and ex-post performance measures, while study design and measurement choices explain additional heterogeneity.
This research aims to make two distinct contributions. First, it advances theoretical understanding by clarifying both the magnitude (i.e., how much) and the conditions under which innovation capabilities are associated with firm performance, thereby refining the Schumpeterian hypotheses with large-scale empirical evidence. Second, it provides the first meta-analytic assessment of this relationship, offering a methodologically robust and generalizable foundation for future theorizing and practice. Practically, the findings guide scholars and practitioners on how to effectively leverage innovation capabilities to enhance firm performance, taking into account the type of Schumpeterian hypothesis, capability type, and performance objectives.
The paper proceeds as follows. First, we review the literature on Schumpeter’s hypothesis and introduce the concepts of innovation capabilities, the proximity of innovation capabilities, and their implications for different dimensions of firm performance. We then develop arguments suggesting that the effect of innovation capabilities on firm performance is likely moderated by factors such as the sample’s uncertainty avoidance, the theoretical frameworks employed, and study design. Next, we describe our search strategy, the samples selected, and the meta-analytic apparatus used in our research. Finally, we report our findings and discus their implications for both scholars and practitioners.
Literature Review and HypothesesA substantial body of research in innovation management, strategy and entrepreneurship highlights innovation as a central mechanism through which firms create and sustain competitive advantage (Guerrero & Siegel, 2024; Piñeiro-Chousa et al., 2020). This literature commonly argues that innovation enables firms to differentiate themselves, respond to market changes, exploit new technological opportunities, and ultimately enhance both financial and competitive outcomes (López-Cabarcos et al., 2019). Yet, despite widespread agreement on the importance of innovation, scholarly perspectives diverge regarding how and under what conditions innovation capabilities lead to superior performance (Wang et al., 2024; Yin et al., 2023). Competing theoretical lenses: such as Schumpeterian views of creative destruction and creative accumulation (Schumpeter, 1934), dynamic capabilities framework (Teece et al., 1997) and resource-based arguments (Barney, 1991), offer differing explanations concerning the origins, nature, and performance consequences of innovative activity. These perspectives suggest that the relationship between innovation capabilities and firm performance is heterogeneous rather than universal, shaped by firm characteristics, capability configurations, environmental conditions, and measurement approaches. Therefore, a systematic theoretical foundation is needed to clarify the specific mechanisms through which innovation capabilities influence firm performance and to motivate the Schumpeterian hypotheses that follow.
Schumpeter’s HypothesisThe Schumpeter’s hypothesis suggests that innovation is the primary driver of economic growth (Schumpeter, 1934). We build on this logic and explain that innovation is the primary driver of competitive advantage at the firm-level. That is, firms that successfully develop and implement new products, processes, or business models can achieve superior performance by disrupting existing markets or improving efficiency (Minviel & Ben Bouheni, 2022). Innovation is therefore source of success but also a mechanism for structural change in the economy, creating a dynamic relationship between entrepreneurial activity, market competition, and economic development (Cevik, 2025; Yin et al., 2023). Therefore:
Hypothesis 1 Innovation capabilities are positively associated with firm performance.
The Schumpeter hypothesis is commonly articulated in two forms. Firm size is an extremely robust and important moderator that influences how innovation capabilities translate into firm performance (Hall et al., 2010). Schumpeter Mark I (i.e., creative destruction) emphasizes the role of small, entrepreneurial firms in driving radical innovations that disrupt established markets, highlighting competition through novelty rather than price (Guichardaz & Pénin, 2025). On the other hand, Schumpeter Mark II (i.e., creative accumulation) focuses on large, established firms, arguing that their resources, R&D capacity, and market power enable sustained innovation, although innovation may eventually diminish with excessive size or bureaucracy (King & Levine, 1993; Sevil et al., 2022). Together, these forms capture the dual view that both small startups and large incumbents can be engines of innovation, but through different mechanisms and under different conditions (Guerrero & Siegel, 2024).
Thus, effect of innovation capabilities on firm performance may depend on firm size (Hall et al., 2010). Smaller firms (Mark I) innovate through agility and disruption, while larger firms (Mark II) rely on resources and structured R&D (Sevil et al., 2022). This suggests the strength of the innovation–performance link varies by firm size. Beyond the impact of the innovation capabilities collectively in Hypothesis 1, we expect the magnitude to differ by firm size. Therefore:
Hypothesis 2 Innovation capabilities are positively associated with firm performance, with the strength of this relationship differing between Mark I and Mark II patterns.
Given the existence of several systematic reviews on the innovation capabilities literature the following findings have emerged. Iddris (2016) found that studies from 2000 to 2015 on innovation capabilities are primarily investigated at the firm-level (90 % of the studies) followed by network or supply chain levels. Furthermore, Mendoza-Silva (2021) reports that majority of empirical studies are subject to two clear biases. First, the studies are influenced by selection bias (i.e., their primary focus is on firms sampled from developed countries). Second, in most studies, the methodological apparatus is influenced by common method bias (i.e., researchers used subjective data collection method to obtain data from questionnaires). This suggests that future research should investigate measures of innovation capabilities by considering both objective indicators (e.g., patents) that are verifiable, as well as subjective indicators (e.g., managerial ratings) that are self-reported. Next, Moreira et al. (2024), based on the studies published until the end of 2022, concluded that there is no one-size-fits-all conceptualization of innovation capabilities dimensions. They suggest that subsequent studies should investigate the ontological, such as the proximity of capabilities we argue here, and axiological dimensions surrounding the intricacies of the innovation capability domain.
Innovation capabilities (at the firm-level) are not simply the sum of individual or interpersonal capabilities (Salvato & Vassolo, 2018). They reflect the capacity to leverage resources, knowledge, and organizational processes to transform inputs (e.g., R&D investments) into outputs (e.g., patents) that enhance performance and competitive advantage (Piñeiro-Chousa et al., 2025). While individual actions influence innovation, firm-level capabilities emerge from dynamic interactions, coordination, and synergy within teams and management, rather than from any single contributor or metric (López-Cabarcos et al., 2019).
This ontological perspective suggests that innovation capabilities are positioned along three dimensions based on their proximity to firm performance. Proximal capabilities are closely linked to measurable innovation outcomes (e.g., patents) and directly reflect a firm’s ability to generate value from innovation (Rubera et al., 2016; Wei & Lau, 2010). Distal capabilities (e.g., investment in knowledge acquisition), on the other hand, represent inputs or antecedent activities that support innovation from distant mechanism compared to proximal capabilities (Menor et al., 2007; Simonin & Özsomer, 2009). While distal capabilities are critical for building the foundation for innovation, their impact on performance is indirectly linked and may depend on how effectively they are translated into tangible innovation outputs. This suggests that there is a stochastic mechanism, where distal capabilities translate into proximal (Stam & Elfring, 2008; Tse et al., 2017; Wu et al., 2020). This translation process is captured by stochastic capabilities, which reflect the combined effect of distal and proximal mechanisms (Zhao & Chadwick, 2014; Zhou & Wu, 2010). Together, these dimensions of proximity capture both the processes and outcomes of a firm’s innovation efforts, but their strength varies by alignment of the mechanism (Vandaie & Zaheer, 2014). Therefore:
Hypothesis 3 The positive association between innovation capabilities and firm performance varies depending on the proximity of innovation capabilities to firm performance.
Innovation capabilities affect different types of firm performance. Firm performance is the extent to which a company achieves its strategic and operational objectives, including both ex-post (i.e., realized) and ex-ante (i.e., expected) outcomes (Souder et al., 2023). Ex-post financial performance metrics capture a firm’s profitability, growth, and value creation (Pecko et al., 2025; Richard et al., 2009), while ex-ante measures reflect investors’ expectations of future competitiveness and firm value (Ellis, 2006). Together, these metrics provide a comprehensive assessment of a firm’s economic health and competitive success, serving as a key outcome for evaluating the effectiveness of capabilities such as innovation (Harris, 2001).
Distal capabilities contribute to ex-post (i.e., realized) firm performance, driving revenue growth and profitability (Lestan et al., 2025; Mihalache et al., 2012). On the other hand, proximal capabilities, such as patents and product launches, enhance ex-ante (i.e., expected) firm performance, increasing market share and customer engagement (Vandaie & Zaheer, 2014).
Souder et. al. (2023) categorized the determinants of firm performance as precisely defined, flexible, scaled or unscaled. This suggests that the effect of innovation capabilities on firm performance may depend on how performance is measured. Metrics that capture precisely defined outcomes (i.e., based on observation), such as profitability are likely to reflect the impact of innovation capabilities more directly than flexible (i.e., based on expectations) measures, such as market capitalization. Therefore:
Hypothesis 4 The positive association between innovation capabilities and firm performance varies depending on whether firm performance is measured using ex-post or ex-ante indicators.
Previous evidence on the innovation capabilities–firm performance is heterogeneous and studies report mixed and inconsistent results, including positive (e.g., Knight & Kim, 2009), non-significant (e.g., Fey, 2005) and negative associations (e.g., Gruber et al., 2008). Therefore, we articulate several study-level moderators to the main effect presented in Hypothesis 1 and collectively we consider them Hypothesis 5.
Studies reporting positive relationship between innovation capabilities and firm performance include Andrade-Rojas et al. (2024); Branzei and Vertinsky (2006); Lee et al. (2001). For example, from 2 independent samples, with a total sample size of 5691 and 3627 firms in France and Great Britain respectively, Ballot et al. (2015) reported a significant and positive relationship. Arunachalam et al. (2018) reported positive and moderate association based on the 190 manufacturing SMEs in India using survey data showing that both number of new products and R&D intensity (i.e., stochastic innovation capabilities) are positively associated with profitability of these firms. Similarly, Deligianni et al. (2017) reports that product diversification and R&D intensity (i.e. distal innovation capabilities), both are positively associated with the revenue growth rate based on longitudinal study of 129 firms in Greece.
Insignificant associations between innovation capabilities and firm performance were reported by studies other than those mentioned above. Díaz-Díaz et al. (2008), found that proximal innovation capabilities and firm performance are non-significant. Gómez et al. (2020), found that some distal dimensions of innovation capabilities have insignificant relationship with firm performance, based on sample of 8554 firms in Spain. Zhang et al. (2019) used China Innovation Survey (CIS) data of 1406 firms between 2005 and 2007 and reported that stochastic innovation capabilities have no impact on profitability of firms and follow nonmonotonic inverted U-shaped pattern.
Negative relationships between innovation capabilities and firm performance are also found in several studies (Hock-Doepgen et al., 2024; Kathuria et al., 2023; Penner-Hahn & Shaver, 2005). Based on 10-K annual reports from 9100 firms between 1994 and 2017, Schäper et al. (2023) found innovation capabilities have significant, strong and negative effect on firm performance. Next, Choi et al. (2011), using cross-sectional design and objective data from the Shanghai Stock Exchange Statistics and Fact Book of Shenzhen Stock Exchange, reports that both R&D intensity and number of patents are negatively associated with ex-post performance measure such as the revenue growth rate based on 548 listed firms in China. Similarly, using the same cross-sectional study design and objective data from S&P’s Compustat database, Saldanha et al. (2017) reported that R&D intensity is negatively associated with revenues of 310 listed firms in the USA. The same negative association pattern, using the stochastic mechanism, was observed in Finland based on the sample 93 small and large firms for the R&D intensity and ex-post growth (Laamanen, 2005).
The inconsistent relationships should be attributed to three potential reasons. First, the diversity of samples also influences the magnitude of the effect size. Shane (1995) noted that uncertainty-accepting societies are particularly highly associated with innovation than uncertainty-avoiding societies because of the greater tolerance for risk. That said, in cultures with low uncertainty avoidance values (i.e., accepting uncertainty), firms could be more willing to take risks and adopt novel strategies, which fosters a more innovative environment (Craighead et al., 2009; Karhade & Dong, 2021; Knight & Cavusgil, 2004). These firms could be more flexible in their approach to problem-solving and less averse to experimentation, enabling them to capitalize on innovation capabilities more effectively. Oppositely, in cultures with high uncertainty avoidance values (i.e., avoiding uncertainty), firms may be more risk-averse and focused on maintaining stability, which could hinder the full utilization of innovation capabilities (Arrighetti et al., 2014; Boehe & Cruz, 2010; Frank et al., 2016).
Second, the inconsistent results may be attributed to theories and frameworks used in the previous studies. Some studies used dynamic capabilities as theoretical framework (Messersmith & Guthrie, 2010; Oh et al., 2012; Rodríguez-Duarte et al., 2007), whereas, other theoretical frames of reference used were appropriability (Kim et al., 2012), organizational design (Koufteros et al., 2002), inducement-opportunities framework (Li & Atuahene-Gima, 2002) or social capital (Maurer et al., 2011).
Third, the inconsistent results may be attributed to research design used in the previous studies. Some studies employed cross-sectional design (Gerasymenko et al., 2015; Hashai, 2011; McKelvie et al., 2018; Vorhies et al., 2011; Zhou et al., 2006, 2014; Zollo & Reuer, 2010), while other focused on the “time” dimensions through a longitudinal setup (Patel et al., 2015; Schubert & Tavassoli, 2020; Tatikonda et al., 2013). Given all these differences in previous studies, we expect that there will be moderation at the study-level. Therefore:
Hypothesis 5 At the study-level, the positive association between innovation capabilities and firm performance varies depending on the (a) uncertainty avoidance of the sample studied, (b) theoretical frameworks employed, and (c) study design.
We systematically conducted a meta-analysis to address our research questions. This method offers the advantage of integrating results from multiple independent studies, many of which individually lack sufficient sample sizes or precise effect estimates (Cooper et al., 2019; Geyskens et al., 2009). By combining zero-order correlations coefficients and accounting for heterogeneity across studies, meta-analysis provides a more reliable estimate of the true relationship (Field & Gillett, 2010). All materials supporting findings of this study are available on the OSF repository (https://osf.io/z4bqx/).
Systematic ReviewSearch StrategyWe followed the latest ‘Preferred Reporting Items for Systematic Reviews and Meta-Analyses’ (PRISMA) guidelines and developed a four-step approach (Page et al., 2021). Adopting the latest search strategy proposed by Bergh et al. (2024), we focused exclusively on empirical studies published in the double-blind reviewed journals (in line with Geyskens et al. 2006 and Vanneste et al. 2014), which are known for their rigorous review process and high-quality publications. This approach ensured a reliable foundation for a robust and credible meta-analysis.
First, we performed topic search (i.e., Title, Keywords and Abstract) in the Web of Science and Scopus registers on September the 11th, 2024. The search string used were: (“innovation” AND “capabilit*” [to capture studies on capability or capabilities] AND “performance,”). This resulted in 1359 identified records from both registers. Then we removed 151 duplicates (identical or redundant2 studies) leaving us with 1208 unique records. Next, we removed 16 records (i.e., 8 systematic reviews, 3 meta-analysis and 5 retracted3 studies) not relevant for our study. This left us with 1192 records for full-text screening. We then screened full text of 1192 records. We initially could not retrieve two studies (Adbi et al., 2022; McGee & Dowling, 1994), but our scholarly network obtained and shared them with us upon request. We excluded 992 records due to several theoretical and methodological reasons, such as wrong study designs (i.e., not empirical), wrong publication types (i.e., professional business magazines), wrong populations (i.e., country- or industry-level studies), and wrong study outcomes (i.e., not firm-level performance). In the first pass with these criteria, we identified 200 studies relevant for our meta-analysis. Next, we excluded 148 studies to maintain quality (see 3.1.2. Inclusion Criteria) and non-independence (see 3.1.3. Non-independence Selection Criteria) because “even one red sock (i.e., bad study) amongst the white clothes (i.e., good studies) can ruin the laundry (Field & Gillett, 2010, p.667)”. This left us with a final sample of 52 primary studies consisting of 54 independent samples, covering studies published in the period between 2001 and 2024, yielding 320,529 unique firms already investigated from 19 countries.
Second, we also searched for unpublished studies to avoid common method bias, using scholarly network of >10,000 scholars. We posted multiple calls for unpublished studies at various research societies and academies, investigating innovation capabilities and firm performance.
Third, since this was the first attempt, we argue here, to meta-analyse the innovation capabilities–firm performance relationship (using number of published meta-analyses as an indicator), we examined the references listed in the previous systematic reviews on innovation capabilities (Iddris, 2016; Mendoza-Silva, 2021; Moreira et al., 2024) and firm performance (Fainshmidt et al., 2016; Karna et al., 2016; Schilke et al., 2018) to find additional studies.
Finally, we used the ‘snowballing’ techniques, including backward (i.e., references) and forward (i.e., citations) tracking of each relevant article identified previously to find additional studies. Fig. 1 portrays the inclusion and study-selection procedure based on the PRISMA framework used in this study.
Inclusion Criteria
We included studies that met three criteria. First, studies had to examine both subjective and objective measures of innovation capabilities with firm performance metrics at the firm-level consistent with prior construct definitions (Ellis, 2006; Richard et al., 2009; Souder et al., 2023). Thus, we included studies with samples of firms (i.e., businesses and corporate entities) and excluded studies with samples of government agencies (Arora & Nandkumar 2012), universities (Slavova et al., 2016) or non-profit organizations (Mendoza-Abarca & Gras, 2019), for example. Additionally, we excluded studies with focal outcomes of the study beyond firm-level such as the alliance-level (Jiang et al., 2010), region-level (Mewes & Broekel, 2022) or industry-level (Mol, 2005). To reduce potential reverse causality, studies had to predict and test Schumpeter’s hypotheses of innovation capabilities with firm performance at least simultaneously (i.e., one-way); studies that explored the effect of firm performance either a priori or a posteriori on innovation capabilities (e.g., Subramaniam & Youndt, 2005) were excluded from our sample.
Second, we included studies that reported the sample size (i.e., number of firms rather than firm-year observations in longitudinal setup) with Pearson rather than Spearman correlations (see Demirkan & Demirkan, 2012, for a practical application of both) because they align better with our meta-analysis, assuming linear one-way relationship between innovation capabilities and firm performance metrics. Sample sizes and correlation coefficients were needed because these values are required for calculating their effect size. Some studies were under-reported and lacked information about their sample size (i.e., number of unique firms) or did not report correlation matrices needed to calculate effect size. We contacted4 authors of these studies and collected missing or under-reported data.
Third, we included studies with between firm-level research design (i.e., inter-firm differences between corporate parents or business units) rather than divisions or departments (i.e., intra-firm differences) within a single firm, as we do not report differences in findings within a single firm and its divisions or departments.
Non-independence Selection CriteriaTo minimize confirmation bias, studies were coded by two experts independently. Coders agreed on 98 percent of initial codes, and discrepancies were resolved via discussion. We used three criteria to ensure that our meta-analysis offer an acceptable level of independence among correlation coefficients in our data set because many studies were based on overlapping datasets. First, for studies with multiple independent samples (Ballot et al., 2015; Zhou et al., 2014); correlation coefficients from each sample were included in our data set. Second, for studies with more than one correlation coefficient for a single relationship (because scholars either tested several performance metrics or involved multiple operationalizations of the same innovation capability constructs, for instance), we included all these correlations in our dataset and aggregated average effect size. Third, when studies relied on the same or overlapping data sources (e.g., Compustat, the Spanish Survey of Business Strategy), we included only one correlation coefficient in our meta-analysis. Specifically, we selected the correlation from the study with the largest sample size. If sample sizes were identical, we prioritized the correlation from the oldest study. However, if overlapping data sources were complemented by unique, self-conducted survey data that linked primary and secondary sources, we included both correlations in our analysis.
These procedures reduced 200 identified records assessed for our meta-analysis to final sample of 52 studies consisting of 54 independent samples. To make a qualitative assessment of the 54 independent samples to show the relevance of conducting a meta-analysis, and to derive suitable moderator variables for the meta-analysis, we present details of the included studies in supplementary Table A1. Collectively, these samples consist of 320,529 independent firms already investigated from 19 countries.
Measurements of Innovation Capabilities and Firm PerformanceAfter reviewing the included studies, we grouped innovation capabilities into three categories: (i) distal, (ii) proximal and (iii) stochastic (i.e., distal and proximal linked in one-way). We focused on how constructs were measured, not labelled. For example, (Deligianni et al., 2017; Zhou & Wu, 2010) use ‘technological capabilities,’ while, (Chittoor et al., 2009) use ’innovation capability’ as their primary construct but these two different constructs were measured using the same indicator: R&D intensity. Therefore, we focus on the indicators of measurement rather than labels. Distal innovation capabilities indicate a measurement that include indicators of innovation inputs, while proximal innovation capabilities indicate a measurement that include dimensions of innovation outputs. A summary of mentioned variables is portrayed in Table 1.
Constructs operationalization of innovation capabilities and firm performance.
| Construct | Measurement | Specific indicator |
|---|---|---|
| Innovation Capabilities | Distal (i.e., inputs or antecedent activities) | - R&D Expenditures (Li & Atuahene-Gima, 2002) |
| - R&D Intensities (Deligianni et al., 2017) | ||
| - R&D Manpower (Arrighetti et al., 2014) | ||
| - Training Expenditures (Simonin & Özsomer, 2009) | ||
| - # of conceptualization projects (Craighead et al., 2009) | ||
| - Investment in machinery and external knowledge (Menor et al., 2007) | ||
| - Expenditure on new products/service development (Kim et al., 2012) | ||
| Proximal (i.e., outputs or measurable outcomes) | - Patent applications (Rodríguez-Duarte et al., 2007) | |
| - Patent citations (Wu et al., 2020) | ||
| - # of Patents (Fey, 2005) | ||
| - Licenses Granted (Frank et al., 2016) | ||
| - Labor Productivity (Patel et al., 2015) | ||
| - Share of new products/services (Zhou & Wu, 2010) | ||
| - # of new products/service (Arunachalam et al., 2018) | ||
| - % of sales of new products/services (McKelvie et al., 2018) | ||
| Firm Performance | Ex-post (i.e., realized based on observed performance) | - Revenues or Sales (Penner-Hahn & Shaver, 2005) |
| - Profit (Tatikonda et al., 2013) | ||
| - Revenue Growth Rate (Hock-Doepgen et al., 2024) | ||
| - Output Growth Rate (Mihalache et al., 2012) | ||
| - Economic Value Added (Branzei & Vertinsky, 2006) | ||
| - Return on Assets (Zhao & Chadwick, 2014) | ||
| - Return on Equity (Deligianni et al., 2017) | ||
| - Return on Sales (Schäper et al., 2023) | ||
| - Total Shareholder Return (Menor et al., 2007) | ||
| Ex-ante (i.e., based on expectations of future performance) | - Dynamic Value (Gerasymenko et al., 2015) | |
| - Tobin’s Q (Schäper et al., 2023) | ||
| - Market-to-Book Value (Zollo & Reuer, 2010) |
We categorized firm performance into three groups: (i) ex-post, (ii) ex-ante (iii) combination of both. We focus on economic financial performance. When scholars focused on non-financial performance (Hulland et al., 2007), we omit these in the interests of being precise. Ex-ante performance indicates a metric that include dimensions of market-based measurement (i.e., Tobin’s Q, Market-to-Book Value, Dynamic value). Ex-post performance indicates a metric that include dimensions of accounting-based measurement (i.e., based on realized observations). Nonetheless, we focus on studies that reflect measures of innovation capabilities and firm performance as simple, univariate single-order measurements or several, multivariate lower-order partials of the objective and subjective measures. We report difference in these measurements.
Meta-Analytical EstimationWe conducted a meta-analysis of zero-order correlations using a random-effects model with Fisher’s (z) transformation of these correlations (Hunter & Schmidt, 2000). First, we calculated the effect sizes by transforming observed zero-order correlation coefficients (ri), that are biased in small samples, negatively skewed, and have a variance formula with inconvenient properties to Fisher’s (z) as expressed in equation one:
Using Fisher’s (z) we obtain an effect size statistic that is less biased and affected by skewness (Borenstein et al., 2009). We then obtain the variance of Fisher’s (z) as expressed in equation two:
We then have the variance of Fisher’s (z), where (n) is the sample size used to compute the correlation. Results were back transformed to the correlation metric for interpretation as expressed in equation three:
For our univariate meta-analysis, we combine the transformed Fisher’s (z) correlations into a mean effect by taking their weighted average, where the weight of a study sample is inversely proportional to its variance as proposed by Hunter and Schmidt (2004). The estimated variance of the weighted mean is the reciprocal of the sum of the weights. We calculated the weighted mean effect using study sample sizes as weights as expressed in equation four:
Next, we quantified the total variability of effect sizes across studies by calculating the observed variance, which reflects both sampling error and true differences between study populations. This observed variance was computed as the weighted sum of squared deviations of each study’s effect from the weighted mean, divided by the total sum of weights as expressed in equation five:
Next, we estimated the portion of variance attributable to sampling error within studies. The mean sampling-error variance quantifies the variation expected purely from within-study error and was calculated as the weighted average of the individual study variances as expressed in equation six:
where vi is the variance of the Fisher’s z-transformed correlation for study (i) and (ni) is its sample size. Subtracting this mean sampling-error variance from the observed variance (Sobs2) isolates the component of variability attributable to true heterogeneity among studies. Next, the between-study variance represents the variability in true effect sizes across studies, beyond what is expected from sampling error. It was calculated as expressed in equation seven:
If the observed variance (Sobs2) is smaller than the mean sampling-error variance (Vsamp‾), (τ2) is set to zero. The square root, (τ), represents the standard deviation of the true correlations across studies. To quantify variability across studies, we computed Cochran’s Q, which tests whether the observed dispersion of effect sizes exceeds what would be expected from sampling error alone as expressed in equation eight:
where (wi) is the weight of study (i) (sample size for Hunter–Schmidt weighting), (zi) is the Fisher’s z-transformed effect size, and (zˉ) is the weighted mean effect. The degrees of freedom are (df=k−1), and (Q) follows a chi-square distribution. The proportion of total variance attributable to true heterogeneity was calculated using the (I²) statistic as expressed in equation nine:Where values near 0 % indicate low heterogeneity, while values closer to 100 % indicate that most of the variability reflects true differences among studies rather than sampling error. We used the function ‘metacor’ from the ‘meta’ package in RStudio for the calculation of the overall effect size (Harrer et al., 2021). It is a standardized measure of the magnitude of observed effects within collected studies in the developed dataset (Field & Gillett, 2010).
Prior studies used different measures, designs and samples; therefore, we used the same ‘metacor’ function while providing additional arguments in the subgroup analyses for the moderator analyses. They allowed us to collectively test hypotheses H2, H3, H4, and H5A to H5C, describing why studies focusing on proximal innovation metrics (e.g., patents) produces lower or higher effects than studies focusing on distal outcomes (e.g., R&D spendings), for example. The data structure included four main subgroups of moderators: (i) decomposed form of Schumpeter’s hypothesis (three levels as Mark I, Mark II or combination of Mark I & II); (ii) capability proximity (three levels as distal; proximal; stochastic); (iii) performance metrics (three levels as ex-post; ex-ante; combination of both); and (iv) study-level moderators, such as uncertainty avoidance of the samples (two levels as high and low5); dynamic capabilities framework in the theoretical background of the study (two levels as yes or no); and study design (two levels as longitudinal and cross-sectional).
Common concern in meta-analyses is that effect sizes in the studies with smaller samples, cross-sectional designs or with subjective measures impact the magnitude of the overall effect size (IntHout et al., 2015; Rothstein et al., 2005). To address this, we conduct supplementary post-hoc tests to gain a deeper understanding of how various contextual considerations (e.g., study-level quality and measures) might influence the observed magnitude of the relationship between the innovation capabilities and firm performance.
ResultsThe Overall Schumpeter's HypothesisIn Hypothesis 1, we predicted that innovation capabilities are positively associated with firm performance (the Schumpeter hypothesis taken as whole). In full support of Hypothesis 1, our meta-analysis results for each study’s effect size, and the aggregate effect size for all studies, are presented graphically as a forest plot in Fig. 2. A total of 54 samples were included, providing 320,529 observations. The results show that while most effect sizes are positive, some studies report small or negligible effects. The aggregate effect size of the overall relationship is positive (r̂ = 0.11; p < 0.001, 95 % CI = [0.09, 0.13]), suggesting that, on average, innovation capabilities are positively associated with firm performance.
However, substantial heterogeneity was observed across studies (Q = 762.34, df = 53, p < 0.001; I² = 93.0 %, 95 % CI = [91.7 %, 94.2 %]), indicating that the strength of this relationship varies depending on study-specific factors such as the type of innovation capabilities assessed, and the measures of firm performance used. The random-effects model accounted for between-study variance (τ² = 0.0022, 95 % CI = [0.0091, 0.0237]; τ = 0.047, 95 % CI = [0.0955, 0.1540]), and all correlations were transformed using Fisher’s z prior to analysis.
The ‘Narrow’ Version of the Schumpeter’s HypothesisIn Hypothesis 2, we predict that innovation capabilities are positively associated with firm performance, with the strength of this relationship differing between Mark I and Mark II hypotheses (i.e., the decomposed form of the Schumpeter hypothesis). Our results provide partial support for Hypothesis 2. We find that the positive effect of innovation capabilities on firm performance was strongest for Mark I hypothesis (i.e., creative destruction), where the aggregated effect size was positive and moderate (r̂ = 0.15; 95 % CI = [0.08, 0.23]; k = 11), with substantial heterogeneity (Q = 86.95, I² = 88.5 %). Studies examining both Mark I & II hypotheses also showed a positive, moderate relationship (r̂ = 0.13; 95 % CI = [0.11, 0.15]; k = 30), though heterogeneity remained high (Q = 519.28, I² = 94.4 %). In contrast, studies focusing exclusively on Mark II hypotheses (i.e., creative accumulation) hypothesis reported essentially no effect (r̂ = −0.0002; 95 % CI = [−0.03, 0.03]; k = 13) with lower heterogeneity (Q = 21.17, I² = 43.3 %). The results for the overall and ‘narrow’ forms of the Schumpeter hypothesis are illustrated in Fig. 3.
A test for subgroup differences indicated that the relationship between innovation capabilities and firm performance varies significantly across the subgroups of Schumpeterian hypotheses (Q = 60.68, df = 2, p < 0.001), suggesting that the type of hypothesis considered is an important moderator of the overall effect. All analyses were conducted using a random-effects model with inverse-variance weighting, Hunter-Schmidt estimation of τ², and Fisher’s z-transformed correlations.
Potential publication or reporting bias in the relationship between innovation capabilities and firm performance was examined using funnel plots. As shown in Fig. 3, the distribution of effect sizes appears reasonably symmetrical across different forms of Schumpeterian hypothesis, providing an initial visual indication of low bias in the included studies.
Proximity of Innovation Capabilities and Firm PerformanceIn Hypothesis 3, we predicted that the positive association between innovation capabilities and firm performance varies depending on the proximity of innovation capabilities (i.e., stochastic, distal, and proximal) to firm performance. In full support of Hypothesis 3, our results and the test for subgroup differences indicated that the strength of the relationship between innovation capabilities and firm performance varies significantly by capability proximity (Q = 7.84, df = 2, p = 0.020), suggesting that proximal innovation capabilities are generally associated with the strongest performance effects (r̂ = 0.21; 95 % CI = [0.04, 0.36]; k = 5), though heterogeneity remained high (Q = 79.57, I² = 95.0 %). The results for each subgroup are presented in Table 2.
Meta-Analyses of the innovation capabilities and firm performance relationship.
Notes. K = number of effect sizes; N = total number of observations; r^ = estimate of correlation; LLCI = lower bound of confidence interval for r^ at 95%; ULCI = upper bound of confidence interval for r^ at 95%; SE = indicates the precision (standard error) of the estimated average effect size; smaller values reflect more precise estimates;τ2 = variance of the true effect size (tau-squared); (Q) = Within-subgroup effect based on Hunter and Schmidt’s (2004); I2 = Percentage (%) of variation across studies that s due to heterogeneity rather than chance; Q = Between-group effect based on Hunter and Schmidt’s (2004) chi-square test for heterogeneity.
For stochastic proximity of innovation capabilities, the aggregate effect size was small but positive (r̂ = 0.07; 95 % CI = [0.05, 0.08]; k = 28), with substantial heterogeneity across studies (Q = 187.86, I² = 85.6 %). Distal proximity of innovation capabilities showed a stronger positive relationship (r̂ = 0.14; 95 % CI = [0.08, 0.20]; k = 21) and higher heterogeneity (Q = 400.88, I² = 95.0 %).
Innovation Capabilities and Measurements of Firm PerformanceIn Hypothesis 4, we predicted that the positive association between innovation capabilities and firm performance varies depending on whether firm performance is measured using ex-post or ex-ante indicators. Our results are illustrated in Table 2. and provide full support for Hypothesis 4. Studies using ex-post measures showed a positive relationship (r̂ = 0.12; 95 % CI = [0.10, 0.14]; k = 51) with substantial heterogeneity (Q = 671.81, I² = 92.6 %). Albeit, one study used an ex-ante measure, reporting a small and non-significant effect (r̂ = 0.05; 95 % CI = [−0.11, 0.20]; k = 1). Two studies combining both measures showed a slightly negative effect (r̂ = −0.03; 95 % CI = [−0.05, −0.01]; k = 2) with low heterogeneity (Q = 1.06, I² = 5.8 %). A test for subgroup differences indicated that the type of performance measure significantly moderates the relationship between innovation capabilities and firm performance (Q = 110.03, df = 2, p < 0.001).
Study-level Moderator AnalysesIn Hypothesis 5, we predicted, at the study-level, that the positive association between innovation capabilities and firm performance varies depending on the (a) uncertainty avoidance of the sample, (b) theoretical frameworks employed, and (c) study design. In partial support of the Hypothesis 5, our results are presented in Table 2. and indicate that only study design shows significant moderation, whereas uncertainty avoidance and theoretical framework not significant.
First, by comparing studies from low- and high-uncertainty avoidance cultures, we used uncertainty avoidance index (UAI). The mean effect size was positive and significant for both high-UAI (r̂ = 0.116, 95 % CI [.071, 0.160]) and low-UAI samples (r̂ = 0.108, 95 % CI [.087, 0.128]). However, the difference between groups was not statistically significant (Q = 0.10, p = .75), suggesting that the relationship between innovation capabilities and firm performance does not vary significantly across samples differing in their uncertainty avoidance.
Second, we examined whether the theoretical framework adopted in primary studies moderated the innovation capabilities–firm performance relationship. Studies using a dynamic capabilities framework (r̂ = 0.113, 95 % CI [.090, 0.137]) and those employing alternative frameworks (r̂ = 0.113, 95 % CI [.076, 0.149]) showed nearly identical effects. The difference between groups was not statistically significant (Q = 0.00, p = .99), indicating that the relationship between innovation capabilities and firm performance is consistent regardless of theoretical background.
Finally, we tested the moderating role of study design. Cross-sectional studies reported a significantly stronger relationship (r̂ = 0.152, 95 % CI [.095, 0.208]) than longitudinal studies (r̂ = 0.083, 95 % CI [.061, 0.104]), with the between-group difference statistically significant (Q = 4.92, p = .03). This suggests that the innovation capabilities–firm performance link tends to appear stronger in cross-sectional settings, potentially reflecting short-term associations, while longitudinal study designs yield more conservative estimates over time.
Post-hoc Tests and Robustness ChecksA post hoc analysis compared studies using unidimensional versus multidimensional measures of innovation capabilities as well as study quality. In the first stance, the results revealed a significant difference between the two groups of dimensionalities (Q = 9.16, p = .003). Studies using unidimensional indicators (r̂ = 0.154, 95 % CI [.098, 0.210]) reported substantially stronger relationships with firm performance than those employing multidimensional measures (r̂ = 0.063, 95 % CI [.047, 0.079]). This suggests that broader, multidimensional operationalizations of innovation capabilities may dilute the direct performance effect observed in narrowly defined, single-dimension indicators.
Next, we analysed whether study quality, based on AJG 2024 journal ranking, moderates the innovation capabilities–firm performance relationship. Studies in level four-star journals (r̂ = 0.100, 95 % CI [.081, 0.119]) reported lower effect sizes compared to lower-ranked journals (level 4: r̂ = 0.182; level 3: r̂ = 0.281), though the between-group difference was marginally non-significant (Q = 5.87, p = .053). This pattern suggests that higher-quality publications may yield more conservative estimates of the innovation capabilities–firm performance link, potentially reflecting more stringent methodological standards.
These post-hoc tests highlight the complexity of the association between innovation capabilities and firm performance and do not fully capture the heterogeneity. Several factors, beyond the study quality and measures, significantly influence both the strength and direction of this association. The findings highlight the importance of considering these additional moderators when interpreting meta-analytic results and suggest that studies incorporating unidimensional measurements of innovation capabilities and reported in lower ranked journals may report stronger associations between innovation and firm performance.
DiscussionThis study set out to clarify how, and under what conditions, innovation capabilities are associated with firm performance. Building on Schumpeterian view, capability proximity logic, and the distinction between ex-ante and ex-post performance metrics, our meta-analytic results provide a robust and comprehensive synthesis of a fragmented body of evidence. Below, we discuss the theoretical meaning of our findings considering prior research and evaluate each of the five predicted hypotheses.
Schumpeter’s Hypothesis and the Overall EffectConsistent with the foundational Schumpeterian view that innovation serves as a key engine of firm success (Cevik, 2025; Guichardaz & Pénin, 2025; Schumpeter, 1934), our results show that innovation capabilities are positively associated with firm performance on average, supporting Hypothesis 1. This aligns with past research demonstrating that innovative activity enables firms to create differentiation, respond to environmental shifts, and generate economic value (Guerrero & Siegel, 2024; López-Cabarcos et al., 2019).
Similarly, our results also reflect the mixed and sometimes contradictory patterns reported in prior studies, including positive (e.g., Andrade-Rojas et al. 2024; Arunachalam et al. 2018), insignificant (e.g., Gómez et al. 2020; Zhang et al. 2019), and even negative associations (e.g., Hock-Doepgen et al. 2024; Schäper et al. 2023). The substantial heterogeneity found in our analyses echoes these divergent findings, reinforcing longstanding arguments that the innovation capabilities–firm performance relationship is not universal but contingent on conceptual, methodological, and contextual factors (Piñeiro-Chousa et al., 2020; Yin et al., 2023). Our moderator analyses therefore provide an important extension of this literature by identifying where these divergences originate.
Moderating EffectsDecomposed Schumpeter’s Hypothesis: Mark I and mark IIOur results provide partial support for Hypothesis 2 by showing that the strength of the innovation capabilities–firm performance association differs across Schumpeter Mark I and Mark II contexts. Specifically, we find a moderate positive association for studies aligned with the Mark I logic of creative destruction, where smaller, entrepreneurial firms rely on agility and experimentation to generate disruptive innovations (Guichardaz & Pénin, 2025). This pattern echoes evidence from Branzei and Vertinsky (2006), and Knight and Kim (2009), which document strong performance effects for innovation indicators in smaller or more flexible firms.
Conversely, the absence of a positive association between innovation capabilities and firm performance in Mark II settings suggests that large, established firms do not consistently convert innovation capabilities into performance gains, despite their greater resource endowments (Hall et al., 2010). This implies that, we argue here, again, entrepreneurial, not big firms alone benefit from innovation in their performance. This aligns with research showing that large firms may experience bureaucratic inertia or diminishing marginal returns on innovation spending (Sevil et al., 2022), and supports the claim that creative accumulation does not guarantee performance improvements (Choi et al., 2011; Kathuria et al., 2023).
Together, our findings confirm that firm size and innovation regime meaningfully impact the innovation capabilities–firm performance link, offering the first meta-analytic substantiation of the dual Schumpeterian logic at the firm-level.
Proximity of Innovation CapabilitiesWe find strong support for Hypothesis 3: the closer innovation capabilities are to actual innovation outputs, the stronger their association with firm performance. This insight directly reflects the ontological argument that capabilities differ in how immediately they translate into value creation (Moreira et al., 2024). Our results show that proximal capabilities (e.g., patents, new products or services) impose the strongest effects. This supports prior work arguing that these indicators are tightly linked to value appropriation and commercialization outcomes (Ceccagnoli, 2009; Rubera et al., 2016). Further, distal capabilities (e.g., knowledge acquisition, or R&D investments) show moderate effects, consistent with research suggesting that they function as foundations for future innovation but do not guarantee immediate performance benefits (Menor et al., 2007; Simonin & Özsomer, 2009).
Finally, stochastic capabilities, reflecting a mixture of distal and proximal indicators, yield the weakest effects, aligning with evidence that their performance impact is uncertain and dependent on conversion mechanisms (Stam & Elfring, 2008; Zhou & Wu, 2010). Our results confirm that innovation capabilities are not homogeneous, that proximity matters, and that ontological distinctions offer a meaningful explanation for inconsistencies reported across primary studies (Mendoza-Silva, 2021; Moreira et al., 2024).
Innovation Capabilities and Ex-post vs. Ex-ante Firm Performance MeasuresOur findings fully support Hypothesis 4: the type of performance metric used influences the observed effect of innovation capabilities. Studies using ex-post, realized performance indicators (e.g., profitability or revenue growth) show stronger and more consistent effects. This aligns with research suggesting that realized outcomes more directly capture the economic consequences of innovative activity (Richard et al., 2009; Souder et al., 2023). By contrast, ex-ante, expectation-based measures (e.g., market capitalization or total shareholder returns) which are both highly relevant for investors, produce much weaker associations. This reinforces the argument that investor expectations and market capitalization incorporate a wider set of intangible or forward-looking signals (Ellis, 2006; Harris, 2001), which may dilute the observable contribution of innovation. Therefore, our results empirically demonstrate that measurement decisions–long acknowledged as a source of variance in the literature–influence the conclusions drawn about whether, at all, innovation capabilities pay off.
Study-Level Moderators: Cultural, Theoretical, and Design DifferencesHypothesis 5 is partially supported. Dynamic capabilities framework and cultural uncertainty avoidance do not significantly differentiate effect sizes, despite long-standing arguments that cultural preferences for risk-taking or stability shape innovative behaviour (Knight & Cavusgil, 2004; Shane, 1995). This suggests that while culture influences innovation inputs, the performance outcomes of innovation capabilities may be more universal than previously assumed.
By contrast, study design significantly moderates effects. Cross-sectional studies show stronger relationships than longitudinal studies, consistent with concerns raised by IntHout et al. (2015), Rothstein et al. (2005) and others that short-term or contemporaneous designs may overestimate, in our case the firm performance benefits of innovation capabilities. Longitudinal studies, which capture time lags and reverse causality concerns (Patel et al., 2015), produce more conservative estimates, indicating that innovation payoffs may materialize gradually. Our finding advances the methodological debate by empirically demonstrating how research design contributes to mixed evidence in the field.
Post-Hoc Tests: Dimensionality and Study QualityOur post-hoc analyses further explain the heterogeneity in prior research. Studies using unidimensional measures of innovation capabilities (i.e., single indicator) report much stronger performance effects than those employing multidimensional constructs (i.e., multiple indicators), suggesting that narrow measures (e.g., R&D intensity alone) may overstate innovation’s contribution, whereas broader measures distribute effects across several underlying dimensions (Piñeiro-Chousa et al., 2025; Salvato & Vassolo, 2018).
Additionally, studies published in higher-ranked journals tend to report smaller effect sizes. This pattern likely reflects stronger methodological rigor and suggests that earlier or lower-ranked studies may have inflated the innovation capabilities–firm performance relationship. This insight is consistent with concerns about common method bias and selection bias in the innovation capabilities literature (Mendoza-Silva, 2021; Moreira et al., 2024).
Theoretical Contributions and Practical ImplicationsFrom the theoretical standpoint, this meta-analysis makes several contributions to the scholarly understanding of the Schumpeterian hypotheses and the broader innovation capabilities–firm performance relationship. First, by synthesizing evidence across more than two decades of empirical work (2001–2024), we provide the most comprehensive test-to-date of the overall Schumpeterian proposition that innovation capabilities are associated with firm performance. Our findings confirm a positive association, thereby offering cumulative support for a foundational assumption in innovation theory. Importantly, the study also clarifies persistent inconsistencies in prior research by demonstrating that the magnitude of the innovation capabilities–firm performance link varies substantially depending on the form of the Schumpeterian hypothesis examined. Whereas previous studies often treated Schumpeter Mark I and Mark II arguments interchangeably, our results show that creative destruction mechanisms (Mark I) tend to yield stronger performance outcomes than creative accumulation mechanisms (Mark II) alone. This distinction advances Schumpeterian theory by empirically validating heterogeneity between its two core traditions.
Second, by integrating capability proximity, performance measurement, and study-level characteristics, our analysis extends theoretical work on how and under what conditions innovation capabilities relate to performance benefits. We show that proximal capabilities–those directly involved in innovation outputs–produce the strongest performance effects, supporting theories that emphasize the value of capabilities closest to commercialization. In contrast, more remote or stochastic capabilities exhibit weaker relationships, clarifying conceptual debates about which types of capabilities matter most for firm success. Similarly, our analysis highlights the importance of measurement choices by demonstrating that ex-post performance indicators more reliably capture innovation impacts than ex-ante indicators.
Finally, the moderating influence of study design and measurement dimensionality reveals that theoretical conclusions in this domain are shaped not only by substantive mechanisms but also by methodological decisions. By identifying these boundary conditions, our study contributes to more precise theorizing about innovation capabilities and offers a structured agenda for future research seeking to explain variance in observed outcomes.
From the practical standpoint, the findings of this study offer several actionable insights for practitioners. First, firms should recognize that not all innovation capabilities are equally impactful: proximal capabilities, which directly generate innovation outputs such as patents or new products, for example, are more strongly linked to performance, whereas distal or stochastic capabilities play a supportive but indirect role. This suggests that managers should prioritize investments and processes that convert innovation inputs into tangible outputs. Second, the results illustrate that the effectiveness of innovation capabilities depends on firm context, including size and strategic orientation. Smaller, entrepreneurial firms may benefit most from agile, disruptive innovation (Mark I), while larger incumbents may need to carefully structure R&D and innovation processes to avoid inefficiencies associated with creative accumulation (Mark II). Finally, performance measurement matters (a lot!): relying solely on expectations-based indicators may understate the benefits of innovation, so managers are encouraged to track both ex-post and ex-ante outcomes to obtain a comprehensive view of innovation effectiveness. These insights can guide firms in designing targeted innovation strategies and inform policymakers seeking to foster competitive, innovation-driven economies.
Overall, this meta-analysis clarifies several long-debated issues surrounding the innovation capabilities–firm performance relationship. Innovation capabilities matter, but their impact depends on proximity, measurement choices, firm size regimes, and research design. By situating these findings within Schumpeterian theory and capability ontology, with this study, we intended to provide a more nuanced and integrated explanation for decades of mixed empirical evidence.
ConclusionThis study aimed to quantitatively determine the overall magnitude of the relationship between innovation capabilities and firm performance, and to explain why and under what conditions this relationship varies across studies. By synthesizing evidence from >320,000 firms through a systematic review and meta-analysis, we examined the Schumpeterian hypothesis in both its broad and ‘narrow’ forms, distinguishing between Mark I (i.e., creative destruction, typically associated with entrepreneurial firms disrupting incumbents) and Mark II (i.e., creative accumulation, typically associated with large firms building on existing capabilities). We also assessed how different proximity of innovation capabilities–from distal processes, such as investment in new knowledge, to proximal outputs, such as patents and new products–relate to performance outcomes. Similarly, we evaluated how different metrics of firm performance measurements vary in the strength of their association with innovation capabilities.
Our findings illustrate that innovation is positively associated with firm performance overall, but the strength of this relationship varies substantially depending on the theoretical, contextual and methodological contingencies. Notably, proximal capabilities, which directly generate tangible innovation outputs, show the most significant performance effects, while Mark I innovation patterns are more strongly associated with performance gains than Mark II patterns alone. These results offer a more nuanced and integrated understanding of the mechanisms through which innovation influence firm performance. Importantly, this research provides a robust empirical foundation for scholars and practitioners seeking clarity on when, how, and why innovation capabilities matter, reinforcing the practical value of targeted innovation investments and advancing theoretical debates surrounding Schumpeter’s legacy in contemporary innovation research.
Despite our contributions, several limitations should be acknowledged. First, the heterogeneity of studies included in the meta-analysis points to a potential area for future research. The observed variability suggests that at least one significant unexamined moderator that we have not investigated but it is influencing the relationship between innovation capabilities and firm performance. Therefore, future research should explore additional contextual factor alongside innovation, such as industry characteristics or technological conditions, that may shape these dynamics.
Second, we acknowledge that all our data are correlational, leaving open the possibility of reverse causality. This implies that firm performance could, in some cases, influence the firm-level capabilities coded as innovation. Although, we controlled for the one-way relationship between innovation capabilities and firm performance, we cannot directly test this alternative explanation. The substantial variation observed in our results suggests that any situational effects of firm performance on innovation capabilities are likely limited. Relatedly, we note that our data are all cross-sectional, making causal conclusions of innovation capabilities on firm-level performance open to critique.
We also note that prior research, based on the number of effect sizes reported, has predominantly focused on distal and stochastic innovation capabilities, in contrast to proximal when testing Schumpeter’s hypotheses. We suggest that future research and analysis, in our view, are needed to fully understand the nuances and implications of these findings within the context of innovation capabilities and firm performance.
To conclude, we hope that our work inspires a generation of scholars to extend our findings in several ways. First, researchers should investigate distinctions in innovation capabilities, including process-level versus outcome-level measures, and investigate longitudinal effects (e.g., cross-lagged meta-analysis) to better understand the timing and persistence of the association between innovation capabilities and firm performance. Second, scholars should expand our cross-country comparisons using richer cultural indices (e.g., long-term orientation) or broader institutional contexts to clarify why some innovations translate more effectively into performance than others. Finally, future research should incorporate multidimensional, dynamic, and context-sensitive measures of innovation capabilities to enhance the explanatory power of the Schumpeterian hypothesis identified in this meta-analysis.
FundingThe authors declare that no funds, grants, or other support were received during the preparation of this manuscript. The authors have no relevant financial or non-financial interests to disclose.
Ethics approvalNot applicable, as this study did not involve human or animal subjects.
ConsentNot applicable, since the study did not involve collection of data from human participants.
Data, material and/or code availabilityThe data from this study comprise statistics extracted from publicly available academic work. The sample and R code for replication of this study is available from the Open Science Framework (https://osf.io/z4bqx/).
CRediT authorship contribution statementFilip Lestan: Conceptualization, Methodology, Software, Formal analysis, Data curation, Writing – original draft, Writing – review & editing. Tommy H. Clausen: Conceptualization, Supervision, Writing – original draft.
The authors declare no competing interests.
We thank two anonymous reviewers and Editor M. Angeles López-Cabarcos for their constructive comments on earlier versions. We are thankful to authors of three under-reported studies who provided data. Finally, we thank participants of the 45th SMS Annual Conference in San Francisco, the 85th Annual Meeting of the Academy of Management in Copenhagen, as well as Professors Dermot Breslin, Michal Kozak, Ut Na and Johan Wiklund for their valuable comments on earlier versions of this research. All remaining errors are ours.
The same study published multiple times, either in different journals or under slightly altered titles.
A study formally withdrawn from the scientific record due to errors, ethical issues, or misconduct.
We remain thankful to the authors of three under-reported studies in our meta-analysis who provided data upon request. The initial contact was 28 October 2024; first reminders were sent on 31 October 2024; and the final (second) reminders were sent on 3 November 2024. We collected missing information for three out of three under-reported primary studies relevant for our meta-analysis leaving us with 100% response rate. First, Schäper et al. (2023) provided sample size (n=9,100) which has not been originally reported in their study. Second, Arrighetti et al. (2014) provided some correlations which have not been originally reported in their study, and Ballot et al. (2015) provided correlations for two independent samples of French and UK firms.
Samples with low uncertainty avoidance culture values consists of: India, Great Britain, Canada, China, USA, Greece, Finland, Sweden, the Netherlands (Hofstede’s Index below 60); on the other hand, countries with high uncertainty avoidance culture values consists of: Mexico, Italy, France, Brazil, Spain, Germany, Israel, South Korea, Turkey, Japan (Hofstede’s Index greater than 60).
articles
- Download PDF
- Bibliography
- Additional material