Three decades of research on innovation and inequality: Causal scenarios, explanatory factors and suggestions

Prompted by rising income inequality (in short, inequality) in advanced economies, a rapidly growing number of studies across various fields and disciplines of social science have, since the 1990s, sought to find out how innovation (as the main engine of economic progress) affects the distribution of income in modern-day capitalist societies. Using the systematic literature review method, this paper provides the first critical review of 166 studies on innovation and inequality published in 114 journals in the last three decades (1990–2019). It is shown that, while the great majority of studies under review concur that innovation induces inequality, this finding is subject to the disciplinary origins of research (e.g., development studies, economics, geography, innovation studies, etc.) and the country under investigation. Furthermore, guided by an original causally holistic analytical framework, the analysis demonstrates that the relationship between innovation and inequality is significantly more causally complex than the most popular theoretical perspective (i.e., skill-biased technological change account) has let us believe; in particular, it is subject to five causal scenarios and a range of explanatory factors (i.e., skill premiums, technological unemployment, international trade, declining union membership, spatial aspects, changing employment conditions, policy, horizontal inequalities, sectoral composition and types of innovation). The paper ends by discussing findings, policy implications and knowledge gaps, one of which concerns the following under-researched question: how, and under what conditions do publicly funded innovation policies reduce (or increase) inequality?


Introduction
What do the contributions of notable thinkers -such as Adam Smith (1776Smith ( /1982, David Ricardo (1891), Karl Marx (1999), Thorstein Veblen (1899), Joseph Schumpeter (1934, 1944 and Werner Sombart (1967) -have in common other than their obvious significance for contemporary socioeconomic thought? In a nutshell, the classics of socioeconomic thought are replete with passages demonstrating that innovation 1 is (bi-)causally related to inequality 2 in capitalist societies. Despite this, innovation scholars had, for several decades of the twentieth century, examined mainly the positive side of the story, particularly the relationship between innovation, employment creation, competitiveness and growth (Fagerberg, 1994;Pianta, 2005;Antonelli, 2009). The question of ameliorates inequality (causal scenario III); inequality hampers innovation (causal scenario IV). Thirdly, the analysis identifies numerous determinants (i.e., skill premiums, technological unemployment, international trade, declining union membership, geographical aspects, changing employment conditions, policy, horizontal inequalities, sectoral composition and types of innovation) that appear to be shaping (in the form of causal mechanisms) the multidimensional direction and strength of causality. Finally, and because of its critical outlook, the paper detects and challenges several prevalent assumptions and methodological practices, such as the following: and Milanovic, 2016;Milanovic, 2016). OECD data confirm that, as measured by the Gini coefficient, 5 inequality has risen considerably in nearly all of the 37 OECD member states: from 0.29 in the mid-1980s to 0.315 in 2013 (OECD, 2015). Similarly, the 90/10 percentile (another widely used measure of inequality) shows that the wealthiest 10% of the population in OECD countries earned ten times more than the remaining 90% of the population in 2013 (OECD, 2015). Compared with the 1980s, this ratio has increased by 37%. Other studies show that the top 1% of income earners (i.e., 99/1 percentile) have made unprecedented income gains (Atkinson et al., 2011;Alvaredo et al., 2013;Dorling, 2019), and this has occurred at a time when some quite old and worrisome social phenomena -such as the 'working rich', 'working poor', 'underpaid and overworked' -have been re-emerging from the dustbin of economic history (Bogliacino, 2009;Lohmann, 2009;Sayer, 2015;Pianta, 2018;Dorling, 2019).
One may wonder why has inequality kept rising during one of the affluent periods in the history of the capitalist system? After all, eminent economists, such as Milton Friedman (2002), as well as the public speeches of iconic conservative politicians (e.g., Margaret Thatcher and Ronald Reagan), have taught us that growing inequality is a transitory social evil on the path to economic equality in capitalist societies (Harvey, 2005;Senker, 2015;Albertson and Stepney, 2020). Similarly, Kuznets's (1955) inverted-U curve hypothesis (also known as the Kuznets curve) predicts that inequality rises in the early stages of economic growth, then it peaks, before subsequently reaching a historic low (cf. Alderson and Nielsen, 2002;Stiglitz, 2012;Piketty, 2014;Milanovic, 2016).
Traditionally, social scientists -especially Marxist economists, geographers and sociologists -have approached the question of rising inequality from the standpoint of the class struggle (e.g., Braverman, 1974;Peet, 1975;Wright, 1994;Smith, 2010;Piketty, 2014;Papaioannou, 2016). From this perspective, inequality is the outcome of (over)exploitation between the two antagonistic social classes in capitalist societies, namely capitalists and labourers. Dissatisfied with the abstract and deterministic outlook of the class struggle perspective, more recent research (circa 1990s) has sought to understand rising inequality in a more theoretically and empirically diverse manner (Neckerman and Torche, 2007;Lemieux, 2008;Vallas and Cummins, 2014;Cavanaugh and Breau, 2018).

Innovation and inequality: review framework
How does innovation shape the distribution of income in contemporary societies? Unfortunately, owing to the predominance of skill-biased technological change (SBTC) research, 6 the broader academic discourse regarding innovation and inequality gives the impression that all that exists, in terms of causality, in the relationship between innovation and inequality is that the former has a significant impact on the latter, mainly through the skill premiums mechanism.
However, a closer examination of the relevant (empirical) literature reveals an entirely different picture. On the one hand, and in line with the SBTC account, several studies suggest that innovation is positively associated with inequality (e.g., Krueger, 1993;Lee, 2011;Breau et al., 2014). On the other hand, and in contrast to the SBTC account, other contributions allude to the fact that innovation lessens inequality (Lundvall, 2002;Heeks et al., 2014;Antonelli and Gehringer, 2017). Yet another line of research demonstrates that it is inequality that affects, either positively or negatively, the development of innovative activities in contemporary societies (e.g., Falkinger and Zweimüller, 1997;Tselios, 2011;Vona and Patriarca, 2011;Woodson et al., 2019). Thus, to offer an eclectic disciplinary overview of the existing empirical literature, as well as to analyse, reconcile and synthesize contradictory research findings, this paper develops an analytical framework (henceforth, review framework). Central to this are five causal scenarios, each of which has its own theoretical origin.
• Causal scenario 0 -absence of causality. Today, it is commonplace to argue that innovation is a major force behind rising rates of labour productivity, employment creation, profitability, growth and standards of living in general (Schumpeter, 1934;Freeman and Louca, 2001;Pianta, 2005;Antonelli, 2009). However, this was not always the case. Neoclassical economists (e.g., Solow, 1956), for instance, had long argued that economic growth is best studied as a function of two factors: capital and labour. This view, among others, was challenged by early neoclassical growth research, particularly by Solow (1957), whose analysis of US growth shows that the variables of capital and labour leave unexplained as much as 90% of the variance in US growth rates. To account for this residual (also known as the 'Solow residual'), innovation was introduced -initially in the form of technical change (a total factor productivity measure) -to a new generation of neoclassical growth models (Fagerberg, 1994;Antonelli, 2009). As far as the relationship between innovation and growth is concerned, neoclassical growth theory implies that rising technological intensity and inequality are two unrelated phenomena (Violante, 2008;Cozzens and Kaplinsky, 2009): innovation is assumed to be exogenous and factor-neutral, meaning that it benefits the skills, marginal productivity and average wages of all economic agents equally. Although no longer influential, the neoclassical perspective on growth raises, in the context of this study, the possibility that innovation and inequality may not always be (bi-)causally related. • Causal scenario I: innovation induces inequality. According to Schumpeter's (1934) theory of economic development, innovation encompasses the development of new products, services, organizational models and markets. In doing so, innovation creates new competences, while gradually destroying those that are no longer needed in the innovation process (Archibugi and Lundvall, 2001;Lundvall, 2002). When the competence-building process is socially exclusive (rather than inclusive), innovation tends to intensify existing socioeconomic inequalities, such as horizontal (gender and racial) inequalities (Gray et al., 1998;Asheim and Gertler, 2005;Cozzens and Kaplinsky, 2009;Juhn et al., 2014;Cheng et al., 2019;Feldman et al., 2021). In a similar manner, the skill-biased technological change (SBTC) account maintains that innovation creates and intensifies skill premiums, i.e., the wage gap among skilled and less skilled employees (Acemoglu, 2002;Violante, 2008), while the more recent version of the SBTC account (i.e., task or routine-biased technological change account) argues that innovation leads to income polarization through both skill premiums and technological unemployment; for instance, by replacing highly routinized job tasks with artificial intelligence and robots (Autor et al., 2003(Autor et al., , 2008Brynjolfsson and McAfee, 2012;Frey and Osborne, 2017;Goos, 2018;Pianta, 2018Pianta, , 2020Cirillo et al., 2021). Furthermore, owing to its highly uncertain and failure-prone character (Schumpeter, 1934, Kline andRosenberg, 1986), innovation can embed an unequal distribution of risks and rewards (Lazonick and Mazzucato, 2013). Thus, when the costs of innovation are collectively undertaken (e.g., state, universities, research institutes), but the benefits of innovation are distributed mainly within the boundaries of the innovative firm (e.g. shareholders, top executives and employees), innovation can lead to (top) income inequality (Lazonick and Mazzucato, 2013;Bapuji, 2015;Aghion et al., 2019;Tomaskovic-Devey and Avent-Holt, 2019;Munir, 2021). • Causal scenario II -inequality stimulates innovation. The idea that inequality shapes the nature and direction of innovative activity has a very long intellectual pedigree in social science. For instance, Karl Marx's (1999) work on social class, Thorstein Veblen's (1899) analysis of the leisure class, Werner Sombart's (1967) theory of economic development, and, more recently, Pierre Bourdieu's (1987) work on social distinction, underline that inequality has a profound effect on innovation (and economic development in general). In a similar manner, neoclassical economists have long believed that inequality provides strong incentives for economic agents (i.e., incentive thesis) to do the 'right things', such as working harder (e.g., productivity gains) and engaging in growth-boosting (Schumpeterian) activities, such as innovation and entrepreneurship (Falkinger and Zweimüller, 1997;Samuelson, 2010;Sayer, 2015;Xavier-Oliveira et al., 2015;Stiglitz, 2012). Therefore, in theory, it is not only that innovation shapes the distribution of income, but also that the latter moulds the former. • Causal scenario III -innovation ameliorates inequality. Traditionally, innovation has been associated with increased standards of living and economic equality (Schumpeter, 1934;Kuznets, 1955;Freeman, 2001;Freeman and Louca, 2001). For instance, in the golden (Fordist) age of capitalism (between the 1940s and the 1970s), innovation-driven growth led to a significant reduction in (male) unemployment and inequality rates (Freeman, 2001;Pianta, 2005;Cozzens and Kaplinsky, 2009;Atkinson et al., 2011). Because of its creative nature, innovation requires the creation of new competences (Archibugi and Lundvall, 2001;Lundvall, 2002). When the competence building process involves marginalized social groups and actors, innovation can mitigate existing horizontal inequalities (Freeman, 2001;Lundvall, 2002;Arndt et al., 2009;Cozzens and Kaplinsky, 2009;Heeks et al., 2014). In addition, by being a creative destructive process (Schumpeter, 1944), innovation undermines the nature of wealth inequality while also fostering social mobility, as when innovators and entrepreneurs belong to marginalized social groups (Heeks et al., 2014;Antonelli and Gehringer, 2017;Kim and De Moor, 2017). Thus, as with the previous causal scenarios, innovation can mitigate inequality through various causal mechanisms and processes. • Causal scenario IV -inequality hinders innovation. In line with Adam Smith's (1776/1982 theory of the division of labour, Schumpeter's (1934) theory of economic development assumes that the entrepreneurial act of innovation reduces inequality and poverty in capitalist societies over time (Freeman, 1994(Freeman, , 2001Antonelli and Gehringer, 2017). However, in his subsequent work, and echoing the work of Marx (1999) and Veblen (1899), Schumpeter (1944 argues that innovation reinforces existing socioeconomic inequalities in capitalist societies. Schumpeter goes as far as to claim that, if unabated, rising inequality erodes the institutional foundations of long-term economic growth in capitalist societies, potentially leading to the displacement of capitalism by socialism (Elliott, 1980;Henrekson and Jakobsson, 2001;Fagerberg, 2003). Rising inequality engenders crime and corruption, both of which can, over time, transform inclusive institutions into extractive ones (Neckerman and Torche, 2007;Acemoglu and Robinson, 2012;Stiglitz, 2012). The latter can reinforce the significance of certain forms of social capital (e.g., bonding social capital), thus prohibiting the formation of alternative forms of social capital (e.g., bridging social capital) among socially and cognitively diverse actors in the innovation process (Archibugi and Lundvall, 2001;Nielsen, 2003;Fragkandreas, 2012;Barnes and Mattsson, 2016). Furthermore, by reducing the overall demand for new products and services (Falkinger and Zweimüller, 1997;Jung et al., 2018) while also increasing social costs (e.g., tensions and frictions) among affluent and less affluent social groups (Cozzens and Kaplinsky, 2009;Juma, 2016), inequality can hinder the adoption of socially desirable radical innovations (e.g., COVID-19 vaccines), sustainable technological transitions and structural change in general (Freeman, 2001;Geels, 2004;Cozzens and Kaplinsky, 2009;Riaz, 2015). Figure 1 provides a graphical representation of the review framework. 7 The remainder of this paper utilizes this framework as a guide to analysing and synthesizing the findings of the existing research on innovation and inequality.

Systematic literature review
As mentioned in the introductory section, there exists a large body of research on innovation and inequality in various fields of social science. This, in turn, begs the following methodological question: how can one identify, select and critically review the most relevant studies on innovation and inequality? To address this question, this paper adopts the systematic literature review (SLR) method (Tranfield et al., 2003;Petticrew and Roberts, 2008).
Originally used in medical studies, the SLR method is increasingly being adopted in the social sciences (Tranfield et al., 2003;Petticrew and Roberts, 2008;Haddaway et al., 2015). As far 7 It is important to note that Figure 1 offers a schematic overview of the five main causal scenarios in the relationship between innovation and inequality. Because it is simplified, the figure purposely leaves out the indirect links, causal mechanisms and conditional factors in each causal scenario. I would like to thank Hans-Jurgen Engelbrecht for encouraging me to bring this issue to the reader's attention.

Figure 1. Innovation and inequality -review framework
Source: own elaboration as innovation research is concerned, SLRs have recently emerged as the methodological norm when it comes to reviewing the current stock of knowledge on innovation (e.g., Martin, 2012;Doloreux and Porto Gomez, 2017;Compagnucci and Spigarelli, 2020;Kalantaridis and Kuttim, 2021). Like traditional (narrative) reviews, SLRs summarize and synthesize the current state of knowledge in a given research topic or field, as well as identifying key weaknesses and opportunities for further research (Tranfield et al., 2003;Weed, 2005;Petticrew and Roberts, 2008;Randolph, 2009). However, and in contrast to narrative reviews, wherein the analytical steps and procedures do not need to be documented, SLRs state in a clear manner the various stages, sampling criteria and method of analysis (Tranfield et al., 2003;Weed, 2005;Petticrew and Roberts, 2008;Haddaway et al., 2015).
Furthermore, and in contrast to other review methods (e.g., meta-analysis and metainterpretive or ethnographic reviews) in which the underlying emphasis is on either quantitative or qualitative research (Weed, 2005;Randolph, 2009;Brannan et al., 2017), SLRs often incorporate the findings of both quantitative (extensive) and qualitative (intensive) studies 8 (Doloreux and Porto Gomez, 2017;Compagnucci and Spigarelli, 2020). Because of their underlying methodological procedures, SLRs can review a much larger number of studies than can narrative reviews, albeit not in an entirely neutral manner (as the work of SLR practitioners implies) (Tranfield et al., 2003;Petticrew and Roberts, 2008). As is the case with any form of scientific analysis, SLRs are theoryladen (Sayer, 2000b;Bhaskar, 2008); thus, their relevance and contribution are contingent upon the theoretical perspective that one takes. As a result of their eclectic nature, a major challenge that SLRs often face is how to synthesize key insights from a very large corpus of studies, especially when the findings are contradictory (Petticrew and Roberts, 2008). To overcome this challenge, this paper uses the review framework as the overall guide to the analysis.

Review sample: collection and analysis
The data in this SLR consist of 166 studies (the review sample) 9 published in 114 journals over the last three decades . 10 The review sample was identified through an iterative search in the Scopus database (https://www.scopus.com/). This database was chosen because it contains 50% more entries than other popular scholarly databases (e.g., Web of Science). A set of keywords was used (in the form of a Boolean equation) to identify the most relevant contributions. These included the following: innovation, technology, technological change, income, wage or earnings inequality, poverty, income distribution and distribution of income, wages and/or earnings. The first search, which was performed in the summer of 2019, identified 1,832 contributions. After excluding conference papers, papers published in predatory journals, 11 conceptual (including formal, mathematical modelling) papers, reviews, book chapters and editorials, as well as after scrutinizing the abstract section of each study for false positives (i.e., articles containing keywords that are relevant but not directly related to the subject), 166 peer-reviewed studies 12 met the following three inclusion criteria: being an empirical study (first inclusion criterion), published in English (second inclusion criterion) and available in a digital form (e.g., PDF) (third inclusion criterion). 8 Following Sayer (2000b) and other critical realist social scientists (e.g., Danermark et al., 2002), this paper refers to qualitative research (e.g., grounded theory, case study research, ethnography, discourse analysis, etc.) as intensive, and to quantitative research (e.g., econometrics, advanced inferential statistics) as extensive. In the critical realist tradition, intensive and extensive research are seen as being both distinct and complementary (e.g., mixed method research) (Downward and Mearman, 2007). 9 For a detailed overview of the sample, see Appendix. 10 This is based on the earliest observation in the data. 11 To do so, a list of predatory journals was used, which was retrieved from the following link: https:// predatoryjournals.net/ (accessed September 2020). 12 The sole focus on peer-reviewed studies is based on the assumption that peer-reviewed published studies often yield reliable and novel findings by applying advanced methodological standards (see Biggi and Giuliani, 2021).
In line with recent reviews on innovation (Doloreux and Porto Gomez, 2017;Compagnucci and Spigarelli, 2020), the review sample was analysed in a systematic manner by using a coding template (see Table 1). This consists of eleven codes. The first six codes (A to F) were developed in the early stages of the review (i.e., a priori coding), whereas the rest of the codes (G to M) emerged from the analysis (i.e., bottom-up coding) in the more mature stages of the review (King and Brooks, 2017). To establish the construct validity (Yin, 2009, p.34) of the coding template, three independent researchers were asked to use the coding scheme to analyse a sample of six studies. As illustrated in Table 1, a very high score of inter-coder reliability was achieved. Finally, following Cooper's (1988) taxonomy of literature reviews, the findings of this review are discussed in a chronological way. As will be shown in the next section, a chronological perspective offers a comparatively rich understanding of the disciplinary origins, development and major findings of three decades of research on innovation and inequality.

Innovation and inequality research: chronological review
One of the earliest observations in this review was that the number of published studies on innovation and inequality has, on average, risen by 220% every ten years ( Figure 2). Even though this number suggests that research on innovation and inequality is growing at a much faster pace than that of research on other topics, 13 the growth in published research did not occur in a linear manner. For instance, while seven studies were published in 2001, this number drops to one study just one year later. Similarly, twelve studies were published in 2009, before this figure fell to six studies in 2010. To capture the ebbs and flows of research on innovation and inequality, the analysis distinguishes among three main research phases: the early phase (1990-9), the growth phase (2000-9) and the expansion phase . The remainder of this section looks more closely at each phase, focusing on key aspects of research, such as bibliometric issues, fields of research, causal scenarios and explanatory themes.

BIBLIOMETRIC INSIGHTS
In the early phase, and unlike the subsequent two phases, research on innovation and inequality was extremely sparse, with less than one published study per year ( Table 2). The paucity of research on innovation and inequality in the early phase reflects key events and developments in the domains of academia, economy and policy. For instance, the advent of free-market capitalism (in short, neoliberalism) as the dominant policy paradigm in the 1980s and 1990s systematically favoured academic discourse and theoretical perspectives that glorify the benefits of extensive economic growth (Harvey, 2005;Smith, 2010;Senker, 2015;Fotaki and Prasad, 2015;Albertson and Stepney, 2020), whereas the negative consequences of growth -such as rising inequality, social exclusion, mental health problems caused by job insecurity, excessive wealth concentration and environmental degradation (Pickett and Wilkinson, 2010;Breau and Essletzbichler, 2013;Sayer, 2015;Biggi and Giuliani, 2021) -were seen as secondary evils that sooner or later would be addressed, in the most efficient manner possible, through the undisturbed operation of (global) markets (Harvey, 2005;Fotaki and Prasad, 2015;Senker, 2015;Albertson and Stepney, 2020). In this context, rising inequality was, initially, seen as a temporary anomaly of the liberal market economies of the US and UK, rather than a general socioeconomic challenge that concerns all market economies equally (Freeman, 2001;Hall and Soskice, 2001;Lundvall, 2002;Piketty, 2014;Dorling, 2019).
Research in the 1990s was ascetic, being based mainly on single author contributions. While single-author contributions were endemic in published research on innovation in the 1990s 13 For instance, bibliometric studies show that the number of published research papers doubles in size every 10-15 years (Bornmann and Mutz, 2015).  (see, for instance, table 1 in Martin, 2012), the early work was highly cited (302 citations per study).
The three most cited studies (i.e., Krueger, 1993;Bernard and Jensen, 1997;Bresnahan, 1999) were published in (mainstream) economic journals. 14 Although the number of citations is by no means a reliable indication of scholarly novelty and quality (Macdonald and Kam, 2011;MacRoberts and MacRoberts, 2018), it nonetheless implies that mainstream economic research has, in one way or another, been highly influential, constituting either an impetus for further research or an object of critique (e.g., Card and DiNardo, 2002;Autor et al., 2008;Lazonick and Mazzucato, 2013;Avent-Holt and Tomaskovic-Devey, 2014;Hanley, 2014). However, as is shown in Table 2, development studies scholars and employment relations scholars were also very active in the 1990s. Thus, unlike previous stock-taking assessments (e.g., Acemoglu, 2002; Acemoglu and Autor, 2011), which give the impression that only mainstream labour economists have investigated the relationship between technological innovation and inequality, research appears to have, from the outset, been significantly more discipline-diverse than previously thought.

RESEARCH CONTEXT, DESIGN, MEASURES AND UNITS OF ANALYSIS
More than 60% of all studies in the 1990s were concerned with the US and UK (Krueger, 1993;Chennells and Reenen, 1998;Bresnahan, 1999), whereas 40% of all studies examined developing countries, such as Indonesia (James and Khan, 1998), Nepal (Thapa et al., 1992) and the Philippines (Otsuka et al., 1990). The focus on the US and the UK can be associated with the fact that most researchers are affiliated with academic organizations in these countries. Innovation was gauged by using narrow measures (e.g., computer usage and R&D intensity) (Krueger, 1993;Machin, 1998), and inequality by utilizing the following measures: percentiles, Gini index and wage gaps (e.g., Thapa et al., 1992;Krueger, 1993;Machin, 1998). More than 92% of studies were extensive, using econometrics and advanced inferential methods, with only one study (Gray et al., 1998), which was published in a heterodox economics journal (Review of Radical Political Economics), using a mixed-method research design. Although no specific methodological reason is stated in all studies under review for the wide use of extensive research designs, this could be linked to the fact that 75% 14 To distinguish between mainstream and heterodox economic journals, Cronin's (2020) list of economic journals was used (available at https://www.hetecon.net/resources/ journal-rankings/).  (Krueger, 1993;Bernard and Jensen, 1997;Bresnahan, 1999) *based on Google Scholar, July 2020 Source: own elaboration, Scopus of all studies in the 1990s were published in economics journals. 15 Employees and individuals are the primary unit of analysis, followed by sectors and countries (cross-country analysis). This, among other things, reveals that, since the beginning, researchers have assumed that the relationship between innovation and inequality is a multilevel one.

CAUSAL SCENARIOS AND EXPLANATORY FACTORS
The majority of studies (approximately 58%) find that innovation induces inequality (causal scenario I) (e.g., Krueger, 1993;Bresnahan, 1999), while 25% opined that there is no causal connection between innovation and inequality (causal scenario 0) (e.g., Colclough andTolbert, 1990, Otsuka et al., 1990;Freebairn, 1995). The remaining studies (17%) imply either that inequality can positively affect innovation (causal scenario II) (e.g., Falkinger and Zweimuller, 1997) or that innovation reduces¨ inequality (causal scenario III) (e.g., James and Khan, 1998). Causal scenario I is mainly attributed to the skill-biased character of technological innovation in general, and to the skill premium mechanism in particular (Krueger, 1993;Bernard and Jensen, 1997;Chennells and Reenen, 1998;Bresnahan, 1999). Krueger (1993), for instance, provides evidence suggesting that US workers who use computers at work earn, on average, 10 to 15% 15 It goes without saying that, unlike the work of classical economists, including the work of other influential economists (e.g., Thorstein Veblen, Werner Sombart, Max Weber, Joseph Schumpeter), modern economistsregardless of being mainstream or heterodox -believe, in one way or another, that econometrics is the most scientifically legitimate method of studying the economic world (Lawson, 1997;Lazear, 2000;Louca, 2007). higher wages. Regarding the possibility that innovation and inequality are not causally related (causal scenario 0), which is the second most popular in the early phase, research does not identify recurrent explanatory factors.
Nonetheless, studies falling within causal scenario 0 raise some interesting questions about the skill-biased technological change hypothesis. For instance, Bernard and Jensen (1997) provide firm-based evidence confirming that while international trade has increased the demand for whitecollar labour in US manufacturing plants, it had no significant impact on the wage gap among white-and blue-collar workers (see also Machin, 1998). Additionally, the analysis of Colclough and Tolbert (1990) raises the possibility that the skill-biased character of technological change may favour the skills, marginal productivity and wages of privileged social groups and actors (e.g., educated, native white men) (see also Echeverri-Carroll et al., 2018;ten Berge and Tomaskovic-Devey, 2021). As will be shown throughout this review, economic studies have paid very little attention to the horizontally biased (i.e., gender and racially biased) character of skill premiums.

BIBLIOMETRIC INSIGHTS
As is the case with the early phase, research in the growth phase consists mainly of single-authored publications: 1.1 authors per study (Table 3). However, unlike the early phase, the number of published studies per year increased tremendously: from 0.8 studies per year to 5.6 studies per year. This can, among other issues, be attributed to a growing interest in the causes and consequences of rising inequality (Neckerman and Torche, 2007;Kim and Sakamoto, 2008;Pickett and Wilkinson, 2010;Cavanaugh and Breau, 2018). Similarly, vivid discussions among mainstream labour economists of the skill-biased character of technological innovation (e.g., Krusell et al., 2000;Card and DiNardo, 2002;Autor et al., 2008), including debates on the impact of international trade and declining union membership upon inequality (Belman and Monaco, 2001;Card and DiNardo, 2002;Autor et al., 2003;Mosher, 2007;Adams, 2008;Autor et al., 2008;Meschi and Vivarelli, 2009), prompted further research on innovation and inequality in the growth phase.

REGARDING FIELDS
No significant change was observed other than that the field of employment relations is the second most active, while the fields of economics and development studies are first and third, respectively. This disciplinary division of research is also reflected in the most popular journals in the period under consideration (e.g., World Development, Journal of Development Economics, Industrial Relations, Labour, Economics of Transition, International Review of Applied Economics, Review of Economics and Statistics, and Industrial and Labor Relations Review). Thus, as is the case with research in the early phase, it is mainly non-economics journals that provided important fora for researchers on innovation and inequality in the growth phase.
*based on Google Scholar, July 2020 Source: own elaboration, Scopus case study research and/or ethnography. This, among other issues, implies that the great majority of researchers in the growth phase see extensive research as ideal in distinguishing what is causal from what is not in the relationship between innovation and inequality. As will be discussed in the concluding section of this paper, the methodological monopoly of extensive research has a number of crucial epistemological consequences for both explanatory research and policy design. One out of three studies (34%) analyses micro-units (e.g., individuals and employees) (e.g., Krusell et al., 2000;McCall, 2000;Englehardt, 2009), while the remainder (64%) focus on the meso-level (e.g., sectors, firms, cities and villages) and/or the macro-level (e.g., countries) (e.g., Kijima, 2006;Bogliacino, 2009;Echeverri-Carroll and Ayala, 2009;Fuchs, 2009;Weinhold and Nair-Reichert, 2009). Thus, unlike research in the 1990s, the more recent research has used sectors and firms as the primary unit of analysis. One possible explanation for this is the availability of firm-level and sectoral data in the 2000s owing to the wide circulation of international (firm-based) surveys on innovation (e.g., Community Innovation Survey) (Smith, 2005;Hong et al., 2012). Nonetheless, innovation is measured in a narrow manner (e.g., computer usage, R&D intensity, patents) (e.g., Xu and Li, 2008;Weinhold and Nair-Reichert, 2009), with only a very small portion of studies using alternative measures, such as the percentage of high-tech employment (e.g., McCall, 2000;MacPhail, 2000), indicators of product and process innovation (e.g., Angelini et al., 2009;Bogliacino, 2009). Additionally, and in line with several studies in the early phase, 25% of all studies treat innovation either as a latent (background) causal factor (e.g., Wheeler, 2005;Kim and Sakamoto, 2008;Xu and Li, 2008;Dustmann et al., 2009) or as export intensity (e.g., Meschi and Vivarelli, 2009). However, since these measures explain very little about the actual nature of innovation (Smith, 2005), several questions are raised as to the construct validity and explanatory power of research in the growth phase. For the measurement of inequality, percentiles (e.g., Cozzens et al., 2002;Kijima, 2006;Borghans and Ter Weel, 2007), the Gini index (e.g., Langer, 2001;Kim and Sakamoto, 2008;Adams, 2008), income and wage gaps (Krusell et al., 2000;McCall, 2000;Bogliacino, 2009) were used widely. The Theil index was also used in some studies (e.g., Cozzens, 2003;Meschi and Vivarelli, 2009) either on its own or in conjunction with other measures, mainly for robustness check purposes.

CAUSAL SCENARIOS AND EXPLANATORY FACTORS
In line with research in the early phase, 70% of all studies in the expansion phase confirm that innovation induces inequality (causal scenario I) (e.g., Krusell et al., 2000;Bogliacino, 2009;Echeverri-Carroll and Ayala, 2009;Wheeler, 2005), whereas the remaining 30% is divided between the rest four causal possibilities: 9% suggest that there is no causality between innovation and inequality (causal scenario 0) (e.g., Brown and Campbell, 2001;Handel and Gittleman, 2004;Belman and Levine, 2004); 8% of studies point out that inequality has a positive impact on innovation (causal scenario II) (e.g., Chakraborty and Bosman, 2005;Englehardt, 2009); 8% has a negative effect on innovation (causal scenario IV); lastly, 5% opine that innovation lessens inequality (e.g., Gibson, 2003;Martin and Robinson, 2007;Mukhopadhyay and Nandi, 2007). In short, as is the case with research in the early phase, the great majority of studies in the growth phase suggest that innovation induces inequality.
Causal scenario I is mainly attributed to skill premiums caused by technological innovation (e.g., Krusell et al., 2000;Wheeler, 2005;Commander and Kollo, 2008;Englehardt, 2009), whereas another much smaller, albeit highly cited, number of studies propose and substantiate empirically the task-biased version of the skill-biased technological change hypothesis, wherein innovation leads to income polarization through skill premiums and technological unemployment mechanisms (e.g., Autor et al., 2003Autor et al., , 2008. Another strand of research suggests that skill premiums are sectorspecific (e.g., high-technology sectors) and geographically confined, occurring mostly in hightechnology sectors and regions (Cozzens, 2003;Wheeler, 2005;Florida, 2007;Angelini et al., 2009;Bogliacino, 2009;Doussard et al., 2009;Echeverri-Carroll and Ayala, 2009). For instance, Echeverri-Carroll and Ayala (2009) find that employees in US cities with a high-technology industry earn, on average, 17% higher salaries than employees in other regions.
Other studies examine the interaction between skills and international trade in both developed and developing countries (Haskel and Slaughter, 2001;Esquivel and Rodrıguez-Lopez, 2003;Attanasio et al., 2004;Baldwin and Cain, 2000;Kijima, 2006;Xu and Li, 2008;Bogliacino, 2009;Meschi and Vivarelli, 2009). The findings of these studies lead to two contradictory conclusions. On the one hand, it is skill-biased technological change, rather than international trade per se, that leads to inequality via the skill premiums mechanism (e.g., Commander and Kollo, 2008). On the other hand, the complementary dynamics among innovation, international trade and organizational factors (e.g., the ownership structure of innovative firms) trigger export-induced skill premiums (e.g., Xu and Li, 2008;Bogliacino, 2009).
In addition to the above, non-economic studies show that skill premiums have a strong horizontal dimension (MacPhail, 2000;Fernandez, 2001;Taylor, 2006). Fernandez (2001), for instance, finds that the adoption of technological innovation in a US food firm led to 'greater racial inequalities in wages' (Fernandez, 2001, p.273). Another line of research raises the possibility that skill premiums may also be induced by such non-market forces as policies, especially policies (a) aimed at boosting high-technology employment and growth in regions (Cozzens et al., 2002;Mukhopadhyay and Nandi, 2007), or (b) reinforcing the intellectual property rights regime (Adams, 2008;Arndt et al., 2009). A relatively small number of mainly employment relations studies (Belman and Monaco, 2001;Brown and Campbell, 2001;Black et al., 2004;Handel and Gittleman, 2004;Mosher, 2007;Doussard et al., 2009) underline that the ability of innovation to induce inequality is subject to both institutional (e.g., declining union membership and collective wage bargaining) and organizational factors (e.g., new employment practices). Belman and Monaco (2001), for instance, show that, thanks to labour market deregulation, the use of advanced technologies (e.g., satellite communication systems) led to a reduction of 21% in the wages of US truck drivers in the 1990s. Lastly, Black et al. (2004) find that new flexible employment practices (e.g., job rotation) are associated with lower employment reductions but higher wage inequality (cf. Handel and Gittleman, 2004).
Regarding the second most observed causal possibility (i.e., absence of causality, causal scenario 0), research in the growth phase provides no clear insight in terms of recurrent explanatory factors. Nonetheless, some of these studies offer a few interesting insights into the explanatory validity of the skill premiums hypothesis. For instance, Kim and Sakamoto (2008) find in their analysis of US manufacturing industries that the adoption of radical technological innovation at work increased wage inequality but not labour productivity as the skill-biased technological change account assumes (Acemoglu and Autor, 2011); in short, skill premiums do not necessarily reflect human capital factors, such as higher labour productivity (see also Hanley, 2014;Tomaskovic-Devey and Avent-Holt;. Other studies (Mishel and Bernstein, 2003;Borghans and Ter Weel, 2007;Xu and Li, 2008) suggest that, since wage inequality has not risen to the same extent in all countries (e.g., OECD, 2015; Kawaguchi and Mori, 2016), the inequality-inducing abilities of innovation (e.g. skill premiums and technological unemployment) seem to be significantly curtailed by non-market factors such as employment strategies, organizational structures and national institutional arrangements (e.g., Card and DiNardo, 2002;Belman and Levine, 2004;Goos et al., 2014;Hanley, 2014;Boyer, 2015;Kawaguchi and Mori, 2016;Croce and Ghignoni, 2020;Tomaskovic-Devey and Avent-Holt, 2019).
Regarding the remaining causal possibilities, namely that inequality stimulates innovation (causal scenario II), inequality hinders innovation (causal scenario IV) and innovation reduces inequality (causal scenario III), research in the growth phase is not especially illuminating. An exception is a few studies that investigate the relationship between existing socioeconomic inequality and the diffusion of innovation (e.g., Gibson, 2003;Chakraborty and Bosman, 2005;Martin and Robinson, 2007). Following this (mainly non-economic) line of research, it seems that existing horizontal inequalities adversely affect the ability of marginalized actors to participate in and take advantage of (digital) innovation activities (Gibson, 2003;Cozzens and Kaplinsky, 2009;Fuchs, 2009;Vona and Patriarca, 2011). Gibson (2003), for instance, examines the use of information and communication technologies (ICT) in Australia. Using data gathered by Australia's national census, the author identifies a significant digital gap among households and territories in Australia. Similarly, like Martin and Robinson's (2007) analysis in the US, as well as Mendonça et al.'s (2015) analysis in Portugal, Chakraborty and Bosman (2005) indicate that digital inequality has a persistent horizontal dimension in the US: 'while income inequalities among PC owners (households) decreased between 1994 and 2001 in all regions and states, the magnitude of this inequality has declined more rapidly among white households compared to African Americans' (Chakraborty and Bosman, 2005, p.395). Overall, research in the growth phase identifies several recurrent explanatory factors in most causal scenarios. As will be shown shortly, research in the expansion phase has, in general, moved along similar lines.

BIBLIOMETRIC INSIGHTS
As with the growth phase, the expansion phase exhibits a significant increase in publications, from 5.6 studies per year to 9.7 studies per year (Table 4). This could be associated with the occurrence of social movements (e.g., Occupy Wall Street and We Are the 99%), including the global financial crisis and the striking income inequalities (e.g., excessive pay compensation packages and bonuses) that were brought to light (Blankenburg and Palma, 2009;Crotty, 2009;Sayer, 2015). All of these have triggered further debates and research on the underlying causes and consequences of inequality (Pickett and Wilkinson, 2010;Stiglitz, 2012;Breau and Essletzbichler, 2013;Bapuji, 2015;Arestis, 2020).
Number of authors per study increases from 1.1 to 1.9. On the one hand, this reflects the broader trend among innovation researchers towards collaboration (Fagerberg et al., 2012;Martin, 2012). On the other hand, this implies that conducting and publishing research on innovation and inequality have become more demanding and time-consuming than previously. Nonetheless, the number of citations per study is lower than in the previous two phases at 118. Arguably, this could be attributed to older studies being more likely than recently published ones to have more citations. Of importance is also the fact that, unlike in the previous two phases, wherein, on average, only 15% of published studies received financial support, more than 32% of published studies were sponsored by academic organizations, think tanks and policy organizations, with the most active non-academic donors being located in the UK (e.g., Economic and Social Research Council, UK Research and Innovation), Europe (e.g., European Commission), the US (e.g., National Science Foundation) and South Korea (National Research Foundation of Korea). Although an in-depth analysis of the power issues and dynamics between sponsors and researchers is beyond the scope of this paper, it is important to mention that external funding activities seem to have reinforced, albeit not necessarily intentionally, certain disciplinary discourses and types of research in the expansion phase, such as the research focus on skill premiums, few countries and research methods (see Table 4).

RESEARCH FIELDS
As for research fields, a significant reshuffle occurred in the expansion phase. Unlike in the growth phase, where the fields of economics, development studies, and employment relations were the three most active, the fields of economics, innovation studies, geography and sociology are the first, second, third and fourth most active in the post-2010 period, respectively. The emergence of innovation studies journals (e.g., Technological Forecasting and Social Change, Industrial and Corporate Change), geographical journals (e.g., Regional Studies) and sociological journals (e.g., American Behavioral Scientist, American Sociological Review) in the list with the most preferred journals is illustrative of this trend.
However, a closer examination of published studies in these journals reveals several important insights and critical observations. Specifically, while at first sight it appears that innovation studies researchers have begun to pay some serious attention to inequality (see, for instance, Faggio et al., 2010, Lazonick andMazzucato, 2013), the rise in published innovation research is attributable to guest editorials (e.g., Coad et al., 2021;Cozzens, 2012) rather than to independent studies. This, among other issues, raises important questions as to the role that the peer review mechanism might play in shaping the research agenda in the field (Macdonald, 2015;Martin, 2016). Questions are also raised with regard to the absence of the flagship journal of the field of innovation studies (i.e., Research Policy) and Prometheus from the list of the most active journals. This is quite surprising because both journals seek, by tradition, to publish critically minded research on innovation (Cozzens, 2003;Fagerberg et al., 2012). Judging from this situation, it seems that, unlike the work of the founding figures of the field (e.g., Christopher Freeman, Dick Nelson and Bengt-Ake Lundvall), where economic and societal challenges (e.g., jobless growth, social inclusion and technological unemployment) figured prominently (e.g., Archibugi and Lundvall, 2001;Lundvall, 2002;Fagerberg et al., 2011), the great majority of contemporary innovation researchers seem to be  (Frey and Osborne, 2017), geographical aspects (Lee, 2011;Consoli et al., 2013;Breau et al., 2014;Florida and Mellander, 2016;Otioma et al., 2019), digital gap (Hilbert, 2010), horizontal inequality (Brouwer and Brito, 2012;Brynin and Perales, 2016;Juhn et al., 2014;Echeverri-Carroll et al., 2018;Cheng et al., 2019), deunionization (Kristal, 2013;Kristal and Cohen, 2017;Stockhammer, 2017), innovation policy (Cozzi and Impullitti, 2010;Lee, 2019), organizational factors (Hanley, 2014), types of innovation (Thakur, 2012;Richmond and Triplett, 2018) Source: own elaboration, Scopus * based on Google Scholar, July 2020 interested in conducting research that primarily reflects the interests of a few select actors (e.g., elite scholars and policymakers) rather than society as a whole (see also Martin, 2016). Geographers have also been quite active in the expansion phase, publishing several wellconducted studies (e.g., Lee, 2011;Consoli et al., 2013;Lee and Rodrıguez-Pose, 2013;Breau et al., 2014;Guo, 2019;Otioma et al., 2019). However, by investigating mainly cities and regions in the US (Lee and Rodrıguez-Pose, 2013), Europe (Lee, 2011;Tselios, 2011) and Canada (Breau et al., 2014), geographical research has extended, yet intensified, our knowledge of a few Englishspeaking countries (e.g., the US, the UK and Canada). While the choice to investigate a certain group of cities and regions over others is determined by data availability (e.g., Lee, 2011;Tselios, 2011), the fact that several geographical studies in the expansion phase received financial support from organizations based in the UK and Europe also seems to have played a role.
Despite being 'too late to join the party', sociological studies have looked mainly at the relationship between innovation and inequality in the US. (e.g., Fernandez, 2001;DiPrete, 2007;Kristal, 2013;Hanley, 2014;Kristal and Cohen, 2017). However, unlike most innovation and geographical studies, which seem to have uncritically adopted the underlying assumptions and hypotheses of the skill-biased technological change account (e.g., Wheeler, 2005;Lee, 2011;Breau et al., 2014;Cirillo et al., 2021), sociological studies tend to problematize, criticize and empirically illustrate that the account in question, including its variants, is misleading and handicapped (e.g., Fernandez, 2001;Kristal, 2013;Hanley, 2014;Kristal and Cohen, 2017). Yet another emerging line of sociological research seeks to develop an alternative explanatory account based on relational inequality theory (Avent-Holt and Tomaskovic-Devey, 2014;Hanley, 2014;Vallas and Cummins, 2014). Nonetheless, despite being equipped with a sophisticated theory of income distribution as a relational-organizational phenomenon (Avent-Holt and Tomaskovic-Devey, 2014;Tomaskovic-Devey, 2014), this sociological line of research lacks -as is the case with SBTC research -an appropriate theory of innovation (see also Lazonick and Mazzucato, 2013;Fragkandreas, 2021).
Regarding popular explanatory factors, no significant change is noticed: causality in its various forms is related to the same explanatory factors as in the growth phase (for more information, see Table 4). However, and unlike in the previous two phases in which the great majority of studies assessed the (statistical) impact of a few explanatory factors (in the form of independent variables), a number of studies in the expansion phase consider competing or alternative explanations for causal scenario I (e.g., Almeida and Afonso, 2010;Jaumotte et al., 2013;Lin and Tomaskovic-Devey, 2013;Kristal, 2013;Kristal and Cohen, 2017;Stockhammer, 2017). For instance, Kristal (2013) and Kristal and Cohen (2017) provide evidence that rising inequality in the US is primarily driven by workers' disempowerment rather than by skill premiums associated with technological change (e.g., Acemoglu et al., 2001). Thewissen et al. (2018) extend this finding by exploring the drivers of earnings inequality at the sectoral level in eight OECD countries. The findings 'provide mixed evidence for the hypothesis that skill-biased technological change increases earnings inequality' (p.1023). On the contrary, Thewissen et al. (2018) show that waning labour union power is an important driver of earnings inequality in the countries under investigation. Similarly, in their study of 109 developing and developed countries, Richmond and Triplett (2018) confirm that the causal association between innovation and inequality is conditioned not only by types of innovation (e.g., product or process innovation) and sectoral technological intensity (see also Angelini et al., 2009;Broccolini et al., 2011), but also by the economic and political characteristics of each country (see also Dell'Anno and Solomon, 2014;Iversen and Soskice, 2015;Goel, 2017;Antonczyk et al., 2018). As will be discussed in the concluding section, identifying concrete configurations of causal factors (i.e., causal mechanisms) that enable (or constrain) certain types of innovation to induce (or reduce) inequality in certain places (e.g., cities, regions and nations), but not in others, constitutes a promising research opportunity.

Main findings
This paper is among the first to identify and review in a critical, systematic manner the extant stock of knowledge on innovation and inequality in various fields of social science. Driven by a novel analytical framework, the analysis yields several novel findings and critical observations. Specifically, and in line with previous reviews (Acemoglu, 2002;Bogliacino, 2014;Lee, 2016), including research on skill-biased technological change (STBC) (Acemoglu and Autor, 2011), the present review confirms that most studies (approximately 71%) find that innovation induces inequality in contemporary capitalist societies (causal scenario I). However, and in contrast to previous reviews, including SBTC research, it was shown that a considerable number of studies (approximately 30%) point to four other causal possibilities (see Table 5). In short, there is much more to be understood about causality in the relationship between innovation and inequality than research on SBTC has assumed. Important also is the fact that, unlike previous contributions that cultivate the impression that it is mostly mainstream economic research that drives our knowledge on innovation and inequality, this review shows that, from a cross-disciplinary standpoint, this view is misleading. While economic studies do, indeed, dominate our knowledge on causal scenario I, development studies and employment relations studies lead our knowledge on causal scenarios 0, II and IV (see Table 6). Similarly, there appears to be a clear disciplinary perspective on explanatory factors (see Table 7). Mainstream economic research attributes causality to market-related factors (e.g., skill premiums, trade and technological unemployment). In contrast, research in other fields, including heterodox economic research, is more likely to examine -in addition to skill premiums -a host of other non-market factors (e.g., deunionization, types of innovation, innovation diffusion process, changing employment conditions, organizational factors, spatial aspects, digital gaps and sectoral change). While this finding showcases distinct specialization of knowledge among different fields,

Knowledge gaps and critical remarks
The review process has detected several essential knowledge gaps that research could address in the years to come (for an overview, see Table 9). Specifically, in all causal scenarios, our knowledge on causal mechanisms is significantly wanting -causal mechanisms remain essentially black boxes that future research needs to unpack. This critical knowledge omission stems from an implicit methodological consensus in the literature that quantification and statistical significance are integral to a sophisticated analysis of causality, despite the fact that how innovation causes inequality 'has nothing to do with the number of times we have observed it happening' (Sayer, 2000b, p.14). This, among other issues, implies that observing a statistical net effect masks a multi-causal reality wherein reinforcing and antagonistic mechanisms shape the relationship between innovation and inequality. Putting it differently, strong, weak or absent statistical associations are, on their own, an unreliable indication of operative causal mechanisms.
Time is also important, namely to what extent are causal mechanisms ephemeral or enduring? Even though this study observed no significant difference in terms of findings among the different types of studies (short-, medium-and long-term), future research needs to make use of intensive research designs and methods (e.g., case study research, grounded theory and ethnography) as a means by which to extend and deepen our knowledge of the enduring nature of causality in general and causal mechanisms in particular (Archer, 2015;Fragkandreas, 2021). The views and experiences of, among others, employees, managers, and policymakers, including marginalized social actors, need to be integral to explanatory causal analysis of innovation and inequality. Otherwise, and because of its excessive, yet naive reliance on secondary statistics, the extant research could be criticized for being externalist (i.e., deliberately detached from the everyday world) and elitist (i.e., based exclusively on the views of researchers rather than the views of social actors).
In addition, research appears to have been neglectful of several key stylized facts regarding our knowledge of innovation (Fagerberg et al., 2012;Martin, 2012). While several decades of innovation research have shown that innovation is a collective activity, often encompassing intense collaboration among a wide array of private and public actors (e.g., firms, suppliers, universities, governmental organizations, laboratories, banks, venture capitalists etc.) (Edquist, 2005;Lundvall, 2013), none of the studies under review have, to date, examined how collectivities of innovative actors (e.g., clusters, networks of innovation and innovation systems) shape the distribution of income. This is a significant knowledge gap as the skill premiums mechanism may, after all, be the result of network fragmentation (e.g., absent or weak university-industry interactions) among focal (triple helix) actors in innovation systems rather than simply the outcome of the supply and demand forces in labour markets (see, for instance, Christopherson and Clark, 2007;Lawton-Smith, 2009;Fragkandreas, 2021).
Furthermore, the great majority of studies under review seem to suffer from 'linear technofetishism' in the sense that innovation is conceptualized and analysed mainly as being a technological, linear, R&D-driven process. Future research needs to go beyond the narrow technological variables of innovation to examine the impact that different types of innovation (e.g., business model, product, incremental, organizational and institutional innovation) have on the distribution of income; for instance, by utilizing data from innovation surveys (Smith, 2005;Hong et al., 2012) and alternative methodological approaches to extensive research (e.g., qualitative comparative analysis) to identify configurations of factors (see, for instance, Greckhamer et al., 2018). This could extend our knowledge not only on the composition of causal mechanisms, but also on the impact that different types of innovation have on the distribution of income, including the reverse (Veblenian) case in which existing inequalities, especially wealth inequality, shape the nature, direction and success of innovation (Cozzens and Kaplinsky, 2009;Piketty, 2014;Rikap and Lundvall, 2021).
In addition, extremely little is known about the distribution of economic rewards among innovative actors (e.g., global innovation networks, value chains and production networks) (Cozzens and Kaplinsky, 2009;Rikap and Lundvall, 2021). Future research needs to examine systematically wage inequality within (and between large and small) innovative firms. For instance, are large firms more unequal than small firms (Cirillo et al., 2017;Song et al., 2019)? Emphasis must also be placed on the (ontological) fact that the income that innovation generates is, primarily, distributed within the legal boundaries of the firm (rather than in labour markets as the bulk of the extant economic literature implies) (Lazonick and Mazzucato, 2013;Tomaskovic-Devey and Avent-Holt, 2019;Rikap and Lundvall, 2021). Research on this issue could also help us to understand better the significantly overlooked relationship between innovation and top income inequality (Lazonick and Mazzucato, 2013;Aghion et al., 2019), particularly how a set of high-income organizational actors (e.g., top executives) manage to convince other organizational actors (e.g., employees, labour union representatives and shareholders) that they deserve a significant share of the value that innovation generates (Kay and Hildyard, 2021;Rikap and Lundvall, 2021), even though, as Lazonick and Mazzucato (2013) emphasize, high-income organizational actors, including larger firms (e.g., hightechnology giants), do not necessarily bear the lion's share of the risks involved in the innovation process (Rikap and Lundvall, 2021). This issue brings to the fore a largely under-researched aspect in the relationship between innovation and inequality, namely the nexus between innovation and wealth inequality: how does innovation affect wealth inequality and vice versa? Relatedly, more research needs to be dedicated to the distributional impact of innovation policies (Cozzens et al., 2002;Zehavi and Breznitz, 2017;Schot and Steinmueller, 2018). In particular, how, and under what conditions, do innovation policies reduce or increase income differentials? This is an essential question with far-reaching implications for both theory and policy.

Policy implications
Despite being a critical stocktaking exercise, the present paper has a few policy implications and recommendations. The analysis in this paper confirms, among other issues, that contemporary innovation scholars and policymakers are right (albeit belatedly) to question the trickle-down thesis whereby innovation-driven growth will over time benefit less affluent individuals and social groups (e.g., OECD, 2011OECD, , 2015Perez, 2013;Soete, 2013;Breznitz, 2021). Instead, at least as far as the experience in liberal market economies is concerned, the pressing question is that of 'innovation for inclusive growth' (e.g., Martin, 2016;Schot and Steinmueller, 2018;Lee, 2019: What kinds of innovation policies need to be in place to ensure that innovation-driven growth is much more inclusive than hitherto? Unfortunately, in this essential question, the existing research on innovation and inequality remains emphatically mute. Other than the main policy implications derived from the SBTC account -i.e., addressing skilled labour shortages could reduce skill premiums and incentivize firms, marginalized and low-skilled employees to invest in education and training (Acemoglu, 2002;Goos, 2018) -the existing research on innovation and inequality appears to be largely policy-irrelevant, despite one third of published research being sponsored by scientific and policymaking organizations. Policymakers (and research donors in general) need to stimulate policy-relevant research on innovation and inequality; for instance, by sponsoring research projects in which the underlying emphasis is on a cross-disciplinary, yet methodologically diverse, analysis geared towards unearthing active causal mechanisms (rather than registering a few statistically significant associations among variables). Funding various forms of interdisciplinary yet methodologically diverse research on innovation and inequality seems to be in the interest of knowledge creation, inclusive policy design and social cohesion. • What are the mechanisms through which innovation induces top income, including wealth, inequality (Lazonick and Mazzucato, 2013;Aghion et al., 2019)? • What strategies do innovative firms adopt to address skill shortages in the innovation process? And how do these strategies impact the (horizontal) distribution of income in innovative firms? • Are large innovative firms more unequal than small firms (Cirillo et al., 2017)?
• How does the collective nature of innovation (e.g., innovation ecosystems and (global) innovation networks) affect the distribution of income (Gray et al., 1998;Fragkandreas, 2021)? • Under what conditions does innovation policy exacerbate inequality (Cozzens et al., 2002)? • (How) Does the sectoral mode of innovation (e.g., science-based sectors, scaleintensive sectors etc.) affect the distribution of risks and rewards in the innovation process (Pavitt, 1984;Lazonick and Mazzucato, 2013)? • Does innovation embed an unequal distribution of risks and rewards (Lazonick and Mazzucato, 2013)? • Which (organizational) actor(s) take(s) the lion's share of risks in the innovation process? And who does capture the rewards? • Are some types of innovation (e.g., product innovation) more inequality-prone than others (e.g., process innovation and organizational innovation) (Angelini et al., 2009;Bogliacino, 2009)? • How do a host of innovation-related factors combine to form causal mechanisms of inequality?
Causal scenario IIinequality stimulates innovation • (How) Does inequality benefit the nature, direction and success of innovative activity (Yanadori and Cui, 2013)? • In what ways does inequality motivate (marginalized) actors to innovate or participate in the innovation process (Xavier-Oliveira et al., 2015)?

Limitations
As is the case with every study, this review could not escape the rule of limitations. By using the scholarly database with the most entries (Scopus), the analysis may have, unintentionally, overlooked a few studies which are not included in this database. Similarly, because of its epistemological aims, methodological criteria and the sheer number of papers under review (166), the paper did not consider conceptual and grey literature (e.g., books, book chapters and policy reports). In addition, the review process made no extensive use of advanced bibliometric methods. This was because a bibliographical coupling and co-citation analysis, which was conducted in the early phases of the review process (albeit not reported in this paper), added very little that was new to the analysis. In fact, it illustrated that, if uncritically applied, an ostensibly neutral method exhibits a systematic bias towards 'the skewed few' (Macdonald and Kam, 2011), namely mainstream economic research on innovation and inequality. Despite this, future reviews could make use of bibliometric tools as one of the means by which to assess the extent to which a narrow mono-disciplinary perspective prevails in the more recent (i.e., post-2020) research. This type of analysis can be performed on policy papers and reports. This could help us determine whether policy documents favour certain disciplinary discourses and research streams. These are a few questions that future reviews on innovation and inequality may consider, among several other issues.

2017
Innovation and within-firm wage inequalities: empirical evidence from major European countries

1990
High technology, work, and inequality in southern labor markets

2001
Transformations of hightech labor markets and socioeconomic inequalities

Sociological Focus
Comin, D., Mestieri, M. 2018 If technology has arrived everywhere, why has income diverged?

American Economic
Journal: Macroeconomics Commander, S., Kollo, J. 2008 The changing demand for skills: evidence from the transition

Economics of Transition
Consoli, D., Vona, F., Saarivirta, T. 2013 Analysis of the graduate labour market in Finland: spatial agglomeration and skill-job match The evolution of wage gaps between STEM and non-STEM graduates in a technological following economy

2011
The impact of agricultural technology adoption on income inequality in rural China: evidence from southern Yunnan Province

2009
After deindustrialization: uneven growth and economic inequality in 'postindustrial' Chicago

2015
Is there less gender inequality in the service sector?
The gender wage-gap in knowledge-intensive services

2009
Revisiting the German wage structure Quarterly Journal of Economics Echeverri-Carroll, E., Ayala, S.

2009
Wage differentials and the spatial concentration of high-technology industries

Papers in Regional Science
Echeverri-Carroll, E., Oden, M., Gibson, D., Johnston, E. 2018 Unintended consequences on gender diversity of high-tech growth and labor market polarization

Research Policy
Engelmann, S. 2014 International trade, technological change and wage inequality in the UK economy Empirica Englehardt, S. 2009 The evolution of skill-biased effects on American wages in the 1980s and 1990s

2013
Increased sorting and wage inequality in the Czech Republic: new evidence using linked employeremployee dataset

2003
Technology, trade, and wage inequality in Mexico before and after NAFTA Journal of Development Economics Faggio, G., Salvanes, K., van Reenen, J.

2010
The evolution of inequality in productivity and wages: panel data evidence Industrial and Corporate Change Falkinger, J., Zweimüller, J.

1997
The impact of income inequality on product diversity and economic growth Technological Forecasting and Social Change Thapa, G., Otsuka, K., Barker, R. 1992 Effect of modern rice varieties and irrigation on household income distribution in Nepalese villages

2018
Taking the sector seriously: data, developments, and drivers of intrasectoral earnings inequality

Social Indicators Research
Toh, R., Tat, H. 2012 Trade liberalization, labor demand shifts and earnings inequality in Singapore Review of Urban and Regional Development Studies Torres, N., Afonso, Ó., Soares, I.

2017
Manufacturing skill-biased wage inequality, natural resources and institutions