The Effect of Workforce Birthplace Diversity on Firms: When do Migrants increase Firms’ Performance?

Growing migration between countries and the sustained trend of globalization are changing business dynamics and creating conditions for increased workforce birthplace diversity within rms. However, few studies investigate the relationships between workforce birthplace diversity and rm performance. We address this, and also study how the impact of workplace birthplace diversity on rm performance is moderated by characteristics of the rms (rm size). We nd that rm performance increases when workforce birthplace diversity increases. While larger rms perform better, smaller rms can make better use of birthplace diversity’s positive impact on rm performance. We analyzed a panel of 33,258 Italian rms operating in the agriculture sector between 2012 and 2017. Theoretical implications of our results are discussed, and further research is recommended to investigate appropriate internal mechanisms to enable rms to take advantage of workforce birthplace diversity. endogeneity, the results are still robust, and a culturally diverse workforce plays a major role in explaining the performance of rms. The ndings also show that the presence of contextual factors, such as larger rm size, is signicantly related to rm performance. Additionally, when assuming persistence of Sales over time, the results did not change. Even considering the past values for rm performance, the role of enclaves and controlling for state dependence (Heckman, 1981), our main hypothesis was supported.


Introduction
The international trend of a sustained level of immigration and globalization of rms has created conditions for growing workforce diversity within rms (Arends -Tóth and Vijver, 2003). Organizations are increasingly attempting to manage these changes in the labor market, because workforce birthplace diversity can heavily affect internal rms' dynamics (Milliken and Martins, 1996) and pose the problem on how successful coordination among diverse workforce can be promoted to boost organizational performance (Lee and Kim, 2020). Indeed, it has been shown that workforce diversity can lead to internal con ict and interpersonal communication issues (Lazear, 1999).
A limited number of studies has analyzed the relationship between rm performance and different types of diversity, such as race and gender (Richard, 2000); however, there is a paucity of analysis on this relationship with other types of workforce diversity, such as birthplace diversity. Birthplace diversity is identi ed by the degree to which employees in the organization vary by area or country of origin (Alesina et al., 2016). Place of origin is an important determinant of performance for the knowledge, perspectives and social network ties people from diverse countries can bring (Lissoni, 2018). So, birthplace diversity may be even more salient and in uential than other demographics (Hambrick et al., 1998). Despite the importance of birthplace and the rapid diffusion of diversity of place of origin within organizational workforces, there is little research on the impact of workforce birthplace diversity on organizational performance (Nielsen and Nielsen, 2013). Studies have shown that birthplace diversity can positively affect performance at multiple levels of analysis: countries (Alesina et al., 2016;Bove and Elia, 2017;Ortega and Peri, 2014), regions (Trax et al., 2015), metropolitan areas and states (Ager and Brückner, 2013;Docquier et al., 2020;Ottaviano and Peri, 2006). At rm level, some authors have considered the degree of diversity of country of origin of team members within teams (Rosenauer et al, 2016), and found that diversity is a positive factor in rm performance (Iranzo et al., 2008). However, independent of the level of analysis used to evaluate the effect of birthplace diversity on performance, research studies have overlooked the in uence of environment in creating conditions for exploiting the potential of complementarity in individual skills and knowledge (Janssens and Zanoni, 2005;Siebers, 2009). In line with Janssens and Zanoni (2005), who suggested that more studies needed to go beyond the analysis of the effect of one socio-demographic characteristic on a particular outcome variable, we assume that the relevance of birthplace diversity is dependent on the organizational context and analyzes how birthplace diversity enacts its effect in speci c organizational contexts. Our study focuses on the impact of birthplace diversity of the workforce on rm performance, expressed in terms of increase in sales, and how this impact is moderated by organizational characteristics, such as rm size. We gathered data on immigrated employees present within rms to construct the measure of birthplace diversity and investigate how it relates to the measure of rm performance. Our regressions use the inverse of the Her ndahl-Hirschman index of fractionalization (a measure of the likelihood that two randomly drawn immigrants are from different countries). To test our hypotheses, we used a panel data analysis, with a data set constructed from the Farm Accountancy Data Network (FADN), an Italian sample of 33,258 rms, observed from 2012 to 2017. The Italian agriculture sector is an excellent case study for several reasons. As with many European countries, Italy has undergone major changes in the composition of its population and has one of the highest immigration in ows in Europe (Frontex, 2018). The agriculture sector is generally characterized by cyclical and seasonal labor demand. Immigrant workers, a highly mobile workforce that is more productive than their home-grown counterparts (Baldoni et al., 2017), are a major force in agriculture, allowing seasonal labor demand to be met. The key role of immigrants in agriculture in providing expertise to achieve increases in productivity in the rural economy has been con rmed by other research studies (Kasimis and Papadopoulos 2005). This makes the Italian agricultural sector and immigrants working in rms particularly suited to our purpose. We found a positive impact of birthplace diversity of workers on rm performance. However, the positive relationship is reduced as the rm size rises, indicating that larger rms are not able to fully exploit the positive effects of workforce birthplace diversity. This result was also seen when we controlled for potential endogeneity, which may arise because of omitted variables and reverse causality. We suggest that this result may be because large rms are less able to create an internal environment that encourages informal communication and incentivizes subgroup creation helping to diffuse more in-depth knowledge helping to overcome prejudice and con icts among diverse workers. This study offers different theoretical contributions. First, it examines the relationship between birthplace diversity at the level of the organizational workforce and rm performance, helping to advance knowledge of the impact of rm-level workforce diversity on rm performance. Second, our study provides a theoretical framework underlying the concept that the bene ts of workforce birthplace diversity are contingent on the context. This article rst reviews the literature on the implications of workforce birthplace diversity on rm performance. It then introduces the hypotheses and methodology and explores the relationship between workforce birthplace diversity and rm performance, in conjunction with selected organizational characteristics of rms (size of rms). Finally, we set out our results and propose new research avenues.

Theoretical background
In line with most academic research (Alesina et al., 2016;Ager and Bruckner, 2013;Böheim et al., 2012;Docquier et al., 2020), the term birthplace diversity describes heterogeneity of individual countries of origin. Past studies have mainly used birthplace diversity to identify the dimension of intrapopulation diversity emerging from the heterogeneity of birthplaces in geographical regions (Trax et al., 2015), metropolitan areas and states (Ager and Brückner, 2013;Docquier et al., 2020;Ottaviano and Peri, 2006). At the rm level, diversity of nationality has been more widely used to mean heterogeneity in individuals' national background (Hambrick et al., 1998) rather than country of origin. Diversity studies have tended to analyze the effect of a diverse set of differences in groups, especially management teams (Nielsen and Nielsen, 2013), on group dynamics and performance. Indeed, research on diversity has devoted attention to examining the impact of the heterogeneity of identities on groups' processes and results (see Milliken and Martins, 1996;van Knippenberg and Schippers, 2007). Research on nationality diversity has considered national culture as an important factor that explains and limits group members' behavior either by collectively programming their minds (Hofstede, 1980) or by imposing the way in which a group of people solve problems and that affects groups' patterns of interaction (see Bochner and Hesketh, 1994;Cox, 1991;Maznevski and Chudoba, 2000). Few studies have considered the effect of diversity at the level of organizational workforce. There is, indeed, very little conceptual support and few empirical results addressing the potentially positive effects of workforce-level diversity on organizational results (Joshi et al., 2011). Richard (2000) and Richard et al (2007) are relevant exemptions that started to theorize and verify the in uence of racial diversity on rm performance, elevating the analysis of the impact of diversity from group to organizational level.

Birthplace diversity and organizational performance
Research on the effects of individual birthplace diversity on regions, countries and states has con rmed a positive relationship between birthplace diversity and economic growth (Audretsch et al., 2010;Ottaviano and Peri, 2006). These studies have tended to assume that individuals born in different countries may have diverse productive skills, because of their exposure to diverse life experiences, education and value systems, and will have developed dissimilar perspectives, enabling them to perceive and address problems in a different way (Alesina et al. 2016). This indication is not found in rm-level studies, where the price of heterogeneity may overcome its advantages (Coad and Timmermans, 2014;Østergaard et al., 2011;). For instance, empirical research studies on workforce ethnic diversity have reported unclear links to rm performance, as a consequence of the complexity of the phenomenon (Herring, 2009;Williams and O'Reilly, 1998) and the relevance of the context. More in general, past studies on workforce diversity have reported both positive and negative performance effects for rms.
They have assumed that a functionally diverse workforce, especially in the ways people perceive and solve problems, positively affects rm performance (Williams and O'Reilly, 1998). Indeed, when there is a problem or a challenge to solve, a varied workforce will produce a larger set of possible solutions and be able to choose the most appropriate one. Other authors, such as Hong and Page (2001), have stated that diverse populations are more likely to generate optimal solutions to di cult problems, despite the fact that each individual agent may not be able to locate a good solution if acting alone.
A functionally diverse work environment will also be able to create conditions for cross-pollinating to create new solutions (Aiken and Hage 1971). Using the same lines of reasoning (but indirectly referring to measures of diversity), Solheim and Fitjar (2016) showed that Norwegian organizations with a more diverse workforce and more foreign workers tend to innovate and start operating internationally more than less diverse rms Individuals born in one country but living in another may be considered as rooted in two cultures, and therefore able to act as between-group brokers (Ottaviano and Peri, 2006). They may have competences, skills, and abilities accumulated from the countries where they were born (Ottaviano and Peri, 2006), and also have tacit knowledge from their host country (Lissoni, 2018). Accordingly, to justify the impact of immigrants at the workforce level, Nathan (2014) described the effect of immigrants as improved labor market matching and facilitation of export, international businesses and serving diverse markets (see also Cox, 2001). Studies have also argued that people born in other countries can use their own networks as an additional source, so that greater heterogeneity can increase performance, either by alleviating informal trade barriers (Rauch, 2001) or by building bridges between rms and overseas partners. Möhlmann and Bakens (2015) con rmed this view by nding helpful evidence demonstrating that diversity heavily supports exporting rms.
Some studies have also linked diversity to a negative impact on performance Nkomo, 1991, Lazear, 1999).
Speci cally, diversity can increase interpersonal con icts and communication issues (McMillan et al., 2003;Sacco and Smith, 2003) and, in turn, lead to staff turnover and absenteeism (Tajfel and Turner, 1985). Such phenomena could impact upon the e ciency and performance of rms (Smulowitz et al., 2019). Along this line, Milliken and Martins (1996) described diversity as a "double-edged sword". However, birthplace diversity at the level of workforce may lead to different dynamics. At level of rm workforce, groups are easily informally formed, often based on shared social networks (such as similar views, physical proximity, or simply those who are familiar). Formally, establishing cross-functional teams, or groups containing members from different functions or parts of the organization can improve the range of information available to the group (Northcraft et al., 1995). At the level of workforce, birthplace diversity turns to affect group composition and therefore their performances. Diversity can produce different effects on performance contingent to the type of diversity analyzed (Brixy et al., 2020), deeplevel (attitudinal) or surface level (demographic) diversity. Deep level diversity can potentially have a positive impact on performance relying on processes of information elaboration (e.g., Kearney et al., 2009). Indeed, when people interact, stereotypes tend to be substituted by a more in-dept knowledge of each other as individuals, which can contribute to reduce prejudice and con ict and lead to greater group cohesiveness (Amir, 1976). Additionally, when tasks demand creativity, innovation, and high-quality decision-making, the cross-fertilization of perspectives, enabled by the diverse birthplace diversity of workforce, may unravel the propitious effects of diversity through the discovery of positive synergies among group members (Kearney et al., 2009). In line with Van Knippenberg et al. (2013), we assume that birthplace diversity bringing added value in the form of nonredundant experience, knowledge, perspectives, and social network ties help to reduce the negative effect of surface level (demographic) diversity. Therefore, despite birthplace diversity is often viewed as surface level diversity, at the rm level it may also improve performance through information elaboration processes (Kearney et al., 2009), derived from the provision of immigrants' diverse knowledge.
Consistently with the idea that the relationship between birthplace diversity and rms' performance is driven by complementarity of skills and knowledge, we thus assume that the social costs related to the birthplace diversity at workforce level are less than the information bene ts that can be derived from it. The possible informational bene ts may explain why diversity induced by the presence of immigrants, as seen above, are always reported as positively affecting productivity.
We therefore formulated the hypothesis: Hypothesis 1: Workforce birthplace diversity is associated with positive performance of rms.
The moderating role of organizational characteristics: rm size Past studies have emphasized the role of the context on the diversity-outcome relationship (Audretsch et al., 2010;2019;Nielsen and Nielsen, 2013). However, there is little research on the in uence of organizational-level factors on the relationship between organizational workforce diversity and rm performance (Homan et al., 2015;Stahl et al., 2010). At team level, studies have begun theorizing the mechanisms under which diversity undermines or helps the functioning of diverse groups. These studies have considered the organizational culture , time (Chatman and Flynn, 2001;Harrison et al., 1998), the type of task (Jehn et al., 1999) and the relevance of task interdependence (Rosenauer et al., 2016) as potential moderators of the relationship between diversity and speci c group outcomes. Other contextual variables moderating the relationship between diversity and group performance have also been highlighted (Bell, 2007), but not analyzed across studies (Jackson et al., 2003).
Extending the logic of the moderation effect of organizational context on the diversity-outcome relationship to the organizational level, we considered the role of rm size in shaping the meaning of birthplace diversity in the workplace. We suggest that this rm-level characteristic may set speci c constraints and opportunities in the workplace and may therefore in uence the effects of workforce diversity on rm performance.
Past studies indicate that intense cooperation among workers, favored by conditions of task interdependence and high exposure to others, may contribute to the formation of a superordinate group identity (Richard et al., 2007). This helps to overcome preconceptions and engage workers in shared goals (Dovidio and Gaertner, 2000;Pettigrew, 1998). Other scholars have underlined that where demographically dissimilar individuals provide identifying information, others' idiosyncratic knowledge about them diminishes the inclination to attach stereotypical categories (Miller and Brewer, 1984). Others' reliance on surface-level demographic diversity as a basis for categorization reduces and tends to shift to the social categories characterizing the work environment.
Typically, larger rms have a more intense level of production than small rms. They tend to standardize their activities and use more complex forms of coordination, involving employees who have to carry out repetitive, 'alienating' activities that do not foster the creation of social relations (Wu et al., 2008). Informal relations between top management and employees are less frequent in large rms (Karahanna and Preston, 2013). Higher levels of technological development, a greater degree of formal communication and formalization of working procedures in larger rms may also result in a perception by minorities in the workforce that they are not supported (Stahl et al., 2010). Conversely, in small rms, access to resources is mainly through informal, private social networks, and not through formal channels (Gibson and Cassar, 2005;Peng, 2003). Indeed, in small rms, informal relationships emerge more easily. Small rms, by their very nature, are also characterized by little or no hierarchical distinction of roles in the production process (Gremler and Gwinner, 2000). This may encourage the development of a positive perception by workers that they are part of a network and give them a sense of con dence in others and mutual obligations that facilitate communication (Rosenauer et al., 2016). In this type of environment, workers may voluntarily build interpersonal ties, creating a situation where social learning and a shared set of complementary skills and knowledge are more likely to diffuse.
Small rms can also more easily create a conduit for individuals to combine their diverse backgrounds and perspectives, favoring the creation of an overall stock of complementary skills and knowledge sets (Audretsch et al., 2010;Nielsen and Nielsen, 2013). This, in turns, reduces the costs typically associated with workforce diversity and limits the potential con icts among heterogenous workers (Boudreaux, 2020). With a more heterogenous workforce, therefore, a smaller rm can give employees the chance to interact with others more easily and to develop perceptions of fair treatment by the management, resulting in full integration (Ely and Thomas, 2001;McLeod and Lobel, 1992;McLeod et al., 1996;Shaw et al., 1998;Sheehan, 2014). We therefore hypothesize: Hypothesis 2: Higher levels of birthplace diversity of workforce will have the strongest positive effect on rms' performance when a rm is small rather than large.

Research Design
The purpose of this study is to examine the interrelationships among workforce birthplace diversity, rm performance and rm size. It attempts to explore the in uence of different workers' countries of origin on rm performance, measured through total sales, as well as the in uence of rm size as a relevant contextual factor in this relationship. This section describes the research setting, data collection and data analysis, and sets out the empirical model for the study, shown in Figure 1.

Research setting
Immigrant workers, who represent a highly mobile workforce (Card 2005(Card , 2001, are able to play a major role in agriculture sector, for two main reasons. First, in sectors characterized by cyclical or seasonal labor demand, as agriculture (Hanson and Bell, 2007), immigrant workers, more incline to displacement given the absence of ties in the host-country, represent a key resource that, on one hand, allow the agriculture seasonal labor demand to be met, and on the other hand, constitute the privileged recruitment area, given the declining employment of home-born citizens (Baldoni et al., 2017). Secondly, the agriculture sector, where jobs are often characterized by manual and low-skilled activities, but required experience and traditional knowledge, immigrant workers, bearers of different traditional skills, often acquired on the job in the home-country, constitute a rare and inimitable knowledge asset. Furthermore, the high mobility and the tendency to be carriers of new knowledge, in the long run may increase the spillover effects between rms. Along this line, Klocker et al. (2018) noted that culturally diverse workforce within rms play a role as "environment builders": workers from different nationalities bring relevant traditional knowledge that acts as a new resource for environmental change in the agriculture sector (Somerville and Sumption, 2009). Finally, Italian agriculture is strongly heterogeneous, with considerable geographical speci city, and may be considered a good representation of the various minorities in the workforce (Macrì et al., 2017). Therefore, agriculture, especially the Italian agriculture sector, would appear to represent a suitable sample for investigating the relationship between birthplace workforce diversity and a rm's performance.

Data collection
The analyses used to test the hypotheses in this study relied on data from an Italian sample from the Farm Accountancy Data Network (FADN). The FADN is a European network providing data on farm incomes and businesses. The FADN data set includes only professional farms that carry out market-oriented activities. The data set collects an average of 1000 data items per farm per year for European Union Member States. Its purpose is to improve knowledge of the agriculture sector, and it gives information on aspects such as land use, animal activities, input-output costs, and farm characteristics. The data set includes only farms that have had an economic size of more than 8000 Euros since 2014, measured as standard gross margin (SGM). This threshold has changed over time but tends to identify the same rm size. The FADN sample therefore contains only rms that are representative of a sub-sample of professional and commercial Italian farms (Baldoni et al., 2017). In this study, data were gathered at the provincial level over a six-year period (2012-2017). The panel includes discontinued repeated observations in time. To deal with multicollinearity in our interaction model, we aggregated and treated them as a mean centering over the above period. This procedure may also enhance the interpretability of data (Midavaine et al., 2016). To validate our initial concepts, we focused on several speci c datasets: 1) farm characteristics; 2) economic data; 3) labor cost; 4) workforce. Relevant information was extracted for each farm, such as general farm characteristics, location, size in hectares, type of farming, and age and gender of farmers. We also included summary information on the workforce, such as total hours worked per farm and total hours worked by immigrants per farm. Important information was included about the economic characteristics of farms, such as total farm revenue, revenue from activities complementary to agriculture, direct costs and current costs.

Dependent variable
The dependent variable, rm performance (Sales), was measured as the total sales of a rm, in Euros, over the period 2012-2017. This measure includes both business sales revenues and operating aid (EU). Total sales are a widely used measurement of rm performance (Christensen et al., 1987;Huselid, 1995). Its main advantage is that it provides a single measure that can be used to compare a rm's performance and enhance comparability with previous studies. In the second step, the logarithm of sales was calculated to deal with skewness. The logarithm of sales also re ects a rm's performance irrespective of variations in employees and capital markets. In previous studies, models have also speci ed rm performance as the logarithm of net sales per employee or net income per employee, although these are less frequently used (Huselid, 1995).

Independent variable
In line with Alesina et al. (2016), Ottaviano and Peri (2006) and Parrotta et al. (2014), we used the concept of birthplace diversity to distinguish between different workers' countries of origin. We focused on all rm employees, distinguishing between native and non-native (immigrant) workers. In line with past research (Smulowitz et al., 2019;Richard et al., 2007), we used the inverse of the Her ndahl-Hirschman index to measure workforce birthplace diversity within a rm. Our independent variable was expressed as: Birthplace Diversity where p is the proportion of workers in each category k. This index measures the degree of birthplace fractionalization within the workforce. It ranges from zero to (k − 1)/k and assumes a maximum value in the presence of maximum heterogeneity, that is, when all countries of origin are equally represented within the workforce. In this study, we considered seven groups by birthplace: native workers, workers from Eastern Europe, workers from elsewhere in Europe, workers from Africa, workers from America, workers from Russia, and workers from Asia. With seven groups, the index has a range of 0 to 0.72, where a value of 0.72 shows more diversity (all categories are equally represented within the rm). All groups were expressed as annual working units (AWU), the total hours worked per rm per annum. In this formula, 1 AWU is equal to 1800 hours worked (Baldoni et al., 2017). Considering the annual working units (AWU) allowed us to take into account the real workforce birthplace diversity within rms, given the role of immigrant workers as a highly mobile workforce in the Italian agricultural sector. Indeed, using total hours worked as a measure, enabled us to show the extent in which workers from different countries of origin, interacting in the workplace and collaborating with each other, provide the rm a stock of traditional skills and knowledge from host countries, which once assembled within rm may represent a rare asset, di cult to acquire elsewhere. The annual working units (AWU) therefore allow us to consider the degree of knowledge transmission between workers and his real effects in a within rm setting. We calculated this measure of heterogeneity for each rm for each year.

Moderating variables
One moderating variable was used, rm size. In line with previous studies, we used the total amount of land per rm, in hectares (Berry 1972;Saini, 1971). Indeed, rm size, expressed in terms of hectares, allow us to take into account the degree of territorial dispersion within farms. In the agriculture sector, where work activities are mainly carried out in the lands, the different territorial extension of rm, in terms of hectares, enables to consider the real extent of social interaction between workers, and consequently, the degree of knowledge transmission. We considered the total amount of land as the main asset of a rm in the agriculture sector.

Control variables
To control for systematic variance that is not due to our main variables, we included all variables that could be confounding factors of the relationship. We therefore controlled for Machine Opportunity Cost (MOC), or the cost of using rm machines within the production processes, in Euros. This variable is cited in many previous studies as having a major impact on rm performance (Koch and McGrath, 1996). It is a measure of both workforce turnover and machinery replacement, so may affect both rm performance and the extent of use of immigrant labor. We also controlled for Organic Firms, Young Entrepreneurs, Family Firms and Wages. Organic Firms was measured by a dummy variable with a value of 1 if the rm used organic processes. In the mainstream literature, the effects of organic processes on rm performance are generally associated with a higher degree of technology innovation (Lansink et al., 2002). This means that organic agriculture is, on average, more e cient but less productive than other (conventional) methods (Bayramoglu and Gundogmus, 2008). Young Entrepreneurs was measured by a dummy variable indicating whether the owner is under 40 years old. The presence of a younger owner is considered a major factor able to affect rm performance, although previous ndings are inconclusive. Young people are often more willing to accept environmental change, both in general and within rms (Jackson et al., 1993) and hence may employ more culturally diverse workers. Young people are also more inclined to bear risks, eliminating old patterns of behavior (Gudmundson and Hartenian, 2000). Another control variable is Family Firms, which is a dummy variable with a value of 1 for family-run rms. Its inclusion is justi ed for two reasons. First, family involvement in a business has an important impact on rm performance (Basu and Altinay, 2002;Chrisman et al., 2016). There is also a positive correlation between rm size and family rms (Lee, 2006), so the introduction of the Family Firms variable allowed us to detect the effects of the rm's size on our main relationship, net of the spurious effects of family management. Finally, we controlled for Wages as a measure of the rm's labor cost. Wages was measured by annual salary received by the worker, expressed in Euros. Immigrant workers are more likely to accept bad working conditions and low wages, so Wages could be a confounding factor, and we controlled for it to eliminate the bias associated with eventual better economic performance given by a lower labor cost at rm level. The variables used to test the hypotheses are summarized in Table 1.

Data analysis
Descriptive statistics, the basic attributes of our sample, and correlations between variables are shown in Table 2. The average land per rm was 24.6 hectares. In the sample, the mean value of workforce heterogeneity showed high levels of workforce diversity within rms (0.33). However, there is an issue of data representativeness with the FADN data: the presence of immigrants in the workforce may well be underestimated in some regions, where there is a greater presence of an irregularly employed workforce, especially for seasonal activities. Even given these data reliability issues, the analysis yields some relevant information about the presence of the immigrant workforce in Italian agriculture. More than 15% of rms were certi ed as organic and more than 43% were family rms. These ndings provide support for the use of the control variables to test the main relationship. To explain within-sample differences, we analyzed the presence of foreign workers by region and farm specialization. These results also suggest that immigrants in the workforce make a major contribution to the Italian agriculture sector, especially for seasonal activities, where they are most frequently employed. The results are shown in Table 3, which also shows the immigrant workforce by region. The regions vary considerably in the concentration of workforce from different countries of origin. Firms in northern Italy have higher levels of immigrant workers. However, these ndings do not reveal the true picture because these workers tend to be regularly employed in the north, whereas in other regions (notably in southern Italy), there is a greater presence of undeclared workers, and the importance of seasonal workers re ects the typical agricultural specializations of such regions. Before we tested our hypotheses, we examined the relationship between the incidence of workforce diversity and rm performance in Italian agriculture (Table 2). There was a correlation between Workforce Diversity and Sales (0.26), which suggested the presence of a possible relation between the presence of workers from different nationalities, in terms of AWU, and rm performance. Data also suggest that these ndings did not reveal a spurious correlation stemming from rm size. We found a highly signi cant correlation between Firm Size and Sales (0.51). The magnitude of these correlation coe cients therefore requires in-depth investigations. Formally, the relation exists and suggests the presence of a link among these variables, although correlation does not imply causation. The direction of this relationship therefore needs to be tested. We also checked for Family Firms. In line with the literature, we found a negative signi cant correlation between Family Firms and Sales (−0.35). Table 2 shows the correlations among the study variables. [1]

Empirical methodology
The estimation of the relationship between workforce birthplace diversity and rm performance in the agricultural sector could be invalidated by endogeneity problems from several causes such as omitted variables (Seifert and Valente, 2018), measurement errors and reverse causality. The presence of unobserved rm-speci c factors, because of, e.g., geographical differences, sub-sector production and the decisions made by farmers can be correlated with both rm performance and the diversity index, creating a spurious relationship between our main variables. Immigrant workers also tend to vary both across rms and over time, as a result of speci c skills and attitudes of the workers and seasonal variations in labor demand. The FADN data misrepresents the true immigrant workforce in the Italian agriculture sector. The number of workers from different countries employed in agriculture could be greatly underestimated because of the presence of undeclared workers. This provides a further problem of endogeneity because of measurement errors. To deal with these causes of endogeneity, we estimated the relationship between rm performance and workforce birthplace diversity by using the xed-effects model at rm-level. This allowed us to control for any unobservable rm characteristics that may affect both the presence of foreign workers and rm performance. We also included time xed effects, in order to consider time-varying aggregate factors, such as an increase of immigration, able to invalidate our estimations. Both measurement errors and omitted variables can be addressed by the inclusion of xed effects at rm-and time-level. Finally, our speci cations include year dummies, altimetry variables and standard errors clustered by rms to account for any other sources of bias (i.e. seasonality effects, geographical features). Previous studies also suggest that immigrants are mostly employed in large (and more productive) rms (Baldoni et al., 2017). This gave us a problem of reverse causality. To solve this problem, we used the Generalized Method of Moments (GMM) model, which can take into account possible endogeneity from reverse causality. We modeled our setting as a dynamic linear process in which we included a lagged dependent variable, to consider the past values of rm performance, controlling for state dependence (Heckman, 1981). We also included the lagged independent variable, to take into account the farmer's historical propensity to hire immigrant workers. Firms that have historically hired immigrant workers are more likely to continue to use immigrants in the production process, because of factors like a lower search process costs (the role of enclaves) (Malchow-Møller et al., 2013). Given extensive data availability, we used the FADN data set with an unbalanced sample of 33,258 rms, observed from 2012 to 2017. As described above, we measured workforce diversity through a fractionalization index, using the inverse of the Her ndahl-Hirschman index.
[1] In order to obtain a measure of how much the variances of estimated parameters are in uenced by multicollinearity, we calculated the variance in ation factors (VIF) scores. Our values are consistently below 1,4. This circumstance allow us to con rm multicollinearity is not a problem.

Results
This section rst provides the results of the regression model related to the research questions. It also describes the methodology adopted to reach our results. We then discuss the main implications of this contribution and de ne the most relevant relationships emerging from our data.

Results of panel data estimations
The empirical results of the static models are shown in Table 4. The xed effects estimates are reported in columns 1-4. The ordinary least squares estimates are shown in column 5. To deal with multicollinearity in the interaction models, we computed the mean centering of the variables in the moderator analysis in all models. In addition to the control variables, all speci cations contain year dummies, altimetry variables and standard errors clustered by rm, to avoid any other sources of bias. The dependent variable was the log of sales (Sales). We rst examined the control variables effects on rm performance (Sales) in column 1. Then we added the main variable (Birthplace Diversity) in column 2. Indeed, in order to take into account the effects coming from rm-speci c characteristics, we considered all variables that could be confounding factors of the relationship. We therefore controlled for Machine Opportunity Cost, Organic Firms, Young Entrepreneurs, Family Firms and Wages. We indeed sustained that rm performance can be in uenced by presence of unobserved rm-speci c factors, e.g., the young entrepreneur's risk inclination or eventual better economic performance given by a lower labor cost. Even adding these controls, results are in line with our hypothesis. Hypothesis 1 was therefore supported in columns 2-4, which showed that employing more heterogeneous workers may enhance levels of sales, improving rm performance. Speci cally, results from FE estimations show that an increase of birthplace diversity in a within rm setting leads to an increase of sales about 18 percentage points (column 2). Hypothesis 2 predicted that the relationship between workforce birthplace diversity and rm performance would be negatively moderated by the presence of large rm size. According to the literature, column 3 indicated that larger rm size is linked to better rm performance (Barkema and Vermeulen, 1998). However, as shown of column 4, the introduction of the moderating term, the multiplication of Firm Size and Birthplace Diversity, was negatively related to performance of rms (Sales). In particular, the estimates coe cient is around -0,25 percent. More speci cally, we nd a negative interaction effect between rm size and workforce birthplace diversity, which means that small rm turns the negative effect into a positive effect of birthplace diversity on rm performance. This implies that, when we consider a proper context (small rm rather than a larger one), an increase in birthplace diversity of workers is associated with a positive variation on rm outcome. Expressed differently, small rms are more able to create an environment that is conducive to fully exploit the bene ts deriving from a greater cultural workforce diversity. Hypothesis 2 was supported in column 4. Finally, we ran an ordinary least squares regression analysis (column 5). As shown, the main results were con rmed by the simple OLS model. However, when we did not control for endogeneity, the main coe cients were higher than the coe cients estimated with the xed effects regression (column 4), where we controlled for positive omitted variables biases. Figure 2 plots out the joint effect of birthplace workforce diversity and rm size on change in performance, considering two illustrative points of rm size distribution: small (-1 SD below the sample mean) and large (+1 SD above the sample mean). Figure 2 suggests that, as assumed in H2, small rms were more able to maximize bene ts deriving from higher levels of birthplace workforce diversity, improving rm's performance. Similarly, higher levels of birthplace diversity of workers were less strongly associated to better performance, when the rm is larger. However, gure 2 also shows that birthplace workforce diversity affects rm's performance, independently of the size of rm. Table 5 includes a number of robustness checks. Column 1 shows the baseline equation; columns 2-5 include additional control variables to test the persistence of a positive contribution of workers from different countries of origin to rm performance. Columns 2-3 include farm specializations and a dummy variable that is equal to one in case of rm's localization in lowland and zero otherwise. We consider that some typical sector characteristics (i.e. differences in productivity between crops, seasonality features) and the speci c location of rms are able to impact on our main relationship. Several studies have indeed considered the in uence of a rm's geographical location on its performance, like potential presence of geographic spillover effects (Shoobridge, 2006). Even considering these additional controls, the positive impact of a culturally diverse workforce on rm performance seems to be con rmed by high values of the statistically signi cant parameter. Column 4 controls for the presence of new investments: the additional control was signi cant and our main relationships did not change. Finally, column 5 includes the size of rms as total employees. Even though this speci cation seems to absorb the signi cance of our main relationships, birthplace diversity continued to play a major role in rm performance and all our hypotheses were supported. As Table 5 shows, our ndings are robust in all speci cations. Table 6 contains robustness checks based on GMM estimators. Estimating the main relationships through a dynamic panel data model with the Arellano-Bond estimator allowed us to consider the possible endogeneity of the workforce diversity not completely eliminated with xed effects at rm level. Column 1 repeats the baseline equation, without control variables, using the lagged dependent variable (Sales t−1; Sales t−2) and the lagged independent variable (Birthplace Diversity t−1) to take into account serial correlation over time. Column 2 repeats the previous equation, introducing the control variables. The results obtained differed slightly from those obtained with the xed-effects speci cations (Table 4). The coe cient for the independent variable (Birthplace Diversity) was positively statistically signi cant in the two columns. This suggests that, even accounting for other sources of endogeneity, the results are still robust, and a culturally diverse workforce plays a major role in explaining the performance of rms. The ndings also show that the presence of contextual factors, such as larger rm size, is signi cantly related to rm performance. Additionally, when assuming persistence of Sales over time, the results did not change. Even considering the past values for rm performance, the role of enclaves and controlling for state dependence (Heckman, 1981), our main hypothesis was supported.

Discussion
In this article, we used workforce birthplace diversity to describe the presence of heterogeneity of employees' birthplaces within the organizational workforce. In line with research from the past decade exploring a distinct type of diversity, birthplace diversity with immigrant diversity in metropolitan areas (Ottaviano and Peri, 2006), we considered the immigrant workforce employed by rms to be a good sample for workforce birthplace diversity. Past empirical studies have reported a positive relationship between urban and regional immigrant diversity and worker productivity, underlining the concept that the advantages of immigrant heterogeneity overcome the costs (Suedekum et al., 2014, Trax et al., 2015. At the workplace level, Cooke and Kemeny (2017) explored the relationship between immigrant diversity and worker productivity and found that the diversity effects on wages were stronger among workers that are heavily appointed to complex problem-solving activities. Drawing on these results, and con rming that birthplace heterogeneity gives tangible economic bene ts, our evidence supports the rst hypothesis showing a positive correlation between birthplace diversity and rm performance. We proved this relationship through an empirical analysis run on a panel data set of 33,258 Italian rms, observed from 2012 to 2017. The results were robust even after having controlled for potential endogeneity. Past studies have reported a robust correlation between workforce immigrant diversity and different measures of productivity, but no one has provided deeper insights into the relationship between workforce birthplace diversity and rm performance. At team-level, current literature has been not able to provide consistent results about the direct impacts of diversity on performance that could be generalizable across contexts (Stewart, 2006;Van Knippenberg and Schippers, 2007). Therefore, some scholars have concluded that all types of diversity can have both positive and negative effect on performances (Van Knippenberg et al., 2013) leaving the knowledge of when and how diversity impacts team performance quite fragmented with some questions unanswered (Van Knippenberg and Schippers, 2007).
Drawing on the idea that individuals construct a preliminary opinion of other people relying on detectable attributes using stereotypes that informal or formal interactions within the company work environment help to replace with deeper knowledge of the cognitive features of other individuals, we proposed that given the value added, in terms of their social ties and background knowledge, individuals originating from different countries can give , birthplace diversity at level of workforce can have positive impact on rm' performance. Indeed, we underlined that individuals originating from diverse countries add to their destination countries different skills, culture, value systems and problem-solving capabilities, which are crucial assets for companies.
Thus, we have explained the positive results of birthplace diversity on rm's performance, introducing the information-decisionmaking perspective and therefore, for this reason, we claim that birthplace diversity can be different from the other types of demographic diversity.
The empirical evidence also shows that the positive effect of birthplace diversity on rm performance depends on the organizational context, as the birthplace diversity can have stronger consequences in environment where individuals can spend more time together and interact more easily. This is consistent with past research that considered that organizational context moderated the relationship between diversity (racial) and performance (Richard, 2000;Richard et al., 2006). These studies also suggested that factors such as organizational structure, culture, and rm strategy acted as moderators of this relationship at both group and rm level. For example, Chatman and Flynn (2001) showed that the interaction between demographic characteristics and organizational culture affected different diversity outcomes. Richard et al. (2006) found that both organizational structure (managerial span of control) and organizational life cycle stage moderated the relationship between racial diversity and rm performance. Earlier, Richard (2000) suggested that business strategy (in terms of growth vs. downsizing) was also a moderating factor in this relationship.
Our results underline the relevance of the organizational context and suggest that the organizational context of human resource diversity can positively affect rm performance. In large rms, the higher coordination costs and degree of formality in the social structure regulating internal activities mean that diversity can lead to higher social costs because of the di culty of creating an environment encouraging spontaneous social ties and informal group formations.
Our study makes several contributions to the literature. First, this is the rst study of which we are aware to extend the level of analysis of birthplace diversity to the organizational level. It complements regional and urban studies that have also shown the effect of birthplace diversity on economic growth. Second, past research on the heterogeneity of employees' places of origin at rm level, with the exception of Cooke and Kemeny (2017), has focused on the effect of diversity, mostly diversity of nationality, on group processes and outcomes. This study also answered the call for more theoretical guidance and empirical studies on the valuable impact of rm-level diversity on organizational outcomes (Joshi et al., 2011).
Third, our study tested the effect of rm size on the relationship between birthplace diversity and rm performance. It therefore addresses the need for more studies about how diversity is affected by organizational and societal context (Joshi and Roh, 2009;Prasad et al., 2006;Siebers, 2009). Past studies on organizational diversity found that organization strategy moderated the effects of organizational diversity (see Dwyer et al., 2003;Richard, 2000;Richard and Shelor, 2002). Richard et al. (2004) proved that entrepreneurial orientation was a positive moderator of the nonlinear relationships between organizational diversity (as gender and racial diversity) and organizational performance. Similarly, Cunningham (2009) reported that having a proactive diversity management strategy (where diversity is encouraged, and the organizational mission statement, policies, procedures, and practices incorporate diversity) positively moderated the relationship between racial diversity and organizational performance. No previous studies included any moderator variables describing organizational context.

Limitations
Our study suggests that organizational characteristics of rms can be a critical asset in the right context and can positively affect the workforce diversity-rm performance relationship. However, conclusions from our ndings must be tempered by the awareness of several study limitations. First, we were unable to completely exclude endogeneity problems, because of measurement errors and reverse causality, and omitted variables issues. Our identi cation strategy included time-varying covariates and a full set of xed effects, but the estimation of the relationship between rm performance and workforce birthplace diversity could still suffer from endogeneity issues. To further take into account these sources of bias, we modelled the generalized method of moments (GMM) model. We are aware that, in some cases, IV estimation could lead to better results (Ager and Brückner, 2013;Ortega and Peri, 2009;Ottaviano and Peri, 2005;Trax et al., 2015), but these strategies do not seem applicable here. The limited availability of data and the poor data quality from the Italian agricultural sector prevent the use of several instrumental variables extensively used in previous studies (Card, 2005;Ottaviano and Peri, 2005). We therefore followed the approach of Parrotta et al., (2014) and used the one-step GMM estimator, because it can account for contemporaneous correlation in the errors of the simultaneous two-equation system and is robust for serial correlation or heteroskedasticity in the error terms (Wooldridge, 2009). In general, the one-step GMM estimator has a larger asymptotic variance than the two-step counterpart, but it allows for a lower loss of e ciency by simultaneously estimating the rst and the second stage conditions. We can therefore reasonably contend that the large number of xed effects, the full set of robustness checks and the GMM estimation are su cient to deal with our endogeneity issues. However, further research could consider implementing an IV estimation, to achieve a more complete understanding of drivers of diversity (Docquier et al., 2020).
Secondly, the use of a single industry might reduce the external validity of our results (Richard, 2000). However, it also allows us to reduce any bias from cross-sectoral variation, reducing endogeneity arising from unobservable confounding factors linked to differences in sectors. The use of one sector may therefore help to better identify the effects of birthplace diversity on rm performance. We are aware that our ndings need to be further validated by analyzing at least one other industry (e.g., manufacturing), to provide con rmation of the role of contingency factors (Richard et al., 2004). Finally, our study did not consider other demographic attributes of diversity of workers (e.g., gender or age), which may interact with the effects of diversity based on birthplace or nationality. This may allow us to explore the complexity of the workforce diversityperformance relationship, in a within-rm setting (Richard et al., 2004).

Conclusions
In this study, we analyzed the relationship between workforce birthplace diversity and organizational performance from a new point of view. We challenged a tacit assumption of past studies that different backgrounds will lead to variation in information, knowledge and cognitive approaches. In larger rms, workers from diverse backgrounds have more opportunity to interact. However, the higher level of coordination, and stronger presence of formal rules mean that these rms may have less opportunity to take advantage of diversity of knowledge and ways of thinking. In smaller rms, there may be more exible interactions among co-workers, with more chances to draw on the combination of backgrounds, information and knowledge.
Our study therefore provided empirical support for a theoretical position, i.e., that birthplace diversity fosters rm performances. These complements studies at regional level and in metropolitan areas. This research is important because of the move towards globalization. Future research could explore the mechanisms underlying the expansion of the pool of knowledge and information deriving from different employee backgrounds at rm level. The lack of operationalization of these topics in previous work means that our results call for additional measures to collect information about the extent to which the combination of different backgrounds leads to new knowledge. The inverse of the Her ndahl-Hirschman index that measure the degree of birthplace fractionalization within the workforce. We considered seven groups by birthplace: native workers, workers from Eastern Europe, workers from elsewhere in Europe, workers from Africa, workers from America, workers from Russia, and workers from Asia, in terms of AWU, that is the total hours worked per rm per annum; 1 AWU is equal to 1800 hours worked.
Total amount of land per rm, in terms of hectares.