Entrepreneurial orientation and firm performance: an updated meta-analysis

2.1 Entrepreneurial orientation and performance

According to Miller (2011), the performance implications of EO are the largest stream of research within the field of entrepreneurship. The predominant argument is that EO plays an important role in enhancing a firm's performance (Rauch et al., 2009). Li, Liu, and Zhao (2006) found that EO motivates companies to aggressively launch product innovations, exploring opportunities and favoring new product development activities. Wong (2014) noted that EO contributes to the success of new products by allowing firms to identify and proactively take advantage of new business opportunities.

Although these theoretical considerations delineate the positive effects of EO, it must be pointed that the literature also mentions negative effects that could result from EO. Naldi, Nordqvist, Sjöberg, and Wiklund (2007) revealed the negative implications to performance caused by the risk-taking dimension of EO. Such dimension shows a negative effect on financial performance in family organizations (Naldi et al., 2007). Yet, Renko et al. (2009) found that the EO construct is not a significant predictor of product innovation. These findings revealed that EO is not necessarily beneficial and viable (Renko et al., 2009).

Regardless the contradictories results, the main stream in literature sustains that the relationship between EO and organizational performance tend to be positive (Li et al., 2006; Wong, 2014). Thus:

2.2 Mediators of the entrepreneurial orientation-performance relationship

2.3 Methodological moderators of the entrepreneurial orientation-performance relationship

3.1 Study identification process

To ensure the representativeness of the empirical studies selected, a search was first performed in the electronic databases of ABI/INFORM Global (ProQuest), Electronic Journals (EBSCO), Emerald Journals, JSTOR, ScienceDirect, SCOPUS and Web of Science. Previous meta-analytic studies (Evanschitzky et al., 2012; Rubera & Kirca, 2012) searched some of these databases, but we investigated all the above-mentioned in order to identify and comprehend the most relevant studies. The keywords examined were “entrepreneurial orientation”, “entrepreneurial proclivity”, “entrepreneurial posture”, “entrepreneurial disposition” and “entrepreneurial intensity”. These terms, chosen as variations of the EO expressions commonly used in literature, were searched in the titles, abstracts and keywords of papers published from 1983 to 2014.

In order to supplement the electronic search, we conducted an issue-by-issue search in the most important journals, according to the ISI Web of Knowledge (2014) – Journal Citation Reports, in management, strategy, innovation, marketing and entrepreneurship: Journal of Management, Academy of Management Journal, Administrative Science Quarterly, Journal of Marketing, Strategic Management Journal, Organization Science, Journal of International Business Studies, Journal of Business Venturing, Journal of the Academy of Marketing Science, Journal of Marketing Research, Entrepreneurship Theory and Practice, Industrial Marketing Management, Management Science, International Journal of Research in Marketing, Journal of Product Innovation Management, Journal of Business Research and Journal of Small Business Management.

Next, we examined references and citations from relevant publications to locate additional studies (Evanschitzky et al., 2012; Rauch et al., 2009). To obtain unpublished work, we used Google Scholar and databases of theses and dissertations (e.g. The DART-Europe E-theses Portal, ProQuest Dissertations and Theses, Open Access Theses and Dissertations – OATD) (Matos, 2009; Moher, Liberati, Tetzlaff, & Altman, 2009).

After that, following the common approach of previous meta-analytic reviews to evaluate the appropriateness of each study identified (Rosenbusch, Brinckmann, & Bausch, 2011; Rubera & Kirca, 2012), we developed a few guiding rules to determine the studies that would be retained for the meta-analysis. The studies needed to:

• address the EO-performance relationship as a major topic of investigation;

• measure both EO and performance at the organizational level; and

• provide the Pearson's correlation coefficient (or variants that can be converted) for the EO-performance relationship.

Some potentially relevant manuscripts were not included in the meta-analysis because they measured EO and/or performance at another level of analysis (ten studies); they did not report the statistics needed to calculate the effect size (six studies); they investigated specific relationships that could not be integrated with other studies (five studies); and, their results were based on samples used in other publications from the data set (six studies). In this regard, it was decided to use the sample only once in the calculations to avoid excessive representation of specific samples (Rosenbusch et al., 2011).

These procedures resulted in 78 studies, with 80 independent samples and 149 effect sizes, with 137 of these for the EO-organizational performance relationship and 12 for the relationship between EO and the mediators.

3.2 Coding and operationalization of variables

A coding scheme was prepared with the information to be extracted from each paper. To facilitate the analysis, the following data was coded: author(s), title, journal, publication date, country and industry of the sample, sample size, descriptions of the respondents, scales and dimensions used in the constructs from the model and statistical information of the relationships studied.

About the EO measurement, it is worth noting that most articles operationalized the construct with the Miller/Covin and Slevin (1989) scale (63 studies). The sub-dimensions most commonly used were: innovativeness, risk-taking and proactiveness (60 studies); innovativeness, risk-taking, proactiveness, competitive aggressiveness, and autonomy (8 studies); risk-taking and proactiveness (2 studies); and, other variations (8 studies) (Calantone et al., 2010; Saeed et al., 2014).

The overall profile of the 78 studies selected according to their year of publication can be seen in Figure 2. In fact, it is possible to infer that there is a greater concentration of studies from 2013 (16 studies) and 2014 (9 studies). This demonstrates a predominance of work conducted on the theme of EO and organizational performance in a more recent period, reflecting the growing importance of the topic for researchers.

Regarding the country factor, the database shows that the country most frequently represented in the studies was the USA (20.0 per cent). Subsequently, approximately 12.5 per cent of the analyzed samples used China as background. The UK (6.3 per cent), Holland (5.0 per cent) and Sweden (5.0 per cent) were also configurations observed in a representative volume of cases. Considering the classifications of Western vs Asian countries, a concentration of 75.0 per cent was found in Western cultures.

3.3 Meta-analytic procedures

First, the statistics from the studies were converted into r correlation coefficients. 16 studies, representing 28 effect sizes (out of a total of 78 and 149 respectively), did not provide the metrics, producing standardized regression (beta) coefficients. Then, following Peterson and Brown’s (2005) suggestions, the correlations were estimated from the beta coefficients, using the formula: r = 0.98β + 0.05γ, where γ is a variable equal to 1 when β is non-negative and 0 when β is negative. No significant difference was found between studies that had correlation coefficients and those where the coefficients were derived from the beta (mean r = 0.249; mean converted r = 0.288; F = 0.976; p > 0.10).

Subsequently, the correlations were corrected for measurement error by dividing the correlation coefficients by the product of the square root of the reliabilities of the two constructs (e.g. EO and organizational performance) (Hunter & Schmidt, 2004). When a study did not indicate its reliability, we computed the average reliability for that construct across the sample and used it for correction purposes (Kirca et al., 2005; Saeed et al., 2014).

With Comprehensive Meta-Analysis®, the corrected correlations were transformed into Fisher's z-coefficients (Rubera & Kirca, 2012). To show the direct effect and confidence intervals results, for example, these Fisher's z-estimates were reconverted into revised correlation coefficients, as proposed by Borenstein, Hedges, Higgins, and Rothstein (2009). The correlations were then weighted by the sample size of each study for correction of sampling error (Rosenbusch et al., 2011).

A 95 per cent confidence interval was calculated to examine the significance of the relationship among the variables (the interval cannot include zero) (Borenstein et al., 2009; Lipsey & Wilson, 2001). Once the significance was determined, the publication bias was evaluated by calculating the fail-safe number (FSN), to indicate the number of studies with non-significant results needed to reduce the cumulative effect size to a level of nonsignificance (Lipsey & Wilson, 2001).

Homogeneity of the effect size distribution was tested by the Q statistics (Borenstein et al., 2009; Lipsey & Wilson, 2001). If there was a result that contradicted the null hypothesis of homogeneity of the relationship (e.g. EO and organizational performance), the correlations would be considered heterogeneous, and this variation could be attributed to potential moderators (Hunter & Schmidt, 2004).

To verify the existence of mediators, a multivariate analysis was conducted with structural equations modeling (Hair, Black, Babin, Anderson, & Tatham, 2009). To examine possible methodological moderations, a multiple regression model was developed (Card, 2012).

We performed the meta-analysis with the random-effects perspective. This approach, unlike the fixed-effects model, does not assume that the studies included in the meta-analysis are identical, i.e. present the same effect size (Borenstein et al., 2009). Therefore, the random-effects approach for calculating the mean values is more reasonable, because it produces more realistic estimates and indicates the real variability of the effect sizes between studies (Borenstein et al., 2009; Saeed et al., 2014).

4.1 Entrepreneurial orientation-performance relationship

Table I summarizes the meta-analytical results for the relationship between EO and organizational performance.

A total of 137 effect sizes were considered, taken from 78 studies, with an accumulated sample of 19,514 participants. The correlations recovered ranged from −0.330 to 0.690, with a mean of 0.240. The results revealed a positive association between EO and organizational performance (r = 0.299), which is consistent with the findings of Rauch et al. (2009). Table I also shows that there is a significant relationship among the constructs, since the 95 per cent confidence interval did not include zero.

The Q statistic produced a significant result (Q = 1,592.06; df = 136; p < 0.001). It indicates a heterogeneous distribution of effect sizes and suggests that this variability may be due to the existence of moderating variables. The fail-safe number was 684. Consequently, new or unpublished studies not included in the meta-analysis did not threaten the validity of the findings (Lipsey & Wilson, 2001). Therefore, H1 is supported.

4.2 Mediation analysis

Following Iacobucci (2010), structural equations modeling was used to test the mediations. First, the model was estimated to verify the mediation of learning orientation. Then, a new analysis with innovativeness was processed. It was decided to assess the two mediating factors separately due to insufficient studies using the constructs at the same time. To address these mediations in AMOS®, a correlation matrix was developed and used as input (Kirca et al., 2005).

Table II shows the correlation matrix designed to investigate the direct and indirect effects (through learning orientation) of EO on organizational performance. To generate the matrix, the mean correlations among constructs were calculated, adjusted by sample size for each pair of factors of the model (Viswesvaran & Ones, 1995).

The model was based on the Maximum Likelihood approach, with the use of bootstrapping (Byrne, 2010). Bootstrapping enables the correction of problems associated with non-normal data, and may increase the statistical power for identification of mediation effects (Wood, Goodman, Beckmann, & Cook, 2008). The results of the mediation of learning orientation are shown in Figure 3.

The results reveal a direct, positive and significant effect of EO on learning orientation (β = 0.55, p < 0.01) and of learning orientation on performance (β = 0.29, p < 0.01). As predicted, there is a direct and positive impact of EO on performance (β = 0.33, p < 0.01). In addition, there is a significant indirect effect of EO on organizational performance through learning orientation (β = 0.16, p < 0.01). These findings provide support for H2.

Based on the assumption that organizations oriented toward entrepreneurship are more inclined to encourage a fertile learning atmosphere, as pointed out by Hakala (2013), it was expected that this environment could have beneficial effects on performance. The EO-performance link was also recapitulated herein, which corroborates the partial mediating of learning orientation in this relationship.

Next, we created the correlation matrix summarized in Table III to examine the possible mediation of innovativeness.

The results obtained are shown in Figure 4. The analysis reveals that the effects between EO and innovativeness (β = 0.55, p < 0.01), innovativeness and organizational performance (β = 0.07, p < 0.01) and EO and performance (β = 0.35, p < 0.01) are direct, positive and significant. The indirect effect of EO on performance through innovativeness also had an adequate level of significance (β = 0.04, p < 0.01). Although it is worth noting the lower magnitude of this effect.

Hence, H3 is supported. EO, through proactiveness and risk-taking, positively influences innovativeness that, subsequently, has a positive impact on the development of advantages that contribute to organizational performance (Rhee et al., 2010). Here, the partial mediation of innovativeness can be seen, which coincides with Hult et al. (2004).

4.3 Moderation analysis

To examine the moderating effects of the sample and measurement characteristics of the studies, a multiple regression model was used. The moderators were coded and added into the analysis as independent variables. The dependent variable was the z-score of the reliability-corrected correlations between EO and organizational performance. The regression model was processed with 100 effects from 63 studies, which provided information on all the moderating and control variables (year of publication and journal scope). Table IV reports the results.

The model was significant (F_(8,91) = 6.067, p < 0.001) and the methodological moderators are responsible for 34.8 per cent of the variation in the correlations between EO and performance. The regression model is free of multicollinearity, since the maximum value of the variance inflation factor (VIF) was 2.038, below the limit recommended by Hair et al. (2009).

The results demonstrate that the strength of the EO-performance relationship does not vary between manufacturing and service firms, not supporting H4 (β = −0.082, p > 0.1). This discovery has implications for the literature on strategic orientation because it goes against the findings of Kirca et al. (2005), in which industry context (i.e. manufacturing and service) is relevant when investigating the relationship between market orientation and performance.

H5, which predicted that the impact of EO on performance would be stronger in a Western country, was not supported (β = 0.052, p > 0.1). Based on the Hofstede (1983) classification, it is reasonable to suppose that the magnitude of the EO-performance relationship can, in fact, be separately influenced by each dimension of the national culture (power distance, uncertainty avoidance, individualism and masculinity).

Likewise, the use of objective vs subjective performance measures did not affect the relationship between EO and organizational performance (β = 0.013, p > 0.1). Thus, H6 was also not supported. This result corroborates previous strategic management papers that found a positive and significant correlation between objective and subjective measures of business performance (Perin & Sampaio, 1999; Venkatraman & Ramanujam, 1987).

Two methodological moderators, consistent with our expectations, demonstrated significance. It was observed that the relationship between EO and performance is stronger for multi-item performance measures (β = 0.248, p < 0.05). This result provides support for H7. It also strengthens Murphy et al. (1996) statement that measuring performance through more than one item is more appropriate.

Additionally, the effect of EO on revenue-based performance measures is stronger than those based on cost, which supports H8 (β = 0.221, p <0.05). This finding confirms the conclusions by Covin et al. (2006) that EO is a construct focused on the growth of the company. Hence, it is more adequate using indicators that reflect this evolution, such as sales growth and market share.

The results also reveal that the EO-performance relationship is not influenced by the EO scale (β = −0.038, p > 0.1). Therefore, we found no support for H9. This is consistent with the meta-analysis by Rauch et al. (2009), which reported that the Covin and Slevin (1989) scale produced a similar correlation to other EO scales for EO-performance relationship.

About the control variables, the model showed that only year of publication (β = 0.290, p < 0.01) significantly moderates the relationship between EO and performance. It was identified that more recent publications present stronger effects.

In conclusion, the analysis from the multiple regression model demonstrate that the variance in the EO-organizational performance relationship can be partially attributed to the methodological differences associated with certain performance measurement characteristics (number of items and revenue or cost-based). Table V includes an overview of the findings.

In summary, the EO-performance relationship was supported (H1). The partial mediations of learning orientation (H2) and innovativeness (H3) were also empirically supported. Importantly, the moderation analysis supported the effect of multi-item performance measures (H7) and revenue-based performance measures (H8).

5.1 Limitations and further research directions

As with all meta-analytic reviews, a few limitations must be addressed. First, we could not include in the database all studies and constructs in the EO literature, especially because of the determined inclusion criteria. This paper attempted to avoid a publication bias. Unfortunately, we had difficulties in accessing the gray literature, even without restricting the identification process to peer-reviewed journal articles.

Our research provides some directions for future research. It might be productive to assess the individual effects of the Hofstede’s (1983) dimensions of national culture (power distance, uncertainty avoidance, individualism, and masculinity) in the EO-performance relationship, to determine if the results vary based on the distinct factors. While uncertainty avoidance and individualism may influence the magnitude of the EO-performance link in one way, power distance and masculinity may perform another kind of impact (Efrat, 2014).

In conclusion, the article contributes to literature by demonstrating the importance of EO to organizational performance with a meta-analysis, reporting the partially mediating variables in this relationship, and seeking to explain the observed inconsistencies, also examining methodological moderating variables. Furthermore, the present research may instigate academics and practitioners to advance the knowledge about the topic, by producing new theoretical and empirical insights.