A Meta-Analysis of Green Supply Chain Management Practices and Firm Performance

Abstract

Integrating green practices into supply chain management is an important issue for companies to combine both environmental responsibility and the goal of increasing profits. In this paper, we present a meta-analysis of the relationship between green supply chain management (GSCM) practices and firm performance. By using robust variance estimation, we can appropriately analyze dependent effect sizes that are common in the studies to be included in this meta-analysis. In this way, more information is extracted from the primary studies and the effects of moderator variables can be examined simultaneously. Based on 408 correlation coefficients (N = 30,568) from 134 studies, we find a significant mean positive relationship between GSCM practices and firm performance in general (r = 0.442), and in particular, for market-based, managerial, and accounting-based performance. However, the correlations vary considerably across studies. To explain this heterogeneity, we analyze the influence of different GSCM practices and organizational characteristics.

1. Introduction

Taking environmental sustainability into account is one of the most serious challenges of the 21st century. As companies in various industries contribute considerably to negative environmental impacts such as greenhouse gas emissions, toxic pollution, and tons of plastic waste, the pressure to integrate sustainable management practices within Industry is growing more and more [1]. Governmental regulations and changing consumer markets force companies and whole supply chains to take over environmental responsibility by implementing green strategies. Hence, organizations integrate sustainable practices in the whole supply chain, including production, organizational processes, and cooperation with suppliers as well as customers [2].

Meanwhile, the relationship between green supply chain management (GSCM) practices and performance has become a key topic among practitioners and researchers. Numerous studies revealed that GSCM and firm performance are positively correlated, and it has been shown that GSCM was able to strengthen an organization’s competitiveness and profit [3,4,5,6,7]. However, a wide range of correlation coefficients between GSCM practices and firm performance has been found. Thus, the moderating effects of various conditions on this relationship have to be taken into account. Applying a meta-analysis is the best method to investigate the relationship between GSCM practices and firm performance and to consider conditions that might affect the extent of this relationship.

Several meta-analyses have already investigated the relationship between GSCM practices and firm performance [3,4,5,6,7]. They uniformly revealed a positive relationship, however, the analysis of moderators provided contradictory results (see below).

In the present meta-analysis, we focus on the relationship between environmental sustainability and economic dimensions of firm performance. Compared to the previous meta-analyses, an updated dataset will be processed. Furthermore, we use meta-analytic methods based on robust variance estimation that are able to cope with dependent effect sizes in the included studies. Thus, through the obtained high amount of effect sizes, we can simultaneously include all moderator variables in one regression equation, which has not been the case in previous meta-analyses.

The remainder of this article is structured as follows: In the second section, we briefly discuss the theoretical and empirical background of the relationship between GSCM and firm performance. In this section, we also briefly present the results of previous meta-analyses. In the third section, the meta-analytic methods used here are outlined. Then, we present the results of the meta-analysis and moderator analyses. Finally, we discuss our findings and the limitations of this meta-analysis.

2. Green Supply Chain Management and Firm Performance

The concept of GSCM has been attracting attention since the early 1990s and there is an enormous publication trend on this topic since 2000 [8,9]. The amount of publications has intensified particularly since 2010 and is still exponentially increasing. The assessment of GSCM and its performance has always been a key issue in this context [1].

2.1. Green Supply Chain Management

Supply chain management (SCM) encompasses a wide range of activities throughout the life cycle of goods, from design to consumption. GSCM can be understood as the integration of environmental issues into supply chain management practices. In the literature, GSCM and sustainable supply chain management (SSCM) are often used interchangeably; however, SSCM is also sometimes used as a broader term encompassing SCM practices including social, environmental, and ethical measures [10], while GSCM activities refer only to improving environmental performance [11]. Here, we consider only the environmental practices within supply chain management that are referred to by the term GSCM.

Several classification systems for GSCM practices have been proposed [3]. Many of them contain similar categories. For the present meta-analysis, we adopt the following classification of GSCM practices based on a synthesis of different approaches in the literature [3,4,5,6,7]:

• Green supplier orientation (i.e., collaboration with suppliers, green purchasing, etc.)
• Eco-design (i.e., use of eco-design, green product innovations, etc.)
• Green production (i.e., green production with reducing waste and preventing pollution, etc.)
• Green customer orientation (i.e., collaboration with customers and logistics)

2.2. Firm Performance

There are various definitions of the concept of firm performance and its measurement [12]. Firm performance can only be as good as its success in exploiting all its abilities to achieve the organization’s goals with the allocated resources [13]. Competitiveness, effectiveness, and efficiency are three important factors that should lead to competitive advantage and finally result in a high firm performance [14]. In line with this approach, Golicic and Smith used the following classification of firm performance in their meta-analysis of the relationship between GSCM and firm performance:

• Market-based performance
• Accounting-based performance and
• Operational-based performance [3].

This classification is a commonly used approach in the research of supply chain management and business research [15] and was also chosen for this meta-analysis. Market-based performance focuses on financial indicators that represent market goals in terms of meeting customer needs and it includes market share, competitive advantage, customer loyalty, brand value, etc. Operational-based performance focuses on aspects of operational efficiency, such as costs, quality, flexibility, and speed. Accounting-based performance reflects overall profitability as expressed by return ratios, profit, and earnings [3]. To measure firm performance, in most cases subjective measures in form of questionnaires are reported. Objective measures like return ratios are rarely used.

In some meta-analyses, e.g., [4,7], two further performance dimensions were used: social performance (e.g., enhanced employee satisfaction or enhanced health and safety) and environmental performance (e.g., reduction of air emissions, reduction in used harmful and toxic materials). Since we do not consider practices that improve social performance, this category was excluded. Clearly, GSCM practices aim at improving environmental measures. However, we consider environmental performance as an independent variable, as the dependent variable we are interested in is economic performance in a narrow sense.

In addition to the type of GSM practices, moderator variables have been investigated in former meta-analyses. Here, we include the most relevant ones that were included in the previous meta-analyses [3,4,5,6,7].

First, the type of organization may influence the relationship between GSCM practices and firm performance. Organizations are faced to differing degrees with the need to introduce GSCM practices. For example, manufacturing companies such as those in the oil, chemical, or car industries are much more responsible for environmental pollution than service industries, such as banks or the hospitality industry. As a result, manufacturing companies are generally more regulated than other sectors. As a consequence, moderation effects from the type of organization are expected.

Larger companies, which have more resources but at the same time face greater pressure to adopt environmental practices, should be more able to adopt GSCM practices and achieve better economic results from them.

Countries are characterized by different laws, regulations, and cultures. These conditions can have a significant impact on GSCM practices and their relationship to firm performance. In this respect, we include the continent as a moderator variable.

Some organizations have adopted ISO certificates such as ISO 14001 or ISO 26000. These firms might be more willing to adopt GSCM practices, e.g., because ISO-certification could lead to more awareness, knowledge, and experience with environmental issues and thus, to the adoption of GSCM practices. Therefore, ISO certification will be included as another moderator variable.

The relationship between GSCM and firm performance may be closer in recent years than in the past. Customers worldwide are increasingly demanding cleaner production. Furthermore, sustainable development policies and frameworks are continuously evolving and updated, e.g., by various UN climate conventions. More-recent studies will usually be based on more developed green supply chain practices. Therefore, we include the publication year of the study, or equivalently, the age of the study as a moderator variable.

Altogether, the following moderator variables will be included in our meta-analysis:

• Type of organization
• Size of organization
• Continent
• ISO-Certification
• Year of the study

2.3. Results of Previous Meta-Analyses about the Relationship between GSCM and Firm Performance

Meanwhile, an enormous number of studies have examined the relationship between GSCM practices and different performance dimensions in various industries and countries, and with different components of a supply chain. In the following, we refer only to previous meta-analytic results including the relationship between GSCM practices and firm performance. To date, five relevant meta-analyses have been published [3,4,5,6,7].

Almost all the primary studies included in these meta-analyses used five- or seven-point Likert scales consisting of a few items, often three to seven, to measure GSCM practices as well as performance dimensions. Questionnaires containing these Likert scales and other questions, e.g., with respect to size or type of organization, were sent to the companies to be completed by an executive familiar with these topics. The relationship between GSCM practices and organizational performance was calculated by product-moment correlation coefficients, which were used as effect measures in all meta-analyses.

2.3.1. Overall Effects

While the categorization of the different GSCM practices and the examined performance dimensions differed slightly, the main finding was similar for all meta-analyses: GSCM practices were significantly positively related to firm performance [3,4,5,6,7]. The meta-analysis of Fang and Zhang reported positive correlations between GSCM practices and operational performance (r = 0.481) as well as economic performance (r = 0.464) [5]. Golicic and Smith found smaller correlations in their meta-analysis, namely with accounting-based performance (r = 0.256), market-based (r = 0.317), and operational-based performance (r = 0.301) [3]. Govindan et al. reported a slightly stronger association between GSCM practices and firm performance in terms of operational and financial performance (r = 0.370) [6]. In the meta-analysis of Qorri et al., GSCM practices were slightly stronger correlated with operational performance (r = 0.457) and economic performance (r = 0.420) [7]. Geng et al. limited their meta-analysis to studies in the manufacturing sector of Asian emerging economies. The relationship between green supply chain management practices and operational performance as well as economic performance resulted in r = 0.370 and r = 0.431, respectively [4].

From the results of the previous meta-analyses, it can be concluded that the correlation coefficients between the studies are very heterogeneous. Thus, applying a random-effects model seems to be appropriate. In addition to the expected values of this distribution, the standard deviation is also of interest to evaluate the heterogeneity of the correlation coefficients. However, information about the heterogeneity of the effect sizes is very sparse. Golicic et al. used a fixed-effects model and therefore do not measure heterogeneity [3]. However, large Q-statistics indicate a high degree of heterogeneity in this meta-analysis. Golicic et al., Geng et al., and Govidan et al. [3,4,6] use the meta-analytic approach of Hunter-Schmidt [16], which is essentially not a random-effects model. From the reported statistics, e.g., Q or I², a considerable amount of heterogeneity can be inferred in these meta-analyses. Only Fang et al. apply an explicit random-effects model and report the variance in the correlation coefficients [5]. The standard deviation of the correlation coefficients between all GSCM practices and economic performance amounts to 0.304 which can be considered a very high degree of heterogeneity. The corresponding standard deviation for the relationship between all GSCM practices and operational performance, SD = 0.205, is somewhat smaller.

In summary, there is a large heterogeneity in effect sizes between studies, but most of the meta-analyses to date do not provide the necessary information about the amount of heterogeneity.

2.3.2. Type of Green Supply Chain Management Practices

The results regarding the influence of different green supply chain management practices are not consistent. For example, while the meta-analysis by Golicic and Smith identified eco-design as the predictor with the greatest influence on market, operational, and economic performance, this result did not occur in the meta-analysis by Fang and Zhang [3,5]. In many cases, the influence of the different categories was not very large. In no case were statistical tests performed to determine whether the differences were significant.

2.3.3. Moderator Variables

Type of Organization

According to Golicic and Smith, the correlation between GSCM and overall firm performance is r = 0.368 for the automobile industry, r = 0.268 for single industry studies, and r = 0.337 for various industry studies [3]. Geng et al. differentiated between the automotive, electronic, and various industries, the corresponding correlation coefficients amounted to r = 0.453, r = 0.377, and r = 0.380, respectively [4]. Fang and Zhang used the same categorization for this moderator, but the number of studies for the automotive and electronic industries was not sufficient to derive meaningful results [5]. The correlation between GSCM and firm performance was higher for manufacturing (r = 0.372) than for services (r = 0.265) in the meta-analysis by Govidan et al. [6]. For multiple-industry studies, the correlation coefficient was r = 0.300. Finally, in the meta-analysis by Qorri et al. these correlation coefficients were very similar for the automotive, electronics, food, shipping, logistics, and other industries, ranging from r = 0.560 to r = 0.610; a higher correlation of r = 0.720 was found for the construction industry [7].

Size of Organization

The effect of company size on the relationship between green practices and performance was not significant in the meta-analysis by Golicic and Smith [3]. According to Geng et al. GSCM, this effect was slightly higher for large companies than for small and medium companies (r = 0.428 vs. r = 0.380). Finally, Qorri et al. found higher correlations for small and medium-sized companies (r = 0.537) compared to large companies (r = 0.420) [7].

Continent

Qorri et al. reported a very small difference between the correlations between GSCM and firm performance for Asia (r = 0.494) and Europe (r = 0.463) [7]. Further differentiation of the geographic region consisting of eight countries revealed correlations between 0.402 and 0.553 for six of the eight countries, with the highest correlation found for South Korea (0.661) and the lowest for the United States (r = 0.287). The corresponding correlations reported by Golicic and Smith were r = 0.451 for Asia, r = 0.195 for Europe, and r = 0.266 for North America [3].

ISO Certification

In the meta-analysis by Govidan et al., the correlation between GSCM and firm performance was higher for ISO-certified companies (r = 0.400) than for those without such certification (r = 0.304) [6]. A negligible difference in this correlation between ISO and non-ISO-certified companies was observed by Qorri et al. (r = 0.443 vs. r = 0.463), while Fang and Zhang found a higher correlation for ISO-certified companies in terms of economic performance (r = 0.519 vs. r = 0.434), but not operational performance (r = 0.463 vs. r = 0.469) [5,7].

Publication Year/Age of Study

Golicic and Smith grouped the pre-2003 and post-2008 studies into one category and compared these studies to those between 2003 and 2008 [3]. The authors hypothesized that it may have been more difficult to achieve results from business initiatives between 2003 and 2008. However, they found a higher correlation of r = 0.341 versus r = 0.267 for the periods classified as more difficult. According to this result, the effect of study age might be not linear.

In their meta-analysis, Govidan et al. compared studies published before 2010 with those published after 2010 [6]. The association between green supply chain management and firm performance was significantly higher in the studies published after 2010 than in the earlier studies (r = 0.368 vs. r = 0.172). Fang and Zhang included the year of publication as a continuous moderator but found no significant overall association between green supply chain management practices and economic as well as operational performance [5]. Qorri et al. found a correlation of r = 0.481 for studies published between 2010 and 2018 and a significantly lower correlation of r = 0.348 for studies published between 1996 and 2009 [7]. It should be noted that these authors used a broader concept of overall company performance.

The results of the moderator analyses of the meta-analyses considered here show an inconsistent picture. Frequently, results are contradictory. This result could be mainly due to the different literature included in these meta-analyses and the different operationalization of the moderators.

2.4. Need for Updating the Meta-Analyses

The need for an additional meta-analysis is to update the previous meta-analyses in terms of the data to be analyzed and the statistical questions. This would allow us to provide more precise and reliable results. First, our dataset includes more recent articles that were not used in previous meta-analyses. We very carefully examined the effect sizes in the articles that were used in previous meta-analyses. In several cases, we could not derive the correlation coefficients that were processed in other meta-analyses. Apparently, in some of the previous meta-analyses, standardized regression coefficients from multiple regressions or standardized path coefficients from structural equation models were equated with correlation coefficients. However, standardized regression or path coefficients are generally not the same as correlation coefficients [17]. Further information, such as correlations between predictors, is usually required to derive correlation coefficients. However, in some cases, we could not find this information in the original articles.

Most previous meta-analyses have used fixed-effects models. However, because of the high degree of heterogeneity in this field, fixed-effects models are not adequate at all. Moreover, random-effects analyses (like fixed-effects analyses) assume independent effect sizes. However, many primary studies report multiple measures, leading to dependencies between effect sizes. When dependencies between effect sizes are ignored in both fixed-effects and random-effects models, test statistics and confidence intervals are generally biased [18]. Therefore, in previous meta-analyses (with one exception), dependent effect sizes have been excluded or averaged. However, much information can be lost by these procedures.

In a meta-analysis with robust variance estimation [18], dependent effect sizes are explicitly taken into account. Thus, all effect sizes from primary studies can be included in a meta-analysis based on this approach. In addition, the assumptions underlying this approach are not as strong as for other meta-analytic techniques (see below). Therefore, we will use this meta-analytic technique in our study.

In general, it is not possible to include all moderator variables simultaneously in a regression equation of a meta-analysis. This is because there are no or only a few observations for many combinations of the levels of the moderator variables. Then, the influence of the moderators must be analyzed separately, as has been the case in previous meta-analyses. However, moderators are often confounded, and separate analyses of the influence of moderator variables may lead to wrong conclusions. By using robust variance estimation in our meta-analysis, we are able to include a sufficiently large number of effect sizes to analyze all moderators simultaneously.

2.5. Research Questions and Hypotheses

The following hypotheses are derived mainly based on the results of previous meta-analyses. First, these previous meta-analyses found positive associations between GSCM practices overall and the various performance dimensions analyzed here [3,4,5,6,7]. This results in the following hypotheses H1a to H1d:

Hypotheses 1a–1d (H1a–H1d).

There is a positive correlation between GSCM practices and (a) overall firm performance, (b) market-based performance, (c) operational-based performance as well as (d) accounting-based performance.

From the results of the previous relevant meta-analyses, it can also be concluded that the different categories of GSCM practices, green supplier orientation, green customer orientation, eco-design, and green production, are positively correlated with overall firm performance, as well as with the three dimensions of firm performance considered here. Hence, hypotheses H2a to H5d are:

Hypotheses 2a–5d (H2a–H5d).

There is a positive correlation between green supplier orientation (H2), green customer orientation (H3), eco-design (H4), as well as green production (H5) and (a) overall firm performance, (b) market-based performance, (c) operational-based performance, and (d) accounting-based performance.

The results of the effects of the moderator variables were very inconsistent throughout the meta-analytic results reported above. No clear picture emerged for any of the moderator variables examined. Therefore, no hypotheses are stated concerning the effects of the included moderator variables.

3. Method

3.1. Inclusion and Exclusion Criteria

To be included in this meta-analysis, the primary studies had to meet several criteria. Only scientific journal articles in English or German reporting quantifiable measures were considered; case studies or theoretical reviews were therefore excluded. According to another inclusion criterion, studies had to report either correlation coefficients or other measures from which correlation coefficients can be calculated. In addition, only primary studies that used measures of firm performance reflecting economic performance rather than social or environmental performance were considered for further investigation. The independent variable had to relate to green or sustainable practices in the supply chain or closely related corporate practices.

3.2. Literature Search

In a preliminary literature search, we first included all articles used in previous relevant meta-analyses. These articles were reviewed to determine whether they met our inclusion criteria. Many of these articles had to be excluded because the supply chain practices or performance measures did not match those specified for our analysis. Additionally, in the case of some articles, we could not see how the authors derived correlation coefficients from other measures, such as standardized regression coefficients.

Then, the following databases were searched: EconLit, GreenFILE, MLA Directory, PsycArticles, PsycInfo, PSYNDEX, and Regional Business News. We used a broad search term to include as many potential articles as possible:

((SU supply chain) AND (SU sustainability OR SU sustainable practices OR SU environmental practices OR SU green OR SU lean OR SU social responsibility OR SU corporate responsibility OR SU environmental management OR SU safety OR SU health OR SU performance OR SU reverse logistics) OR (SU closed loop)) OR ((SU closed-loop supply chains OR SU eco-logistics OR SU eco-supply chains OR SU environmentally friendly logistics OR SU environmentally friendly supply chains OR SU green logistics OR SU green reverse logistics OR SU green reverse supply chains OR SU green supply chains OR SU responsible supply chains OR SU reverse logistics OR SU reverse supply chains OR SU sustainable supply chains) AND (SU performance OR SU consequences OR SU outcome))

This search yielded 3227 articles (the last search was on 22 November 2021). First, the titles and then abstracts were screened to select articles that appeared suitable. From this subset, the abstracts of those articles that had not been included in previous meta-analyses were examined. Finally, this search yielded 134 studies with 137 independent samples from which 408 correlation coefficients (N = 30,568) were extracted. (For a flowchart, see Appendix A. The extracted studies are listed in Appendix B).

3.3. Coding Procedure

The coding of the relevant variables from the primary studies was based on a detailed coding scheme. Two staff members of the Unit of Statistics and Methodology of the Institute of Psychology at the University of Münster coded the studies independently of each other. In the case of divergent ratings, a joint decision was made in consultation with one of the authors. General study information such as authors, journal name, and publication year, as well as sample characteristics such as sample size and moderator variables, were coded.

While most of these variables and their values were carried over as they were reported in the original study, some variables were coded based on newly created categories. In cases where company size was reported in terms of the number of employees of organizations, the organizational size was assigned to one of the following categories: small < 50 employees, medium = 50–249 employees, large = 250–999 employees, very large = 1000–4999 employees, and extremely large > 5000 employees. These categories were based on the KMU (small and medium-sized enterprises)-recommendation of the EU-commission [19] for defining small and medium firms.

As correlation coefficients were used as effect sizes, the reported correlation coefficients between the examined GSCM practices and the performance measures were extracted for the primary studies. Moreover, the reliability of both the measure of the GSCM practice and the performance measure was noted in order to correct the correlation coefficients for measurement errors.

The independent variable GSCM practice and the dependent variable were coded using the above categories, i.e., green supplier orientation, green customer orientation, eco-design, and green production for GSCM practices, and operations-related, market-related, and accounting-related performance to categorize the performance measures. The moderator variables were coded as follows:

• Type of organization: Automotive, Electronics, Food, Textiles, various
• Size of organization: small, medium, large, very large, extremely large, not specified
• Continent: Asia, Europe, North America, various
• ISO certification: yes, partially, not specified
• Age of the study (in years)

3.4. Statistical Procedure

Since our database consists of 408 correlation coefficients based on 137 independent samples, the studies included in this meta-analysis contribute on average three coefficients per sample. Because the correlation coefficients from one sample are dependent, the standard meta-analytic setup, i.e., the inverse variance meta-analysis, cannot be applied. As mentioned above, neglecting the dependencies of the effect sizes leads to erroneous standard errors of the estimates and thus, to biased confidence intervals and test statistics.

A meta-analysis based on robust variance estimation can be successfully applied to our data. Here, the covariances of the sampling errors of effect sizes per sample are explicitly taken into account by specifying the correlation between the dependent effect sizes within the studies. This correlation has to be guessed, but such a guess is quite robust against misspecification. Additionally, its adequacy can be checked by a sensitivity analysis.

Using a multivariate meta-analysis allows for the inclusion of dependent effect sizes that come from the same study. However, this method requires knowledge of the covariance matrices of the sampling errors per study. The information used to derive these covariances is almost never published in original studies and is therefore not available in almost all primary studies that address the relationship between GSCM practices and firm performance.

The robust variance estimation is based on a weighted least squares approach which leads to strongly consistent estimators of the parameters under a wide range of conditions. Thus, it is not necessary that the effect sizes be assumed to be normally distributed as in the standard meta-analytic models. As already pointed out, unlike the multivariate meta-analysis model, the correlation of the sampling errors within samples has not to be known for the estimation of the parameters.

In addition to the sampling error, effect sizes may be impaired by other sources, particularly measurement inaccuracy in the independent and dependent variables. When applying the correction for attenuation $r_{\left(\mathrm{adj}\right)}= \frac{r}{(\sqrt{reliabilit{y}_{\mathrm{IV}}} \ast \sqrt{reliabilit{y}_{\mathrm{DV}}})}$ [16] the reported correlation coefficients were adjusted to correct for impaired measurement precision that might attenuate the true correlation effect. When studies did not report the reliabilities of their measurements, the mean reliability of all studies that did report such a value was used for the missing reliability coefficients, specifically, $\alpha$ = 0.85, was used. Sensitivity analyses were performed to ensure the robustness of this value. Since the studies included in our meta-analysis are very heterogeneous and a fixed-effects analysis is not useful, we will always perform a random-effects meta-analysis.

In meta-analyses, Pearson product-moment correlation coefficients are usually transformed to the Fisher´s z-metric with the formula $z= \frac{1}{2} \ln \left(\frac{1+r}{1-r}\right)$. When applying this transformation the sampling distribution of the correlation coefficients will be better approximated by a normal distribution. However, such a distributional assumption is not necessary for a meta-analysis based on robust variance estimation. Furthermore, the correction for attenuation, which we will apply, leads to difficult distributional implications, which raises doubts about whether Fisher’z transformation is beneficial at all. Therefore, we do not use this transformation for our analyses.

To test the proposed hypotheses, effect sizes with the respective 95% confidence interval, variances of the effect sizes, and p-values were calculated. An $\alpha$-level of 0.05 was used for all significance tests. The present meta-analysis was conducted in R using the packages metafor, robumeta, and clubSandwich [20,21,22,23].

Since the moderators are confounded (see below), they should be included simultaneously as covariates. As mentioned earlier, the corresponding results are then more adequate than when the moderators are included separately. In meta-analyses, moderators are usually analyzed separately because information on the joint distribution of moderators is sparse. However, the robust variance estimation approach allows us to include all effect sizes in our analyses, including dependent ones, and thus we are able to consider all moderators collected in our study simultaneously.

4. Results

4.1. Descriptive Statistics

Overall, 408 correlation coefficients from 137 independent samples and 134 studies (N = 30,568) were used for this meta-analysis.

Table 1. Relative frequency distribution of the variables included in the meta-analysis.

Variable	Level	%
Firm performance	operational-based	35.8
	market-based	22.5
	accounting-based	41.7
GSCM practices	eco design	14.7
	green supplier orientation	21.1
	green production	24.4
	green customer orientation	15.4
	various	24.4
Type of organization	automotive	7.6
	electronics	7.8
	food	5.6
	textiles	2.5
	various	76.5
Size of organization	small	6.4
	medium	31.1
	large	19.1
	very large	5.9
	extremely large	5.6
	not specified	31.9
Continent	Asia	59.8
	Europe	23.5
	North America	5.4
	various	11.3
ISO certification	yes	9.1
	partially	5.9
Age of publication (in years)	0–2	17.4
	2–4	14.0
	4–6	13.1
	6–8	19.4
	8–10	12.5
	more than 10	5.6
	not specified	18.1

contains the relative frequency distribution of the variables included in the meta-analysis.

Most effect sizes are based on accounting-based performance followed by operational-based and market-based performance. Green supplier orientation and green production are the most often occurring GSCM practices, while about 25% of all practices reported in the primary studies could not be clearly assigned to one of the four categories of this variable.

Most of the organizations operated in various industries (76.5%), the automotive industry (7.6%) and the electronics industry (7.8%) appeared most frequently as a single industry. A total of 37.5% of the retrieved coefficients were taken from samples with a small or medium employee number and 30.6% from larger organizations. In 31.9% of all cases, the size of the organization was not reported.

Most correlation coefficients came from Asian (59.8%) or European (23.5%) samples, a small amount (5.4%) from North America, while 11.3% were classified as “various”. The country could not be specified for 31.9% of the effect sizes. ISO certification was known for 9.1% of the effect sizes and partial ISO certification for 5.9%. However, information on this variable could not be obtained for 80.5% of all cases. While 18.1% of the effect sizes were retrieved from studies that were older than 8 years, i.e., published before 2013, the majority (81.9%) of the coefficients were retrieved from studies that were published after 2013.

Considering these results, it should be mentioned that for many variables, the proportion of the level “various” or “not specified” was quite high. This must be taken into account for the moderator analysis.

Further important information about the studies is provided by bivariate analyses of GSCM practices and moderator variables. Associations between publication year and the other variables are measured with η, all other associations with Cramer’s V. First, the associations were determined on the basis of all levels of the variables (upper values presented in

Table 2. Bivariate analysis.

Variable	GSCM Type	Organizational Type	Organizational Size	ISO Certification	Age of Study	Continent
GSCM type	-
Organizational type	0.142 **	-
	0.210
Organizational size	0.156 **	0.432 ***	-
	0.133	NA
ISO certicifation	0.112	0.458 ***	0.163 *	-
	0.212	NA	NA
Age of study	0.081	0.204 ***	0.323 ***	0.142 **	-
	0.067	0.350 ***	0.377 ***	0.301 *
Continent	0.110	0.246 ***	0.430 ***	0.194 ***	0.187 **	-
	0.130	NA	0.600 ***	NA	0.164 **

*** p < 0.001, ** p < 0.01, * p < 0.05. Associations between the year of publication and the other variables are measured by η, all other associations by Cramers V and reported in the upper row. The associations by excluding the levels “various” or “not specified” are given in the lower row. When the associations could not be determined due to the sparsity of the tables NA is reported.

). Because of the high proportion of the levels “various” or “not specified”, the associations were additionally calculated without these values (lower values presented in Table 2).

The associations when excluding the levels “various” or “not specified” are mostly somewhat higher than those including these categories. But, often Cramers’ V could not be computed in this case since too many cells with zeros occurred. Several association coefficients yielding values greater than 0.30 indicate a considerable degree of confounding. By including all independent variables in the moderator analysis, this confounding is accounted for.

Further insight into the confounding can be gained by inspecting the contingency tables. It can be seen that, actually, all GSCM categories occurred most frequently in Asian countries, followed by European countries or the USA. Moreover, all GSCM practice types were predominantly studied in medium- or large-sized firms or in those where the employee size was not specified. However, it is noteworthy that production-related practices were frequently used in medium-sized firms as well as in unspecified firms, while green customer-related practices as well as a combination of several practices (category: various) were mostly used in large-sized, medium-sized, and unspecified firms. While green supplier and green customer orientation practices were most frequently applied in the automotive industry or various industries, production- and design-related practices occurred most frequently within the electronics industry or in various industries.

The moderator variable “continent” showed considerable associations with the type and size of the organization and ISO certification. In European countries, the food industry and various industries were the most common. The Asian samples were predominantly from the automotive or electronics industry or various industries. While the studies from Asia predominantly examined medium-sized companies, large companies, or companies with no specification of the number of employees, the American studies mainly dealt with large companies, and the European studies primarily with small or medium-sized companies or companies with an unknown number of employees. In terms of ISO certification, only Asian companies indicated that they had implemented these certifications. For all other continents, with the exception of the partial adoption of certificates in Europe, no information was provided on the status of ISO certificate adoption. In addition, a considerable association between ISO certification and firm size was found in the sense that all certified organizations were of medium size, or their size was not specified. Almost all companies that reported implementing ISO certifications were in the electronics industry. While the automotive industry included primarily large and very large companies, the companies considered in the electronics industry were of medium, small, or unspecified size. All companies in the food industry could be classified as small.

These results again indicate substantial confounding. By including all moderator variables in a regression equation, the influence of the individual moderator can be measured.

4.2. Meta-Analytic Results of the Relationship between GSCM Practices and Firm Performance

First, we will test hypotheses H1a–H1d. According to this, there is a positive correlation between any GSCM practices and (a) overall firm performance (OA), (b) market-based performance (M), (c) operational-based performance (O), as well as (d) accounting-based performance (A).

The first four rows in

Table 3. Results of the meta-analysis with robust variance estimation of correlations between GSCM practices and firm performance.

Ind	Dep	$\widehat{\mathit{E}}\mathbf{\left(}\mathit{\rho }\mathbf{\right)}$	SE	LBCI	UBCI	$\widehat{\mathbf{\tau }}$	LBCR	UBCR	k	n
All	OA	0.442	0.018	0.407	0.476	0.287	0.071	0.812	408	137
	M	0.429	0.035	0.359	0.500	0.302	0.033	0.826	92	41
	O	0.481	0.025	0.431	0.530	0.233	0.177	0.784	170	63
	A	0.389	0.026	0.338	0.440	0.237	0.080	0.697	146	68
GSO	OA	0.436	0.025	0.385	0.486	0.193	0.183	0.688	85	54
	M	0.403	0.046	0.305	0.501	0.188	0.142	0.665	15	15
	O	0.481	0.034	0.410	0.551	0.193	0.223	0.738	44	32
	A	0.413	0.038	0.335	0.492	0.181	0.169	0.658	26	23
ED	OA	0.473	0.029	0.414	0.532	0.241	0.157	0.788	59	48
	M	0.465	0.063	0.328	0.602	0.268	0.091	0.839	14	14
	O	0.541	0.035	0.468	0.614	0.154	0.330	0.751	21	19
	A	0.429	0.040	0.346	0.511	0.158	0.213	0.645	24	22
GP	OA	0.462	0.030	0.403	0.521	0.260	0.123	0.801	98	57
	M	0.479	0.057	0.357	0.602	0.252	0.132	0.826	19	16
	O	0.513	0.050	0.409	0.616	0.210	0.227	0.798	40	23
	A	0.414	0.039	0.335	0.494	0.202	0.143	0.686	39	26
GCO	OA	0.501	0.029	0.443	0.560	0.246	0.179	0.824	62	43
	M	0.492	0.048	0.388	0.596	0.194	0.222	0.762	15	14
	O	0.509	0.048	0.409	0.609	0.255	0.165	0.853	28	21
	A	0.465	0.034	0.393	0.537	0.156	0.252	0.678	19	18

Ind = independent variable, Dep = dependent variable, $\widehat{E}\left(\rho \right)$ = estimated expectation of the correlation coefficient corrected for attenuation, SE = Standard Error of $\widehat{E}\left(\rho \right)$, LBCI = lower bound of 95% confidence interval of $E\left(\rho \right)$, UBCI = upper bound of 95% confidence interval of $E\left(\rho \right)$, $\widehat{\tau }$ = standard deviation of the effect sizes, LBCR = lower bound of 80% credibility interval of ρ, UBCR = upper bound of 80% credibility interval of ρ, k = number of correlation coefficients used, n = number of studies. OA = overall, M = Market-based, O = operational-based, A = accounting-based, GSO = Green supplier orientation, ED = Eco Design, GP = Green production, and GCO = green customer orientation.

correspond to the meta-analytic results for these hypotheses. This table contains the estimated mean (or more precisely the expectation) of the distribution of the correlation coefficients $\widehat{E}\left(\rho \right)$, its standard error (SE), and the lower and upper bound of the 95% confidence interval (LBCI and UBCI, respectively) and 80% credibility interval (LBCR and UBCR, respectively), as well as the estimated standard deviation of the effect sizes ($\widehat{\tau }$) and the number of samples (k) and studies (n). All mean correlations are significantly positive, and hypotheses H1a–H1d are confirmed. However, the estimated standard deviations of the distribution of the correlation coefficients are very large, ranging from 0.233 to 0.302. In this respect, the lower and upper bounds of the 80% credibility intervals show a very high amount of heterogeneity.

In addition, a Wald test was conducted to test whether the association between GSCM practices in general and market-, operational- and accounting-based performance (corresponding to rows 2–4 in Table 3) are identical or different. As can be seen in

Table 4. Results of multiple regression of overall firm performance on the GSCM practices and moderators.

Variable	Level	b	SE	t	df	p	CILB	CI.UB
	intercept	0.265	0.850	3.122	27.29	0.004	0.091	0.439
GSCMP	ED	0.070	0.046	1.525	60.390	0.133	−0.022	0.163
	GCO	0.058	0.045	1.276	59.470	0.207	−0.033	0.148
	GSO	0.040	0.042	0.965	68.800	0.338	−0.043	0.123
	GP	0.032	0.045	0.708	78.150	0.481	−0.058	0.121
Organizational type	automotive	0.014	0.055	0.265	9.120	0.797	−0.109	0.138
	electronics	−0.001	0.097	−0.073	8.160	0.943	−0.230	0.216
	food	0.003	0.062	0.052	5.910	0.961	−0.150	0.157
	textiles	−0.127	0.114	−1.118	1.130	0.447	−1.226	0.972
Organizational size	small	0.061	0.062	0.983	7.610	0.356	−0.083	0.205
	medium	0.078	0.047	1.663	66.740	0.101	−0.016	0.171
	large	0.053	0.048	1.096	56.820	0.278	−0.044	0.150
	very large	−0.019	0.086	−0.224	8.780	0.828	−0.213	0.176
	extremely large	−0.010	0.122	−0.082	5.670	0.938	−0.312	0.292
Continent	Asia	0.179	0.062	2.900	21.990	0.008	0.051	0.308 ***
	Europe	0.054	0.067	0.808	28.460	0.426	−0.083	0.190
	North America	0.004	0.133	0.032	11.820	0.975	−0.285	0.293
ISO certification	partially	−0.068	0.062	−1.101	8.830	0.300	−0.207	0.072
	yes	−0.000	0.064	−0.001	12.450	1.000	−0.139	0.139
Age of study (in years)		0.001	0.006	0.242	37.510	0.810	−0.011	0.013

b: Estimated regression coefficient; SE = Standard Error of regression coefficient; 95% CI = 95% confidence interval of regression coefficient; t: t-value of the test that regression coefficient = 0, df: degrees of freedom of t, p: p-value of the test that regression coefficient = 0, UBCI = upper bound of 95% confidence interval of regression coefficient, LBCR = lower bound of 95% confidence interval of regression coefficient, *** = p < 0.001.

, this test revealed no significant difference (F(2,89) = 2.03, p = 0.138).

According to hypotheses H2a–H5d, there is a positive correlation between green supplier orientation, green customer orientation, eco-design, as well as green production, and overall firm performance and the three performance dimensions distinguished here. Rows 5–19 in Table 3 contain the results related to these hypotheses. Again, all hypotheses are confirmed and also a considerable amount of heterogeneity is found. All estimates of the mean correlations are higher than 0.40 and quite similar. Moreover, all standard deviations of the correlation coefficients are close to the median of about 0.20.

Again, Wald tests for testing the associations of each type of GSCM practices with the three kinds of performances did not yield a significant result in any of the four cases.

The parameter $\rho$ for the correlation of sampling errors was set to 0.8 for all analyses. A sensitivity analysis with the values 0.2, 0.4, and 0.6 showed no significant influence on any parameter estimate.

4.3. Multiple Regression of Overall Firm Performance on the GSCM Practices and Moderators

Table 4 contains the results of the multiple regression of overall performance as the dependent variable on the GSCM practices (GSCMP) and moderator variables as predictor variables. All predictors were simultaneously included in the regression equation. Age of study represents a continuous variable, while all other predictors are dummy-coded categorical variables. The reference category for the ISO certification variable was “not specified” and the reference category for all other categorical variables was “various”.

With one exception, there is no significant regression coefficient. The mean correlation for Asia of r = 0.18 is higher than the mean correlations for the reference category and, thus, for North America. The second largest (absolute) value for the regression coefficient of the “Textiles” category is −0.127, but it is not significant.

The variance ${\tau }^2$ for the regression equation to predict the (overall) firm performance when only an intercept is included amounts to 0.082 and decreases to 0.062 when all predictors are included. This represents a reduction of about 0.02 by including the moderators. However, this difference is not significant (F(19,20.5) = 1.04, p = 0.451), as indicated by a Wald test for linear contrasts using a sandwich estimator and a correction for small sample p-value.

In addition, Wald tests were performed to check that the influence of the levels of the categorical moderator variables was identical or different (see

Table 5. Results of Wald tests.

Variable	F	df_num	df_denom	p
GSCMP	0.355	3	69.70	0.786
Organizational type	0.316	3	4.18	0.814
Organizational size	0.358	4	18	0.835
Continent	4.270	2	17.3	0.0311
ISO certification	0.767	1	15.7	0.394

F: F-statistic of Wald test, df_num: degrees of freedom on numerator of F-statistic, df_denom: degrees of freedom of denominator of F-statistic, p: p-value of Wald test.

). A significant difference was only found for the variable “continent”, as expected from the previous result.

The results of the regression analyses may be influenced by the sometimes high proportion of the categories “various” or “not specified”. Simultaneous inclusion of all independent variables without these categories in a regression analysis failed because of the small number of observations for many combinations of the levels of the predictor variables. Therefore, each categorical predictor without the level “various” or “not specified” was included in a regression analysis with all other predictors including all levels. However, these analyses did not change the previous result regarding significance.

4.4. Publication Bias

Publication bias can occur when some studies testing a particular hypothesis are unpublished or inaccessible. As a result, both effect sizes and variances may be incorrectly estimated if studies confirming the null hypothesis are missing. In our meta-analysis, publication bias may exist if there are too many studies with negative or nonsignificant positive effect sizes and a corresponding asymmetry in effect sizes is observed. As a test for this assumption, we perform a one-sided Egger sandwich test that accounts for the dependence of effect sizes [24]. This test regresses the effect sizes on the sampling error (SE) using robust variance estimation. The negative regression coefficient of −4.923 (t = −7.88, p < 0.001) indicates that there is no publication bias. A funnel plot and trim-and-fill analysis for dependent effect sizes are not currently available. Performing these two approaches without considering dependence as exploratory methods indicates, contrary to our hypothesis, that some studies with positive effect sizes are lacking.

5. Discussion

Understanding the relationship between GSCM practices, firm performance, and moderators can be considered to be crucial for finding strategies for organizations to take on social and environmental responsibility without neglecting the goal of increasing profit. By providing a comprehensive meta-analytic update on this relationship, the present work aimed to contribute to this seminal issue that is part of the challenge of mitigating climate change. By using meta-analysis based on robust variance estimation, considering publication bias, integrating a larger and substantially updated database, and simultaneously analyzing the impact of moderator variables, the present meta-analysis extended previous meta-analyses [3,4,5,6,7] in crucial ways. Based on these extensions, the aim of this meta-analytic approach was to confirm previous meta-analytic results, to shed light on mixed meta-analytic findings, and to offer more precise estimations of the relationship between GSCM practices and firm performance and its moderating effects. In the following, the above-presented results will be discussed in detail.

5.1. The Relationship between GSCM Practices and Firm Performance

With respect to the first hypothesis H1a, it was expected that there would be a significant positive relationship between GSCM practices and overall firm performance, based on the results of previous meta-analyses. Given the present result ($\widehat{\rho }$ = 0.402), this hypothesis was supported. According to Cohen´s classification of effect sizes [25], the strength of the overall relationship could be considered medium-high. It is stronger than the overall effect size in the previous meta-analysis of Golicic and Smith [3] (r = 0.294) which is conceptually comparable to that one in our analysis. For the other meta-analyses, we compare our overall effect with the average of the subset of the available effect sizes for market, operational, and economic performance. Fang and Zhang as well as Govidan et al. arrive at a smaller overall effect size (r = 0.255 and r = 0.370, respectively) while Geng et al. and Qorri et al. found a larger overall effect (r = 0.432 and r = 0.440, respectively) [3,4,5,6,7]. Altogether, our overall effect size of about r=.402 seems to be a reasonable result.

As expected in hypotheses H1b-H1d, significant positive relationships between GSCM practices and market-based (r = 0.400), operational-based (r = 0.427), and accounting-based performance (r = 0.359) were found. The difference between these correlations was not significant. Comparing our results with those based on similar independent and dependent variables in the other meta-analyses, again the correlations reported in the meta-analysis of Golicic et al. and Fang and Zhang are smaller, while they are slightly higher in the meta-analyses of Geng et al., and Qorri et al. [3,4,5,6,7]. It should be taken into account that the databases are different and that, in addition, there could be more or fewer differences in the operationalization of categories of GSCM practices and firm performance.

In summary, this meta-analysis was able to confirm previous meta-analytic findings in showing that there is substantial evidence for a positive relationship between GSCM practices and firm performance. Furthermore, no significant difference was found between the impact of the GSCM practices on market-, operational- and accounting-based performance.

5.2. Moderators and Type of GSCM Practices Explaining Differences in the Relationship between GSCM Practices and Overall Firm Performance

To explain the differences in effect sizes, the present meta-analysis examined the influence of five moderator variables and the type of GSCM practices. Although it may be of great interest to both researchers and practitioners which of the applied GSCM practices is most beneficial, this question is probably one of the most difficult to answer.

As previous meta-analyses predominantly discussed results on descriptive differences in the effectiveness of specific GSCM practices, the present approach aimed to also determine whether significant differences in the impact of different GSCM practices on firm performance occur. The different types of GSCM practices were not found to be a significant effect in explaining the variance in the correlation coefficients found. One explanation for this result could be that it is difficult to categorize the range of GSCM practices reported in the primary studies into the selected GSCM aspects considered in both this and the other meta-analyses. This appraisal is consistent with Sellitto et al. who emphasized that differentiation among the wide range of GSCM practices was difficult [26]. Although there was no evidence of significant differences in effect sizes between the different GSCM practices, each category showed a significant positive relationship with overall performance as well as with each performance dimension.

Only one moderator variable showed a statistically significant result, the relationship between GSCM practices and firm performances is different for the continents included in this meta-analysis. Such a result was also found by Golicic and Smith [3], but without conducting a statistical test. In view of these results, one might wonder why this pattern is found even though Europe seems to have focused most attention and regulations on this issue. However, the fact that the focus on sustainability emerged later in Asia than in the more industrialized countries of Europe [27,28] could also be the starting point for understanding Asia’s current strength in improving corporate performance through applied green practices. First of all, many international companies have moved their production sites to Asia, or new companies have been set up there, motivated by cheap labor costs [29]. Accordingly, Asian companies have been under strong pressure to improve their economic performance and gain a competitive advantage [30]. In addition, with the increasing global attention to environmental issues, there was pressure to consider environmental sustainability [4]. The given pressure and expectations of various stakeholders may have been strong drivers for the adoption of GSCM practices in Asia [1]. Accordingly, many purchasing companies demanded that their suppliers adopt green practices [31,32].

It stands to reason that GSCM practices could have different effects on company performance depending on the type of industry in which the companies operate. Comparing the results of the subgroup analysis descriptively, the relationship between GSCM practices and firm performance was lower in the textile industry than in the other industries, which had similar coefficients. However, no significant differences were observed.

As previous meta-analyses provided inconsistent evidence on whether organizational size has an impact on how strongly GSCM practices are related to firm performance, the present meta-analysis attempted to shed light on this issue. This moderator effect was not significant. This is at odds with the plausible conclusion that larger companies that adopt GSCM practices more frequently achieve better economic outcomes as a result of these practices [33,34]. Moreover, some studies have found that firm size is a factor influencing effectiveness in adopting GSCM practices [35,36] or even moderating the relationship between environmental innovation and financial performance [37]. However, the literature also found no positive influence of firm size on the adoption of GSCM practices [28]. This result is consistent with Golicic and Smith’s meta-analytic finding that firm size has no effect on the relationship between GSCM practices and firm performance [3], as well as with previous primary studies that also demonstrated the absence of an effect [38,39]. A methodological explanation for the present result could be that the coding of the given samples was challenging due to the company size of most organizations that were part of a sample. As the information provided in the primary studies was sometimes sparse, it was not always easy to form the categories.

Previous meta-analytical results showed mixed evidence of effects on ISO certificate acquisition. The effect sizes identified in our analysis did not differ significantly for ISO-certified, partially ISO-certified, or non-specified organizations. This is not consistent with some previous primary studies that have found a positive effect of ISO adoption on the feasibility of GSCM practices and even on improved organizational performance [40]. Previous meta-analytic results were controversial in that Fang and Zhang found a descriptively higher effect size for the ISO-certified subgroup than for the unspecified subgroup [5], while Geng and colleagues even found a descriptively stronger effect for the unspecified subgroup [4]. The present results thus shed light on this issue, as no significant moderation effect was found. Moreover, the descriptive difference between the certified and the non-specified group seems to be negligible in the present meta-analysis. However, the number of studies that did not specify whether their samples had adopted considerably exceeded the number of studies that reported having studied ISO-certified companies. This pattern was also found in other meta-analyses [4,5]. However, a plausible explanation for the association between green practices and firm performance not being influenced by ISO certification could be the following: Even if companies are ISO-certified, this does not necessarily mean that they really apply all recommended GSCM practices. This was empirically demonstrated in a study where only 17% of the surveyed ISO-certified companies in Malaysia were really ambitious in implementing several GSCM practices [41]. Moreover, this relationship can also be reversed. Companies that consistently implement GSCM practices can improve their performance results without being ISO certified [4]. In particular, manufacturing companies in Asia are very dependent on overseas markets and are not forced to obtain ISO certification but to comply with expected green policies within the supply chain [4,41].

Finally, the influence of the age of the publication on the relationship between GSCM practices and company performance was examined. However, no significant moderator effect was found when this variable was included in the regression equation. According to the results of Golicic et al., there could be a non-linear effect. However, a scatterplot gave no evidence of such an effect. Including an additional quadratic effect of this moderator variable also resulted in a non-significant effect. It can therefore be concluded that a moderating effect of the age of the publications does not exist.

5.3. Publication Bias

An analysis of publication bias, as performed in this meta-analysis, was not included in most previous meta-analyses. Only Govindan et al. applied a different approach to analyze the presence of publication bias [6]. Our results are consistent with those of Govidan et al. in that there is also no publication bias in our analyses for the overall association between GSCM and firm performance in the sense that effect sizes are overestimated. Nevertheless, the presence of publication bias should be investigated in further research.

5.4. Implications for Management and Policy Making

Consistent with previous meta-analyses, our updated meta-analytic results show clear evidence of a substantial relationship between GSCM practices and firm performance. This is not only a relevant scientific finding but also an important message for managers and companies as well as for decision-makers in regulatory authorities and political institutions. In this way, potential reservations and prejudices can be reduced. Companies should consider implementing GSCM practices throughout the supply chain. Our findings confirm that all activities within the supply chain, from supplier interactions to design, production, and customer coordination, improve business performance. Moreover, in addition to overall profitability, which is measured by return on investment ratios, profit, and revenue, financial indicators that represent market targets or relate to quality, flexibility, and speed are also improved. Our research also provided significant empirical evidence that GSCM practices are positively related to firm performance, regardless of the type or size of the firms and the continent in which they are located. Managers could establish monitoring and control systems that measure the impact of GSCM practices based on specific indicators of those practices and firm performance. Finally, policymakers should not shy away from developing efficient shelf systems that protect the environment, as companies also benefit from such regularities.

5.5. Limitations and Future Directions

Although the present meta-analysis was carefully conducted, some limitations should be noted. Some of them have already been addressed and will be comprehensively presented below.

First, some limitations regarding the research framework applied should be discussed. In order to examine the relationship between GSCM practices and organizational performance, categories had to be formed to include primary studies in this meta-analysis. The categorization of GSCM applied was quite broad. Therefore, GSCM practices did not provide the opportunity to identify the impact of more specific practices on performance outcomes. Although there could be other reasons for not finding significant differences in the degree of influence between green practices, this could be due to this very broad categorization. For example, the category ‘green supplier orientation’ included practices such as supplier collaboration and green purchasing. Although these practices certainly have similar aspects, they could have different impacts on performance outcomes. Future research should address this limitation and include more specific GSCM practices in examining the relationship with organizational performance. It would also be interesting to investigate whether certain combinations of different GSCM practices are necessary or more effective in increasing organizational performance than single practices or combinations that have been analyzed so far. Future research could therefore compare organizations using different combinations of GSCM practices.

When looking at the three distinguished performance dimensions, namely market-based, operational-based, and accounting-based performance, the same concerns arise as mentioned above. The relatively broad performance dimensions fulfill the objective of providing a comprehensive overview of the relationship between practices and performance and represent the performance measures used in the primary studies. However, in the future, it may be interesting to conduct studies that examine the impact of green practices on single performance metrics. For example, while a particular GSCM practice may increase sales growth, it may also decrease profits but increase a company’s overall economic performance [4]. In addition, the use of objective measures of performance metrics should be encouraged, as they provide a more valid and reliable perspective and facilitate comparisons between practices adopted that are not dependent on the researcher’s judgment.

In addition, several moderator variables that are hypothesized to influence the relationship between practice and performance were included. Such variables were selected that were used as control variables in the primary studies and shown as influential or were identified by previous meta-analyses as relevant moderator variables. Although there are several other variables that could influence this relationship, they could not be included in the analysis because the primary studies did not provide sufficient information about them. For example, the maturity of a company’s GSCM practices or the antecedents of GSCM practices, including stakeholder pressure or other driving forces, could be relevant [33,40,42]. Future research in different industries and countries should conduct longitudinal studies to examine the evolution of the relationship between specific green practices and performance at different stages of maturity of GSCM practices, from adoption to frequent use. In addition, it should be clarified whether certain cultural differences influence the success of the adoption of green strategies.

In the future, studies should be conducted that take into account greater variation within the geographical region of the samples and at least attempt to include a wider range of industries and companies of different sizes within different countries. Finally, the levels of the moderator variables should always be described in sufficient detail so that coding with the categories “various” or “not specified” can be avoided.

Finally, the independent and dependent variables should be measured in more detail. Typically, one member of an organization responded to Likert scales consisting of a few items to measure GSCM practices and dimensions of performance. These measures have led to a “global” assessment of the different constructs, but not to a differentiated picture. For example, the nature, intensity, or duration of the GSCM practices assessed by respondents may have varied considerably between the organizations studied. Additionally, respondents may have different ideas about firm performance in mind. Therefore, it could be beneficial to use more objective indicators to operationalize company performance.

5.6. Conclusions

Given the urgent need to take action to address climate change, this study contributes to this challenging topic by providing a comprehensive meta-analytical update of the relationship between GSCM practices and firm performance. Based on a much larger database and methodological extensions, this meta-analysis extended the findings of previous meta-analyses using robust variance estimation. It supports previous findings by showing significant moderate relationships between GSCM practices and market-, operational-, and accounting-based performance. Considering that no publication bias was identified, this work makes an important contribution to the research area in question, as practitioners can be strongly encouraged to view the integration of green practices as an advantage rather than an obstacle. Moreover, previous meta-analytical findings could be confirmed by identifying the continent as a significant moderator of the relationship between GSCM practices and firm performance. However, no evidence of any other moderating variable was found. This result is not surprising given the previous, often contradictory, findings.