A statistically significant association between CSR and CFP was examined using stakeholder theory, which guided the research project and supplied the framework. As a theoretical framework for analyzing CSR, Freeman (15) stakeholder theory centered on stakeholders’ viewpoints about company practices (16). According to Donaldson and Preston (19), stakeholder theory provides a framework for studying how stakeholder management practices relate to the success of corporations. According to stakeholder theory, organizations are just distinct interest groups. According to stakeholder theory, businesses have responsibilities to various groups, including shareholders, consumers, suppliers, workers, and communities (20, 21).

The core tenet of Agency Theory, which was prevalent in the 1970s (22, 23), was that managers’ primary duty was to run the company as usual, without regard for the interests of stakeholders or shareholders outside the company. Economic literature from the 1960s and 1970s established a distinction between an organization’s agent (the management) and principal (the owner) and the power to delegate decision-making to the agent (24–26). In addition, as stated in Namazi (26) and Kuo et al. (27), the literature has shed light on the distinct function of the agent, also known as management, as the party involved in a contractual arrangement with every other party involved.

Stakeholder theory, developed by Freeman (15), broadened the scope of corporate viewpoint beyond agency theory to include all relevant parties, including workers, suppliers, creditors, community members, investors, and even the government. According to Ullmann (28), external stakeholders are organizations’ primary responsibilities. These stakeholders might be individuals or groups that indirectly or directly impact the organization’s accomplishments. This definition is based on Freeman’s stakeholder theory. According to Ullmann, CSR efforts measure a company’s success. There is a good correlation between social and economic performance, ROE, and earnings per share (EPS), as shown in research studies. Stock market performances are often used to assess economic performances. As a result, Ullmann laid the groundwork for meeting stakeholder expectations via CSR disclosure and a planned organizational structure. Among many other traits, CSR disclosure is linked to organizations’ openness and visibility (28, 29). Researchers have also shown that factors including managerial style, business structure, cost of capital, and firm size are significant (30, 31).

Companies voluntarily pledge to become responsible stewards of society, the environment, and their stakeholders via what is known as CSR (32–34). Stakeholder theory (15), CSR (32, 35), and the long-term viability of businesses have all been the subject of much academic investigation. According to stakeholder theorists, organizations should prioritize satisfying stakeholders’ needs as much as their shareholders. Accordingly, owners and investors are among the larger stakeholder groups impacted by organizational managers’ activities (15, 36). Recognized as the model of firm sustainability, the interaction between the organization and stakeholder groups is acknowledged to have a favorable influence on attaining the organization’s goals (15, 34, 37). The most recent definition of CSR offered by the corporate responsibility organization (CRO) was used for the sake of this research study. CSR is a tool that companies use to improve their image, meet their stakeholder obligations, show that they are accountable, increase share value, strengthen their commitment to CSR, and ensure long-term sustainability (Magazine, 2015, 38).

According to Marens (39) and Bowen (40), social responsibility of the businessman was the first to formally recognize the existence of CSR, which had its roots in the 1950s. Bowen intended that managers willingly take on a certain amount of responsibility rather than feeling pressured. Regarding issues such as pollution and car safety, he said that CSR may help remedy society’s shortcomings (41, 42). Archie Carroll proposed a three-pronged model of CSR in the 1970s, which included social concerns, business social responsiveness, and social duties. According to Carroll (43), organizations have responsibilities not just to society but also to the economy and the law.

The scope of CSR extends far beyond the executive suite, and a company’s ethical standards are an essential part of that framework. In addition, organizations shall guarantee the fulfillment of 41 specific duties via CSR operations. The CSR pyramid that Schwartz and Carroll (44) said may help businesses succeed. There are four parts to corporate citizenship as follows: (a) Making a profit, which is the bedrock of all others; (b) following the law, which is society’s code of conduct; (c) acting ethically, which means you have a responsibility to do the right thing; and (d) philanthropic, which means you give back to the community (32, 45).

The evaluated literature shows a need for further study on CSR because of the mixed conclusions drawn from the many existing qualitative and quantitative studies on the subject (46). This analysis reduced the variance in results by illustrating comparable variables and similar organizations. Moreover, despite the large number of controlling and predicting factors used in prior studies, additional research is required to link and explain the connection between CSR activities and financial success (47, 48).

A quantitative, nonexperimental, cross-sectional research methodology was used in this study. Finding patterns in data by gathering and analyzing it statistically is the primary goal of quantitative research. The objectives of quantitative analysis include theory development, prediction, explanation, and validation of connections (49). Using measurement, description, and conclusions to examine connections was the criterion for selecting a quantitative research design (50, 51). According to Swanson and Holton (52), this method is positivist, descriptive, and quantitative. Consequently, this quantitative research strategy was deemed the most suitable for bolstering and gathering pertinent facts to answer the study issue. To clarify, the purpose of the study question was to determine whether there is a correlation between the corporate social responsibility index (CSRI) and the financial performance metrics of corporations, namely, ROA, ROE, and ROI. This study used SPSS version 23 to analyze secondary data extracted from the Russell 1000 database.

This study was theoretically grounded on stakeholder theory. According to Freeman (15), who developed stakeholder theory, companies should shift the focus from shareholders to stakeholders if they want to alter who has decision-making authority and benefits. Communities, society, and the economy all benefit from stakeholders’ capacities, according to Stieb (53). This study’s research topic was resolved as follows:

Using descriptive notation, the research question’s hypotheses were presented as follows:

With the use of statistical notation, we examined the following hypotheses on the study question:

The 1000 American businesses from the Russell 1000 database, made available by the investment company Russell Investments, made up the population for this study. We chose this database because it contains information on major publicly listed firms in the United States. This choice makes sense since the 1000 firms are all involved in some CSR, as shown by their CSRI ratings.

One thousand American businesses were used as a sampling frame, with their information culled from the Russell 1000 database (54). Participation in CSR initiatives by publicly listed American firms was a criterion for inclusion. Earnings ROI, ROAs, and ROE are all publicly available metrics obtained from the website of S&P Dow Jones Indices. The research study’s minimal sample size was determined using G*Power 3.1.9.2. Figure 1 graphically shows the process of calculating the sample. The adequacy of the sample size was checked using this instrument. The program showed that at a 95% confidence interval (CI), the sample size was at least 119 out of 1000. Out of a total population of 1000 businesses, 130 were selected at random using the random number generator in Microsoft Excel. Table 1 gives the protocol of power analyses.

Two sets of financial performance metrics—ROI, ROE, and CSR indices—and a set of CSR indices were culled from secondary data sources for this study. This investigation made use of the Russell 1000 database. While the Russell 1000 database is not accessible to the public, permission has been given to utilize it. Each company’s CSRI scores (DV) were supplied from the database. CSR’s seven subcategories employee relations, environment, climate change, finances, human rights, charity, and governance—contributed to the final CSRI score. A trio of profitability ratios ROA, ROE, and ROI—served as independent variables. The research study used a database with ratios for all companies. The independent variables, which were measures of financial success, were accessible to the public. Many factors contributed to the Russell 1000 database’s reputation as an accurate resource. The CSR categories that are used to determine the scores are thought of as factors that influence the importance of the organization.

Traditional multiple linear regression (MLR) was the tool of choice for data analysis in this investigation. According to Tabachnick et al. (55), MLR was a suitable multivariate statistical method for investigating and clarifying linear correlations between a single dependent and several independent variables. It was necessary to explore the data by looking for outliers, missing values, and assumptions before data analysis could begin. For this study, we used the following MLR model: The equation is given by

where i = 1, 2, …, n, where n is the sample size; y is the CSR index; b₀ is a constant; b₁ is the ROA regression coefficient; b₂ is the ROE regression coefficient; b₃ is the ROI regression coefficient; and e is the error term.

The data were analyzed using Statistical Package for the Social Sciences (SPSS), version 23.0. There was a search for missing data in the dataset. We did not find any missing data and did not need to make any modifications. Outliers, independence, normalcy, homoscedasticity, linearity, and appropriate multicollinearity were all examined to ensure that the data met the requirements of MLR.

Certain assumptions must be met for the MLR model (55, 56). The following assumptions must be satisfied to draw reasonable conclusions from the analysis: independence, normalcy, homoscedasticity, linearity, and reasonable multicollinearity. Eliminating outliers is essential, as is demonstrating homoscedasticity of variance, having a linear relationship between IVs and DV, and avoiding significant multicollinearity.

Finding outliers was a part of the exploratory data analysis testing process. Finding outliers is possible with the use of box plots or histograms. A box depicts the interquartile range between the 25th and 75th quartiles of the data (57). It is recommended to exclude data that falls outside of these quartiles. After removing the outliers, the sample size decreased from 130 to 125. Extreme data points, known as outliers, may skew parameter estimations (56). Using SPSS, version 23.0, box and whisker plots were used to identify outliers. As a result, the CSRI scores variable did not include glaringly high or low numbers. Figure 2, a box plot, shows no extreme values.

Both ROA and ROE showed three extreme values as shown in Figures 3 and 4, respectively. The ROI variable included four extreme cases (Figure 5). After removing the records with outliers, the sample size was reduced from 130 to 125.

No missing values were discovered in the sample when the data set was pre-screened for missing data before hypothesis testing. No missing data points were found during the research, as shown in Table 2.

The Durbin–Watson test was used to evaluate the assumption of residual independence (Table 3). The residuals of an MLR are independent variables that were shown using the Durbin–Watson test. A score of 2 indicates the absence of autocorrelation in the sample, whereas values between 0 and 4 are acceptable (56). The lack of autocorrelation in the model is confirmed by the Durbin–Watson values, which come out to 1.853.

Suppose the residuals did not originate from a normally distributed population. In that case, we may reject the null hypothesis and employ tools like histograms, Q–Q plots, and the KS test. Whether the null hypothesis [F(e) = N(μ, σ²)] or the alternative hypothesis [F(e) ≠ N(μ, σ²)] was supported for the normal distribution was determined using these tests. The normalized residuals’ histogram (Figure 1) provided evidence that the residuals may not follow a normal distribution, which is the alternative hypothesis. Because of the discrepancy, more analysis was necessary since the histogram revealed a bimodal distribution. This led to the execution of the KS test.

To determine whether the sample represents a specific distribution, statisticians use the KS test, sometimes called a goodness-of-fit test (56). To accept the null hypothesis, the sample must have originated from a known distribution, such as a normal distribution. The alternative hypothesis is obtained if the sample is not from a regularly distributed set. The empirical evidence for the nonnormal distribution of the residuals [F(e) ≠ N(μ, σ²)] is supported by the very significant results of the KS and Shapiro-Wilk (S-W) tests (Table 4) for the dependent variable CSRI Scores, where [(p = 0.001) < (α = 0.05)]. Both the tests by Fisher and Tabachnick et al. (55) might nevertheless fail to support the null hypothesis, even if they depend on normally distributed residuals [F(e) = N(μ, σ²)]. To determine whether the sample represents a particular distribution, statisticians use the KS test, sometimes called a goodness-of-fit test (56). To accept the null hypothesis, the sample must have originated from a known distribution, such as a normal distribution. The alternative hypothesis is obtained if the sample is not from a regularly distributed set. The empirical evidence for the nonnormal distribution of the residuals [F(e) ≠ N(μ, σ²)] is supported by the very significant results of the KS and SW tests (Table 5) for the dependent variable CSRI scores, where [(p = 0.001) < (α = 0.05)]. Both the tests by Fisher and Tabachnick et al. (55) might nevertheless fail to support the null hypothesis, even if they depend on normally distributed residuals [F(e) = N(μ, σ²)].

The skewness standard error (SE) is 217. With a skewness of −0.369, the CSRI variable is almost symmetrical. The center peak’s height concerning the standard distribution’s bell curve is called kurtosis. The findings also showed 430 SEs of kurtosis. In this case, CSRI had a kurtosis of −0.956. The results are all somewhat near zero, suggesting everything is normal.

A linear connection between the dependent and independent variables is assumed by the assumption of linearity (56). According to Tabachnick et al. (55), the forecasts’ accuracy depends on whether or not the assumption of linearity is satisfied. We used scatter plots and Pearson’s correlation to determine whether there were any correlations between the variables.

A zpred vs zresid scatter plot was executed to examine the hypothesis of homoscedasticity. The predicted scores (zpred) and the prediction errors (zresid) are shown on opposite sides of a residual scatter plot, which allows for visual analysis of the homoscedasticity assumption (55). The most reliable method for assessing skewed data is this nonparametric test applied to ranking data. Field (56) states that the distribution is normal if 95% of the standardized residuals lie within the range of −2 to +2. The results showed that the ROA predictor had a VIF of 4.76, indicating a modest association level—a moderate association, as demonstrated by the VIF of 1.73 for the ROE predictor. The ROI predictor’s VIF was 4.92, indicating a reasonable level of correlation. As a result, the VIF values are below the ten thresholds Field (56) suggested. The corresponding tolerance levels for the independent variables ROI77, ROI, and ROA are 210, 580, and 0.203.

Table 1 displays the Pearson correlations. Field (56) states that the ROE value is below the acceptable limit of 0.8, although it was not statistically significant for a two-tailed test (a/2 = 0.025). However, there is no statistical significance between the ROA (p = 0.812) and ROI (p = 0.895) values [(p = 0.001 < α = 0.05)] goes beyond the threshold of 0.8, which is considered optimal. Thus, the alternative hypothesis is supported by the variance inflation factor (VIF) and tolerance statistics for all variables. According to the correlation coefficients, only the ROE variable lent credence to the null hypothesis.

The goal of exploratory data analysis is to glean insights from the data that go beyond what can be derived from formal models, assumptions, or the outcomes of hypothesis tests.

Table 6 shows that the average CSRI score (DV) was 438.31. The range of possible scores ranged from 41 up to 728.1. The ratings went from very low to high since the sample was randomly chosen. The dispersion of the data was 180.66. There was an average ROA (IV) of 0.064. Scores ranged from 0.002 (the lowest possible) to 0.19 (the highest possible). The dispersion of the data was 0.042. There was an average ROE (IV) of 0.158. The score ranged from a low of 0.008 to a high of 0.406 among all 79 items. The dispersion of the data was 0.089. The average ROI was 0.088. Scores ranged from 0.003 (the lowest possible) to 0.229 (the highest possible). The dispersion of the data was 0.052.

According to Field (56), residuals are calculated as the discrepancy between the actual value of the dependent variable (y) and its anticipated value (w). The standardized residual is a statistical measure of the dispersion of the observed values with respect to the predicted values. According to Tabachnick et al. (55), the standardized residual is calculated by dividing the residual by the standard deviation (SD). A value of −2.405 is the rest. When the residuals were less than −2, the actual frequency was lower than predicted. The statistical significance of the MLR model’s predictive validity (model fit) was tested using the following assumptions, using a significance threshold of α = 0.05:

where ρ² stands for the population’s coefficient of determination.

To determine whether the model was fit, we used a significance threshold of α = 0.05 to test our hypotheses.

Tables 6 and 7 show a p-value of 0.086 for the model fit. The fact that p = 0.086 > α = 0.05 indicates that the predictor variables do not have a statistically significant impact on the dependent variable lends credence to the null hypothesis H₀: ρ² = 0.

In addition, the results show that the regression accounts for only 5.3% of the variation in CSRI scores, lending credence to the null hypothesis (H₀: ρ² = 0). The findings of the model fit test are supported by the scatter plots which indicate no evident linear connection between the dependent and independent variables. We tested the hypotheses for the individual population regression coefficients despite finding support for the null hypothesis for model fit (H₀: ρ² = 0). Table 8 displays the outcomes.