Kasvuryhmä provided their member company list with growth rates for the selection of the high-growth companies, together with a non-disclosure agreement. As a condition for the availability of the data, it was agreed that the research would be carried out anonymously. Hence, neither company names nor their official IDs can be disclosed in this research.
Finnish Technology Industries is an association for the technology branch that provides lobbying and employer services to Finnish technology companies. They provided a list of their member companies free of charge. The principle of anonymity was applied to both the control group and the industry control group. The Finnish rating company, Asiakastieto Oy, provided the financial information of all the companies in the dataset for years 2014 to 2018.
Missing financial data reduced the size of the sample
Although submitting the company financial information to the Finnish Trade Register is compulsory and should be done annually (Finnish Trade Register, 2020), there was missing data for some of the financial information. If a year was totally missing from the five-year period, these companies were rejected from the research.
Another reason for rejection was a severe discontinuity in relation to company history as a result of mergers and/or acquisitions or company restructuring. All the years were thus not comparable, which led to rejection. These rejections reduced the number of G and M companies from 40 to 37 and S-companies from 90 to 76 from the companies originally selected prior to the availability of complete financial data.
Hence, the total size of the sample was 150 companies. The net impact data from one company was missing, reducing the total sample size to 149 companies.
The test group companies were labelled from G01 to G37, control group companies from M01 to M37 and industry control group companies from S01 to S76.
All the variables are listed in Appendix 1.
3.2.1 Dependent variables
One of the limitations of using regression has been that the dependent variable should not be a sum of any other components used in the analysis (George &
Mallery, 2019). However, in the CSP literature, the de facto standard for measuring CSP is MSCI KLD 400 Social Index (Waddock & Graves, 1997; Perrault & Quinn, 2018; Kasperczyk, 2019). The MSCI KLD is a sum of both its positive and negative components and, in many publications, the aggregate value of MSCI KLD has been used as the dependent variable (Mattingly, 2017; Chen & Delmas, 2011). However, since the dimensions of net impact are very different from one another, the more interesting result of the study—rather than the net impact—is how the impact of different dimensions, e.g. from the impact lag point of view, differ from one another.
The above basis from previous research was used when concluding that sums of dimensions can be regressed.
The original dependent variables were: 1) net impact (NScoreA), the sum of four subcategories, 2) the sum of environmental impacts (EA), 3) the sum of health impacts (HA), 4) the sum of societal impacts (SA) and 5) the sum of knowledge impacts (KA). The CSP measurement method was the net impact method (upright.com, 2019) and the results were received for the dataset from upright.com.
Absolute values for impacts were used in this research study. Relative values were used in the robustness checks.
3.2.2 Independent variables
The financial information of the sample was received, pro bono, from the Finnish rating company, Asiakastieto Oy, which provides financial and credit information for Finnish companies. The information consists of profit and loss statements, balance sheets and some precalculated key figures for the sample companies from financial years 2014, 2015, 2016, 2017 and 2018. The fiscal year for all these companies was the calendar year (Todd & Taylor, 1993).
The independent variables consist mainly of the above financial data of the sample.
In addition, two dummy variables were constructed for the regression models to 1) test the impact of growth companies on the regression models and 2) to test whether a product company impact is different from that of a service company impact. All the variables are presented in Appendix 1.
Correlations between the independent and dependent variables were run with the long lists of independent variables so that the annual independent variables from years 2015 to 2018 were run against the dependent variables, measured in 2018.
One of the assumptions of multiple regression is related to the sample size—the sample should be large enough to allow about 20 cases per independent variable.
Therefore, the lists of independent variables had to be reduced 1) by finding multicollinearity between the independent variables and 2) by finding those independent variables that do not correlate at all with the dependent variables. To meet the first objective, intercorrelations for every independent variable were also run against each other, one by one. Alfa = 0.01 and Alfa = 0.05 were used as the flagging criteria for the significance of the correlation. Multicollinear and non-correlating variables were removed from the model.
Using the above procedure, the list of independent variables was reduced to:
Personnel growth % (PG1514, PG1615, PG1716 and PG1814), Turnover growth
% (TOG1514, TOG1615, TOG1716, TOG1814), Turnover per person € (TOP15, TOP16, TOP17, TOP18), Value add person € (VA15, VA16, VA17, VA18) and Return on Assets % (ROA15, ROA16, ROA17, ROA18). All the variables are defined in Appendix 1. According to the theoretical part of this research, personnel growth and turnover growth are growth-based measures commonly used in company growth research. Turnover per person, value add per person and ROA are accounting-based measures, out of which the return on assets is often used in CSP research and can be used as a slack resource proxy measure.
Two dummy variables, GrtwthCo (GrwthCo = 1 if a growth [test group] company, else GrwthCo = 0) and ProdServ (ProdServ = 1 if a product company [across the whole sample], else ProdServ = 0) were added to the list of independent variables.
A company was considered to be a product company if the percentual share of its product sales was more than 50%, while that of a service company is less than 50%
(Finnish Technology Industries, 2019; Cheung et al., 2013; Huang & Yang, 2014).
3.2.3 Control variable(s)
Company size and branch are the most commonly used control variables in CFP–
CSP and growth company research (Makni et al., 2008; Waddock & Graves, 1997;
Barbero et. al., 2011; Baum & Bird, 2010; Moreno & Castillas, 2007; Coad et al., 2014; Presutti & Odoricci, 2018). Company age is a common variable in growth company research but not in CSP–CFP research.
For company size, two alternatives were available in the data set, the personnel count and the company turnover in euros (€). These obviously correlated strongly with one another in the correlation tests of the previous section. Both were also used as input parameters for the net impact method (upright.com, 2019). In the explorative tests, it was discovered that the personnel count dominated as a predictor in the regression of societal impacts using absolute impact data. This is the reason why company size (neither personnel nor turnover) was not used as a control variable. Branch or industry sector is built into the net impact method through product taxonomy (uprightproject.com, 2019). Hence, branch was not used as a control variable.
Company age in years—calculated between the official registration date of the company with the Finnish corporate register and the respective ending date of the fiscal year—was selected as the only control variable (Age15, Age16, Age17 and Age18).