Within this chapter the chosen method is described. First, the sampling technique is presented, which is followed by data collection. Moreover a description of the statistical tests that are used to analyse the data are described.
The study investigate larger firms in Sweden, from the NASDAQ OMX Stockholm Large Cap and Mid Cap. The study is limited to Swedish companies as they follow the same rules and regulations, along with the same norms and customs. The results there-fore contain less noise. The study does not include banks and insurance companies be-cause they operate under different regulations (Cooke, 1989; Principe, 2004).
The sampling technique used is random sampling. All companies that are listed on Large Cap and Mid Cap, with the exception of banks and insurance companies, was given a number. The numbers assigned do not indicate any order of the companies, it is used only for identification since ordering them may create bias (Bryman, 2012). This is a total of 124 companies, from which a sample was selected using a website1 recom-mended by Bryman (2012). It randomly picked 70 numbers and the companies with the chosen numbers were selected into the sample. The name of the companies included in the sample are presented in Appendix 1. A sample of 70 is deemed to be sufficient to fulfil the purpose, as there are previous studies with a similar sample size (Bukh et al, 2005; Whiting & Woodcock, 2011; Yau et al, 2009). Since it is more than half of the population, it should be large enough for the results to be generalized to Swedish listed companies that are not in the financial sector.
In order to perform statistical tests, the companies were divided into different groups. For the test of hypothesis H1, industry type, the companies in the sample were divided into two groups, high tech and low tech. When determining to which sector a company belongs, the classification on NASDAQ has been followed. The high tech group is companies from the sectors Healthcare, Telecom and Technology and the low tech group is companies form Oil & gas, Materials, Industrials, Consumer goods, Consumer services and Utilities. As mentioned before, in this study companies from the financial sector are excluded and the companies classified as Financials was therefore omitted. This classification gave 13 companies in the high tech sector and 57 companies in the low tech sector. As the division of the companies into different industry types are col-lected from an external party, it is objective and easy to access for anyone.
It appeared that the most logical method for dividing the sample when testing (H2- H6), would be to split it at the median which will result in two equally large groups. This technique has been used before by Liao et al. (2013). In order to test the age characteris-tic, the companies was ranked by the amount of years they have been listed. This meas-ure was also used by Prencipe, (2004), Haniffa and Cooke (2002), Li et al. (2008) and Liao et al. (2013). Companies that had merged was counted from the year of the merger. This resulted in one group of older companies and one group with relatively younger companies. When the sample was split by the median there were companies with the listing age of 16 in both groups. In order to avoid companies with the same age being in different groups two tests were conducted. One where the firms in the older group were 16 years and older and one other where they were 17 years and older. This resulted in a division of the sample where the group with older firms consisted of 38 companies for one test and 33 for the other. When testing for size, the companies were ranked accord-ing to their market capitalization as was used by White et al. (2007). The ownership concentration was measured by how large part of the shares that are owned by the three largest shareholders. This measure was also used in a previous study by Oliveira et al. (2006). In order to rank the companies after leverage their total liabilities was divided by their total assets, the same was used by White et al, (2007). The measure that was used to measure profitability is return on equity. This was achieved by dividing the companies’ net income by the shareholders equity, as was used by Haniffa and Cooke (2002) and Eng and Mak (2003). The information used to form the groupings are gath-ered from audited annual reports, and thus is objective enough to make sure the infor-mation is interpreted the same way regardless of who is observing it.
Accounting narratives can be found in everything from formal documents, websites, and press releases to social media posts (Beattie, 2014). It is not possible to go through all that information in one study. The information in this study is therefore retrieved from annual reports since it is the main communication channel for companies (White et al, 2007). To gather the data, the consolidated annual reports from the year end 2013 in electronic form has been collected from the companies’ websites. For companies that have a broken financial year, the annual report from 2013/2014 is used. The content of the annual reports was scrutinized and the voluntary disclosure was quantified in order to analyse it. In order to reduce the risk that something was overlooked the researchers had a checklist of relevant search words. The list was created so that there were differ-ent search words covering all the 78 items. A list on where to find the annual reports is presented in Appendix 2, and a list of the search words is found in Appendix 3.
A disclosure index that was developed by Bukh et al. (2005) was used to score the an-nual reports in order to quantify the amount of voluntary disclosure. The index has been effectively used in other previous studies (Rimmel et al., 2009; White et al., 2007), which indicates that it is a good way to measure voluntary disclosure. It consist of 78 items that are divided into six different categories, which are employees, customers, IT, processes, research and development, and strategic statements. For the complete disclo-sure index, see Appendix 4. If an item is disclosed in the annual report of a company, it got a score 1 and if it is not disclosed the score was 0. Using this method, the amount of voluntary disclosure was quantified and statistical tests possible to perform.
This thesis uses content analysis as method, where subjectivity of the coders are always present (Husin et al., 2012; Marston & Shrives, 1991). To ensure the consistency of the coding between the two independent researchers, both analysed the annual reports inde-pendently to see if there were any differences between the coding. According to Smith, (2011) there is no level that is agreed upon to be satisfactory, but it is suggested a con-sistency level above 80% should be achieved. When comparing the results after going through the annual reports a consistency of 92, 3% was achieved and therefore the inter-observer consistency is judged to be at a satisfactory level.
In order to see if there are significant differences between companies that have certain characteristics and the amount of voluntary disclosure, statistical tests was performed. Parametric tests are the most powerful and are therefore the preferred tests to use. They do however rest on an underlying assumption about the data, which is that the data is drawn from normal distributed data (Smith, 2011). Non-parametric tests are not as pow-erful, but they do not make such assumptions and therefore they can be used even if the data cannot be assumed to be normally distributed (Smith, 2011).
Normal distribution can be assumed if the sample size is 30 or higher (Aczel, 2009). When testing H2-H6, there are two equal groups of 35 companies in each. This is enough to assume normal distribution. However, when testing for H1, industry type, the two groups was not equal and the group with high-tech companies were lower than 30. Therefore a non-parametric test was used when testing H1. For all the tests a signifi-cance level of 95 % was used, which is the convention among social researchers (Bry-man, 2012).
In order to test all the hypotheses, except for H1, analysis of variance (ANOVA) was used. This test is used to see if the samples differ from each other so much that they can be judged to come from different populations (Smith, 2011). ANOVA tests the differ-ence between the sample means in relation to the variation within the samples. If the difference between the samples are bigger than within the samples, it is assumed that the samples come from different populations and thus, the null hypothesis is rejected (Smith, 2011).
In order to test the hypotheses H1 industry type, normal distribution cannot be assumed and therefore the Mann-Whitney U-test was used. It answers the same question as the ANOVA: Could the samples have been drawn from the same population or are the dif-ferences between them so large that it is unrealistic? The difference between the two is that Mann-Whitney U-test is a non-parametric test and therefore it is based on the rank order of the values within the groups instead of the variance as is the case in a paramet-ric test.
In order to measure the association between certain company characteristics and the amount of voluntary disclosure the correlation was calculated by using Pearson’s r. It shows if there is a relationship between variables, the direction of it and how much of the variation in one variable that is explained by the other (Bryman, 2012).
This section of the thesis starts by providing descriptive statistics of the sample. There-after, the results of the hypotheses tests are presented.
Above is a table of the sample and the different sectors it consists of. The highest mean value (28,5) was achieved by the technology sector. The technology sector in this sam-ple consist only of two companies, which is too few to draw generalizations. But if we include the telecom and healthcare sector which has the second and third largest means in in the sample, the total mean score for the high tech firms is 25,5. This should be enough to make a generalization that high tech companies in Sweden provide more vol-untary information than low tech firms. The lowest mean value can be found in the Oil Gas sector with a mean of 13, consisting of only two companies. The total mean val-ue for the low tech companies is 18,91.
The average disclosure when including all companies is 26% (20 items). Looking at the range in the sample, the lowest score was attained by Fenix Outdoor of 7,5% (6 items). The highest score attained was by Active Biotech of 50% (39 items). Looking at the companies’ disclosure in the sample one can see that most information was provided about the companies’ employees, averaging 32% over the sample. Second comes strate-gic statements where on average 28% was disclosed. When looking at the sample one can see that the more knowledge intensive industries such as technology, telecom and healthcare have disclosed a higher amount of information about IT and R&D. This is not surprising since these sectors depend on IT and R&D to a larger extent than the oth-ers.
Results from hypothesis testing
The detailed results from SPSS are presented in Appendix 5.
H1: Industry Type. When testing for industry type a Mann Whitney U-test was per-formed. Type of industry was found to be significant with a p-value of 0,004. The re-sults therefore indicate that the high tech companies in the Swedish Mid Cap and Large Cap disclose more information than the low tech companies. The sample of high tech companies in this study had a mean rank of 50,08 while the low tech companies had a mean rank of 32,18.
H2: Age. When testing for the variable age we performed two ANOVA tests. The first test with 38 large and 32 small companies resulted in a p-value of 0,041, with a mean of 21,68 for the large firms compared to a mean of 18,31 for the smaller firms. The second test with 33 large and 37 small companies resulted in a p-value of 0,036, with a mean of 21.97 for the older firms and a mean of 18,51 for the younger firms. The results indicate that there are a statistical difference between the groups’ mean scores. Also when per-forming a correlation test, a positive correlation of 0,362 between company age and voluntary disclosure was found. The correlation was significant with a p-value of 0,002.
H3: Size. When testing for company size, the ANOVA test resulted in a p-value of 0,008. Therefore it can be concluded that there is a significant difference between the groups mean scores and therefore it is unlikely that they come from the same popula-tion. The group of large companies had a mean score of 22,31 and the group of small companies had a mean score of 17,97 indicating that larger companies provides more voluntary information. A Pearson’s’ r test was performed and a positive correlation of 0,178 was found. However, it had a p-value of 0,140. Since the p-value is greater than 0,05 the Pearson’s r test is not significant because it cannot be proved that the correla-tion was not found by chance.
H4: Ownership concentration. Ownership concentration was found to be not significant with a p-value of 0,430 from the ANOVA test. This means that H4 that claims that companies with more dispersed shares disclose more information is not supported by the test. The correlation test is also insignificant with a p-value of 0,961.
H5: Leverage. The ANOVA does not show any support for H5 that states that compa-nies that are more leveraged disclose more information than less leveraged firms. The p-value for the ANOVA is 0,242 and is therefore not significant. The correlation test showed a p-value of 0,270 and is also not significant.
H6: Profitability. The ANOVA test for the hypothesis showed that there was not a sig-nificant relationship between profitability and voluntary disclosure with a p-value of 0,560. Companies that was more profitable had a mean score of 19,66 and therefore disclosed insignificantly less information than less profitable companies that had a mean score of 20,63. The correlation test was also insignificant with a p-value of 0,089 and hence, no conclusions can be drawn from it.
Analysis of company characteristics’ influence on voluntary disclosure
In this chapter the results are analysed, starting with H1 moving on and ending with the analysis of H6. The results are interpreted with the use of relevant theories and previ-ous research.
H1: Industry type. The impact of industry type confirm the theoretical assumptions made from Legitimacy Theory, Stakeholder Theory and Signalling Theory. Stakeholder Theory suggest that all stakeholders should be treated equally and get as much infor-mation they need and it can increase the confidence between the firm and its stakehold-ers. Signalling Theory stipulates that signals sent by voluntary disclosure enhance the quality of the firm. Knowledge intensive companies has a need to show their intellectual capital in order to satisfy the stakeholders’ need for information and to signal their val-ue. The information is particularly important for investors when making investment de-cisions about industries that rely on intellectual capital (Breuggen et al., 2009). This is because there is a larger difference between the book value and market value in knowledge intensive companies (Guo et al. 2005). These companies are forced to use voluntary disclosure to a larger extent since a large part of their assets are intangible and cannot be accounted for according to IAS 38. Therefore they need to legitimize their status through voluntary disclosure, as they cannot do it through fixed assets (Whiting
Woodcock, 2011). This thesis shows that these assumptions ring true in a Swedish context as well. Most of the previous studies have also confirmed industry differences to be a significant factor.
The results are in line with previous studies performed by Breuggen et al. 2009; Bukh et al, 2005; Nurunnabi et al. 2011; Rimmel et al. 2009; Whiting & Woodcock, 2011. However, the Swedish study by Broberg et al, (2010) found that research intensive in-dustries such as healthcare and telecommunication disclose less voluntary information. They argue that their results could be explained by the assumption that companies in this field need to shield themselves from competitors and therefore have a reason to dis-close less information. The difference of results with Broberg et al, (2010) can be ex-plained by factors such as time difference, there are 8 years between the observations.
The studies use different disclosure items in the scoreboards. This study has 27 items concerning employees while Broberg et al, (2010) have none. Broberg et al, (2010) use a larger sample and also has a larger target population. This study only depicts the situa-tion in Swedish Large Cap and Mid Cap while their study consider all listed firms. Therefore they have a higher amount of smaller and relatively newly incorporated firms in their sample which might have a different disclosure behaviour.
H2: Age. The results from the test of listing age are in line with the studies performed by Hossein and Hammami (2009), Prencipe (2004) and White et al. (2007) who also found that older companies disclose more information. That is also coherent with the expectations on the base of Stakeholder Theory and Legitimacy Theory. As corpora-tions need to be perceived as legitimate by society in order to survive (Deegan, 2006), older firms have incorporated this in their culture. Drastic changes in a firm’s disclosure strategy can cause tensions between the firm and its stakeholders (Roberts, 1992). It is more expensive for younger firms to produce information on intellectual capital (With-ing and Woodcook, 2011), also older companies have developed their reporting practic-es over time which might also explain a higher amount of disclosure (Liao et al, 2013).
The results from this study contradict the findings by Bukh et al. (2005) where there was no significant relationship. Also, Rimmel et al. (2009) concluded that younger firms disclosed more. These differences could be explained by the fact that those studies were performed on Initial Public Offerings (IPO’s), which are not yet listed firms. This implies that their sample would all be considered ‘young’ firms with the measurement used in this study, which is listing age. This could also be a reason for the contradicting results. Since the majority of studies performed on listed firms confirm the hypothesis while the opposite is true for IPO studies one can conclude that the disclosure behaviour is different between IPO’s and listed firms. It is also supported by Branswijk and Evera-ert (2012) who found that companies have a higher amount of voluntary disclosure in their prospectus than in their following annual report.
H3: Size. As expected, the results from the ANOVA imply that larger firms have a higher amount of disclosure in accordance with Stakeholder Theory and Agency Theo-ry. Large companies have the resources to disclose more and are also more scrutinized by the public, since they have a larger impact on society. Large companies also have more intellectual capital to disclose and therefore size can have an explanatory effect (Petty & Cuganensan, 2005). These findings are in line with previous studies done by Watson et al. (2002), Guthrie et al. (2006) and White et al. (2007). As can be seen in ta-ble 3-1, the mean value of the market capitalization is 44, which is considerably closer to the lowest value than the highest. This means the market capitalization of most of the companies are closer to the minimum value than the maximum, which can also be seen in the scatter plot in Appendix 6. This may explain that the correlation is not very strong or significant even though the ANOVA shows that size is an influential factor.
Table of Contents
2 Theoretical framework
2.1 Intellectual capital
2.2 Relevant theories
2.3 Previous research
2.4 Hypothesis development
3.2 Data collection
3.3 Data analysis
4.1 Descriptive statistics
4.2 Results from hypothesis testing
5 Analysis of company characteristics’ influence on voluntary disclosure
6 Conclusion and Limitations
6.2 Limitations and future research
List of references
GET THE COMPLETE PROJECT
Company characteristics and voluntary disclosure of intellectual capital A study on Swedish listed companies