To what extent does companies’ use of
graphs in loss-making years differ
compared to profitable years?
Name: Mieke Schouten Student number: 10060820 Version: Draft
Date: June 15, 2015
Program: Accountancy and Control – Accountancy track Institution: Amsterdam Business School
Faculty of Economics and Business University of Amsterdam
Thesis supervisor: Prof. dr. V.S. Maas Word count: 14,808
Statement of originality
This document is written by student Mieke Schouten who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
The purpose of this paper is to see if there is a link between companies’ profitability and the extent of their use of graphs. Following existing literature on imaging in annual reports where there’s proof that images are used to create a more positive assessment of the company, it’s predicted that the same would be the case for graphs. From mixed model and t-test regressions conducted on ten years of hand collected data for 20 companies listed in the Netherlands (2004-2013), no significant effect of profitability on the total number of graphs was found. For the number of financial graphs, on the contrary, it was found that loss-making companies use fewer financial graphs than profitable companies.
This research contributes to existing literature on the use of graphs by being the first to investigate the effect profitability has on it and providing both investors and regulators with valuable information. For investors information hereon can be beneficial in order to take precautions against being misled by the primary monitoring tool they have (annual reports), for regulators the information is worthwhile in order to help assess the potential need for regulation. This study’s biggest limitation is that only Dutch companies have been investigated.
Table of Contents
1. Introduction 5
2. Literature Review and Hypotheses Development 7
2.1. Theory 8
2.2. The Nature of and Reason for Graphs in Annual Reports 9
2.3. Imaging in Annual Reports 10
2.4. The Use of Graphs in Annual Reports 11
2.5. Hypotheses 14 3. Research Methodology 15 3.1. Sample 15 3.2. Research Design 16 4. Findings 18 4.1. Descriptive Statistics 18 4.2. Data Validity 25 4.3. Results 26
4.4. Comparison Between the Different Regression Models 29
4.5. Discussion 30
5. Conclusion 33
6. References 35
1. Introduction
Over the last years, annual reports have increasingly become more graphic, designed and ‘glossy.’ Instead of merely discussing the year’s financial results, the reports now include many colourful pages with descriptions of the business, business tactics, risks and the company’s strategy. Apart from that, the report usually displays quite a number of images, under which graphs (Ball, 2011). The reports and especially their narrative sections have grown in size and generally use quite a lot of financial jargon, the use of which is claimed to make them hard to comprehend for readers (Dempsey et al., 2012, p. 451 and Jones and Shoemaker, 1994).
The main reason to include graphs in annual reports, has always been to aid investors in comprehending this difficult financial language by having a visual aid and thus decreasing the language barrier (Dempsey et al., 2012 and Mather, Ramsay and Serry, 1996). Additionally, Mather et al. state that graphs emphasize important information, summarize information, indicate trends and save investors time in analysing data. Another reason to include graphs in annual reports, is to emphasize certain things that the company considers to be important. This can be derived from the fact that graphs are very often a repeat of information that has already been presented elsewhere in the report, either in tabulated or in textual form (Lothian, 1976).
Due to the increased use of imagery in financial reporting that was mentioned before, there have been quite a number of investigations in this field and it has been found by both Graves, Flesher and Jordan (1996) and by Preston, Wright and Young (1996) that pictures in annual reports can and may be used to create a more positive appearance of the company than the actual situation (impression management). Graphs are kinds of images, which on the same time, unlike many ordinary images, have a very specific, content related purpose because of which they are included in the report. Being images, graphs are likely to also be used to create a positive company appearance (Beattie and Jones, 1992). Companies can for instance accomplish this through selectivity, which means bias in disclosures by only selecting favourable items to be graphed (Birnberg, Turopolec and Young, 1983). On the other hand, since graphs serve a very specific purpose of communicating information and enhancing it over purely numerical information (Desanctis and Jarvenpaa, 1989), it may be the case that companies find them too important to misuse.
The use of graphs in annual reports has been investigated by Beattie and Jones quite a lot, both on a national and on an international level. In various studies, they look at whether companies include graphs (1992; 1997; 2001), how many they include (1992; 1997; 2001), which variables have been graphed (1997; 2001; 2008), in order to subsequently compare this between countries. Even though, in 2008, they review the ways that graphs can be misused, Beattie and Jones don’t look at the difference between companies performing either good or bad, but at the ways graphs are used by companies in
general. In their 1992 research on use of graphs in UK companies that ‘consider their own performance to be good,’ however, they do lay the base for investigating the difference between profitable and loss-making companies. The results of their study show that companies considering their own performance to be ‘good’ relative to an absolute benchmark, are more likely than others to use at least one graph in their report. This may lead one to wonder, for instance, whether profitable companies use the graphs in their annual reports to try to hide their losses and emphasize their profits.
As can be seen above, just like the use of imagery, the use of graphs in annual reports has previously been researched. However, it has never been empirically investigated before whether there are differences in the use of graphs in annual reports between profitable and loss-making companies, which is precisely the aim of this research. Resulting, the question that will be investigated is “to what extent does companies’ use of graphs differ in loss-making compared to profitable years?” Companies’ use of graphs in the narrative section of their annual reports will be examined for a period of ten years, comparing annual reports in years that the company is profitable to ones in which they’re loss-making, so that the difference can be judged optimally. The data gathered in those years are panel data that will be explored using multiple suitable regression analyses.
The difference in use of graphs between profitmaking and loss-making companies has never been investigated before which leads this study to contribute to the literature by being the first study to empirically do so. The study thus extends current research on use of graphs, which hasn’t gone so far yet. Additionally, a time-series analysis concerning use of graphs, like this study does, has never been conducted over a period as long as ten years before. Furthermore, Beattie and Jones (2008) state that even though prior research provides us with an image of how financial graphs are used, there is still a lot of unknown information.
From a social point of view this study is interesting as well. Since graphs are often a repeat of information that has been tabulated or described on a different location in the report (Lothian, 1976), graphs show which information management considers important to emphasize in a handy, easily comprehensible format. There thus is managerial discretion over which information to stress and attract readers’ attention to, which can focus them towards certain data and material (Mather, Ramsay and Serry, 1996). Furthermore, Frownfelter-Lohrke and Fulkerson (2001) find evidence that companies worldwide present graphics that may be misleading to users. The authors find that this is more present for non-US than for US companies. That graphs may be used to mislead readers, is also stated by Fries and Jud (1997) as well as Taylor and Anderson (1986), the latter of which additionally show how graphs on firm performance can be distorted to create more advantageous graphical displays for firms. The study shows this can be done by discarding one or more of seven guidelines for graphical usage that the
authors’ literature review has suggested Also, even if the graphs are prepared according to key-preparation guidelines, for instance as the previously mentioned ones set by Taylor and Anderson (1986), still graphical displays may bias decision makers who view it (Amer, 2005). As a result, it is valuable for investors to know whether companies use their graphical discretion to their advantage so that they can acknowledge it and keep it in mind when reading and judging annual reports. By doing so, they can allow themselves to take precautions so that the introduced bias influences them to a lesser extent than it would otherwise.
Furthermore, this study may be interesting for regulators, partly because, so far, the use of graphs is unregulated. Companies wanting to use graphs properly, without bias, can follow guidelines set for this purpose, which have been introduced by prior literature. However, since there is nothing with more power and substance than guidelines, it’s fairly easy for companies that have less good or even bad intentions to manipulate information sent out in their annual reports. By the results of this study, regulators could start deciding whether regulatory action is necessary or not.
The results found don’t provide a significant difference regarding the total number of graphs used when a company is loss-making rather than profitable. However, when solely looking at financial graphs, a significant difference is found, which leads to conclude that management indeed uses fewer financial graphs with the intention of trying to hide bad performance. The difference is found using two separate models, which strengthens the results.
The remainder of the paper will be as follows: after first introducing economic theory that lays the base for this research, prior literature on the use of both images and graphs will thoroughly be researched and the hypotheses will emerge from this. The third paragraph will consecutively introduce the sample and describe the research design after which the fourth paragraph will extensively attend to the research findings. This fourth part includes descriptive statistics on the different variables, years in the sample and economic situations, provide some information about the validity of the data and furthermore describe the actual results, make a comparison between the different regression models and discuss said results further. Finally, the conclusion will be presented.
2. Literature Review and Hypotheses Development
This part of the research will lay the theoretical base the paper intends to extend. Firstly, the economic theories agency theory and obfuscation theory will be explained alongside the psychological dual coding theory. Second, the nature of graphs and motivations for including them in annual reports will be discussed, followed by literature on the use of imaging in annual reports provided that graphs are kinds of images and could therefore, logically, be used in a similar way. Inherently, this paragraph will
proceed with discussing literature that has so far been published on the use of graphs in annual reports, only to be concluded by developing the hypotheses used in this research.
2.1. Theory
The theory that is primarily built on, is agency theory. Agency theory deals with the difference in goals between the principal (being investors) and the agent (being management). Investors want the best for the firm because by investing in it they’re its owners and they’ll make the most profit out of the acts that are most beneficial for the firm. Management, on the contrary, is expected to act in its own best interest rather than the firm’s by pursuing its personal objectives and thereby harming firm value. As a result, the principal requires mechanisms to monitor the agent, for example the publication of an annual report (Jensen & Meckling, 1976). This report reflects on management’s performance so investors can judge whether management has conducted operations well. As is publicly known, management may try to have its performance reflected better (or less bad) than it actually is (Birnberg, Turopolec and Young, 1983), so that investors will judge their performance as good and being in the company’s best interest and reward the managers with a bonus (managers’ personal objective).
Information asymmetry between principals and agents, being investors and managers respectively, where managers have an information advantage over investors, leads to the ‘obfuscation hypothesis.’ This is explained by Adelberg (1979) as a situation where it is ‘perfectly natural’ to expect that managers use their complete control of the accounting communication process which monitors their performance to obfuscate their failures and emphasize their successes. Consistently, Bloomfield (2002) mentions that when it’s harder for principals to extract certain information from public disclosures, there’s a larger incentive for managers to hide bad firm performance. The information in annual reports is often said to be too difficult and too much for investors to comprehend all at once (Courtis, 1986), which, as said, increases the likelihood that managers try to hide bad performance (Courtis, 1995). This has been researched quite a lot, for instance by Li (2008), and Subramanian, Insley and Blackwell (1993) and the results are similar to one another.
Subramanian, Insley & Blackwell’s results (1993), for example, show that annual reports of companies with bad performance are more difficult to read than these of companies with good performance. Likewise, Courtis (1995 and 1998) found that the readability of annual reports decreases as companies have more bad news to disclose to investors. Li (2008) also found that annual reports of firms with poor performance are harder to read than those of firms with good performance. Additionally, he found that these firms’ results are less persistent than those of firms with good performance. As agency theory predicts and the papers just mentioned confirm, managers will use their influence on
annual reports to their own advantage by trying to hide bad performance and being more clear and open in reporting good performance.
Dual coding theory explains how people process graphics and words together and why combination of the two facilitates comprehending (and remembering) the matter better than when only one of the two is used. The explanation is separated into two systems: a verbal system which deals with language and a non-verbal system which deals with processing objects and events. According to this theory, when information is presented in two ‘codes’ (formats), an example of which being text and graphs, it’s both easier to comprehend and to remember (Sadoski, Goetz and Fritz, 1993, Clark and Paivio, 1991 and Mayer and Sims, 1994). Sadoski, Goetz and Fritz further mention that readability and content familiarity are factors that modify the extent of comprehension and remembrance. Something else that leads to remembering information better, is repeating it. Paivio (1991) states that according to the dual coding theory, when information is once presented in text and once in a picture, this repetition contributes more to remembering the information than does presenting the information in the same format twice.
As dual coding theory explains, using graphs in annual reports helps users understand the information more easily compared to just using plain text. Also, as mentioned before, graphs are a repeat of information presented elsewhere in the report which contributes to remembering the information better. Therefore graphs reflect which information agents want the principals (and other stakeholders) to comprehend and remember. It is very likely that due to the agency theory previously described, management will want to emphasize their positive achievements and thus their profits, not their losses. This can be explained by the self-serving bias. As stated in the International Encyclopedia of the Social Sciences, written by Darity (2008), a self-serving bias occurs when an individual (or, in respect to a firm, a body such as management) gives oneself more credit than he deserves due to the need to maintain or enhance self-esteem. With regard to organizations, Clatworthy and Jones (2003) explain how this works in the firm itself and especially in their annual reports’ accounting narratives. They say companies try to emphasize the positive aspects of their performance, for instance by emphasizing good news and taking credit for it, while simultaneously blaming the external environment for bad news (Aerts, 2005 and Bhana, 2009). Management serves itself by putting the organization in the best light possible, thereby increasing share prices from which management itself will often benefit through for example bonus schemes (Healy, 1985 and Holthausen, Larcker and Sloan, 1995).
2.2. The Nature of and Reason for Graphs in Annual Reports
Graphs in annual reports typically summarize financial data in an easily comprehensible manner. As annual reports have increasingly become large documents including a lot of financial jargon due to
expansion of the narratives, they’ve become harder to read for users (Courtis, 1995). Since humans are naturally better at using their dominant visual sense ‘sight,’ visual representations, like graphs, that can be ‘seen’ rather than observed otherwise, are more easily comprehended than for instance written text (Lewandowsky and Spence, 1989).
There are different types of graphs, which each serve specific purposes. The type of graph used, usually depends on the message the company wants to send. Common, frequently used graph types in annual reports, as explained by Beattie and Jones (2008), are line graphs, column graphs, bar graphs and pie graphs. Line graphs show a line that connects dots representing data points in order to show a trend therein. Together with column graphs, this is the type of graph that is most often used to represent a time-series of data. Column graphs consist of standing columns, meaning the columns are presented vertically, whereas bar graphs consist of lying columns, meaning the columns are presented horizontally. The last common type of graph, a pie graph, is a little different from the previously mentioned graph types. This type of graph is used to represent only one categorical variable at a time. This is a variable that can take on one of a limited, and often fixed, number of possible values.
According to Beattie and Jones (2008) there are two basic motivations for including graphs in an annual report: altruistic and self-serving. The paper introduces six main altruistic reasons for companies to include graphs in their annual reports. The first they mention is that due to a lack of regulatory constraints on the use of graphs, management is able to present information in graphs in a flexible manner. This means management has discretion over the content provided and can therefore signal valuable information. Secondly, graphs stand out. They catch the reader’s eye and information presented in graphs is therefore seen earlier, more easily and more frequent than other information. The third main reason to include graphs is that they are an attractive format to summarize, refine and communicate information. The fourth reason Beattie and Jones mention, is that graphs rely on humans’ spatial intelligence for their interpretation. This is relatively easy to use for humans and therefore decoding graphs is easier than decoding text, which means the information contained in graphs is comprehended sooner and better than that presented in textual form. Fifth, information from a graph is memorized better than plain numbers, because it’s displayed in a picture. Finally, graphs are egalitarian, impartial, equal. This is said to be the case for two reasons, the first of which being that inexperienced users can comprehend graphs just as well as experienced users. The second reason is that graphs are international since they lack language barriers.
2.3. Imaging in Annual Reports
It is a fact that, in a text, images attract the reader’s attention. This is something that has been exploited for a long time past, for example by advertisers and in newspapers. In annual reports, images may be
used for the same purpose. The use of imagery has been investigated by Preston, Wright and Young (1996), who stress that images can either truly reflect reality, mask it, mask the absence of it or create an image of reality that doesn’t truly reflect the actual situation. They’re saying that some companies purposely use images to create a perception of reality that is different from the actual reality. In this latter way of using images, the images have been produced with the intent of constituting reality instead of merely representing it. Images in annual reports, by creating a new or different reality, therefore, influence readers’ perception of the company’s overall performance.
Graves, Flesher and Jordan (1996) follow the same basic reasoning in their study on imaging. They argue that the images in corporate reports can constitute a form of rhetoric based on which readers assert the trustworthiness and the “truth claims” of a report. In other words, the function of the pictures in annual reports is to persuade readers of the truth in the accounts. According to the authors, readers’ judgements on these truth claims of images, rub off on whether they assess the numbers in the actual accounts to be a true reflection of reality. According to the research, positive images create a more positive reflection of reality and using positive images thus leads to annual report users assigning a higher credibility to the reports.
On the topic of positioning the organization in the most positive way, Stanton, Stanton and Pires (2004) research companies’ impression management. This is defined by Schlenker (1980, p. 6) as “the conscious or unconscious attempt to control images that are [either] real or imagined in social interactions” and explained by both Stanton et al. and Gardner and Martinko (1988) as organizations avoiding images that are expected to be negatively value loaded, while they choose to display as many images as possible that are expected to have a positive value for readers. The researchers consider this to be the case indeed, and David (2001) also shows how pictures in annual reports are carefully selected to portray a certain, desired image.
As can be seen above, prior literature suggests that regarding images there is the possibility for companies to use these to their advantage in masking or creating a new reality and that this possibility is seized. According to Simpson (1999), images are used both to woo readers and to detract their attention from other information in the report.
2.4. The Use of Graphs in Annual Reports
In a way, graphs and graphical portrays, are kinds of images as well. Just like images, graphs are a reproduction of something and add some livelihood to plain text. Therefore it can be argued that they may be used in a similar way as images, for instance by reflecting reality or by masking it to some extent. Financial graphs, just like other graphs, attract the reader’s attention by their nature. This fact can be exploited by managers in order to emphasize their positive performance when the company is
making profit (Lothian, 1976). Also, when companies are loss-making, managers may want to use (financial) graphs less often so that the reader’s attention isn’t attracted towards the negative numbers in a similar way as companies only using images with a positive value and avoiding negatively loaded ones.
Beattie and Jones are leading researchers in the field of the use of graphs. In various studies, some previously mentioned in this paper, they look at whether companies include graphs, how many they include and which variables they include. Subsequently, they compare this between countries. For example in their 1992 study they investigate the way in which 240 listed UK companies ‘use and abuse’ graphs in financial reports, meaning that they also look at the ways companies misuse graphs to benefit themselves, possibly at the expense of investors. They find that the average number of graphs per report is 5.9 in 1989, with 65% of companies graphing at least one important financial variable and that firms who perform ‘good’ relative to some absolute benchmark are more likely to use at least one graph than others. Additionally, they find that there is a significant number of measurement distortion present in the displayed graphs.
Measurement distortion occurs when the representation of a graph’s numbers isn’t in proportion to the actual numbers (Beattie and Jones, 2008, pp. 5-6). This breaches the constitutional principle of graph construction that graphs schematically and fairly represent numbers and may arise for instance by not using axes starting with zero or because the graphical specifiers, being the means by which the numerical values are transferred to the user (e.g. lines, bars or curves), are not drawn to scale (Beattie and Jones, 2008). With a level of 10% distortion as the boundary of what is considered material by Beattie and Jones (1992), 20% of the graphs are concluded to be materially distorted. The high number of firms distorting their graphs in combination with firms’ intention to emphasize good performance, leads the authors to conclude that graphs included in annual reports are actively being manipulated. This harms the neutrality that is considered to be so important in financial reporting by the Financial Accounting Standards Boards.
In their 1997 study, Beattie & Jones compare the use of graphs between 91 major UK and 85 leading US companies in their 1990 financial reports. In this study they find that US companies include graphs more often and include a higher number of them than UK companies do. Their data show that 92% of US companies include graphs compared to 80% of UK companies, the numbers on average being 13.0 and 7.7 respectively. The aggregate variables that were graphed most often, being sales, an absolute earnings measure, earnings per share (EPS) and dividends per share (DPS), were the same for both countries. It should however be noted that sales was graphed many more times by US than by UK companies, which means that US companies seemingly consider sales to be more important to signal to
annual report users. Earnings, on the contrary, was relatively considered to be more important by UK companies, which may mean that these companies focus more on short-term earnings compared to a long-term growth perspective of US companies. In this study, again, the authors find that graphs provide the possibility for management to manipulate the financial signal they sent to readers. The level of distortion is found to be significantly greater for US companies (with graphs being 16% distorted) than for UK companies (7% distortion).
In 2001, Beattie and Jones take their research a step further and compare the use of graphs in annual reports by firms in the US, UK, Australia, France, Germany and the Netherlands. They find that the percentage of companies using graphs doesn’t differ much, but there is a difference in the variables graphed. There are six main performance variables that were graphed very often (as opposed to the four identified in the authors’ 1997 study), namely sales, earnings, DPS, EPS, Return on Capital Employed (ROCE) and cash flow. Dutch and French companies were the only ones sticking out on graphing cash flows, while, roughly, only US companies graph ROCE.
A significant difference is found between micro-based and macro-based countries. Australia, the Netherlands, the UK and the US are considered to be micro-Anglo Saxon where France and Germany are considered to be macro-continental accounting practices, as classified by Nair and Frank (1980) and corroborated by Nobes (1983). This result is important since macro-based countries are considered to be very different from micro-based countries. Countries with micro-based accounting practices, as opposed to macro-based ones, which show opposite characteristics, typically have weak governmental influence on accounting, relatively strong accounting professions and comparatively active equity markets. The different results between these groups, however, was driven most by German companies, since these are most different from the others, typically graphing solely sales. Furthermore, Beattie and Jones claim that the differences between micro and macro-based companies have seemingly grown smaller in the years prior to their research. It is unclear from prior literature what the current status is.
Frownfelter-Lohrke and Fulkerson (2001), like Beattie and Jones, investigate the nature and extent of the use of graphs and whether or not there is compliance with guidelines for good use set in previous literature (Taylor and Anderson, 1986), but then for 270 US firms as opposed to an equal number of non-US firms. They do this by a matched pair study, where every US firm that has been investigated, is matched and compared to a non-US firm.
Frownfelter-Lohrke and Fulkerson’s study shows that readers of annual reports rely on aggregate information as presented in graphs and that they’re unable to correct for distortions of data in graphical charts on the short time. This is important to note, as their study also found that financial
graphs are distorted to the large extent of 81% for US companies and even 173% for US, non-Canadian companies. According to Arunachalam, Pei & Steinbart (1999), distortions larger than 5% are material and those larger than 100% can actually have effect on readers’ decision-making. Beattie and Jones (2002), state that to avoid distorting users’ perceptions, graphs shouldn’t be distorted over 10%. The large distortions, according to Frownfelter-Lohrke and Fulkerson, are possible due to companies’ non-compliance with many of the guidelines for good use of graphs set forth in prior literature. This non-compliance, in its turn, is possible because its only guidelines provided, not regulations.
As can be seen above, the use of graphs has been investigated quite a bit in the past years. However, it has never been empirically investigated whether profitable companies include more graphs than loss-making companies do.
2.5. Hypotheses
As Lothian (1976) found, graphs reflect management’s perception upon importance of information since graphs are very often an attractive and accessibly formatted repeat of information that is either tabulated or written elsewhere in the report. It is likely that management considers it important to repeat profitable numbers and good performance in the annual report and not to repeat numbers on bad performance, like losses. Concerning this, there are two likely scenarios. The first is that companies could try to hide bad performance by drastically increasing the number of graphs they use in annual reports, and then specifically graphs on other subjects than financials so that the financial graphs are overshadowed by other types of graphs. The other scenario is that companies do the exact opposite: decreasing the total number of graphs used in order of attracting as little attention as possible to the bad numbers. Since the direction of the hypothesised relationship between profitability and the number of graphs used is unclear, H1 will be formulated in the form of a null hypothesis. This results in the following hypothesis:
H1: Loss-making companies use the same number of graphs as profitable companies.
As such, if proof is found that the number of graphs in annual reports is different for loss-making and profitable companies, H1 will be rejected. The direction of the difference found, i.e. loss-making companies use more graphs or loss-making companies use fewer graphs, will then naturally determine the direction of the alternative hypothesis that is accepted. Concerning financial graphs specifically, it’s expected that companies will decrease the number they use when making a loss, so as to attract as little attention as possible to the bad numbers. Therefore, concerning financial graphs, I hypothesize the following:
H2: Loss-making companies use fewer financial graphs than profitable companies.
3. Research Methodology
This paragraphs will firstly introduce details about the sample. For instance about what sort of firms are included in it, what caused firms to be excluded from it and how data were collected. After that, the models used for the research will be introduced and the variables used in them will be explained.
3.1. Sample
The sample consists of firms listed in the Netherlands (on the AEX, AMX and AScX indices) on January 1st, 2015. The sample period is 2004-2013, over which a time series analysis will be conducted for multiple firms in the mentioned indices, which leads to panel data. Only firms that make at least one loss and one profit over the sample period will be taken into account. Every firm that has fused, merged, or split will be removed from the sample in order to keep observations comparable to one another. Firms that have merely changed their name but kept their business the comparable to the situation before the name change will be kept in the sample. Other reasons for excluding firms are unavailability of one or multiple annual reports, not including graphs in annual reports at all, only publishing standardized Form 20-F annual reports. Why such firms are excluded will be further explained in paragraph 4.1. Behr AEX, AMX and AScX (2015) provides data on which firms are listed on the respective indices.
Previous literature nearly always looks at the use of graphs by companies in the US and the UK. Since the US and UK have been investigated quite a lot on their use of graphs, be it that different aspects have been investigated, these countries won’t be the focus of this research in order of adding as much to existing literature as possible. As Nair and Frank (1980) classified national accounting practices, they characterized the Netherlands to be in the same category as the US and the UK, namely as micro-Anglo Saxon practices as opposed to France and Germany being macro-continental accounting practices (corroborated by Nobes, 1983). Due to the Netherlands’ accounting practice being of a similar nature to that in the US and UK, it is a logical choice to investigate this country since research on the Netherlands’ reporting is more likely to be comparable with existing literature than for example research on Germany’s reporting would be. Beattie and Jones (2001) support this, as can be read in the literature review: they compare the use of graphs in annual reports between the US, the UK, Australia, France, Germany and the Netherlands. Evidently, there is enough prior information on the use of graphs within the Netherlands, which makes it even more suitable as a subject of this research.
Accounting narratives are becoming increasingly important in the annual report (Clatworthy and Jones, 2003) and this section is viewed to be a crucial element for achieving reporting quality (Beattie, McInnes and Fearnley, 2004 and Clatworthy and Jones, 2003). Both Courtis (1998) and Aerts (2005)
conclude that the accounting narratives aren’t neutral and are used as impression management, and Aerts additionally states that both the content and the form of the narratives should be fully taken into account. The accounting narratives section of the annual report is used to provide readers with (background) information about the company and its performance. Naturally, most of the graphs that are included in the annual report, are therefore to be found in the narrative section. For those reasons, the section of the annual report that will be researched in this thesis is the narrative section, defined as the whole section between the cover and the financial statements section.
The data necessary for the regression will be hand collected from the aforementioned narratives. For every company, for every year observation, the number of graphs that companies include in the narrative section will be counted and classified into the categories ‘financial’ and ‘other’. When financials are segmented and just the distribution of the item, not the height of the amount is displayed (for instance graphing revenue distribution over the parts of the world) the graphs will be classified under ‘other’ instead of financials. The reason for this is that in such a case the reader’s attention isn’t really being attracted to the financial number since it hasn’t been presented. In the revenue example, attraction would be directed more to what parts of the world or what countries are important for the company’s business instead of to the height of the revenue. Apart from collecting data on the number of graphs, the profitability of the company will also be determined per year.
3.2. Research Design
A regression model will be used to assess whether companies’ use of graphs is different when they make a profit compared to when they make a loss. The dependent variable is the number of graphs and the independent variable is the economic situation (profit/loss), which will be a Dummy variable. The main regression model that will be used for testing is as follows:
Yi,t = ai,t + b1i,t * D_Prof + ti + λt + εi,t ; where: Y i,t = number of graphs in the narrative section
a i,t = constant b1i,t = effect of profit
D_Prof = Dummy variable for profit; value equals 1 if a company is profitable and 0 otherwise ti = time fixed effect
λt = company fixed effect ε = error term
The model is a mixed model that will be tested in SPSS Statistics 22. Additionally, to test the results separate from this mixed model, a group mean centered independent samples t-test will be executed. This makes sure the findings won’t just be found because of the way the model is built and
will also be found separate from it, i.e. using a different model. In group mean centering, the companies’ yearly observations will be deducted from their own mean number of graphs used over the sample period in order to eliminate differences between companies’ general graphical preferences. One company for instance includes an average of ten graphs per annual report whereas another may prefer to include a number of around thirty every year. When differences between this standard ‘level’ are filtered out of the test, solely the remaining differences occurring, are tested against the influence of profitability.
In the mixed model, control variables are irrelevant since correction for both differences over time and between companies are built in through the differences-in-differences method and no other control variables seem necessary. When using the independent samples t-test, it would normally make sense to include a control variable for the trend in the use of graphs over time. Doing this, however, approaches the mixed model and would not provide us with valuable new information. Therefore, no control variables will be included in this model, it will just be a test of whether the results hold up in a simpler model than the mixed one. Consequently, the model for the t-test will be the simplified Yi = ai,t + b1i,t * D_Prof + εi,t, where all variables have the same meaning as in the mixed model.
Since this research is essentially about finding out whether companies try to shift their focus away from financials when they make a loss and as made evident in the hypotheses, additional tests will be executed on solely financial graphs rather than all graphs in the narrative section. Just like for the influence on total number of graphs, the influence of profitability on the use of financial graphs will be tested with both the mixed model and the separate t-test. The mixed model is Yi,t = ai,t + b1i,t * D_Prof + ti + λt + εi,t and the t-test model will be Yi,t = ai,t + b1i,t * D_Prof + εi,t. In both these models, the dependent variable (Yi,t) will be the number of financial graphs in the narrative section rather than the total number of graphs, and the other variables will be the same as in the previous regression models.
For the mixed models, the ‘subject’ is specified as ‘companies’ so that the measurements per company will be grouped together and intercompany habits regarding the use of graphs can be eliminated from the influence of profitability. D_Prof is a fixed effect and the year of time will be random. The latter means that, in general, what year in time it is essentially doesn’t influence the use of graphs, however, per company there is a different effect of what year it is on its use of graphs. Concerning the topic of this study, this makes sense because of the specific yearly circumstances a company encounters. Above assumptions provide the model with the best Akaike’s Information Criterion (AIC) and Schwarz’s Bayesian Criterion (BIC) where the validity of the results can be ascertained. The absolute values of the AIC and BIC don’t have any interpretive meaning but a decline of 2 is good evidence for a better balance between complexity and good fit of the model (Seltman,
2008). The mixed models with the assumptions above, show a decline of over 2 on both the AIC and BIC compared to any other assumed mixed model on the data in question and are therefore most suitable for analysis of the data.
As for the somewhat rigorous judgement of profitability as a dichotomous variable in this research rather than seeing it as a continuous variable, one might argue that a very small profit isn’t necessarily a very good achievement when the company normally makes large profits and therefore wonder whether management will still want to emphasize even that slight profit. This distinction is mainly made due to prior literature showing the large importance that is placed on the difference for companies between making a very small loss and a very small profit (Burgstahler and Dichev, 1997). Degeorge, Patel and Zeckhauser (1999) even state that reporting a profit as opposed to a loss is the most important reason for firms to engage in earnings management. Burgstahler and Dichev (1997) additionally report that the difference between companies disclosing either a very small loss or a very small profit has a huge effect on the share price and every slightly profitable year is considered as a ‘good’ one by shareholders.
4. Findings
This paragraph includes descriptive statistics on the different variables used in the regression. These are specified separately for the different years in the sample and also for the state of profitability a firm is in: profitable versus loss-making. Of course, aggregate descriptives will be shown as well. After the descriptive statistics, validity of the data will be discussed, followed by the actual results of the regression models. The different regression models used will subsequently be compared, just like the difference in results between the regressions on all the graphs in the narrative section and those on solely the financial graphs. The paragraph concludes with a discussion, in which possible explanations for the results will be sought, noteworthy observations from data gathering will be discussed, limitations are introduced and topics for future research are suggested.
4.1. Descriptive Statistics
The AEX, AMX and AScX indices each list 25 companies, which brings the original sample size before eliminations to a total of 75 companies. Of these 75 companies, there are 51 that make at least one loss and at least one profit over the sample period. Of these 51, there were 13 where one or multiple full annual reports were missing that, therefore, couldn’t be included in the research. Furthermore, 4 companies were eliminated due to having one or multiple reports only available in the standardized Form 20-F that requires including only certain graphs and 5 due to not including graphs in their reports at all. The reason for this is that in these situations, even though not including any graphs can be a choice
just like including a certain number of graphs is, it’s very hard to judge whether this actually is a choice or not since the reports not showing any graphs are often standardized reports, just like Form 20-F reports are. Another 6 companies were eliminated for taking part in a fusion, merger or split off for the reason that this harms comparability between the companies’ own annual reports over the sample period. Finally, 3 companies were eliminated due to having an outlier outside the range of 1.5 times the interquartile range of the sample before outlier elimination. The final sample includes 20 companies with ten year-observations per company which brings us to a total of 200 year-observations. The included and eliminated companies are presented in appendix 1 with, if applicable, the reason for elimination included.
The mean number of graphs observed per annual report is 14.46, with the lowest mean per year being 11.85 in 2013 and the highest being 16.95 in 2011. Up until 2008, the mean number of graphs saw a yearly rise. In subsequent years there was some fluctuation before the mean rose to its aforementioned mean peak in 2011, after which it decreased to its lowest point in, as said, 2013. The lowest number of graphs observed in an annual report was zero and the highest was 46.0. Of the company years observed, 146 are profitable and 54 are loss-making. Contrary to what one may logically expect it’s not 2008, the year that indicates the start of the financial crisis, in which most losses are made but the year 2012, with 14 companies making a loss. In the years 2005 through 2007 all 20 companies in the sample made a profit. All mentioned statistics and more details per year are to be found in Table 1. The mean for the Dummy on profitability is 0.73, with a standard deviation of 0.445 (neither has been tabulated). This indicates that 73% of the observed company years is a profitable year, which corresponds to the 146 profitable years observed.
TABLE 1
Descriptives on total number of graphs per annual report - per year
Year N Mean Std.dev. Minimum Maximum
Number of profitable companies Number of loss-making companies 2004 20 12.25 10.213 0.0 38.0 17 3 2005 20 13.80 9.639 0.0 39.0 20 0 2006 20 14.10 9.910 1.0 37.0 20 0 2007 20 15.60 11.038 1.0 41.0 20 0 2008 20 15.75 10.676 1.0 41.0 11 9 2009 20 15.50 12.551 1.0 46.0 12 8 2010 20 15.90 11.397 1.0 45.0 18 2 2011 20 16.95 11.732 1.0 45.0 10 10 2012 20 12.95 9.367 1.0 32.0 6 14 2013 20 11.85 7.741 1.0 32.0 12 8 All 200 14.46 10.397 0.0 46.0 146 54
Table 2 summarizes descriptives per economic situation: profit versus loss. It shows us that profitable firms see a mean of 14.19 graphs per annual report whereas 15.20 is measured for loss-making companies, which is a difference of 1.01. Both groups see a minimum of zero and the maximum only differs by one, with 45.0 graphs being the highest observed value for profitable years and 46.0 for loss-making ones.
TABLE 2
Descriptives on total number of graphs per annual report - per economic situation
Economic
situation N Mean Std.dev. Minimum Maximum
Profitable 146 14.19 10.026 0.0 45.0
Loss-making 54 15.20 11.406 0.0 46.0
All 200 14.46 10.397 0.0 46.0
After group mean centering data, the group mean is zero by definition. When separated per year, naturally the lowest and highest means occur in the same years as in the non-group mean centered
sample, respectively 2013 and 2011. In 2013, the mean is -2.62 indicating that on average over the sample, companies deviate -2.62 from their personal ten year average. In 2011, the mean is 2.49, indicating a similar relationship, except that it’s 2.49 graphs that companies use more rather than 2.62 fewer. The largest negative company deviation from its own average is 24.3. This occurred in both 2012 and 2013 and means the respective company used 24.3 graphs fewer than it did on average over the ten year sample period. The largest positive deviation is 17.6, occurring in 2010, meaning the respective company used 17.6 graphs more in that year than it did on average over the sample. All mentioned statistics and more details per year are to be found in Table 3.
TABLE 3
Descriptives on group mean centered number of graphs - per year
Year N Mean Std.dev. Minimum Maximum
Number of profitable companies Number of loss-making companies 2004 20 -2.22 7.524 -21.7 10.3 17 3 2005 20 -0.67 6.441 -14.7 8.3 20 0 2006 20 -0.37 6.278 -14.7 10.3 20 0 2007 20 1.14 6.107 -8.2 15.3 20 0 2008 20 1.29 5.849 -7.1 15.3 11 9 2009 20 1.04 7.548 -11.3 15.9 12 8 2010 20 1.44 8.463 -12.7 17.6 18 2 2011 20 2.49 7.073 -6.7 16.9 10 10 2012 20 -1.52 8.930 -24.3 16.0 6 14 2013 20 -2.62 7.415 -24.3 10.3 12 8 All 200 0.00 7.251 -24.3 17.6 146 54
Table 4 summarizes descriptives on the group mean centered data per economic situation: profit versus loss. Profitable firm years see a mean of 0.07, indicating that when making a profit, firms positively deviate 0.07 graphs from their company average. In loss-making years, on the contrary, on average companies use 0.18 graphs fewer than their personal ten year average. Both groups see a minimum of -23.4, profitable firms see a maximum of 17.6 and loss-making firms of 16.0. This means that the largest deviation from a firm’s personal average that occurred, was 23.4. This was a negative deviation: a number of 23.4 graphs below firm average was used, both by a profitable and by a loss-making
company. The values of 17.6 and 16.0 represent the highest positive deviations: 17.6 and 16.0 graphs more were used than the company averages by a profitable and a loss-making firm respectively.
TABLE 4
Descriptives on group mean centered number of graphs - per economic situation
Economic
situation N Mean Std.dev. Minimum Maximum
Profitable 146 0.07 6.886 -24.3 17.6
Loss-making 54 -0.18 8.223 -24.3 16.0
All 200 0.0000 7.251 -24.3 17.6
As mentioned before, Tables 2 and 4 specify the descriptive statistics separately for profitable and loss-making years. Both tables show only little differences between the mean values of the two groups, 1.01 regarding the total number of graphs and 0.25 (unrounded: 0.067 minus -0.182, which equals 0.2486, not tabulated) regarding the group mean centered number of graphs to be precise. This already indicates that there’s not a strong relationship between D_Prof and the dependent variable ‘number of graphs per annual report’. The weak scores of a -0.043 Pearson correlation and a -0.032 non-parametric Spearman correlation for the regular data and a 0.015 Pearson and 0.038 Spearman correlation for the group mean centered data, all four of which being far from significant (0.542, 0.652, 0.830 and 0.591 respectively), furthers this into stating that there is no significant correlation between the two.
Table 5 shows a mean of 6.41 financial graphs being included in companies’ annual reports, with the highest mean being 8.00 in 2007, meaning that in this year, companies averagely use 8.00 financial graphs. The highest number of graphs used by one company in this year is 24.00, which is also highest number used by any company in the whole sample period. The lowest number in 2008 is 1, whereas in all other years but 2013, where it is 1 as well, the lowest number is zero. The lowest mean observed, is 5.15 in 2013, meaning that in that year, on average, companies used 5.15 financial graphs in their narratives. Striking is that 2012, the year that sees the lowest maximum of financial graphs used (15.0), is also the year in which most companies make a loss. Unfortunately it’s not very meaningful to state something like that about the highest maximum when there’s only very few profitable firms in a year, since, by coincidence or if the alternative hypothesis is correct, having only a single profitable firm can lead to a very high maximum.
TABLE 5
Descriptives on number of financial graphs per annual report - per year
Year N Mean Std.dev. Minimum Maximum
Number of profitable companies Number of loss-making companies 2004 20 5.80 6.263 0.0 24.0 17 3 2005 20 6.25 5.250 0.0 22.0 20 0 2006 20 6.75 6.214 0.0 22.0 20 0 2007 20 8.00 5.903 0.0 21.0 20 0 2008 20 7.75 5.928 1.0 24.0 11 9 2009 20 6.70 4.975 0.0 16.0 12 8 2010 20 6.25 4.471 0.0 16.0 18 2 2011 20 6.05 4.904 0.0 15.0 10 10 2012 20 5.45 5.145 0.0 15.0 6 14 2013 20 5.15 4.727 1.0 18.0 12 8 All 200 6.41 5.361 0.0 24.0 146 54
With the descriptives found in Table 6, the difference in mean between profitable and loss-making years can be calculated. With a mean of 6.74 for profitable years and 5.54 for loss-making ones, the difference equals 1.20 financial graphs that profitable firms use more than loss-making firms according to the data from the sample. Another interesting calculation is on the maximum number of graphs used for both economic situations. Where the maximum observation and thus the highest number of financial graphs used in an annual report was 16.0 for loss-making firms, that of profitable firms was 24.0. This is no less than 50% higher and provides a minor indication, but of course no proof, that profitable firms use more financial graphs than loss-making firms do.
TABLE 6
Descriptives on number of financial graphs per annual report - per economic situation
Economic
situation N Mean Std.dev. Minimum Maximum
Profitable 146 6.74 5.535 0.0 24.0
Loss-making 54 5.54 4.797 0.0 16.0
Table 7, separating the descriptives for the group mean centered financial graphs per year, naturally shows a mean of 0.00 due to the deviations averaging each other out. The highest mean found is 1.59 in 2007, meaning that in this year, companies averagely use this number of financial graphs more than their personal averages. The lowest mean is -1.27 in 2013, implicating that on average, in this year, companies use 1.27 financial graphs fewer than they do over the sample period. The highest observation was in 2008, when a company used 14.1 financial graphs more than its personal average, the lowest in 2010, when there was a company that used 11.8 fewer than its personal average.
TABLE 7
Descriptives on group mean centered number of financial graphs - per year
Year N Mean Std.dev. Minimum Maximum
Number of profitable companies Number of loss-making companies 2004 20 -0.62 5.234 -10.2 11.2 17 3 2005 20 -0.17 4.241 -10.2 9.2 20 0 2006 20 0.34 4.684 -10.2 9.2 20 0 2007 20 1.59 3.532 -4.2 11.1 20 0 2008 20 1.34 3.799 -3.6 14.1 11 9 2009 20 0.29 4.061 -7.8 9.4 12 8 2010 20 -0.17 4.010 -11.8 5.9 18 2 2011 20 -0.37 2.475 -3.9 4.9 10 10 2012 20 -0.97 3.870 -7.9 6.4 6 14 2013 20 -1.27 3.187 -6.6 7.8 12 8 All 200 0.00 3.982 -11.8 14.1 146 54
Table 8 provides the descriptive statistics for the group mean centered data on financial graphs and shows us that, on average, companies use 0.32 graphs more than their personal average when they are profitable compared to 0.87 fewer when they’re loss-making. It also shows us that the highest deviations from company average occur when companies are profitable. A 11.8 graph deviation is the highest negative deviation from company average by profitable firms, compared to loss-making companies maximally using 10.2 financial graphs fewer than their personal average. The maximum positive
deviation is 14.1, compared to loss-making companies maximally using 6.4 financial graphs more than they, themselves, do on average.
TABLE 8
Descriptives on group mean centered number of financial graphs - per economic situation
Economic
situation N Mean Std.dev. Minimum Maximum
Profitable 146 0.32 4.134 -11.8 14.1
Loss-making 54 -0.87 3.424 -10.2 6.4
All 200 0.00 3.982 -11.8 14.1
Just like the correlations between profitability and the total number of graphs, those between profitability and the number of financial graphs are weak. The Pearson correlation is 0.100 with a 2-tailed significance of 0.159 and the non-parametric Spearman correlation is 0.089 with a 2-2-tailed significance of 0.208 for the original data on financial graphs. This would indicate that the number of financial graphs in a company’s annual report can’t be said to be significantly correlated with profitability. The differences in means between the profitable and loss-making groups have grown bigger compared to the models on all the graphs, though, with a difference of 1.20 for the original number of financial graphs (compared to 1.01 for total graphs; +19%) and 1.20 (unrounded: 1.1923) for the group mean centered data on financial graphs (relative to 0.25 rounded and 0.2486 unrounded for total graphs; both +380%). Furthermore, for the group mean centered data on financial graphs, the Pearson and Spearman correlations are both 0.133 with a 0.06 p-value (significant on the 10% level). Of course, some companies have different graphical preferences or habits concerning how many financial graphs they ‘normally’ use in their annual reports which all influence the correlation. For the group mean centered data on financial graphs, these have been taken into account and filtered out. Judging by the significance (albeit on the 10% level) of the correlations after filtering this out, there essentially is a correlation between a company being profitable or not and the number of financial graphs it uses (relative to the number it generally uses).
4.2. Data Validity
For all four models, Levene’s test for equality of variances is insignificant (0.267 and 0.713 for the mixed models on the number of total and financial graphs respectively and 0.257 and 0.260 for the group mean centered models on the number of total and financial graphs respectively; none tabulated). This means that for all four models equality of variances can be assumed. In other words,
homoscedasticity can be assumed for all. Multicollinearity is out of the question since there is only one covariate, being year. The residuals are approximately normally distributed, with the data on the normality plot being somewhat positively skewed and the mean of both the residuals and the standardized residuals being zero. This holds for both the total number of graphs and the number of financial graphs.
The distribution of the data on the total number of graphs can’t be assumed to be perfectly normal since results on the Kolmogorov-Smirnov test are significant. The normality tests were executed separately for the profitable and the loss-making groups. The non-normality occurred in both groups and is caused by a positive and significant value for skewness, which indicates that data accumulate on the left of the distribution rather than having a perfect bell-shape. For financial graphs, the same situation exists: non-normality occurs in both groups, caused by positive and significant values for skewness.
For the t-test on the group mean centered data, not all data can be assumed to be perfectly normal either. The only category for which normality can be assumed, is for financial graphs in loss-making years since there is no significant score on the Kolmogorov-Smirnov test (p > 0.05). There are two main reasons for the profitable firms deviating from normality in the mean centered data on the financial graphs. The first is that the data are pointy around zero, i.e. there are many observations where there is zero difference between the respective year-observation and the mean. This can be concluded due to a positive and significant score for the kurtosis statistic. The other reason is that there are outliers in the data. Those won’t be deleted for the t-test since the companies and company years observations should be the same in both the mixed model test and the t-test to obtain the best comparable results. For the total number of graphs, there is no pointiness. Here, the deviation from normality by both the profitable and loss-making group is caused solely by the outliers.
None of the deviations from normality mentioned above are problematic since none are severe deviations. This is indicated by the fact that none of the results for skewness and/or kurtosis amount to the (absolute) value of one or higher and that the histogram with normality curve looks regular.
4.3. Results
Results of the mixed model regression can be found in Table 9. The two bigger columns represent the results from respectively the full final sample mixed model and the, more specific, financial graphs mixed model. In the first mixed regression, the coefficient for the Dummy is 1.1661, meaning that the test predicts that firms use this number of graphs more when they make a profit compared to a loss. However, the coefficient has a significance of 0.325 due to which the results for the two groups are, by far, not significantly different from each other. So even though the means in the two groups differ a bit
(1.01 to be precise), it can’t be said that the groups are significantly different and furthermore no difference can be generalized to and thus predicted for firms outside the sample.
In the second mixed model regression, the F-value is 5.279 and the Dummy’s coefficient 1.4307 with a significance of 0.023. This means the second test is significant at the 5% level, resulting in evidence that companies include a different number of financial graphs in their narratives in years they make a loss compared to years in which they make a profit and, therefore, rejection of H0. The 95% confidence interval lies between 0.2017 and 2.6597. This means that, based on this sample where profitable and loss-making companies have a mean difference of 1.20, one can say with 95% certainty that firms use between 0.2017 and 2.6597 financial graphs fewer when they make a loss than they do on average.
TABLE 9
Results of the mixed model regression on the effect of profitability on the number of graphs in annual reports
F Estimate t Sig. F Estimate t Sig.
Intercept 29.234 14.3449 5.784 (<0.0005)***1 23.026 6.9922 5.501 (<0.0005)***1 D_Prof 0.974 1.1661 0.987 (0.325) 5.279 1.4307 2.298 (0.023)** ***/**/* = significant at the 1%/5%/10% level
1 = transformed to the right method of displaying significance
With regard to the t-tests performed in order to corroborate the results found in the mixed models, a check for equality of variances is needed. Due to the insignificant score on the Levene’s test mentioned before (0.257 for total graphs and 0.260 for financial graphs, neither has been tabulated), equal variances in the two groups (profitable and loss-making) can be assumed for both t-tests, which leads to the t-test results tabulated in Table 10.
TABLE 10
T-test results for group mean centered difference in the use of graphs
t-value Mean Profitable value Loss-making Difference Sig. 2-tailed Graphs total 0.215 0.0671 -0.1815 0.2486 0.830 Financial graphs 1.892 0.3219 -0.8704 1.1923 0.060* ***/**/* = significant at the 1%/5%/10% level
For the first t-test (on the mean centered data for total graphs), the results show that there is only a very slight difference in companies’ deviation from their regular use of graphs in years that they make a profit compared to years in which they make a loss (0.2486). Furthermore, the found values are insignificant (0.830) which can’t lead us to conclude that there actually is a significant difference at all. For the second t-test (on financial graphs), a difference of 1.1923 is found between the after-group mean centering data of profitable years versus loss-making years. This means that the difference in ‘within company deviation from the mean’ between profitable and loss-making companies is a lot bigger for financial graphs than for total number of graphs (+380%).
The difference of 1.1923 on the second t-test is significant on the 10% level. Thus, on the basis of the t-test on the group mean centered data, the alternative hypothesis that companies use a different number of financial graphs when they make a loss compared to when they make a profit can be accepted with a likelihood of 90% after rejecting the null hypothesis. Just like in the mixed model, the results indicate that companies indeed deviate from the number of financial graphs they normally use in their annual reports when they make a loss rather than a profit. The term ‘normally’ can be used since making profit is more ‘normal’ than making a loss, considering that all companies in the sample have more profitable years than loss-making years (with the exception of Wessanen which has an equal number of both) and 11 of the companies only have 1 or 2 loss-making years compared to 9 or 8 profitable ones respectively.
The 90% confidence interval of the difference of 1.1923 has a lower boundary of 0.1510 and an upper boundary of 2.2336. This means that on the basis of this last t-test, one can state with 90% certainty that when firms are loss-making rather than profitable, they use an amount between 0.1510 and 2.2336 financial graphs fewer than their own average than they normally would (normal previously having been determined the profitable state).
4.4. Comparison Between the Different Regression Models
For neither the mixed model nor the t-test on the total number of graphs, are the results significant. In other words, no significant relationship is found between a company’s profitability and the total number of graphs used in the narrative section of the annual report. The null hypothesis can’t be rejected and therefore it can’t be concluded that making a loss leads a company to include a different number of graphs in its annual report compared to making a profit. For the tests on financial graphs on the contrary, a statistically significant result is found using both models, although for the t-test it’s only at the 10% level versus the 5% level for the mixed model. The 95% confidence interval resulting from the mixed model predicts loss-making companies to use between 0.2017 and 2.6597 financial graphs fewer than profitable ones, and the 90% confidence interval resulting from the t-test predicts them to use between 0.1510 and 2.2336 fewer.
A mixed model makes use of a differences-in-differences approach, where differences between subjects (companies) and differences over time are filtered out of the results. In the t-test, due to group mean centering the data, differences between companies are filtered out. A reason that the results in the t-test on financial graphs are different (that is, a lower difference and a lower significance) than the ones in the financial graphs mixed model, is that the t-test just looks at the difference in means between the two groups after manually eliminating intercompany differences. No correction for the time trends has been included. Therefore, the t-test shows us that companies deviate from the number of financial graphs they normally use, regardless of any time trends. To include a correction for the time trend in the t-test, would approximate the mixed model itself. If a very prominent time trend was present, that would of course influence the deviations from the mean greatly compared to only having a minor time trend or none at all. Since the mixed model corrects for this, the mixed model is more trustworthy than the t-test. However, still finding the same results with the t-test (albeit with a lower significance), does corroborate the results by showing that they are due to a difference in use of financial graphs actually being present as opposed to the difference resulting from the choice of model.
Another fact that might explain the lower significance of the t-test with group mean centered data on financial graphs, is that in this test the outliers haven’t been excluded. The reason for this is to keep the companies and observations taken into account in the test comparable to the ones taken into account in the mixed model. In my opinion, having data on the same companies and company years makes the results more comparable than eliminating outliers and then comparing the t-test data on a smaller number of companies with the mixed model that then includes some companies that aren’t included in the t-test.
A significant difference being found for financial graphs at all, despite not finding significant Pearson and Spearman correlations, can be contributed to the fact that it’s very common to always use
