In Good Times and Bad: framing of Organization Types During Economic Prosperity and Crises

In Good Times and Bad:

Framing of Organization Types During Economic Prosperity and Crises

Tamara Raats (11356375)

University of Amsterdam Graduate School of Communication Research Master’s Program: Communication Science

Master’s Thesis

Supervised by Dr. A.C. Kroon Date of Completion: 30 January 2019

Abstract

As communication research on how organization types are framed in an economic context is scarce, this study aims to fill this gap. This study looks, firstly, at how organization types differ in how strongly they are associated with frames that depict contrasting economic circumstances and, secondly, whether the economic climate has an effect on to what extent organizations are associated with these frames. Associations are measured by using word embedding models that are trained on sixteen years’ (2000-2015) worth of news articles from five established Dutch newspapers, resulting into a sample of 3,316,494 articles. This study finds that companies have stronger associations with both frames compared to governmental organizations and NGOs. No effects of economic climate were found on the association strengths between organizations and the frames. The results show that organization types differ amongst one another when it comes to what economic context they appear in and that companies dominate the economic discourse in news media compared to GOs and NGOs. Not only does this tell us about how the news media frames different organizations in times of crisis and prosperity, but also that media practitioners might want to adjust their media strategies to increase association with more positive economic frames in the news.

Acknowledgements

First and foremost, I would like to thank my supervisor Dr. Anne Kroon for guiding me throughout the process of creating this thesis. Her enthusiasm has been very inspiring to me

along the way and I am very grateful that she provided me detailed and quick feedback.

I would also like to thank Dr. Jeroen Jonkman, Dr. Damian Trilling and Dr. Rens Vliegenthart for helping me work out some methodological difficulties.

Finally, I would like to thank my parents for always being there for me and supporting me throughout my studies.

In Good Times and Bad: Framing of Organization Types During Economic Prosperity and Crises

Appearances of organizations in news media are very important for organizations and their reputation. Whenever a company, governmental organization (GO) or NGO is

mentioned in the news, elements like the contextual topic, layout and language contribute to the shaping of a frame that is being communicated to the media consumer (Entman, 1993; Cawley, 2012). These frames influence the organization’s reputation. An organization’s reputation not only impacts the way the public perceives the organization, but also how stakeholders react to the organization (Jacobs & Wonneberger, 2017). A change of an organization’s reputation might result into stakeholders adjusting their involvement in the organization or their association to it. News frames can thus be indirectly responsible for the performance of an organization (Schafraad, van Zoonen, & Verhoeven, 2015).

Previous communication research has pointed out differences in news coverage of the organization types of corporations, GOs and NGOs (Jacobs & Wonneberger, 2017;

Wonneberger & Jacobs, 2017). They not only differ in how frequently they appear in mass media, but also in relation to what topics and what tone is being used when they are being discussed. These differences are important to look at, as each organization type has a

distinctive societal role and also interacts differently with the media based on what their aim or cause is (Wonneberger & Jacobs, 2017).

Yet, a lot more examination is required to reveal more differences, especially in regard to framing in relation to a specific context, like the economy. Organizations are often mentioned in economic or financial news (Verhoeven, 2009), and are assigned a certain role that they play in the economic status of a country. The public and private sector, for example, are often put against one another in a frame that assigns responsibility to either one of them during times of a weak economy (Cawley, 2012; Von Scheve, Zink, & Ismer, 2016; Falasca,

2014). It is thus worthwhile to know for communication researchers and media practitioners whether these contexts associate organizations with economic prosperity or economic downturn as this can affect the organization’s trustworthiness, reputation and, ultimately, their performance (Meadows & Meadows, 2016; Schafraad, van Zoonen, & Verhoeven, 2015).

Although some research has presented findings about how certain organizations, such as companies and GOs, have been framed during the most recent economic crisis (Damstra & Vliegenthart, 2016; Cawley, 2012; Rafter, 2014; Manning, 2012; van Scheijen, 2015),

general differences amongst the organization types in times of both economic decline and economic flourishing have not been investigated yet. Furthermore, most of these studies cover only a rather short time period of the crisis and sample only a few newspapers or organizations which, consequently, does not represent the majority of the media climate. This is mainly due to the methods that are used to conduct these content analyses, that often rely on a lot of manual work. Because of looking at smaller timeframes and case studies during the economic crisis, it is unclear to what extent framing of organizations during crisis times differs from framing in times of a stable or growing economy. It is thus difficult to determine how deviating the frames in crisis times are compared to non-crisis economic circumstances. More general claims about organization framing in an economic context are in this case more difficult to make.

A natural language processing method that uses word embeddings allows for research that resembles a big data approach (Mikolov, Corrado, Chen, & Dean, 2013; Mikolov, Yih, & Zweig, 2013). Word embeddings models can handle a large corpus of text (i.e. millions of news articles) and have been used more and more in communication research to expose the usage of certain frames, agendas and biases in human-made content (Kroon, Trilling, & Raats; 2018). This study will use word embeddings models to calculate association strengths

between organizations and the usage of two economic frames in Dutch news media: a frame of economic prosperity and a frame of economic decline. These frames are characterized by terms and words that carry sentiment or a certain attitude often used in economic and

financial news to describe economic changes and circumstances. These frames are selected to research as they are expected to reveal how organizations are being spoken about in an economic context.

This study aims to contribute to organizational communication and framing research, by trying to answer the questions of 1) how organization types differ in how strongly they are associated with frames that depict contrasting economic circumstances, and 2) whether the economic climate has an effect on to what extent organizations are associated with these frames. Additionally, this study will also add to the still scarce section of organizational communication that use NLP and big data-based methodologies, such as word embeddings in particular.

First, general differences amongst the three organization types of companies, GOs and NGOs will be compared by looking at their relationship with an economic prosperity and an economic decline frame by training a word embeddings model on a sixteen-year period’s worth of news articles from five different Dutch newspapers. This time period features times of a good economic climate and that of economic crisis. The second part of the analysis looks at whether Dutch economic climate indicators have an effect on these association strengths by using the same sample of news articles, but now the organizations’ association strengths with each frame are calculated on a yearly level.

Theoretical background Framing

Frames shape the way people interpret the information being given in a text. Entman (1993) discusses the factors of “selection” and “salience” in the process of framing (p. 52).

Some elements of an actual situation, event or issue are chosen to be prominent in a text, to foster an adoption of a certain belief, opinion or evaluation about it. In media, this can be done through the usage or repetition of specific words, using an overall tone throughout the text or through headlines, images or referenced sources (Entman, 1993; Cawley, 2012). Additionally, framing also occurs through absence of specific information (Cawley, 2012). Entman (1993) emphasizes the notion of “power”: “[…] the concept of framing consistently offers a way to describe the power of a communicating text” (p. 51). According to him, research on framing exposes “[...] the precise way in which influences over a human

consciousness is exerted by the transfer (or communication) of information from one location – such as a speech, utterance, news report, or novel – to that consciousness” (p. 51-52). Analyzing certain frames, especially those in news media, thus can help us gain

understanding why and how there are certain belief systems about specific topics present or pervasive in our society.

In regard to news coverage about organizations, framing is important to look at, as some produced belief systems might impact the way stakeholders view the organization and how they choose to be involved with them (Jacobs & Wonneberger, 2017; Wonneberger & Jacobs, 2017). As news media is often regarded as reliable sources of information (Meadows & Meadows, 2016) an organization’s reputation and consequently, performance, is often dependent on the way they are framed in (news) media (Schafraad, van Zoonen, &

Verhoeven, 2015; Jacobs & Wonneberger, 2017). The appearance of certain frames in reports about corporations were found to be successful as a study discovered that the presence of the attributes “corporate vision” and “leadership” in news articles were positively correlated with a company’s financial performance (Kiousis, Popescu, & Mitrook, 2007, p. 161). So

society as a whole accepts it, stakeholders are likely to adjust their association with the organization.

Many organizations, especially corporations and NGOs, thus use (news) media as an instrument to gain some influence on how they will be represented to the public.

Corporations do this by for example acting as information sources themselves, by sending out press releases and doing interviews and press conferences (Meadows & Meadows, 2016; Kiousis, Popescu, & Mitrook, 2007). Furthermore, many public relation departments have strong liaisons with journalists and media companies and their work goes beyond providing information subsidies. Interviewed UK PR practitioners have admitted that they themselves sometimes prepare entire news articles for news outlets, in which they obviously apply frames that promote their company’s agenda (Jackson & Moloney, 2015). NGOs like to receive media attention for the purpose of promoting their organization’s cause and receiving more public awareness for it (Wonneberger & Jacobs, 2017). Lack of organizational

influence and not very coherent organizational structures result often into less coverage than they desire. GOs prefer to keep themselves more on the media background (Deacon & Monk, 2001) and have trouble in general to use PR to their advantage for a more positive reputation as they are political institutions (Wonneberger & Jacobs, 2017). It is, however, important to keep into account that many newspaper articles may contain frames that are not necessarily applied by journalists solely, but that organizations themselves exert influence to a certain degree on how is written about them in the mass media (Boumans, 2018).

Framing of organizations in news media

The scarce previous research that does exist comparing news media coverage of different organization types, mainly provides a general overview of how salient each

organization type is and what tone is used. Looking at Dutch media, Wonneberger and Jacobs (2017) found that companies are covered significantly more often in the news than GOs and

NGOs. In the same study, they report that NGOs receive more positive frames than

companies and GOs while “[…] the coverage of corporations and public organizations was more balanced consisting of negative and positive journalistic evaluations” (Wonneberger and Jacobs, 2017, p. 363). It was expected that due to being in charge of public

responsibilities and expenditure, they would be greatly analyzed in the media. The level of examination, however, was found to be modest. Yet in another study that only discusses GOs, media coverage of GOs was claimed to not be as balanced (Schillemans & Jacobs, 2012). Although most news reports that featured GOs were neutral of tone, GOs appear more often in negative news (40%) than in positive news (14%).

In terms of what kind of topics each organization type is put in context to, it is found that NGOs distinctively differ from corporations and GOs. The articles that companies and GOs are featured in, were more about organizational topics, while those that featured NGOs were more about substantial topics related to the cause of the NGO (Wonneberger & Jacobs, 2017). Verhoeven (2009) presented that in broadcasting news, one third of companies’ appearances was related to business and economics. These findings could be a sign of successful PR efforts by the organizations as the context they are put into correspond to the above-mentioned agendas.

Framing of organizations in news media in an economic context

Surprisingly, positive framing towards corporations seems to be present during times of an economic crisis for many organizations in different countries. Damstra and Vliegenthart (2016) found that media often follow the elitists’ (financial) standpoints as Dutch media coverage of the economic crisis hardly gave any attention towards perspectives that did not reiterated those of neo-liberalists. This study supports pre-existing ideas that the financial private sector mainly dominated the economic discourse during the crisis and that journalists did not act as watchdogs during the economic downturn (Cawley, 2012; Rafter, 2014;

Manning, 2012). Another study that looked at the coverage of the Dutch financial crisis in 2008 found similar results for what information sources were used in the news, but did show that the reliance on a specific group of actors for information differed per newspaper. One of the country’s leading newspapers De Telegraaf primarily used corporate actors as

information sources that spread the message of trusting in the recovery of the Dutch economy (van Scheijen, 2015). Another prominent newspaper NRC Handelsblad relied more on

governmental sources and journalists and focused on economic and financial consequences through analyses.

A study about Irish media reports during the beginning years of the crisis (July 2008 to June 2010) found that that a “division/oppositional”-frame was prevalent in the news media (Cawley, 2012, p. 613). The private sector was in general framed more favorably than the public sector. The public sector was “other[ed]” and not regarded capable of handling the crisis adequately (p. 610). A neo-liberal approach towards bettering the economy was favored amongst journalists: less state involvement and a smaller public sector.

In contrast, the Swedish media actually reversed the roles. A study about how the Swedish news framed the financial crisis from September 2008 until December 2008 (when the crisis started in Sweden), shows that government officials were the main sources for news about the economy for most of this period (Falasca, 2014). This resulted into the media actually framing the government more positively in comparison to banks for the first two phases of the crisis. The crisis was framed by the government as originating from the corporate banking sector. Consequently, evaluative reporting of the government in the news was also more positive during the crisis, while banks were evaluated more negatively. The labor market frame that put the government in a negative light, arising from the opposition, only began to make its way into the news in the last stage of crisis reporting, when the crisis was already fading from the media agenda.

In Germany, newspapers were generally quite negative towards organizations. It was shown that out of individuals, (belief) systems and abstract “network structures”, “collective and corporate actors” were found to be framed to carry the strongest direct deliberate

responsibility for the crisis (Von Scheve, Zink, & Ismer, 2016, p. 645). Although the study unfortunately did not provide a further categorization of these organizational actors, it does show that, overall, organizations were put in a negative economic frame during the crisis.

Although study outcomes do vary to a certain extent, it seems that the most prevalent results point to corporations having the overhand in positive economic framing and GOs being mostly associated with economic decline, at least in the Netherlands, in general and during times of crisis. So as most research indicates that companies are regarded to be more in line with the elitist discourse about their position in the economy and are evaluated more positively in general, which is even emphasized in times of crisis, it is expected that

companies are more brought into the context of economic prosperity than GOs and NGOs.

H1A: Companies are more strongly associated with the economic prosperity frame than GOs and NGOs.

As most research suggests that GOs have in general been receiving quite negative framing and also in economic times do not fit the neo-liberal capitalist discourse, it is expected that that they are more associated with economic decline:

H1B: GOs are more strongly associated with the economic decline frame than companies and NGOs.

Based on research that suggests that companies and GOs are more featured in economic news in general than NGOs and NGOs are mostly brought into context with news that relates to their cause and rather substantial matters, it is expected that association with both frames for NGOs will be lower than for companies and GOs:

H1C: NGOs have an overall lower economic decline frame and economic prosperity frame association strength than companies and GOs.

The last hypotheses focus on the effect of the fluctuating economic climate on organizations’ economic prosperity frame and economic decline association strengths. It is expected that the changes in GDP rate and whether the economy is in a crisis impact how high the

organizations’ association strengths with the frames are. Logically, when the indicators signify a time of economic prosperity, association strengths with the economic prosperity frame are assumed to be higher and those for the economic decline lower. In contrast, when it is a time of economic decline, the predictions are expected to be vice versa.

H2A: GDP rate has a positive effect on the economic prosperity frame association strengths of organizations and has a negative effect on the economic decline frame association strengths of organizations.

H2B: The presence of an economic crisis has a negative effect on the economic prosperity frame association strengths of organizations and a positive effect on the economic decline frame association strengths of organizations

Methods

This section will elaborate on how the data was retrieved with word embeddings by applying the word2vec toolkit from Gensim, how the organizations were sampled and how the words

were selected to represent the two frames of economic prosperity and economic decline. The variables in the final dataset were aggregated using SPSS and hypotheses were tested using Stata. Graphs were created in Word Excel.

Word embeddings

Word2vec is a Natural Language Processing (NLP) tool developed in 2013 (Mikolov, Corrado, Chen, & Dean, 2013; Mikolov, Yih, & Zweig, 2013) in which word embeddings are used to depict an association strength between the meaning of different words. This word embedding cosine similarity score is retrieved by calculating word vector distances in a corpus of text, on which the word embedding algorithm is trained on. The word vector distances are based on the placement of the words within the text corpus. Text corpuses can be social media data, newspaper media, press releases etc., but have to be large enough to be appropriate for training.

In communication science, the method of NLP with word embeddings is continuously used more to conduct sentiment analyses (Rudkowsky et al., 2018), and also to detect biases in texts (Caliskan, Bryson, & Narayanan, 2017; Garg, Schiebinger, Jurafsky, & Zou, 2018; Kroon, Trilling, & Raats, 2018). Especially newspaper media text corpora lend themselves perfectly to detect (covert) biases or frames about specific topics, issues or events. Word embeddings are very suitable for framing research, as the method concentrates on how close or far away words are placed from one another. It can, therefore, pick up subtle nuances in texts and can show to what extent certain words are placed in the same context. So far, no research has been conducted using word embedding vectors to detect framing of particular organizations in news media.

The data and trained word embeddings model for this study have been used

previously in Kroon et al.’s (2018) research about ethnical stereotypes in Dutch news media. For further explanation of the data cleaning procedure and accuracy of the model, please refer

to this paper. The corpus of newspaper texts features five Dutch national broadsheet

newspapers: Algemeen Dagblad, NRC Handelsblad, De Telegraaf, Trouw and de Volkskrant. The time period is from January 2000 to December 2015 (N = 3,316,494). The model was trained on this corpus in two different ways. For hypotheses that concerned overall

differences between organization types, the model was trained on the entire 16-year corpus featuring 562,042 distinct words. This corpus thus includes media coverage during periods of recessions, economic crises but also economic prosperity or stability. To test the hypotheses that measure the influence of economic climate, the entire corpus was split on a yearly level and the model was trained on the texts of each year resulting into a minimum of 108,000 words per year. The word embedding score between every organization’s name and economic frame word will be different for each year, as there will be a different vector distance between the words based on how it has been contextualized in each year by the five newspapers.

Word embeddings work ‘best’ mainly to retrieve words that belong to the same part of speech. In general, the highest association strengths exist between pronouns and pronouns, nouns and nouns, adjectives and adjectives etc. Looking at this study in particular, the words that have the highest association strengths with organizations were found to be,

unsurprisingly, names of other organizations (that often belonged to the same sector). As this does demonstrate the model’s accuracy, it does imply that association strengths with other words that are not organization names are inherently lower. So far, no efficient way has been developed to filter particular words from the corpora. As this study, however, mainly

compares similarity scores amongst organization types, this should not be as much of an issue as the vector distance between two words is expected to remain constant across levels of association strength when using the same trained model.

There were several requirements for an organization to be featured in this study. Firstly; the organization must have existed during the period of January 2000 to December 2015. Secondly; the organization has not undergone a name change during this period. Thirdly; the name of the organization consists of one word or features one word that is regarded to be recognizable or characteristic of the organization (i.e. ABN is chosen for ABN AMRO; and the abbreviations of the GOs). Fourthly; the name is not ambiguous or

synonymous with another organization’s name (i.e. the omitted Ministry of General Affairs is called ‘AZ’ when abbreviated in Dutch, which can potentially be confused with soccer club AZ). The last requirement of an organization to be Dutch is only met for companies and GOs. Originally Dutch NGOs were regarded to be too ‘unknown’ to provide us with media

coverage comparable to the selected companies and NGOs.

The sample of Dutch companies is derived from the Brand Finance Netherlands 50 list which has been published yearly since 2011. Brand Finance (BF) is a consultancy concerned with brand valuation. The list resembles the Elsevier Top 500; which has been used to select a sample of companies in previous corporate marketing research (Wonneberger & Jacobs; 2017). Whereas the Elsevier Top 500 is compiled on the basis of which companies had the highest turnover; the BF list measures brand value; which also incorporates the value of the trademark and how the business is valued in its sector (Brand Finance, 2018). As this study is focused on media coverage and framing of organizations, it was deemed more relevant to look at organizations that are most likely to have the highest coverage in the media. Hence why a list focusing on organizations that have high brand value was used instead of the Elsevier top 500, as the Elsevier list also features companies that are generally unknown, but yet have high position on the list due to a high turnover.

To minimize the possibility that sector differences might influence the results, it was chosen to select three companies per sector from the highest valued sectors. Sectors were

already defined in the BF list; but some sectors were merged as they were thought to provide a comparable product or service (i.e. ‘food’; ‘beers’ and ‘retail’ became the sector

‘food/retail’ and ‘oil & gas’ and ‘chemicals’ became ‘oil/chemicals’). KPN was originally defined by BF as a telecom company, but this can also be considered to be a commercial service; hence why KPN has been labelled to fall under the sector ‘commercial services’.

The selection of NGOs is based on the sample used by Boumans (2016). Five

organizations were added to reach a sample (N = 10) comparable to the number of companies (N = 18) and GOs (N = 10). The final selection of organizations with their respective sector can be found in Table 1 in Appendix A.

Words used for economic prosperity and economic decline frames

As previous research on economic framing of organizations is scarce, especially in ACA research, no consensus exists regarding word lists appropriate for calculating association strengths with organizations. The word lists used to depict the economic prosperity and economic decline frames were compiled by using the Harvard IV-4 dictionary; which has been included in the General Inquirer (GI) text analysis tool. The GI has been used previously in semantics and sentiment analysis research (Devitt & Ahmad; 2007). Although this study in essence does not focus on associations with general positive or negative sentiment, the frames are regarded to have to consist of words related to economic situations that possess an inherent positive or negative meaning.

In a case study that compares different dictionaries for sentiment analysis, the GI is regarded to be representative of American English speech due to containing the frequently employed Brown Corpus and Thorndike and Lorge word lists (Devitt & Ahmad; 2007). Besides the Harvard IV-4 dictionary; the GI features the better known Lasswell dictionary. Each word is tagged in different categories. Not only positive/negative semantic meanings

are noted down, but also whether the word has meanings related to emotions; expressiveness; institutions; social categories etc.

The institution category features two overlapping subcategories; ‘Econ@’ and ‘Econ’; which in total consists of 743 words that are related to economy; business; industry and money. Some of these words are tagged to carry positive or negative meaning. As the GI is in English; all ‘Econ@’ and ‘Econ’ words that were tagged to carry sentiment were translated to Dutch. Some words turned out to have an awkward or ambiguous meaning in Dutch or not be used commonly in the Dutch language. Finally, 12 positively tagged words and 11 negatively tagged words were selected for the framing word lists (Appendix B). To make sure both word lists consist of the same number of words; 8 positive and 9 negative words were added that were regarded to also frequently be used in either positive or negative economic context. Each list consists thus of twenty words.

Variables

Association strength with economic frames. For the first analysis, the association strengths with both the economic prosperity frame and the economic decline frame are calculated by using the word embeddings model that is trained on the corpus comprising all newspaper articles over the entire period of 2000-2015. The word embedding scores constitute the distance between the organization’s name and the economic prosperity frame words or the economic decline frame words (see Appendix B). A cosine similarity score is calculated for each organization and each word in the word lists. Using SPSS, these scores were then aggregated to the level of each organization for each frame. The higher the score, the higher the association strength is with the frame.

The association strengths with both economic frames in the second analysis are trained on a yearly level on the same corpus of newspaper articles for the period 2000-2015. For each frame, the cosine similarity scores are derived from the yearly trained embedding

models between all organizations and the words in the word lists corresponding with each frame after which these scores are aggregated to a yearly level.

Organization type. In the first analysis, three organization types are being compared: 18 companies, 10 GOs and 10 NGOs. In the second analysis that tests the effect of economic climate on the frame association strengths, two dummy variables are created to also look at the effects of the organization types on a yearly level and see whether the results of the first analysis uphold in data that is aggregated to a yearly level. One dummy specifies whether an organization is a company (1) or non-company (0) and the other whether an organization is a GO (1) or non-GO.

Netherlands GDP annual growth rate. The gross domestic product (GDP) annual growth rate in percentages for the period 2000-2015 is displayed in the graph in Figure 1, Appendix C. The data is retrieved from Statistics Netherlands (2018). This growth rate concerns an increase or decrease of the GDP in comparison with the year before and is often used to measure economic growth (M = 1.34, SD = 1.94).

Economic crisis. A dummy variable was created to indicate whether a year was a year of economic crisis: yes (1) or no (0). The years 2001 and 2007-2013 were all regarded to be years in which the Netherlands was found to be in a state of economic crisis, based on the outline of economic events as brought forward in the literature regarding media coverage of the most recent economic crisis (Cawley, 2012; Damstra & Vliegenthart, 2016). This means that half of the years are regarded to be years of economic crisis in the Netherlands.

Analysis

To test hypotheses 1A, 1B and 1C that concern differences amongst the organization types’ association strengths with the economic prosperity frame and the economic decline frame, a one-way ANOVA was performed. To test hypotheses 2A and 2B, two autoregressive distributed lag (ADL) models were performed (each one using a different

economic frame as a dependent variable) as we are dealing with time-series data. The models include all independent variables and a lagged dependent variable. As this study’s data resembles the structure of panel data, the Im, Pesaran and Shin (IPS) test is used for checking whether the data is stationary. This test is specifically designed to detect unit roots in panel data and follows the Dickey-Fuller approach (Hassan, Bakar, & Abdullah, 2014). It is found that all variables are stationary, except for the independent variable economic crisis. In order to make the series completely stationary, this variable would normally have to be differenced. Differencing, however, leads to the loss of a lot of variation, especially as differencing of one variable results into having to difference all variables (including the dependent variable). It is chosen to show the results of the non-differenced model, as the results of the differenced model do not considerably vary, and more information could be kept this way.

Results

Differences between organization type association strengths with economic frames for the entire period of 2000-2015

The one-way ANOVA shows statistically significant differences are present between the organization types (Table 2, Appendix A) in regards to association strength with

economic prosperity: F(2) = 14.18, p<0.01. The post-hoc Bonferroni test showed that companies (M = 0.26, SD = 0.06) have statistically significant higher association strengths than GOs (M = 0.18, SD = 0.08) (p = 0.013) and NGOs (M = 0.13, SD = 0.06) (p<0.01). Hypothesis 1A is thus accepted.

Statistical significant differences also exist between the groups when looking at the association strengths with the economic decline frame: F(2) = 12.85, p<0.01. No difference, however, was between GOs (M = 0.19, SD = 0.08) and companies (M = 0.24, SD = 0.05) (p<0.01). Hypothesis 1B is, consequently, rejected. The statistical differences were between NGOs (M = 0.11, SD = 0.07) and the other two organization types. The Bonferroni test

showed that NGOs had statistically weaker associations with the frame than companies (M = 0.24, SD = 0.05) (p<0.01) and GOs (M = 0.19, SD = 0.08) (p = 0.017). As for the economic prosperity frame (F(2) = 14.18; p<0.01), only one mean difference between NGOs and the other organization types was statistically significant: the one between NGOs and companies as was already mentioned before. The mean difference between NGOs and GOs was not statistically significant (p = 0.214), so we can only partially accept hypothesis 1C.

These findings show that over the period of 2000 until 2015, companies have been put significantly more in a context of economic prosperity than GOs and NGOs. Additionally, NGOs have lower associations with this same context than GOs and have weaker

associations with a frame of economic decline than companies.

Effects of economic climate on organization’s association strengths with economic prosperity and economic decline frames for the period of 2000-2015

Firstly; the over-time framing scores of each type of organization in the period of 2000-2015 will be plotted and analyzed. Looking at the trends of each organization type might already give some insight into what time periods were pivotal for the development of the organizations’ association strengths with each frame. The graphs in Figure 2 and Figure 3 (Appendix D) depict how much each type of organization was associated with the economic prosperity and economic decline frame. Looking at both graphs; one can see that companies have the highest association strengths with both frames from the three types of organizations. It appears that the association strengths with economic prosperity frames are generally higher for companies. These results are in line with the findings from the ANOVA.

Figure 3 shows that the biggest increase of the economic decline framing association strength in the total period of sixteen years for companies took place in the year 2008; when the crisis unfolded. After a decrease in 2011; companies have been increasingly being brought more into context of the economic decline frame in the period 2012-2014; reaching

its highest point yet in 2014. Figure 2 shows us, in contrast, that positive framing association strengths did not fluctuate as much as negative framing strengths during the entire period of 2000-2015, with no sudden peak or plunge before or right after the economic crisis.

Economic decline framing of GOs and NGOs decreased in the period 2007-2009 quite tremendously (Figure 3). It appears that the economic crisis has not resulted into bringing these types of organizations into a context of economic downturn as much. Interesting to note is that the trend of association strength scores with the economic prosperity frame of GO and NGO fluctuate more during the sixteen year-time period than that of company scores (Figure 2). During the economic crisis (2007-2013); a continuous decrease in economic prosperity framing can be noticed for GOs and NGOs, except for 2010, but does not look very deviating from the fluctuating pattern of these organization types before 2007.

Secondly, the ADL model will be evaluated. Table 3 shows the results of the ADL models. The table displays the unstandardized and standardized coefficients, as well as the standard errors for each frame. The significant standardized coefficients of the lagged dependent variable show us that the news of the year before had a significant effect on the news published after, which in essence means that what is published last year influences what is published this year. Additionally, when looking at the dummy organization type variables, it can be seen that companies (in comparison to GOs and NGOs) has a significant positive relationship with the association strengths with both frames. The results of the ANOVA, of which the word embeddings model was trained on the entire sixteen-year corpus, partially support this finding, as it showed that companies’ association strengths with the economic prosperity differed significantly from those of GOs and NGOs. It is logical that more

variation exists in the data in which the association strengths are aggregated to a yearly level than in the data which contains cosine scores that comprise all years.

Table 3

Regression models to predict economic decline frame association strengths and economic

prosperity association strengths over time (N = 570).

Note: * p<.05. ** p<.01. *** p<.001.

Now we turn to the expectations of the effects that the indicators of GDP rate and the presence of an economic crisis had on the organizations’ association strengths with each frame. Hypothesis 2A proposes that GDP rate positively effects economic prosperity frame association strengths and negatively affects the economic decline frame association strengths. No significant relationship of both frame association strength exists. Hypothesis 2A is not accepted.

For hypothesis 2B that suggests the existence of a negative effect of the state of an economic crisis on the organization’s economic prosperity frame association strengths and a

Economic decline frame Economic prosperity frame

B b* SE B b* SE Constant .080 - .009 .074 - .009 Lagged dependent variable .749 .751*** .271 .713 .720*** .028 GDP rate -.0005 -.010 .001 .0004 .008 .001 Economic crisis -.003 -.016 .005 -.0006 -.003 .006 Company .021 .124*** .006 .028 .151*** .006 GO .011 .054 .006 0.003 .016 .007 R2 .619 .620 F 183.54*** 184.17***

positive effect on the economic decline frame association strengths, no significant effects are found either. Hypothesis 2B is also rejected.

Discussion and Conclusions

This study intended to answer the questions of how organization types differ in how strongly they are associated with frames that represent opposing economic circumstances and whether a country’s economic climate influences to what extent organizations are associated with these frames. To do this, association strengths of organizations, that belong to a variety of sectors, with an economic prosperity frame and an economic decline frame have been compared both on a time constant and a time-variant level. The results support partly the phenomena as put forward by previous framing research. This study’s main finding is that companies are found to be more strongly associated with economic prosperity and economic decline than GOs and NGOs. The analysis using data that was trained on the entire time period of 2000-2015 news coverage showed that the differences between companies and GOs was not significant. The analysis that used data with yearly calculated association strength scores, however, showed that companies had a significant positive effect on both frames. Economic climate did not seem to have an effect on organizations’ association strengths with both frames.

As companies are profitable organizations, it was expected that they appear in

economic and financial news more often than GOs and NGOs, as previous research has found much of the news that they appear in is about economics, business and about the

organizations themselves (Verhoeven, 2009; Wonneberger & Jacobs, 2017). NGOs are generally put into context with societal issues, for which NGOs seek the media to raise awareness for (Jacobs & Wonneberger, 2017). The expectation that NGOs therefore had lower association strengths with both the economic prosperity and decline frame was, unexpectedly, only partially met.

For the economic prosperity frame, no substantial difference was found between NGOs’ association strengths and those of GOs. A possible reason why GOs’ association strengths with both frames were lower than expected is that GOs are found to be often featured in neutral-toned news (Schillemans & Jacobs, 2012). Almost one-third of this study’s sampled GOs (Statistics Netherlands, CPB and UWV) perform (statistical) research that is often used as a source for factual information, also in but not only confined to articles discussing the state of the economic climate (Schillemans & Jacobs, 2013). This can lead to appearances in neutral-toned news that might be not about the economy per se (Schillemans & Jacobs, 2012). If this is the case, it is not surprising that association strengths with both frames are quite low as words related to other topics and frames compete with the economic frame terms such as those in this study’s word lists.

The results using data that was trained on a yearly level, showed that companies have a significant effect on both frames, which is surprising. That companies are more put into a context of economic prosperity, also during economic crises, can be explained by the notion that newspaper media follow and perpetuate the elitist discourse that forwards neo-liberal economic ideals (Damstra & Vliegenthart, 2016). Another explanation based on previous literature could be that particularly in times of economic crisis, frames that put companies directly against GOs in the benefit of companies seem to be more frequently used in the news (Cawley, 2012). No effect of GDP rate or the presence of an economic crisis on association strengths with both the economic frames, however, were found.

Grounded in literature, it was thus expected that GOs would be more dominantly put in context with economic decline, but this study showed that companies led this frame as well. Theoretically, we can again turn to the beforementioned argument that companies simply appear more in economic or finance related news. This would, however, indicate that this study’s word lists both represent a more general economic context, and not as much

strongly oppositional positive and negative economic circumstances. Although the frames for this study were intentionally not explicitly meant to be absolute opposites, their

distinctiveness was still based on sentiment. The words in the prosperity frame word list were selected to hint at positive economic terms and those in the decline word list to refer to a negative economic situation. These differences might, however, not has been as evident and both frames might actually point towards a more general economic context. This would be a limitation as the frames may not be as representative of the intended economic

circumstances. More research on language and sentiment in financial, corporate and economic news is needed to develop valid lexicons that are language specific.

Another limitation concerns working with word embeddings: spatial context of only single words is considered when calculating the cosine similarity score. Consequently, some words that in general are regarded to imply economic growth might actually have been used in negative form to indicate economic deterioration (i.e. ‘incomes did not increase’ or ‘no profit was made’). The economic prosperity frame words increase and profit might still have a close spatial placement to an organization’s name but the negative form in which the sentence was written actually indicates an opposite frame. Although this impact might be marginal, these subtle differences might have contributed to companies having the highest associations with both economic frames. Future research could possibly implement the method of word embeddings but then use sentences or multiple words to account for instances of opposite framing.

The last limitations involve the generalizability to other populations by looking at the sampling of the organizations and newspapers. The literature already showed the variety of results in media coverage amongst different countries and different media sources. It is likely that resembling results only can be found in countries in which corporations and neo-liberal ideals dominate the economic discourse. Results might also still differ between news media,

because of their political standing. Perhaps leftist newspapers frame corporations and the banking sector more negatively than rightist newspapers as previous literature already indirectly indicated such framing differences between Dutch newspapers (van Scheijen, 2015). These differences could unfortunately not be detected from this study’s results as the sample of newspapers was quite balanced in terms of political position. Lastly, it was too ambitious for this study to focus on economic framing differences amongst diverse sectors. Sectors were mentioned in this study to make sure the samples of organizations per type were comparable and varied, and that organization types were not overrepresented by one or two particular sectors that could skew the results. Future research could possibly also detect differences in economic framing of organizations between different sectors.

Although it is evident that working with NLP methods that use word embeddings still needs more careful evaluation in order to increase their efficacy and validity for optimal usage in communication science research, this study did show its potential. The findings show that there is certainly more to learn by using word embedding models about how organizational actors are framed in mass media on the scale that this study did. On a

theoretical level, this study did show the pronounced position that companies take in Dutch economic related news compared to governmental institutions and NGOs. Additionally, all organization types seem to be more brought into a context with words that indicate economic prosperity than those that signify economic decline. Times of economic crisis do not seem to substantially alter the usage of each frame for organizations in general.

For media practitioners, these results imply that the media places organizations into frames of economic prosperity and economic decline. Although associations with economic prosperity are higher for all organization types compared to the associations with the

economic decline frame, they could still try to improve their association with positive economic news if they desire to do so, especially GOs. Being put into a context of economic

prosperity is obviously more favorable for an organization’s reputation than being framed with words that connote economic deterioration. As using certain positive attributes in reports can help a company’s performance (Kiousis, Popescu, & Mitrook, 2007), increasing the association with the economic prosperity frame could be done by incorporating words that signify economic prosperity (such as those used in this study) in their press releases or quotes.

Additionally, organizations do not necessarily have to be afraid of being very strongly associated with negative economic circumstances when there is a financial crisis. Although not significant, for companies a slight increase in negative economic coverage can be detected since the crisis started in 2007 (Figure 2, Appendix D), while GOs and NGOs have decreased or continued their pre-crisis association with economic decline. Although the association strengths since 2015 are not known, it seems unlikely that these have gone down to the point of where they were before the recession started in 2007. Media strategists of companies could thus focus on trying to lower associations with economic decline which will return companies to the level that is comparable to that of GOs and NGOs before the

recession.

Overall, this study has shown there are clear differences in how different organization types are framed in an economic context in the Dutch news, taking a big-data approach. Additionally, it has shown how promising word embeddings models can be to detect

contextual linkages in news articles between names of organizations and words that represent a certain frame. The results found in this study indicate that there is more to uncover about how different organization types are framed in economy related news, in times of prosperity and in times of crisis.

References

sources and news agencies in the contemporary news landscape. Brand Finance. (2018). Brand Finance Netherlands 50 2018. Retrieved from

http://brandfinance.com/images/upload/brand_finance_netherlands_50_locked.pdf Caliskan, A., Bryson, J. J., & Narayaran, A. (2017). Semantics derived automatically from

language corpora contain human-like biases. Science,356(6334), 183-186. doi:10.1126/science.aal4230

Cawley, A. (2012). Sharing The Pain Or Shouldering The Burden? Journalism Studies,13(4), 600-615. doi:10.1080/1461670x.2011.629108

Damstra, A., & Vliegenthart, R. (2016). (Un)covering the Economic Crisis? Journalism Studies,19(7), 983-1003. doi:10.1080/1461670x.2016.1246377

Deacon, D., & Monk, W. (2001). ‘New managerialism’ in the news: Media coverage of quangos in Britain. Journal of Public Affairs,1(2), 153-166. doi:10.1002/pa.60

Devitt, A., & Ahmad, A. (2010). The Language of Emotion and Financial News, presented at XVIth European Symposium on Language for Special Purposes (LSP), Hamburg, 2007. Frankfurt am Main: Peter Lang.

Entman, R. M. (1993). Framing: Toward Clarification of a Fractured Paradigm. Journal of Communication,43(4), 51-58. doi:10.1111/j.1460-2466.1993.tb01304.x

Falasca, K. (2014). Framing the Financial Crisis: An unexpected interaction between the government and the press. Observatorio Journal,8(1), 001-021. doi:

10.15847/obsOBS812014752 .

Garg, N., Schiebinger, L., Jurafsky, D., & Zou, J. (2018). Word embeddings quantify 100 years of gender and ethnic stereotypes. Proceedings of the National Academy of Sciences,115(16). doi:10.1073/pnas.1720347115

Retrieved January 24, 2019, from

https://statline.cbs.nl/Statweb/publication/?DM=SLEN&PA=84105eng&D1=0- 70&D2=0,2&D3=29,34,39,44,49,54,59,64,69,74,79,84,89,94,99,104,112-113,115-117&LA=EN&HDR=G1,G2&STB=T&VW=T

Hassan, S., Bakar, N.N.A., & Abdullah, H. (2014). Analysis of FDI Inflows into China from ASEAN-5 Countries: A Panel Cointegration Approach. Journal of Economic

Cooperation and Development,35(3).

Jackson, D., & Moloney, K. (2015). Inside Churnalism. Journalism Studies,17(6), 763-780. doi:10.1080/1461670x.2015.1017597

Jacobs, S., & Wonneberger, A. (2017). Did we make it to the news? Effects of actual and perceived media coverage on media orientations of communication

professionals. Public Relations Review,43(3), 547-559. doi:10.1016/j.pubrev.2017.03.010

Kiousis, S., Popescu, C., & Mitrook, M. (2007). Understanding Influence on Corporate Reputation: An Examination of Public Relations Efforts, Media Coverage, Public Opinion, and Financial Performance From an Agenda-Building and Agenda-Setting Perspective. Journal of Public Relations Research,19(2), 147-165.

doi:10.1080/10627260701290661

Kroon, A.C., Trilling, D., & Raats, T. (2019). Guilty by Association: Using Word

Embeddings to Measure Ethnic Stereotypes in News Coverage. Manuscript submitted for publication.

Manning, P. (2012). Financial journalism, news sources and the banking crisis. Journalism: Theory, Practice & Criticism,14(2), 173-189. doi:10.1177/1464884912448915 Meadows, C. Z., & Meadows, C. W. (2016). Corporate Reputation on Mass Media, Public

Opinion, and YouTube: Examining the Factors Influencing Reputation of Fortune 500 Companies. Corporate Reputation Review,19(4), 345-356. doi:10.1057/s41299-016-0011-2

Mikolov, T., Corrado, G., Chen, K., & Dean, J. (2013). Efficient estimation of word representations in vector space. ArXiv, 1–12.

Mikolov, T., Yih, W., & Zweig, G. (2013). Linguistic Regularities in Continuous Space Word Representations, (June), 746–751.

Rafter, K. (2014). Voices in the crisis: The role of media elites in interpreting Ireland’s banking collapse. European Journal of Communication,29(5), 598-607.

doi:10.1177/0267323114537787

Rudkowsky, E., Haselmayer, M., Wastian, M., Jenny, M., Emrich, Š, & Sedlmair, M. (2018). More than Bags of Words: Sentiment Analysis with Word

Embeddings. Communication Methods and Measures,12(2-3), 140-157. doi:10.1080/19312458.2018.1455817

Schafraad, P., van Zoonen, W., & Verhoeven, P. (2016). The news value of Dutch corporate press releases as a predictor of corporate agenda building power. Public Relations Review,42(3), 451-458. doi:10.1016/j.pubrev.2015.11.014

Scheve, C. V., Zink, V., & Ismer, S. (2016). The Blame Game: Economic Crisis Responsibility, Discourse and Affective Framings. Sociology,50(4), 635-651. doi:10.1177/0038038514545145

Schillemans, T., & Jacobs, S. (2012). Beeldenstrijd. Framing van en door publieke organisaties in de media. Bestuurskunde,4, 21-32.

Schillemans, T., & Jacobs, S. (2013). Autoriteit in beeld. Tijdschrift Voor Toezicht,4(3), 6-25. doi:10.5553/tvt/187987052013004003002

van Scheijen, S. (2015). Financiële crises in de media. Tijdschrift voor Mediageschiedenis, 14(1).

Wonneberger, A., & Jacobs, S. (2017). Media positioning: Comparing organizations’ standing in the news. Corporate Communications: An International Journal,22(3), 354-368. doi:10.1108/ccij-11-2016-0075

Appendix A

Table 1

Organization Types and the Respective Organizations and Sectors.

Companies (N = 18) Governmental Institutions (N=10) NGOs (N=10)

Organization Sector Organization Sector Organization Sector

ING banking BZK domestic policy Consumentenbond consumer protection

Rabobank banking CPB economics Greenpeace environment

ABN AMRO banking OCW education Milieudefensie environment

Aegon insurance SZW employment WWF environment

NN insurance UWV employment KWF health

ASR insurance Rijkswaterstaat environment Unicef humanitarian aid

Shell oil/chemicals Belastingdienst finance Vluchtelingenwerk humanitarian aid

AkzoNobel oil/chemicals IND foreign policy FNV trade unions

DSM oil/chemicals VWS public health CNV trade unions

Philips tech CBS statistic research ANWB travelling

ASML tech NXP tech KPMG commercial services KPN commercial services Randstad commercial services Heineken food/retail Unilever food/retail Ahold food/retail

Table 2

Economic Prosperity Frame Association Strengths per Organization Type trained on Entire Corpus of Newspaper Articles from 2000-2015 (N = 3,316,494).

Organization type Economic prosperity frame association strength

Economic decline frame association strength

Company 0.26 0.23

Governmental institution 0.18 0.19

Appendix B

Economic prosperity frame: arbeidsproductiviteit; *beloning; consumentenvertrouwen; duurzaamheid; *groei; *hoogconjunctuur; investeren; investering; *klandizie; *opbrengst; *productiviteit; *spaargeld; stabiel; *subsidie; *toenemen; *vergoeding; welvaart; *winst; *winstgevend; *zekerheid

Economic decline frame: bezuinigen; bezuinigingen; *boete; *crisis; *depressie; *failliet; *faillissement; *inflatie; *kosten; laagconjunctuur; *onrendabel; *ontslagen; *onzekerheid; *recessie; schuld; schulden; stagnatie; *tekort; *uitgaven; werkloosheid

English translations (includes the original entries from the Harvard IV-4 dictionary): Economic prosperity frame (English translations): work productivity; *reward; consumer confidence; sustainability; *growth; *boom; invest; investment; *patronage; *profit#1; *productivity; *savings; stabile; *subsidy; *accrue; *allowance; prosperity; *profit#2; *profitable; *security

Economic decline frame (English translations): cut back; reductions in expenditures; *fine; *crisis; *depression; *bankrupt; *bankruptcy; *inflation; *cost#1; low economic activity; *unprofitable; layoffs; insecurity *recession; debt; debts; stagnation; *deficit; *cost#2 unemployment

*Entries from the Harvard IV-4 dictionary. Sometimes the dictionary featured multiple versions of the same word which is signified with a ‘#’.

Appendix C

Figure 1 Dutch GDP annual growth rate in percentages (Statistics Netherlands, 2018).

4,2 2,1 0,1 0,3 2 2,2 3,5 3,7 1,7 -3,8 1,4 1,7 -1,1 -0,2 1,4 2,3 -5 -4 -3 -2 -1 0 1 2 3 4 5 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 G DP ann u al g ro w th r ate in % Year

Appendix D

Figure 2 Association strengths with the economic prosperity (‘e.p.’ in graph) frame per type of organization per year between 2000-2015

Figure 3 Association strength with the economic decline (‘e.d.’ in graph) frame per type of organization per year between 2000-2015

0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5 A s s oc ia tio n s tren gth w ith e .p . fram e Year Companies Governmental Organizations NGOs 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5 A s s oc ia tio n s tren gth w ith e .d . fram e Year Companies Governmental Organizations NGOs