Towards A Typology of Conflict Frames

Conflict-framing literature has suffered from a lack of consistent conceptualization and consequently the aim of this paper is to unravel the dimensionality of conflict frames. We conducted a content analysis (N = 1536) to develop a typology of conflict news coverage. Using factor analysis, different aspects of conflict news are distinguished with level of substantiveness and interventionism as the main dimensions. Furthermore, the study focuses on the occurrence of these different types of conflict frames in the news and the role of contextual factors. For example, results indicate that articles with conflict frames contain more interventionism and less strategic framing online, more journalistic intervention in quality newspapers and less substantive news in election times when compared to routine periods. The study contributes to the literature by offering a more nuanced view of the presence of different types of conflict news in the media.


Introduction
Conflict is central to politics, because the essence of democracy is a clash of political ideas (Schattschneider 1975) and a struggle over which is the best option (Sartori 1987). This importance of conflict also translates into news-framing research, and "conflict frames" are one of the most frequently used frames in political communication research today (Vliegenthart, Boomgaarden, and Boumans 2011). Conflict frames are particular in that they "emphasize conflict between individuals, groups, or institutions as a means of capturing audience interest" (Semetko and Valkenburg 2000, 95). Empirically, this definition has been translated to include elements such as disagreement, tension between different sides, incompatibility between viewpoints and politicians attacking each other in the media (Putnam and Shoemaker 2007). However, a review of the literature shows inconsistent operational definitions of conflict framing in content research, as well as differential findings within framing effects research. This promises that conflict frames are multilayered and complex constructs in the news, rather than one single entity. In this paper, we propose that there are two dimensions that are particularly important for conflict framing: (1) the level of substantiveness and (2) the level of journalistic intervention of a conflict frame.
First, conflict news can focus on policy issues and "hard" information (e.g., Hänggli and Kriesi 2010), as opposed to non-substantive, strategic aspects of politics (e.g., Adriaansen, Van Praag, and De Vreese 2010). Conflict news can also be characterized by incivility between politicians, instead of civil and courteous political deliberation (Mutz and Reeves 2005;Forgette and Morris 2006).
Second, conflict frames may vary by the extent to which journalists intervene during the news production process. Previous research shows that journalists have a propensity to intervene in the conflict frame-building process (Bartholomé, Lecheler, and de Vreese 2015). This intervention in conflict framing is exhibited in journalistic speculation concerning consequences of political conflicts, active journalistic framing of conflict news by juxtaposing political actors, and by journalistic interpretations and analysis of political disagreements. Journalists can also be a part of the political conflict themselves. Journalists can also intervene by adding direct commentary within a news report or taking a stance on issues (Hanitzsch 2007). Yet, the extent to which journalistic intervention is visible in actual coverage of conflict frames in political news is unknown.
Hence, this study uses an extensive content analysis to develop a typology of conflict frames, depending on (1) the level of substantiveness and (2) the level of journalistic intervention within the frame. We contribute to the literature by taking the generic definition of conflict framing as a starting point and from this generic definition make more nuanced sub-distinctions of different types of conflict frames.

Understanding Conflict Frames in the News
This paper focuses on news framing (e.g., De Vreese 2005;Matthes 2012). Framing can be explained as a process in which certain aspects and considerations of a political debate are highlighted over others. This entails a selection to be made concerning the information that is shared and certain judgments, decisions and behaviours that are suggested to the individual receiving the message (e.g., Nelson, Clawson, and Oxley 1997;Scheufele 1999). A frame is often defined as a "central organizing idea or storyline that provides meaning to an unfolding strip of events" (Gamson and Modigliani 1987, 143). One of the most commonly used journalistic frames is the conflict frame (Semetko and Valkenburg 2000). Following the traditional definition, a conflict frame reflects disagreement between individuals, institutions or countries (Cappella and Jamieson 1997). In political news, conflict frames are showcased through critique or attacks from and towards political actors, or the visibility of diverging political views in the media. The popularity of conflict frames in the Western media may be explained by the fact that Western journalists greatly value objective reporting (Schudson 2001). Part of this objectivity norm is the routine of balancing between competing positions (Neuman, Just, and Crigler 1992). Journalists juxtapose positions and viewpoints of actors to give a fair and unbiased account of a story (McManus 1994). Because balancing competing voices often entails the inclusion of critique or attacks on the other actors, or on the viewpoints of the other actors, this media routine contributes to the emergence of conflict frames in the media.
The generic definition of conflict frames as used in most studies has not led to accumulative research, but has allowed researchers to operationalize the concept of conflict frames in different ways in their studies. Conflict frames have been associated with a number of negative outcomes on citizens. For instance, public approval of politicians decreases after exposure to conflict-laden television coverage (Forgette and Morris 2006). Furthermore, studies have found conflict frames to contribute to the erosion of political trust (Mutz and Reeves 2005). However, there are studies that stress the democratic potential of conflict frames, which is connected to the idea that, if conflict is present, citizens are more likely to realise the importance of political decision-making (De Vreese and Tobiasen 2007;Schuck et al. 2013). Conflict frames are also found to cause an increase in support for certain policies (Vliegenthart and Schuck 2008), and a positive impact on turnout (Min 2004). At closer look, it seems that previous effect studies have actually taken different types of conflict into account, all published under the label of conflict news or conflict framing (Min 2004;Lee, McLeod, and Shah 2008). These studies show that exposure to different types of conflict frames leads to different outcomes on citizens.
Surprisingly, most content analyses towards conflict frames fail to take into account different types of conflict frames and look at the more general definition of conflict framing (Semetko and Valkenburg 2000) or conflict and co-operation (Vliegenthart, Boomgaarden, and Boumans 2011). To summarize, the wide array of conceptualizations and thus operationalizations of conflict frames makes it hard to generalize results and compare studies on this subject. Also, different variations of conflict frames have led to different outcomes on citizens. This calls for a clear and coherent conceptualization of conflict framing that takes into account the dimensionality of the concept. Furthermore, the occurrence of these different types of questions in actual news content is of interest. Certain types of conflict frames can be more prevalent than others. Hence, we pose the following overarching research questions for this paper.

RQ1:
To what extent can different types of conflict frames be distinguished?
RQ2: To what extent are these different types present in conflict news coverage?

Distinguishing Different Types of Conflict Frames
We use previous research on conflict framing to make a distinction between different types of conflict frames. We specifically focus on two possible dimensions of conflict frames: interventionism and substantiveness. The concept of interventionism alludes to the level of journalistic intervention in a news message (Strömbäck and Dimitrova 2011). This aspect of conflict framing alludes to the level in which the journalistic voice is visible in conflict frames in the media. In contrast, non-interventionist conflict frames allude to a more detached and factual description of political conflict as it unfolds between political or societal actors. Furthermore, we also distinguish between substantive and non-substantive conflict frames. Substantive conflict frames focus on issues and ideas mainly, whereas nonsubtantive conflict frames focus on concepts such as strategy (Gross and Brewer 2007) and incivility (Mutz and Reeves 2005). Using substantiveness and interventionism as a subdimension allows us to incorporate different theoretical concepts that are often used under the umbrella of conlict framing in one more comprehensive theoretical framework.

Conflict Frames Depend on Journalistic Interventionism
First, we consider the level of journalistic intervention in a conflict frame. The importance of conflict as a news value makes it attractive for politicians to put forward conflicting viewpoints to gain media attention. Conflict frames that emerge can be the product of frames put forward by political or societal actors. However, conflict frames in the news are also a result of active journalistic intervention in the frame-building process TOWARDS A TYPOLOGY OF CONFLICT FRAMES 1691 (Bartholomé, Lecheler, and de Vreese 2015). Journalistic intervention reflects the other dimension of conflict frames proposed in this study. Interventionism, or "the media's discretionary power", refers to the degree to which the media take a formative role in shaping the agenda of election campaigns (Semetko et al. 1991, 3). To consider the involvement of journalists lies at the core of determining who shapes conflict coverage.
Two aspects of interventionism are of importance for conflict framing. Firstly, interventionism determines the degree to which journalists are visible in a news item (Strömbäck and Dimitrova 2011). Non-interventionist reporting resonates with a descriptive style, which offers a detached and factual description of events. In contrast, an interpretative style is focused on meaning of events that goes beyond facts and statements of sources (Falasca 2014), as well as journalistic explanations and analysis (Salgado and Strömbäck 2011). Journalistic visibility entails the adaptation of a more interpretative style of reporting (Hanitzsch 2007) and "journalists reporting about political news in their own words, scenarios and assessments" (Esser 2008, 403). This interpretative style has become an important characteristic of political news reporting, both in the United States and in Europe (Iyengar and McGrady 2007;Strömbäck and Dimitrova 2011).
Secondly, journalists can perform their watchdog role and evaluate or attack politicians themselves (Weaver and Wilhoit 1996). Here, interventionism is interpreted as the extent to which journalists pursue particular missions and values instead of functioning as a detached, objective observer (Hanitzsch et al. 2011). This aspect of interventionism differs from journalistic intervention in the form of interpretation, because it includes a reference to the journalist's explicit personal perspective. In contrast, an interpretative style entails the explanation of a fact, without necessarily including a value judgment (Benson and Hallin 2007). This dimension closely resembles the "watchdog role" (Weaver and Wilhoit 1996). The journalist functions as an adversarial to politicians and organizations. In summary, political conflicts in the news can be characterized by a detached and more neutral style or a more interventionist style, with more journalistic interpretation and evaluations of politicians involved in the conflict.

Conflict Frames Vary in Their Substantiveness
Asides from interventionism, this study proposes that conflict frames themselves can differ in their level of substantiveness. Some studies described conflict frames as being part of the game frame or horse-race coverage (e.g., Hänggli and Kriesi 2010;Takens et al. 2013). Conflict frames also have been operationalized as fundamentally non-substantive (Gross and Brewer 2007). In this study, we depart from the theoretical notion that conflict frames are necessarily part of a more process-oriented approach or a game frame. Rather, conflict frames can also be substantive and focus on the issues or the ideological clash around which the disagreement revolves (Lawrence 2000). Hence, this study differentiates specifically between substantive conflict frames that focus mainly on political ideas, policy issues, ideological issues and values, and non-substantive conflict frames that address mainly the political process, politics as a game or personal attacks.
The relevance of such a distinction in conflict framing lies in potential outcomes on citizens. Outside the context of conflict framing, scholars have expressed the concern that strategic news as such becomes more prevalent, often at the cost of substantive news (e.g., Patterson 1993;Brants and Van Praag 2006). Furthermore, the use of this strategic game frame is associated with eroding political trust and an increase in political cynicism (Cappella and Jamieson 1997;Jackson 2011). For example, Elenbaas and De Vreese (2008) found that strategic framing positively affected political cynicism. In contrast, exposure to issue frames can reduce political cynicism among young voters (Adriaansen, Van Praag, and De Vreese 2010). Differences in effects between substantive and non-substantive news content have also been found in the context of conflict frames. Findings by Min (2004) indicate that while conflict news focused on policy items increased turnout, conflicts consisting of personal attacks depressed likelihood of turnout. Similarly, Lee, McLeod, and Shah (2008) found that when compared to substantive news, non-substantive news on controversial issues discounts partisan affiliation as a primary consideration. Thus, differences in substantiveness can be used to characterize conflict frames in political news and these differences can be important for outcomes on citizens.

What Determines the Occurrence of Different Conflict Frames?
Any framing process is influenced by the context in which it takes place. Substantiveness and level of journalistic interventionism in conflict frames is likely influenced by both (1) the type of media outlet that produces such frames, as well as (2) the political context within which a conflict frame emerges.

The Impact of Contextual Features on Interventionism
Regarding media type, a first distinction is conflict framing in online versus offline news media. Previous research showed that online journalists value a disseminator role more than a watchdog or interpretative role, as compared to journalists from traditional media (Cassidy 2005). This is arguably a result of the importance of speed of publication in online media, which affects news production processes (Singer 2005). The speed in which online journalists often have to operate likely causes a less-active and interventionist approach in the conflict frame-building process. A lack of time may refrain journalists from being visible and play a formative role in reporting on conflict news. Previous research outside the Netherlands suggests that online news contains generally fewer interpretative news items (Salgado et al. 2016). Therefore, we hypothesize that the level of interventionism in conflict frames is lower in online news than in newspapers.
H1a: Online news articles containing the conflict frame contain a lower level of interventionism than print newspaper articles.
A second factor that likely affects the presence of different types of conflict frames in the media are simply which type of news outlet a journalist works in (Schuck et al. 2013). Different news organizations have different organizational goals and different target audiences. In this study, a comparison will be made between tabloid newspapers and more quality newspapers. Previous research suggests that an interpretative style is more prevalent in quality titles (Falasca 2014). Quality outlets may want to give readers more than just the news and adapt an active approach, evaluating politicians and provide readers with interpretations and backgrounds. Therefore, we hypothesize that quality newspapers contain higher levels of interventionism as opposed to tabloid newspapers.
H1b: News articles containing the conflict frame have a higher level of interventionism in quality news outlets when compared to tabloid newspapers.

TOWARDS A TYPOLOGY OF CONFLICT FRAMES 1693
The final contextual factor that will affect interventionism in conflict news relates to whether a frame emerges during an election context versus during a non-election context. Research towards framing is generally focused on election times (Van Aelst and De Swert 2009). Routine periods of news are often overlooked. However, agenda-setting studies have shown that political news reporting will differ in election periods as opposed to routine news periods (Walgrave and Van Aelst 2006). It is known that time factors can explain variations in media interventionism in news coverage (Esser 2008). However, it is not yet clear how these differences play out in the case of conflict frames. Therefore, differing levels of interventionism during different time periods will be treated as an open empirical question.
RQ3: Is the level of interventionism in articles containing the conflict frame higher in election times when compared to non-election times?
The Impact of Contextual Features on Level of Substantiveness Some researchers suggest that the higher commercialized online environment increases an audience-centred approach associated with increased focus on sensationalism and entertainment (e.g., Fenton 2010). However, on the other hand, research suggests that there are strong similarities between news content online and offline, because journalistic norms are stable, even when the medium changes (e.g., Mitchelstein and Boczkowski 2009). This inconclusiveness in the literature, paired with the lack of research on content-specific differences between online and traditional news, problematizes predictions on whether online news includes more or less substantive conflict frames than print news. Previous studies do suggest that online news articles are shorter than print news (Maier 2010). This may indicate there is less room for elaborations on policy issues. However, we formulate an open research question to investigate whether online or offline news will be characterized by more or less substantive news coverage.
RQ4: Are conflict frames in online news more or less substantive than in print news?
Quality newspapers may differ in their level of substantiveness in conflict framing when compared to tabloid newspapers. Tabloids generally have a greater economic motivation; they are inclined to present news in a format that has a significant entertainment and interest value, sometimes at the cost of civic or educational value (Jackson 2011). Earlier research suggested that quality newspapers tend to focus more on substantive news than tabloid newspapers (Norris 2000;Strömbäck and Van Aelst 2010). Therefore, the expectation is that conflict-laden news elicits higher levels of substantiveness in quality news outlets as opposed to tabloid newspapers.
H2a: Articles framed in terms of conflict have higher levels of substantiveness in quality newspapers as opposed to tabloid newspapers.
Coverage of political conflicts in election times might be more focused on non-substantive features such as personal attacks, attacks on competences and strategic aspects. Indeed, outside the context of conflict framing, previous studies indicate that strategic and game framing are mostly important during election times (Aalberg, Strömbäck, and De Vreese 2012). Furthermore, Binderkrantz and Green-Pedersen (2009) indicate that during election times, political news is more focused on processes than on issues. Horserace frames are also more often used when an election is approaching (Vliegenthart and Van Aelst 2010). For the specific case of conflict framing, these findings suggest that in the build-up towards an election, with candidates struggling for votes and rising stakes, the emphasis on politics as a game, strategy and personal attacks are more prevalent than in routine times when there is no election on the horizon. Hence, in this paper it is expected that during election times, political conflict in the news is less substantive than during routine times. This leads to the following hypothesis: H2b: Levels of substantiveness in articles containing the conflict frame are lower in election times when compared to non-election times.

Sample
To test these hypotheses and research questions, we conducted a content analysis with news articles as the unit of analysis. We assessed the occurrence of conflict frames in political news in the four largest Dutch newspapers, including two quality newspapers (De Volkskrant and NRC Handelsblad) and two tabloid newspapers (De Telegraaf and Algemeen Dagblad). Furthermore, online news from the four largest online news websites in terms of readership was included in the sample: the online versions of one quality newspaper (Volkskrant.nl) and one tabloid (Telegraaf.nl), as well as the news website of the Dutch public broadcasting (NOS.nl) and the major Dutch commercial news website (Nu.nl).
We used automated content analysis for sampling purposes; only those political news articles that included conflict frames were selected. For this purpose, we used an automated content analysis method based on supervised machine learning (Burscher et al. 2014). Two yes/no questions are used in this method to determine whether an article entails conflict frame. The first question is: Does the item reflect disagreement between parties, individuals, groups or countries? The second is: Does the item refer to two sides or more than two sides of the problem? When one of these indicators was positive, the article was included in sample. These items are often used to measure conflict frames in the news and originate from the work of Semetko and Valkenburg (2000). Subsequently, five coders were trained to collect data from the eventual sample of conflict frame articles.
Fourteen random news days were selected during a routine news period where no extraordinary political event occurred. The days included in these weeks were randomly selected from the Dutch parliamentary year of 2014-2015. Constructing a random constructed week as a sample is considered an adequate method for news sampling (Hester and Dougall 2007). To ensure a sample that is as varied as possible, 14 days leading up to the elections period of 2015 were also sampled. In total, the sample consisted of 1536 news articles. These included 1042

Measures
In order to measure and validate the dimensionality of conflict framing in political news, a total of 14 yes/no questions were included in the codebook. Several training sessions were organised to ensure the five coders could reliably code the items included in TOWARDS A TYPOLOGY OF CONFLICT FRAMES 1695 this study. This approach resulted in indicators that investigate the presence of the concepts measured. A reliability check was also conducted for this purpose: each coder coded a subset of 100 articles randomly selected from the full sample.
Interventionism. Interventionist reporting style was assessed using two main indicators. First, we assessed the concept of journalistic visibility. We coded actor type for the first three actors in each news article. The presence of one or more journalistic actors among these three actors indicated journalistic visibility. An actor was coded as journalistic when a journalist or journalistic medium was specifically mentioned or when the writer of the article gave a clear evaluation of an actor involved in the political conflict in the article. Second, a measure was included to assess interpretative journalism. Coders indicated whether journalistic style of the article was predominantly descriptive or predominantly interpretative (Salgado and Strömbäck 2012). Third, two items were included to tap adversarial journalism or the watchdog role in journalism. The first item measured whether a journalistic evaluation of a politician was present in the article (yes/no). For the second item, coders indicated whether the writer or another journalistic actor expresses doubts regarding the competence of political actors (yes/no). Overall, the average intercoder reliability for these variables was 0.70 (Krippendorf's alpha).
Substantiveness. To measure the extent to which the conflict frames in news articles were focused on substantive issues or not, three items were included, based on the work of Cappella and Jamieson (1997). These items assessed whether the article focused predominantly on politically substantive issues (yes/no), laws and policies (yes/no), and the presence of substantive views on policy issues by political actors (yes/no). Nonsubstantiveness was measured by assessing the presence of the strategic game frame elements in conflict news coverage, specifically sub-dimensions of the strategic game frame. These include two items tapping strategic framing (e.g., Is the article focused on political strategy?). Also included are three items assessing a focus on politics as a game (e.g., Does the story deal with politicians winning and losing?). In order to assess personal attacks and incivility, three items were included (e.g., Does the article include attacks on personality and style?). These items were based on Aalberg, Strömbäck, and De Vreese (2012). Finally, the concept of incivility was measured by using two items: insulting language and belittling language. These items were based on Sobieraj and Berry (2011). The average intercoder reliability for the substantiveness items was acceptable at 0.67 (Krippendorf's alpha). All item wordings and item-specific reliability rates can be found in Table 1.

Results
We analyse the data using the following steps. First, we aim to examine to what extent different types of conflict frames can be distinguished by conducting an exploratory factor analysis (EFA). Second, we further corroborate the factor structure by conducting a confirmatory factor analysis (CFA) using structural equation modelling (SEM). We opted for this approach, because using EFA as a precursor for a CFA has been shown to be a good method to validate a factor structure (Gerbing and Hamilton 1996). Third, we look at the occurrence of the different types of conflict frames in the general sample. Finally, we look at the impact of contextual factors using ordinary least squares (OLS) regressions.

Exploratory Factor Analysis
In the first part of the Results section, we focus on RQ1. We want to assess to what extent different types of conflict frames can be distinguished. Hence, we first conducted a principal component analysis with varimax rotation to corroborate the proposed factor dimension. 1 As seen in Table 2, the items form five separate factors that generally follow the structure of the dimensions that were expected beforehand. A total of 66.02 per cent of variance is explained by these factors. All items have a factor loading of higher than 0.50, which is usually considered to be an adequate threshold by statistical researchers (e.g., Hair et al. 1998). The first factor that emerges in the factor analysis is the interventionism dimension. This factor explains 24.39 per cent of the variation. Together, the items in the interventionism dimension have a Cronbach's alpha of 0.79 (four items). The second factor that emerges is the substance factor, which explains 12.9 per cent of the variance. The three items of the substance dimension have a Cronbach's alpha of 0.76 (three items). The three next factors represent the three sub-dimensions of non-substantive conflict coverage. The first of these factors is personal attacks. This factor explains 11.47 per cent of the variance (three items). Within this dimension, the items have a Cronbach's alpha of 0.7. The second of these non-substantive factors is strategy. This factor explains 9.78 per cent of the variance (three items). These items have a Cronbach's alpha of 0.54 2 and a correlation of r = 0.28 (p < 0.001). Finally, the incivility factor explains 7.41 per cent of the variation (two items). The Cronbach's alpha for these items is 0.7 with a correlation of r = 0.55 (p < 0.001). These results indicate that, initially, five different dimensions of conflict framing can be distinguished.

Confirmatory Factor Analysis
A CFA was deployed to further validate the factor structure. This was done by conducting SEM. 3 Usually, the chi-square statistic is used to test SEM models with significance indicating bad model fit. However, a drawback of using chi-square is the sensitivity that large samples nearly always lead to model rejection (Jöreskog and Sörbom 1993). Since the sample that will be the subject of analysis in this study is relatively large (N = 1536), three alternative fit indices will be assessed to determine model fit: the Root Mean Square Error of Approximation (RMSEA), the Comparative Fit Index (CFI) and the Standardized Root Mean Square Residual (RSMR).
The model fit of the factor structure depicted in Figure 1 was implemented in the initial model. As expected, the values for the chi-square were significant, χ 2 (88) = 1119.13, p < 0.0001. However, this is unsurprising given the large sample size. Nonetheless, the relevant statistics indicate the hypothesized factor structure did not fit the data well: RMSEA = 0.092, where a value of 0.08 or lower indicates good fit; the SRMR also indicated poor model fit (SRMR = 0.06), here a value of 0.05 or lower indicates acceptable fit; finally, the CFI is too low as well (CFI = 0.837).
In the next part of the SEM analysis, the standardized residual covariances and modification indices of the model were assessed to see if adjustments could be made for the model. Error terms with high modification indices were covaried in the new model. This was only done for error terms that load on the same latent sub-dimension (Schreiber et al. 2006). Furthermore, modification indices indicated a problem with the C4B variable, with generally high modification indices on other items. Hence, this variable, which also showed low factor loading, was removed. This variable tapped attacks on the competence of politicians. The remaining two items of the personal attacks dimension deal with style, personality and performance. After this, the model was again tested for model fit. For the new model, the chi-square was again significant: χ 2 (65) = 529.4, p < 0.0001. After these modifications, the model yielded a much better fit to the data (CFI = 0.929, RMSEA = 0.07, SRMR = 0.04). Compared to traditional cut-off criteria, all indicators but the chisquare generated an acceptable goodness-of-fit.
To compare the initial and modified models further, we also used the Akaike information criterion (AIC) and the Bayesian information criterion (BIC). The absolute value of the models is irrelevant. However, if values of these criteria lower after model adjustment, this indicates an improvement in model fit. For the first model the AIC and the BIC have values of 20,368.3 and 20,677.69, respectively (see Figure 1). For the second model, depicted in Figure 2, these values are 17,936.99 and 18,225.05, respectively. This indicates that the model fit has improved greatly. In sum, the results from the CFA confirm the factor structure found in the EFA. We can thus answer RQ1 by arguing that different types of conflict frames can be distinguished. Specifically, we find five dimensions of conflict framing, namely interventionism, substance, personal attacks, strategic game and incivility.

Level of Interventionism in Conflict Articles
Next, to answer RQ2, we assess to what extent interventionism is present in conflict frame news coverage. For this purpose, the scores of the individual items in each of the dimensions were combined into a multi-item scale variable. The values of these scales range from 0.00 (dimension not present) to 1.00 (dimension present). Hence, the closer to 1.00 the score of these scale variables is, the more indicators of that particular dimension are present in that news articles.
The first main dimension that can be identified for conflict news is interventionism. In Table 3, the mean values of the multi-item interventionism scale variable are shown per outlet. The results show that the level of interventionism in conflict news is highest in the quality newspaper Volkskrant, both in the online (0.27) and print version (0.29). The other quality newspaper NRC also shows high levels of interventionism (0.22). Surprisingly, NOS.nl, which was considered a more quality and sober online outlet, has a low mean on  journalistic intervention (0.08). Furthermore, articles on online websites generally have lower levels of interventionism than traditional newspaper articles, with the exception of Volkskrant.nl which has higher levels of interventionism than the other websites, including the website of the public broadcasters. As shown in Table 3, the highest level of interventionism is found in both the newspaper and the website of Volkskrant. The lowest level of intervention is on the online news site Nu.nl. For our research question, this means that level of journalistic intervention is highest in the print newspapers, with the quality titles showing generally higher levels of interventionism. Furthermore, interventionism is mostly higher in print titles when compared to online titles.

Level of Substantiveness
In Table 4, the mean values of the multi-item scale variables for the substantiveness subdimensions are shown per outlet. The results indicate that the highest level of substance in conflict news articles can be found on the news site of the Dutch public broadcaster: NOS.nl. With an average of 0.67, this is the highest mean score where the other outlets range from 0.51 to 0.63. Hence, political conflicts reported upon by the NOS.nl generally include more information on policy issues than political conflict news in other outlets. When looking at the personal attacks dimension, a clear difference between online and traditional outlets appears. Generally, online outlets focus more on personal attacks than traditional media. In online titles, the mean score ranges from 0.52 to 0.59, whereas in traditional media the means range from 0.37 to 0.44. These results seem to indicate that conflict news in online media outlets is more likely to include personal attacks than conflict news in traditional media outlets. On the strategy dimension, the highest average is found for the newspaper version of Volkskrant, with an average of 0.44, followed closely by the traditional newspaper NRC with 0.32 and Telegraaf which has a mean value of 0.45. The incivility dimension was the least frequent of all these dimensions. Both in online and traditional news, conflict coverage was hardly characterized by framing conflicts in terms of games and winners and losers. Values range from 0.10 to 0.18. Most incivility was found in the newspaper version of Telegraaf (0.18), followed closely by Volkskrant.nl with a mean score of 0.17.

The Effects of Contextual Factors
In the final part of the analysis, we conducted OLS regressions on the separate dimensions to examine to what extent contextual factors influence the variation in two main dimensions of conflict coverage. These factors include medium type (newspaper versus online), time period (election coverage versus routine coverage) and type of outlet ("tabloid" versus "quality").
Effects of contextual factors on interventionism. First, we look at the interventionist level. Table 5 displays a series of OLS regressions performed for the multi-item scale variable that represents journalistic interventionism found in conflict frame news items. First, the results  indicate that in online news articles, a significantly lower degree of online journalistic interventionism is found. Hence, the first hypothesis (H1a) is corroborated. In online coverage of political conflicts, journalists intervene on a lesser scale than in traditional newspaper coverage of political conflicts. H1b is also supported. Results show that the level of journalistic interventionism is significantly lower for more tabloid media outlets. This suggests that in quality newspapers, journalistic intervention in the form of evaluations and an interpretative style is present more often than in tabloid newspapers.
Finally, the results indicate that in election time, the degree of journalistic interpretation is not significantly more visible. Journalists did not become more visible in conflict news in election coverage, when compared to routine coverage. Hence, regarding RQ3, it can be concluded that there is no difference in journalistic intervention when comparing routine coverage with election coverage.
Effects of contextual factors on substance. The next part of the analysis focuses on the effect of contextual factors on the level of substantiveness. As seen in Table 6, results indicate that during election times news content is significantly less substantive. This is in line with H2a. In routine coverage, the news is significantly more focused on substance than in election coverage. Furthermore, election conflict coverage differs from routine conflict coverage in the sense that there are more personal attacks, more strategic coverage and more depiction of politics as a game. When looking at the factors internal to media, results indicate that tabloid newspapers are less likely to include substantive content. However, contrary to expectations, the strategy and personal attacks dimensions are significantly more prevalent in more quality newspapers.
Outlet type had a significant effect on the level of substance in conflict news. However, the results indicate lower degrees of strategy reporting in populist newspapers. Hence, H2b can only be partially confirmed.
Medium type also did not influence the level of substance across all subdimensions. Nevertheless, higher degrees of strategy reporting were found for online news articles. Furthermore, the strong effect of medium on the prevalence of personal attacks deserves attention. Personal attacks are more prominently featured in online news than in traditional news.

Discussion
The aim of this article was to develop a more comprehensive and fine-grained understanding of conflict framing in political news coverage. Our results indicate that articles framed in terms of conflict indeed vary along a number of dimensions. The scale developed to measure interventionism was confirmed as a dimension. Second, level of substantiveness was measured by looking at level of substance in the form of policy news and a focus on legislation. Three dimensions reflected non-substantive news coverage: strategic coverage, personal attacks and incivility. Furthermore, during election times, articles with conflict frames were found to be less substantive, and deal more with strategy, personal attacks and incivility when compared to a routine news period. Second, interventionist reporting, substance, strategy and personal attacks were lower for sensationalist newspapers when compared to quality newspapers. Finally, online articles showed lower levels of interventionism, substance and strategy when compared to print articles. However, the level of personal attacks was higher for online titles.
Framing theory has long been associated with unclear operationalizations and definitions (Scheufele and Tewksbury 2007). The criticism aimed towards framing mainly stems from the fuzziness of framing conceptualization and incoherent applications, illustrated in this paper by the example of conflict framing (Matthes 2009). However, rather than abandoning the concept of framing, this paper suggests that an alternative solution would be to strengthen research towards emphasis frames. This entails (1) providing clearer and more consistent definitions, and (2) unravelling underlying dimensions which can explain variations within generic frames. The findings in this paper point to the relevance of such variations for the specific case of conflict framing and apply pre-existing theories to identify and measure these underlying dimensions. Previous studies define conflict framing by definition as strategic (e.g., Gross and Brewer 2007), characterized by incivility (Mutz and Reeves 2005;Forgette and Morris 2006), or make the distinction between attacks on style and integrity or personal attacks and attacks based on issues (e.g., Ansolabehere and Iyengar 1995;Hänggli and Kriesi 2010). This paper aimed to combine these elements. Furthermore, we adhered to earlier claims in the literature that conflicts can also be substantive (e.g., Lawrence 2000;Min 2004), and we included interventionism, which is a sub-dimension of conflict that has not been taken into account before. The theoretical framework we propose can be used to distinguish between types of conflict framing within political news that adhere to this more general definition. This definition can then be used for cross-country comparisons and cumulative research based on consistent conceptualizations. Focusing on two important aspects of conflict framing, our study provides a good starting point for future studies. Nonetheless, there may be more characteristics of conflict that may be of importance. For example, what if a conflict frame in the news focuses on a one-sided attack? Two sides of an issue can come forward in a news article that gives one point of view or political actor considerably more room to attack or discredit the political opponent, or to bring forward arguments for his or her side of the issue. Another example could be emotionalization. The extent to which conflict coverage is characterized by emotional language may be another way to distinguish different types of conflict frames.
Our results also show that contextual factors internal to the media, such as medium type and outlet type, affect conflict coverage. For example, the findings indicated that less journalistic interventionism is more prevalent in online coverage when compared to traditional coverage. This finding may be explained by differing journalistic routines and the emphasis on speed that characterizes online journalism (Singer 2005). Another explanation that can be found is the move towards more quality journalism by traditional media, who seek to differentiate themselves from the large numbers of free content by providing interpretative journalism and performance of the watchdog role; future research should focus on explaining the differences between online and offline content.
The findings also show the impact of election coverage versus routine coverage on the manifestation of substantive news in the media. While there is a lot of research done on election coverage content (e.g., Strömbäck and Kaid 2008) and also on routine periods (e.g., Entman 2004), research comparing these two different time periods is scarce. Earlier research did not find proximity to the elections to be a predictor of generic conflict frames (Schuck et al. 2013). Our study shows that proximity to the election does explain variations of differing types of conflict coverage. This difference between these different time periods is in line with findings from Falasca (2014), which suggest that levels of mediatization vary between periods of election and routine coverage. During election times there are several other factors that might affect contextual characteristics of contentious political news coverage that should be taken into account in future studies. One example of these factors is issue types. Different issues that are present may lead to different characteristics of conflict reporting, due to journalists reporting differently on these issues, but also frames provided by political actors may differ on more emotionally laden issues.
Our study also has a number of limitations. First, the findings of this paper should be seen as a first step towards a better understanding of conflict framing. While the results of the CFA are encouraging, we did make use of modification indices and standardized residual covariances. Follow-up work testing this approach and factor structure is necessary to develop further and confirm the theoretical framework proposed in this study.
Second, we measure journalistic visibility as a proxy for journalistic interventionism. For the purposes of this study, we considered this method to be most reliable as it provides a manifest indicator of journalistic intervention as it is visible to the audience in a media text TOWARDS A TYPOLOGY OF CONFLICT FRAMES 1705 and hence represents distinguishable ways of reporting on political disagreement. However, there are other, more subtle ways in which journalists intervene in the framing processes (Bartholomé, Lecheler, and de Vreese 2015). The journalistic news production process is complex and not easily observed in a content analysis alone. For instance, the differences between offline and online news articles found in this study raise interesting questions regarding the causes of these differences. We can speculate on how to explain differing content in, for example, online versus traditional media, but future research should focus on journalistic processes and differing routines for online and offline journalists. Sometimes it is argued that the lines between online and offline media content are blurry and old journalistic values prevail (O'Sullivan and Heinonen 2008). However, recent work indicates that there are differences in terms of the visual architecture when comparing online news sites and newspapers, and the differences even have effects on readers (Kruikemeier, Boyer, and Lecheler 2016). Our findings indicate that there can also be a difference in characteristics of the news content. An explanation for this may be that the enhanced time pressure resulting from the constant flow of news in online titles could affect the ability of journalists to intervene. Furthermore, the interactive nature of online news which enhances possibilities for the audience in terms of selective exposure might affect reporting by journalists as well, especially since journalists are known to be affected by audience perceptions (Bennett 2005).
The findings of this study are derived from political news in the Netherlands, which is an example of a democratic corporatist media system with a strong history of public broadcasting and a multiparty politics system (Van Aelst et al. 2008). This makes the findings relevant and likely translatable to countries with similar media systems, such as Denmark and Germany (Hallin and Mancini 2004). Nonetheless, future studies should also be conducted in other countries. Particularly in countries with different media systems and different party systems, such as the United States, where, for instance, the tone of conflict coverage might be different from the tone in democratic corporatist countries.

Conclusion
Our findings contribute scholastically by illustrating the different characteristics of conflict coverage in political news coverage. Our typology provides a starting point for a better and more nuanced understanding of conflict framing. Furthermore, the results highlight the importance of distinguishing between different types of conflict frames.
Future research must disentangle whether exposure to these different manifestations of conflict framing in the news affects relevant political behaviour of citizens. This particularly calls for effect studies that take into account different aspects of conflict framing, both in experimental and in panel settings. Nevertheless, conflict framing as a generic concept is still valuable as a framing device that is used in journalism across all reporting contexts. Our results merely add to the concept identifying multiple layers that can be subsumed within the broader concept of conflict framing and that are distinctively present across reporting contexts and media outlets.