Biases in news media as reflected by personal pronouns in evaluative contexts

Original Article

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Biases in News Media as Reflected

by Personal Pronouns in Evaluative Contexts Marie Gustafsson Sendén,1 Torun Lindholm,2 and Sverker Sikström3 1

Stockholm University, Sweden, 2Uppsala University, Sweden, 3Lund University, Sweden

Abstract. This paper examines whether pronouns in news media occurred in evaluative contexts reflecting psychological biases. Contexts of pronouns were measured by computerized semantic analysis. Results showed that self-inclusive personal pronouns (We, I) occurred in more positive contexts than self-exclusive pronouns (He/She, They), reflecting self- and group-serving biases. Contexts of collective versus individual pronouns varied; We occurred in more positive contexts than I, and He/She in more positive contexts than They. The enhancement of collective relative to individual self-inclusive pronouns may reflect that media news is a public rather than private domain. The reversed pattern among self-exclusive pronouns corroborates suggestions that outgroup derogation is most pronounced at the category level. Implications for research on language and social psychology are discussed.

Keywords: pronouns in social categorization, language bias, intergroup bias, self-serving bias, latent semantic analysis, implicit attitudes

‘‘Everything we do as Reuters journalists has to be independent, free from bias, and executed with the utmost integrity.’’

The quote above is the first sentence in the ‘‘Handbook of Journalism’’ published by Reuters (2008). Although the intentions are clear, previous research has shown that the media tend to be biased in reporting about different social categories, such as gender or minority groups (Dixon & Linz, 2000; Geschke, Sassenberg, Ruhrmann, & Sommer, 2010; Koivula, 1999). For example, sports news reproduce gender stereotypes (Koivula, 1999), and minority groups are often presented more negatively than the majority group (Dixon & Linz, 2000). Moreover, using abstract or concrete words in describing behavior of groups and individuals, might reflect and influence evaluative attitudes (Geschke et al., 2010). In this article, we study biases between general social categories as reflected by how personal pronouns (e.g., We, I, He/She, and They) are used in evaluative semantic contexts.

Language can be seen as a vessel for transferring information about intergroup relations (Sutton, 2010), either explicitly or implicitly. Explicit information may involve evaluative adjectives or traits in person or group descriptions (e.g., ‘‘Swedes are beautiful’’). Research also shows how implicit markers of intergroup relations can be expressed by variations in linguistic abstractness (e.g., ‘‘he falls’’ for ingroup members vs. ‘‘he is clumsy’’ for outgroup members; Maass, Salvi, Arcuri, & Semin, 1989; Semin & Fiedler, 1988). Plenty of studies have demon-

2013 Hogrefe Publishing

strated intergroup biases in evaluative judgments and behavioral descriptions (see, e.g., Brewer, 1979, 2007; Mullen, Brown, & Smith, 1992; Tajfel & Turner, 1979, 1986). However, evaluations of people or behavior can be expressed in language by means of other than direct descriptions. For example, describing a specific person in a positive context (e.g., together with beautiful people at a nice party) might implicitly convey a message that this person should be evaluated positively. In a similar vein, we suggest that evaluative differences between social categories can be expressed in language, not only when groups or persons are described or compared, but also as a general bias in the semantic context of social categories. Moreover, the focus in previous studies on intergroup biases in language has typically been on words that denote specific groups or identities (e.g., immigrants, men, women).

However, there are generic words – such as the pronouns (We, They, I, He/She) – that are markers of social categories. In the current study, we extend previous research on linguistic biases, by examining whether pronouns occur in contexts of different valence in patterns that reflect systematic self- and group-serving biases. We do this by using latent semantic analyses (LSA; Landauer, 1998, 1999; Landauer, McNamara, Dennis, & Kintsch, 2007) in a large corpora of media news. The LSA is a multidimensional computerized method that analyzes the meaning of words by how they are used in the context. The LSA is completely data-driven and allows analyses of large amounts of data where manual coding would be highly time-consuming,

Social Psychology 2014; Vol. 45(2):103–111 DOI: 10.1027/1864-9335/a000165

104

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

as well as analyses of subtle differences that would be difficult to detect by human coders.

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Pronouns in Social Categorization Pronouns have previously been used in psychological research as markers for ingroups and outgroups (Perdue, Gurtman, Dovidio, & Tyler, 1990), gender (Lenton, Sedikides, & Bruder, 2009; Twenge, Campbell, & Gentile, 2012), and as indicators of psychological states (Campbell & Pennebaker, 2003; Pennebaker, Mehl, & Niederhoffer 2003). Pennebaker and colleagues argue that personal pronouns are so-called function words that reveal the cognitive structures and mental states of the sender (Chung & Pennebaker, 2007). From an evolutionary perspective, it seems reasonable that personal pronouns evolved as short and efficient markers for important categories in human life (Dunbar, 1993, 1997; Nettle, 1999).

Pennebaker and his group have studied whether frequencies of pronoun use correlate with an individual’s personality, status, or mood (see, e.g., Campbell & Pennebaker, 2003; Pennebaker, 2011; Sexton & Helmreich, 2000), and whether different situations increase the use of one type of pronoun over others (Slatcher, Chung, Pennebaker, & Stone, 2007; Stone & Pennebaker, 2002; Tausczik & Pennebaker, 2010). For example, they found that depression was associated with higher frequencies of I (Campbell & Pennebaker, 2003), and that high status was correlated with increased use of We (Sexton & Helmreich, 2000; Tausczik & Pennebaker, 2010).

Extending these findings, the current study focuses on how social evaluations may be conveyed in language by the semantic contexts of the pronouns. In this research, two dimensions of pronouns are integrated. In the inclusiveness dimension, first personal pronouns (I, We) are categorized as self-inclusive, and third personal pronouns (He, She, They) as self-exclusive. The other dimension reflects the categorization of people as individuals or collectives, such that singular pronouns (I, He, She) correspond to individual levels in social categorization, and plural pronouns (We, They) reflect collective levels (see Table 1). We label this mapping the Pronouns in Social Categorization model (Gustafsson Send n, Lindholm, & Sikstrom, in press), and below we set out hypotheses along these two dimensions.

The Inclusiveness Dimension There is ample evidence of self- and group-serving biases in the literature. People rate themselves more positively than other people, and their own groups more positively than other groups (see, e.g., Brewer, 2007; Cortes, Demoulin, Rodriguez, Rodriguez, & Leyens, 2005; Critcher, Helzer, & Dunning, 2011; Sedikides & Gregg, 2008; Tajfel & Turner, 1986).

Social Psychology 2014; Vol. 45(2):103–111

Table 1. Pronouns in Social Categorization model

Individual (singular pronouns) Collective (plural pronouns)

Self-inclusive

Self-exclusive

(1st personal pronouns)

(3rd personal pronouns)

I

He, She

We

They

Studies on linguistic biases have shown that people tend to use either abstract or concrete words depending on the group involved and the valence of the behavior described (Maass et al., 1989; Semin & Fiedler, 1988). According to the linguistic category model (LCM; Semin & Fiedler, 1988), abstract behavior descriptions lead to inferences about personal dispositions, whereas concrete ones lead to inferences about the situation. Studies on the LIB (Maass et al., 1989) have applied the notion of the psychological implications of abstract and concrete statements, and examined how words are chosen in descriptions of intergroup behaviors. These studies show that positive behaviors by the ingroup more often are described by abstract, traitimplying words (e.g., kind), whereas the same behavior by outgroup members is described by concrete words suggesting situational attributions (e.g., helping). However, in describing negative behaviors, people use concrete words for the ingroup (e.g., slap) and abstract words (e.g., aggressive) for the outgroup (Maass, 1999; Maass et al., 1989; Wigboldus & Douglas, 2007). These asymmetries in abstraction and valence across groups subtly, yet effectively, communicate group-serving biases, such that positive and stable dispositions become associated with the ingroup, and stable negative traits with the outgroup. The LIB has also been demonstrated in news reports. For example, in international sports news, a linguistic bias was found in descriptions of the national team; and in political news, descriptions of intergroup relations reflected linguistic biases in favor of the political position of the newspaper (Maass, 1999; Wigboldus & Douglas, 2007).

While most studies on linguistic biases have focused on verbs and adjectives, recent research has examined the role of nouns referring to social categories in expressions of intergroup biases (Carnaghi et al., 2008; Graf, Bilewicz, Finell, & Geschke, 2012). On the surface, nouns (e.g., ‘‘Kim is a Swede’’) and adjectives (e.g., ‘‘Kim is Swedish’’) seem to convey the same content, but research suggests that nouns have more inductive potential, and more strongly indicate group membership than adjectives (Carnaghi et al., 2008). Studies also demonstrate that using nouns compared to adjectives when describing nationalities enhances stereotypic inferences and ingroup bias (Graf et al., 2012). These findings are particularly pertinent in the context of the current study, since nouns and pronouns are semantically, as well as syntactically, more similar than pronouns and verbs or adjectives. Thus, pronouns (e.g., We)

2013 Hogrefe Publishing

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

and nouns (e.g., Swedes) both denote categories to which people belong rather than characteristics of the categories, and they can be substituted in sentences without changing the syntax (e.g., We can be substituted by Swedes in the sentence We are beautiful). Hence, it could be assumed that as nouns, pronouns may be particularly potent in eliciting group-related biases.

Self-enhancement and ingroup protection have been suggested as underlying motivations behind LIB (Maass, Ceccarelli, & Rudin, 1996; Maass, Milesi, Zabbini, & Stahlberg, 1995). Supporting this notion, studies show that these biases are stronger in social comparison situations (Maass et al., 1995, 1996), when there is coherence in a group (Moscatelli, Albarello, & Rubini, 2008; Rubini, Moscatelli, & Palmonari, 2007), or when a group is threatened (Maass et al., 1996).

In experiments using pronouns as priming stimuli, Perdue and colleagues (1990) showed an ingroup bias for plural but not for individual pronouns. These authors found that participants responded faster to positive trait adjectives after subliminal priming of We in comparison to They. Participants also rated nonsense syllables more positively when previously paired with We than with They (Perdue et al., 1990).

In a recent study, it was demonstrated that individuals selected contexts of difference valence in self-generated messages including pronouns (Gustafsson Send n et al., in press). In a set of experiments, participants constructed three-word sentences including a pronoun, a verb, and an evaluative adjective. The results showed that participants more often selected positive adjectives in combinations with self-inclusive pronouns than with self-exclusive pronouns.

The lack of effect for the individual pronouns in the study by Perdue et al. (1990) may be due to this study using You for this category, which might refer to both individual and collective categories. Hence, in the current study, first and third personal pronouns (I, He/She) are used to assess the individual in terms of the inclusiveness dimension.

In line with the reviewed findings, it is assumed that individuals choose words contributing to positive evaluations of themselves and their own group. Thus, the first hypothesis is that self-inclusive pronouns, both individual (I), and collective (We), will occur in a more positive evaluative context than self-exclusive pronouns (He/She and They) in media reports.

The Individual and Collective Dimension In comparison to the inclusiveness dimension, differences in evaluative contexts along the individual/collective dimension are more difficult to predict. Research shows that most people have a strong tendency to enhance themselves (Critcher et al., 2011; Sedikides & Gregg, 2008), and that the individual self (I) rather than the collective (We) is the primary motivational basis for self-definition (Gaertner, Sedikides, Vevea, & Iuzzini, 2002), at least in Western societies (Gaertner et al., 2012; Markus & Kitayama, 1991; Na & Choi, 2009). While the individual self is usually motivationally primary (Gaertner et al., 2012), there 2013 Hogrefe Publishing

105

are indications that situational aspects might change this.

Research shows that while people prefer to enhance I relative to We in an individual, confidential situation, the reverse is true in public or interpersonal settings (Baumeister & Ilko, 1995; Miller & Schlenker, 1985). For example, when people judge a group performance in a confidential setting, they exaggerate their own importance over the group, whereas in a public setting, the group is boasted about (Baumeister & Ilko, 1995; Miller & Schlenker, 1985). In a recent study, it was also found that when people in a confidential setting were asked to construct sentences including pronouns, they included I in sentences with more positive words than sentences including We, whereas in a collaborative setting, the evaluative context of I and We did not differ (Gustafsson Send n et al., in press). Communicating positively about the ingroup in public settings may be motivated by self-presentational concerns (e.g., to appear unselfish), but may also be used as a means to facilitate the activation of a common ground with the listener (Aronsson & S tterlund-Larsson, 1987; Brown & Levinson, 1987). Based on these findings, together with the fact that media represent a public rather than a private domain, it is tentatively hypothesized that in media reports, the context of We will be at least as positive as the context for I. With regard to evaluative differences in the context of individual and collective self-exclusive pronouns (He/She & They), notions from research on prejudice and social cognition are used to guide our predictions (e.g., Allport, 1954/ 1979; Dovidio et al., 2005; Fiske, Lin, & Neuberg, 1999).

Allport’s contact theory (1954/1979) suggests that contact with individual outgroup members can reduce intergroup hostility and derogation of the outgroup (Miller, 2002).

Studies have also shown that information-processing based on individual characteristics leads to more positive evaluations of a person, compared to processing based on categorical information (Fiske et al., 1999). Hence, research suggests that outgroup derogation is more pronounced at the group compared to the individual level. This leads us to hypothesize that for the self-exclusive level, collective pronouns will occur in more negative contexts than individual pronouns.

Taken together, our second hypothesis is that for individual and collective pronouns, the evaluative contexts will depend on the inclusiveness dimension, such that We will occur in contexts at least equally positive as the context for I, whereas They will appear in more negative contexts than He/She. Our hypotheses will be tested on a large corpus of short media news statement. This corpus will be subjected to computational sentiment analyses. Because the method is rather new, we allocate some space to explain it on a conceptual level.

Method Within studies of artificial intelligence there has been development of mathematical ways to analyze similarities between words and contexts that do not depend on human Social Psychology 2014; Vol. 45(2):103–111

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

106

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

judges (Campbell & Pennebaker, 2003; Landauer, McNamara, Dennis, & Kintsch, 2007; Pitts & Nussbaum, 2006). One such promising technique is LSA (Landauer & Dumais, 1997; Landauer, 1998). The LSA has been suggested both as a theory and a method for how semantic meaning is derived by the contexts in which words are used (Landauer, 1999; Landauer & Dumais, 1997; Landauer et al., 2007). The exact calculations in LSA are beyond the scope of this article, however, LSA relies on singular value decomposition which is akin to factor analyses and multidimensional scaling (Landauer & Dumais, 1997; Landauer et al., 2007). The LSA algorithm takes a huge text corpus from a given language as an input, where the size of this corpus typically is in the magnitude of gigabytes.

The computation of LSA results in a huge matrix, where the column of this matrix represents words in the corpus, and the column the semantic dimensions. Thus, each word in the corpus will be associated with a vector describing a semantic representation of the word. The representation of a word should be understood in relation to all other words in the space, where words with similar representations have similar meaning or synonyms. For detailed descriptions, we refer to methodological articles on LSA (see e.g., Foltz, Kintsch, & Landauer, 1998; Landauer, 1998; Landauer & Dumais, 1997; Landauer et al., 2007), and articles where LSA has been implemented in psychological empirical research, for example in studies on psychological health (Campbell & Pennebaker, 2003), gender stereotypes (Lenton, Sedikides, & Bruder, 2009), changes in self- and object representations (Arvidsson, Sikstrçm, & Werbart, 2011), and happiness (Garcia & Sikstrçm, 2013).

Because LSA measures patterns of word co-occurrences across a vast number of local contexts, it does not only measure first-order co-occurrences between words (i.e., two words occur in the same context) but also higher-order co-occurrences, i.e., matches two words occurring in two similar, but different, contexts (Landauer, 1998; Landauer & Dumais, 1997; Lenton et al., 2009). For example, since words such as doctor and physician are used in the same contexts, the word-vectors for physician and doctor will be located on a proximal distance to each other in the semantic space. Moreover, the meaning of physicians as computed in LSA is represented by a vector with words that are proximal in the space representing synonyms, inflections, and the most common contexts of physicians, for example, doctor, hospital, nurse, surgery, and so forth.

The most common way of using LSA is to measure similarities between words or documents. However, it is also possible to make a semantic evaluation, for example to evaluate the valence of the words constituting the context of pronouns, which is the purpose of this study.

Material To create a semantic space with media news, a corpus with approximately 800,000 thousand (404 MB) Reuters news messages published in 1996 and 1997 was used. When the current project was initiated, this dataset was released Social Psychology 2014; Vol. 45(2):103–111

publically by Reuters, and it was one of the richest samples available for research. Reuters articles are short news messages covering, for example, business, politics, entertainment, and sports (http://www.reuters.com). Two sentences are presented here to show how pronouns might be used in a news context. The first includes We in a positive context about resources or sharing: ‘‘We are Chinese; we share the belief there should be one country and there should be peaceful reunification between China and Taiwan.’’ The word ‘‘peaceful’’ is explicitly positive and used close to the self-inclusive pronouns. This sentence can be contrasted with an example including They in a more negative or ambiguous context: ‘‘The US State Department sent letters to the five Domos executives on Monday saying they and some of their closest family members would be barred from entry to the United States’’ (see e.g., www.reuters.com).

The INFOMAP software (http://www.infomap-nlp. sourceforge.net) was used to perform the LSA algorithm.

The quality of the semantic space (based on how closely synonyms are located in the space) is improved if the words used to build the matrix contain meaningful semantic content. Therefore, about 750 high-frequency, non-content words (e.g., and, but, etc.) were removed before reduction of the matrix (Landauer et al., 2007). From the original dataset the remaining 15,000 most common words were used to build a space with 100 dimensions.

The ambition in this study is to code each pronoun occurrence with a value that represents the averaged valence of the context in which the pronoun occurs. The context of the pronoun was set to a window consisting of 15 words preceding and succeeding the pronoun. Using human coders in such a project would be highly resourceand time-consuming. Moreover, due to the difficulty of achieving valid criteria that could be used by human coders to evaluate media text intended to be objective in nature, one would expect low inter-rater reliability.

To assess valence in the semantic space, we used the Affective Norms for English Words (ANEW; Bradley & Lang, 1999) as a dependent measure. The ANEW word list contains more than one thousand nouns ranked for valence by human participants (on a Likert scale from 1 to 9). Multiple linear regression was used to create a vector across the 100 dimensions in the semantic space that best fitted the valence of the words in the ANEW word list. The quality of this valence prediction depends on the number of dimensions used as predictors. Therefore, we used cross-validating by the leave-one-out-method (Picard & Cook, 1984) to find the best model and decide the number of dimensions. A single observation from the original sample is used as the validation data, and the remaining observations as the training data. This is repeated so that each observation in the sample is used once as the validation data. Using this method, the best model was found for the 85 first dimensions in the space. The correlation between ANEW and LSA was R2 = .48 for the cross-validated sample, and R2 = .62 for the total sample. Differences in how LSA estimates similarities in a text have previously been compared to human inter-rater congruence, with similar results in rating by LSA or by human experts (Landauer, Laham, & 2013 Hogrefe Publishing

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

107

Foltz, 2003; Landauer, Laham, Foltz, Shermis, & Burstein, 2003; Landauer et al., 2007).

Finally, the valence of each pronoun occurrence was assessed by averaging the 15 words preceding and following each pronoun. The means and the standard deviations of the context associated with each pronoun could then be calculated.

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Results The extraction and calculation of valence of the context of the pronouns were computationally extensive. To facilitate the data processing, a subset consisting of 50% (400,000) of the news articles was randomly selected from the original dataset. This subset included approximately 205,000 subjective personal pronouns (NI = 26.863; NWe = 46.340; NHe/She = 90.704; NThey = 41.363). The average valence of all the words in the semantic representation (N = 15,000) was 6.52 (SD = 1.38).

We computed a 2 (self-inclusive/self-exclusive) ¡ 2 (individual/collective) ANOVA with context-valence as the dependent measure. A main effect of the inclusiveness dimension, F(1, 205269) = 4,951.29, p < .001, gp2 = .024, showed that self-inclusive pronouns (M = 6.88, SD = 1.24) were associated with more positive contexts than selfexclusive pronouns (M = 6.63, SD = 1.34). There was also a main effect of the individual/collective dimension, F(1, 205269) = 539.23, p < .001, gp2 = .003, such that individual pronouns (M = 6.66, SD = 1.31) appeared in more positive contexts than collective pronouns (M = 6.60, SD = 1.38). However, this main effect was moderated by a two-way interaction, F(1, 205269) = 1,007.26, p < .001, gp2 = .005. For self-inclusive pronouns, simple effects analysis showed that collective pronouns (M = 6.91, SD = 1.30) were associated with more positive contexts than individual pronouns (M = 6.85, SD = 1.44), F(1, 205269) = 28.98, p < .0001, gp2 = .001. In contrast, among self-exclusive pronouns, the analysis showed that individual pronouns (M = 6.60, SD = 1.35) were associated with more positive contexts than collective ones (M = 6.25, SD = 1.38), F(1, 205269) = 2016.45, p < .0001, gp2 = .01 (see Figure 1).

Finally, t-tests were used to examine whether the pronoun categories differed in valence from the average of all 1,500 words in the space (M = 6.52). The contexts of individual and collective self-inclusive pronouns were significantly more positive than the average valence in the space, MI,Diff = .33, t(26862) = 47,44, p < .001, MWe,Diff = .39, t(46339) = 64.12, p < .001. Among selfexclusive pronouns, the contexts for the individual pronouns were more positive than the mean valence for all words: MHe/She,diff = .08, t(90703) = 18.946, p < .001, whereas the semantic contexts for the self-exclusive, collective pronouns were more negative than the average, MThey,Diff = .27, t(41362) = 39.17, p < .001.

2013 Hogrefe Publishing

Figure 1. The evaluative contexts of the personal pronouns.

Discussion Much theory and research on language use as reflecting social evaluations has focused on traits and descriptions of explicit persons and groups (e.g., Abele & Wojciszke, 2013; Asch, 1946; Fiske, 2012; Maass et al., 1989; Semin & Fiedler, 1988). The current findings provide an important extension of previous studies by demonstrating that language also enables evaluative communication about social categories by subtle variations in the semantic context in which markers of social groups occur. Specifically, this study shows that the semantic context of pronouns in news media varies systematically in valence, in ways that reflect self- and group-serving biases (Mullen et al., 1992; Sedikides & Gregg, 2008). Our analyses were not limited to specific words in the context, nor to any specified groups occurring in these contexts, but included all words surrounding generic markers of social categories. Importantly, these evaluative differences were expressed by intermediary journalists who do not refer to their personal identities in writing the articles, and who, according to directives, should aim at neutral and unbiased language and content.

The evaluative contexts of the self-inclusive relative to exclusive pronouns matched previous research on self- and group-serving biases (e.g., Mullen et al., 1992; Sedikides & Gregg, 2008). For the self-inclusive level, the evaluative context of individual versus collective pronouns was consistent with research on communication in public settings, demonstrating the enhancement of the collective over the individual (Baumeister & Ilko, 1995). At the self-exclusive level, this evaluative pattern in the context of individual versus collective pronouns was reversed, corroborating earlier notions that outgroup derogation is more pronounced at the category level of person perception.

Social Psychology 2014; Vol. 45(2):103â&#x20AC;&#x201C;111

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

108

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

Overall, the pattern of findings validates the two dimensions used in our model of pronouns in social categorization. The combination of interpersonal and intergroup dimensions in this model opens up new areas of potential research on linguistic biases. Thus, rather than investigating either the interpersonal (Carnaghi et al., 2008; Wigboldus, Semin, & Spears, 2000), or the intergroup domain (Graf et al., 2012; Maass et al., 1989), both dimensions could be included simultaneously by using pronouns. For example, it could be tested whether the LIB occurs also in pronoun use, such that positive words around We are more abstract than positive words around They, or whether biases as shown in these research paradigms are stronger for collective than for individual pronouns.

Group-serving biases are often referred to as a consequence of social categorization (Maass et al., 1996; Tajfel & Turner, 1979), group cohesion (Gaertner & Schopler, 1998; Menegatti & Rubini, 2012), or intergroup conflict (Brewer, 1999; Sherif et al., 1961) instigated by a motivation for self-esteem or ingroup protection (Critcher et al., 2011; Maass et al., 1996). As recently noted by Menegatti and Rubini (2013), what has been largely missing in extant research is a communication perspective including both speaker and listener. When communicating, speakers are motived to present themselves in positive light (Goffman, 1959), but also to influence their listeners (Austin, 1962; Lee & Pinker, 2010). For example, communicating We in more positive contexts than They could be a strategy to attract listeners to the speaker’s group and to increase motivation and commitment among ingroup members. Future research should examine whether different motivations to influence the listener or group relations affect the evaluative communication patterns described here.

Pennebaker and colleagues have demonstrated associations between the frequency in people’s pronoun use, and their personality, status, or situation (see e.g., Campbell & Pennebaker, 2003; Stone & Pennebaker, 2002). For example, depressed people showed an overrepresentation of I in communication (Campbell & Pennebaker, 2003) and high status people an overrepresentation of We (Tausczik & Pennebaker, 2010). An interesting extension of this work would be to examine whether groups that have been found to differ in pronoun frequencies (e.g., depressed vs. nondepressed, men vs. women, high vs. low status groups), may also differ in their use of evaluative contexts for pronouns.

Cross-cultural comparisons might also be facilitated by using the pronouns in social categorization model, since most languages have the same categories of pronouns.

One obvious issue in this context would be to test whether individualistic and collective cultures (Markus & Kitayama, 1991; Na & Choi, 2009) show similar or contrasting patterns in the context of individual and collective pronouns.

Limitations Although we found significant differences in the contexts for different pronouns in this research, the effect sizes were Social Psychology 2014; Vol. 45(2):103–111

small (gp2 = .001–.024). Field studies, as compared to laboratory experiments, often have smaller effect sizes (Evans, 1985; McClelland & Judd, 1993), since they include many sources of variance. McClelland and Judd (1993) estimated that field experiments are 80% less efficient than experiments in a laboratory setting. Reuters news includes both positive topics such as sports and business, and negative topics such as war and crime, and possibly, the topics are more important predictors of contextual valence than pronouns are. Hence, large effect sizes should not be expected when using this material. However, people are soaked in news from different media throughout the day, and given this massive exposure and constant repetition, small and subtle evaluative differences are also likely to have implications for how people construe their social world (Graf et al., 2012). A vital issue for future research is to examine whether, and to what extent evaluative differences in semantic context of pronouns actually influence people’s perceptions of, and behavior toward members of different social categories.

When the current project was initiated, this Reuters corpus from 1996–1997 was one of the richest samples available for research. While there are no apparent reasons for assuming that evaluative contexts for pronouns in media have changed in the last decade, future research should replicate the findings in more recent corpora.

Conclusion In the current study, we showed that evaluative differences between social categories are expressed in language as a subtle bias in the semantic context chosen for markers of social categories. Specifically, these biases were conveyed, not in descriptions or comparisons of specified groups, but in the context of generic self-inclusive and self-exclusive pronouns. These biases are likely to go unnoticed, since differences are subtle, and communicated by a source that most people rely on as unbiased. In our study, we examined biases in news media texts, where there is an explicit goal to avoid biases and evaluations in descriptions of people and groups. It seems reasonable that the evaluative pattern in semantic contexts could be more strongly expressed in spoken or written language where no such explicit criteria exist.

The scarcity of language studies within social psychology could partially be due to the lack of available tests measuring evaluations of social groups in language. LSA, and the model of pronouns in social categorization appear to be a promising combination for furthering the development of studies on social psychological biases in language.

References Abele, A. E., & Wojciszke, B. (2013). The Big Two in social judgment and behavior. Social Psychology, 44, 61–62. doi: 10.1027/1864-9335/a000137

2013 Hogrefe Publishing

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

Allport, G. W. (1954/1979). The nature of prejudice. Cambridge, MA: Addison-Wesley.

Aronsson, K., & S tterlund-Larsson, U. (1987). Politeness strategies and doctor-patient communication: On the social choreography of collaborative thinking. Journal of Language and Social Psychology, 6, 1–27.

Arvidsson, D., Sikstrçm, S., & Werbart, A. (2011). Changes in self and object representations following psychotherapy measured by a theory-free, computational, semantic space method. Psychotherapy Research, 21, 430–446. doi: 10.1080/10503307.2011.577824 Asch, S. E. (1946). Forming impressions of personality. Journal of Abnormal and Social Psychology, 41, 258–290. doi: 10.1037/H0055756 Austin, J. L. (1962). How to do things with words. The William James lectures delivered at Harward University in 1955.

Oxford, UK: Clarendon.

Baumeister, R. F., & Ilko, S. A. (1995). Shallow gratitude – Public and private acknowledgment of external help in accounts of success. Basic and Applied Social Psychology, 16, 191–209.

Bradley, M. M., & Lang, P. J. (1999). Affective norms for English words (ANEW): Stimuli, instruction manual and affective ratings. Gainesville, FL: The NIMH center for the study of emotion and attention, University of Florida.

Brewer, M. B. (1979). In-group bias in the minimal intergroup situation – Cognitive-motivational analysis. Psychological Bulletin, 86, 307–324. doi: 10.1037/0033-2909.86.2.307 Brewer, M. B. (1999). The psychology of prejudice: Ingroup love and outgroup hate? Journal of Social Issues, 55, 429– 444. doi: 10.1111/0022-4537.00126 Brewer, M. B. (2007). The importance of being we: Human nature and intergroup relations. American Psychologist, 62, 726–738. doi: 10.1037/0003-066X.62.8.728 Brown, P., & Levinson, S. C. (1987). Politeness: Some universals in language usage. New York, NY: Cambridge University Press.

Campbell, R. S., & Pennebaker, J. W. (2003). The secret life of pronouns: Flexibility in writing style and physical health.

Psychological Science, 14, 60–65. doi: 10.1111/14679280.01419 Carnaghi, A., Maass, A., Gresta, S., Bianchi, M., Cadinu, M., & Arcuri, L. (2008). Nomina sunt omina: On the inductive potential of nouns and adjectives in person perception.

Journal of Personality and Social Psychology, 94, 839–859. doi: 10.1037/0022-3514.94.5.839 Chung, C. K., & Pennebaker, J. W. (2007). The psychological functions of function words. In K. Fiedler (Ed.), Social Communication (pp. 343–359). New York, NY: Psychology Press.

Cortes, B. P., Demoulin, S., Rodriguez, R. T., Rodriguez, A. P., & Leyens, J. P. (2005). Infrahumanization or familiarity?

Attribution of uniquely human emotions to the self, the ingroup, and the outgroup. Personality and Social Psychology Bulletin, 31, 243–253.

Critcher, C. R., Helzer, E. G., & Dunning, D. (2011). Selfenhancement via redefinition: Defining social concepts to ensure positive views of the self. In M. D. Alicke & C. Sedikides (Eds.), Handbook of self-enhancement and selfprotection (pp. 69–91). New York, NY: Guilford Press.

Dixon, T. L., & Linz, D. (2000). Overrepresentation and underrepresentation of African Americans and Latinos as lawbreakers on television news. Journal of Communication, 50, 131–154.

Dovidio, J. F., Glick, P., & Rudman, L. A. (2005). On the nature of prejudice. Fifty years after Allport. Malden, MA: Blackwell.

2013 Hogrefe Publishing

109

Dunbar, R. I. M. (1993). Co-evolution of neo-cortex, group size and language in humans. Behavioral and Brain Sciences, 16, 681–735. doi: 10.1017/S0140525X00032325 Dunbar, R. I. M. (1997). Grooming, gossip and the evolution of language. New York, NY: Faber & Faber.

Evans, M. G. (1985). A Monte-Carlo study of the effects of correlated method variance in moderated multiple-regression analysis. Organizational Behavior and Human Decision Processes, 36, 305–323. doi: 10.1016/0749-5978(85)90002-0 Fiske, S. T. (2012). Warmth and competence: Stereotype content issues for clinicians and researchers. Canadian Psychology/Psychologie Canadienne, 53, 14–20. doi: 10.1037/A0026054 Fiske, S. T., Lin, K., & Neuberg, S. L. (1999). The continuum model: Ten years later. In S. Chaiken & Y. Trope (Eds.), Dual process theories in social psychology (pp. 231–254).

New York, NY: Guilford Press.

Foltz, P. W., Kintsch, W., & Landauer, T. K. (1998). The measurement of textual coherence with latent semantic analysis. Discourse Processes, 25, 285–307.

Gaertner, L., & Schopler, J. (1998). Perceived ingroup entitativity and intergroup bias: An interconnection of self and others.

European Journal of Social Psychology, 28, 963–980.

Gaertner, L., Sedikides, C., Luke, M., O’Mara, E. M., Iuzzini, J., Jackson, L. E., . . . Wu, Q. (2012). A motivational hierarchy within: Primacy of the individual self, relational self, or collective self? Journal of Experimental Social Psychology, 48, 997–1013.

Gaertner, L., Sedikides, C., Vevea, J. L., & Iuzzini, J. (2002).

The ‘‘I’’ the ‘‘we’’, and the ‘‘when’’: A meta-analysis of motivational primacy in self-definition. Journal of Personality and Social Psychology, 83, 574–591. doi: 10.1037/ 0022-3514.83.3.574 Garcia, D., & Sikstrçm, S. (2013). A collective theory of happiness: Words related to the word ‘‘Happiness’’ in Swedish online newspapers. Cyberpsychology, Behavior, and Social Networking, 27, 1–4.

Geschke, D., Sassenberg, K., Ruhrmann, G., & Sommer, D. (2010). Effects of linguistic abstractness in the mass media: How newspaper articles shape readers’ attitudes toward migrants. Journal of Media Psychology: Theories, Methods, and Applications, 22, 99–104. doi: 10.1027/1864-1105/ a000014 Goffman, E. (1959). The presentation of self in everyday life.

New York, NY: Anchor Books.

Graf, S., Bilewicz, M., Finell, E., & Geschke, D. (2012). Nouns cut slices: Effects of linguistic forms on intergroup bias.

Journal of Language and Social Psychology, 32, 62–83. doi: 10.1177/0261927x12463209 Gustafsson Send n, M., Lindholm, T., & Sikstrom, S. (in press).

Selection bias as reflected by choice of words: The evaluations of ‘‘I’’ and ‘‘We’’ differ between communication contexts but ‘‘They’’ are always worse. Journal of Language and Social Psychology.

Koivula, N. (1999). Gender stereotyping in televised media sport coverage. Sex Roles, 41, 589–604. doi: 10.1023/ a:1018899522353 Landauer, T. K. (1998). Learning and representing verbal meaning: The latent semantic analysis theory. Current Directions in Psychological Science, 7, 161–164.

Landauer, T. K. (1999). Latent semantic analysis: A theory of the psychology of language and mind. Discourse Processes, 27, 303–310.

Landauer, T. K., & Dumais, S. T. (1997). A solution to Plato’s problem: The latent semantic analysis theory of acquisition, induction, and representation of knowledge. Psychological Review, 104, 211–240.

Social Psychology 2014; Vol. 45(2):103–111

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

110

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

Landauer, T. K., Foltz, P. W., & Laham, D. (1998). An introduction to latent semantic analysis. Discourse Processes, 25, 259–284.

Landauer, T. K., Laham, D., & Foltz, P. (2003). Automatic essay assessment. Assessment in Education, 10, 295–308.

Landauer, T. K., Laham, D., Foltz, P. W., Shermis, M. D., & Burstein, J. (2003). Automatic essay scoring: A crossdisciplinary perspective. Mahwah, NJ: Erlbaum.

Landauer, T. K., McNamara, D. S., Dennis, S., & Kintsch, W. (2007). Handbook of latent semantic analysis. Mahwah, NJ: Erlbaum.

Lee, J. J., & Pinker, S. (2010). Rationales for indirect speech: The theory of the strategic speaker. Psychological Review, 117, 785–807. doi: 10.1037/a0019688 Lenton, A., Sedikides, C., & Bruder, M. (2009). A latent semantic analysis of gender stereotype-consistency and narrowness in American English. Sex Roles, 60, 269–278. doi: 10.1007/s11199-008-9534-z Maass, A. (1999). Linguistic intergroup bias: Stereotype perpetuation through language. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 31, pp. 79–121). San Diego, CA: Academic Press.

Maass, A., Ceccarelli, R., & Rudin, S. (1996). Linguistic intergroup bias: Evidence for in-group-protective motivation. Journal of Personality and Social Psychology, 71, 512–526. doi: 10.1037/0022-3514.71.3.512 Maass, A., Milesi, A., Zabbini, S., & Stahlberg, D. (1995).

Linguistic intergroup bias – Differential expectancies or ingroup protection. Journal of Personality and Social Psychology, 68, 116–126.

Maass, A., Salvi, D., Arcuri, L., & Semin, G. R. (1989).

Language use in intergroup contexts – The linguistic intergroup bias. Journal of Personality and Social Psychology, 57, 981–993.

Markus, H. R., & Kitayama, S. (1991). Culture and the self – Implications for cognition, emotion, and motivation.

Psychological Review, 98, 224–253.

McClelland, G. H., & Judd, C. M. (1993). Statistical difficulties of detecting interactions and moderator effects. Psychological Bulletin, 114, 376–390.

Menegatti, M., & Rubini, M. (2012). From the individual to the group: The enhancement of linguistic bias. European Journal of Social Psychology, 42, 36–40. doi: 10.1002/ Ejsp.856 Menegatti, M., & Rubini, M. (2013). Convincing similar and dissimilar others: The power of language abstraction in political communication. Personality and Social Psychology Bulletin, 39, 596–607. doi: 10.1177/0146167213479404 Miller, N. (2002). Personalization and the promise of contact theory. Journal of Social Issues, 58, 387–410.

Miller, R. S., & Schlenker, B. R. (1985). Egotism in group members – Public and private attributions of responsibility for group-performance. Social Psychology Quarterly, 48, 85–89.

Moscatelli, S., Albarello, F., & Rubini, M. (2008). Linguistic discrimination in minimal groups – The impact of status differentials. Journal of Language and Social Psychology, 27, 140–154. doi: 10.1177/0261927x07313652 Mullen, B., Brown, R., & Smith, C. (1992). Ingroup bias as a function of salience, relevance, and status – an integration.

European Journal of Social Psychology, 22, 103–122.

Na, J., & Choi, I. (2009). Culture and first-person pronouns.

Personality and Social Psychology Bulletin, 35, 1492–1499. doi: 10.1177/0146167209343810

Social Psychology 2014; Vol. 45(2):103–111

Nettle, D. (1999). Using social impact theory to simulate language change. Lingua, 108, 95–117.

Pennebaker, J. W. (2011). The secret life of pronouns: What our words say about us. New York, NY: Bloomsbury.

Pennebaker, J. W., Mehl, M. R., & Niederhoffer, K. G. (2003). Psychological aspects of natural language use: Our words, our selves. Annual Review of Psychology, 54, 547–577.

Perdue, C. W., Gurtman, M. B., Dovidio, J. F., & Tyler, R. B. (1990). Us and them – Social categorization and the process of intergroup bias. Journal of Personality and Social Psychology, 59, 475–486. doi: 10.1037/0022-3514.59.3.475 Picard, R. R., & Cook, R. D. (1984). Cross-validation of regression models. Journal of the American Statistical Association, 79, 575–583. doi: 10.2307/2288403 Pitts, M. J., & Nussbaum, J. F. (2006). Integrating the past and paving the future. Examining current trends and extending boundaries of language and social psychology research.

Journal of Language and Social Psychology, 25, 197–202.

Reuters. (2008). Handbook of journalism, Retrieved from http:// handbook.reuters.com/index.php?title=Main_Page Rubini, M., Moscatelli, S., & Palmonari, A. (2007). Increasing group entitativity: Linguistic intergroup discrimination in the minimal group paradigm. Group Processes and Intergroup Relations, 10, 280–296. doi: 10.1177/1368430207 075156 Sedikides, C., & Gregg, A. P. (2008). Self-enhancement food for thought. Perspectives on Psychological Science, 3, 102– 116. doi: 10.1111/j.1745-6916.2008.00068.x Semin, G. R., & Fiedler, K. (1988). The cognitive functions of linguistic categories in describing persons – Social cognition and language. Journal of Personality and Social Psychology, 54, 558–568. doi: 10.1037/0022-3514.54.4.558 Sexton, J. B., & Helmreich, R. L. (2000). Analyzing cockpit communications: The links between language, performance, error, and workload. Human Performance in Extreme Environments, 5, 63–68.

Sherif, M., Harvey, O. J., White, B. J., Hood, W. R., & Sherif, C. V. (1961). Intergroup conflict and cooperation: The Robbers cave experiment. Norman, OK: University of Oklahoma Book Exchange.

Slatcher, R. B., Chung, C. K., Pennebaker, J. W., & Stone, L. D. (2007). Winning words: Individual differences in linguistic style among US presidential and vice presidential candidates. Journal of Research in Personality, 41, 63–75. doi: 10.1016/j.jrp.2006.01.006 Stone, L. D., & Pennebaker, J. W. (2002). Trauma in real time: Talking and avoiding online conversations about the death of Princess Diana. Basic and Applied Social Psychology, 24, 173–183.

Sutton, R. M. (2010). The creative power of language in social cognition and intergroup relations. In H. Giles, S. Reid, & J. Harwood (Eds.), The dynamics of intergroup communication (pp. 105–115). New York, NY: Peter Lang.

Tajfel, H., & Turner, J. (1979). An integrative theory of intergroup conflict. In W. G. Austin & S. Worchel (Eds.), The social psychology of intergroup relations (pp. 33–47).

Monterey, CA: Brooks-Cole.

Tajfel, H., & Turner, J. C. (1986). The social identity theory of inter-group behavior. In S. Worchel & L. W. Austin (Eds.), Psychology of intergroup relations. Chicago, IL: Nelson-Hall.

Tausczik, Y. R., & Pennebaker, J. W. (2010). The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29, 24–54.

2013 Hogrefe Publishing

M. Gustafsson Send n et al.: Personal Pronouns in Evaluative News Media Context

This document is copyrighted by the American Psychological Association or one of its allied publishers.

This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Twenge, J. M., Campbell, W. K., & Gentile, B. (2012). Male and female pronoun use in US books reflects women’s status, 1900–2008. Sex Roles, 67, 488–493. doi: 10.1007/ s11199-012-0194-7 Wigboldus, D. H. J., & Douglas, K. (2007). Language, stereotypes, and intergroup relations. In K. Fiedler (Ed.), Social communication (pp. 79–106). New York, NY: Psychology Press.

Wigboldus, D. H. J., Semin, G. R., & Spears, R. (2000). How do we communicate stereotypes? Linguistic biases and inferential consequences. Journal of Personality and Social Psychology, 78, 5–18.

111

Marie Gustafsson Send n Department of Psychology Stockholm University 106 91 Stockholm Sweden Tel. +46 707 644 144 Tel. +46 8 163-946 E-mail mgu@psychology.su.se

Received February 18, 2013 Revision received September 10, 2013 Accepted September 13, 2013 Published online November 15, 2013

2013 Hogrefe Publishing

Social Psychology 2014; Vol. 45(2):103–111