This article was downloaded by: [University of Southern Queensland] On: 13 March 2015, At: 21:13 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Cognition and Emotion Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/pcem20
The language of worry: Examining linguistic elements of worry models a
Elena M. C. Geronimi & Janet Woodruff-Borden
a
a
Department of Psychological and Brain Sciences, University of Louisville, Louisville, KY, USA Published online: 20 May 2014.
Click for updates To cite this article: Elena M. C. Geronimi & Janet Woodruff-Borden (2015) The language of worry: Examining linguistic elements of worry models, Cognition and Emotion, 29:2, 311-318, DOI: 10.1080/02699931.2014.917071 To link to this article: http://dx.doi.org/10.1080/02699931.2014.917071
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
COGNITION AND EMOTION, 2015 Vol. 29, No. 2, 311–318, http://dx.doi.org/10.1080/02699931.2014.917071
BRIEF REPORT The language of worry: Examining linguistic elements of worry models Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
Elena M. C. Geronimi and Janet Woodruff-Borden Department of Psychological and Brain Sciences, University of Louisville, Louisville, KY, USA
Despite strong evidence that worry is a verbal process, studies examining linguistic features in individuals with generalised anxiety disorder (GAD) are lacking. The aim of the present study is to investigate language use in individuals with GAD and controls based on GAD and worry theoretical models. More specifically, the degree to which linguistic elements of the avoidance and intolerance of uncertainty worry models can predict diagnostic status was analysed. Participants were 19 women diagnosed with GAD and 22 control women and their children. After participating in a diagnostic semi-structured interview, dyads engaged in a free-play interaction where mothers’ language sample was collected. Overall, the findings provided evidence for distinctive linguistic features of individuals with GAD. That is, after controlling for the effect of demographic variables, present tense, future tense, prepositions and number of questions correctly classified those with GAD and controls such that a considerable amount of the variance in diagnostic status was explained uniquely by language use. Linguistic confirmation of worry models is discussed. Keywords: Worry language; Linguistic; GAD; Worry models.
Generalised anxiety disorder (GAD) has a high prevalence rate; approximately 5% of people will develop GAD during their lifetime (APA, 2013). The fundamental feature of GAD is exaggerated, chronic worry (APA, 2013). Worry has been defined primarily as a verbal mental process with high frequency of linguistic thought and low frequency of imagery, and this pattern has been supported in several studies (Behar, Zuellig, & Borkovec, 2005;
Borkovec & Inz, 1990; Borkovec & Lyonfields, 1993; Freeston, Dugas, & Ladouceur, 1996). Freeston et al. (1996) found that, when asked to characterise their worry, participants identify a higher frequency of thoughts (70.4%) as opposed to images (24.8%). In addition, excessive worriers tend to report higher percentages of thoughts compared to low worriers (Freeston et al., 1996). Using a different method and a clinical sample,
Correspondence should be addressed to: Elena M. C. Geronimi, Department of Psychological and Brain Sciences, University of Louisville, 317 Life Sciences, Louisville, KY 40292, USA. E-mail: [email protected] © 2014 Taylor & Francis
311
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
GERONIMI AND WOODRUFF-BORDEN
Borkovec and Inz (1990) asked participants to engage in worry and relaxation. During relaxation, control participants reported higher frequency of imagery than thoughts, while individuals with GAD reported equal amounts of imagery and thoughts. In addition, during worry, both groups reported higher frequency of thoughts and, after receiving treatment, individuals with GAD reported a decrease in thought activity during relaxation. These findings suggest that thought activity characterises both worry activity and individuals who have GAD. Further evidence for the verbal nature of worry was found in Behar et al.’s (2005) study. Participants’ reports supported the idea that worry is predominantly a verbal thinking process, while trauma recall is an imaginal process. Behar et al. (2005) speculate that, during worry activity, threatening images are substituted by verbal content as a way of avoiding stronger somatic and emotional reactions (more associated to images) and thus maintaining worry through negative reinforcement. Despite strong evidence that worry is a verbal process, studies examining linguistic features in individuals with GAD are lacking. An empirical way to further understand worry and GAD is through the analysis of linguistic elements derived from the worry and GAD theoretical models. For instance, applying Behar et al.’s (2005) speculation to a purely verbal chain of thoughts, it could be expected that some linguistic elements would be preferred, over others, by individuals with GAD, as a way of avoiding stronger emotional reactions. If this is the case, differences in the speech between individuals with GAD and controls should be expected based on the worry theoretical models. Several conceptual models have been proposed as an attempt to understand predisposing and maintaining factors of GAD and worry (see Behar, DiMarco, Hekler, Mohlman, & Staples, 2009, for a review). The avoidance function of worry is well established as the fundamental feature of the avoidance worry model and is also a common feature across other conceptual models of worry (Behar et al., 2009). The avoidance model conceptualises worry as an avoidant strategy that is reinforced in a variety
312
COGNITION AND EMOTION, 2015, 29 (2)
of ways (Borkovec, 1994; Borkovec, Alcaine, & Behar, 2004). A core feature of this model is the assumption that worry is primarily verbal in nature and that its verbal features are key in preventing the experience of somatic and emotional symptoms that are more closely associated to imagery (Behar et al., 2005; Borkovec & Inz, 1990). In fact, physiological findings suggest that worry is associated with poor autonomic modulation (Thayer, Friedman, & Borkovec, 1996). Combined, these findings reveal that worry may inhibit the processing of threatening information and, thus, maintain the cognitive and affective fear response patterns, despite repeated exposures (Foa, Huppert, & Cahill, 2006). Considering the avoidance model of worry and applying its principles to the language domain, high worriers, in comparison to low worriers, should differ in their use of linguistic elements to help to perpetuate the worry. For instance, it could be expected that individuals who worry would use more connectors in their speech as a way to extend worry periods. Further, according to Borkovec’s (1994) view, since the future threat exists only in the minds of the individuals who worry, engaging in worry is one of the few possibilities left as an attempt to prevent hypothetical negative events from happening. Thus, another linguistic characteristic that could derive from this process is that worriers would tend to focus their speech in the future, as opposed to the present. If the worry is about a topic that cannot be addressed in the present moment, simply maintaining attention in the future, as opposed to present, may be a good strategy for maintaining this activity and contributing to a false perceived sense of control over the threat. In recent studies, intolerance of uncertainty (IU) has been linked to exaggerated worry, and thus, has received growing attention in the field (e.g., Buhr & Dugas, 2002, 2006). IU consists of negative responses when facing uncertain and ambiguous situations (Buhr & Dugas, 2006). Buhr and Dugas (2002) suggested that IU is composed of four main factors that highly overlap: uncertainty is stressful and upsetting, it makes it difficult to act, events that are uncertain should be avoided and being uncertain
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
THE LANGUAGE OF WORRY
is unfair. IU is a fundamental component of worry (Buhr & Dugas, 2002) and has been conceptualised as a central piece in theoretical models of worry and GAD (Dugas, Gagnon, Ladouceur, & Freeston, 1998). In an attempt to test a conceptual model, Dugas et al. (1998) found that IU, beliefs about worry, poor problem orientation and cognitive avoidance discriminated between individuals with GAD and controls. Most importantly, IU was the variable that most contributed to the model. In the proposed model, the presence of IU functions as a fundamental cognitive piece in initiating questions (e.g., ‘what if’) even when a precipitating stimulus is not present (Dugas et al., 1998). A linguistic way of assessing such patterns in individuals with GAD would be through assessing the frequency of questions asked by high worriers in comparison to low worriers. Although findings derived from worry model studies are extremely important in understanding the phenomenon, most of these studies relied on self-report measures (Behar et al., 2009). The use of other methodological tools is needed in further investigating worry and GAD. Considering the verbal nature of worry, language analyses could be an important contribution in this process. It has been found that individuals who are more prone to worry tend to have certain cognitive characteristics (e.g., perseverative thought process) that are independent of content and valence (Davey & Levy, 1998). In addition, Davey and Levy (1998) argue that the ability to catastrophise—a process that highly relates to worry—about a new topic may be driven by permanent cognitive features that the worrier applies to the catastrophising process. Similarly, assessing worriers’ linguistic style could be a way of identifying cognitive features that are unique to individuals with high levels of worry. Although some studies have investigated the worry process and its content (e.g., Molina Borkovec, Peasley, & Person, 1998), to our knowledge, there are no studies that have examined specifically linguistic characteristics of individuals with GAD. However, linguistic-related studies conducted with depressed and socially anxious individuals have revealed that language can be a meaningful assessment of cognitive structures as well as theoretical
models of mental pathology (Hofmann, Moore, Gutner, & Weeks, 2012; Rude, Gortner, & Pennebaker, 2004). The aim of the present study is to investigate language characteristics in individuals diagnosed with GAD and controls based on GAD and worry theoretical models. It is well established that there are differences in cognitive activities between those with GAD and controls. For instance, individuals with GAD tend to think more in semantic thoughts than images compared to controls (e.g., Borkovec & Inz, 1990). But there may also be specific linguistic characteristics that are unique to individuals with GAD and that contribute to the precipitation and maintenance of worry activity. More specifically, based on the verbal nature of worry and on the proposed language characteristics of the worry models, we expect that language use will distinguish between those with GAD and controls. We hypothesise that present tense, future tense, prepositions and number of questions combined will correctly classify those with a GAD diagnosis and those not meeting criteria for a diagnosis. In addition, we also predict that each of the language categories will uniquely predict diagnostic status.
METHOD Participants Participants were selected as part of a broader study that investigated familial anxiety. Families were recruited through distribution of flyers to schools, day care programmes, churches, YMCA, clinics and through referrals from mental health agencies. The flyers described the benefits in participating in the study (i.e., diagnostic assessment for both parent and child and a small toy for the child). Participants for the current study were women from 41 dyads (19 mothers with GAD— as a primary diagnosis—and 22 control mothers and their children—ages 3–5 years; 18 boys and 23 girls). Thirty mothers (73.2%) were EuropeanAmerican, 10 (24.3%) were African-American and 1 (2.5%) was Native American. The majority of mothers were married (n = 27; 66%), seven (17%) were divorced, six (14.5%) were never married and COGNITION AND EMOTION, 2015, 29 (2)
313
GERONIMI AND WOODRUFF-BORDEN
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
one (2.5%) was widowed. Secondary diagnoses for mothers with GAD were as follows: social phobia (n = 5), major depressive disorder (n = 4), posttraumatic stress disorder (n = 2), panic disorder without agoraphobia (n = 1), panic disorder with agoraphobia (n = 1) and hypochondriasis (n = 1). Family’s median gross income was $65,000 and the mother’s median education level was college graduate. Control mothers did not meet criteria for any diagnosis according to a semi-structured interview (see details below).
Materials and procedure Diagnostic interviews Diagnoses were determined by the administration of the Anxiety Disorders Interview ScheduleFourth Edition (ADIS-IV; Brown, Di Nardo, & Barlow, 1994). The ADIS-IV is a semi-structured interview that assesses the presence and severity of anxiety and related disorders based on the criteria defined on the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV; APA, 1994) criteria. The ADIS-IV has been shown to have good-to-excellent reliability for most mood and anxiety disorders assessed (Brown, DiNardo, Lehman, & Campbell, 2001). Diagnostic interviews were administered by advanced doctoral students who met reliability criteria of three consecutive diagnostic matches within one point of severity rating. One-third of the interviews were rated by a second interviewer to determine inter-rater reliability. Kappa for diagnostic agreement was .90. Interaction task Parent and child were given uniform instructions to enrol in a free-play task for 10 minutes. No other instructions were given to the dyads; the goal was to mimic a daily interaction between them, providing an approximation of mother’s typical linguistic patterns. They had free access to toys, puzzles and books during that time, and all interactions were videotaped.
314
COGNITION AND EMOTION, 2015, 29 (2)
Language sample and analysis The mothers’ words during the interactions were transcribed verbatim into text files and the Linguistic Inquiry and Word Count (LIWC2007; Pennebaker, Chung, Ireland, Gonzales, & Booth, 2007) software was used to analyse linguistic content. The LIWC2007 uses a hierarchical model to classify words in different word groups (e.g., pronouns and negations). An output file is generated by the software containing the frequencies of words in several language categories. In this study, four language categories were analysed: present tense, future tense, prepositions and number of questions. The categories used were standard categories offered by LIWC2007. The LIWC2007 has good internal reliability and external validity (Kahn, Tobin, Massey, & Anderson, 2007; Pennebaker et al., 2007; Tausczik & Pennebaker, 2010).
RESULTS Demographics and language variables Prior to conducting statistical analyses, data were explored regarding any effects of demographic variables and collinearity between language variables. A t-test indicated that women with GAD and control women did not differ in terms of age. However, significant correlations were found between parent’s age and present tense (r = .41, p = .008). In addition, Pearson’s chi-square tests indicated no differences between the groups in terms of ethnicity, marital status and education. However, groups differed in terms of income [χ2(9, 41) = 18.81, p = .027]. Further, income was correlated with future tense (r = .35, p = .027) and education was correlated with a number of questions (r = −.36, p = .020). In addition, language use in the categories analysed did not differ relative to child sex, and there were no significant correlations between language use and child’s age. Correlations between language categories failed to reach significance.
THE LANGUAGE OF WORRY
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
Language predicting diagnostic status Given our hypothesis that language would predict diagnosis, we examined how much of the variance in diagnostic grouping would be accounted for by language use. Thus, the following categories were included in a logistic regression: present tense, future tense, prepositions and number of questions. In addition, considering the demographic findings, demographic variables that were identified as possible confounds (i.e., parent income, age and education; see demographic and language variables section above) were entered in the first block. The model was significant in both Block 1 (p = .031) and Block 2 (p < .001). Differences in Cox & Snell R2 and Nagelkerke R2 values between Block 1 and Block 2 revealed that the linguistic categories alone explained 34.80%–46.40% of the variance in mothers’ diagnosis. Analyses of the contribution of each word category separately revealed that present tense (W = 5.72, p = .017, OR = .40), future tense (W = 4.37, p = .037, OR = 23.84) and number of questions (W = 4.49, p = .034, OR = 1.68) were significant predictors of mothers’ diagnoses. In addition, B values suggest that the use of future tense and questions increased the likelihood of having a diagnosis of GAD. Contrarily, using more present tense verbs decreased the likelihood of presenting with GAD (see Table 1 for the logistic regression model). Furthermore, the model in Block 2 (linguistic categories) had 82.9% of accuracy in diagnosis classification, as opposed to 68.3% of accuracy in Block 1. More specifically, in Block 2, 86.4% of controls and 78.9% of the individuals with GAD were correctly classified (see additional details in Table 2).
DISCUSSION In the present study, linguistic features based on the avoidance and IU worry models were assessed to predict diagnostic status in individuals with GAD and controls. As expected, language use successfully predicted diagnostic status. After controlling for the effect of demographic variables, present tense, future tense, prepositions and number of questions
combined correctly classified those with GAD and controls such that a considerable amount of the variance in diagnostic status was explained uniquely by language use. Furthermore, present tense, future tense and number of questions separately were significant predictors of diagnostic status. These findings suggest that higher use of future tense and lower use of present tense increased the likelihood of GAD. This pattern of verb tense use may suggest an attempt to keep distance from immediate emotional processing by those with GAD, consistent with the avoidance worry model (Borkovec, 1994). Interestingly, in a previous study, worry, the characteristic feature of GAD, was associated with a lower frequency of present statements compared to neutral periods (Molina et al., 1998). Although Molina et al. (1998) adopted a different method by classifying statements of anxious and dysphonic participants during worry and neutral periods, as opposed to looking specifically to linguistic categories, a lower use of present-oriented periods seems consistent in both worry periods and the speech of individuals with GAD. The present findings also presented support to the IU model of worry. That is, those diagnosed with GAD used more questions than did controls, possibly as a linguistic attempt to decrease uncertainty. The present findings also suggest that linguistic features of those with GAD may be less concrete in nature (i.e., increased likelihood of using future tense and questions and decreased likelihood of using present tense verbs) compared to those of controls. Based on the worry reduced-concreteness theory, Stöber (1998) suggested that concrete verbal content generates more vivid and faster images compared to more abstract material. If worry is a cognitive attempt to avoid the processing of imagery (Behar et al., 2005; Borkovec & Inz, 1990), the language associated with worry as well as with those with GAD should also be less concrete. In fact, although they did not examine specific linguistic elements, Stöber and Borkovec (2002) found that worry in GAD is associated with reduced concreteness. Future studies should try to test this theory more directly through linguistic analyses, for instance, it could be hypothesised that those with GAD would use more subjunctive and COGNITION AND EMOTION, 2015, 29 (2)
315
GERONIMI AND WOODRUFF-BORDEN
Table 1. Language use as a predictor of diagnostic status
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
Variable Demographics Block 1 Model (at Block 1) Parent income Parent age Parent education Language categories Block 2 Model (at Block 2) Present Future Questions Prepositions
χ2
P
8.9 8.9
.031 .031
R2
.194; .260
Cognition and Emotion Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/pcem20
The language of worry: Examining linguistic elements of worry models a
Elena M. C. Geronimi & Janet Woodruff-Borden
a
a
Department of Psychological and Brain Sciences, University of Louisville, Louisville, KY, USA Published online: 20 May 2014.
Click for updates To cite this article: Elena M. C. Geronimi & Janet Woodruff-Borden (2015) The language of worry: Examining linguistic elements of worry models, Cognition and Emotion, 29:2, 311-318, DOI: 10.1080/02699931.2014.917071 To link to this article: http://dx.doi.org/10.1080/02699931.2014.917071
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
COGNITION AND EMOTION, 2015 Vol. 29, No. 2, 311–318, http://dx.doi.org/10.1080/02699931.2014.917071
BRIEF REPORT The language of worry: Examining linguistic elements of worry models Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
Elena M. C. Geronimi and Janet Woodruff-Borden Department of Psychological and Brain Sciences, University of Louisville, Louisville, KY, USA
Despite strong evidence that worry is a verbal process, studies examining linguistic features in individuals with generalised anxiety disorder (GAD) are lacking. The aim of the present study is to investigate language use in individuals with GAD and controls based on GAD and worry theoretical models. More specifically, the degree to which linguistic elements of the avoidance and intolerance of uncertainty worry models can predict diagnostic status was analysed. Participants were 19 women diagnosed with GAD and 22 control women and their children. After participating in a diagnostic semi-structured interview, dyads engaged in a free-play interaction where mothers’ language sample was collected. Overall, the findings provided evidence for distinctive linguistic features of individuals with GAD. That is, after controlling for the effect of demographic variables, present tense, future tense, prepositions and number of questions correctly classified those with GAD and controls such that a considerable amount of the variance in diagnostic status was explained uniquely by language use. Linguistic confirmation of worry models is discussed. Keywords: Worry language; Linguistic; GAD; Worry models.
Generalised anxiety disorder (GAD) has a high prevalence rate; approximately 5% of people will develop GAD during their lifetime (APA, 2013). The fundamental feature of GAD is exaggerated, chronic worry (APA, 2013). Worry has been defined primarily as a verbal mental process with high frequency of linguistic thought and low frequency of imagery, and this pattern has been supported in several studies (Behar, Zuellig, & Borkovec, 2005;
Borkovec & Inz, 1990; Borkovec & Lyonfields, 1993; Freeston, Dugas, & Ladouceur, 1996). Freeston et al. (1996) found that, when asked to characterise their worry, participants identify a higher frequency of thoughts (70.4%) as opposed to images (24.8%). In addition, excessive worriers tend to report higher percentages of thoughts compared to low worriers (Freeston et al., 1996). Using a different method and a clinical sample,
Correspondence should be addressed to: Elena M. C. Geronimi, Department of Psychological and Brain Sciences, University of Louisville, 317 Life Sciences, Louisville, KY 40292, USA. E-mail: [email protected] © 2014 Taylor & Francis
311
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
GERONIMI AND WOODRUFF-BORDEN
Borkovec and Inz (1990) asked participants to engage in worry and relaxation. During relaxation, control participants reported higher frequency of imagery than thoughts, while individuals with GAD reported equal amounts of imagery and thoughts. In addition, during worry, both groups reported higher frequency of thoughts and, after receiving treatment, individuals with GAD reported a decrease in thought activity during relaxation. These findings suggest that thought activity characterises both worry activity and individuals who have GAD. Further evidence for the verbal nature of worry was found in Behar et al.’s (2005) study. Participants’ reports supported the idea that worry is predominantly a verbal thinking process, while trauma recall is an imaginal process. Behar et al. (2005) speculate that, during worry activity, threatening images are substituted by verbal content as a way of avoiding stronger somatic and emotional reactions (more associated to images) and thus maintaining worry through negative reinforcement. Despite strong evidence that worry is a verbal process, studies examining linguistic features in individuals with GAD are lacking. An empirical way to further understand worry and GAD is through the analysis of linguistic elements derived from the worry and GAD theoretical models. For instance, applying Behar et al.’s (2005) speculation to a purely verbal chain of thoughts, it could be expected that some linguistic elements would be preferred, over others, by individuals with GAD, as a way of avoiding stronger emotional reactions. If this is the case, differences in the speech between individuals with GAD and controls should be expected based on the worry theoretical models. Several conceptual models have been proposed as an attempt to understand predisposing and maintaining factors of GAD and worry (see Behar, DiMarco, Hekler, Mohlman, & Staples, 2009, for a review). The avoidance function of worry is well established as the fundamental feature of the avoidance worry model and is also a common feature across other conceptual models of worry (Behar et al., 2009). The avoidance model conceptualises worry as an avoidant strategy that is reinforced in a variety
312
COGNITION AND EMOTION, 2015, 29 (2)
of ways (Borkovec, 1994; Borkovec, Alcaine, & Behar, 2004). A core feature of this model is the assumption that worry is primarily verbal in nature and that its verbal features are key in preventing the experience of somatic and emotional symptoms that are more closely associated to imagery (Behar et al., 2005; Borkovec & Inz, 1990). In fact, physiological findings suggest that worry is associated with poor autonomic modulation (Thayer, Friedman, & Borkovec, 1996). Combined, these findings reveal that worry may inhibit the processing of threatening information and, thus, maintain the cognitive and affective fear response patterns, despite repeated exposures (Foa, Huppert, & Cahill, 2006). Considering the avoidance model of worry and applying its principles to the language domain, high worriers, in comparison to low worriers, should differ in their use of linguistic elements to help to perpetuate the worry. For instance, it could be expected that individuals who worry would use more connectors in their speech as a way to extend worry periods. Further, according to Borkovec’s (1994) view, since the future threat exists only in the minds of the individuals who worry, engaging in worry is one of the few possibilities left as an attempt to prevent hypothetical negative events from happening. Thus, another linguistic characteristic that could derive from this process is that worriers would tend to focus their speech in the future, as opposed to the present. If the worry is about a topic that cannot be addressed in the present moment, simply maintaining attention in the future, as opposed to present, may be a good strategy for maintaining this activity and contributing to a false perceived sense of control over the threat. In recent studies, intolerance of uncertainty (IU) has been linked to exaggerated worry, and thus, has received growing attention in the field (e.g., Buhr & Dugas, 2002, 2006). IU consists of negative responses when facing uncertain and ambiguous situations (Buhr & Dugas, 2006). Buhr and Dugas (2002) suggested that IU is composed of four main factors that highly overlap: uncertainty is stressful and upsetting, it makes it difficult to act, events that are uncertain should be avoided and being uncertain
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
THE LANGUAGE OF WORRY
is unfair. IU is a fundamental component of worry (Buhr & Dugas, 2002) and has been conceptualised as a central piece in theoretical models of worry and GAD (Dugas, Gagnon, Ladouceur, & Freeston, 1998). In an attempt to test a conceptual model, Dugas et al. (1998) found that IU, beliefs about worry, poor problem orientation and cognitive avoidance discriminated between individuals with GAD and controls. Most importantly, IU was the variable that most contributed to the model. In the proposed model, the presence of IU functions as a fundamental cognitive piece in initiating questions (e.g., ‘what if’) even when a precipitating stimulus is not present (Dugas et al., 1998). A linguistic way of assessing such patterns in individuals with GAD would be through assessing the frequency of questions asked by high worriers in comparison to low worriers. Although findings derived from worry model studies are extremely important in understanding the phenomenon, most of these studies relied on self-report measures (Behar et al., 2009). The use of other methodological tools is needed in further investigating worry and GAD. Considering the verbal nature of worry, language analyses could be an important contribution in this process. It has been found that individuals who are more prone to worry tend to have certain cognitive characteristics (e.g., perseverative thought process) that are independent of content and valence (Davey & Levy, 1998). In addition, Davey and Levy (1998) argue that the ability to catastrophise—a process that highly relates to worry—about a new topic may be driven by permanent cognitive features that the worrier applies to the catastrophising process. Similarly, assessing worriers’ linguistic style could be a way of identifying cognitive features that are unique to individuals with high levels of worry. Although some studies have investigated the worry process and its content (e.g., Molina Borkovec, Peasley, & Person, 1998), to our knowledge, there are no studies that have examined specifically linguistic characteristics of individuals with GAD. However, linguistic-related studies conducted with depressed and socially anxious individuals have revealed that language can be a meaningful assessment of cognitive structures as well as theoretical
models of mental pathology (Hofmann, Moore, Gutner, & Weeks, 2012; Rude, Gortner, & Pennebaker, 2004). The aim of the present study is to investigate language characteristics in individuals diagnosed with GAD and controls based on GAD and worry theoretical models. It is well established that there are differences in cognitive activities between those with GAD and controls. For instance, individuals with GAD tend to think more in semantic thoughts than images compared to controls (e.g., Borkovec & Inz, 1990). But there may also be specific linguistic characteristics that are unique to individuals with GAD and that contribute to the precipitation and maintenance of worry activity. More specifically, based on the verbal nature of worry and on the proposed language characteristics of the worry models, we expect that language use will distinguish between those with GAD and controls. We hypothesise that present tense, future tense, prepositions and number of questions combined will correctly classify those with a GAD diagnosis and those not meeting criteria for a diagnosis. In addition, we also predict that each of the language categories will uniquely predict diagnostic status.
METHOD Participants Participants were selected as part of a broader study that investigated familial anxiety. Families were recruited through distribution of flyers to schools, day care programmes, churches, YMCA, clinics and through referrals from mental health agencies. The flyers described the benefits in participating in the study (i.e., diagnostic assessment for both parent and child and a small toy for the child). Participants for the current study were women from 41 dyads (19 mothers with GAD— as a primary diagnosis—and 22 control mothers and their children—ages 3–5 years; 18 boys and 23 girls). Thirty mothers (73.2%) were EuropeanAmerican, 10 (24.3%) were African-American and 1 (2.5%) was Native American. The majority of mothers were married (n = 27; 66%), seven (17%) were divorced, six (14.5%) were never married and COGNITION AND EMOTION, 2015, 29 (2)
313
GERONIMI AND WOODRUFF-BORDEN
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
one (2.5%) was widowed. Secondary diagnoses for mothers with GAD were as follows: social phobia (n = 5), major depressive disorder (n = 4), posttraumatic stress disorder (n = 2), panic disorder without agoraphobia (n = 1), panic disorder with agoraphobia (n = 1) and hypochondriasis (n = 1). Family’s median gross income was $65,000 and the mother’s median education level was college graduate. Control mothers did not meet criteria for any diagnosis according to a semi-structured interview (see details below).
Materials and procedure Diagnostic interviews Diagnoses were determined by the administration of the Anxiety Disorders Interview ScheduleFourth Edition (ADIS-IV; Brown, Di Nardo, & Barlow, 1994). The ADIS-IV is a semi-structured interview that assesses the presence and severity of anxiety and related disorders based on the criteria defined on the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV; APA, 1994) criteria. The ADIS-IV has been shown to have good-to-excellent reliability for most mood and anxiety disorders assessed (Brown, DiNardo, Lehman, & Campbell, 2001). Diagnostic interviews were administered by advanced doctoral students who met reliability criteria of three consecutive diagnostic matches within one point of severity rating. One-third of the interviews were rated by a second interviewer to determine inter-rater reliability. Kappa for diagnostic agreement was .90. Interaction task Parent and child were given uniform instructions to enrol in a free-play task for 10 minutes. No other instructions were given to the dyads; the goal was to mimic a daily interaction between them, providing an approximation of mother’s typical linguistic patterns. They had free access to toys, puzzles and books during that time, and all interactions were videotaped.
314
COGNITION AND EMOTION, 2015, 29 (2)
Language sample and analysis The mothers’ words during the interactions were transcribed verbatim into text files and the Linguistic Inquiry and Word Count (LIWC2007; Pennebaker, Chung, Ireland, Gonzales, & Booth, 2007) software was used to analyse linguistic content. The LIWC2007 uses a hierarchical model to classify words in different word groups (e.g., pronouns and negations). An output file is generated by the software containing the frequencies of words in several language categories. In this study, four language categories were analysed: present tense, future tense, prepositions and number of questions. The categories used were standard categories offered by LIWC2007. The LIWC2007 has good internal reliability and external validity (Kahn, Tobin, Massey, & Anderson, 2007; Pennebaker et al., 2007; Tausczik & Pennebaker, 2010).
RESULTS Demographics and language variables Prior to conducting statistical analyses, data were explored regarding any effects of demographic variables and collinearity between language variables. A t-test indicated that women with GAD and control women did not differ in terms of age. However, significant correlations were found between parent’s age and present tense (r = .41, p = .008). In addition, Pearson’s chi-square tests indicated no differences between the groups in terms of ethnicity, marital status and education. However, groups differed in terms of income [χ2(9, 41) = 18.81, p = .027]. Further, income was correlated with future tense (r = .35, p = .027) and education was correlated with a number of questions (r = −.36, p = .020). In addition, language use in the categories analysed did not differ relative to child sex, and there were no significant correlations between language use and child’s age. Correlations between language categories failed to reach significance.
THE LANGUAGE OF WORRY
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
Language predicting diagnostic status Given our hypothesis that language would predict diagnosis, we examined how much of the variance in diagnostic grouping would be accounted for by language use. Thus, the following categories were included in a logistic regression: present tense, future tense, prepositions and number of questions. In addition, considering the demographic findings, demographic variables that were identified as possible confounds (i.e., parent income, age and education; see demographic and language variables section above) were entered in the first block. The model was significant in both Block 1 (p = .031) and Block 2 (p < .001). Differences in Cox & Snell R2 and Nagelkerke R2 values between Block 1 and Block 2 revealed that the linguistic categories alone explained 34.80%–46.40% of the variance in mothers’ diagnosis. Analyses of the contribution of each word category separately revealed that present tense (W = 5.72, p = .017, OR = .40), future tense (W = 4.37, p = .037, OR = 23.84) and number of questions (W = 4.49, p = .034, OR = 1.68) were significant predictors of mothers’ diagnoses. In addition, B values suggest that the use of future tense and questions increased the likelihood of having a diagnosis of GAD. Contrarily, using more present tense verbs decreased the likelihood of presenting with GAD (see Table 1 for the logistic regression model). Furthermore, the model in Block 2 (linguistic categories) had 82.9% of accuracy in diagnosis classification, as opposed to 68.3% of accuracy in Block 1. More specifically, in Block 2, 86.4% of controls and 78.9% of the individuals with GAD were correctly classified (see additional details in Table 2).
DISCUSSION In the present study, linguistic features based on the avoidance and IU worry models were assessed to predict diagnostic status in individuals with GAD and controls. As expected, language use successfully predicted diagnostic status. After controlling for the effect of demographic variables, present tense, future tense, prepositions and number of questions
combined correctly classified those with GAD and controls such that a considerable amount of the variance in diagnostic status was explained uniquely by language use. Furthermore, present tense, future tense and number of questions separately were significant predictors of diagnostic status. These findings suggest that higher use of future tense and lower use of present tense increased the likelihood of GAD. This pattern of verb tense use may suggest an attempt to keep distance from immediate emotional processing by those with GAD, consistent with the avoidance worry model (Borkovec, 1994). Interestingly, in a previous study, worry, the characteristic feature of GAD, was associated with a lower frequency of present statements compared to neutral periods (Molina et al., 1998). Although Molina et al. (1998) adopted a different method by classifying statements of anxious and dysphonic participants during worry and neutral periods, as opposed to looking specifically to linguistic categories, a lower use of present-oriented periods seems consistent in both worry periods and the speech of individuals with GAD. The present findings also presented support to the IU model of worry. That is, those diagnosed with GAD used more questions than did controls, possibly as a linguistic attempt to decrease uncertainty. The present findings also suggest that linguistic features of those with GAD may be less concrete in nature (i.e., increased likelihood of using future tense and questions and decreased likelihood of using present tense verbs) compared to those of controls. Based on the worry reduced-concreteness theory, Stöber (1998) suggested that concrete verbal content generates more vivid and faster images compared to more abstract material. If worry is a cognitive attempt to avoid the processing of imagery (Behar et al., 2005; Borkovec & Inz, 1990), the language associated with worry as well as with those with GAD should also be less concrete. In fact, although they did not examine specific linguistic elements, Stöber and Borkovec (2002) found that worry in GAD is associated with reduced concreteness. Future studies should try to test this theory more directly through linguistic analyses, for instance, it could be hypothesised that those with GAD would use more subjunctive and COGNITION AND EMOTION, 2015, 29 (2)
315
GERONIMI AND WOODRUFF-BORDEN
Table 1. Language use as a predictor of diagnostic status
Downloaded by [University of Southern Queensland] at 21:13 13 March 2015
Variable Demographics Block 1 Model (at Block 1) Parent income Parent age Parent education Language categories Block 2 Model (at Block 2) Present Future Questions Prepositions
χ2
P
8.9 8.9
.031 .031
R2
.194; .260
