- Email: [email protected]
Interpretive cues and ambiguity in generalized anxiety disorder
Behaviour Research and Therapy 42 (2004) 881–892 www.elsevier.com/locate/brat
Interpretive cues and ambiguity in generalized anxiety disorder Holly Hazlett-Stevens a,, T.D. Borkovec b b
a Department of Psychology/298, University of Nevada, Reno, NV 89557, USA Department of Psychology, Pennsylvania State University, University Park, PA 16802, USA
Received 23 January 2003; received in revised form ; accepted 7 July 2003
Abstract The current study investigated whether generalized anxiety disorder (GAD) individuals rely on antecedent information to interpret ambiguity and whether reliance on such preceding cues persists in the absence of potential threat. Twenty-six GAD and 23 nonanxious control college students performed a lexical decision task, using homographs (i.e. words with multiple meanings) as ambiguous primes. In half the trials, a homograph prime that possessed both threat-related, as well as neutral meanings was followed by a target word related to one of these two meanings. In addition, each ambiguous prime was immediately preceded by a series of four antecedent words that were either: (a) associated with the threatening meaning of the prime; (b) associated with the neutral meaning of the prime; or (c) unrelated to either meaning of the homograph, as well as the target. Homographs for which both meanings were neutral in valence comprised the other half of the trials. Effect size statistics suggest that GAD participants utilized the antecedent words to interpret the homograph primes with threat-related meanings, unlike their nonanxious counterparts (p < 0:06). When both meanings of the homograph prime were neutral in valence, the GAD group appeared deficient in the use of preceding information to interpret the ambiguous prime. # 2003 Elsevier Ltd. All rights reserved. Keywords: Generalized anxiety disorder; Information processing; Interpretation of ambiguity
Anxious individuals are more likely than nonanxious individuals to interpret ambiguous stimuli as threatening. In a comprehensive recent review of this literature, MacLeod (in press)
Corresponding author. Tel.: +1-775-784-6828; fax: +1-775-327-5043. E-mail address: [email protected] (H. Hazlett-Stevens).
0005-7967/$ - see front matter # 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0005-7967(03)00204-3
882
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
described several studies finding evidence for an interpretive threat bias among GAD samples. However, methodological limitations including experimenter demand characteristics, reliance on participant self-report, potential response selection biases, and possible memory biases were identified. Experimentally rigorous investigations of trait-anxious participants have addressed these concerns, but interpretive biases of GAD individuals have not been explored with these superior methods (MacLeod, in press; Aikins & Craske, 2001). In one such trait-anxiety study, Richards and French (1992) utilized homographs as ambiguous stimuli in a lexical decision task. In this experimental paradigm, words with both threatrelated and neutral meanings (threat/neutral homographs) were presented on a computer screen as ambiguous primes (e.g. batter, shot). Threat/neutral homograph primes were immediately followed by target words in the context of a lexical decision task (i.e. a task in which participants are instructed to decide whether the target string of letters is an actual word or a nonword). Each word target was consistent with one of the two meanings of the homograph or unrelated to the homograph. The speed with which participants responded to each target indicated the degree to which their performance on that trial was facilitated by the prime. Richards and French found that when presented with homograph primes containing both neutral and threatening possible meanings, high trait-anxious individuals accessed the threatening meanings more readily than their low trait-anxious counterparts (as evidenced by faster reaction times to the threat-related targets relative to unrelated targets). Similar lexical decision task results were found for test anxiety participants when ambiguous sentences served as primes (Calvo, Eysenck, & Estevez, 1994). MacLeod and Cohen (1993) presented high and low trait-anxious participants with sentence pairs, administered one sentence at a time at a self-paced rate. The first sentence of each pair was ambiguous, while the subsequent sentence disambiguated the first sentence consistent with either its threatening or its non-threatening meaning. Each trial was preceded by a cue word reflecting either meaning of the ambiguous sentence or by a cue word unrelated to the ambiguous sentence. Comprehension latency results suggested that high trait-anxiety participants were more prone than low trait-anxiety participants to interpret the ambiguous sentence as threatening. Of particular interest, MacLeod and Cohen also found a general slowing in comprehension latencies when the cue word suggested one meaning for an ambiguous sentence, but the continuation sentence reflected its other meaning. If anxious individuals are more likely than nonanxious individuals to interpret such ambiguity as threatening, they might also rely more on available preceding information to make their interpretations. Such a finding would extend previous interpretive bias research to more naturalistic conditions, as naturally occurring ambiguity in the environment rarely appears in the absence of preceding interpretive cues. The current investigation examined whether GAD participants were more likely than nonanxious participants to rely on preceding verbal cues when interpreting ambiguous stimuli with potentially threatening meanings. If GAD individuals are prone to interpret threat in ambiguous stimulus situations, they may also depend upon available antecedent information to make their interpretations when a threatening interpretation is possible. Second, if this reliance on available cues exists and functions to prepare an individual for threat, then GAD participants would not necessarily rely on such cues to interpret ambiguity under different conditions where no threat could be interpreted. Therefore, an additional aim was to investigate GAD and non-
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
883
anxious participants’ use of preceding cues when a threatening interpretation of the ambiguous stimulus was not possible. Rigorous methods similar to those used in trait-anxiety experiments were chosen. A modified version of the lexical decision task developed by Richards and French (1992) was designed. Crucially, the current study presented antecedent words before the homograph prime to examine their effects on the subsequent interpretation of this ambiguous word. In some trials, antecedent cues prepared participants for one meaning of the ambiguous word yet contained a target word reflecting its other meaning. A second set of trials in which both meanings of the ambiguous word were neutral in valence addressed whether an increased reliance on antecedent words for the GAD group was present only under conditions of potential threat.
1. Method 1.1. Participants A total of 1131 male and female undergraduate students enrolled in psychology courses at the Pennsylvania State University completed the Generalized Anxiety Disorder Questionnaire during a group testing session (GAD-Q-IV; Newman et al., 2002). The GAD-Q-IV is a reliable and valid self-report diagnostic measure of GAD that accurately identifies individuals meeting DSM-IV GAD criteria with 89% specificity and 83% sensitivity when compared to a clinician administered semi-structured diagnostic interview, the Anxiety Disorders Interview Schedule (ADIS-IV; Brown, DiNardo, & Barlow, 1994). Thus, the demonstrated reliability between the GAD-Q-IV and the ADIS-IV is comparable to that reported for two independently administered ADIS-IV interviews (Newman et al., 2002; Brown, DiNardo, Lehman, & Campbell, 2001). In addition, the GAD-Q-IV reliably distinguishes GAD from panic disorder and from social phobia (Newman et al., 2002). A total of 26 (three male, 23 female) participants meeting all DSM-IV GAD diagnostic criteria and 23 participants (two male, 21 female) meeting no diagnostic criteria on the GAD-Q-IV participated in the present study. All 49 participants identified English as their first language. Mean age was 19.98 years (SD ¼ 1:55) and mean education was 14.71 years (SD ¼ 1:21). 1.2. Materials (homograph word task)1 A pool of 120 threat/neutral homographs (i.e. homographs for which one meaning is threatening and one meaning is neutral in valence) were selected from the norms complied by French and Richards (1992) to be used as ambiguous primes. Each threat/neutral homograph had two meanings that differed significantly in terms of threat ratings from 12 independent judges (French & Richards, 1992). For each threat/neutral homograph, four words associated with its threatening meaning, four words associated with its neutral meaning, and four words that were unrelated in meaning to the homograph yet neutral in valence were selected as antecedent words. Trials in which the antecedent words were unrelated to the 1
Examples of stimuli from the homograph word task are available from the first author upon request.
884
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
homograph prime were included for replication of the naturally occurring interpretive threat bias reported by Richards and French (1992). Finally, for each threat/neutral homograph, one additional word associated with its threatening meaning and one additional word associated with its neutral meaning were selected for use as threat and neutral valence targets, respectively. Threat or neutral target words immediately followed and disambiguated each ambiguous (homograph) prime. A separate pool of 120 neutral/neutral homographs (i.e. homographs for which both meanings are neutral in valence) was selected from the norms compiled by Nelson, McEvory, Walling and Wheeler (1980) for use as neutral ambiguous primes. For each neutral/neutral homograph, four words associated with its primary meaning (i.e. the meaning which appears first in the compilation of norm data; Nelson et al., 1980), four words associated with its secondary meaning (i.e. a meaning which appears following the first definition in the compilation of norm data; Nelson et al., 1980), and four words that were unrelated to any meaning of the homograph yet neutral in valence were selected for use as antecedent words to be presented immediately before the neutral/neutral homograph prime. Finally, for each neutral/neutral homograph, one additional word associated with its primary meaning and one additional word associated with its secondary meaning were selected for use as primary and secondary targets, respectively. Primary or secondary target words immediately followed and disambiguated each ambiguous (homograph) prime. The two target words for each homograph were matched for frequency and word length. The neutral/neutral homographs were matched to the threat/neutral homographs for frequency and word length such that there were no significant differences between these two sets of homographs. Printed frequency data for each of these words were obtained from Francis and Kucera (1982). In total there were 240 trials per participant, with each trial consisting of a series of four antecedent words followed by a homograph prime followed by a target word or non-word. Nonword targets were used in half of the trials. For non-word trials, one letter of each target was changed to produce pronounceable non-word targets. The total 120 word trials consisted of 60 threat/neutral homograph primes and 60 neutral/neutral homograph primes. For the 60 threat/neutral homograph prime word trials, 10 trials were presented for each of six different combinations: (a) the neutral antecedent words, followed by the threat/neutral homograph prime, followed by the neutral target; (b) the neutral antecedent words, followed by the threat/neutral homograph prime, followed by the threat target; (c) the threat antecedent words, followed by the threat/neutral homograph prime, followed by the threat target; (d) the threat antecedent words, followed by the threat/ neutral homograph prime, followed by the neutral target; (e) the unrelated antecedent words, followed by the threat/neutral homograph prime, followed by the neutral target; and (f) the unrelated antecedent words, followed by the threat/neutral homograph prime, followed by the threat target. For the 60 neutral/neutral homograph prime word trials, 10 trials were presented for each of the following six different combinations: (a) the primary definition antecedent words, followed by the neutral/neutral homograph prime, followed by the primary definition target; (b) the primary definition antecedent words, followed by the neutral/neutral homograph prime, followed by the secondary definition target; (c) the secondary definition antecedent words, followed by
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
885
the neutral/neutral homograph prime, followed by the secondary definition target; (d) the secondary definition antecedent words, followed by the neutral/neutral homograph prime, followed by the primary definition target; (e) the unrelated antecedent words, followed by the neutral/neutral homograph prime, followed by the primary definition target; and (f) the unrelated antecedent words, followed by the neutral/neutral homograph prime, followed by the secondary definition target. For these stimuli, the ‘primary’ vs. ‘secondary’ definition simply referred to the order in which that meaning was presented in Nelson et al. (1980). Stimuli were originally counterbalanced such that each threat/neutral and neutral/neutral homograph would appear in each of its six possible trial variations equally across each group of six participants. Participants were randomized to counterbalancing group in blocks of six to maximize the balance of stimuli across the experiment as a whole as much as possible. Because there were three more participants in the GAD group, one extra GAD participant was run in three of the six counterbalancing order groups. The 240 trials were presented in a randomized order for each participant. Stimuli were presented and lexical decision reaction times were recorded using SuperLab program software on a Power Mac model Macintosh computer. 1.3. Procedure Participants initially completed a screening packet provided in their psychology classroom, consisting of the GAD-Q-IV, the Penn State Worry Questionnaire (PSWQ; Meyer, Miller, Metzger, & Borkovec, 1990), the trait version of the State-Trait Anxiety Inventory (STAI-T; Spielberger, Gorsuch, & Lushene, 1970), and the Beck Depression Inventory (BDI; Beck, 1978). The experimental session was conducted in a laboratory with a large writing table in the central room and a smaller adjoining room containing the computer. Each participant was tested individually for a 90-min session and remained in the experimental room throughout the experiment. One of three experimenters conducted each experimental session and was blind to each participant’s GAD/nonanxious group status. After informed consent was obtained, the state version of the STAI (Spielberger, Gorsuch & Lushene, 1970) was administered to provide an assessment of the anxiety level experienced at the time of the experimental session. Participants were then seated at the computer for the homograph word task. The experimenter explained that participants would be presented with strings of words on a computer screen, one word at a time. Participants were told that their task was to read each of the first five words silently to themselves, and then to indicate whether the sixth word was an actual word or not by pressing the corresponding key on the computer keyboard as quickly as possible. Participants performed a practice series of trials prior to the experiment proper. Each trial began with an asterisk in the center of the screen to serve as a signal for the participant to fixate centrally. After the participant hit the spacebar key, each of the four antecedent words were consecutively displayed centrally for 200 ms each, immediately followed by a homograph prime. Each homograph prime was presented centrally for 750 ms. This difference in presentation time was used so that participants could clearly discriminate
886
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
between the antecedent words and the homograph prime, both of which preceded the target. Immediately after the homograph prime, the target (word or non-word) appeared on the screen, overwriting the prime, and remained there until the participant made a manual response by pressing keys on the computer keyboard (right hand for words, left hand for non-words). Stimuli were presented in lower-case letters. Participants were given short breaks after each 80 trials. After the homograph word task was completed, participants were administered the Reading subtest of the Wide Range Achievement Test-Revised (WRAT-R) as a general measure of reading ability and overall intellectual ability. Responses only to the word trials of the lexical decision task were used in data analyses. Response times less than 300 ms or greater than 3000 ms were counted as errors. In addition, reaction times that exceeded 2.5 standard deviations above the mean for a given participant were discarded as outliers. Response errors and outliers were not included in data analyses, resulting in a loss of 4.4% of the data.
2. Results 2.1. Self-report anxiety and depression questionnaire measures To verify expected group differences on self-report measures, a multivariate analysis of variance (MANOVA) on the three screening measures (PSWQ, STAI-Trait version, BDI) and the experimental session questionnaire (STAI-State version) was performed with GAD group status as the between-subjects factor. Results indicated a significant multivariate effect, F ð4; 44Þ ¼ 33:08; p < 0:001. Univariate analyses revealed that this effect reflected a statistically significant difference between GAD and nonanxious groups on each of the four anxiety and depression questionnaire measures in the predicted direction: PSWQ, F ð1; 47Þ ¼ 118:02; p < 0:001; STAI-Trait, F ð1; 47Þ ¼ 50:52; p < 0:001; BDI, F ð1; 47Þ ¼ 21:23; p < 0:001; STAIState, Fð1; 47Þ ¼ 55:74; p < 0:001 (see Table 1 for means and standard deviations). 2.2. Wide range achievement test-revised To establish equivalent intellectual ability between the two groups, a t-test on the raw scores from the Reading subtest of the WRAT-R with GAD group status as the between-subjects factor was conducted. No significant differences between the groups on this control measure were found [tð1; 47Þ ¼ 0:41; p > 0:5]. Table 1 Means and standard deviations for self-report anxiety and depression questionnaire measures for GAD and nonanxious control groups Group
PSWQ M
GAD 65.96 Nonanxious 35.57
STAI-Trait
BDI
STAI-State
SD
M
SD
M
SD
M
SD
9.43 10.15
52.46 34.30
9.35 8.38
16.38 4.30
11.28 5.88
48.15 32.13
8.60 6.01
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
887
2.3. Homograph word task reaction times 2 2.3.1. Unrelated antecedent word trials Analysis was first conducted for those trials involving only unrelated antecedent words plus the threat/neutral homographs to determine whether GAD participants responded faster to threat targets than to neutral targets when compared to nonanxious participants. Interpretation of the ambiguous homograph prime was not expected to be impacted by the preceding four antecedent words for these trials. This approach allowed for replication of the interpretive threat bias reported by Richards and French (1992). Reaction times in ms for correct responses to word targets were averaged over trial type for each participant. A target (threat, neutral) by group (GAD, nonanxious) repeated measures ANOVA was conducted on the mean reaction times of correct responses to word targets. Neither a main effect of target (p > 0:2) nor a group target interaction effect (p > 0:5) was found. A main effect of group approached significance [F ð1; 47Þ ¼ 3:31; p < 0:08, partial eta squared g2 ¼ 0:066, observed power ¼ 0:429], in which GAD participants responded slower (M ¼ 939:04; SD ¼ 316:99) than nonanxious participants (M ¼ 793:85; SD ¼ 227:97) to the targets overall. To verify an absence of interpretive differences for those trials involving only unrelated antecedent words plus the neutral/neutral homographs, a target (primary, secondary) by group (GAD, nonanxious) repeated measures ANOVA was conducted on the mean reaction times of correct responses. As expected, no main effect of target and no group x target interaction effect were found (both p’s > 0:3). However, a main effect of group approached significance [F ð1; 47Þ ¼ 3:44; p < 0:07, partial eta squared g2 ¼ 0:068, observed power ¼ 0:444], in which GAD participants again responded slower (M ¼ 941:89; SD ¼ 323:14) than nonanxious participants (M ¼ 788:78; SD ¼ 242:58) to the targets overall. 2.3.2. Related antecedent word trials To test the main study hypotheses regarding the effects of the antecedent words on subsequent ambiguous homograph prime interpretation, a series of analyses were conducted for the threat and neutral antecedent word trials of the threat/neutral homograph trials and for the primary and secondary antecedent word trials of the neutral/neutral homograph trials. It was predicted that for the threat/neutral homograph trials, the GAD group would show greater disruption than the nonanxious group on trials containing a mismatch between the meaning implied by the antecedent words and the meaning suggested by the target (compared to trials in which the antecedent words and target suggested the same meaning). Thus, the GAD participants were expected to exhibit a slowing on threat/neutral homograph trials when the antecedent word and target meanings were different than when the meanings reflected in each were the same, regardless of valence. Reaction times for correct responses to word targets were aver2
Inspection of histograms revealed that reaction time values used in these analyses were moderately positively skewed. Therefore, data were transformed using the square root data transformation procedure described by Tabachnick and Fidell (2001), and repeated measures ANOVA analyses were repeated with transformed data. Results obtained were similar to those reported above. This similarity between untransformed data results and transformed data results was not surprising, given that all reaction time variables were skewed in the same direction to about the same moderate extent (see Tabachnick & Fidell, 2001: 81).
888
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
aged over trial type for each participant. T-tests confirmed no difference between neutral antecedent words/neutral target trials and threat antecedent words/threat target trials and no difference between neutral antecedent words/threat target trials and threat antecedent words/neutral target trials for either group (all p’s > 0:4). Therefore, trials were collapsed into either same antecedent word-target trials or into different antecedent word-target trials to reflect the relationship between the meaning suggested by the antecedent words and that suggested by the target. An antecedent word (same, different) by group (GAD, nonanxious) repeated measures ANOVA on the mean reaction times for correct responses to word targets presented following the threat/neutral homograph primes revealed a marginally significant main effect for group status [Fð1; 47Þ ¼ 3:67; p < 0:06], qualified by an antecedent word group interaction approaching significance [F ð1; 47Þ ¼ 3:59; p < 0:06, partial eta squared g 2 ¼ 0:071, observed power ¼ 0:458]. Due to the observed limitations in power, three follow-up comparisons were conducted to examine the nature of this interaction trend. First, it had been predicted that the GAD participants would exhibit a slowing for different antecedent word trials compared to same antecedent word trials. However, the difference in this predicted direction only approached significance (p < 0:08). Second, it had been predicted that the nonanxious group would not exhibit any differences between same and different antecedent word trials for the threat/neutral homograph trials. Indeed, no such effect was found (p > 0:4). Third, it had been predicted that the difference in reaction time between different and same antecedent word trials would be greater for the GAD group than for the nonanxious group. While this effect only approached statistical significance [tð1; 47Þ ¼ 1:89; p < 0:06Þ], calculations of effect size revealed a mediumsize effect, d ¼ 0:528 (Cohen, 1988) (see Table 2 for mean reaction times, standard deviations, and difference scores). An analogous series of analyses were then conducted for related antecedent word trials from the neutral/neutral homograph trials. It had been predicted that for the neutral/neutral homograph trials, in which the ambiguous homograph prime had no threatening meanings, the nonanxious group would show greater disruption than the GAD group on trials containing a mismatch between the meaning implied by the antecedent words and the meaning suggested by the target (compared to trials in which the antecedent words and target suggested the same meaning). Thus, the nonanxious participants were expected to exhibit a slowing on neutral/neutral homograph trials when the antecedent words and target meanings were different than when the meanings reflected in each were the same. T-tests confirmed no difference between primary Table 2 Means and standard deviations of reaction times and mean difference scores (in ms) for the same and different antecedent words plus the threat/neutral homograph prime trials for GAD and nonanxious control groups Group
GAD Nonanxious
Same meaning for antecedent words and target
Different meaning for antecedent words and target
Different meaning–same meaning difference score
M
SD
M
SD
M
907.22 789.67
260.75 239.56
944.22 777.46
324.00 205.60
37.00 12.20
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
889
Table 3 Means and standard deviations of reaction times and mean difference scores (in ms) for the same and different antecedent words plus the neutral/neutral homograph prime trials for GAD and nonanxious control groups Group
GAD Nonanxious
Same meaning for antecedent words and target
Different meaning for antecedent words and target
Different meaning–same meaningdifference score
M
SD
M
SD
M
921.71 760.00
340.00 230.23
908.66 807.25
278.32 263.54
13.05 47.26
meaning antecedent words/primary meaning target trials and secondary meaning antecedent words/secondary meaning target trials for either group, and no difference between primary meaning antecedent words/secondary meaning target trials and secondary meaning antecedent words/primary meaning target trials for either group (all p’s > 0:2). Therefore, trials were collapsed into either same antecedent words-target trials or into different antecedent words-target trials to reflect the relationship between the meaning suggested by the antecedent words and that suggested by the target. An antecedent word (same, different) by group (GAD, nonanxious) repeated measures ANOVA was conducted on the mean reaction times for correct responses to word targets presented following the neutral/neutral homograph primes. A significant antecedent word x group interaction effect emerged [Fð1; 47Þ ¼ 6:00; p < 0:05, partial eta squared g2 ¼ 0:113, observed power ¼ 0:669]. Three follow-up analyses were conducted to explore the nature of this interaction effect. First, it had been predicted that the nonanxious participants would exhibit a slowing for different antecedent word trials compared to same antecedent word trials, and this hypothesis was supported (p < 0:001). Second, it had been predicted that the GAD group would not exhibit any differences between same and different antecedent word trials for the neutral/neutral homograph trials. Indeed, no such effect was found (p > 0:5). Third, it was predicted that the difference in reaction time between the different and same antecedent word trials would be greater for the nonanxious group than for the GAD group, and this effect was also found [tð1; 47Þ ¼ 2:45; p < 0:05; d ¼ 0:667) (see Table 3 for mean reaction times, standard deviations, and difference scores).
3. Discussion Richards and French (1992) demonstrated that trait-anxious participants show an interpretive bias in favor of threatening meanings when confronted with threat/neutral homograph primes as ambiguous stimuli. The present study aimed to replicate this effect in a GAD sample and to extend previous findings by presenting interpretive cue words immediately before the ambiguous prime. The current investigation did not find evidence of an interpretive bias when potentially threatening homograph primes were preceded merely by unrelated neutral antecedent words. This discrepancy between the Richards and French study and the current experiment may have resulted from methodological differences, as the unrelated antecedent words (or the embedded
890
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
trials involving related antecedent words) may have inadvertently influenced responses to the unrelated antecedent word trial targets. Effect size statistics from the trials involving related antecedent words plus the threat/neutral homographs suggested that GAD individuals may have utilized the antecedent words more so than their nonanxious counterparts to interpret the ambiguity of the threat/neutral homographs. Unlike the nonanxious group, the GAD group exhibited a slowing when the meaning reflected in the antecedent words differed from the meaning reflected by the target. Although this effect was of medium size (d ¼ 0:528; Cohen, 1988), it failed to reach statistical significance (p < 0:06). It therefore cannot be determined whether this finding reflects an actual phenomenon that went undetected due to insufficient statistical power or whether the null hypothesis provides a more accurate interpretation. Even if this finding were reliable, it does not address whether the GAD group’s tendency to rely on preceding information to interpret ambiguity persists beyond the defensive preparation for potential threat. Therefore, neutral/neutral homograph trials have been included to examine whether the GAD participants’ tendency to utilize interpretive cues (if found) was specific to stimuli conditions of potential threat interpretation. When the neutral/neutral homograph data are considered, a different picture emerges. Analyses here revealed that reaction times for nonanxious participants depended on whether the antecedent words reflected the same meaning or a different meaning of the ambiguous word as the target did. In contrast, GAD participants failed to utilize the antecedent words to interpret the neutral/neutral homographs, as evidenced by equivalent reaction times when the meaning reflected in the antecedent words differed from the meaning reflected by the target compared to trials in which the antecedent words and target meanings were consistent. This pattern of results suggests that while the GAD group may have used the antecedent words to interpret ambiguous words to anticipate threat, this strategy to interpret ambiguity was abandoned when the potential for threat was removed. Equivalent scores on the WRAT-R Reading subtest suggest that these group differences are not a function of reading ability or overall intellectual functioning. In addition to statistical power concerns, other limitations should be considered. First, participants were selected on the basis of a self-report diagnostic measure (GAD-Q-IV) rather than a clinician-administered diagnostic interview (e.g. ADIS-IV). Many clinicians and researchers consider clinician-administered diagnostic interviews the gold standard of clinical assessment. However, the GAD-Q-IV yields strong reliability when compared to the ADIS-IV (Newman et al., 2002), equivalent to that reported for two independently administered ADIS-IV interviews (Brown, DiNardo, Lehman & Campbell, 2001). Furthermore, PSWQ scores for the GAD group (M ¼ 65:96; SD ¼ 9:43) were comparable to those reported for treatment-seeking clinical GAD samples diagnosed by ADIS-R interviews (M ¼ 67:66; SD ¼ 8:86; Molina & Borkovec, 1994). Nevertheless, the current sample consisted of college students seeking course credit rather than treatment for their participation. Due to differences in motivation and demographic composition, these results may not generalize to treatment-seeking community populations. A second limitation concerns our inability to discriminate between two possible explanations for the effects of the antecedent words. This experimental approach assumes that a mismatch between the meaning reflected in the antecedent words and that reflected in the target should slow reaction time to the extent the participant used the antecedent words to interpret the ambiguous homograph prime. It is possible that these antecedent words thus served as contex-
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
891
tual stimuli, and that the GAD group relied on the experimental context less than the nonanxious group in the absence of potential threat interpretation (i.e. neutral/neutral homograph trials). Word stimuli were presented such that participants could clearly discriminate between the antecedent words and the homograph prime, both of which preceded the target. The alternative hypothesis is that the antecedent words provided an additional priming effect for the homograph prime itself, resulting in a double-priming effect in the related antecedent word trials. Thus, GAD and nonanxious individuals may have differed in the degree to which this further priming effect was present across the two homograph types. GAD may be characterized by increased use of interpretive cues when facing potentially threatening ambiguous situations, but the use of such information to interpret other ambiguous situations appears deficient for these individuals. If most daily life situations for a given GAD individual cannot be interpreted as threatening, a failure to process other relevant non-threatening information may be as important to the maintenance of anxiety as his or her periodic interpretive threat biases. This could leave anxious individuals at a disadvantage in numerous everyday situations that do not involve potential threat.
Acknowledgements Preparation of this manuscript was supported in part by an American Psychological Association Dissertation Research Award and a Penn State Research and Graduate Studies Dissertation Support Grant awarded to the first author and by National Institute of Mental Health Grant MH-39172 awarded to the second author. This research is based on the dissertation of the first author. Portions of this paper were presented at the 36th annual convention of the Association for Advancement of Behavior Therapy (AABT), Reno, NV, November, 2002. The authors gratefully acknowledge Andrew Mathews for his helpful comments on an earlier draft, Judy Kroll for her assistance with the design and implementation of the lexical decision task, Mia Peteritas for her assistance with the collection and management of data, and Michael Mitchell for his assistance with data analyses.
References Aikins, D. E., & Craske, M. G. (2001). Cognitive theories of generalized anxiety disorder. The Psychiatric Clinics of North America, 24, 57–74. Beck, A. T. (1978). Depression inventory. Philadelphia: Center for Cognitive Therapy. Brown, T. A., DiNardo, P. A., & Barlow, D. H. (1994). Anxiety disorders interview schedule for DSM-IV (ADIS-IV). San Antonio, TX: The Psychological Corporation. Brown, T. A., DiNardo, P. A., Lehman, C. L., & Campbell, L. A. (2001). Reliability of DSM-IV anxiety and mood disorders: Implications for the classification of emotional disorders. Journal of Abnormal Psychology, 110, 49–58. Calvo, M. G., Eysenck, M. W., & Estevez, A. (1994). Ego-threat interpretive bias in test anxiety. Cognition and Emotion, 8, 127–146. Cohen, J. (1988). Statistical power analysis for behavioral sciences. (2nd ed.). New York: Academic Press. Francis, W. N., & Kucera, H. (1982). Frequency analysis of English usage. Boston: Houghton Mifflin. French, C. C., & Richards, A. (1992). Word association norms for a set of threat/neutral homographs. Cognition & Emotion, 6, 65–87.
892
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
MacLeod, C. (in press). Information processing approaches to generalized anxiety disorder: Assessing the selective functioning of attention, interpretation and memory in GAD patients. In R. G. Heimberg, C. L. Turk, & D. S. Mennin (Eds.), Generalized anxiety disorder: Advances in research and practice. New York: Guilford Press. MacLeod, C., & Cohen, I. L. (1993). Anxiety and the interpretation of ambiguity: A text comprehension study. Journal of Abnormal Psychology, 102, 238–247. Meyer, T. J., Miller, M. L., Metzger, R. L., & Borkovec, T. D. (1990). Development and validation of the Penn State Worry Questionnaire. Behaviour Research and Therapy, 28, 487–495. Molina, S., & Borkovec, T. D. (1994). The Penn State Worry Questionnaire: Psychometric properties and associated characteristics. In G. Davey, & F. Tallis (Eds.), Worrying: Perspectives on theory, assessment and treatment (pp. 265–283). New York: Wiley. Nelson, D. L., McEvory, C. L., Walling, J. R., & Wheeler, J. W. (1980). The University of South Florida homograph norms. Behavior Research Methods & Instrumentation, 12, 16–37. Newman, M. G., Zuellig, A. R., Kachin, K. E., Constantino, M. J., Przeworski, A., Erickson, T., & CashmanMcGrath, L. (2002). Preliminary reliability and validity of the GAD-Q-IV: A revised self-report diagnostic measure of generalized anxiety disorder. Behavior Therapy, 33, 215–233. Richards, A., & French, C. C. (1992). An anxiety-related bias in semantic activation when processing threat/neutral homographs. The Quarterly Journal of Experimental Psychology, 45A, 503–525. Spielberger, C. D., Gorsuch, R. L., & Lushene, R. E. (1970). Manual for the state-trait anxiety inventory. Palo Alto, CA: Consulting Psychologist Press. Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics. (4th ed.). Needham Heights, MA: Allyn & Bacon.
Interpretive cues and ambiguity in generalized anxiety disorder Holly Hazlett-Stevens a,, T.D. Borkovec b b
a Department of Psychology/298, University of Nevada, Reno, NV 89557, USA Department of Psychology, Pennsylvania State University, University Park, PA 16802, USA
Received 23 January 2003; received in revised form ; accepted 7 July 2003
Abstract The current study investigated whether generalized anxiety disorder (GAD) individuals rely on antecedent information to interpret ambiguity and whether reliance on such preceding cues persists in the absence of potential threat. Twenty-six GAD and 23 nonanxious control college students performed a lexical decision task, using homographs (i.e. words with multiple meanings) as ambiguous primes. In half the trials, a homograph prime that possessed both threat-related, as well as neutral meanings was followed by a target word related to one of these two meanings. In addition, each ambiguous prime was immediately preceded by a series of four antecedent words that were either: (a) associated with the threatening meaning of the prime; (b) associated with the neutral meaning of the prime; or (c) unrelated to either meaning of the homograph, as well as the target. Homographs for which both meanings were neutral in valence comprised the other half of the trials. Effect size statistics suggest that GAD participants utilized the antecedent words to interpret the homograph primes with threat-related meanings, unlike their nonanxious counterparts (p < 0:06). When both meanings of the homograph prime were neutral in valence, the GAD group appeared deficient in the use of preceding information to interpret the ambiguous prime. # 2003 Elsevier Ltd. All rights reserved. Keywords: Generalized anxiety disorder; Information processing; Interpretation of ambiguity
Anxious individuals are more likely than nonanxious individuals to interpret ambiguous stimuli as threatening. In a comprehensive recent review of this literature, MacLeod (in press)
Corresponding author. Tel.: +1-775-784-6828; fax: +1-775-327-5043. E-mail address: [email protected] (H. Hazlett-Stevens).
0005-7967/$ - see front matter # 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0005-7967(03)00204-3
882
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
described several studies finding evidence for an interpretive threat bias among GAD samples. However, methodological limitations including experimenter demand characteristics, reliance on participant self-report, potential response selection biases, and possible memory biases were identified. Experimentally rigorous investigations of trait-anxious participants have addressed these concerns, but interpretive biases of GAD individuals have not been explored with these superior methods (MacLeod, in press; Aikins & Craske, 2001). In one such trait-anxiety study, Richards and French (1992) utilized homographs as ambiguous stimuli in a lexical decision task. In this experimental paradigm, words with both threatrelated and neutral meanings (threat/neutral homographs) were presented on a computer screen as ambiguous primes (e.g. batter, shot). Threat/neutral homograph primes were immediately followed by target words in the context of a lexical decision task (i.e. a task in which participants are instructed to decide whether the target string of letters is an actual word or a nonword). Each word target was consistent with one of the two meanings of the homograph or unrelated to the homograph. The speed with which participants responded to each target indicated the degree to which their performance on that trial was facilitated by the prime. Richards and French found that when presented with homograph primes containing both neutral and threatening possible meanings, high trait-anxious individuals accessed the threatening meanings more readily than their low trait-anxious counterparts (as evidenced by faster reaction times to the threat-related targets relative to unrelated targets). Similar lexical decision task results were found for test anxiety participants when ambiguous sentences served as primes (Calvo, Eysenck, & Estevez, 1994). MacLeod and Cohen (1993) presented high and low trait-anxious participants with sentence pairs, administered one sentence at a time at a self-paced rate. The first sentence of each pair was ambiguous, while the subsequent sentence disambiguated the first sentence consistent with either its threatening or its non-threatening meaning. Each trial was preceded by a cue word reflecting either meaning of the ambiguous sentence or by a cue word unrelated to the ambiguous sentence. Comprehension latency results suggested that high trait-anxiety participants were more prone than low trait-anxiety participants to interpret the ambiguous sentence as threatening. Of particular interest, MacLeod and Cohen also found a general slowing in comprehension latencies when the cue word suggested one meaning for an ambiguous sentence, but the continuation sentence reflected its other meaning. If anxious individuals are more likely than nonanxious individuals to interpret such ambiguity as threatening, they might also rely more on available preceding information to make their interpretations. Such a finding would extend previous interpretive bias research to more naturalistic conditions, as naturally occurring ambiguity in the environment rarely appears in the absence of preceding interpretive cues. The current investigation examined whether GAD participants were more likely than nonanxious participants to rely on preceding verbal cues when interpreting ambiguous stimuli with potentially threatening meanings. If GAD individuals are prone to interpret threat in ambiguous stimulus situations, they may also depend upon available antecedent information to make their interpretations when a threatening interpretation is possible. Second, if this reliance on available cues exists and functions to prepare an individual for threat, then GAD participants would not necessarily rely on such cues to interpret ambiguity under different conditions where no threat could be interpreted. Therefore, an additional aim was to investigate GAD and non-
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
883
anxious participants’ use of preceding cues when a threatening interpretation of the ambiguous stimulus was not possible. Rigorous methods similar to those used in trait-anxiety experiments were chosen. A modified version of the lexical decision task developed by Richards and French (1992) was designed. Crucially, the current study presented antecedent words before the homograph prime to examine their effects on the subsequent interpretation of this ambiguous word. In some trials, antecedent cues prepared participants for one meaning of the ambiguous word yet contained a target word reflecting its other meaning. A second set of trials in which both meanings of the ambiguous word were neutral in valence addressed whether an increased reliance on antecedent words for the GAD group was present only under conditions of potential threat.
1. Method 1.1. Participants A total of 1131 male and female undergraduate students enrolled in psychology courses at the Pennsylvania State University completed the Generalized Anxiety Disorder Questionnaire during a group testing session (GAD-Q-IV; Newman et al., 2002). The GAD-Q-IV is a reliable and valid self-report diagnostic measure of GAD that accurately identifies individuals meeting DSM-IV GAD criteria with 89% specificity and 83% sensitivity when compared to a clinician administered semi-structured diagnostic interview, the Anxiety Disorders Interview Schedule (ADIS-IV; Brown, DiNardo, & Barlow, 1994). Thus, the demonstrated reliability between the GAD-Q-IV and the ADIS-IV is comparable to that reported for two independently administered ADIS-IV interviews (Newman et al., 2002; Brown, DiNardo, Lehman, & Campbell, 2001). In addition, the GAD-Q-IV reliably distinguishes GAD from panic disorder and from social phobia (Newman et al., 2002). A total of 26 (three male, 23 female) participants meeting all DSM-IV GAD diagnostic criteria and 23 participants (two male, 21 female) meeting no diagnostic criteria on the GAD-Q-IV participated in the present study. All 49 participants identified English as their first language. Mean age was 19.98 years (SD ¼ 1:55) and mean education was 14.71 years (SD ¼ 1:21). 1.2. Materials (homograph word task)1 A pool of 120 threat/neutral homographs (i.e. homographs for which one meaning is threatening and one meaning is neutral in valence) were selected from the norms complied by French and Richards (1992) to be used as ambiguous primes. Each threat/neutral homograph had two meanings that differed significantly in terms of threat ratings from 12 independent judges (French & Richards, 1992). For each threat/neutral homograph, four words associated with its threatening meaning, four words associated with its neutral meaning, and four words that were unrelated in meaning to the homograph yet neutral in valence were selected as antecedent words. Trials in which the antecedent words were unrelated to the 1
Examples of stimuli from the homograph word task are available from the first author upon request.
884
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
homograph prime were included for replication of the naturally occurring interpretive threat bias reported by Richards and French (1992). Finally, for each threat/neutral homograph, one additional word associated with its threatening meaning and one additional word associated with its neutral meaning were selected for use as threat and neutral valence targets, respectively. Threat or neutral target words immediately followed and disambiguated each ambiguous (homograph) prime. A separate pool of 120 neutral/neutral homographs (i.e. homographs for which both meanings are neutral in valence) was selected from the norms compiled by Nelson, McEvory, Walling and Wheeler (1980) for use as neutral ambiguous primes. For each neutral/neutral homograph, four words associated with its primary meaning (i.e. the meaning which appears first in the compilation of norm data; Nelson et al., 1980), four words associated with its secondary meaning (i.e. a meaning which appears following the first definition in the compilation of norm data; Nelson et al., 1980), and four words that were unrelated to any meaning of the homograph yet neutral in valence were selected for use as antecedent words to be presented immediately before the neutral/neutral homograph prime. Finally, for each neutral/neutral homograph, one additional word associated with its primary meaning and one additional word associated with its secondary meaning were selected for use as primary and secondary targets, respectively. Primary or secondary target words immediately followed and disambiguated each ambiguous (homograph) prime. The two target words for each homograph were matched for frequency and word length. The neutral/neutral homographs were matched to the threat/neutral homographs for frequency and word length such that there were no significant differences between these two sets of homographs. Printed frequency data for each of these words were obtained from Francis and Kucera (1982). In total there were 240 trials per participant, with each trial consisting of a series of four antecedent words followed by a homograph prime followed by a target word or non-word. Nonword targets were used in half of the trials. For non-word trials, one letter of each target was changed to produce pronounceable non-word targets. The total 120 word trials consisted of 60 threat/neutral homograph primes and 60 neutral/neutral homograph primes. For the 60 threat/neutral homograph prime word trials, 10 trials were presented for each of six different combinations: (a) the neutral antecedent words, followed by the threat/neutral homograph prime, followed by the neutral target; (b) the neutral antecedent words, followed by the threat/neutral homograph prime, followed by the threat target; (c) the threat antecedent words, followed by the threat/neutral homograph prime, followed by the threat target; (d) the threat antecedent words, followed by the threat/ neutral homograph prime, followed by the neutral target; (e) the unrelated antecedent words, followed by the threat/neutral homograph prime, followed by the neutral target; and (f) the unrelated antecedent words, followed by the threat/neutral homograph prime, followed by the threat target. For the 60 neutral/neutral homograph prime word trials, 10 trials were presented for each of the following six different combinations: (a) the primary definition antecedent words, followed by the neutral/neutral homograph prime, followed by the primary definition target; (b) the primary definition antecedent words, followed by the neutral/neutral homograph prime, followed by the secondary definition target; (c) the secondary definition antecedent words, followed by
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
885
the neutral/neutral homograph prime, followed by the secondary definition target; (d) the secondary definition antecedent words, followed by the neutral/neutral homograph prime, followed by the primary definition target; (e) the unrelated antecedent words, followed by the neutral/neutral homograph prime, followed by the primary definition target; and (f) the unrelated antecedent words, followed by the neutral/neutral homograph prime, followed by the secondary definition target. For these stimuli, the ‘primary’ vs. ‘secondary’ definition simply referred to the order in which that meaning was presented in Nelson et al. (1980). Stimuli were originally counterbalanced such that each threat/neutral and neutral/neutral homograph would appear in each of its six possible trial variations equally across each group of six participants. Participants were randomized to counterbalancing group in blocks of six to maximize the balance of stimuli across the experiment as a whole as much as possible. Because there were three more participants in the GAD group, one extra GAD participant was run in three of the six counterbalancing order groups. The 240 trials were presented in a randomized order for each participant. Stimuli were presented and lexical decision reaction times were recorded using SuperLab program software on a Power Mac model Macintosh computer. 1.3. Procedure Participants initially completed a screening packet provided in their psychology classroom, consisting of the GAD-Q-IV, the Penn State Worry Questionnaire (PSWQ; Meyer, Miller, Metzger, & Borkovec, 1990), the trait version of the State-Trait Anxiety Inventory (STAI-T; Spielberger, Gorsuch, & Lushene, 1970), and the Beck Depression Inventory (BDI; Beck, 1978). The experimental session was conducted in a laboratory with a large writing table in the central room and a smaller adjoining room containing the computer. Each participant was tested individually for a 90-min session and remained in the experimental room throughout the experiment. One of three experimenters conducted each experimental session and was blind to each participant’s GAD/nonanxious group status. After informed consent was obtained, the state version of the STAI (Spielberger, Gorsuch & Lushene, 1970) was administered to provide an assessment of the anxiety level experienced at the time of the experimental session. Participants were then seated at the computer for the homograph word task. The experimenter explained that participants would be presented with strings of words on a computer screen, one word at a time. Participants were told that their task was to read each of the first five words silently to themselves, and then to indicate whether the sixth word was an actual word or not by pressing the corresponding key on the computer keyboard as quickly as possible. Participants performed a practice series of trials prior to the experiment proper. Each trial began with an asterisk in the center of the screen to serve as a signal for the participant to fixate centrally. After the participant hit the spacebar key, each of the four antecedent words were consecutively displayed centrally for 200 ms each, immediately followed by a homograph prime. Each homograph prime was presented centrally for 750 ms. This difference in presentation time was used so that participants could clearly discriminate
886
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
between the antecedent words and the homograph prime, both of which preceded the target. Immediately after the homograph prime, the target (word or non-word) appeared on the screen, overwriting the prime, and remained there until the participant made a manual response by pressing keys on the computer keyboard (right hand for words, left hand for non-words). Stimuli were presented in lower-case letters. Participants were given short breaks after each 80 trials. After the homograph word task was completed, participants were administered the Reading subtest of the Wide Range Achievement Test-Revised (WRAT-R) as a general measure of reading ability and overall intellectual ability. Responses only to the word trials of the lexical decision task were used in data analyses. Response times less than 300 ms or greater than 3000 ms were counted as errors. In addition, reaction times that exceeded 2.5 standard deviations above the mean for a given participant were discarded as outliers. Response errors and outliers were not included in data analyses, resulting in a loss of 4.4% of the data.
2. Results 2.1. Self-report anxiety and depression questionnaire measures To verify expected group differences on self-report measures, a multivariate analysis of variance (MANOVA) on the three screening measures (PSWQ, STAI-Trait version, BDI) and the experimental session questionnaire (STAI-State version) was performed with GAD group status as the between-subjects factor. Results indicated a significant multivariate effect, F ð4; 44Þ ¼ 33:08; p < 0:001. Univariate analyses revealed that this effect reflected a statistically significant difference between GAD and nonanxious groups on each of the four anxiety and depression questionnaire measures in the predicted direction: PSWQ, F ð1; 47Þ ¼ 118:02; p < 0:001; STAI-Trait, F ð1; 47Þ ¼ 50:52; p < 0:001; BDI, F ð1; 47Þ ¼ 21:23; p < 0:001; STAIState, Fð1; 47Þ ¼ 55:74; p < 0:001 (see Table 1 for means and standard deviations). 2.2. Wide range achievement test-revised To establish equivalent intellectual ability between the two groups, a t-test on the raw scores from the Reading subtest of the WRAT-R with GAD group status as the between-subjects factor was conducted. No significant differences between the groups on this control measure were found [tð1; 47Þ ¼ 0:41; p > 0:5]. Table 1 Means and standard deviations for self-report anxiety and depression questionnaire measures for GAD and nonanxious control groups Group
PSWQ M
GAD 65.96 Nonanxious 35.57
STAI-Trait
BDI
STAI-State
SD
M
SD
M
SD
M
SD
9.43 10.15
52.46 34.30
9.35 8.38
16.38 4.30
11.28 5.88
48.15 32.13
8.60 6.01
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
887
2.3. Homograph word task reaction times 2 2.3.1. Unrelated antecedent word trials Analysis was first conducted for those trials involving only unrelated antecedent words plus the threat/neutral homographs to determine whether GAD participants responded faster to threat targets than to neutral targets when compared to nonanxious participants. Interpretation of the ambiguous homograph prime was not expected to be impacted by the preceding four antecedent words for these trials. This approach allowed for replication of the interpretive threat bias reported by Richards and French (1992). Reaction times in ms for correct responses to word targets were averaged over trial type for each participant. A target (threat, neutral) by group (GAD, nonanxious) repeated measures ANOVA was conducted on the mean reaction times of correct responses to word targets. Neither a main effect of target (p > 0:2) nor a group target interaction effect (p > 0:5) was found. A main effect of group approached significance [F ð1; 47Þ ¼ 3:31; p < 0:08, partial eta squared g2 ¼ 0:066, observed power ¼ 0:429], in which GAD participants responded slower (M ¼ 939:04; SD ¼ 316:99) than nonanxious participants (M ¼ 793:85; SD ¼ 227:97) to the targets overall. To verify an absence of interpretive differences for those trials involving only unrelated antecedent words plus the neutral/neutral homographs, a target (primary, secondary) by group (GAD, nonanxious) repeated measures ANOVA was conducted on the mean reaction times of correct responses. As expected, no main effect of target and no group x target interaction effect were found (both p’s > 0:3). However, a main effect of group approached significance [F ð1; 47Þ ¼ 3:44; p < 0:07, partial eta squared g2 ¼ 0:068, observed power ¼ 0:444], in which GAD participants again responded slower (M ¼ 941:89; SD ¼ 323:14) than nonanxious participants (M ¼ 788:78; SD ¼ 242:58) to the targets overall. 2.3.2. Related antecedent word trials To test the main study hypotheses regarding the effects of the antecedent words on subsequent ambiguous homograph prime interpretation, a series of analyses were conducted for the threat and neutral antecedent word trials of the threat/neutral homograph trials and for the primary and secondary antecedent word trials of the neutral/neutral homograph trials. It was predicted that for the threat/neutral homograph trials, the GAD group would show greater disruption than the nonanxious group on trials containing a mismatch between the meaning implied by the antecedent words and the meaning suggested by the target (compared to trials in which the antecedent words and target suggested the same meaning). Thus, the GAD participants were expected to exhibit a slowing on threat/neutral homograph trials when the antecedent word and target meanings were different than when the meanings reflected in each were the same, regardless of valence. Reaction times for correct responses to word targets were aver2
Inspection of histograms revealed that reaction time values used in these analyses were moderately positively skewed. Therefore, data were transformed using the square root data transformation procedure described by Tabachnick and Fidell (2001), and repeated measures ANOVA analyses were repeated with transformed data. Results obtained were similar to those reported above. This similarity between untransformed data results and transformed data results was not surprising, given that all reaction time variables were skewed in the same direction to about the same moderate extent (see Tabachnick & Fidell, 2001: 81).
888
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
aged over trial type for each participant. T-tests confirmed no difference between neutral antecedent words/neutral target trials and threat antecedent words/threat target trials and no difference between neutral antecedent words/threat target trials and threat antecedent words/neutral target trials for either group (all p’s > 0:4). Therefore, trials were collapsed into either same antecedent word-target trials or into different antecedent word-target trials to reflect the relationship between the meaning suggested by the antecedent words and that suggested by the target. An antecedent word (same, different) by group (GAD, nonanxious) repeated measures ANOVA on the mean reaction times for correct responses to word targets presented following the threat/neutral homograph primes revealed a marginally significant main effect for group status [Fð1; 47Þ ¼ 3:67; p < 0:06], qualified by an antecedent word group interaction approaching significance [F ð1; 47Þ ¼ 3:59; p < 0:06, partial eta squared g 2 ¼ 0:071, observed power ¼ 0:458]. Due to the observed limitations in power, three follow-up comparisons were conducted to examine the nature of this interaction trend. First, it had been predicted that the GAD participants would exhibit a slowing for different antecedent word trials compared to same antecedent word trials. However, the difference in this predicted direction only approached significance (p < 0:08). Second, it had been predicted that the nonanxious group would not exhibit any differences between same and different antecedent word trials for the threat/neutral homograph trials. Indeed, no such effect was found (p > 0:4). Third, it had been predicted that the difference in reaction time between different and same antecedent word trials would be greater for the GAD group than for the nonanxious group. While this effect only approached statistical significance [tð1; 47Þ ¼ 1:89; p < 0:06Þ], calculations of effect size revealed a mediumsize effect, d ¼ 0:528 (Cohen, 1988) (see Table 2 for mean reaction times, standard deviations, and difference scores). An analogous series of analyses were then conducted for related antecedent word trials from the neutral/neutral homograph trials. It had been predicted that for the neutral/neutral homograph trials, in which the ambiguous homograph prime had no threatening meanings, the nonanxious group would show greater disruption than the GAD group on trials containing a mismatch between the meaning implied by the antecedent words and the meaning suggested by the target (compared to trials in which the antecedent words and target suggested the same meaning). Thus, the nonanxious participants were expected to exhibit a slowing on neutral/neutral homograph trials when the antecedent words and target meanings were different than when the meanings reflected in each were the same. T-tests confirmed no difference between primary Table 2 Means and standard deviations of reaction times and mean difference scores (in ms) for the same and different antecedent words plus the threat/neutral homograph prime trials for GAD and nonanxious control groups Group
GAD Nonanxious
Same meaning for antecedent words and target
Different meaning for antecedent words and target
Different meaning–same meaning difference score
M
SD
M
SD
M
907.22 789.67
260.75 239.56
944.22 777.46
324.00 205.60
37.00 12.20
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
889
Table 3 Means and standard deviations of reaction times and mean difference scores (in ms) for the same and different antecedent words plus the neutral/neutral homograph prime trials for GAD and nonanxious control groups Group
GAD Nonanxious
Same meaning for antecedent words and target
Different meaning for antecedent words and target
Different meaning–same meaningdifference score
M
SD
M
SD
M
921.71 760.00
340.00 230.23
908.66 807.25
278.32 263.54
13.05 47.26
meaning antecedent words/primary meaning target trials and secondary meaning antecedent words/secondary meaning target trials for either group, and no difference between primary meaning antecedent words/secondary meaning target trials and secondary meaning antecedent words/primary meaning target trials for either group (all p’s > 0:2). Therefore, trials were collapsed into either same antecedent words-target trials or into different antecedent words-target trials to reflect the relationship between the meaning suggested by the antecedent words and that suggested by the target. An antecedent word (same, different) by group (GAD, nonanxious) repeated measures ANOVA was conducted on the mean reaction times for correct responses to word targets presented following the neutral/neutral homograph primes. A significant antecedent word x group interaction effect emerged [Fð1; 47Þ ¼ 6:00; p < 0:05, partial eta squared g2 ¼ 0:113, observed power ¼ 0:669]. Three follow-up analyses were conducted to explore the nature of this interaction effect. First, it had been predicted that the nonanxious participants would exhibit a slowing for different antecedent word trials compared to same antecedent word trials, and this hypothesis was supported (p < 0:001). Second, it had been predicted that the GAD group would not exhibit any differences between same and different antecedent word trials for the neutral/neutral homograph trials. Indeed, no such effect was found (p > 0:5). Third, it was predicted that the difference in reaction time between the different and same antecedent word trials would be greater for the nonanxious group than for the GAD group, and this effect was also found [tð1; 47Þ ¼ 2:45; p < 0:05; d ¼ 0:667) (see Table 3 for mean reaction times, standard deviations, and difference scores).
3. Discussion Richards and French (1992) demonstrated that trait-anxious participants show an interpretive bias in favor of threatening meanings when confronted with threat/neutral homograph primes as ambiguous stimuli. The present study aimed to replicate this effect in a GAD sample and to extend previous findings by presenting interpretive cue words immediately before the ambiguous prime. The current investigation did not find evidence of an interpretive bias when potentially threatening homograph primes were preceded merely by unrelated neutral antecedent words. This discrepancy between the Richards and French study and the current experiment may have resulted from methodological differences, as the unrelated antecedent words (or the embedded
890
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
trials involving related antecedent words) may have inadvertently influenced responses to the unrelated antecedent word trial targets. Effect size statistics from the trials involving related antecedent words plus the threat/neutral homographs suggested that GAD individuals may have utilized the antecedent words more so than their nonanxious counterparts to interpret the ambiguity of the threat/neutral homographs. Unlike the nonanxious group, the GAD group exhibited a slowing when the meaning reflected in the antecedent words differed from the meaning reflected by the target. Although this effect was of medium size (d ¼ 0:528; Cohen, 1988), it failed to reach statistical significance (p < 0:06). It therefore cannot be determined whether this finding reflects an actual phenomenon that went undetected due to insufficient statistical power or whether the null hypothesis provides a more accurate interpretation. Even if this finding were reliable, it does not address whether the GAD group’s tendency to rely on preceding information to interpret ambiguity persists beyond the defensive preparation for potential threat. Therefore, neutral/neutral homograph trials have been included to examine whether the GAD participants’ tendency to utilize interpretive cues (if found) was specific to stimuli conditions of potential threat interpretation. When the neutral/neutral homograph data are considered, a different picture emerges. Analyses here revealed that reaction times for nonanxious participants depended on whether the antecedent words reflected the same meaning or a different meaning of the ambiguous word as the target did. In contrast, GAD participants failed to utilize the antecedent words to interpret the neutral/neutral homographs, as evidenced by equivalent reaction times when the meaning reflected in the antecedent words differed from the meaning reflected by the target compared to trials in which the antecedent words and target meanings were consistent. This pattern of results suggests that while the GAD group may have used the antecedent words to interpret ambiguous words to anticipate threat, this strategy to interpret ambiguity was abandoned when the potential for threat was removed. Equivalent scores on the WRAT-R Reading subtest suggest that these group differences are not a function of reading ability or overall intellectual functioning. In addition to statistical power concerns, other limitations should be considered. First, participants were selected on the basis of a self-report diagnostic measure (GAD-Q-IV) rather than a clinician-administered diagnostic interview (e.g. ADIS-IV). Many clinicians and researchers consider clinician-administered diagnostic interviews the gold standard of clinical assessment. However, the GAD-Q-IV yields strong reliability when compared to the ADIS-IV (Newman et al., 2002), equivalent to that reported for two independently administered ADIS-IV interviews (Brown, DiNardo, Lehman & Campbell, 2001). Furthermore, PSWQ scores for the GAD group (M ¼ 65:96; SD ¼ 9:43) were comparable to those reported for treatment-seeking clinical GAD samples diagnosed by ADIS-R interviews (M ¼ 67:66; SD ¼ 8:86; Molina & Borkovec, 1994). Nevertheless, the current sample consisted of college students seeking course credit rather than treatment for their participation. Due to differences in motivation and demographic composition, these results may not generalize to treatment-seeking community populations. A second limitation concerns our inability to discriminate between two possible explanations for the effects of the antecedent words. This experimental approach assumes that a mismatch between the meaning reflected in the antecedent words and that reflected in the target should slow reaction time to the extent the participant used the antecedent words to interpret the ambiguous homograph prime. It is possible that these antecedent words thus served as contex-
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
891
tual stimuli, and that the GAD group relied on the experimental context less than the nonanxious group in the absence of potential threat interpretation (i.e. neutral/neutral homograph trials). Word stimuli were presented such that participants could clearly discriminate between the antecedent words and the homograph prime, both of which preceded the target. The alternative hypothesis is that the antecedent words provided an additional priming effect for the homograph prime itself, resulting in a double-priming effect in the related antecedent word trials. Thus, GAD and nonanxious individuals may have differed in the degree to which this further priming effect was present across the two homograph types. GAD may be characterized by increased use of interpretive cues when facing potentially threatening ambiguous situations, but the use of such information to interpret other ambiguous situations appears deficient for these individuals. If most daily life situations for a given GAD individual cannot be interpreted as threatening, a failure to process other relevant non-threatening information may be as important to the maintenance of anxiety as his or her periodic interpretive threat biases. This could leave anxious individuals at a disadvantage in numerous everyday situations that do not involve potential threat.
Acknowledgements Preparation of this manuscript was supported in part by an American Psychological Association Dissertation Research Award and a Penn State Research and Graduate Studies Dissertation Support Grant awarded to the first author and by National Institute of Mental Health Grant MH-39172 awarded to the second author. This research is based on the dissertation of the first author. Portions of this paper were presented at the 36th annual convention of the Association for Advancement of Behavior Therapy (AABT), Reno, NV, November, 2002. The authors gratefully acknowledge Andrew Mathews for his helpful comments on an earlier draft, Judy Kroll for her assistance with the design and implementation of the lexical decision task, Mia Peteritas for her assistance with the collection and management of data, and Michael Mitchell for his assistance with data analyses.
References Aikins, D. E., & Craske, M. G. (2001). Cognitive theories of generalized anxiety disorder. The Psychiatric Clinics of North America, 24, 57–74. Beck, A. T. (1978). Depression inventory. Philadelphia: Center for Cognitive Therapy. Brown, T. A., DiNardo, P. A., & Barlow, D. H. (1994). Anxiety disorders interview schedule for DSM-IV (ADIS-IV). San Antonio, TX: The Psychological Corporation. Brown, T. A., DiNardo, P. A., Lehman, C. L., & Campbell, L. A. (2001). Reliability of DSM-IV anxiety and mood disorders: Implications for the classification of emotional disorders. Journal of Abnormal Psychology, 110, 49–58. Calvo, M. G., Eysenck, M. W., & Estevez, A. (1994). Ego-threat interpretive bias in test anxiety. Cognition and Emotion, 8, 127–146. Cohen, J. (1988). Statistical power analysis for behavioral sciences. (2nd ed.). New York: Academic Press. Francis, W. N., & Kucera, H. (1982). Frequency analysis of English usage. Boston: Houghton Mifflin. French, C. C., & Richards, A. (1992). Word association norms for a set of threat/neutral homographs. Cognition & Emotion, 6, 65–87.
892
H. Hazlett-Stevens, T.D. Borkovec / Behaviour Research and Therapy 42 (2004) 881–892
MacLeod, C. (in press). Information processing approaches to generalized anxiety disorder: Assessing the selective functioning of attention, interpretation and memory in GAD patients. In R. G. Heimberg, C. L. Turk, & D. S. Mennin (Eds.), Generalized anxiety disorder: Advances in research and practice. New York: Guilford Press. MacLeod, C., & Cohen, I. L. (1993). Anxiety and the interpretation of ambiguity: A text comprehension study. Journal of Abnormal Psychology, 102, 238–247. Meyer, T. J., Miller, M. L., Metzger, R. L., & Borkovec, T. D. (1990). Development and validation of the Penn State Worry Questionnaire. Behaviour Research and Therapy, 28, 487–495. Molina, S., & Borkovec, T. D. (1994). The Penn State Worry Questionnaire: Psychometric properties and associated characteristics. In G. Davey, & F. Tallis (Eds.), Worrying: Perspectives on theory, assessment and treatment (pp. 265–283). New York: Wiley. Nelson, D. L., McEvory, C. L., Walling, J. R., & Wheeler, J. W. (1980). The University of South Florida homograph norms. Behavior Research Methods & Instrumentation, 12, 16–37. Newman, M. G., Zuellig, A. R., Kachin, K. E., Constantino, M. J., Przeworski, A., Erickson, T., & CashmanMcGrath, L. (2002). Preliminary reliability and validity of the GAD-Q-IV: A revised self-report diagnostic measure of generalized anxiety disorder. Behavior Therapy, 33, 215–233. Richards, A., & French, C. C. (1992). An anxiety-related bias in semantic activation when processing threat/neutral homographs. The Quarterly Journal of Experimental Psychology, 45A, 503–525. Spielberger, C. D., Gorsuch, R. L., & Lushene, R. E. (1970). Manual for the state-trait anxiety inventory. Palo Alto, CA: Consulting Psychologist Press. Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics. (4th ed.). Needham Heights, MA: Allyn & Bacon.