Interpretation Bias Featured in Military Personnel with High Trait Anxiety Measured by a Novel Paradigm

Background: Anxiety has become one of the most common psychological problems affecting the combat effectiveness of soldiers. As the generation, maintenance, and recurrence of anxiety have an important interaction with interpretation bias, yet none proof was for the existence of interpretation bias in military personnel. Methods: 112 military ocers and soldiers were recruited. Based on scores of the Trait-anxiety Inventory, participants were divided into the high trait anxiety group and the low trait anxiety group. the Picture Sentence Association Paradigm comprised of military-simulated ambiguous scenarios and emotional facial expressions was used to test the differences of the interpretation bias between the two groups. Results: Military personnel with high trait anxiety showed interpretation bias by endorsing more negative valence to the ambiguous scenarios and reject the positive valence. Especially in a self-related scenario, the interpretation bias was more remarkable. Conclusion: This study revealed the existed interpretation bias in military personnel with high trait anxiety using a new paradigm and highlighted the need for further researches to improve the measurement of interpretation bias. Moreover, the picture sentence association paradigm could provide plausible methods for cognitive bias modication to decrease the soldiers’ anxiety.

attention quickly (attention bias), their interpretation will be biased towards danger (interpretation bias), and they will be primed in memory (memory bias) according to the theory of cognitive bias [8,9]. For instance, the model for social anxiety has been researched widely, which posits that individuals with elevated social anxiety tend to demonstrate negative biases in processing social cues that are indicative of negative evaluation [9].
Namely, these cues from social interaction soliciting neutral or positive emotion in normal people would be partially visualized as fear or worry for socially anxious people. In laboratory studies, the subjects are presented with a kind of vague story material or life scene (employing sentences, picture or sound or others as ambiguous stimuli) and then they are invited to make a tendentious explanation of the ambiguous situation through the self-report method (selection of different explanations, grading of different explanations, open questionnaire or interview, etc.). The results of these studies showed that anxious people were more likely to make threatening inferences than non-anxious ones who were more likely to anticipate positive outcomes [10]. Clinic researches have been focused on cognitive bias modi cation (CBM), based on the correction of interpretation bias of anxious people proved effective in decrease the trait anxiety level within a long-time intervention [5]. Mathews, Ridgeway, Cook, and Yiend increased CBM from a single session to four sessions and assessed trait anxiety one week later [11]. High trait anxious individuals completed a CBM program that presented ambiguous scenarios, each of which resolved in an increasingly positive manner over the four sessions, while the control group completed only a pre-assessment and post-assessment two weeks later. Results showed that the active group's interpretation was more positive and less negative than the control group at post-assessment. More importantly, one week following the post-assessment the active group had signi cantly lower trait anxiety scores than the control group. Thereby, supported by laboratory studies and clinic practices, it is safe to say the generation, maintenance, and recurrence of anxiety disorder have signi cant interaction with interpretation bias.

Trait anxiety related to interpretation bias
In Spielberg's view, anxiety is dimerized by trait anxiety and state anxiety according to the variability and stability of anxiety [12]. Trait anxiety is a stable personality trait, while state anxiety is a temporarily emotional situation affected by the autonomic nervous system [13]. People with trait anxiety perceive the surrounding environment as a threat and induce more anxious feelings through self-evaluation and they are in uenced by an individual's internal psychological stress, however, the intensity of state anxiety is not so stable and is more vulnerable to the external environment [14]. Trait anxiety can be understood as a generalization of the frequency and intensity of past state anxiety [14]. Noteworthy, Beard and Amir (2010) investigated whether interpretation bias mediated the relationship between trait social anxiety and state anxiety in response to a social evaluative threat [15]. They invited undergraduate students with high social anxiety to attend experimental sessions where students completed measures of trait social anxiety and an Interpretation Questionnaire followed by an impromptu speech and a state anxiety rating. Results revealed that participants' rankings of the negative interpretations of ambiguous social scenarios mediated the relationship between trait social anxiety and state anxiety in response to the impromptu speech. Meanwhile, in the context of the military, a study has shown the negative cognitive bias correlated with mental health and trait anxiety of Chinese plateau military personnel, indicating the importance of the interpretation bias in the biased cognitive processing of soldiers. Hence, it is plausible to hypothesize that trait anxiety is reciprocally related to interpretation bias.
The adaptation of design for measuring interpretation bias Page 4/16 Most of the studies and treatment have been focused on adolescents or college students and none proof for the existence of interpretation bias in military personnel who undoubtedly experience different stress environment engendering trait anxiety and vulnerability to anxiety disorder. Therefore, it is still unfolded the relation between trait anxiety of military personnel and interpretation bias in ambiguous scenarios on the background of military environments. Furthermore, whether intervention on interpretation bias of the military personnel with high trait anxiety could be effective to guide the cognitive bias modi cation into military training, which aims to reduce the level of trait anxiety and improve the treatment for the anxiety disorder, is far from certain. Hence, to nd whether trait anxiety would be associated with interpretation bias in ambiguous scenarios on the background of military environments and provide clinical implications for psychological help for anxious military personnel, we adopted a new method to perform this research. Referred to the previous studies, the measures of interpretation bias have been designed primarily on the assessment of potential threat (e.g., whether the stimulus/scenario is negative, and the likelihood of a negative outcome), which repeatedly revealed that individuals with anxiety interpret ambiguous scenarios more negatively than do controls [16]. Later, Beard and Amir designed the Word Sentence Association Paradigm (WSAP) to study the role of interpretation bias in anxiety by asking subjects to complete the last word of emotional valence for a paragraph describing an ambiguous scenario [15]. Since then, WSAP becomes a typical paradigm for the following studies to measure the interpretation bias. Nonetheless, recently researches found in the process of interpreting ambiguous information the mental image would affect the mood and emotional valence [17], therefore, other paradigms also combined picture or daily experiences as the reaction choice, such as the cognitive bias modi cation based on imagery (CBM-I), and Picture Sentence Association Paradigm (PSAP). PSAP requires participants to identify whether the followed facial expressions (positive or negative) matched with scenarios instead of judgment to words. The facial expressions of positive or negative emotions are allowed for strong external validity and a ne-grained analysis of interpretation biases, which renders PSAP could advantage WSAP in exploring the association between trait anxiety and interpretation bias when resolving valence ambiguity of scenarios based on the intercourse of people. To ensure the validity of our research, we adopted PSAP and redesigned the scenarios on the background of military environments. Notably, the self-involvement in the scenarios was associated with the triggering of interpretation bias and effects on mental imagery [18,19]. To distinguish self-related and non-self-related ambiguous situations, most studies used "you" in the description of self-related scenes to increase the subject's self-involvement [16,20], while the subject was modi ed to refer to a speci c name of another person in non-self-related scenes. In this study, "I" was used in the self-related military-simulated ambiguous scenarios, and "company" or "comrade in arms" were used to refer to non-self-related military-simulated ambiguous scenarios, aiming to explore interpretation bias in the two different scenarios and the effect of self-involvement on interpretation.
As a special study, we invited the military personnel as our research subjects and explore the characteristics of interpretation bias to ambiguous scenarios in soldiers with trait anxiety. We hypothesized that trait anxiety is closely related to the interpretation bias of military personnel. We hope this exploration can not only enrich and improve theoretical knowledge of interpretation bias but also provide the cognitive processing model of anxious soldiers through novel experimental methods.

Participant
The convenient sampling method was used in recruitment. 112 o cers and soldiers from a certain group army and a certain coastal defense brigade were selected to take the trait anxiety questionnaire. The speci c population composition is shown in Table 1. High social anxiety and low social anxiety groups were identi ed from this screening sample based on Trait-anxiety scale scores on the State-Trait Anxiety Inventory (TAI).
Individuals who scored in the top 27% (TAI total > = 45) were recruited as "high trait anxiety"( 5 people tied for 45 were all included in the high group), individuals who scored in the bottom 27% (TAI total < = 36) were recruited as "low trait anxiety" participants (5 people tied for 36 were all included in the low group). In this way, 34 people were enrolled in the "high trait anxiety" group; 33 people were enrolled in the low one. Research has indicated that using such analog groups based on trait anxiety measures is a viable means for studying processes present in anxious symptoms [21]. The sample was all males in the troop with an average age of 20.
Groups did not differ in age, education, position, marriage, and family background. The demographic information for the two groups was presented in Table 1. In this process of performing tests, three participants in the high anxiety group were interrupted because of duty call, while four of the low anxiety group was interrupted. Thus, we eliminated the seven subjects' experimental data. The high anxiety group was consisted of 31 people, with 29 people in the low anxiety group. All subjects were male, right-handedness, with normal vision or corrected vision, and without mental illness. Table 1 Demographic information and interpretation differences in the high and low trait anxiety group. comments, the scenarios were revised and again assessed the ambiguity by 285 soldiers on a -5 to 5 scale (-5 was equal to the most negative meanings and 5 for the most positive meanings). Eventually, according to the recommended criterion reported by Zhu et al. [20], 81 ambiguous scenarios were selected with the emotion valence scores between 3.05-4.9 (standard error between 0.447 and 1.930), for example, "The commander told me to go to his o ce" appraised 4.9 points, "Before the training, the monitor said to discuss some problems with me after the dismissal" appraised 4.15, and "Commanders and instructors often disagree" appraised 3.05. The Cronbach coe cient of the questionnaire was 0.958 indicating good internal consistency.
In the following study, 40 representative sentences were selected from the ambiguous scenarios, among which 20 were self-related and 20 were non-self-related. Meanwhile, the split-half reliability of the questionnaire was 0.912.

Emotional facial expression
Emotional faces were selected from the Chinese Facially Emotional Picture System (CFAPS) revised by Bai et al [22]. There were 200 negative, neutral, and positive faces, with 100 male and 100 female faces. The faces have been proved to have high reliability in emotional aspects of pleasure, arousal, dominance, and attraction, and it is a good picture material for domestic local emotion research and cross-cultural emotion comparison research. All the images in the system are black and white with a size of 6.5 cm*7.5 cm and a resolution of 102 pixels/inch.

Trait Anxiety Level
Participants completed the trait form from the State-Trait Anxiety Inventory (STAI) [13,23], which consists of 20 items assessing symptoms of trait anxiety and has adequate psychometric properties (ranges from .73 to .86). The Cronbach coe cient of the trait-anxiety questionnaire was 0.751.

Experiment Paradigm
The experimental Paradigm was the "Picture Sentence Association Paradigm". The experiment was presented by E-Prime-2.0 software. The speci c procedure was as follows: in the screen, the "+" sign was rst presented to arouse the attention of the subject within 500 ms, and then a military-simulated ambiguous scenario was

Procedure
The experiment was carried out in the psychological relaxation rooms of troops (controlled as the experimental site) with a mild and suitable temperature and dimly illuminated. Three psychologist assistants maintained the order of the site, ensured the experimental environment to be quiet, and no distraction in and out of each room where one participant was performing the computer tasks. To ensure the consent from the participants, the experimenter would inform the subjects of the anonymous experimental task was designed for investigation about their anxiety trait and further contribution to bene t the work of psychological aid for soldiers. After the informed consent was attained from the soldiers, a total task of 45 cycles was conducted, including ve cycles for the practice sessions to ensure subjects familiar with the experiment before entering the experimental block. The experimental computer screen uni ed the black background and white character, the picture was black-white. Since it was not easy to recruit the participant again in the army, and better to reduce the disturbance to their daily training in the army, we asked the subjects to experiment directly after the questionnaire test, and only analyzed the experimental data of the selected subjects in the later stage when all of them nished the task. The resolution of the experimental computer screen was 1024*768hz. The

Results
The interpretation differences between the high and low trait anxiety group From the analysis of the comparison in the two groups showed in Table 1, in the responses to the ambiguous scenarios, there was no signi cant difference in the endorsement ratio of positive faces (calculated by the frequency of pressing the "F" button when the positive-valence face presented after the described scenarios) between the high trait anxiety group and the low trait anxiety group (F = 3.539, p = 0.065); however, as for the endorsement ratio of the negative face (calculated by the frequency of push the "F" button when the negativevalence face presented after the described scenarios), the high trait anxiety group rated signi cantly higher than that of the low trait anxiety group (F = 5.878, p = 0.018). A signi cant difference between the two groups also existed in the rejection ratio, which was calculated by frequency of pressing the "J" button when the valence face presented, either for positive (F = 4.488,p = 0.039) or negative faces (F = 4.799, p = 0.033 ). These signi cances indicated that the high trait group tended to interpret the ambiguous scenarios with negative emotion valences as they showed more endorsement and lower rejection for negative faces and higher rejection of positive faces than the low anxiety group did.
As for the reaction time, one adopted concept of bias scores advocated by Bear and Amir were applied and was proved to provide a more convenient way to compare reaction time and self-report indexes. The calculated bias scores for the ambiguous scenarios were formed as below steps: Rt represents the reaction time for pressing the bottom "F" or "J". In this sense, the larger bias scores are the more tendency toward negative interpretations and away from positive interpretations. On the whole, there was a signi cant difference in the positive bias score between the high anxiety group and the low anxiety group (t=-2.217, p = 0.031), but no signi cant difference in the negative bias score (t=-0.984, p = 0.329) as shown in Fig. 2, suggesting the higher anxiety group had more tendency to reject positive interpretations and less likely to endorse them instead of biasing negative interpretation.

The effect of self-involvement in the military-simulated ambiguous scenarios on the relation between anxiety and interpretation bias
The ANOVAs were conducted with Group (higher and lower trait anxiety) as the between-group factor and Emotional face valence (positive and negative) and Scenario type (self-related, non-self-related) as withingroup factors. Meanwhile, the frequency of pressing "F" related to the scenarios by participants was calculated and deemed as the dependent variable. The attained results were presented in Table 2. Mauchly's Test of Sphericity showed Mauchly's W for the interaction was equal to 1 indicating the error covariance matrix of the orthonormallized transformed dependent variables was proportional to an identity matrix.  Table 3 showed These suggested military personnel with low trait anxiety showed positive interpretation bias while those with high trait anxiety did not possess the positive bias but also showed a tendency to endorse the negative valence of ambiguous scenarios; however, all these signi cances were con ned in the self-related scenarios. Table 2 Analysis of variance of endorsement ratio of the emotional face in two groups based on different selfinvolvement ambiguous scenarios. Table 3 The simple effect of the Group on the interaction between self-involvement and emotion face valence.

Discussion
This study examined interpretation bias in the military personnel with different levels of anxiety using a novel paradigm that combined military ambiguous scenarios as backdrops and emotional faces as responding stimuli. Both the analysis of responses and reaction time supported the existence of interpretation bias in military personnel with high trait anxiety. Especially, the results of responses to emotion faces, calculated by the ratios of endorsement and rejection for different emotion valence (positive and negative), were signi cantly different in interpretation patterns between the high trait anxiety group and the low trait anxiety group. In line with the previous studies [24][25][26], military personnel with high trait anxiety endorsed more negative valence to the ambiguous scenarios and had more di culty in rejecting the negative interpretation, besides, they were also vulnerable to rejecting the positive emotion valence for the scenarios. Notably, the study utilized the facial expressions of positive and negative emotion for strong external validity and conducted a ne-grained analysis of interpretation biases to ambiguous scenarios adapted to the military environment which facilitated our subjects' understanding and full imagery. This is a novel study to prove that the interpretation bias to the self-related scenario was remarkable in military personnel with high trait anxiety and our result is in line with previous ndings that negative interpretation is closely related to anxiety [27,28].
Prior studies have acknowledged that anxious people experience an enhanced sense of insightfulness but greater pessimism about positive events and generate fewer effective solutions to interpersonal problems and positive responses to imagined problems [29,30]. Besides, pieces of evidence from memory tasks with thought-induction procedure proved that anxious-related disruption was found in remembering following the self-focused but not the other-focused thought induction [31]. In the non-self-related scenarios, the effect of interpretation with emotion valence was not such strong, suggesting in our cohort trait anxiety had no signi cant effect on understanding emotion face valence and empathizing with others. Similar studies also proved that if only emotional facial expression from other people with happy or disgust was presented to high social anxiety participants, their sensitivity to perceiving negative evaluation did not demonstrate [32].
Moreover, if the scenarios were not related to self-interaction with other people like some ambiguous stimuli with homographs, there was no signi cant negative interpretation bias in the social anxiety group [7].
Therefore, we proposed that the military personnel with high trait anxiety seemed to have more self-focused thought in the interpretation of environmental stimuli and predisposed to generate the negative bias which in turn generated the state anxiety and enhanced the trait anxiety. However, as an on-line measurement, it is di cult to directly compare the current results to previous studies because the reaction times were obtained through different tasks and re ect different processes. Thus, it is not surprising that the current results differ from previous studies, as they suggest that differences in response time regarding positive and negative interpretations are important in social anxiety. However, we calculated the reaction time data with the bias score to indicate the expected interpretation bias, and we were novel to nd that only bias scores for the positive face were signi cantly associated with an anxious level in our cohort, but the negative bias scores were not signi cantly different in the two groups. These explained that the anxious people's cognitive de ciency to recognize the positive cues, which was acknowledged in previous studies about social anxiety [25,33,34].
As a novel paradigm of combining ambiguous scenarios with positive and negative faces, procedural differences were inevitable. In terms of mixed reaction time ndings, we did not present a positive or negative prime like the WASP procedure which presented a threat or benign prime followed by an ambiguous sentence.
In real life, anxious people usually do not have a prime before they encounter various stimuli. The prime activated cognitive processes involved in interpretation (e.g. negative beliefs) that then in uenced the interpretation of an ambiguous sentence and the difference in reaction time data were signi cantly presented in the experiment [15,35,36]. However, in our study, the reaction data re ected the interpretation of the scenarios and the recognition of the emotion faces. Participants were allowed unlimited time to judge the relatedness of the scenarios and faces, therefore, it was di cult to demonstrate the on-line results of interpretation bias and control all the extremum in the responses. In this case, separated reaction times were not as meaningful as the compared results of bias scores. However, within the high trait anxiety group, the average reaction time also supported the negative interpretation bias as readiness to the negative faces and slow to the positive faces. It is also important to mention that using faces instead of words or other forms as response simpli ed the cognitive process and shorten the time for understanding, therefore, the reaction data of recognizing the face way would be closer to the real situation. Assumed that 50% represents a baseline endorsement level, the low trait anxiety group's negative endorsement was low (41%) and their positive endorsement was high (53%), while the rejection ratios were reversed for the negative (58%) and positive (47%). The high trait anxiety group's negative and positive endorsement levels (54% and 44% respectively) and rejection levels (47% and 56%) were both closer to baseline. These ndings suggest that the lack of positive bias and the presence of negative bias should be conceptualized as separate constructs, which are also advised by Beard [15] and Huppert [27]. Therefore, it may be more accurate to associate control status with a lack of a negative bias and the presence of a positive bias, rather than associating trait anxiety with bias. Besides, the results from our study also suggested that treatments should target both negative interpretation bias and the lack of benign interpretation bias rather than target exclusively threat interpretations. People with high trait anxiety might bene t from endorsing the positive interpretation of a situation similar to they would reject the positive interpretation, especially the situation or problem with their issues. It also suggested CBM-I with the interference for self-focused thinking could be more e cient to mediate the generation of anxiety.

Limitations
Our study has several limitations that could be addressed in future research. First, the current study did not examine the speci city of the observed biases to trait anxiety rather than to depression. Although controlling for depression is common, we chose not to control it because current models of anxiety and depression suggest that these two constructs are conceptually related and co-occur for meaningful reasons. Separating them may result in spurious data [37]. Second, we used emotion faces as the provided choices for better validity and understanding, however, we found the material from CFAPS multiple emotions. Although they could be divided into two categories-positive and negative, the emotional valences of different positive emotions or different negative emotions were not speci ed, for instance, surprise & happiness and disgust & angry. Besides, the matching of scenarios and emotion valences was not particularly appropriate, which may affect the reaction time of the subjects. Therefore, in future studies, the paradigm should be improved by control the valence of the positive and negative emotion presented after the scenarios or ask the subjects to evaluate the degree of the endorsement or rejection of the emotional valence. Moreover, we cannot be certain that participants were reading words or ambiguous sentences. For example, participants were only responding to the faces, rather than determining the relatedness of the word to the scenarios. Finally, the coverage of military living situation was not wide enough, especially for different types of military troops and positions.
These limits need to be improved in the later stage.

Conclusion
In summary, the current compared design revealed the existed interpretation bias in military personnel with high trait anxiety by the method of PSAP comprised of military-simulated ambiguous scenarios and emotional facial expressions. In the instant information-processing stage, the readiness for negative interpretation of soldiers with high trait anxiety was higher than that of soldiers with low trait anxiety, and the negative interpretation was only in self-related situations. Hence, we concluded soldiers with high trait anxiety lack the positive interpretation bias and prefer self-related negative interpretation bias. Besides, the need for further research to improve the control of emotion valences in the emotional faces used in PSAP and plausible methods of cognitive bias modi cation to decrease the soldiers' anxiety was highlighted.

Consent for publication
All the authors consented to the publication.

Availability of data and materials
The datasets analyzed and materials used in this study are available from the the corresponding author on a reasonable request.

Competing interests
The authors declare that they have no competing interests Funding This research was supported in part by a grant from a Major military medical research project under the 12th ve-year plan (Award number 14CXZ002) Authors' contributions QY contributed to the writing of this article and part of statistical analysis. DG led the whole study, including putting forward this study, getting source, and carrying out the study, and was the corresponding author. CW and ZM contributed to revising this article and part of statistical analysis. CB and SX contributed to perform the investigation and collection of all data. We are all accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. We all agree to submit our research result in the article to this journal. All authors read and approved the nal manuscript.