This study examines whether a proactive thought control intervention can reduce negative core beliefs, cognitive biases, and anxiety symptoms in university students with social anxiety. Participants with elevated social anxiety (screened via the Liebowitz Social Anxiety Scale) are randomly assigned to either a proactive thought control group or a reactive control group. Both groups complete two computerized tasks - a Free Association Task and a Sentence Completion Task - across 180 trials. The proactive group is trained to generate only positive or neutral associations to socially threatening cues and receives real-time AI-powered sentiment feedback, while the reactive group responds freely without sentiment-based guidance. Outcomes including negative core beliefs, interpretation bias, attentional bias, state anxiety, and trait anxiety are assessed before and after the intervention using standardized measures (CBQ, WSAP, Dot Probe Task, STAI). The study uses a parallel-group randomized controlled trial design with repeated measures and aims to establish preliminary effect size estimates for future, larger-scale trials.
Background and Rationale Social anxiety disorder (SAD) is characterized by intense fear of negative evaluation in social situations, leading to significant avoidance and distress. Cognitive models of SAD emphasize the role of negative core beliefs (e.g., beliefs about being inferior or unacceptable to others) and cognitive biases - particularly interpretation bias and attentional bias toward threat - in the maintenance of the disorder. While cognitive bias modification (CBM) approaches have shown promise, most are reactive in nature, modifying biases after they arise. Proactive cognitive control, by contrast, involves the sustained, goal-directed maintenance of positive or neutral mental sets before threatening stimuli are encountered. This study tests whether a proactive approach to thought control, delivered through modified free association and sentence completion paradigms, can more effectively shift core beliefs and cognitive biases compared to a reactive control condition. Study Design This is a parallel-group randomized controlled trial (RCT) with a repeated-measures (pre-post) design. Participants are randomly assigned using the fishbowl block randomization technique to either: (1) the Proactive Thought Control group, or (2) the Reactive Control group (active sham condition). Counterbalancing is applied to the order of stimuli in the WSAP and Dot Probe Task to control for practice effects. Participants Participants are university students aged 18-24 years, recruited via purposive sampling using the Liebowitz Social Anxiety Scale - Self Report (LSAS-SR). Individuals scoring above 30 on the LSAS-SR are eligible. Exclusion criteria include current psychological treatment, psychiatric or medical comorbidities, and medication use. Of 204 initially screened, 149 met the LSAS cutoff; 54 were enrolled and randomized. Final analyses include 37-39 participants who completed both pre- and post-assessments. As a pilot trial, this sample size is consistent with recommended ranges of 15-30 participants per group for feasibility and preliminary effect size estimation. Measures Pre- and post-intervention assessments include: Core Belief Questionnaire (CBQ): Measures negative core beliefs about self and others in social anxiety (trait and other subscales used). Spielberger State-Trait Anxiety Inventory - Short Form (STAI): Assesses state and trait anxiety (5 items each; 4-point Likert scale). Word Sentence Association Paradigm (WSAP): Computerized task measuring interpretation bias; participants judge relatedness of threatening/benign word-sentence pairs. Bias scored as proportion of threatening endorsements. Dot Probe Task: Computerized task measuring attentional bias toward threat using reaction time differences (Attentional Bias Index = mean RT threat-incongruent minus mean RT threat-congruent trials). Screening measures include the LSAS-SR and DASS-12 (for baseline distress comparability). Intervention The intervention comprises 180 trials divided into six blocks (framed as "game levels"): three blocks of Free Association Task (FAT) and three blocks of Sentence Completion Task (SCT). Each block contains 25 socially threatening stimuli and 5 positive/neutral stimuli, presented in randomized order. Free Association Task (FAT): Participants see a single threatening or neutral word cue and respond with the first word that comes to mind. The proactive group is instructed to produce only positive or neutral single-word associations and cannot advance until doing so. The reactive group responds freely. Sentence Completion Task (SCT): Participants complete unfinished sentence stems (e.g., "They stared, and I felt \_\_\_"). The proactive group must provide positive or neutral completions; the reactive group responds without restriction. A real-time AI feedback system powered by DistilBERT-base-uncased-finetuned-SST-2-English (a lightweight transformer model fine-tuned for binary English sentiment classification) evaluates responses. Both groups receive quality-based feedback (for repetitions, spelling errors, and invalid inputs). The proactive group additionally receives sentiment-based reinforcement: +2 points and "Correct" feedback for positive/neutral responses, and corrective guidance for negative responses. The reactive group receives no sentiment feedback. This gamified point system reinforces goal-directed positive thinking in the proactive group. Procedure Participants are recruited in-person (paper-based classroom administration) and online (Google Form). Following informed consent and LSAS/DASS screening, eligible participants complete pre-assessment (CBQ, STAI, WSAP, Dot Probe Task) via OpenSesame software, then complete the intervention tasks, followed immediately by post-assessment using the same battery. Data Analysis Data cleaning and scoring are conducted using Excel and R (packages: jsonlite, readxl, dplyr, tidyr, stringr, writexl). Statistical analyses are conducted in SPSS and include: chi-square for attrition analysis; descriptive statistics for demographics; Mixed ANOVA for main and interaction effects (Group × Time); paired-sample t-tests for within-group pre-post comparisons; independent-sample t-tests for between-group comparisons; and ANCOVA to isolate intervention effects from baseline differences.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
TREATMENT
Masking
SINGLE
Enrollment
54
A computerized behavioral intervention delivered across 180 trials in six blocks. The first three blocks use a Free Association Task (FAT), where participants respond to single threatening or neutral word cues. The last three blocks use a Sentence Completion Task (SCT), where participants complete socially threatening or neutral sentence stems. Participants in the proactive group must generate positive or neutral single-word responses and cannot advance until doing so. Real-time sentiment feedback is delivered via DistilBERT (an AI language model), awarding +2 points for positive/neutral responses and providing corrective guidance for negative ones. The reactive control group completes identical tasks but responds freely, receiving only neutral quality-based feedback without sentiment reinforcement. Both groups receive feedback for repeated, misspelled, or invalid entries. Each block contains 25 threatening and 5 positive/neutral stimuli presented in randomized order.
A computerized behavioral sham condition delivered across 180 trials in six blocks, identical in structure to the experimental intervention. The first three blocks use a Free Association Task (FAT) and the last three use a Sentence Completion Task (SCT), both involving socially threatening and neutral stimuli. Participants respond freely with any single-word association without restriction on response valence. No sentiment-based feedback or scoring is provided. Participants receive only neutral quality-based feedback for repeated, misspelled, or invalid entries. This condition controls for nonspecific factors including task engagement, time-on-task, the Hawthorne effect, and demand characteristics, while isolating the active ingredient of proactive sentiment-directed training present in the experimental arm.
GIFT University
Gujranwala, Punjab Province, Pakistan
Core Beliefs: Change in Negative Core Beliefs as Measured by the Core Belief Questionnaire (CBQ)
The Core Belief Questionnaire (CBQ; Wong et al., 2017) measures negative core beliefs in socially anxious individuals across three versions: trait (beliefs about self), contingent (beliefs about self when negatively evaluated), and other (beliefs about others). Each scale contains 17 items rated on a 6-point Likert scale (1 = strongly disbelieve to 6 = strongly believe). Total scores range from 17 to 102, with higher scores indicating stronger negative core beliefs. In this study, the trait and other subscales are used. Change scores from pre- to post-intervention are compared between the proactive thought control group and the reactive control group.
Time frame: Baseline (pre-intervention) and immediately after intervention completion (post-intervention)
Interpretation Biases: Change in Interpretation Bias as Measured by the Word Sentence Association Paradigm (WSAP)
The WSAP (Beard \& Amir, 2009) is a computerized task measuring interpretation bias toward threatening versus benign interpretations of ambiguous social scenarios. Each trial presents a word cue followed by an ambiguous sentence; participants judge relatedness by pressing Y or N. Bias is scored as the proportion of trials in which participants endorse threatening word-sentence pairings. Higher proportions indicate stronger negative interpretation bias. The task includes 30 trials with threatening and benign word-sentence pairs presented in randomized, counterbalanced order across pre- and post-assessments.
Time frame: Baseline (pre-intervention) and immediately after intervention completion (post-intervention)
Attentional Biases: Change in Attentional Bias as Measured by the Dot Probe Task
The Dot Probe Task (MacLeod et al., 1986) measures attentional bias toward threatening stimuli using reaction times. Each trial presents a threatening and a neutral word simultaneously; a probe then replaces one stimulus. The Attentional Bias Index (ABI) is calculated as: mean RT on threat-incongruent trials minus mean RT on threat-congruent trials. Positive ABI scores indicate vigilance toward threat; negative scores indicate avoidance; scores near zero indicate no bias. Trials with RTs below 200 ms or above 1500 ms and incorrect responses are excluded. The task includes 30 trials in counterbalanced, randomized order across assessments.
Time frame: Baseline (pre-intervention) and immediately after intervention completion (post-intervention)
State Trait Anxiety:Change in State-Trait Anxiety as Measured by the STAI - Short Form (STAI-State Subscale)
The short-form State-Trait Anxiety Inventory (Zsido et al., 2020) state subscale consists of 5 items assessing temporary anxiety symptoms such as tension and apprehension. Items are rated on a 4-point Likert scale (0 = not at all to 3 = very much so), with total scores ranging from 5 to 20. Higher scores indicate greater state anxiety. Change from pre- to post-intervention is compared between groups. The short-form State-Trait Anxiety Inventory (Zsido et al., 2020) trait subscale consists of 5 items assessing stable individual differences in anxiety proneness. Items are rated on a 4-point Likert scale (0 = not at all to 3 = very much so), with total scores ranging from 5 to 20. Higher scores reflect greater trait anxiety. Change from pre- to post-intervention is compared between the proactive thought control and reactive control groups.
Time frame: Baseline (pre-intervention) and immediately after intervention completion (post-intervention)
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.