Latent profile analysis of delusion-like experiences among emerging adults of Kashmir

Abstract

Delusion-like experiences (DLEs) are understood not as a categorical marker of psychosis but as a dimensional phenomenon distributed across the general population. Contemporary models of psychosis suggest that variations in the frequency and phenomenology of such experiences may reflect meaningful subgroups within nonclinical samples. Against this backdrop, the present study examined the latent profiles of emerging adults DLEs using Peter’s Delusions Inventory (PDI-21).

A cross-sectional convenience sample of 498 emerging adults of Kashmir valley completed the PDI-21, and latent profile analysis was conducted on the total score as well as the Distress, Preoccupation, and Conviction dimensions of the scale. Model fit indices supported a 3-profile solution (AIC = 12,736; BIC = 12,921; Entropy = 0.719) comprising weak (29.7%), moderate (45.6%), and strong endorsement profiles (24.7%), with high classification certainty (posterior probabilities = 0.95–0.96). Welch’s ANOVA and Post Hoc analyses confirmed significant differences across all profiles with respect to TPDI and its subscales (all p < .001). Further Chi-square analyses indicated that profiles didn’t differ by gender, education, family type, domicile or SES (all p > .05), suggesting that all profiles were not shaped by socio-demographic variables.

Findings of this research indicate that DLEs vary in degree among emerging adults while also revealing a separate subgroup at the upper end of severity. This gradient observed across the profiles provides support for the dimensional view of psychosis, but the marked elevation in distress and conviction dimensions among the strongest endorsement groups aligns with hybrid models proposing a continuous variation with increased psychological impact. These subgroups help clarify when normative experiences may differ in intensity, without implying clinical risk or diagnosis status.

Introduction

Delusions are traditionally considered defining features of psychotic disorders. At the same time, a substantial body of evidence suggests that delusion-like experiences (DLEs) are present across the general population, supporting the view that psychotic experiences lie on a continuum rather than a categorical disease state [1–3] Studies using broad community samples consistently show DLEs, such as ideas of reference, suspiciousness or magical thinking, without accompanying distress or functional impairment [4, 5]. This aligns with evidence that DLEs are not inherently pathological; rather, their clinical significance depends on associated dimensions such as conviction, preoccupation, and distress, which differentiate nonclinical from clinical presentations [5].

Epidemiological evidence indicates that 10–15% of the general population reports psychotic-like experiences at some point, with higher prevalence observed in adolescents and emerging adults [2, 6, 7]. Although most individuals with such experience do not transit to psychosis, DLEs are associated with affective dysregulation, trauma exposure, cognitive biases such as jumping to conclusions, and elevated stress sensitivity [8–11]. These associations highlight the importance of characterizing heterogeneity within nonclinical DLEs rather than treating them as a uniform construct.

Despite this, current research has focused on variable-centered approaches and assumed population-level homogeneity with obscure meaningful subgroups [12]. In comparison to it, person-centered methods, such as Latent Profile Analysis (LPA), focus on unobserved subgroups formed by the configuration of symptoms or beliefs within an individual [13]. Similar methods have revealed subgroups of differing severity, risk, and functional correlates of psychosis-spectrum traits [14, 15]. Advanced modeling approaches (including factor mixture models) further show that continuous variation is essential to DLEs and delusional ideation, with categorical distinctions providing minute support for severity continua [16]. Yet, fewer studies have used these methodologies, particularly for DLEs within the nonclinical population, creating heterogeneity of these experiences and scope for further study.

Identifying heterogeneity in DLEs within the nonclinical population can advance understanding of psychosis, refine predictive models of risk [17], and inform culturally specific intervention approaches. This is particularly relevant in under-studied context such as South Asian emerging adults, where cultural and religious belief structures may influence the expression and interpretation of DLEs [18]. Profiling heterogeneity among South Asian emerging adults may therefore offer both conceptual and practical contributions by illuminating culturally specific pathways of psychosis liability and resilience.

Methods

Participants

We conducted a cross-sectional study with 503 emerging adults, 18–30 years of age, including both males and females, between March and October 2024 in Kashmir. Participants were emerging adults enrolled in several colleges and universities in Kashmir using convenience sampling. The data was collected both online (N = 234) and offline (N = 269) from various college campuses and universities of Kashmir by the researchers. Respondents had to complete the questionnaire within 30–45 min as per the guidelines. Out of the total, only 498 (online = 232 & offline = 266) were considered for final analysis on the basis of statistical outliers. 5 participants were excluded on the basis of visual inspection and descriptive statistics; no predefined statistical cut-offs or influence diagnostic were applied.

The mean age of the participants was 23.01 years with SD 2.39. As shown in the Tables 1 and 55.2% of the participants were females, regarding educational qualification, 238 (47.8%) participants were undergraduates (UG), while 260 (52.2%) were postgraduates (PG). In terms of family structure, 334 (67.1%) belonged to nuclear families, and 164 (32.9%) belonged to joint families. With respect to domicile, 234 (47%) resided in rural areas, whereas 264 (53%) were from urban areas. Considering SES, 184 (36.9%) participants were from below poverty line (BPL) households, while 314 (63.1%) belonged to above poverty line (APL) households.

Table 1.

Descriptive statistics and demographic characteristics of the sample (N = 498)

Variable		N (%)
Gender	Male	223 (44.8%)
	Female	275 (55.2%)
Education	UG	238 (47.8%)
	PG	260 (52.2%)
Family Type	Nuclear	334 (67.1%)
	Joint	164 (32.9%)
Domicile	Rural	234 (47.0%)
	Urban	264 (53.0%)
SES	BPL	184 (36.9%)
	APL	314 (63.1%)

SES  Socioeconomic Status, BPL  Below Poverty Line, APL  Above Poverty Line

Participants were eligible for inclusion if they were between 18 and 30 years of age, studying in the institution or university, able to read and understand the language of assessment, and provided informed consent to participate in the study. Participants with a current or past diagnosis of major psychiatric disorder or currently taking psychiatric medication, reported a serious neurological condition or substance abuse, or were unwilling to complete the study questionnaires, were not included in the study.

Measures

We used the 21-item Peter’s Delusion Inventory (PDI-21) to assess DLEs in emerging adults, a nonclinical population. The PDI-21 measures the presence of unusual beliefs using a dichotomous YES/NO endorsement format. For each endorsed belief, participants additionally rate three core phenomenological dimensions namely Distress, Preoccupation, and Conviction, using 5-point Likert-type scales, with higher scores indicating greater severity. A total PDI endorsement score (TPDI) was calculated by summing endorsed items, while dimensional scores were computed by summing the respective Likert ratings across endorsed items. The PDI-21 scale was administered in English. Although English is not the first language of participants, it is the primary medium of instruction of all the participants. Therefore, no formal translation or adaptation was considered necessary. In previous research, this scale has demonstrated good internal consistency, with Cronbach’s α typically ranging from 0.79 to 0.90 and strong convergent validity with other measures of psychotic-like experiences and schizotypy [5, 19]. In the present sample, internal consistency was high for the Scale PDI-21 (α = 0.90) and acceptable to good for the Distress (α = 0.83), Conviction (α = 0.85), and Preoccupation (α = 0.82) subscales.

Statistical analyses

We computed descriptive statistics for the sample and PDI-21, LPA was then performed on the TPDI and three-dimensional subscale scores, namely Distress, Preoccupation, and Conviction. LPA is a probabilistic clustering approach that identifies unobserved subgroups of individuals based on similarity in multivariate response patterns [20]. Analyses were conducted in Jamovi 2.6 using the snowRMM module [21]. The estimation routines are based on the tidyLPA R engine, which implements Gaussian mixture modeling using the mclust framework [22]. Four theoretically standard covariance specifications are available (Models 1, 2, 3, and 6), and the diagonal covariance specification (analogous to Model 6) was employed. Multiple candidate profile solutions (2-class, 3-class, 4-class) were fitted, and model comparison was guided by information criteria (AIC, BIC, SABIC, AWE, CAIC, CLC, KIC), entropy, parsimony, and theoretical interpretability (For 2-class and 4-class, see related files). Including both the TPDI and subscale scores allowed the model to capture overall symptom endorsement as well as patterns of severity across dimensions, facilitating the identification of nuanced profiles. The 2-class profile was too coarse and the 4-class profile added a small and poorly interpretable class (< 5% of the sample), suggesting over extraction. Based on this holistic evaluation, the 3-class profile was retained for substantive interpretation.

Finally, we compared the three profiles on the basis of various socio-demographic variables using ANOVA and chi-square test. An alpha of 0.05 was used.

Results

Mean, SD and skewness-kurtosis of scores on PDI-21

Descriptive statistics for the study variables are presented in Table 2. The mean TPDI score on the PDI-21 was 10.50 (SD = 4.23). The three subscales—Distress (M = 30.73, SD = 16.26), Preoccupation (M = 33.08, SD = 16.37), and Conviction (M = 36.89, SD = 16.99)—all showed positive skewness (ranging from 1.77 to 5.81), indicating that most participants reported relatively low levels of DLEs, with a smaller subset reporting higher scores. Kurtosis values ranged from − 2.76 to 1.55, suggesting predominantly platykurtic distribution, although the distress subscale showed mild leptokurtosis. Such distributional characteristics are typical in clinical/psychological constructs and do not meaningfully bias descriptive inferences [23, 24]. In addition, these deviations do not violate assumptions relevant to latent profile modelling, which requires conditional multivariate normality within profiles, not global normality across the entire sample [25, 26]. In the present study, both the TPDI scores and subscales were treated as continuous indicators in LPA, consistent with the standard practice in Gaussian mixture modelling.

Table 2.

Descriptive analysis of PDI-21 and dimensions

Variable	M (SD)	Skewness	Kurtosis
TPDI	10.50 (4.23)	1.49	–1.24
Distress	30.73 (16.26)	5.81	1.55
Preoccupation	33.08 (16.37)	4.11	–2.7
Conviction	36.89 (16.99)	1.77	–2.76

Latent profile analysis

LPA was conducted using participant scores on TPDI, Distress, Preoccupation and Conviction. Eight models were estimated and compared using several information criteria, including Akaike Information Criteria (AIC), Bayesian Information Criteria (BIC), Sample-size-Adjusted BIC (SABIC), and classification entropy.

Model class indices indicated that the 3-class solution for Model 6 (AIC = 12,736; BIC = 12,921; SABIC = 12,782) provided the best overall fit for the data, demonstrating the lowest AIC and BIC among all competing models. Although the 4-class solution demonstrated a slightly lower AIC, the 3-class solution showed lower BIC and SABIC values, higher entropy (Entropy = 0.719), and yielded theoretically meaningful and well-separated profiles with adequate class sizes. Additionally, an analytic hierarchy process (AHP), incorporating multiple indices (AIC, AWE, BIC, CLC and KIC) identified 3-class Model 6 as optimal [27]. Therefore this model was retained for further interpretation.

Profile characteristics

The final three profiles reflected a distinct pattern of endorsement across the four PDI indicators (see Table 3 and Fig. 1). Profile 1 exhibited lowest scores on all variables (TPDI = 5.79, Distress = 13.8, Preoccupation = 15.3, Conviction = 18.2), indicating minimal endorsement of DLEs and associated distress.

Table 3.

Latent profile model fit indices of the sample

Model	Classes	AIC	BIC	aBIC	Entropy	BLRT (p)	Class Proportions (%)
1	2	14298.00	14352.68	14314.00	0.890	0.0099	61.0/39.0
1	3	13666.52	13742.32	13685.18	0.905	0.0099	29.7 / 24.7 / 45.6
6	3 (Best)	12736.00	12921.39	12782.00	0.719	< 0.001	29.7 / 24.7 / 45.6
6	4	12712.00	12959.72	12773.00	0.676	< 0.001	26.1/22.3/38.0/13.7

AIC Akaike Information Criteria, BIC Bayesian Information Criteria, aBIC Sample size adjusted BIC, Entropy measure of classification accuracy, BLRT (p) Bootstrap Likelihood Ratio Test p-value

Fig. 1.

Profile of delusion-like experiences among the sample

Profile 2 demonstrated the highest scores in all dimensions (TPDI = 15.8, Distress = 52.2, Preoccupation = 54.7, Conviction = 58.1), representing participants who strongly endorsed DLEs accompanied by significant distress and conviction and Profile 3 showed intermediate scores (TPDI = 10.7, Distress = 30.2, Preoccupation = 32.9, Conviction = 37.6), indicating moderate endorsement of DLEs and preoccupation These three profiles form a clear gradient from low to high endorsement of DLEs.

Also, posterior classification probabilities were high (range = 0.95–0.96), indicating good classification certainty. Profile membership frequencies were as follows: Profile 1 (Weak Endorsement): 148 participants (29.7%), Profile 2 (Strong Endorsement): 123 participants (24.7%), and Profile 3 (Moderate Endorsement): 227 participants (45.6%).

Between profile differences

Welch’s one-way analyses of variance were conducted to examine whether the Weak, Moderate, and Strong endorsement profiles differed significantly on TPDI score, distress, preoccupation, and conviction. Levene’s tests indicated violations of the homogeneity of variance assumption for all indicators (p < .05), so Welch’s ANOVA was used. The results revealed significant between-profile differences for all four indicators: TPDI, F (2, 267) = 663.00, p < .001, ηp² = 0.83; distress, F (2, 268) = 692.00, p < .001, ηp² = 0.84; preoccupation, F (2, 264) = 754.00, p < .001, ηp² = 0.85; and conviction, F (2, 277) = 701.00, p < .001, ηp² = 0.84. Games–Howell post hoc comparisons indicated that all three profiles differed significantly from one another on each indicator (p < .001). Specifically, participants in the Strong Endorsement profile had the highest scores, followed by the Moderate Endorsement profile, with the Weak Endorsement profile having the lowest scores across all PDI indicators (Table 4).

Table 4.

Welch’s ANOVA showing differences between the indicators

Outcome Variable	Welch’s F	df1	df2	p	Pairwise Comparison	Mean Difference	p	ηp2
TPDI	663.00	2	267	< 0.001	Weak vs. Strong	−9.99	< 0.001	0.83
					Weak vs. Moderate	−4.92	< 0.001
					Strong vs. Moderate	5.07	< 0.001
Distress	692.00	2	268	< 0.001	Weak vs. Strong	−38.40	< 0.001	0.84
					Weak vs. Moderate	−16.40	< 0.001
					Strong vs. Moderate	22.00	< 0.001
Preoccupation	754.00	2	264	< 0.001	Weak vs. Strong	−39.40	< 0.001	0.85
					Weak vs. Moderate	−17.60	< 0.001
					Strong vs. Moderate	21.80	< 0.001
Conviction	701.00	2	277	< 0.001	Weak vs. Strong	−39.90	< 0.001	0.84
					Weak vs. Moderate	−19.40	< 0.001
					Strong vs. Moderate	20.50	< 0.001

Chi-square tests were conducted to examine whether profile membership varied by demographic variables, including gender, education, family type, domicile, and SES. None of the associations were significant (all p > .05), suggesting that the latent profiles were not demographically driven (Table 5).

Table 5.

Demographic comparisons of three profiles

Variable	Weak Endorsement(n = 148)	Moderate Endorsement(n = 227)	Strong Endorsement(n = 123)	χ²	p
Gender				1.45	0.485
Male	70(47.2%)	95 (41.8%)	58(47.1%)
Female	78 (52.7%)	132(58.2%)	65 (52.8%)
Education level				1.18	0.554
Undergraduate	69 (46.6%)	105 (46.2%)	64(52.0%)
Postgraduate	79 (53.3%)	122 (53.7%)	59(47.9%)
Family Type				1.70	0.427
Nuclear	95 (64.1%)	151 (66.5%)	88 (71.5%)
Joint	53 (35.9%)	76 (33.4%)	35 (28.4%)
Domicile				2.83	0.243
Urban	78 (52.7%)	100 (44.0%)	56 (45.5%)
Rural	70 (47.2%)	127 (56.0%)	67 (54.4%)
SES				1.85	0.397
BPL	48 (32.4%)	88 (38.7%)	48 (39.0%)
APL	100 (67.5%)	139 (61.3%)	75 (61.0%)

BPL  Below Poverty Line, APL Above Poverty Line

In summary, the LPA identified three distinct latent profiles characterized by Weak, Moderate, and Strong endorsement of DLEs. The model demonstrated good statistical fit, meaningful interpretability, and clear differentiation among profiles, while demographic factors did not significantly influence profile membership (Fig. 1).

Discussion

We identified three patterns of DLEs among emerging adults using PDI-21 using LPA. These three profiles represented weak, moderate and strong endorsement of DLEs. Our findings contribute to the literature on DLEs in the nonclinical population, while also providing a cautious and nuanced interpretation. Our approach differs in focusing specifically on DLEs in South Asian context while allowing us to capture meaningful variations, as well as replications of the results in previous studies.

Consistent with the previous research, our sample exhibited positive skewness, with most participants reporting relatively low levels of DLEs [19]. Such skewness is typical in community samples and does not contradict the continuum model of psychosis, which proposes that psychosis or psychotic experiences are distributed in a graded manner [28] in comparison to the categorical approach where the categories are discrete [29].

The three profile solution demonstrated meaningful distinction across all PDI dimensions, supporting theoretical models that suggest that distressing beliefs with preoccupation and conviction are central to delusional ideation and ultimately psychotic experiences [30–32]. These findings suggest that variation in DLEs within a nonclinical population is primarily quantitative rather than qualitatively distinct. However, this pattern does not imply that the phenomenological structure of delusions in clinical psychosis is identical to that observed in the general population, a point emphasized in alternative theoretical accounts [33].

Out of the three profiles the Weak Endorsement profile represents individuals with minimal unusual beliefs, corresponding to the healthy majority, commonly identified in population based studies [34]. The Moderate Endorsement profile, which comprises nearly half of the sample, aligns with the studies proposing many young adults experience transient or low intensity DLEs, often associated with developmental instability, stress and normative cognitive intrusions [35]. However, an alternate perspective suggests that moderate scores may not necessarily indicate meaningful delusion ideation. Particularly in South Asian cultural settings, including Kashmir and broader regions of India, religious and supernatural beliefs often function as socially shared symbolic frameworks through which individuals interpret ambiguous or unusual experiences. Therefore, endorsement of certain PDI items may reflect culturally sanctioned belief systems rather than delusion ideation. Prior research has shown that several PDI items overlap with normative perceptual variability or imaginative engagement, suggesting moderate endorsement may partly reflect normal psychological experiences rather than incipient pathology [19]. Overall, this profile likely represents a heterogeneous mixture of benign, culturally tangible experiences rather than a coherent subclinical state.

The Strong Endorsement profile exhibited substantially higher distress, conviction and preoccupation, features that are risk markers and common among subclinical groups [36, 37]. Individuals scoring high on distress and conviction, in particular, have been shown to predict later functional impairment and help-seeking [38]. Nevertheless, the present study does not permit conclusions regarding clinical risk. Without longitudinal or diagnostic data the Strong Endorsement profile should not be interpreted as a proxy for psychosis or clinical delusions. Elevated scores on PDI-21 may also reflect acute emotional states such as anxiety, sleep disturbance, or academic stress, which are known to be associated with psychotic experiences and any substantial decision should be made while being aware that psychotic experiences exist in broader domains of psychopathology as well as trans-diagnostic cognitive and affective processes [39, 40]. Accordingly, while the Strong Endorsement profile is statistically distinct, its clinical significance remains provisional.

A notable finding is the absence of demographic difference across three latent profiles. This aligns with the findings that the DLEs occur across gender, SES groups, and residential background [7]. However, the homogenous age range and student population may have restricted variability, potentially obscuring subtle demographic influences. Moreover, demographic variables often operate indirectly through exposure to stress, trauma and social adversity [41], variables that are not measured in the current study. Hence, this lack of demographic effects in our results should not be understood as an indication of irrelevancy of such factors for DLEs.

Methodologically, this study demonstrates strong entropy and high posterior classification probabilities, indicating reliable subgroup differentiation. This study also underscores the importance of meaningful heterogeneity in the experience of DLEs. The identification of Strong Endorsement profile and its significant difference from other profiles in Welch’s ANOVA highlights the need for proactive mental health screening in university settings, not as a means of identifying clinical psychosis, but as an opportunity to detect elevated distress associated with unusual beliefs. Early identification of individuals experiencing DLEs and related distress may help prevent adverse psychological outcomes, consistent with the preventative frameworks in psychosis research [42].

The pattern of profiles also informs the categorical versus continuum debate. The profiles formed a clear severity gradient, without evidence of qualitatively distinct symptom configuration. All PDI indicators increased in parallel, suggesting quantitative rather than structural discontinuities. This provides strong support for the dimensional models of psychotic experiences [2, 3, 43]. At the same time, the statistical stability of the Strong Endorsement profile and its elevated distress and conviction suggests a meaningful subgroup residing at the upper end of the continuum. This aligns with the quasi-dimensional or hybrid models, which posit continuous variations but acknowledge thresholds indicative of increased psychological vulnerability [44]. Importantly, this interpretation does not imply that nonclinical profiles replicate the structure of clinical psychosis, but rather that they reflect the variation in shared underlying processes, expressed at differing levels of severity, consistent with findings reported by Cella and colleagues [15].

In conclusion, the study identified three latent profiles reflecting weak, moderate and strong endorsement of DLEs among emerging adults. While these profiles provide empirical support to dimensional models of psychosis and offer meaningful distinctions within a nonclinical population, they are shaped by cultural context and methodological considerations that warrant thoughtful interpretation. This variation in itself undermines any uniform interpretation [45]. Importantly, the latent profiles identified here are not intended to represent direct analogues of delusions in clinical population, but rather inform dimensional model of psychotic phenomena. Integrating these findings with longitudinal, clinical and cross-cultural research will be essential for clarifying how DLEs relate to broader models of psychosis and mental health.

Moreover, the reliance on convenience sampling of students may limit generalizability and reduce demographic variability. Also, the absence of formal clinical assessment restricts the interpretation of subclinical conditions and prevents conclusions regarding clinical risk. Finally, these findings reflect patterns of heterogeneity rather than prevalence estimates for the general population.

Authors’ contributions

1. Amir Sultan: Conception of the idea designed and supervised the study. Also, Amir conducted data analysis and prepared and revised the study report of this research. 2. Sadiya Sajad: Data collection, literature review and preparing the report.

Funding

No funding was received for conducting this study.

Data availability

The data can be made available by the corresponding author on a reasonable request.

Declarations

Ethics approval and consent to participate

Indira Gandhi National Open University (IGNOU), Institutional Ethical Committee (IEC), (IGNOU-IEC-266 of 09/08/2023) New Delhi, approved the research protocol to collect data and perform the analysis for the study. Voluntary participation and informed consent was required from the participants while the researchers maintained confidentiality. Besides, this study was conducted in accordance with the ethical standards of the Helsinki Declaration.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.