Distinguishing Personality Functioning,Normal-Range Traits,and Maladaptive Traits in Offenders

Distributions

As described above, normal-range traits are theoretically bipolar, meaning that they capture variance at both the high and low levels of each domain. In contrast, maladaptive traits are unipolar and positively skewed, meaning that they focus on variance at the tails of traits that are nomothetically related to problems in living. Although the developers of the LPFS have emphasized that it is meant to capture the full spectrum of health (Bender et al., 2011), it serves as a diagnostic indicator, and its indicators focus primarily on dysfunction. Thus, it is possible that the personality-functioning distribution is similar to that of maladaptive traits. Thus, we hypothesized the following:

Stability

Recent research indicates that both personality functioning and personality traits can change over time, and some evidence suggests that maladaptive personality features are less stable than normal-range traits (Bleidorn et al., 2022; Haehner, Sleep, et al., 2024; A. G. C. Wright et al., 2016; A. J. Wright et al., 2026). There are multiple types of stability; we focused on three: rank-order stability, mean-level change, and individual differences in change.

Intercorrelations among normal-range and maladaptive traits

Normal-range-trait domains were originally thought to be orthogonal, although subsequent research suggested that relatively modest correlations among the traits could be modeled as higher-order factors (Digman, 1997). If maladaptive traits are simply variants of normal-range traits, they should also be modestly intercorrelated. However, if they reflect a combination of normal-range traits and personality dysfunction, they would be more strongly intercorrelated because personality dysfunction would be shared across traits. Emery et al. (2024) recently found that the average correlation for maladaptive traits was .49 for the Personality Inventory for DSM-5 (PID-5; Krueger et al., 2012) and .55 for the Computer Adaptive Test for Personality Disorder (CAT-PD; Simms et al., 2011), whereas the average correlation for normal-range traits measured with the Big Five Inventory (BFI; John & Srivastava, 1999) was .27. Building on these findings, we hypothesized the following:

Personality functioning as an explanation for intercorrelation differences between maladaptive and normal-range traits

As mentioned earlier, stronger intercorrelations among maladaptive traits may arise from their common characteristics related to personality dysfunction. If personality functioning explains the higher intercorrelation among maladaptive traits relative to normal-range traits, controlling for the LPFS in the maladaptive traits should reduce their intercorrelations and make them more similar to those of normal-range traits. For instance, Emery et al. (2024) found that controlling for personality functioning reduced the intercorrelations among maladaptive traits from .49 to .42 for the PID-5 and from .55 to .43 for the CAT-PD. In both cases, although maladaptive traits were still more correlated than normal-range traits, removing personality functioning reduced those correlations and made them more similar to normal-range traits. Meanwhile, Emery et al. used an interview measure for the LPFS and questionnaires to measure maladaptive and normal-range traits. Although there are general advantages to multimethod research, in this case, method variance shared with LPFS and maladaptive traits could not be eliminated, reducing the power of this test. Overall, their findings suggest that saturation with personality functioning partly explains their high intercorrelations:

Discriminant validity of maladaptive traits

In theory, maladaptive traits should have high correlations with corresponding normal-range traits but low correlations with the rest of the normal-range traits (i.e., good discriminant validity). However, Morey, Good, and Hopwood (2022) found that maladaptive traits had moderate to strong cross-correlations with noncorresponding normal-range traits, indicating low discriminant validity. For example, detachment not only had a high correlation with its corresponding normal-range trait (i.e., extraversion, average |r| = .62) but also had high correlations with neuroticism (average |r| = .51), agreeableness (average |r| = .49), and conscientiousness (average |r| = .43). After variance associated with LPFS was removed from maladaptive traits, the discriminant validity of maladaptive traits significantly improved to .19, .23, and .15, respectively. However, Emery et al. (2024) did not replicate this effect in their multimethod study, perhaps because of the method effect described above. We hypothesized the following:

Maladaptive traits as a joint function of normal-range traits and personality dysfunction

The Morey, Good, and Hopwood (2022) article and Emery et al. (2024) replication study both tested the hypothesis that maladaptive traits would be a joint function of normal-range traits and personality dysfunction. This gets to the heart of the “hybrid” concept that inspired the combination of A and B Criteria in the AMPD, that is, that personality disorder is a function of what the person is like in general (personality traits) and the person’s overall capacity to adapt to life’s stressors and challenges (personality dysfunction; Morey et al., 2011). In both studies, the hypothesis was tested using a regression model that examined one trait domain at a time. In these models, normal-range traits and personality functioning were used to predict the corresponding maladaptive traits simultaneously. The results showed that both normal-range traits and personality functioning consistently had significant regression coefficients on maladaptive traits.

We tested this effect in three ways with the general expectation that normal-range traits and personality dysfunction would jointly predict maladaptive traits. First, we expected the following:

Second, we expected the following:

This provided the first test of whether changes in both personality functioning and normal-range traits mutually explain changes in maladaptive traits. Third, we examined the joint function of within-persons changes. Nearly all research so far in this field has been concerned with how people are different from each other. However, within-persons questions about how people change relative to themselves are rated by clinicians as at least as important than between-persons questions for clinical practice (Hopwood et al., 2025). We expected the following:

Cross-lagged within-persons associations

One of the important underlying questions about personality traits and dysfunction in the literature has to do with which is primary. One perspective is that traits influence functioning such that changes in traits should precede/cause changes in functioning (Clark & Ro, 2014). For instance, if people become less conscientious, they might have a more difficult time adapting to their life situation and thus experience decreases in personality functioning. A different perspective is that functioning is primary such that changes in functioning should precede changes in traits (Zimmermann et al., 2015). For instance, if people mature and are better able to adapt to a range of circumstances, they will adopt the kinds of roles and habits that fit their circumstances better, which might involve being more conscientious. A middle perspective is that personality functioning and traits transact. For example, impairments in empathy (functioning) could lead to increased distrust and withdrawal (traits). These traits, in turn, may limit relational learning opportunities, further reinforcing the original impairment. Haehner, Sleep, et al. (2024) found that changes in personality functioning led to changes in normal-range traits more than the other way around:

Gender differences

Some previous research has identified gender differences in the internal structure of maladaptive personality features (McIntyre, 2010), external associations (Silberschmidt et al., 2015), stability across adulthood (Vergauwe et al., 2023), and treatment engagement (Bozzatello et al., 2024). Gender differences in levels of certain traits are also established (Gomez et al., 2023; Weisberg et al., 2011). Although we anticipated that results from hypothesis tests listed above will replicate across samples, we also explored potential gender differences in hypothesized effects.

Transparency and openness

This registered report was based on data from two four-wave data sets collected with Chinese offenders. Stage 1 of this registered report is available at https://osf.io/x5qjf/. The analysis plan for this study was preregistered; however, the data collection was not preregistered. By the time Stage 1 of the registered report was accepted, data collection for all four waves had already been completed. However, at that point, only the Time 1 (T1) data for both samples had been digitized. All the code, materials, de-identified data, and supplemental materials are available at https://osf.io/x5qjf/. Because of data-security requirements imposed by the collaborating institution, we were granted access only to a screened data set prepared specifically for the purposes of this study. As a result, the analysis scripts begin from the stage of formal data analysis rather than from raw data preprocessing. In addition, demographic information was not uploaded together with the other variables because it contains personal information about the offenders and is subject to confidentiality restrictions. We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. This study was approved by the ethics committee of Guangzhou University (Protocol No. 2024 [056]), and all procedures were in accordance with the Declaration of Helsinki.

Procedure

We recruited two samples from prisons for men (Sample 1) and women (Sample 2) in southwestern China. All the participants are Chinese. Participants completed paper-and-pencil tests in groups of 50 to 100 people, supervised by researchers. A total of four measurements were conducted; Time 2 (T2) to Time 4 (T4) measurements were about 17, 34, and 53 weeks after T1 for Sample 1 and about 18, 35, and 52 weeks after T1 for Sample 2. Participants received daily necessities (valued at 35 RMB, or about $5) in exchange for their participation.

Participants

All participants had completed at least primary school and were able to read and understand the survey. Sample sizes were determined by participant availability in the prison. In Sample 1, a total of 1,395 participants had valid data for at least one time point; valid sample sizes across the four waves were 1,173, 916, 910, and 714, respectively. At T1, mean age was 39.68 ± 10.33 years. Education levels were primary school (n = 281), junior high (n = 652), high school (n = 223), college or above (n = 110), and unknown (n = 129). Average incarceration duration was 4.34 ± 4.30 years; 367 participants were violent offenders, 395 were people convicted of drug-related crimes, 129 had unknown offenses, and the remaining were convicted of fraud, theft, and property crimes. In Sample 2, a total of 1,017 participants had valid data for at least one time point; four-wave valid sample sizes were 887, 805, 696, and 680, respectively. At T1, mean age was 39.35 ± 9.71 years. Education levels were primary school (n = 152), junior high (n = 477), high school (n = 218), college or above (n = 148), and unknown (n = 22). Average incarceration duration was 4.64 ± 3.26 years; 111 were violent offenders, 474 were convicted of drug-related crimes, 21 offenses were unknown, and the remaining were convicted of fraud, theft, and property crimes.

Measures

All McDonald’s omega values were above .58 and had a median of .77 across the two samples (Table S1 in the Supplemental Material available online). For the results of the confirmatory factor analysis for the LPFS, Big Five Inventory–2 Extra-Short Form (BFI-2-XS), and PID-5-Brief Form (PID-5-BF), see Table S2 in the Supplemental Material. Across all models, comparative-fit-index (CFI) values ranged from .823 to .934 (Mdn = .897), root mean square error of approximation (RMSEA) ranged from .058 to .120 (Mdn = .083), and mean standardized root mean squared residual (SRMR) ranged from .038 to .075 (Mdn = .059). These indices are in the desirable range of values observed for other common personality-assessment measures (Hopwood & Donnellan, 2010).

Analysis plan

The Stage 1 registered report and supplementary materials are available at https://osf.io/x5qjf/. Because of issues such as model convergence and redundancy, there were some differences between this article and preregistered plan regarding Hypotheses 3 and 4 (see details in the sections on Hypotheses 3 and 4 and Supplemental Material). We report only the adjusted models and results in the main text. For the results based on the preregistered methods and the reasons for deviations from the preregistration, see Tables S6 through S8 in the Supplemental Material.

All analyses were conducted in R Studio. We conducted Hypothesis 2 and Hypotheses 5 through 8 in a latent-variable framework with full information maximum likelihood estimation. The three items for each normal-range trait and the five items for each maladaptive trait were used to estimate latent variables. For the LPFS, four item parcels (three items per parcel) were created based on its four elements—empathy, intimacy, self-direction, and identity—to improve model estimation (Little, 2013). For Hypotheses 3, 4, 9, and 10, because of model complexity and convergence difficulties at the item level, we applied the analysis using total scores to estimate the constructs directly. Cross-sectional analyses used T1 through T4 data, but only T1 results are reported in the article because it had the largest sample size; results from T2 to T4 are provided in the Supplemental Material. Unless otherwise noted, all coefficients are standardized, and a significance level of α = .05 was used for all hypothesized effects. We focused on effect sizes and the overlap of 95% confident intervals (CIs) to interpret gender differences (cf. Cumming, 2014). A nonoverlapping 95% CI is roughly equivalent to p = .01 in the traditional null-hypothesis-significance-testing framework (Cumming, 2009).

Measurement invariance

We estimated configural, metric, and scalar models to ensure that scores are comparable across occasions. We used three indices to evaluate metric and scalar invariance: ΔCFI < .01, ΔMcDonald’s Noncentrality Index (NCI) < .02, ³ and ΔRMSEA < .015 (Chen, 2007; Cheung & Rensvold, 2002). Across the two samples, we tested 66 longitudinal invariance models involving 11 variables (configural, metric, and scalar invariance), all of which met at least two criteria; 61 models satisfied all three criteria (Table S3 in the Supplemental Material). These results support comparisons of mean levels and associations across assessment waves. However, measurement invariance was not fully supported for some variables across samples (Table S4 in the Supplemental Material), complicating gender comparisons.

Hypothesis 1: descriptive analysis and distributions

We computed mean, standard deviations, and skewness of personality functioning, normal-range traits, and maladaptive traits. We used R package Boot (Canty & Ripley, 2024) to calculate the 95% CI of skewness differences between two variables. Consistent with Hypothesis 1, normal-range traits had more normal distributions than maladaptive traits in general, although not all differences were statistically significant (see Table 1). The LPFS distribution was between normal-range and maladaptive traits in Sample 1 but was more skewed in Sample 2. For distributions for each variable, see Figures S1 and S2 in the Supplemental Material.

OPEN IN VIEWER

Table 1.

Descriptive Statistics for Personality Functioning, Normal-Range Traits, and Maladaptive Traits at Time 1

	Sample 1			Sample 2
Variable	M	SD	Skewness	M	SD	Skewness
Extraversion.r	3.24 [3.20, 3.28]	0.77	−0.01 [−0.15, 0.12]ab	3.24 [3.18, 3.30]	0.87	−0.17 [−0.30, −0.06]ab
Agreeableness.r	2.44 [2.39, 2.49]	0.88	0.37 [0.27, 0.47]b	1.97 [1.92, 2.02]	0.70	0.44 [0.24, 0.65]ab
Conscientiousness.r	2.71 [2.66, 2.75]	0.76	0.00 [−0.13, 0.12]ab	2.65 [2.60, 2.71]	0.77	0.13 [−0.01, 0.26]ab
Neuroticism	2.65 [2.60, 2.70]	0.87	0.17 [0.06, 0.28]	2.68 [2.61, 2.75]	1.07	0.32 [0.24, 0.41]b
Openness	2.73 [2.69, 2.78]	0.83	0.07 [−0.04, 0.18]	2.25 [2.19, 2.31]	0.89	0.35 [0.23, 0.48]ab
Detachment	1.01 [0.98, 1.05]	0.65	0.30 [0.20, 0.40]	0.87 [0.83, 0.91]	0.62	0.80 [0.67, 0.93]a
Antagonism	0.90 [0.86, 0.93]	0.61	0.49 [0.38, 0.60]ab	0.62 [0.59, 0.66]	0.49	0.84 [0.69, 1.01]a
Disinhibition	1.04 [1.00, 1.08]	0.65	0.35 [0.24, 0.46]a	0.92 [0.88, 0.96]	0.63	0.60 [0.48, 0.72]ab
Negative affectivity	1.14 [1.10, 1.18]	0.68	0.21 [0.11, 0.31]	1.11 [1.06, 1.15]	0.67	0.41 [0.29, 0.53]b
Psychoticism	1.11 [1.08, 1.15]	0.64	0.20 [0.09, 0.31]	0.95 [0.90, 0.99]	0.64	0.67 [0.55, 0.79]ab
LPFS	1.98 [1.95, 2.02]	0.63	0.24 [0.12, 0.34]	1.84 [1.80, 1.88]	0.61	0.86 [0.71, 1.00]

Note: Values in brackets are 95% confident intervals. Values that differ by gender are in italics. Letter “a” indicates significant difference between normal-range trait and its corresponding maladaptive traits. Letter “b” indicates significant difference with LPFS. Reversed scores are indicated with “.r” at the end of the variable name. LPFS = Level of Personality Functioning Scale.

Hypotheses 2 through 4: stability

We compared the stability of personality functioning, normal-range traits, and maladaptive traits to test Hypotheses 2 to 4. We first fit univariate models to examine rank-order stability, mean-level change, and individual differences in mean-level change. We then estimated nested bivariate models to test differences in stability. All comparisons were pairwise, comprising (a) personality functioning and a normal-range trait, (b) personality functioning and a maladaptive trait, and (c) a normal-range trait and the corresponding maladaptive trait.

Hypothesis 2: rank-order stability

We examined rank-order stability using latent test-retest correlations between T1 and T4 (illustrated in Fig. S3a in the Supplemental Material) to detect stability over longer time frames. As shown in Tables 2 and S9, with the exception of openness and agreeableness in Sample 1 (below .55) and negative affectivity in Sample 2 (.76), rank-order stabilities ranged from .61 to .72.

OPEN IN VIEWER

Table 2.

Summary of Stability Indicators for Each Construct

	Sample 1			Sample 2
Constructs	Rank-order stability (r)	Mean of latent change (M)	Variance of latent change (SD)	Rank-order stability (r)	Mean of latent change (M)	Variance of latent change (SD)
Extraversion.r	.61 [.51, .72]	−.00 [−.02, .01]b	.60 [.56, .65]a	.70 [.62, .77]	.00 [−.01, .02]b	.53 [.48, .58]b
Agreeableness.r	.54 [.46, .62]b	.02 [−.00, .04]b	.74 [.68, .80] ab	.64 [.57, .71]	.04 [.03, .06]ab	.35 [.31, .39]a
Conscientiousness.r	.62 [.54, .71]	.00 [−.02, .02]ab	.63 [.57, .68]b	.70 [.63, .77]	−.00 [−.02, .01]a	.35 [.31, .39]
Neuroticism	.64 [.56, .71]	−.03 [−.05, −.01]	.78 [.72, .84]ab	.69 [.63, .75]	−.07 [−.09, −.05]b	.67 [.61, .73]ab
Openness	.51 [.42, .60]ab	.02 [.00, .04]ab	.79 [.73, .85]ab	.65 [.59, .72]	.04 [.02, .06]ab	.52 [.47, .57]ab
Detachment	.67 [.61, .74]	−.01 [−.03, .01]b	.70 [.65, .76]ab	.62 [.55, .69]	.00 [−.01, .02]b	.48 [.43, .53]b
Antagonism	.63 [.56, .71]	−.00 [−.02, .02]b	.61 [.56, .66]ab	.67 [.59, .75]	−.02 [−.03, −.00]a	.29 [.26, .32]a
Disinhibition	.69 [.63, .75]	−.02 [−.04, −.01]a	.65 [.60, .71]b	.71 [.65, .77]	−.04 [−.06, −.03]ab	.37 [.34, .41]b
Negative affectivity	.69 [.63, .75]	−.02 [−.04, −.01]	.70 [.64, .75]ab	.76 [.70, .82]	−.05 [−.07, −.04]b	.43 [.39, .47]ab
Psychoticism	.72 [.66, .78]a	−.03 [−.05, −.02]a	.65 [.60, .70]ab	.68 [.62, .75]	−.07 [−.09, −.05]ab	.39 [.35, .44]ab
LPFS	.66 [.60, .71]	−.04 [−.06, −.02]	.54 [.49, .59]	.71 [.66, .76]	−.02 [−.04, −.00]	.32 [.28, .35]

Note: Values in brackets are 95% confident intervals. Values that differ by gender are in italics. Letter “a” indicates significant difference between normal-range trait and its corresponding maladaptive traits. Letter “b” indicates significant difference with LPFS. Bold values indicate significant difference from 0. Reversed scores are indicated with “.r” at the end of the variable name. LPFS = Level of Personality Functioning Scale.

We used nested models to test differences in rank-order stability (illustrated in Fig. S3b in the Supplemental Material). We compared models in which latent correlations (r₁ and r₂ in Fig. S3b in the Supplemental Material) were freely estimated or restricted to be equal using χ²-difference test (Satorra & Bentler, 1994). As shown in Table 2, most comparisons were nonsignificant, with three exceptions. In Sample 1, openness and agreeableness were both less stable than LPFS, and openness was less stable than psychoticism. Overall, stabilities were similar across variables, and thus Hypothesis 2 was not supported.

Hypothesis 5: intercorrelations among normal-range traits and maladaptive traits

We computed latent correlations among normal-range traits, maladaptive traits, and LPFS (Table 3). For Hypotheses 5 to 7, we used Fisher’s r to z transformation before averaging correlations and used Fisher’s Z back-transformation after averaging. The average absolute intercorrelations for normal-range traits and maladaptive traits were .50 and .88 in Sample 1 and .34 and .68 in Sample 2, respectively. The differences exceeded our preregistered benchmark of .10 (.16 and .20, respectively), supporting Hypothesis 5. ⁴ In addition, the LPFS showed moderate correlations with normal-range traits (.30 for Sample 1 and .26 for Sample 2) and higher correlations with maladaptive traits (.70 for Sample 1 and .66 for Sample 2). For results for T2 through T4, see Tables S12 to S14 in the Supplemental Material. For results based on observed scores, see Tables S15 to S18 in the Supplemental Material.

OPEN IN VIEWER

Table 3.

Zero-Order and Partial Latent Correlation Matrices for Normal-Range Traits, Maladaptive Traits, and LPFS at Time 1

Sample 1												Sample 2
Zero-order correlation
	E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY	LPFS		E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY	LPFS
E.r	1.00											E.r	1.00
A.r	.58	1.00										A.r	.31	1.00
C.r	.76	.88	1.00									C.r	.52	.56	1.00
N	.27	.18	.35	1.00								N	.23	.03	.31	1.00
O	−.59	−.25	−.40	.24	1.00							O	−.51	−.24	−.46	.07	1.00
DET	.39	.46	.52	.71	.09	1.00						DET	.50	.21	.23	.45	−.14	1.00
ANT	.05	.50	.42	.49	.27	.85	1.00					ANT	−.03	.38	.34	.41	.26	.50	1.00
DIS	.25	.43	.51	.76	.11	.85	.75	1.00				DIS	.01	.28	.56	.55	−.14	.39	.60	1.00
NA	.24	.36	.45	.84	.21	.93	.78	.91	1.00			NA	.12	.18	.45	.88	−.02	.56	.67	.84	1.00
PSY	.23	.34	.43	.76	.24	.93	.84	.87	.95	1.00		PSY	.08	.06	.22	.77	.25	.68	.73	.69	.89	1.00
LPFS	.20	.32	.42	.78	.25	.85	.76	.85	.89	.91	1.00	LPFS	.16	.17	.41	.84	.08	.62	.68	.73	.93	.92	1.00
Controlling for LPFS
	E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY			E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY
E.r	1.00											E.r	1.00
A.r	.56	1.00										A.r	.29	1.00
C.r	.76	.87	1.00									C.r	.50	.55	1.00
N	.18	−.12	.03	1.00								N	.18	−.21	−.08	1.00
O	−.67	−.36	−.57	.08	1.00							O	−.53	−.25	−.54	.00	1.00
DET	.43	.38	.33	.16	−.23	1.00						DET	.52	.14	−.04	−.16	−.24	1.00
ANT	−.16	.41	.18	−.24	.14	.62	1.00					ANT	−.18	.37	.09	−.41	.28	.14	1.00
DIS	.16	.32	.31	.30	−.19	.45	.31	1.00				DIS	−.15	.24	.41	−.17	−.28	−.12	.21	1.00
NA	.14	.18	.19	.52	−.03	.72	.37	.66	1.00			NA	−.07	.07	.19	.46	−.27	−.04	.15	.66	1.00
PSY	.13	.13	.13	.20	.04	.73	.54	.43	.73	1.00		PSY	−.17	−.24	−.44	−.05	.45	.36	.38	.07	.24	1.00
Controlling for negative affect
	E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY			E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY
E.r	1.00											E.r	1.00
A.r	.57	1.00										A.r	.31	1.00
C.r	.74	.89	1.00									C.r	.51	.58	1.00
N	.16	.14	.26	1.00								N	.20	.03	.19	1.00
O	−.67	−.29	−.48	.12	1.00							O	−.53	−.24	−.52	−.02	1.00
DET	.34	.49	.48	.47	−.06	1.00						DET	.49	.22	.16	.38	−.19	1.00
ANT	−.06	.50	.36	.25	.20	.82	1.00					ANT	−.06	.40	.28	.29	.23	.45	1.00
DIS	.14	.48	.47	.51	−.05	.74	.68	1.00				DIS	−.04	.31	.50	.43	−.21	.32	.54	1.00
NA	.12	.41	.42	.62	.07	.89	.75	.84	1.00			NA	.08	.20	.36	.82	−.12	.52	.63	.81	1.00
PSY	.11	.37	.38	.47	.13	.89	.82	.75	.89	1.00		PSY	.03	.07	.09	.67	.21	.66	.70	.61	.85	1.00
Controlling for LPFS and negative affect
	E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY			E.r	A.r	C.r	N	O	DET	ANT	DIS	NA	PSY
E.r	1.00											E.r	1.00
A.r	.59	1.00										A.r	.29	1.00
C.r	.78	.87	1.00									C.r	.50	.56	1.00
N	.15	−.04	.09	1.00								N	.18	−.19	−.10	1.00
O	−.68	−.36	−.57	.06	1.00							O	−.54	−.25	−.55	−.03	1.00
DET	.43	.39	.33	.17	−.23	1.00						DET	.52	.13	−.03	−.14	−.23	1.00
ANT	−.14	.39	.16	−.16	.15	.64	1.00					ANT	−.18	.37	.10	−.40	.29	.13	1.00
DIS	.15	.36	.33	.24	−.21	.45	.36	1.00				DIS	−.15	.24	.41	−.17	−.28	−.12	.21	1.00
NA	.11	.25	.24	.42	−.05	.76	.48	.64	1.00			NA	−.07	.07	.19	.47	−.28	−.03	.16	.66	1.00
PSY	.11	.17	.16	.10	.03	.74	.61	.40	.71	1.00		PSY	−.17	−.24	−.44	−.06	.45	.37	.38	.07	.24	1.00

Note: Reversed scores are indicated with “.r” at the end of the variable name. E = extraversion; A = agreeableness; C = conscientiousness; N = neuroticism; O = openness; DET = detachment; ANT = antagonism; DIS = disinhibition; NA = negative affectivity; PSY = psychoticism; LPFS = Level of Personality Functioning Scale.

Hypothesis 6: personality functioning as an explanation for intercorrelation differences between maladaptive traits and normal-range traits

We sought to examine whether the intercorrelations of maladaptive traits would decrease after controlling for the LPFS. Following a reviewer’s suggestion, to confirm that the reduction in intercorrelations is primarily attributable to LPFS rather than general factors, we controlled for negative affect to account for general distress or response style and then controlled for LPFS (i.e., controlling for both negative affect and LPFS) to capture the specific reduction attributable to LPFS. When controlling only for LPFS but not negative affect, we found that the averaged intercorrelations of the maladaptive traits decreased from .88 and .68 to .57 and .25 in two samples, respectively. As shown in Table 3, after controlling only for negative affect, we found that the intercorrelations of maladaptive traits changed very little, decreasing from .88 to .82 in Sample 1 and from .68 to .63 in Sample 2. However, when LPFS was controlled for in addition to negative affect, the averaged intercorrelations of maladaptive traits further decreased to .60 and .25, respectively, and both reductions exceeded the preregistered benchmark of .10. Overall, Hypothesis 6 is confirmed.

Hypothesis 7: discriminant validity of maladaptive traits

We computed the latent correlation matrix between normal-range traits and maladaptive traits to evaluate their discriminant validity. In the lower-left box of Table 3, the off-diagonal coefficients represent correlations between nonmatching normal-range traits and maladaptive traits, which theoretically should not be high; thus, lower absolute values indicate higher discriminant validity. When controlling for LPFS, we found that the average absolute off-diagonal zero-order correlation in Sample 1 decreased from .40 to .20 and that those in Sample 2 decreased from .26 to .20. Furthermore, in Sample 1, the average absolute off-diagonal zero-order correlation was .40, decreasing to .30 after controlling for negative affect and further to .20 when both negative affect and LPFS were controlled. In Sample 2, the corresponding values were .26, .23, and .20. Taken together, these results partly confirm Hypothesis 7 that controlling for LPFS can improve the discriminant validity of maladaptive traits.

Hypotheses 8 through 10: joint prediction of maladaptive traits by personality functioning and normal-range traits

Hypothesis 8: cross-sectional regressions

We estimated five cross-sectional regression models in each sample with personality functioning and the corresponding normal-range trait as predictors of each maladaptive trait (as shown in Table 4). Across both samples, LPFS and normal-range traits jointly predicted maladaptive traits, except openness did not significantly predict psychoticism in Sample 1. LPFS showed consistently larger effect sizes than normal-range traits. Results at later time points replicated the T1 findings (see Table S19 in the Supplemental Material). Overall, Hypothesis 8 is supported.

OPEN IN VIEWER

Table 4.

Independent Predictions of Personality Functioning and Normal-Range Traits on Maladaptive Traits

Level	Sample	Predictor	E.r to DET	A.r to ANT	C.r to DIS	N to NA	O to PSY
Cross-sectional level (Hypothesis 8, with Time 1 data)	Sample 1	Normal trait	.25 [.18, .32]	.28 [.22, .34]	.16 [.09, .22]	.39 [.32, .47]	.04 [−.02, .09]
		LPFS	.71 [.66, .76]	.59 [.54, .64]	.70 [.66, .75]	.53 [.46, .61]	.82 [.78, .86]
	Sample 2	Normal trait	.38 [.31, .45]	.26 [.17, .34]	.29 [.21, .37]	.43 [.32, .55]	.19 [.12, .25]
		LPFS	.48 [.41, .55]	.51 [.43, .58]	.53 [.46, .59]	.52 [.41, .63]	.82 [.78, .86]
Longitudinal change score (Hypothesis 9)	Sample 1	Normal trait	.00 [−.03, .04]	.04 [.01, .08]	.04 [.01, .08]	.16 [.12, .19]	.04 [.01, .07]
		LPFS	.32 [.28, .36]	.28 [.24, .32]	.35 [.31, .39]	.26 [.21, .30]	.37 [.33, .41]
	Sample 2	Normal trait	−.02 [−.06, .03]	.08 [.04, .12]	.07 [.02, .11]	.14 [.10, .19]	.05 [.01, .09]
		LPFS	.26 [.22, .31]	.29 [.24, .33]	.28 [.23, .32]	.26 [.21, .31]	.35 [.29, .41]
Within-persons level (Hypothesis 10)	Sample 1	Normal trait	.07 [.03, .12]	.09 [.04, .13]	.06 [.02, .1]	.21 [.17, .25]	.00 [−.04, .05]
		LPFS	.37 [.33, .41]	.36 [.32, .40]	.42 [.38, .46]	.34 [.29, .38]	.45 [.41, .49]
	Sample 2	Normal trait	.04 [.00, .09]	.10 [.05, .14]	.10 [.05, .14]	.20 [.16, .25]	.06 [.02, .11]
		LPFS	.35 [.30, .39]	.37 [.32, .41]	.36 [.32, .41]	.33 [.29, .38]	.45 [.41, .49]

Note: Values in brackets are 95% confident intervals. Significant values (p < .05) are in bold. Values that differ by gender are in italic. Reversed scores are indicated with “.r” at the end of the variable name. E = Extraversion; A = Agreeableness; C = Conscientiousness;, N = Neuroticism; O = Openness; DET = Detachment; ANT = Antagonism; DIS = Disinhibition; NA = Negative affectivity; PSY = Psychoticism; LPFS = Level of Personality Functioning Scale.

Hypothesis 11: cross-lagged within-persons associations

Using RI-CLPMs, we estimated the standardized cross-lagged effects (c₁–c₆ in Fig. S9 in the Supplemental Material) to explore cross-lagged within-persons associations between personality functioning, normal-range traits, and maladaptive traits. These effects show how one construct predicts another from one occasion to the next at the within-persons level, serving as the test for Hypothesis 11. We fixed cross-lagged effects to be equal across intervals to obtain an average effect size. ⁷ Few (13/60) cross-lagged effects were significant, and only one significant path was replicated in both samples (Table 5). In Sample 1, we found four significant paths from LPFS to traits, three significant paths from traits to LPFS, and negative reciprocal association between openness and psychoticism. In Sample 2, the four significant cross-lagged effects all reflected LPFS predicting traits. This suggests that the cross-lagged effects among the three constructs are weak and less stable. When they were evident, it was more likely that changes in personality functioning preceded changes in traits than the other way around, providing evidence for the exploration of Hypothesis 11.

OPEN IN VIEWER

Table 5.

Cross-Lagged Effects Estimated With Trivariate Random-Intercept Cross-Lagged Panel Models

Model	Cross-lagged effect	Sample 1			Sample 2
		T1 → T2	T2 → T3	T3 → T4	T1 → T2	T2 → T3	T3 → T4
1	LPFS → extraversion.r	−.03 [−.10, .05]	−.03 [−.11, .05]	−.03 [−.10, .05]	.09 [.01, .17]	.09 [.01, .16]	.07 [.01, .13]
	LPFS → detachment	.07 [−.01, .14]	.08 [−.01, .17]	.07 [−.01, .16]	.10 [.01, .19]	.09 [.01, .17]	.07 [.00, .13]
	Extraversion.r → LPFS	.01 [−.06, .07]	.01 [−.06, .08]	.01 [−.06, .08]	−.07 [−.14, .01]	−.05 [−.11, .01]	−.04 [−.08, .01]
	Extraversion.r → detachment	.03 [−.04, .09]	.03 [−.03, .09]	.03 [−.04, .10]	−.05 [−.13, .03]	−.04 [−.09, .02]	−.03 [−.07, .02]
	Detachment → LPFS	−.01 [−.07, .06]	−.01 [−.09, .07]	−.01 [−.09, .07]	.02 [−.06, .10]	.02 [−.06, .10]	.02 [−.05, .08]
	Detachment → extraversion.r	.03 [−.04, .10]	.03 [−.04, .10]	.03 [−.04, .10]	−.03 [−.11, .04]	−.03 [−.10, .04]	−.03 [−.09, .03]
2	LPFS → agreeableness.r	.03 [−.05, .10]	.03 [−.06, .12]	.03 [−.05, .10]	.03 [−.06, .12]	.02 [−.04, .09]	.02 [−.04, .08]
	LPFS → antagonism	.11 [.04, .18]	.14 [.04, .23]	.12 [.04, .20]	.06 [−.03, .14]	.04 [−.03, .12]	.04 [−.02, .10]
	Agreeableness.r → LPFS	−.01 [−.08, .06]	−.01 [−.08, .06]	−.01 [−.07, .06]	.01 [−.06, .09]	.01 [−.06, .09]	.01 [−.06, .08]
	Agreeableness.r → antagonism	.02 [−.05, .09]	.02 [−.05, .09]	.02 [−.04, .08]	−.05 [−.12, .03]	−.04 [−.10, .02]	−.04 [−.11, .03]
	Antagonism → LPFS	.03 [−.04, .10]	.03 [−.05, .11]	.03 [−.04, .10]	.02 [−.06, .10]	.02 [−.05, .10]	.02 [−.05, .08]
	Antagonism → agreeableness.r	−.01 [−.08, .06]	−.01 [−.08, .07]	−.01 [−.07, .06]	−.05 [−.14, .03]	−.04 [−.10, .02]	−.04 [−.11, .02]
3	LPFS → conscientiousness.r	−.06 [−.14, .02]	−.07 [−.15, .02]	−.06 [−.14, .02]	.09 [.01, .18]	.07 [.00, .14]	.07 [.00, .13]
	LPFS → disinhibition	.05 [−.02, .13]	.06 [−.03, .15]	.06 [−.03, .15]	.06 [−.03, .14]	.05 [−.03, .13]	.04 [−.02, .11]
	Conscientiousness.r → LPFS	.02 [−.05, .08]	.02 [−.05, .09]	.02 [−.06, .09]	−.00 [−.08, .08]	−.00 [−.08, .07]	−.00 [−.07, .07]
	Conscientiousness.r → disinhibition	.04 [−.02, .10]	.04 [−.03, .10]	.05 [−.03, .12]	−.00 [−.08, .08]	−.00 [−.08, .08]	−.00 [−.07, .07]
	Disinhibition → LPFS	.06 [−.01, .13]	.07 [−.01, .15]	.07 [−.01, .14]	−.02 [−.10, .07]	−.01 [−.09, .06]	−.01 [−.07, .05]
	Disinhibition → conscientiousness.r	.06 [−.01, .13]	.06 [−.01, .13]	.06 [−.01, .13]	.03 [−.06, .11]	.02 [−.04, .09]	.02 [−.04, .08]
4	LPFS → neuroticism	.14 [.06, .21]	.16 [.07, .24]	.14 [.06, .23]	.04 [−.05, .13]	.03 [−.05, .11]	.03 [−.04, .10]
	LPFS → negative affectivity	.07 [.00, .15]	.09 [−.00, .18]	.08 [−.00, .17]	.07 [−.02, .16]	.06 [−.02, .15]	.05 [−.02, .11]
	Neuroticism → LPFS	.07 [.01, .13]	.08 [.01, .16]	.09 [.01, .17]	.06 [−.02, .14]	.06 [−.02, .13]	.05 [−.02, .11]
	Neuroticism → negative affectivity	−.02 [−.08, .05]	−.02 [−.09, .06]	−.02 [−.10, .06]	.01 [−.07, .09]	.01 [−.07, .08]	.01 [−.06, .07]
	Negative affectivity → LPFS	.05 [−.02, .12]	.06 [−.02, .14]	.06 [−.02, .13]	−.06 [−.14, .02]	−.06 [−.13, .02]	−.04 [−.10, .01]
	Negative affectivity → neuroticism	.02 [−.05, .09]	.02 [−.05, .09]	.02 [−.05, .10]	.03 [−.05, .11]	.03 [−.05, .10]	.02 [−.04, .09]
5	LPFS → openness	.04 [−.04, .11]	.05 [−.04, .14]	.04 [−.04, .13]	−.04 [−.13, .06]	−.03 [−.11, .05]	−.02 [−.08, .04]
	LPFS → psychoticism	.14 [.07, .21]	.17 [.08, .27]	.16 [.07, .25]	.10 [.01, .19]	.09 [.01, .18]	.07 [.00, .14]
	Openness → LPFS	−.06 [−.12, −.00]	−.07 [−.14, −.00]	−.07 [−.14, −.00]	−.02 [−.10, .05]	−.02 [−.09, .05]	−.02 [−.08, .04]
	Openness → psychoticism	−.10 [−.16, −.03]	−.10 [−.16, −.04]	−.10 [−.17, −.04]	−.05 [−.12, .02]	−.04 [−.11, .02]	−.04 [−.11, .02]
	Psychoticism → LPFS	.08 [.01, .14]	.10 [.01, .18]	.09 [.01, .17]	.05 [−.04, .14]	.04 [−.04, .12]	.03 [−.03, .09]
	Psychoticism → openness	−.12 [−.19, −.05]	−.14 [−.21, −.06]	−.13 [−.20, −.05]	.04 [−.05, .14]	.03 [−.04, .11]	.02 [−.03, .08]

Note: Values in brackets are 95% confidence intervals. Significant values (p < .05) are in bold. Reversed scores are indicated with “.r” at the end of the variable name. T1 = Time 1; T2 = Time 2; T3 = Time 3; T4 = Time 4; LPFS = Level of Personality Functioning Scale.

Similarities and differences between personality functioning, normal-range traits, and maladaptive traits

First, whereas normal-range traits have normal distributions, maladaptive traits are positively skewed, and the LPFS is somewhere in between. This is important because although one tail of each trait (as described by maladaptive traits) is more likely to be related to personality-related impairments, it is possible for tails not captured by maladaptive traits to be related to impairment for some people (Widiger & McCabe, 2020). For example, obsessive-compulsive problems may involve extreme conscientiousness (Samuel & Widiger, 2011) or low openness (Mike et al., 2018). Differences between maladaptive traits and normal-range traits in basic distributional properties also complicate the translation of basic personality science to personality-disorder assessment because nearly all evidence about personality in nonclinical research comes from measures with normally distributed scales.

Second, maladaptive traits are more strongly correlated with one another than normal-range traits, but correcting for personality functioning significantly reduces these intercorrelations. Third, controlling for personality functioning improved otherwise problematic discriminant validity of maladaptive traits in relation to normal-range traits. This pattern of results closely replicates previous findings (Emery et al., 2024; Morey, Good, & Hopwood, 2022) and further suggests that the skewed distributions of maladaptive traits may be driven by their saturation with personality dysfunction, which, in turn, contributes to the empirical overlap between maladaptive traits and personality dysfunction and negatively affects the discriminant validity of maladaptive traits. Fourth, in most cases, personality functioning and normal-range traits jointly predicted maladaptive traits across cross-sectional analyses, longitudinal models, and within-persons changes. These findings replicate results previously observed at the cross-sectional level (Emery et al., 2024; Morey, Good, & Hopwood, 2022) and to our knowledge, represent the first study to provide evidence for this perspective from longitudinal changes and within-persons fluctuations.

Taken together, these results provide a robust demonstration that maladaptive traits reflect a combination of normal-range traits and personality dysfunction. Consistent with this observation, the associations between LPFS and maladaptive traits were relatively high, and their associations with normal-range traits were only moderate. Perhaps the most important conclusion from these results—and from this study more generally—is that a significant source of overlap between personality functioning and traits in the AMPD is due to the trait measures focusing on maladaptive content. Replacing maladaptive traits with normal-range traits would make the A and B Criteria more distinct, allow for a more comprehensive assessment of personality style that capture both trails of each trait, and better connect Criterion B to basic personality research (Morey, McCredie, et al., 2022). However, this does not seem to be the direction that the AMPD or personality-disorder diagnosis is likely to go in (Sharp et al., 2025), and thus, overlap and redundancy between traits and functioning is likely to continue to be a problem.

Clinical and theoretical implications

The findings of this study regarding the associations between LPFS and traits carry important clinical and theoretical implications. The observation that LPFS was more skewed than normal-range traits is consistent with the LPFS being a clinical variable that reflects dysfunction, in contrast to traits that represent variation in personality across individuals regardless of clinical status. This is consistent with the view that personality functioning represents an intervention target, although it may also reflect features of these specific samples or the measure used in this study or the AMPD more generally.

Finding that changes in personality functioning were more likely to precede changes in traits than vice versa addresses an important theoretical question about the nature of personality disorder. This finding is consistent with the view that personality functioning is expressed through behaviors that can be described by traits, and thus changes in personality functioning lead to changes in trait expression (Haehner, Sleep, et al., 2024; Zimmermann et al., 2015). This may suggest that changing level of personality functioning could have the effect of improving maladaptive traits, reinforcing the idea of targeting personality functioning in therapeutic interventions.

Several findings in this study were consistent with the perspective that maladaptive traits represent a combination of personality functioning and normal-range traits (Emery et al., 2024; Morey, Good, & Hopwood, 2022). This helps explain empirical overlap between the two criteria of the AMPD. Clinically, it suggests that measuring personality functioning and normal-range traits will provide more comprehensive and useful information and that having done that, there is likely to be limited additional value to measuring maladaptive traits.

In contrast to some previous research (Bleidorn et al., 2022; Haehner, Sleep, et al., 2024), we did not find consistent support for stability differences in study variables. Even when differences have been found in previous research, they have tended to be small and inconsistent. One possible conclusion is that the stability of personality functioning, maladaptive traits, and normal-range traits is largely the same; another might be that differences are difficult to detect and highly sensitive to methodological factors, such as sampling and timescale. For example, in this study, there was relatively little mean-level change in general, perhaps because differences may be more likely to emerge in studies with more plausible reasons for change, such as life events or psychotherapy.

Population generalizability and implications on forensic practice

This study was conducted in two offender samples in China, providing a valuable opportunity to examine the generalizability of personality-disorder research in a diverse context. Findings were largely consistent across samples and with those of previous studies, generally supporting the applicability of personality assessment among Chinese offenders. Moreover, although both samples consisted of offenders, they differed substantially in demographic characteristics—most notably in gender and types of crimes—yet the overall pattern of results was replicated well. This strengthens confidence in the robustness and stability of the findings.

One of the central functions of mental-health care in prisons is to facilitate offenders’ reintegration into society, and many offenders are also motivated to pursue such change. Personality-related problems are also very common and challenging issues in prisons (Barrett & Byford, 2012; Putkonen et al., 2003). Overall, although effect sizes were modest, both personality functioning and maladaptive traits showed declines over 1 year, suggesting that the routine rehabilitative practices in the Chinese prison system may have a generally positive influence on personality pathology. At the same time, our findings on rank-order stability and individual differences in mean-level change revealed substantial heterogeneity in trajectories of change. This highlights the need for prison practitioners to attend to such individual variability in rehabilitation efforts and for researchers to investigate the factors underlying these differences to better facilitate improvements in offenders’ personality functioning.

Limitations and directions for future research

There were three main limitations to this study. First, all data were from self-report questionnaires. The shortcomings of self-report measures in general and these measures in particular (e.g., similarities in item content) could inflate associations and generally provide only one perspective on personality. Self-report is the most widely used approach in personality research and many clinical and forensic settings, and questionnaires perform comparably with or even outperform other measurement methods, such as interviews (Hopwood et al., 2008; Kaplan et al., 1994), behavioral measures (Kormos & Gifford, 2014), or implicit tests (Corneille & Gawronski, 2024). Thus, this approach provides a strong basis for testing study hypotheses, but there are multiple advantages to multimethod assessment for research and practice. For example, the correlations among maladaptive traits and between some of the Big Five traits were somewhat high, which may be related to response styles that could be amplified in prison settings. Specifically, in the prison environment, offenders’ defensive response tendencies when reporting maladaptive traits may constitute a common factor, thereby inflating the intercorrelations among maladaptive traits.

Second, the use of offender samples, although advantageous in certain respects, introduces unique challenges. Although the offender sample offers valuable insights into a specific population, results derived from this sample may not necessarily translate to nonoffender populations. Moreover, because both samples were Chinese offenders, any differences between these results and those from other studies may be related to Chinese culture, the offender population, or sampling error, and these three factors could not be distinguished in this study.

Third, patterns observed in longitudinal research are often influenced by the timescale of measurement (Hopwood et al., 2022). This study used four waves of data collected over 1 year with intervals of approximately 4 months. Although such timescales may be adequate to detect changes in personality functioning and traits, this was ultimately more of a practical than a theoretically informed decision. Different patterns would be likely to emerge at longer or shorter timescales.