Using a cross-lagged panel model, Wan et al. interpreted their findings as evidence for reciprocal longitudinal relationships between illness-related stigma and loneliness among stroke survivors. However, estimates from cross-lagged models can reflect statistical artifacts rather than true prospective influences. In the present study, we reexamined this conclusion by fitting several alternative models to data simulated to reproduce the correlational structure reported by Wan et al. Across models, estimates varied substantially, yielding both positive and negative prospective effects between stigma and loneliness. When these divergent estimates were synthesized using meta-analytic procedures, the overall effects were not statistically distinguishable from zero. These findings suggest that the reported longitudinal effects may not represent genuine influences. Consistent with multiverse principles, we argue that conclusions should be informed by results across multiple plausible analytic specifications rather than a single modeling approach.

Based on statistically significant adjusted effects in multiple regression models, Sak et al. concluded that parental education and occupational status have independent additive effects on children’s intelligence over and above the contribution of the other. However, we show that the same data could be interpreted to support the paradoxical conclusion that low parental education or occupational status may counteract a low value on the other and help offspring to have the same intelligence as children with parents with higher education or occupational status. This is an example of the regression paradox which indicates that observed adjusted regression effects may be spurious. Hence, the conclusions by Sak et al. may be challenged. We recommend researchers to scrutinize their adjusted regression effects. If analyses identify the presence of the regression paradox, caution is advised and confident conclusions should be postponed.

Chen et al. examined temporal relations among maternal health literacy, psychological empowerment, and health-promoting behaviors and interpreted their findings as evidence of reciprocal longitudinal influences. However, significant cross-lagged coefficients do not necessarily demonstrate true prospective effects, as such estimates may arise from methodological and statistical artifacts. Using simulated data constructed to match the sample size and correlation matrix reported by Chen et al., we assessed the stability of the reported findings across several analytically plausible models. The results differed markedly depending on the specification employed. Some models suggested positive longitudinal effects, whereas others indicated negative or non-significant associations. After combining all estimates using meta-analytic methods, no overall effect reached statistical significance. These findings indicate that the conclusions drawn by Chen et al. should be interpreted with caution and that the reported cross-lagged associations may not reflect genuine prospective influence. In accordance with multiverse analytic principles, researchers should routinely evaluate multiple reasonable model specifications and synthesize evidence across models before drawing substantive conclusions.

Fu et al. (2026) recently reported significant associations in a cross-lagged panel network (CLPN) analysis and interpreted these findings as evidence of mutually reinforcing relationships among harsh parental discipline, negative automatic thoughts, and depressive symptoms. Yet it is well established that cross-lagged coefficients can arise from statistical artifacts rather than genuine prospective influences. To evaluate the robustness of these conclusions, we fitted a set of alternative models to data simulated to approximate the characteristics of the dataset analyzed by Fu et al. Depending on the analytic specification, estimated relationships among the study variables were either positive or negative. When these alternative estimates were synthesized meta-analytically, the resulting effects generally did not differ significantly from zero. Consequently, the associations reported by Fu et al. may reflect spurious rather than substantive prospective relationships, rendering strong interpretations premature. Researchers should remain cautious when interpreting correlations, including cross-lagged effects, obtained from observational data. We therefore recommend evaluating findings across multiple analytic specifications and drawing conclusions only after considering the consistency of results across models.

Recently, Zhang et al. concluded decreasing prospective effects between meaning in life and ostracism based on findings from the cross-lagged panel model (CLPM). However, it is well known that effects in the CLPM may be spurious. Here, we fitted alternative models to data and found discrepant decreasing, increasing, and null effects of meaning in life on subsequent change in ostracism, and vice versa, depending on the analyzed model. Hence, we conclude that the findings by Zhang et al. may have been spurious and their conclusions premature. It is important to bear in mind that correlations, including cross-lagged effects, may be spurious in order not to overinterpret findings. We recommend, in line with multiverse methodology, researchers to fit alternative models to data and to base conclusions on an aggregation of findings.

Using results from a cross-lagged panel model, Peng et al. argued that internet gaming disorder (IGD) and sleep problems exert reciprocal longitudinal influences on one another during adolescence. However, statistically significant cross-lagged associations do not necessarily indicate genuine prospective effects and may instead arise as artifacts of the underlying data structure. We fitted several alternative models to data simulated to reproduce the correlation structure reported by Peng et al. The analyses yielded conflicting results, with some models indicating increasing effects and others suggesting decreasing effects between IGD and sleep problems. Furthermore, a model assuming spurious longitudinal associations provided an adequate representation of the data without requiring direct effects between the constructs over time. Consequently, the findings reported by Peng et al. may reflect statistical artifacts rather than true prospective influences. Consistent with multiverse methodology, we recommend evaluating competing models and drawing conclusions from the collective pattern of evidence rather than from a single analytic approach.

In a recent study, Nagata et al. concluded prospective within-individual associations between problematic social media use (PSMU) and ADHD symptoms. In the present reanalyses, we report discrepant increasing and decreasing effects of PSMU on subsequent change in ADHD symptoms, and vice versa, depending on the model used to analyze data. Meta-analytic aggregations of these divergent effects did not differ significantly from zero. Hence, we conclude that the findings by Nagata et al. may have been spurious and their conclusions premature. It is important to bear in mind that correlations, including cross-lagged effects, in observational data may be spurious in order not to overinterpret findings. For increased analytic rigor, we recommend researchers to fit, as we did here, alternative models to data to verify whether results are robust under different justifiable analytical strategies.

Xu and Li recently reported prospective associations between mobile phone addiction, bedtime procrastination, and physical activity based on analyses using cross-lagged panel models (CLPMs). However, estimates obtained from CLPMs are known to be vulnerable to statistical artifacts and may not necessarily reflect genuine prospective influences. We reanalyzed the reported correlational structure using alternative analytic specifications and obtained markedly different conclusions depending on model choice, including positive, negative, and null effects. When these discrepant estimates were meta-analytically synthesized, the overall effects were indistinguishable from zero. These findings suggest that the reported associations may not provide convincing evidence of prospective influence and that the conclusions drawn by Xu and Li may be premature. More generally, the present commentary highlights the importance of evaluating findings across multiple plausible analytic approaches and avoiding strong inferences based on a single model specification.

Cross-lagged panel network (CLPN) models estimate prospective effects between several nodes (e.g., symptoms) while adjusting for initial scores on the outcome variables. However, it is well established that such adjusted cross-lagged effects may be spurious due to correlations with residuals and regression toward the mean. Here, we recommend that researchers conduct multiverse analyses, where cross-lagged effects are estimated with alternative models. Then, conclusions can be based on meta-analytic averaging of the estimated effects. Multiverse analyses will add rigor and transparency to analyses by acknowledging and incorporating, instead of ignoring, uncertainty due to the analyzed model. In an application of this methodology, we found that most cross-lagged effects between indicators of psychological flexibility and inflexibility, reported in a recent study, did not survive scrutiny.

Based on longitudinal associations, Geng et al. concluded that parental PTSD symptoms may impact children’s PTSD symptoms and the effect may be mediated by parenting style and children’s emotion regulation abilities. However, correlations do not prove causality. Here, we show that the same covariance structure could be due to a correlation between children’s latent general negativity and parental PTSD symptoms and this correlation could, for example, be due to common genetic vulnerability. Hence, the conclusions by Geng et al. should not be taken at face value.

Recently, Kapel Lev-ari et al. concluded, based on effects in cross-lagged panel models (CLPM), that betrayal is central in the development and maintenance of PTSD and depression. However, it is well known that cross-lagged effects in the CLPM may be spurious. Here, we reanalyzed data generated to resemble the data used by Kapel Lev-ari et al. We report contradicting increasing and decreasing prospective effects between sense of betrayal, PTSD, and depression, depending on the analyzed model, and statistically non-significant meta-analytic aggregations of these discrepant effects. Hence, the findings by Kapel Lev-ari et al. may have been spurious and their conclusions may be challenged. We recommend researchers to scrutinize prospective effects by fitting alternative models to data and to base conclusions on an aggregation of findings.

Drawing on the covariance matrix published by Ye et al., we reexamined the evidence underlying their proposed mediation model linking adverse childhood experiences to suicide risk in adolescents with depression. We show that the observed associations can be reproduced almost exactly by a different latent-variable model that does not rely on the directional assumptions advanced in the original study. Within this alternative framework, self-reported adverse childhood experiences, anhedonia, self-hate, and suicide risk are treated as observable manifestations of a broader construct reflecting core self-evaluation. The results illustrate the challenge of model underdetermination in observational research, whereby competing theoretical accounts may provide equally good representations of the same data. These findings suggest that alternative model specifications should be considered and that correlational evidence alone offers limited support for specific influencing mechanisms.

Mercier and Lubart analyzed data (N = 1384) with necessary condition analysis (NCA) and concluded that creativity self-efficacy and, to a lesser degree, creative process engagement, creative personal identity, openness to experience, and creative personality appeared necessary for creativity in the workplace. However, it has been established that necessity effects in NCA may be spurious due to correlations between the variables. Here, we estimated ranges of spuriousness across 1000 necessity effects estimated in data generated to have the same sample size and correlations between variables as in the data used by Mercier and Lubart. The necessity effects reported by Mercier and Lubart did not fall above these ranges of spuriousness, meaning that the reported effects may have been spurious and the conclusions by Mercier and Lubart premature. It is important for users of NCA to be aware that necessity effects in NCA do not prove necessity any more than correlations prove causality. We recommend researchers using NCA to scrutinize their findings by estimating, as we did here, ranges of spuriousness and to require that necessity effects fall above this range before claiming necessity.

The Dunning-Kruger effect describes a phenomenon where individuals with low ability allegedly tend to overestimate their ability more than individuals with higher ability. According to a contemporary operationalization of the Dunning-Kruger effect, individuals with low measured ability are predicted to have higher self-rated ability in a LOESS (locally estimated scatterplot smoothing) model compared with a linear regression model. In simulations we show that a Dunning-Kruger effect can appear due to an impact of disturbance on measured ability, even when self-rated ability is a perfect measure of true ability. A higher self-rated than measured ability may be due to measured ability underestimating true ability rather than due to self-rated ability being an overestimation. Hence, Dunning-Kruger effects do not prove that individuals with low measured ability overestimate their true ability.

Based on positive cross-lagged effects in data (N = 357, 51.5% female) from two waves of measurement, Swingler et al. concluded that violence exposure in early adolescence represents a prospective risk factor for conduct problems. However, it is well known that cross-lagged effects may be spurious. We fitted alternative models to data simulated to resemble the data used by Swingler et al. and found discrepant increasing, decreasing, and null effects of violence exposure on subsequent change in conduct problems depending on the analyzed model. A meta-analytic pooling of these discrepant effects did not differ significantly from zero. Hence, the findings by Swingler et al. may have been spurious and their conclusion premature. It is important for researchers to bear in mind that associations, including cross-lagged regression effects, in observational (i.e., non-experimental) data may be spurious in order not to overinterpret findings. We recommend researchers to scrutinize cross-lagged effects by fitting alternative models to data and to base conclusions on a juxtaposition of findings.

There are several models for estimating prospective within-individual effects between constructs. Researchers in psychology usually pick one model and ignore the others. The objective of the present study was to show how multiverse analyses with meta-analytic aggregation can be used for assessing prospective effects. We fitted the random-intercept cross-lagged panel model (RI-CLPM), the latent change score model (LCSM), the stable trait, autoregressive trait, and state (STARTS) model, a reversed version of the RI-CLPM, as well as corresponding multilevel models (MLM) on data on trust, loneliness, and life satisfaction. The fitted models suggested diametrically different prospective effects. Meta-analytic aggregations, on the other hand, indicated increasing prospective within-individual effects between loneliness and trust and between loneliness and life satisfaction and decreasing prospective effects between trust and life satisfaction. However, a good fit of the model of spurious longitudinal associations (MoSLA) suggested that the effects may have been spurious. Analyses of within-individual prospective effects may suggest diametrically different results depending on used model. For increased rigor and transparency, we recommend researchers to use multiverse analyses with meta-analytic aggregation and the MoSLA.

Based on findings in cross-lagged panel network (CLPN) models, Ma et al. concluded prospective effects between indicators of mindfulness and mental health problems. Here, we used multiverse methodology and found discrepant increasing, decreasing, and null effects depending on the used model, and meta-analytic aggregations of these discrepant effects did not differ significantly from zero. Hence, the conclusions by Ma et al. can be challenged. It is important for researchers to bear in mind that correlations, including cross-lagged effects in CLPN models, do not prove causality in order not to overinterpret findings, something that may have happened to Ma et al. For increased analytic rigor, we recommend researchers to fit, as we did here, alternative models to data and to juxtapose findings.

Based on statistically significant effects in cross-lagged panel network (CLPN) models, Ning and Zou concluded bidirectional prospective influences between negative self-attention and depression. However, cross-lagged effects may be spurious rather than due to genuine associations. In reanalyses of the data used by Ning and Zou, we found paradoxical increasing, decreasing, and null prospective effects between symptoms of negative self-attention and depression, depending on the analyzed model, and meta-analytic aggregations of these discrepant effects did not differ significantly from zero. Hence, the findings by Ning and Zou may have been spurious and their conclusions premature. It is important for researchers to bear in mind that correlations, including cross-lagged effects, in observational (i.e., non-experimental) data may be spurious in order not to overinterpret findings. We recommend researchers to fit alternative models to data and base conclusions on a juxtaposition of findings.

Wang et al. reported significant associations in a cross-lagged panel network (CLPN) model and interpreted these findings as evidence that meaning in life and internalizing symptoms may mutually diminish one another over time. Because cross-lagged coefficients obtained from observational data can arise from statistical artifacts rather than genuine prospective processes, we revisited this conclusion using a multiverse analytic framework. Applying several alternative longitudinal models to data simulated from the correlation structure reported by Wang et al., we observed patterns that varied substantially across specifications. Some models suggested protective effects, others indicated adverse effects, and several yielded negligible associations. When estimates were synthesized using meta-analytic aggregation, the resulting effects did not differ from zero. These findings imply that the reported CLPN effects may not provide robust evidence for reciprocal longitudinal influences between meaning in life and internalizing symptoms. More broadly, the results highlight the need for caution when interpreting cross-lagged associations from non-experimental data and underscore the value of evaluating conclusions across multiple analytic specifications.