arxiv: 2605.06989 · v1 · submitted 2026-05-07 · 📊 stat.AP · cs.AI· cs.LG· stat.ME

Recognition: 1 theorem link

· Lean Theorem

Drawing Lines in Psychological Space: What K-means Clustering Reveals in Simulated and Real Psychometric Data

Luiz Carlos Serramo Lopez, Maria Jullyanna Ferreira Marques, Pedro Henrique Ramos Pinto

Pith reviewed 2026-05-11 00:49 UTC · model grok-4.3

classification 📊 stat.AP cs.AIcs.LGstat.ME

keywords k-means clusteringpsychometric datacontinuous distributionslatent subgroupspsychological profilesclustering artifactssimulated datasurvey responses

0 comments

The pith

K-means clustering can generate stable, coherent groups even when the data has no true subgroup structure.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines the behavior of K-means clustering in psychological research by applying it first to simulated datasets drawn from continuous Gaussian distributions and then to responses from a large international psychometric survey. K-means works by partitioning space around centroids using geometric distances rather than testing for the existence of latent categories. The simulations and survey analysis both produce stable and visually coherent clusters despite the absence of any built-in subgroup structure. This matters for psychologists who rely on clustering to define profiles or typologies, because the method can impose apparent divisions on data that is fundamentally continuous. The contrast between the controlled simulations and the real survey data supports the conclusion that such clusters may arise as geometric artifacts.

Core claim

K-means clustering partitions multidimensional space into regions around centroids, favoring compact, approximately spherical clusters defined by geometric distance. In a sequence of controlled simulated datasets drawn from continuous Gaussian latent spaces and in the SMARVUS survey of university student responses across 35 countries, this partitioning produces stable and visually coherent solutions even though no true subgroup structure is present.

What carries the argument

K-means partitioning of space by minimizing within-cluster sums of squared distances to centroids, which imposes compact spherical geometry regardless of the underlying distribution.

Load-bearing premise

The chosen simulated datasets and the international survey responses are representative enough of typical psychometric data for the observed clustering behavior to hold more generally.

What would settle it

Replicating the Gaussian simulation in the same dimensionality as the survey items, running K-means with varied random starts, and finding that the resulting partitions are unstable across runs or lack visual coherence in low-dimensional projections.

read the original abstract

K-means clustering is widely used in psychological and psychometric research to identify profiles, subgroups, and potential typologies, yet its classical formulation does not test whether such groups exist as latent psychological categories. Instead, K-means partitions multidimensional space into regions around centroids, favoring compact, approximately spherical clusters defined by geometric distance. In this paper, we examine this limitation through a sequence of controlled simulated datasets. We then extend the analysis to the SMARVUS dataset, a large international psychometric dataset comprising survey responses from university students across 35 countries, to evaluate whether similar geometric partitioning patterns emerge in empirical psychological data. By contrasting simulated and empirical data, this paper argues that K-means can produce stable and visually coherent clustering solutions even in continuous Gaussian latent spaces without true subgroup structure.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

K-means will split continuous Gaussian data into apparently stable clusters with no subgroups present, and the paper shows this in simulations plus SMARVUS responses but adds no new theory.

read the letter

The main point is straightforward: K-means clustering partitions space around centroids and will produce compact groups even when the data is generated from a single multivariate Gaussian with no latent structure. The paper runs controlled simulations to demonstrate this and then applies the same procedure to the SMARVUS international student survey data to check whether similar patterns appear in real psychometric responses. That contrast is the core of the work. The simulations are the stronger part. Generating data without built-in subgroups by construction lets them isolate the algorithm's geometric bias cleanly, and showing that the resulting partitions can still look stable and visually coherent is a useful reminder for anyone using K-means to hunt for psychological types. The SMARVUS application serves as a practical check rather than a broad claim, which keeps the scope reasonable. The limitation is that this behavior of K-means is already documented in the clustering literature. The paper does not derive a new result or propose a fix; it mainly transplants the known issue into a psychometric context. The abstract gives little detail on simulation parameters, stability measures, or how they handled the real-data preprocessing, so it is difficult to judge robustness without the full methods. The real-data section is framed as an existence demonstration, not a generalization, which is appropriate but also limits how far the caution can be pushed. This is the sort of paper that psychometricians who routinely run K-means on survey data to identify profiles should see. It would be worth a referee's time because the demonstration is direct and the warning is relevant to common practice in the subfield, even though the underlying observation is not novel. I would send it for review with requests to cite the prior clustering work more explicitly and to report quantitative stability metrics from the simulations.

Referee Report

2 major / 3 minor

Summary. The paper demonstrates through controlled simulations that K-means clustering can produce stable and visually coherent partitions in data drawn from a single multivariate Gaussian distribution without any latent subgroup structure. It then applies the same clustering approach to the SMARVUS dataset of psychometric survey responses from university students in 35 countries, observing similar patterns, and concludes that K-means may impose apparent structure on continuous psychological data.

Significance. This provides a clear cautionary tale for the use of clustering methods in psychological research. By showing that the method can yield interpretable-looking clusters even when none exist by construction in the simulations, and that analogous results appear in real data, the paper underscores the need for caution in interpreting K-means results as evidence of psychological typologies. The controlled nature of the simulations is a strength, offering a direct counterexample to the assumption that coherent clusters imply discrete categories.

major comments (2)

§3 (Simulations): The simulation setup is central to the existence claim, but the manuscript does not provide sufficient detail on the data generation process, such as the specific covariance structure, number of dimensions, or sample sizes used in the Gaussian simulations. Without these, it is difficult to assess how general the finding is or to replicate the exact conditions under which stable clusters emerge.
§4 (SMARVUS application): There is no quantitative assessment of cluster stability (e.g., using metrics like the adjusted Rand index across multiple runs or bootstrap resampling). The reliance on visual inspection alone makes the claim of 'stable' solutions less rigorous, especially since the central argument hinges on stability in the absence of true structure.

minor comments (3)

The abstract could more explicitly state the number of clusters tested or the criteria used for choosing K.
Figure legends should include more detail on what the axes represent and any preprocessing applied to the data points.
A brief discussion of alternative clustering methods (e.g., Gaussian mixture models) that might better handle continuous data would contextualize the findings.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their supportive summary and for identifying areas where additional detail and rigor would strengthen the manuscript. We address each major comment below and will incorporate revisions to improve reproducibility and quantitative support for our claims.

read point-by-point responses

Referee: §3 (Simulations): The simulation setup is central to the existence claim, but the manuscript does not provide sufficient detail on the data generation process, such as the specific covariance structure, number of dimensions, or sample sizes used in the Gaussian simulations. Without these, it is difficult to assess how general the finding is or to replicate the exact conditions under which stable clusters emerge.

Authors: We agree that the simulation details require expansion for reproducibility and to clarify the scope of the findings. In the revised version we will add explicit descriptions of the covariance matrices (identity or low-variance diagonal structures), the dimensionality (5–10 variables chosen to mimic typical psychometric scales), and the sample sizes (n = 500–2000 per replication). We will also deposit the exact data-generation code in a public repository and reference it in the text. revision: yes
Referee: §4 (SMARVUS application): There is no quantitative assessment of cluster stability (e.g., using metrics like the adjusted Rand index across multiple runs or bootstrap resampling). The reliance on visual inspection alone makes the claim of 'stable' solutions less rigorous, especially since the central argument hinges on stability in the absence of true structure.

Authors: We accept that quantitative stability metrics would make the argument more rigorous. While the original manuscript relied on repeated visual inspection of partitions and silhouette widths, the revision will include adjusted Rand indices computed across 50 random initializations of K-means for both the simulated and SMARVUS data, together with a brief bootstrap resampling exercise. These additions will be presented alongside the existing figures without changing the substantive conclusion. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical demonstration rather than a derivation: it generates data from a single multivariate Gaussian (by construction, no latent subgroups), applies K-means, and shows that stable partitions still appear; it then applies the same procedure to the SMARVUS survey responses. No equations, fitted parameters, or predictions are claimed; the central existence claim follows directly from the controlled simulation design and the real-data application. No self-citations, ansatzes, or uniqueness theorems are invoked as load-bearing steps. The analysis is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the assumption that the chosen simulation distributions and the SMARVUS sample adequately represent the continuous nature of typical psychological data; no new mathematical axioms or invented entities are introduced.

pith-pipeline@v0.9.0 · 5442 in / 1081 out tokens · 51301 ms · 2026-05-11T00:49:33.609916+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear
K-means can produce stable and visually coherent clustering solutions even in continuous Gaussian latent spaces without true subgroup structure.

Reference graph

Works this paper leans on

4 extracted references · 4 canonical work pages

[1]

L., Heinrichs, N., Kim, H.-J., & Hofmann, S

Baker, S. L., Heinrichs, N., Kim, H.-J., & Hofmann, S. G. (2002). The Liebowitz Social Anxiety Scale as a self-report instrument: A preliminary psychometric analysis. Behaviour Research and Therapy, 40 (6), 701–715. DOI: 10.1016/s0005-7967(01)00060-2 Benson, J., & El-Zahhar, N. (1994). Further refinement and validation of the Revised Test Anxiety Scale. S...

work page doi:10.1016/s0005-7967(01)00060-2 2002
[2]

J., Cortes, R

DOI: 10.1038/s41398-026-03832-x Daker, R. J., Cortes, R. A., Lyons, I. M., & Green, A. E. (2020). Creativity anxiety: Evidence for anxiety that is specific to creative thinking, from STEM to the arts. Journal of Experimental Psychology: General, 149 (1), 42–57. DOI: 10.1037/xge0000630. de la Fuente-Tomás, L., Arranz, B., Safont, G., Sierra, P., Sánchez-Au...

work page doi:10.1038/s41398-026-03832-x 2020
[3]

DOI: 10.1186/s12888-021-03648-7 Lötsch, J., & Ultsch, A. (2024). Comparative assessment of projection and clustering method combinations in the analysis of biomedical data. Informatics in Medicine Unlocked, 50 , 101573. DOI: 10.1016/j.imu.2024.101573 Orchard C., Lin E., Rosella L., Smith P. M. (2024). Using unsupervised clustering approaches to identify c...

work page doi:10.1186/s12888-021-03648-7 2024
[4]

G., & Kim, J

DOI: 10.5334/jopd.80 Kwak, S. G., & Kim, J. H. (2017). Central limit theorem: the cornerstone of modern statistics. Korean J Anesthesiol. 2017 Apr;70(2):144-156. DOI: 10.4097/kjae.2017.70.2.144

work page doi:10.5334/jopd.80 2017