Pith. sign in

REVIEW 2 major objections 7 minor 238 references

Gain rule on variational bound recovers true factor count

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

2026-07-09 18:35 UTC pith:HKWL6CEU

load-bearing objection Solid, practical methodology paper. The hard/soft selection framework and gain rule are genuinely useful. Main concern is the variational gap assumption, but it's bounded. the 2 major comments →

arxiv 2607.07159 v1 pith:HKWL6CEU submitted 2026-07-08 stat.ME cs.CL

Recovering Latent Structures after Variational Bayesian Variable Selection: Fit Assessment and Factor-Number Selection in Partially Exploratory Factor Analysis

Jinsong Chen , Yi Jin This is my paper
classification stat.ME cs.CL
keywords Bayesian variable selectionspike-and-slab priorsvariational inferencefactor analysismodel fit assessmenteffective degrees of freedom
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

When researchers use Bayesian variable selection to discover the loading structure of a factor model, the converged solution carries inclusion probabilities for every candidate loading but no standard way to assess fit or decide how many latent factors exist. This paper converts the variational posterior into a conventional covariance model in two ways: hard selection thresholds inclusion probabilities into a fixed sparse pattern, while soft selection retains them as fractional weights yielding an effective parameter count in the spirit of Bayesian model comparison. From these two representations the authors derive degrees of freedom, absolute fit indices (RMSEA, SRMR, CFI, TLI), and relative criteria (AIC, BIC, ELBO) that respect the selection step. The central practical contribution is a scale-free gain rule for choosing the number of factors: retain factors only up to the point where the marginal improvement in a criterion drops below a fixed fraction (default 10%) of the largest improvement observed, with a sustained-drop guard requiring the decline to persist for two consecutive steps. Simulations show that absolute indices track loading recovery and flag under-factoring but are blind to over-factoring, that raw relative criteria systematically over-factor because they cannot distinguish genuine from nuisance dimensions, and that the gain rule applied to the variational evidence lower bound (ELBO) recovers the true dimensionality in 95% of high-dimensional replications versus 57% for the raw ELBO optimum. An empirical analysis of the 100-item PID-5 personality inventory shows that two completely disjoint backbone specifications recover the same 22-factor major structure, fitting better than the confirmatory 25-facet model despite a plug-in evaluation that penalizes the exploratory solution.

Core claim

The scale-free gain rule with sustained-drop guard converts the over-factoring tendency of raw relative criteria into accurate dimensionality recovery by thresholding marginal criterion gains against the largest gain in the window rather than locating a global optimum. Applied to the ELBO, this rule recovers the true number of factors across a wide range of nuisance structure, sample sizes, and specification levels where raw ELBO, AIC, and BIC optima all drift upward. The mechanism is that genuine factors produce large marginal improvements while nuisance factors produce negligible ones, so a ratio threshold on gains cleanly separates the two without requiring any calibrated absolute cutoff.

What carries the argument

The gain rule computes marginal gains g(K) = c(K) - c(K-1) for a criterion c oriented so larger is better, finds the maximum gain g_max, and selects the last K whose gain exceeds delta% of g_max, with a sustained-drop guard requiring s-1 subsequent gains to also fall below the threshold. Hard selection thresholds inclusion probabilities at tau (default 0.50) into a binary loading pattern with nominal or rank-adjusted parameter counts; soft selection retains probabilities as effective weights yielding posterior-expected parameter counts analogous to p_D in DIC or p_WAIC.

Load-bearing premise

The ELBO gain rule depends on the variational gap the difference between the ELBO and the true log evidence being roughly comparable across candidate models with different numbers of factors. If this gap shrinks or grows systematically with K, the marginal ELBO gains no longer reflect genuine evidence gains, and the rule's accuracy in the simulations could be partly coincidental rather than a property of the criterion itself.

What would settle it

A simulation or empirical setting where the variational gap varies systematically with K, causing the ELBO gain path to separate at the wrong dimensionality. More directly: if the gain rule were applied to exact log-marginal-likelihood gains (removing the variational approximation) and selected a different K than the ELBO gain rule in a substantial fraction of cases, the variational gap assumption would be falsified.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • The gain rule principle is criterion-agnostic and could be applied to any model-selection path where adding parameters yields diminishing returns, including mixture model component selection, network edge selection, or tree depth in gradient boosting.
  • The hard-soft parameter-count gap (the difference between thresholded and effective parameter counts) serves as a built-in diagnostic for selection uncertainty that could be reported alongside any fit index to warn analysts when thresholding discards substantial posterior information.
  • The finding that stronger specification can reduce selection accuracy because it frees added factors to capture genuine nuisance covariance suggests that partial specification has a non-monotonic relationship with recovery quality, which has implications for how much prior structure to impose in exploratory settings.
  • The identification boundary that unspecified collinear clusters are resolved arbitrarily while well-separated major factors are recovered robustly provides a formal target for future work on partial identification in exploratory latent variable models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the variational gap (KL divergence between the variational approximation and the true posterior) varies systematically with the number of factors, ELBO differences across K values would not reflect genuine evidence differences, and the gain rule's accuracy could be a consequence of the gap's behavior rather than the criterion's fidelity to model evidence. The paper acknowledges this is untestab
  • The gain rule's sustained-drop guard with s=2 costs one extra factor of over-factoring headroom. In settings where factor estimation is expensive (e.g., large-scale genomic or neuroimaging data), the optimal s may trade off computational cost against robustness differently than in the simulations studied.
  • The finding that CFI and TLI cleanly separate under-factoring from adequate factoring while RMSEA and SRMR are uninformative at high dimensions (J=80) suggests that conventional cutoff recommendations for RMSEA and SRMR may need recalibration as the number of variables grows, independent of the variable-selection machinery proposed here.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 7 minor

Summary. This paper develops post-selection fit assessment tools for the regularized variational approximation for partially confirmatory factor analysis (PCFA-VA), extending its use to partially exploratory factor analysis (PEFA) where both the loading structure and the number of factors are weakly specified. The core contributions are: (1) converting converged variational solutions into covariance models via hard selection (thresholding inclusion probabilities) or soft selection (retaining them as effective parameter weights); (2) deriving corresponding degrees of freedom, absolute fit indices (RMSEA, SRMR, CFI, TLI), and relative criteria (AIC, BIC, ELBO); and (3) proposing a scale-free gain rule with a sustained-drop guard for selecting the number of factors. Two simulation studies calibrate the tools, and an empirical analysis of the 100-item PID-5 demonstrates the workflow. The gain rule, applied to the ELBO path, is shown to recover true dimensionality far more robustly than raw criteria, which systematically over-factor.

Significance. The paper addresses a genuine methodological gap: Bayesian variable selection in factor analysis produces inclusion probabilities, but the field lacks principled post-selection fit statistics that respect the selection step. The hard/soft selection distinction and the connection to effective degrees of freedom (p_D, p_WAIC, generalized d.f.) is well-motivated and correctly positioned in the literature. The scale-free gain rule with a formal correctness condition (Proposition 1) is a clean, falsifiable contribution. The reproducible code and simulation archives, the exact bracketing condition, and the disjoint-backbone robustness test in the empirical example are all commendable. The finding that the ELBO gain is most robust under heavy nuisance, while raw criteria collapse, is practically important for researchers working with large personality inventories.

major comments (2)
  1. The ELBO comparability assumption (Eqs. 31-32 and surrounding text) is the most load-bearing assumption in the paper. ELBO differences equal log-evidence differences minus differences in the KL divergence term across candidate K values. The paper states this is 'untestable directly' and relies on simulations. Two specific concerns: (1) The simulations use data generated from linear Gaussian factor models fit with normal-theory variational inference—a well-specified regime where the variational approximation is likely good and the KL gap relatively stable. The PID-5 empirical application uses 4-category items treated as continuous, introducing misspecification that could shift the KL gap behavior across K. (2) In Study 2 under heaviest nuisance (3w+3m), ELBO gain accuracy drops from 99.3% to 88.3% (Appendix A, Table A1 note). The paper does not disentangle whether this degradation stems (
  2. The plug-in nature of the fit statistics (Section on Absolute and Relative Fit Indices) creates a two-sided bias that is acknowledged but not fully bounded. The paper notes that the plug-in direction is conservative (shrunken variational estimates bound the minimized discrepancy from above) while the selection direction is optimistic (hard-selected d.f. are conditional on a data-chosen pattern). The hard-soft parameter-count gap is offered as a diagnostic, not a correction. However, the paper does not provide any simulation evidence on the magnitude of the post-selection optimism in T_H (Eq. 24) or its downstream effect on RMSEA_H, CFI_H, TLI_H. Since these indices are proposed as the primary reporting tools, even a brief simulation reporting the mean and SD of the gap between the plug-in T_H and the ML-refit T_H on the same selected pattern would substantially strengthen the claim that
minor comments (7)
  1. The notation switches between PCFA-VA and PCFA V A (with a space) throughout the manuscript. Consistent use of PCFA-VA would improve readability.
  2. In the paragraph following Eq. (6), the notation qvar(·) is introduced as the variational density, but the subscript notation in bπ_jk and bλ_jk uses a hat that is not defined. Clarifying that the hat denotes posterior quantities would help.
  3. Table 1: the K_fit=6 row shows RMSE=0.073, which is higher than K_fit=5 (0.052) but lower than K_fit=4 (0.109). The text attributes this to over-factoring absorbing nuisance, but the non-monotonicity (K_fit=7 has RMSE=0.077, K_fit=8 has 0.085, then K_fit=9 has 0.082) could use a brief explanation.
  4. Figure 5A: the y-axis label 'ELBO marginal gains g(K)' could be more informative if the units (nats) were indicated, as the text references '≈700 nats' but the figure axis is unlabeled in this respect.
  5. The default δ=10 is described as sitting at the 'under-selection-averse end of the stable region,' but Table A2 shows that for backbone a, δ=25 moves the selection from 22 to 20. A brief discussion of why δ=10 rather than, say, δ=15 (which also selects 22 for both backbones) is the default would be helpful.
  6. The paper cites Jin & Chen (2025) for the PCFA-VA method, which shares authors with the present paper. This is standard and appropriate, but a brief note in the introduction explicitly stating that the present paper extends the authors' prior work would be transparent.
  7. Appendix Table B1: the ELBO values are negative (e.g., -313,321), which is expected for a log-likelihood bound, but a note clarifying that these are negative and that larger (less negative) is better would prevent momentary confusion.

Circularity Check

0 steps flagged

No significant circularity: the gain rule and fit indices are validated against known truth in simulations, not against the authors' own prior results.

full rationale

The paper's central contributions—the post-selection fit indices (RMSEA, CFI, TLI, AIC, BIC, ELBO), the hard/soft selection framework, and the scale-free gain rule with sustained-drop guard—are logically independent of the estimation method's correctness. The gain rule (Eq. 37) is a general elbow-detection heuristic applied to any criterion path; its performance is validated against known true dimensionality in two simulation studies (99.4% and 94.8% accuracy), not against the authors' own prior results. The fit indices are derived from standard SEM formulas (Eqs. 23-36) applied to the converged variational solution, with parameter counts defined independently (Eqs. 12-22). The self-citation to Jin & Chen (2025) for the PCFA-VA estimation machinery is not load-bearing for the present paper's claims: the post-selection assessment framework would work with any variable-selection method that produces inclusion probabilities. The ELBO (Eq. 31-32) is the algorithm-native objective whose properties follow from variational inference theory generally, not from the authors' prior work. Proposition 1 (Appendix A) provides an independent correctness condition for the gain rule based on gain separation, with a self-contained proof. No step in the derivation chain reduces to its own inputs by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The paper introduces no new entities, particles, or forces. All constructs (spike-and-slab priors, ELBO, fit indices, gain rule) are established concepts recombined for the PEFA context. The free parameters are methodological choices (thresholds, hyperparameters) rather than physical constants fitted to data.

free parameters (5)
  • tau (threshold) = 0.50
    Hard-selection threshold for inclusion probabilities; default value chosen to parallel prior work, not fitted to data.
  • delta (gain threshold) = 10
    Percentage of largest gain used as cutoff in the gain rule; default value chosen empirically from simulation sensitivity analysis.
  • s (sustained-drop guard) = 2
    Number of consecutive sub-threshold gains required to confirm elbow; chosen to balance over-factoring headroom against noise protection.
  • v0, v1 (spike/slab variances) = not specified in text
    SSVS prior hyperparameters inherited from PCFA-VA; control the spike-and-slab separation but values not reported in this paper.
  • rho (Bernoulli inclusion prob) = 0.5
    Prior inclusion probability for SSVS; fixed at 0.5 when no prior loading activeness information is available.
axioms (4)
  • domain assumption Variational gap comparability across candidate K
    ELBO differences treated as evidence differences assume KL{q_var||p} is comparable across models with different K (Eq. 31-32). Stated as untestable directly.
  • domain assumption Local identification of selected loading pattern
    Nominal degrees of freedom formula (Eq. 14) assumes the selected pattern is locally identified; rank check (Eq. 16-17) provides fallback but is noted as potentially slow for large J.
  • domain assumption Pseudo chi-square interpretation of T_H
    The statistic T_H (Eq. 24) is evaluated at regularized variational estimates after selection; treated as descriptive rather than exact ML likelihood-ratio. Conventional cutoffs treated as heuristics.
  • domain assumption Gain separation at true K
    Proposition 1 requires that genuine factors produce gains exceeding the threshold while nuisance factors do not. This is an empirical condition verified in simulations but not guaranteed a priori.

pith-pipeline@v1.1.0-glm · 26534 in / 2582 out tokens · 407241 ms · 2026-07-09T18:35:14.519150+00:00 · methodology

0 comments
read the original abstract

In partially exploratory factor analysis (PEFA), the loading structure and factor numbers are weakly specified. The regularized variational approximation for partially confirmatory factor analysis (PCFA VA) recovers this structure via Bayesian variable selection, using spike and slab priors to assign inclusion probabilities to unspecified loadings. This research introduces a post selection assessment framework for this approach. We convert converged solutions into covariance models using either hard selection (thresholding probabilities into a sparse pattern) or soft selection (retaining them as weights for effective parameter counts). We derive the resulting degrees of freedom, absolute fit diagnostics (RMSEA, SRMR, CFI, TLI), and relative criteria (AIC, BIC, ELBO). To determine factor numbers, we propose a scale free gain rule with a sustained drop guard. Simulations show absolute indices successfully track loading recovery and flag under factoring. While raw criteria over factor, our gain rule accurately recovers true dimensionality, with the ELBO variant proving most robust. Finally, a 100 item PID 5 example demonstrates that our model fits better than a confirmatory 25 facet model and concordantly recovers major structures across disjoint specifications.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

238 extracted references · 238 canonical work pages · 11 internal anchors

  1. [1]

    and Derringer, Jaime and Markon, Kristian E

    Krueger, Robert F. and Derringer, Jaime and Markon, Kristian E. and Watson, David and Skodol, Andrew E. , journal=. Initial construction of a maladaptive personality trait model and inventory for. 2012 , publisher=

    work page 2012

  2. [2]

    PsyArXiv preprint , year=

    The SAPA Personality Inventory: An empirically-derived, hierarchically-organized self-report personality assessment model , author=. PsyArXiv preprint , year=. doi:10.31234/osf.io/sc4p9 , publisher=

    work page doi:10.31234/osf.io/sc4p9

  3. [3]

    2024 , note=

    psychTools: Tools to Accompany the 'psych' Package for Psychological Research , author=. 2024 , note=

    work page 2024

  4. [4]

    A generalized partially confirmatory factor analysis framework with mixed

    Chen, Jinsong , journal=. A generalized partially confirmatory factor analysis framework with mixed. 2022 , publisher=

    work page 2022

  5. [5]

    A partially confirmatory approach to scale development with the

    Chen, Jinsong and Guo, Zhihan and Zhang, Lijin and Pan, Junhao , journal=. A partially confirmatory approach to scale development with the. 2021 , publisher=

    work page 2021

  6. [6]

    Journal of machine Learning research , volume=

    Latent dirichlet allocation , author=. Journal of machine Learning research , volume=

    work page

  7. [7]

    BERTopic: Neural topic modeling with a class-based TF-IDF procedure

    BERTopic: Neural topic modeling with a class-based TF-IDF procedure , author=. arXiv preprint arXiv:2203.05794 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  8. [8]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Bert: Pre-training of deep bidirectional transformers for language understanding , author=. arXiv preprint arXiv:1810.04805 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    Advances in neural information processing systems , volume=

    Language models are few-shot learners , author=. Advances in neural information processing systems , volume=

    work page

  10. [10]

    Multimedia tools and applications , volume=

    Latent Dirichlet allocation (LDA) and topic modeling: models, applications, a survey , author=. Multimedia tools and applications , volume=. 2019 , publisher=

    work page 2019

  11. [11]

    Information Systems , volume=

    A review of topic modeling methods , author=. Information Systems , volume=. 2020 , publisher=

    work page 2020

  12. [12]

    ACM computing surveys (csur) , volume=

    Generalizing from a few examples: A survey on few-shot learning , author=. ACM computing surveys (csur) , volume=. 2020 , publisher=

    work page 2020

  13. [13]

    , author=

    How to determine the number of factors to retain in exploratory factor analysis: A comparison of extraction methods under realistic conditions. , author=. Psychological Methods , volume=. 2019 , publisher=

    work page 2019

  14. [14]

    An empirical

    Braeken, Johan and Van Assen, Marcel ALM , journal=. An empirical. 2017 , publisher=

    work page 2017

  15. [15]

    , author=

    Evaluating the use of exploratory factor analysis in psychological research. , author=. Psychological Methods , volume=. 1999 , publisher=

    work page 1999

  16. [16]

    2023 , publisher=

    Principles and practice of structural equation modeling , author=. 2023 , publisher=

    work page 2023

  17. [17]

    Transactions of the Association for Computational Linguistics , volume=

    A primer in BERTology: What we know about how BERT works , author=. Transactions of the Association for Computational Linguistics , volume=. 2021 , publisher=

    work page 2021

  18. [18]

    Springer google schola , volume=

    Pattern recognition and machine learning , author=. Springer google schola , volume=

    work page

  19. [19]

    Fundamentals of artificial intelligence , pages=

    Natural language processing , author=. Fundamentals of artificial intelligence , pages=. 2020 , publisher=

    work page 2020

  20. [20]

    2009 , publisher=

    Natural language processing with Python: analyzing text with the natural language toolkit , author=. 2009 , publisher=

    work page 2009

  21. [21]

    Upper Saddle River, NJ: Prentice Hall , year=

    Speech and Language Processing: An introduction to speech recognition, computational linguistics and natural language processing , author=. Upper Saddle River, NJ: Prentice Hall , year=

    work page

  22. [22]

    2021 , publisher=

    Transformers for Natural Language Processing: Build innovative deep neural network architectures for NLP with Python, PyTorch, TensorFlow, BERT, RoBERTa, and more , author=. 2021 , publisher=

    work page 2021

  23. [23]

    2019 , publisher=

    Text analytics with Python: a practitioner's guide to natural language processing , author=. 2019 , publisher=

    work page 2019

  24. [24]

    LLaMA: Open and Efficient Foundation Language Models

    Llama: Open and efficient foundation language models , author=. arXiv preprint arXiv:2302.13971 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  25. [25]

    RoBERTa: A Robustly Optimized BERT Pretraining Approach

    Roberta: A robustly optimized bert pretraining approach , author=. arXiv preprint arXiv:1907.11692 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1907

  26. [26]

    Advances in neural information processing systems , volume=

    Attention is all you need , author=. Advances in neural information processing systems , volume=

    work page

  27. [27]

    2014 , publisher=

    Factor analysis: Classic edition , author=. 2014 , publisher=

    work page 2014

  28. [28]

    Multivariate Behavioral Research , volume=

    An overview of analytic rotation in exploratory factor analysis , author=. Multivariate Behavioral Research , volume=. 2001 , publisher=

    work page 2001

  29. [29]

    Proceedings of the 27th international conference on computational linguistics , pages=

    Contextual string embeddings for sequence labeling , author=. Proceedings of the 27th international conference on computational linguistics , pages=

    work page

  30. [30]

    BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension

    Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension , author=. arXiv preprint arXiv:1910.13461 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  31. [31]

    Utilizing BERT for Aspect-Based Sentiment Analysis via Constructing Auxiliary Sentence

    Utilizing BERT for aspect-based sentiment analysis via constructing auxiliary sentence , author=. arXiv preprint arXiv:1903.09588 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  32. [32]

    Rethinking complex neural network architectures for document classification , author=. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pages=

    work page 2019

  33. [33]

    Advances in Information Retrieval: 42nd European Conference on IR Research, ECIR 2020, Lisbon, Portugal, April 14--17, 2020, Proceedings, Part II 42 , pages=

    Keyphrase extraction as sequence labeling using contextualized embeddings , author=. Advances in Information Retrieval: 42nd European Conference on IR Research, ECIR 2020, Lisbon, Portugal, April 14--17, 2020, Proceedings, Part II 42 , pages=. 2020 , organization=

    work page 2020

  34. [34]

    2008 , publisher=

    Introduction to information retrieval , author=. 2008 , publisher=

    work page 2008

  35. [35]

    2007 , publisher=

    The text mining handbook: advanced approaches in analyzing unstructured data , author=. 2007 , publisher=

    work page 2007

  36. [36]

    NLTK: The Natural Language Toolkit

    Nltk: The natural language toolkit , author=. arXiv preprint cs/0205028 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  37. [37]

    Robustness in statistics , pages=

    Robustness in the strategy of scientific model building , author=. Robustness in statistics , pages=. 1979 , publisher=

    work page 1979

  38. [38]

    Psychometrika , volume=

    Exploratory bi-factor analysis , author=. Psychometrika , volume=. 2011 , publisher=

    work page 2011

  39. [39]

    Psychometrika , volume=

    The development of hierarchical factor solutions , author=. Psychometrika , volume=. 1957 , publisher=

    work page 1957

  40. [40]

    Multivariate Behavioral Research , volume=

    The rediscovery of bifactor measurement models , author=. Multivariate Behavioral Research , volume=. 2012 , publisher=

    work page 2012

  41. [41]

    International Conference on Learning Representations , year=

    Long Range Arena : A Benchmark for Efficient Transformers , author=. International Conference on Learning Representations , year=

    work page

  42. [42]

    Approaches for

    George, Edward I and McCulloch, Robert E , journal=. Approaches for. 1997 , publisher=

    work page 1997

  43. [43]

    Guoqiang, Lin , journal=. A

    work page

  44. [44]

    The annals of mathematical statistics , volume=

    On information and sufficiency , author=. The annals of mathematical statistics , volume=. 1951 , publisher=

    work page 1951

  45. [45]

    Variational inference:

    Blei, David M and Kucukelbir, Alp and McAuliffe, Jon D , journal=. Variational inference:. 2017 , publisher=

    work page 2017

  46. [46]

    Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Automatic text scoring using neural networks , author=. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=. 2016 , organization=

    work page 2016

  47. [47]

    International Journal of Advanced Computer Science and Applications , volume=

    An empirical analysis of BERT embedding for automated essay scoring , author=. International Journal of Advanced Computer Science and Applications , volume=

    work page

  48. [48]

    PNAS Nexus , year=

    Meaning modulations and stability in large language models: An analysis of BERT embeddings for psycholinguistic research , author=. PNAS Nexus , year=

    work page

  49. [49]

    Science , volume=

    Semantics derived automatically from language corpora contain human-like biases , author=. Science , volume=

    work page

  50. [50]

    South , author=

    Leveraging the alignment between machine learning and intersectionality: Using word embeddings to measure intersectional experiences of the nineteenth century U.S. South , author=. Poetics , volume=

    work page

  51. [51]

    and Shaukat, A

    Arshad, F. and Shaukat, A. and Latif, S. , title =. Proceedings of the 2024 International Conference on Frontiers of Information Technology , year =

    work page 2024

  52. [52]

    and Suhartono, D

    Christian, H. and Suhartono, D. and Chowanda, A. and Zamli, K. Z. , title =. Journal of Big Data , volume =

    work page

  53. [53]

    and Du, J

    Luo, Y. and Du, J. S. and Qiu, L. , title =. Proceedings of the IEEE International Conference on Behavioral and Social Computing (BESC) , year =

    work page

  54. [54]

    and Yang, L

    Zhou, X. and Yang, L. and Fan, X. and Ren, G. and Yang, Y. and Lin, H. , title =. Lecture Notes in Computer Science , year =

    work page

  55. [55]

    and Long, Y

    Huang, Z. and Long, Y. and Peng, K. and Tong, S. , title =. Journal of Intelligence , volume =

    work page

  56. [56]

    Wulff, D. U. and Mata, R. , title =. Nature Human Behaviour , year =

    work page

  57. [57]

    Political Analysis , volume =

    Rheault, Luke and Cochrane, Christopher , title =. Political Analysis , volume =

    work page

  58. [58]

    Proceedings of the National Academy of Sciences , volume =

    Garg, Nikhil and Schiebinger, Londa and Jurafsky, Dan and Zou, James , title =. Proceedings of the National Academy of Sciences , volume =

    work page

  59. [59]

    Applied and Computational Engineering , volume =

    Shu, Xiaoxue , title =. Applied and Computational Engineering , volume =

    work page

  60. [60]

    Mathematics , volume =

    Faseeh, Muneeb and Jaleel, Abdul and Iqbal, Nadeem and Ghani, Abdullah and Abdusalomov, Abduvohid and Mehmood, Asad and Cho, Young , title =. Mathematics , volume =

    work page

  61. [61]

    Rytting, C. M. and Sorensen, T. and Argyle, L. and Busby, E. and Fulda, N. and Gubler, J. and Wingate, D. , title =. arXiv preprint , year =

    work page

  62. [62]

    arXiv preprint , year =

    Li, Lin and Li, Jincheng and Chen, Cheng and Gui, Feng and Yang, Hongyi and Yu, Chenyu and Wang, Zefan and others , title =. arXiv preprint , year =

    work page

  63. [63]

    arXiv preprint , year =

    Fitz, Samuel , title =. arXiv preprint , year =

    work page

  64. [64]

    arXiv preprint , year =

    Petukhova, Anna and Matos-Carvalho, Joao Pedro and Fachada, Nuno , title =. arXiv preprint , year =

    work page

  65. [65]

    arXiv preprint , year =

    Nie, Zihan and Feng, Zhen and Li, Ming and Zhang, Cheng and Zhang, Yichen and Long, Donghong and Zhang, Rui , title =. arXiv preprint , year =

    work page

  66. [66]

    arXiv preprint , year =

    Tao, Chao and Shen, Tao and Gao, Shangsong and Zhang, Jun and Li, Zhiyuan and Tao, Zhanxing and Ma, Shuming , title =. arXiv preprint , year =

    work page

  67. [67]

    Structural Equation Modeling: A Multidisciplinary Journal , volume=

    Regularized Variational Approximation for Partially Confirmatory Factor Analysis , author=. Structural Equation Modeling: A Multidisciplinary Journal , volume=. 2025 , publisher=

    work page 2025

  68. [68]

    2024 , publisher=

    Exploring Psychometric Analysis of Textual Data with Large Language Models: Chances and Challenges , author=. 2024 , publisher=

    work page 2024

  69. [69]

    Dynamic Documents with

    Yihui Xie , publisher =. Dynamic Documents with. 2015 , edition =

    work page 2015

  70. [70]

    1977 , author =

    Exploratory Data Analysis , publisher =. 1977 , author =

    work page 1977

  71. [71]

    1987 , author =

    Empirical Model-Building and Response Surfaces , publisher =. 1987 , author =

    work page 1987

  72. [72]

    1907 , journal=

    Vox populi , author=. 1907 , journal=

    work page 1907

  73. [73]

    Investigating variation in replicability:

    Klein, Richard A and Ratliff, Kate A and Vianello, Michelangelo and Adams Jr, Reginald B and Bahn. Investigating variation in replicability:. Social psychology , year=

    work page

  74. [74]

    2011 , publisher=

    Data analysis: A model comparison approach , author=. 2011 , publisher=

    work page 2011

  75. [75]

    2014 , publisher=

    Doing Bayesian data analysis: A tutorial with R, JAGS, and Stan , author=. 2014 , publisher=

    work page 2014

  76. [76]

    Wellcome Open Research , VOLUME =

    Allen, M and Poggiali, D and Whitaker, K and Marshall, T R and Kievit, R A , TITLE =. Wellcome Open Research , VOLUME =. 2019 , NUMBER =

    work page 2019

  77. [77]

    Biometrika , pages=

    The probable error of a mean , author=. Biometrika , pages=. 1908 , publisher=

    work page 1908

  78. [78]

    the probable error of a mean

    On student's 1908 article "the probable error of a mean"" , author=. Journal of the American Statistical Association , volume=. 2008 , publisher=

    work page 1908

  79. [79]

    Synthetic horizontal branch morphology for different metallicities and ages under tidally enhanced stellar wind

    Sample Quantiles in Statistical Packages , volume =. The American Statistician , author =. 1996 , pages =. doi:10.1080/00031305.1996.10473566 , language =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1080/00031305.1996.10473566 1996

  80. [80]

    Psychonomic bulletin & review , volume=

    The fallacy of placing confidence in confidence intervals , author=. Psychonomic bulletin & review , volume=. 2016 , publisher=

    work page 2016

Showing first 80 references.