pith. machine review for the scientific record. sign in

arxiv: 2605.04457 · v2 · submitted 2026-05-06 · 📊 stat.ME · stat.AP· stat.CO

Recognition: unknown

Penalized KLIC Model Selection for the Generalized Method of Moments in Longitudinal Data with Time-Dependent Covariates

Mahmud Hasan , Mathias Nthiani Muia , Mous-Abou Hamadou , Niloofar Ramezani
Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:38 UTC · model grok-4.3

classification 📊 stat.ME stat.APstat.CO
keywords KLICmodel selectiongeneralized method of momentslongitudinal datapenalized criteriatime-dependent covariatesGMMover-parameterized models
0
0 comments X

The pith

Penalized KLIC criteria add terms for parameters and moment conditions to curb over-selection of complex models in GMM longitudinal analysis.

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

The paper proposes two penalized versions of the Kullback-Leibler Information Criterion to improve model selection when estimating generalized method of moments models on longitudinal data with time-dependent covariates. Original KLIC tends to favor models with more parameters or more valid moment conditions as those numbers grow. The new MPPP-KLIC multiplies a penalty by the product of parameter count and moment count, while LP-KLIC adds a logarithmic penalty on those quantities. Simulations with binary and continuous responses show the penalties increase correct model distinction and decrease selection of over-parameterized alternatives. In the Filipino Child Morbidity data the criteria produce stable rankings and identify age as the leading predictor of morbidity.

Core claim

The Moment-Parameter Product Penalty KLIC and Logarithmic Penalty KLIC supply a theoretically motivated balance between model fit and complexity by incorporating explicit penalties for both the number of parameters and the number of valid moment conditions, thereby reducing the original KLIC's tendency to select overly complex GMM models in longitudinal settings with time-dependent covariates.

What carries the argument

The two penalized KLIC variants (MPPP-KLIC, which multiplies a penalty by the product of parameter count and valid moment count, and LP-KLIC, which adds a logarithmic penalty on those counts) that are added to the standard KLIC score to penalize complexity.

If this is right

  • Improved ability to distinguish among competing GMM models in both binary and continuous response longitudinal data.
  • Lower rates of selecting over-parameterized models when the number of valid moment conditions grows.
  • Stable and interpretable model rankings in applied longitudinal studies such as the Filipino Child Morbidity dataset.
  • Consistent identification of key predictors like age in child health outcomes.

Where Pith is reading between the lines

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

  • The same penalty structure could be tested in other GMM applications beyond longitudinal health data, such as econometrics or survival analysis.
  • Alternative penalty forms that scale with sample size might further improve performance in very high-dimensional moment settings.
  • The criteria might be combined with existing moment-selection procedures to create a two-stage model and moment selection workflow.

Load-bearing premise

The added penalties will consistently prevent selection of overly complex models across binary and continuous longitudinal settings without introducing new biases or overlooking important predictors.

What would settle it

A simulation study in which the penalized criteria select a known over-parameterized model more often than the unpenalized KLIC, or select an incorrect model while the original KLIC selects the true one.

Figures

Figures reproduced from arXiv: 2605.04457 by Mahmud Hasan, Mathias Nthiani Muia, Mous-Abou Hamadou, Niloofar Ramezani.

Figure 1
Figure 1. Figure 1: Comparison of MPPP–KLIC and LP–KLIC across candidate models under binary view at source ↗
read the original abstract

Model selection plays an important role in longitudinal data analysis, especially when models are estimated using the generalized method of moments (GMM) in the presence of time-dependent covariates. In this setting, the number of valid moment conditions can grow quickly and may lead to over-parameterized models. The Kullback--Leibler Information Criterion (KLIC) has been proposed as a model-selection tool for this framework; however, the original KLIC criterion may favor overly complex models when the number of parameters or valid moment conditions increases. To address this limitation, this study proposes two penalized versions of KLIC that incorporate penalties based on both the number of model parameters and the number of valid moment conditions. The proposed criteria are referred to as the Moment--Parameter Product Penalty KLIC (MPPP--KLIC) and the Logarithmic Penalty KLIC (LP--KLIC). These criteria provide a theoretically motivated mechanism for balancing model fit and model complexity in GMM-based longitudinal models. Through an extensive simulation study involving both binary and continuous response settings, the proposed criteria are shown to improve the ability of KLIC to distinguish among competing models and to reduce the selection of over-parameterized models. The performance of the proposed methods is further illustrated using the Filipino Child Morbidity dataset, a longitudinal study of child health in the Philippines. The results show that the proposed penalized criteria provide stable and interpretable model rankings and consistently identify age as the most important predictor of child morbidity. Overall, the proposed penalized KLIC criteria offer practical and theoretically grounded tools for model selection in GMM-based longitudinal data analysis with time-dependent covariates.

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.

Referee Report

2 major / 2 minor

Summary. The paper proposes two penalized extensions of the KLIC criterion—M PPP-KLIC (using a product penalty on the number of parameters and valid moment conditions) and LP-KLIC (using a logarithmic penalty)—for model selection in GMM estimation of longitudinal data models with time-dependent covariates. The goal is to correct the tendency of unpenalized KLIC to favor over-parameterized models as the number of valid moments grows. Support is provided via simulation experiments in binary and continuous response settings and an application to the Filipino Child Morbidity dataset, where the penalized criteria are reported to yield more stable model rankings and consistently identify age as the leading predictor.

Significance. If the penalties are robust, the work supplies a practical tool for GMM-based longitudinal model selection where moment conditions proliferate. The simulation-plus-real-data design offers applied value, but the absence of consistency proofs or exhaustive coverage of growing-moment regimes limits the theoretical advance and generalizability.

major comments (2)
  1. [Simulation Study] The simulation study does not systematically vary the ratio of valid moment conditions to sample size T or include misspecified working-correlation structures, both of which are explicitly flagged in the introduction as the primary settings where standard KLIC over-selects complex models; without these regimes the reported gains in model distinction and reduced over-parameterization cannot be taken as general.
  2. [Proposed Penalized Criteria] No analytic consistency or asymptotic expansion result is supplied for either MPPP-KLIC or LP-KLIC; the penalties are presented as theoretically motivated yet the manuscript provides only heuristic justification and finite-sample simulation evidence, leaving open whether the criteria remain consistent when the number of moments grows with T.
minor comments (2)
  1. [Abstract] The abstract and introduction refer to an 'extensive simulation study' but supply no table or text listing the grid of T, n, number of moments, or strength of time dependence; this omission makes it difficult to judge coverage of the motivating regime.
  2. [Methods] Notation for the penalty terms (e.g., the exact functional form multiplying p and the number of moments in MPPP-KLIC) should be stated explicitly in an equation rather than described in prose only.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments, which help clarify the scope and limitations of our work. We respond to each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: The simulation study does not systematically vary the ratio of valid moment conditions to sample size T or include misspecified working-correlation structures, both of which are explicitly flagged in the introduction as the primary settings where standard KLIC over-selects complex models; without these regimes the reported gains in model distinction and reduced over-parameterization cannot be taken as general.

    Authors: We agree that the current simulation design does not fully cover the regimes highlighted in the introduction. In the revised manuscript we will expand the simulation study to include (i) systematic variation of the ratio of valid moments to sample size T and (ii) misspecified working-correlation structures. New tables and figures will be added to demonstrate the behavior of MPPP-KLIC and LP-KLIC under these conditions. revision: yes

  2. Referee: No analytic consistency or asymptotic expansion result is supplied for either MPPP-KLIC or LP-KLIC; the penalties are presented as theoretically motivated yet the manuscript provides only heuristic justification and finite-sample simulation evidence, leaving open whether the criteria remain consistent when the number of moments grows with T.

    Authors: The manuscript supplies only heuristic motivation and finite-sample evidence; no consistency proof or asymptotic expansion is derived. We will revise the discussion section to explicitly acknowledge this gap, clarify that the criteria are proposed as practical tools supported by simulation, and identify the development of asymptotic theory under growing moments as an important direction for future research. revision: partial

Circularity Check

0 steps flagged

No circularity: proposed penalties are new definitions validated externally via simulation

full rationale

The paper identifies a limitation of the original KLIC (favoring complex models as parameters or moments grow) and introduces two new penalized criteria, MPPP-KLIC and LP-KLIC, whose functional forms are explicitly defined in terms of the number of parameters p and valid moments m. These definitions are presented as motivated additions rather than derived from or fitted to the target selection performance. The simulation study and real-data application serve as external validation of the new criteria's behavior, not as inputs that the criteria are constructed to reproduce. No self-citation chain, self-definitional loop, or renaming of known results is indicated in the provided text; the derivation chain therefore remains self-contained and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

With only the abstract available, no specific free parameters, axioms, or invented entities can be identified from the text. The new penalty terms likely involve choices in functional form that are not detailed here.

pith-pipeline@v0.9.0 · 5614 in / 1272 out tokens · 83033 ms · 2026-05-08T17:38:07.885089+00:00 · methodology

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

42 extracted references

  1. [1]

    Akaike, H. (1973). Information theory and an extension of the maximum likelihood principle. InSecond International Symposium on Information Theory, pages 267–281

    1973

  2. [2]

    Akaike, H. (1974). A new look at the statistical model identification.IEEE Transactions on Automatic Control, 19(6):716–723

    1974

  3. [3]

    Shane, M. N. (2019).Model Selection for Longitudinal Data With Time-Dependent Co- variates Using Generalized Method of Moments. PhD Dissertation, University of North- ern Colorado. Available at:https://digscholarship.unco.edu/dissertations/643

    2019

  4. [4]

    and Zeileis, A

    Kleiber, C. and Zeileis, A. (2008).Applied Econometrics with R. Springer

    2008

  5. [5]

    Qu, A., Lindsay, B. G. and Li, B. (2000). Improving generalized estimating equations using quadratic inference functions.Biometrika, 87, 823–836

    2000

  6. [6]

    Lindsay, B. G. and Qu, A. (2003). Inference functions and quadratic score tests.Statis- tical Science, 18, 394–410

    2003

  7. [7]

    and Kalbfleisch, J

    Neuhaus, J. and Kalbfleisch, J. (1998). Between- and within-cluster covariate effects in the analysis of clustered data.Biometrics, 54(2), 638–645

    1998

  8. [8]

    Ferguson, T. S. (1958). A method of generating best asymptotically normal estimates with application to the estimation of bacterial densities.The Annals of Mathematical Statistics, 29, 1046–1062

    1958

  9. [9]

    Hansen, L. P. (2007). Generalized method of moments estimation. InThe New Palgrave Dictionary of Economics. Palgrave Macmillan

    2007

  10. [10]

    Lai, T. L. and Small, D. S. (2007). Marginal regression analysis of longitudinal data with time-dependent covariates: A generalized method-of-moments approach.Journal of the Royal Statistical Society: Series B (Statistical Methodology), 69(1):79–99

    2007

  11. [11]

    L., Wilson, J

    Lalonde, T. L., Wilson, J. R., & Yin, J. (2014). GMM logistic regression models for longitudinal data with time-dependent covariates and extended classifications.Statistics in Medicine, 33(27), 4756–4769

    2014

  12. [12]

    (ed.) (1999),Generalized Method of Moments Estimation, New York: Cam- bridge University Press

    Mátyás, L. (ed.) (1999),Generalized Method of Moments Estimation, New York: Cam- bridge University Press

    1999

  13. [13]

    Neyman, J. (1949). Contribution to the theory of theχ2 test.Proceedings of the Berkeley Symposium on Mathematical Statistics and Probability, pp. 239–273. Berkeley: Univer- sity of California Press

    1949

  14. [14]

    Pan, W. (2001). Akaike’s information criterion in generalized estimating equations. Biometrics, 57(1):120–125

    2001

  15. [15]

    Nielsen, H. B. (2005). Generalized method of moments estimation. Econometrics 2 Lec- ture, Department of Economics, University of Copenhagen: Copenhagen, Denmark. 29

    2005

  16. [16]

    Zivot, E. (2015). Generalized method of moments. Econometrics 583 Lecture at the University of Washington: Seattle, WA

    2015

  17. [17]

    G., and Segal, L

    Altonji, J. G., and Segal, L. M. (1996). Small-sample bias in GMM estimation of co- variance structures.Journal of Business & Economic Statistics,14(3), 353–366

    1996

  18. [18]

    (1975).I-divergence geometry of probability distributions and minimization problems

    Csiszár, I. (1975).I-divergence geometry of probability distributions and minimization problems. The Annals of Probability, 3(1), 146–158

    1975

  19. [19]

    (1982).Maximum likelihood estimation of misspecified models

    White, H. (1982).Maximum likelihood estimation of misspecified models. Econometrica, 50(1), 1–25

    1982

  20. [20]

    Modeling the health of Filipino children.Journal of the Royal Statistical Society: Series A (Statistics in Society), 157(3), 417–432, (1994)

    Bhargava, A. Modeling the health of Filipino children.Journal of the Royal Statistical Society: Series A (Statistics in Society), 157(3), 417–432, (1994)

    1994

  21. [21]

    E., and Haddad, L

    Bouis, H. E., and Haddad, L. J. Effects of agricultural commercialization on land tenure, household resource allocation, and nutrition in the Philippines.Research Report 79, International Food Policy Research Institute, Washington, DC, (1990)

    1990

  22. [22]

    Wiley, New York (1990)

    Agresti, A.Categorical Data Analysis. Wiley, New York (1990)

    1990

  23. [23]

    Information theory and an extension of the maximum likelihood principle

    Akaike, H. Information theory and an extension of the maximum likelihood principle. InSecond International Symposium on Information Theory, 267–281 (1973)

    1973

  24. [24]

    A new look at the statistical model identification.IEEE Transactions on Automatic Control, 19, 716–723 (1974)

    Akaike, H. A new look at the statistical model identification.IEEE Transactions on Automatic Control, 19, 716–723 (1974)

    1974

  25. [25]

    Y., Zeger, S.Analysis of Longitudinal Data

    Diggle, P., Heagerty, P., Liang, K. Y., Zeger, S.Analysis of Longitudinal Data. Oxford University Press (2002)

    2002

  26. [26]

    A caveat concerning independence estimating equations with multivari- ate binary data.Biometrics, 51, 309–317 (1995)

    Fitzmaurice, G. A caveat concerning independence estimating equations with multivari- ate binary data.Biometrics, 51, 309–317 (1995)

    1995

  27. [27]

    Wiley, New York (2011)

    Fitzmaurice, G., Laird, N., Ware, J.Applied Longitudinal Analysis. Wiley, New York (2011)

    2011

  28. [28]

    Hansen, L. P. Large sample properties of generalized method of moments estimators. Econometrica, 50, 1029–1054 (1982)

    1982

  29. [29]

    D.Longitudinal Data Analysis

    Hedeker, D., Gibbons, R. D.Longitudinal Data Analysis. Wiley, New York (2006)

    2006

  30. [30]

    Chapman & Hall/CRC (2009)

    Hilbe, J.Logistic Regression Models. Chapman & Hall/CRC (2009)

    2009

  31. [31]

    M., Tsai, C

    Hurvich, C. M., Tsai, C. L. Regression and time series model selection in small samples. Biometrika, 76, 297–307 (1989)

    1989

  32. [32]

    An information-theoretic alternative to generalized method of moments estimation.Econometrica, 65, 861–874 (1997)

    Kitamura, Y., Stutzer, M. An information-theoretic alternative to generalized method of moments estimation.Econometrica, 65, 861–874 (1997)

    1997

  33. [33]

    Kullback, S., Leibler, R. A. On information and sufficiency.Annals of Mathematical Statistics, 22, 79–86 (1951). 30

    1951

  34. [34]

    L., Small, D

    Lai, T. L., Small, D. Generalized method of moments for longitudinal data with time- dependent covariates.Biometrika, 94, 501–515 (2007)

    2007

  35. [35]

    Y., Zeger, S

    Liang, K. Y., Zeger, S. L. Longitudinal data analysis using generalized linear models. Biometrika, 73, 13–22 (1986)

    1986

  36. [36]

    Akaike’s information criterion in generalized estimating equations.Biometrics, 57, 120–125 (2001)

    Pan, W. Akaike’s information criterion in generalized estimating equations.Biometrics, 57, 120–125 (2001)

    2001

  37. [37]

    S., Anderson, G

    Pepe, M. S., Anderson, G. L. A cautionary note on inference for marginal regression models with longitudinal data and general correlated response data.Communications in Statistics, 23, 939–951 (1994)

    1994

  38. [38]

    Estimating the dimension of a model.Annals of Statistics, 6, 461–464 (1978)

    Schwarz, G. Estimating the dimension of a model.Annals of Statistics, 6, 461–464 (1978)

    1978

  39. [39]

    Further analysis of the data by Akaike’s information criterion and the finite corrections.Communications in Statistics, 7, 13–26 (1978)

    Sugiura, N. Further analysis of the data by Akaike’s information criterion and the finite corrections.Communications in Statistics, 7, 13–26 (1978)

    1978

  40. [40]

    L., Liang, K

    Zeger, S. L., Liang, K. Y. Longitudinal data analysis for discrete and continuous out- comes.Biometrics, 42, 121–130 (1986)

    1986

  41. [41]

    L., Liang, K

    Zeger, S. L., Liang, K. Y., Albert, P. Models for longitudinal data: a generalized esti- mating equation approach.Biometrics, 44, 1049–1060 (1988)

    1988

  42. [42]

    L., Liang, K

    Zeger, S. L., Liang, K. Y. An overview of methods for the analysis of longitudinal data. Statistics in Medicine, 11, 1825–1839 (1992). 31

    1992