pith. sign in

arxiv: 1906.10221 · v1 · pith:CBJ2KU5Qnew · submitted 2019-06-24 · 📊 stat.ME

Parametric versus Semi and Nonparametric Regression Models

Hamdy F. F. Mahmoud This is my paper

Pith reviewed 2026-05-25 16:52 UTC · model grok-4.3

classification 📊 stat.ME
keywords parametric regressionsemiparametric regressionnonparametric regressionmodel selectionestimation methodsrobust estimationregression modeling
0
0 comments X

The pith

Regression model selection depends on how much prior information is available about the relationship form and error distribution.

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

This review distinguishes parametric, semiparametric, and nonparametric regression models according to the quantity of known information about the functional relationship between response and explanatory variables and about the random error distribution. Parametric models assume a fully specified form and distribution, semiparametric models relax some of those assumptions, and nonparametric models estimate the relationship with minimal prior structure. The paper presents differences among the approaches, common estimation procedures, robust variants, and example applications, along with R code for replication. A reader cares because matching model class to available knowledge reduces the risk of incorrect assumptions in analysis.

Core claim

The type of modeling used is based on how much information are available about the form of the relationship between response variable and explanatory variables, and the random error distribution. The article introduces differences between models, common methods of estimation, robust estimation, and applications.

What carries the argument

Information-based classification of regression models according to the amount of prior knowledge on functional form and error distribution.

If this is right

  • Full prior specification of form and distribution supports parametric models for efficient estimation.
  • Partial prior information supports semiparametric models as an intermediate option.
  • Little prior information requires nonparametric models that derive the relationship from the data.
  • Robust estimation procedures apply across all three model classes to address outliers or heavy tails.

Where Pith is reading between the lines

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

  • In applied work the framework would direct analysts to first inventory their substantive knowledge before examining data volume.
  • Fields with uncertain functional forms would see greater use of semiparametric defaults when prior information is intermediate.
  • The inclusion of R code indicates the distinctions are intended to be immediately usable by practitioners.

Load-bearing premise

The primary determinant of model choice is the amount of prior information about relationship form and error distribution, with no other factors such as sample size or computational cost entering the decision at a structural level.

What would settle it

A demonstration that practitioners routinely select models according to sample size or computational limits rather than the stated amount of prior information on form and distribution.

Figures

Figures reproduced from arXiv: 1906.10221 by Hamdy F. F. Mahmoud.

Figure 1
Figure 1. Figure 1: A scatter plot of age and strontium ratio (a), age versus log of wage (b), income [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A scatterplot of age and log(wage) along with different polynomial regression [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Residual plots for the linear model, y = β0 + β1x. 12.5 13.0 13.5 −2 0 1 Fitted values Residuals Residuals vs Fitted 147 19570 −3 −2 −1 0 1 2 3 −4 0 2 Theoretical Quantiles Standardized residuals Normal Q−Q 147 170195 12.5 13.0 13.5 0.0 1.0 2.0 Fitted values Stand ardize d residuals Scale−Location 14719570 0.00 0.10 0.20 −4 0 2 Leverage Standardized residuals Cook's distance 1 0.5 0.5 Residuals vs Leverage… view at source ↗
Figure 4
Figure 4. Figure 4: Residual plots for the 4th degree polynomial model, y = β0+β1x+β2x 2+β3x 3+β4x 4 . 8 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Scatter plot and kernel smoothing at h smoothing parameter (a), and estimating the unknown function by Kernel smoothing using Ksmooth function at different value of smoothing parameter, h. Bandwidth has an impact on the estimation of the nonparametric function [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Estimate of the unknown function (a), and its derivative with 95% confidence [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Estimate of the unknown function using optimal [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Estimate of the unknown function (a), and its derivative with 95% confidence [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Estimated relationship between eduacation, income and prestige using smoothing [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Estimated relationship between eduacation, income and prestige using smoothing [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 13
Figure 13. Figure 13: 14 [PITH_FULL_IMAGE:figures/full_fig_p014_13.png] view at source ↗
Figure 11
Figure 11. Figure 11: Scatter plots of wage data 6 Robust Nonparametric Estimation Outliers may affect kernel or spline nonparametric estimation. So robust estimation is needed in this case. In kernel smoothing, at a point x, the smoothing is obtained from fitting the pth-degree polynomial model using weighted least squares with kernel weights as described in equation (7). The kernel function K is usually taken to be a symmetr… view at source ↗
Figure 12
Figure 12. Figure 12: Smoothed estimated functions of wage data for the model [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Smoothed estimated single index function of wage data for the model [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Simulated data from f(x) = {1 + e −20(x−0.5)} (−1) along with two added outliers points, (0.8, 0.6) and (0.75, 0.62), and the smoothed function by kernel regression, spline and robust nonparametric regression at the same bandwidth, h = 0.046. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_14.png] view at source ↗
read the original abstract

Three types of regression models researchers need to be familiar with and know the requirements of each: parametric, semiparametric and nonparametric regression models. The type of modeling used is based on how much information are available about the form of the relationship between response variable and explanatory variables, and the random error distribution. In this article, differences between models, common methods of estimation, robust estimation, and applications are introduced. The R code for all the graphs and analyses presented here, in this article, is available in the Appendix.

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

1 major / 3 minor

Summary. The manuscript is a review article introducing the distinctions among parametric, semiparametric, and nonparametric regression models. It states that model choice depends on the amount of prior information available about the functional form relating the response to explanatory variables and about the error distribution; the paper then covers differences between the approaches, common estimation methods, robust estimation, applications, and supplies R code for all presented graphs and analyses.

Significance. The paper restates a conventional taxonomy without new derivations, theorems, or original empirical claims. Its main strength is the explicit provision of reproducible R code for all examples, which supports transparency in a review setting. If the central framing were qualified to reflect additional practical determinants of model choice, the manuscript could serve as a basic pedagogical reference, but in its current form its significance for the statistical literature is limited.

major comments (1)
  1. Abstract: the claim that 'the type of modeling used is based on how much information are available about the form of the relationship ... and the random error distribution' presents this as the primary determinant. This framing is load-bearing for the paper's organization yet omits the structural roles of sample size and computational cost, both of which are standard considerations that can render nonparametric methods impractical even when prior information is limited.
minor comments (3)
  1. Abstract, first sentence: grammatical error ('how much information are available' should read 'is available').
  2. The manuscript would benefit from explicit section headings or a table that systematically contrasts the three model classes on the dimensions of assumed form, assumed error distribution, and typical estimators; the current narrative presentation makes these distinctions harder to extract.
  3. No references are cited for the 'common methods of estimation' or 'robust estimation' sections; adding a short, targeted reference list would improve utility without altering the review character.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and constructive suggestion regarding the abstract. We address the comment below and will make the indicated revision to improve the manuscript's framing as a pedagogical reference.

read point-by-point responses

  1. Referee: Abstract: the claim that 'the type of modeling used is based on how much information are available about the form of the relationship ... and the random error distribution' presents this as the primary determinant. This framing is load-bearing for the paper's organization yet omits the structural roles of sample size and computational cost, both of which are standard considerations that can render nonparametric methods impractical even when prior information is limited.

    Authors: We agree that the abstract presents the amount of available prior information on the functional form and error distribution as the basis for model choice, and that this framing is central to the paper's organization. While the review focuses on this distinction as its primary pedagogical theme, we acknowledge that sample size and computational cost are important practical determinants that can make nonparametric approaches infeasible even with limited prior information. We will revise the abstract to qualify the statement, for instance by adding: 'While the choice is primarily guided by the amount of prior information available, practical considerations such as sample size and computational resources also play a role.' This change will be reflected in the next version without shifting the manuscript's core emphasis. revision: yes

Circularity Check

0 steps flagged

Review article with no derivations or predictions

full rationale

The manuscript is a review article that restates the conventional taxonomy of regression models: parametric models assume fully specified functional form and error distribution, semiparametric relax one or the other, and nonparametric assume neither. This framing is drawn from standard literature and contains no new derivation, theorem, equation, or empirical claim whose validity could reduce to its own inputs by construction. No fitted parameters, predictions, or self-citations are load-bearing for any novel result. The abstract and body confirm the absence of original research content that could exhibit circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

As a review article the work draws on standard statistical knowledge without introducing new free parameters, axioms, or invented entities.

pith-pipeline@v0.9.0 · 5599 in / 972 out tokens · 41507 ms · 2026-05-25T16:52:54.252759+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Locally Near Optimal Piecewise Linear Regression in High Dimensions via Difference of Max-Affine Functions

    stat.ML 2026-05 unverdicted novelty 7.0

    ABGD parametrizes piecewise linear functions as difference of max-affine functions and converges linearly to an epsilon-accurate solution with O(d max(sigma/epsilon,1)^2) samples under sub-Gaussian noise, which is min...

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages · cited by 1 Pith paper

  1. [1]

    Fan, J., and Yao, Q. (2003). Nonlinear time series: nonparametric and parametric methods . Springer: New York

    work page 2003

  2. [2]

    S., Sahu, P

    Dhekale, B. S., Sahu, P. K., Vishwajith, K. P., Mishra, P., and Narsimhaiah, L. (2017). Application of parametric and nonparametric regression models for area, production and productivity trends of tea (Camellia sinensis) in India. Indian Journal of Ecology , 44(2), 192-200

    work page 2017

  3. [3]

    Ichimura, H. (1993). Semiparametric least squares (SLS) and weighted SLS estimation of single-index models. Journal of Econometrics , 58, 71-120. Jialiang Li, Chao Huang, Zhub Hongtu, for the Alzheimers Disease Neuroimaging Initiative. (2017). A functional varying-coefficient single-index model for functional response data. Journal of the American Statistic...

    work page 1993

  4. [4]

    Lin, W., and Kulasekera, K. B. (2007). Identifiability of single index models and additive index models. Biometrika, 94, 496-501

    work page 2007

  5. [5]

    Loader, C. (1999). Bandwidth selection: classical or plug-in?. The Annals of Statistics , 27(2), 415-438

    work page 1999

  6. [6]

    Mahmoud, H. F. F., Kim, I., and Kim, H. (2016). Semiparametric single index multi change points model with an application of environmental health study on mortality and tem- 18 perature. Environmetrics, 27(8), 494-506

    work page 2016

  7. [7]

    Mahmoud, H. F. F., and Kim, I. (2019). Semiparametric spatial mixed effects single index models. Computational Statistics & Data Analysis , 136, 108-112

    work page 2019

  8. [8]

    Nadaraya, E. A. (1964). On estimating regression. Theory of probability and its applications , 9, 141-142. Qin. J., Yu, T, Li, P., Liu, H., and Chen, B. (2018). Using a monotone single index model to stabilize the propensity score in missing data problems and causal inference. Statistics in Medicine. 38(8) 1442-1458

    work page 1964

  9. [9]

    and Parmar, R

    Rajarathinan, A. and Parmar, R. S. (2011). Application pf parametric and nonparametric regression models for area, production and productivity trends of castor corn. Asian Journal of Applied Sciences , 4(1), 42-52

    work page 2011

  10. [10]

    P., and Carrol, R

    Ruppert, D., Wand, M. P., and Carrol, R. J. (2003). Semiparametric regression. New York: Cambridge University Press

    work page 2003

  11. [11]

    Wang, Y. (2011). Smoothing splines: methods and applications . FL: CRC Press, Boca Raton

    work page 2011

  12. [12]

    P., and Jones, M.C

    Wand, M. P., and Jones, M.C. (1995). Kernel smoothing . London; New York: Chapman and Hall

    work page 1995

  13. [13]

    Watson, G. S. (1964). Smooth regression analysis. Sankhya, Series A , 26, 359-372. 19 Appendix library ( SemiPar ) ; library ( np ) ; library ( car ) ; library ( s t a t s ) library ( graphics ) ; data ( f o s s i l ) # Load f o s s i l d a t a attach ( f o s s i l ) f i t = spm( strontium . r a t i o ˜ f ( age ) ) plot ( f i t ) points ( age , strontium ...

    work page 1964