pith. sign in

arxiv: 2604.14460 · v2 · submitted 2026-04-15 · 💻 cs.HC · cs.LG

Bias in Surface Electromyography Features across a Demographically Diverse Cohort

Aditi Agrawal , Celine John Philip , Giancarlo K. Sagastume , Marcus A. Battraw , Wilsaan M. Joiner , Jonathon S. Schofield , Lee M. Miller , Richard S. Whittle This is my paper

Pith reviewed 2026-05-10 11:56 UTC · model grok-4.3

classification 💻 cs.HC cs.LG
keywords surface electromyographysEMGdemographic biasgesture decodingmachine learningneural interfacessignal featuresassistive devices
0
0 comments X p. Extension

The pith

Demographic traits like age and body mass associate with 33% of standard sEMG features.

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

The paper investigates how personal differences influence surface electromyography signals during hand gestures. It extracts 147 common features from 81 demographically varied participants and applies mixed-effects models plus partial least squares analysis to demographic variables including age, sex, height, weight, skin properties, fat, and hair. The analysis identifies significant associations in 49 features, or one third of the set. This matters for machine-learning gesture decoders used in prosthetics and interfaces because unaddressed biases could produce unequal performance across users. The results indicate that feature selection or correction steps may be required for equitable deployment of sEMG-based systems.

Core claim

In data from 81 individuals performing discrete hand gestures, 33% (49 of 147) of commonly extracted sEMG features exhibit statistically significant associations with demographic characteristics when tested via mixed-effects linear models and partial least squares regression that incorporate age, sex, height, weight, skin properties, subcutaneous fat, and hair density.

What carries the argument

Mixed-effects linear models and partial least squares analysis applied to demographic predictors to quantify associations with each of the 147 sEMG features.

If this is right

  • Gesture-decoding models may require explicit demographic normalization or stratified training to maintain consistent accuracy across users.
  • Certain sEMG features with weaker demographic links could be prioritized for cross-population applications.
  • Assistive device developers should collect and report participant demographics when validating new sEMG algorithms.
  • Personalization pipelines for neural interfaces may need to include demographic covariates to reduce performance gaps.

Where Pith is reading between the lines

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

  • Retraining classifiers after removing or correcting the 49 affected features could measurably improve fairness metrics on held-out demographic groups.
  • The same analysis pipeline could be applied to other biosignals such as EEG or force myography to check for comparable demographic sensitivities.
  • Electrode placement or signal preprocessing choices not varied in this study might interact with the observed associations and warrant targeted follow-up experiments.

Load-bearing premise

The chosen statistical models and demographic variables capture the true sources of feature variability without leaving substantial residual confounding from unmeasured factors or from the specific feature extraction methods.

What would settle it

An independent replication study that applies identical feature extraction, mixed-effects models, and partial least squares analysis to a new cohort of at least 80 demographically matched participants and finds fewer than 15% of the same features showing significant demographic associations.

Figures

Figures reproduced from arXiv: 2604.14460 by Aditi Agrawal, Celine John Philip, Giancarlo K. Sagastume, Jonathon S. Schofield, Lee M. Miller, Marcus A. Battraw, Richard S. Whittle, Wilsaan M. Joiner.

Figure 1
Figure 1. Figure 1: Top sEMG Features by number of significant Demographic associations [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Distribution of Effect Sizes Across Demographics [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Significant Feature–Demographic Associations [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: PLS clustered image map of sPLS component 1 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
read the original abstract

Neuromotor decoding from upper-limb electromyography (sEMG) can enhance human-machine interfaces and offer a more natural means of controlling prosthetic limbs, virtual reality, and household electronics. Unfortunately, current sEMG technology does not always perform consistently across users because individual differences such as age and body mass index, among many others, can substantially alter signal quality. This variability makes sEMG characteristics highly idiosyncratic, often necessitating laborious personalization and iterative tuning to achieve reliable performance. This variability has particular import for sEMG-based assistive devices and neural interfaces, where demographic biases in sEMG features could undermine broad and fair deployment. In this study, we explore how demographic differences affect the sEMG signals produced and their implications for machine learning-based gesture decoding. We analyze the data set provided by, in which we derive 147 common sEMG features extracted from 81 demographically diverse individuals performing discrete hand gestures. Using mixed-effects linear models and partial least squares (PLS) analysis, which take into consideration demographic variables (including age, sex, height, weight, skin properties, subcutaneous fat, and hair density), we identify that 33\% (49 of 147) of commonly used sEMG features show significant associations with demographic characteristics. These results may help guide the development of fair and unbiased sEMG-based neural interfaces across a diverse population.

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 / 2 minor

Summary. The manuscript analyzes 147 common surface electromyography (sEMG) features extracted from 81 demographically diverse participants performing discrete hand gestures. Using mixed-effects linear models and partial least squares (PLS) regression that incorporate demographic variables (age, sex, height, weight, skin properties, subcutaneous fat, and hair density), the authors report that 33% (49 of 147) of these features exhibit statistically significant associations with demographic characteristics. The work positions these findings as guidance for developing fairer sEMG-based neural interfaces and machine-learning gesture decoders.

Significance. If the numerical claim is robust, the result supplies concrete empirical evidence that standard sEMG feature sets are not demographically neutral, which is directly relevant to equitable deployment of upper-limb prosthetics, VR controllers, and assistive devices. The study benefits from a relatively large and diverse cohort together with the use of mixed-effects modeling (to handle repeated measures) and PLS (to address multicollinearity among predictors); these methodological choices are appropriate for the question and constitute a strength.

major comments (1)
  1. [Abstract] Abstract: The central claim that 49 of 147 features show significant demographic associations rests on mixed-effects linear models, yet the abstract (and the provided description) gives no indication of multiplicity correction, effect-size reporting, or model diagnostics. With 147 features tested against multiple correlated demographics, an uncorrected nominal p < 0.05 threshold would be expected to produce roughly 7 false positives under independence; the reported 33% figure could therefore be inflated. Please state the exact correction procedure (Bonferroni, FDR, etc.), supply adjusted p-values or q-values, and report standardized effect sizes or R^{2} values for the flagged associations.
minor comments (2)
  1. [Abstract] Abstract: The sentence describing the data source is truncated ('the data set provided by, in which we derive'); supply the complete citation or repository reference.
  2. The manuscript should clarify how the 147 features were selected and whether any sensitivity analysis was performed with respect to alternative window lengths, filtering choices, or normalization procedures.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful and constructive review of our manuscript. We address the single major comment below and will incorporate the requested clarifications into the revised version.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that 49 of 147 features show significant demographic associations rests on mixed-effects linear models, yet the abstract (and the provided description) gives no indication of multiplicity correction, effect-size reporting, or model diagnostics. With 147 features tested against multiple correlated demographics, an uncorrected nominal p < 0.05 threshold would be expected to produce roughly 7 false positives under independence; the reported 33% figure could therefore be inflated. Please state the exact correction procedure (Bonferroni, FDR, etc.), supply adjusted p-values or q-values, and report standardized effect sizes or R^{2} values for the flagged associations.

    Authors: We agree that the abstract should explicitly describe the statistical procedures used to support the central claim. In the Methods section, each of the 147 features was modeled separately with mixed-effects linear regression that included the demographic predictors as fixed effects and participant as a random effect. To control the family-wise error rate across the 147 tests, we applied the Benjamini-Hochberg false-discovery-rate procedure and retained only associations with FDR-adjusted q < 0.05; the 49 features reported are those that survived this threshold. We will revise the abstract to state: “After Benjamini-Hochberg FDR correction for multiple comparisons, 33 % (49 of 147) of features exhibited significant demographic associations.” Standardized effect sizes (partial R² for each demographic predictor) and the corresponding adjusted q-values are provided for all significant features in Supplementary Table S3; we will add a sentence to the abstract summarizing the median partial R² across the retained associations. Model diagnostics (residual normality via Q-Q plots, homoscedasticity via scale-location plots, and variance-inflation factors < 5) were examined and are described in the Methods; we will insert a one-sentence summary of these checks into the abstract as well. These additions will make the statistical foundation of the 33 % figure transparent without altering the reported count of significant features. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical associations derived from raw signals and demographics

full rationale

The paper extracts 147 sEMG features from raw signals of 81 subjects, then applies mixed-effects linear models and PLS regression treating demographic variables (age, sex, height, weight, skin properties, fat, hair density) as independent predictors. The reported 33% (49/147) significant associations are direct outputs of these statistical tests on the data; no equation, self-citation chain, or ansatz reduces the count or the associations to a definition or fit of the target result itself. The derivation chain is self-contained against external benchmarks and does not invoke uniqueness theorems or rename known patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper applies standard statistical techniques to an existing dataset; no new free parameters, axioms beyond routine model assumptions, or invented entities are introduced in the abstract.

axioms (2)
  • standard math Standard assumptions of mixed-effects linear models hold (normality of residuals, appropriate random effects structure).
    Implicit in the use of mixed-effects models to test demographic associations.
  • domain assumption The listed demographic variables (age, sex, height, weight, skin properties, subcutaneous fat, hair density) capture the primary sources of sEMG variability.
    Used directly as predictors in the models.

pith-pipeline@v0.9.0 · 5577 in / 1400 out tokens · 58905 ms · 2026-05-10T11:56:11.755900+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

12 extracted references · 12 canonical work pages

  1. [1]

    Evaluation of Surface EMG-based Recognition Algorithms for Decod- ing Hand Movements

    Sara Abbaspour et al. “Evaluation of Surface EMG-based Recognition Algorithms for Decod- ing Hand Movements”. In:Medical & Biological Engineering & Computing58.1 (Jan. 2020), pp. 83–100.issn: 1741-0444.doi:10.1007/s11517-019-02073-z

    work page doi:10.1007/s11517-019-02073-z 2020

  2. [2]

    Electromyography Data for Non-Invasive Naturally-Controlled Robotic Hand Prostheses

    Manfredo Atzori et al. “Electromyography Data for Non-Invasive Naturally-Controlled Robotic Hand Prostheses”. In:Scientific Data1 (2014), p. 140053.issn: 2052-4463.doi: 10.1038/sdata.2014.53

    work page doi:10.1038/sdata.2014.53 2014

  3. [3]

    Surface Electromyography Evaluation for Decoding Hand Motor Intent in Children with Congenital Upper Limb Deficiency

    Marcus A. Battraw et al. “Surface Electromyography Evaluation for Decoding Hand Motor Intent in Children with Congenital Upper Limb Deficiency”. In:Scientific Reports14.1 (Dec. 2024), p. 31741.issn: 2045-2322.doi:10.1038/s41598-024-82519-z

    work page doi:10.1038/s41598-024-82519-z 2024

  4. [4]

    Surface Electromyography Signal Processing and Classification Techniques

    Rubana H. Chowdhury et al. “Surface Electromyography Signal Processing and Classification Techniques”. In:Sensors (Basel, Switzerland)13.9 (Sept. 2013), pp. 12431–12466.issn: 1424- 8220.doi:10.3390/s130912431

    work page doi:10.3390/s130912431 2013

  5. [5]

    The Loss of Skeletal Muscle Strength, Mass, and Quality in Older Adults: The Health, Aging and Body Composition Study

    B. H. Goodpaster et al. “The Loss of Skeletal Muscle Strength, Mass, and Quality in Older Adults: The Health, Aging and Body Composition Study”. In:The Journals of Gerontology Series A: Biological Sciences and Medical Sciences61.10 (Oct. 2006), pp. 1059–1064.issn: 1079-5006, 1758-535X.doi:10.1093/gerona/61.10.1059. 14

    work page doi:10.1093/gerona/61.10.1059 2006

  6. [6]

    A Database of Upper Limb Surface Electromyogram Signals from Demographically Diverse Individuals

    Harshavardhana T. Gowda et al. “A Database of Upper Limb Surface Electromyogram Signals from Demographically Diverse Individuals”. In:Scientific Data12.1 (Mar. 2025), p. 517.issn: 2052-4463.doi:10.1038/s41597-025-04825-z

    work page doi:10.1038/s41597-025-04825-z 2025

  7. [7]

    Sex Differences in Human Skeletal Muscle Fatigue:

    Audrey L. Hicks, Jane Kent-Braun, and David S. Ditor. “Sex Differences in Human Skeletal Muscle Fatigue:” in:Exercise and Sport Sciences Reviews29.3 (July 2001), pp. 109–112.issn: 0091-6331.doi:10.1097/00003677-200107000-00004

    work page doi:10.1097/00003677-200107000-00004 2001

  8. [8]

    Pushing the Frontiers of Unconstrained Face Detection and Recog- nition: IARPA Janus Benchmark A

    Brendan F. Klare et al. “Pushing the Frontiers of Unconstrained Face Detection and Recog- nition: IARPA Janus Benchmark A”. In:2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, MA, USA: IEEE, June 2015, pp. 1931–1939.isbn: 978-1-4673-6964-0.doi:10.1109/CVPR.2015.7298803

    work page doi:10.1109/cvpr.2015.7298803 2015

  9. [9]

    The Effect of Subcutaneous Fat on My- oelectric Signal Amplitude and Cross-Talk

    T. A. Kuiken, M. M. Lowery, and N. S. Stoykov. “The Effect of Subcutaneous Fat on My- oelectric Signal Amplitude and Cross-Talk”. In:Prosthetics & Orthotics International27.1 (Apr. 2003), pp. 48–54.issn: 0309-3646.doi:10.3109/03093640309167976

    work page doi:10.3109/03093640309167976 2003

  10. [10]

    Effects of Muscle Fiber Type and Size on EMG Median Frequency and Conduction Velocity

    E. J. Kupa et al. “Effects of Muscle Fiber Type and Size on EMG Median Frequency and Conduction Velocity”. In:Journal of Applied Physiology79.1 (July 1995), pp. 23–32.issn: 8750-7587, 1522-1601.doi:10.1152/jappl.1995.79.1.23

    work page doi:10.1152/jappl.1995.79.1.23 1995

  11. [11]

    2024.doi:10.48550/ARXIV.2412.02725

    Sasha Salter et al.Emg2pose: A Large and Diverse Benchmark for Surface Electromyographic Hand Pose Estimation. 2024.doi:10.48550/ARXIV.2412.02725

    work page doi:10.48550/arxiv.2412.02725 2024

  12. [12]

    The Effects of Obesity on Skeletal Muscle Contractile Function

    Jason Tallis, Rob S. James, and Frank Seebacher. “The Effects of Obesity on Skeletal Muscle Contractile Function”. In:Journal of Experimental Biology221.13 (July 2018), jeb163840. issn: 1477-9145, 0022-0949.doi:10.1242/jeb.163840. 15 Appendix Table 1: Table A1: Time-Domain (TD) Features Abbreviation Description MAV Mean Absolute Value: Average of absolute...

    work page doi:10.1242/jeb.163840 2018