May 2005
Volume 46, Issue 13
ARVO Annual Meeting Abstract  |   May 2005
Nonlinear Mixed Effects Modeling as an Estimation Procedure for Sparse MNREAD Data
Author Affiliations & Notes
  • C.S. Kallie
    Psychology, University of Minnesota, Minneapolis, MN
  • S.H. Cheung
    Psychology, University of Minnesota, Minneapolis, MN
  • G.E. Legge
    Psychology, University of Minnesota, Minneapolis, MN
  • C. Owsley
    Ophthalmology, University of Alabama, Birmingham, AL
  • G. McGwin
    Ophthalmology, University of Alabama, Birmingham, AL
  • Footnotes
    Commercial Relationships  C.S. Kallie, None; S.H. Cheung, None; G.E. Legge, Lighthouse Low Vision Products P; C. Owsley, None; G. McGwin, None.
  • Footnotes
    Support  NIH Grant EY02934, NIH P30 AG11684, NIH R21 EY14071, RPB, and EyeSight Foundation of Alabama
Investigative Ophthalmology & Visual Science May 2005, Vol.46, 4589. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      C.S. Kallie, S.H. Cheung, G.E. Legge, C. Owsley, G. McGwin; Nonlinear Mixed Effects Modeling as an Estimation Procedure for Sparse MNREAD Data . Invest. Ophthalmol. Vis. Sci. 2005;46(13):4589.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Abstract: : Purpose: MNREAD is a clinical vision test that measures reading speeds at different print sizes. Range compression due to low vision or limited testing time can result in sparse response data. Sparse data can be difficult or impossible to fit with traditional modeling procedures. Nonlinear mixed effects (NLME) modeling provides parameter estimates for a population and individuals simultaneously – borrowing information from each individual to stabilize parameter estimations. The current study compares a NLME model to traditional individual–based least–squares (LS) models on MNREAD data. Methods: MNREAD data are often characterized by a rising branch at small print sizes, and an asymptotic maximum reading speed at large print sizes. Consequently, two 3–parameter functions – a compressive exponential function with an asymptote (ASYM), and a bilinear function – were compared. Individual LS fitting of both functions to MNREAD curves from 48 normally sighted people fit equally well. NLME modeling of ASYM was then compared with the LS fitting of ASYM in 3 different sets of MNREAD data: (1) Normal: binocular testing on two different MNREAD test versions from 48 normally sighted people; (2) Sparse Normal: 50% randomly sub–sampled data from the Normal data set (producing artificially sparse data); and (3) ICOM: clinical data from the ICOM cataract study containing MNREAD data from monocular testing of 341 patients aged 56 to 86 – with and without cataracts. Results: NLME and LS fits to the Normal data produced nearly identical results. NLME modeling outperformed LS modeling in both Sparse Normal data and ICOM data. In these two data sets, NLME produced reasonable estimates for all subjects, however LS modeling produced reasonable estimates for only 73% of the Sparse Normal data, and only 84% of the ICOM data. Conclusions: The NLME model had superior performance with sparse data compared to individually fit LS models. NLME models could improve parameter estimations in sparse MNREAD or other data, when data from individuals or eyes sampled from the same population can be analyzed together.

Keywords: reading • low vision • visual acuity 

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.