Home > Research > Publications & Outputs > Developing robust statistical scoring methods f...

Electronic data

  • 2018gichuruphd

    Final published version, 4.64 MB, PDF document

    Available under license: CC BY-ND: Creative Commons Attribution-NoDerivatives 4.0 International License

Text available via DOI:

View graph of relations

Developing robust statistical scoring methods for use in child assessment tools

Research output: ThesisDoctoral Thesis

Unpublished
Publication date2018
Number of pages290
QualificationPhD
Awarding Institution
Supervisors/Advisors
  • Lancaster, Gillian, Supervisor, External person
  • Titman, Andrew, Supervisor
Publisher
  • Lancaster University
<mark>Original language</mark>English

Abstract

Timely and accurate diagnosis of developmental disability reduces its detrimental effect on children. Most of the current scoring methods do not appropriately remove the effect of age on development scores. This frustrates both disability status classification and comparison of scores across different child populations because their age dependent development profiles are usually quite different. Hence, the key objective of this research is to develop robust statistical scoring methods that appropriately correct for age using a) item by item age estimation methods that provide the expected age of achieving specific developmental milestones and b) overall score norms independent of the age effect using all the responses of a child to give one score across the entire domain for each child. Using data from 1,446 healthy and normally developing children (standard group) from the 2007 Malawi Development Assessment Tool (MDAT) study, a review of classical methods including generalised linear models, simple sum, Z-score, Log Age Ratio and Item Response Theory scoring methods in this child development context using binary responses only was carried out. While evaluating the pros and cons of each method, extensions to the current scoring methods using more flexible and robust methods including smoothing to reduce score variability are suggested. The results show that; a) the suggested generalized additive model extensions used for age estimation were more suited to deal with skewed item pass rate response distributions, b) smoothing of Z-scores was especially beneficial when variability in certain age groups is high due to low sample sizes, c) the more complex methods accounting for item response correlation or increase in item difficulty resulted in reliable and generalisable normative scores d) the extended overall scoring approaches were able to effectively correct for age achieving correlation coefficients of less than +0.25 between age and scores. The suggested overall scoring extensions improved the accuracy of detecting delayed development both in the disabled and even in the harder to classify malnourished children achieving sensitivity values of up to 98% and 85% respectively.