IOE EPrints

Optimal modelling of hearing impairment in middle age in relation to hearing in childhood as measured by audiograms

Ecob, R (2008) Optimal modelling of hearing impairment in middle age in relation to hearing in childhood as measured by audiograms. Centre for Longitudinal Studies, Institute of Education, University of London, London. ISBN 1-898453-69-1

PDF (Ecob2008Optimal.pdf)
Ecob2008Optimal.pdf - Other

Download (796Kb) | Preview


NCDS (National Child Development Survey) data include detailed audiograms at childhood ages 7,11,16 years and a ‘cut down’ audiogram at 45 years(frequencies 1 and 4 kHz only). A range of models relating data in childhood to adulthood is examined. Hearing losses at all childhood ages contribute independently and with comparable relationships to adult hearing loss: such models should include adjustments at all ages and at all frequencies in combination. This is possible through polynomial contrasts between frequencies at each age. Models which use the logarithm of the adult hearing measure as dependent variable are compared with those which use the raw score. Residuals from raw score models are found to have unacceptable distributional properties which are avoided by models of the log hearing loss. However, care needs to be taken in the interpretation of the coefficients in these models which can be easily transformed into an additive contribution due to the effects of further explanatory variables at specified values of existing variables. For categorical variables such as gender this will be a particular category and for continuous variables, such as childhood hearing, usually the mean. A range of multiple imputation models is compared with a complete case analysis. The complete case analysis excludes more than half of the data in models. In these cases I recommend the use of multiple imputation models. These models are can be estimated using recent macros in ‘off-the-peg’ statistical software (e.g. Mice in STATA). Mixed-effects and latent trajectory growth models are alternatives to the conditional models used here. These allow each individual to have a unique (growth) trajectory over time. Their advantages are limited in the case of these data due to the concentration of hearing loss data in childhood resulting from the absence of measurements of hearing loss in early adulthood

Item Type: Book
Depositing User: IOE Repository Editor (2)
Date Deposited: 07 Jun 2010 10:01
Last Modified: 29 Jan 2015 10:22
View Item View Item