BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:MRC Biostatistics Unit Seminars
SUMMARY:Congenial multiple imputation of partially observe
d covariates within the full conditional specifica
tion framework - Jonathan Bartlett\, LSHTM.
DTSTART;TZID=Europe/London:20120515T143000
DTEND;TZID=Europe/London:20120515T153000
UID:TALK37217AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/37217
DESCRIPTION:Missing covariate data is a common issue in epidem
iological and clinical research\, and is often dea
lt with using multiple imputation (MI). When the a
nalysis model is non-linear\, or contains non-line
ar (e.g. squared) or interaction terms\, this comp
licates the imputation of covariates. Standard sof
tware implementations of MI typically impute covar
iates from models that are uncongenial with such a
nalysis models. We show how imputation by full con
ditional specification\, a popular approach for pe
rforming MI\, can be modified so that covariates a
re imputed from a model which is congenial with th
e analysis model. We investigate through simulatio
n the performance of this proposal\, and compare i
t to passive imputation of non-linear or interacti
on terms and the `just another variableâ€™ approach.
Our proposed approach provides consistent estimat
es provided the imputation models and analysis mod
els are correctly specified and data are missing a
t random. In contrast\, passive imputation of non-
linear or interaction terms generally results in i
nconsistent estimates of the parameters of the mod
el of interest\, while the `just another variable'
approach gives consistent results only for linear
models and only if data are missing completely at
random. Furthermore\, simulation results suggest
that even under imputation model mis-specification
our proposed approach gives estimates which are s
ubstantially less biased than estimates based on p
assive imputation. The proposed approach is illust
rated using data from the National Child Developme
nt Survey in which the analysis model contains bot
h non-linear and interaction terms.
LOCATION:Large Seminar Room\, 1st Floor\, Institute of Pub
lic Health\, University Forvie Site\, Robinson Way
\, Cambridge
CONTACT:Dr Jack Bowden
END:VEVENT
END:VCALENDAR