Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

III. Data Layout

Basic data layout for correlated
analysis:

Ksubjects

niresponses for subjecti

The basic data layout for a correlated analysis

is presented to the left. We consider a longitu-

dinal dataset in which there are repeated mea-

sures forK subjects. Theith subject has ni

measurements recorded. Thejth observation

from theith subject occurs at timetijwith the

outcome measured asYij, and withpcovariates,

Xij 1 ,Xij 2 ,...,Xijp.

Subjects are not restricted to have the same

number of observations (e.g.,n 1 does not have

to equaln 2 ). Also, the time interval between

measurements does not have to be constant

(e.g., t 12 t 11 does not have to equal

t 13 t 12 ). Further, in a longitudinal design, a

variable (tij) indicating time of measurement

may be specified; however, for nonlongitudinal

designs with correlated data, a time variable

may not be necessary or appropriate.

The covariates (i.e.,Xs) may be time-indepen-

dent or time-dependent for a given subject. For

example, the race of a subject will not vary, but

the daily intake of coffee could vary from day to

day.

Presentation: III. Data Layout 499