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

Infant Care Study: Sample Data

From three infants: five (of nine)
observations listed for each

IDNO MO OUTCOME BIRTHWGT GENDER DIARRHEA
00282 1 0 2000 Male 0
00282 2 0 2000 Male 0
00282 3 1 2000 Male 1
   
00282 8 0 2000 Male 1
00282 9 0 2000 Male 0
....................................
00283 1 0 2950 Female 0
00283 2 0 2950 Female 0
00283 3 1 2950 Female 0
   
00283 8 0 2950 Female 0
00283 9 0 2950 Female 0
....................................
00287 1 1 3250 Male 1
00287 2 1 3250 Male 1
00287 3 0 3250 Male 0
   
00287 8 0 3250 Male 0
00287 9 0 3250 Male 0
....................................

IDNO: identification number

MO: observation month (provides
order to subject-specific measure-
ments)

OUTCOME: dichotomizedz-score
(values can change month to month)

Independent variables:

1. Time-dependent variable: can
   vary month to month within a
   cluster
   DIARRHEA: dichotomized
   variable for presence of
   symptoms


2. Time-independent variables:
   do not vary month to month
   within a cluster
   BIRTHWGT
   GENDER

On the left, we present data on three infants to

illustrate the layout for correlated data. Five

of nine monthly observations are listed per

infant. In the complete data on 136 infants,

each child had at least 5 months of observa-

tions, and 126 (92.6%) had complete data for

all 9 months.

The variable IDNO is the number that identi-

fies each infant. The variable MO indicates

which month the outcome measurement was

taken. This variable is used to provide order for

the data within a cluster. Not all clustered data

have an inherent order to the observations

within a cluster; however, in longitudinal stud-

ies such as this, specific measurements are

ordered over time.

The variable OUTCOME is the dichotomized

weight-for-heightz-score indicating the pres-

ence or absence of wasting. Notice that the

outcome can change values from month to

month within a cluster.

The independent variable DIARRHEA can also

change values month to month. If symptoms of

diarrhea are present in a given month, then the

variableiscoded1;otherwiseitiscoded0.DIAR-

RHEA is thus atime-dependentvariable. This

contrasts with the variables BIRTHWGT and

GENDER, which do not vary within a cluster

(i.e., do not change month to month).BIRTHWGT

and GENDER aretime-independentvariables.

494 14. Logistic Regression for Correlated Data: GEE