Table 3.

Hypothetical example of data with five imputed datasets

Imputed Dataset	ID	Age (yr)	Woman	BMI (kg/m2)			PPRA (%)		Stroke	Years Followed
				15–18.5	25–30	30–45	11–80	80–100
1	1	39	No	0	1	0	0	1	No	8.4
2	1	39	No	0	1	0	0	1	No	8.4
3	1	39	No	0	1	0	0	1	No	8.4
4	1	39	No	0	1	0	0	1	No	8.4
5	1	39	No	0	1	0	0	1	No	8.4
1	2	44	Yes	0.35a	0.34a	−0.21a	0	1	Yes	10.9
2	2	44	Yes	0.12a	0.45a	0.03a	0	1	Yes	10.9
3	2	44	Yes	0.21a	0.27a	−0.47a	0	1	Yes	10.9
4	2	44	Yes	−0.01a	0.97a	−0.44a	0	1	Yes	10.9
5	2	44	Yes	0.38a	0.80a	0.64a	0	1	Yes	10.9
1	4	67	No	0	0	0	−0.21a	0.08a	No	11.6
2	4	67	No	0	0	0	0.04a	−0.33a	No	11.6
3	4	67	No	0	0	0	0.25a	0.21a	No	11.6
4	4	67	No	0	0	0	0.31a	−0.04a	No	11.6
5	4	67	No	0	0	0	0.69a	0.07a	No	11.6

BMI=18.5–25 (normal) and PPRA=0–10 (normal) are used as reference groups and represented by a zero in all dummy variables pertaining to each variable. ID, identification; PPRA, panel reactive antibody.

^aMultiple imputation may lead to data that are not consistent with the original format; in this case, values imputed for missing observations of categorical (binary) data are continuous. Furthermore, although original categories of a variable may be mutually exclusive, imputed data may not be mutually exclusive, which is appropriate, because the imputed values, per se, do not have any meaning.