We have longitudinal data collected for a cohort (n=356) of patients attending the South-Verona community-based service on three occasions: 1994 (baseline), 1996 (first wave) and 2001 (second wave). The variables pertain socio-demographical, clinical and service utilization informations (available from the case register) and five instruments (BPRS, GAF, DAS, LQL and VSSS: they are all measured on ordinal scales, but we use them just as they were continuos variables) on each follow-up occasion. We are facing with missing value problem: infact a certain number of patients completed all the instruments on each occasion; other patients completed the instruments on two out of three occasions; at the end, another part of the cohort completed the instruments on only one occasion. For example, if we consider all the five instruments, 261 patients have complete information at baseline, 182 out of 261 have complete information at 1996 assessment and only 79 out of 182 have complete data at 2001 assessment. If we consider each instrument, the complete data pattern is better than the previous one.
We want to make an attempt to estimate missing data values, possibly manteining a longitudinal perspective for each subject. Our future statistical analysis will regard linear and logistic multiple regressions, analysis of variance, factor analysis and so on.
I know that Schafer (1997) proposed a multiple imputation procedure for missing data. In you opinion, is it possible to apply successfully this method to our database? Are there procedures that permit a better solution of data missing problem?
On the other side, are there statistical models (for example random regression models) that allow for missing data?
Dr. Chiara Bonetto, Dr.Stat., Ph.D.
Statistician
Department of Medicine and Public Health
Section of Psychiatry
WHO Collaborating Centre for Research and Training in Mental Health and Service Evaluation
University of Verona
P.le L.A. Scuro, 10
37134 Verona, Italy
Tel: +39 045 8074441
Fax: +39 045 585871
E-mail: [log in to unmask]
Web page: http://www.medicina.univr.it/~psymed/
|