Type I and Type II error rates in the last observation carried forward method under informative dropout

Chandan Saha and Michael P. Jones

Journal of Applied Statistics, 2016, vol. 43, issue 2, 336-350

Abstract: Dropout is a persistent problem for a longitudinal study. We exhibit the shortcomings of the last observation carried forward method. It produces biased estimates of change in an outcome from baseline to study endpoint under informative dropout. We developed a theoretical quantification of the effect of such bias on type I and type II error rates. We present results for a setup where a subject either completes the study or drops out during one particular interval, and also under the setup in which subjects could drop out at any time during the study. The type I error rate steadily increases when time to dropout decreases or the common sample size increases. The inflation in type I error rate can be substantially high when reasons for dropout in the two groups differ; when there is a large difference in dropout rates between the control and treatment groups and when the common sample size is large; even when dropout subjects have one or two fewer observations than the completers. Similar results are also observed for type II error rates. A study can have very low power when early recovered patients in the treatment group and worsening patients in the control group drop out even near the end of the study.

Date: 2016
References: View complete reference list from CitEc
Citations:

Downloads: (external link)
http://hdl.handle.net/10.1080/02664763.2015.1063112 (text/html)
Access to full text is restricted to subscribers.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:taf:japsta:v:43:y:2016:i:2:p:336-350

Ordering information: This journal article can be ordered from
http://www.tandfonline.com/pricing/journal/CJAS20

DOI: 10.1080/02664763.2015.1063112

Access Statistics for this article

Journal of Applied Statistics is currently edited by Robert Aykroyd

More articles in Journal of Applied Statistics from Taylor & Francis Journals
Bibliographic data for series maintained by Chris Longhurst ().