Miss­ing data? The TARMOS frame­work for acces­si­ble, repro­du­ci­b­le analysis


Although there is incre­asing guidance on how to hand­le miss­ing data, prac­ti­ce is chan­ging slow­ly and misappre­hen­si­ons abound. Important­ly, the lack of trans­pa­ren­cy around metho­do­lo­gi­cal decis­i­ons is threa­tening the vali­di­ty and repro­du­ci­bi­li­ty of modern research.

I will descri­be how the recent­ly published ‚Treat­ment And Report­ing of Miss­ing data in Obser­va­tio­nal Studies‘ (TARMOS) frame­work seeks to address these concerns, arguing that multi­ple impu­ta­ti­on (which can be used in an ever incre­asing range of settings) provi­des a natu­ral tool to imple­ment this frame­work. The need to assist analysts with imple­men­ting multi­ple impu­ta­ti­on has moti­va­ted recent work deve­lo­ping ‘mi.doc’ – an expert system in the statis­ti­cal soft­ware R, which I will brief­ly demonstrate. 

The ideas will be illus­tra­ted with an analy­sis of data from the UK Avon Longi­tu­di­nal Study of Parents And Child­ren (ALSPAC), explo­ring whether there is a causal link between smoking at 14 years and educa­tio­nal attain­ment at 16 years.


Carpen­ter, J. R., Bart­lett, J. W., Morris, T. P., Wood, A. M.,Quartagno, M. and Kenward, M. G. (2023) Multi­ple Impu­ta­ti­on and its Appli­ca­ti­on (second editi­on). Wiley. 

Lee, K. J., Tilling, K. M., Cornish, R. P. Litt­le, R. J. A., Bell, M. L., Goetghe­be­ur, E., Hogan, J. W. and Carpen­ter J. R. on behalf of the STRATOS initia­ti­ve. Frame­work for the treat­ment and report­ing of miss­ing data in obser­va­tio­nal studies: The Treat­ment And Report­ing of Miss­ing data in Obser­va­tio­nal Studies frame­work. Jour­nal of Clini­cal Epide­mio­lo­gy (2021), vol 134, pages 79–88.


James Carpenter is profes­sor of medi­cal statis­tics at the London School of Hygie­ne & Tropi­cal Medi­ci­ne and MRC Inves­ti­ga­tor in Trials Metho­do­lo­gy at the MRC Clini­cal Trials Unit at UCL, London. His prin­ci­pal rese­arch inte­rests are coping with miss­ing data in clini­cal trials and complex hier­ar­chi­cal models (espe­ci­al­ly using multi­ple impu­ta­ti­on metho­do­lo­gy), esti­man­ds and sensi­ti­vi­ty analy­sis, meta-analysis and novel trial designs.


Datum: Mitt­woch, 20 Septem­ber
Beginn: 10:00 – 11:00 Uhr