From Controlled Trials to Big Data and Back
Statistical data analyses are sometimes classified as being either exploratory or confirmatory, while the reality of statistical practice often lies in between. This middle territory is exemplified by “model selection” issues and Frank Harrell’s famous words: “Using the data to guide the data analysis is almost as dangerous as not doing so”.
The most accomplished confirmatory statistical analyses are conducted in the context of controlled (clinical) trials, where regulations and guidelines are to ensure a fully protocoled and planned statistical analysis. On the other hand, we are now living in the era of “big data” and “data science”, where extreme forms of exploratory data analyses are encouraged with the hope that data quantity prevails over data quality.
While data science is currently in vogue, there is also some perception that “those who ignore statistics are condemned to reinvent it”, as Brad Efron once said. It might be a time to return from the big data paradigm towards more classical approaches and concerns, and to land somewhere between the two extremes of the purely confirmatory and purely exploratory data analyses.
The XXXIst ROeS statistical conference will be a timely occasion to try to define what this “middle ground” should or could be to best meet the expectations of scientists.
- Sander Greenland (University of California)
- Stephen Senn (Luxembourg Institute of Health)
- Stefan Wäger (University of Stanford)
Confirmed invited speakers
- Jan Beyersmann (University of Ulm)
- Florian Frommlet (Medical University of Vienna)
- Els Goetghebeur (University of Gent)
- Martin Huber (University of Fribourg)
- Marcus Hudec (University of Vienna)
- Rianne Jacobs (University of Groningen)
- Markus Lange (Novartis, Basel)
- Christoph Lippert (University of Postdam)
- Kaspar Rufibach (Roche, Basel)
- Georgia Salanti (University of Berne)
- Martin Spindler (University of Hamburg)
- Maarten van Smeden (University of Leiden)
- Manuela Zucknick (University of Oslo)