4. 1 Missing focus on prediction
Bzdok, 2017 „Classical Statistics and Statistical Learning in Imaging Neuroscience“
Shmueli, 2010 „To explain or to predict?“
Yarkoni & Westfall, 2016. Choosing prediction over explanation in psychology: Lessons from machine learning.
5. 1 Missing focus on prediction
Bzdok, 2017 „Classical Statistics and Statistical Learning in Imaging Neuroscience“
Shmueli, 2010 „To explain or to predict?“
Yarkoni & Westfall, 2016. Choosing prediction over explanation in psychology: Lessons from machine learning.
7. 2 P-values cannot be applied to single patients
Null hypothesis testing Out-of-sample generalization
Bzdok, Krzywinski, Altman, 2017 „Machine learning: A primer“ Nature Methods
Casella & Berger, 2002 „Statistical Inference“
8. 3 P-values cannot be applied to future individuals
Casella & Berger, 2002 „Statistical Inference“
9. 4 Hypothesis testing was not made for high-n setting
Berkson, 1938. „Some difficulties of interpretation encountered in the application of the chi-square test“
Bzdok, Nichols, Smith 2019. „Towards Algorithmic Analytics for Large-Scale Datasets“ Nature Machine Intelligence
Miller, et al., 2016. „Multimodal population brain imaging in the UK Biobank prospective epidemiological study.“
10. 4 Hypothesis testing was not made for high-n setting
Gelman and Carlin, 2017
Gelman and Carlin, 2017
11. 5 Hypothesis testing was not made for high-p setting
Breiman, 2001 (page 203)
Breiman, 2001. „Statistical Modeling: The two cultures“
Bzdok, Nichols, Smith 2019. „Towards Algorithmic Analytics for Large-Scale Datasets“ Nature Machine Intelligence
12. 5 Hypothesis testing was not made for high-p setting
James et al., 2013 (page 247)
James, 2013. „Introduction to Statistical Learning“
13. 6 Inferential thinking is often dichotomic in nature
Casella & Berger, 2002 „Statistical Inference“
Casella & Berger, 2002
14. 7 Small tradition in modeling many outcomes at once
ADHD vs. healthy
working memory performance ,
attention scores, IQ, socioeconomic
status, significant life events,
social interaction capacities,
gender, smoking behavior, caffeine
intake, mood assessment,…
Caruana, 1998. „Multitask learning“
16. 9 Precision medicine will be driven by observational data
Henke, et al., 2016. „The age of analytics: Competing in a data-driven world“ McKinsey Global Institute
Manyika, et al., 2015. „Unlocking the Potential of the Internet of Things“ McKinsey Global Institute
vs
Controlled experiment Internet of Things
17. 10 Small tradition in phenotype discovery
Insel, et al., 2015. „Brain disorders? Precisely.“ Science
Bengio et al., 2013. „Representation learning: A review and new perspectives“
18. Bzdok et al., 2018 Nature Methods
Prediction != inference in biomedicine
21. Bzdok & Ioannidis, 2019 Trends in Neurosciences
Prediction != inference in biomedicine
22. n=500
Automatic ranking of predictive cognitive domains in schizophrenia
Karrer, Bassett, …, Bzdok, 2019 Human Brain Mapping
23. Karrer, Bassett, …, Bzdok, 2019 Human Brain Mappingn=500
Automatic ranking of predictive cognitive domains in schizophrenia
24. n=1,300
A continuous ADHD-ASD axis can be learned from brain data
Kernbach, Satterthwaite, …, Bzdok, 2018 Translational Psychiatry
25. n=1,300
A continuous ADHD-ASD axis can be learned from brain data
Kernbach, Satterthwaite, …, Bzdok, 2018 Translational Psychiatry
26. A continuous ADHD-ASD axis can be learned from brain data
n=1,300 Kernbach, Satterthwaite, …, Bzdok, 2018 Translational Psychiatry
27. Mehdi, Bzdok, Thirion, Varoquaux, 2017 NeuroImage
Multi-output models do better in deeply phenotype brain data
n=500
28. Mensch, Mairal, Bzdok, et al., 2017 Neural Information Processing Systemsn=80-1,200
Deep learning extensions for transferring patterns between datasets
29. n=80-1,200 Mensch, Mairal, Bzdok, et al., 2017 Neural Information Processing Systems
Deep learning extensions for transferring patterns between datasets
30. Conclusion
Not a matter of tradition or taste.
Different statistical regimes serve distinct research goals.