SDS 581: Bayesian, Frequentist, and Fiducial Statistics in Data Science — with Xiao-Li Meng

Data Quality

The importance of data quality in statistical analysis cannot be overstated. Xiao-Li Meng emphasizes the need for a thorough understanding of data origins, collection methods, and potential biases before analysis. He warns against the "garbage in, garbage out" phenomenon, where poor data quality leads to misleading conclusions 1. Jon Krohn agrees, noting that without proper data "minding," analysts risk missing the true signals within datasets 2.

   

Big Data Paradox

The big data paradox reveals how large datasets can lead to misleading outcomes if not carefully assessed. Xiao-Li Meng illustrates this with the 2016 election predictions, where vast amounts of data resulted in inaccurate forecasts due to non-response biases 3. He explains that as data size increases, the confidence in incorrect conclusions can also grow, likening it to tasting a poorly mixed soup 4.

   

Data Confession

Data confession is a practice that encourages transparency and integrity in research. Xiao-Li Meng advocates for openly discussing data defects and limitations to enhance reproducibility and scientific progress 5. He notes that the current incentive system often discourages such transparency, as revealing flaws can lead to criticism rather than constructive dialogue.