Dr.Prashant Mukkannavar, Ph.D’s Post

Are we guilty of using statistics to reinforce our opinions without questioning them? Most of us are. Laziness is the most basic reason for this mistake, regardless of the source's reliability. It's time to think about how accurate the figures we use are. Check out this presentation that explores this issue and learn how to question sources better in the future. Here's the link: https://lnkd.in/gCpU6fwh

"The fundamental analytical act in statistical reasoning is to answer the question 'Compared with what?'" (Edward Tufte, Beautiful Evidence) Show numbers over time. Show numbers against a target. Show numbers for the values of a dimension. But always compare them with something.

On the statistical Lie groups of normal distributions https://lnkd.in/eEPXf4WU

Simple method for calculating decision criteria weights: Improving human decision intelligence with Julia code https://lnkd.in/gHtdz8Ga

Good reference sheet for statistical tests.

https://lnkd.in/gMgX43M6 A delightful article about reframing using complementary thinking… a must read in this day and age of misinformation. Starts off with very simple probability principles, but the latter examples show the power of the complement!

Closed form of the Fisher information metric is a good example of application of Riemannian metric and Levi-Civita connection. You ought to love differential geometry.

Want to disentangle causal effects? Can the average treatment effect be partially explained as a mediated effect through another variable? Explain causal effects using a mediating variable with the new mediate command in #Stata18. Learn more: https://buff.ly/3UdmTTu

Reproducibility in independent cohorts/experiments, and convergence of different approaches, should be the gold standard for science getting done and published. (That is how my teams have done our studies for the last 25 years). Statistical significance by itself could be just overfitting to the particular cohort you are studying. I wrote a paper about that years ago called “ The p-value Illusion”( the original title had the word Delusion instead of Illusion). If you are going to use statistics as part of a comprehensive approach ( and you should), Bayesian approaches tend to be more informative and more reproducible than Fischerian ones. In addition, longitudinal studies are more powerful than cross-sectional ones, especially if you include within-subject designs. Of course, they take time and effort to do and collect your own data. (People prefer to analyze instantly other people’s data, and trumpet p-values in a larger N of a hodgepodge of aggregated imperfect datasets. The findings then do not reproduce).

Our August edition of IWH News is out! 👇 This issue includes: ➡Five things we think you should know ➡Severe work-related injury rates not decreasing in Ontario ➡Estimating the worker and job groups most likely to be affected by machine learning ...and more! Access the full issue here: https://lnkd.in/gtb8DKbx