New Era of Statistics

Pure-prediction algorithms and how they interact with and relate to classical statistics.

Blogposts

Today we discuss the new era of pure-prediction algorithms, the differences that these algorithms have from more classical statistical approach, and how the landscape is changing to possibly accomodate for them. Below are a short list of different algorithms that each serve a differnt purpose.

  • Prediction: Random Forests. Boosting, Support Vector Machines, Neural Nets, Deep Learning
  • Estimation: Odinary Least Squares (OLS), Logistic Regression, Generalized Linear Model: Maximum Likelihood Estimation
  • Attribution (Significance): ANOVA, Lasso, Neyman-Pearson

Perhaps if you are familiar with buzzwords from the data science world, you know some (or all) of the algorithms listed under the "Prediction" catagory. These have gotten amazing public attention, and while still stemming from traditional regression theory, they can operate on an enormous scale and have many popular successes. Equivalently, if you are familiar with statistics, you may recall hearing some of the algorithms listed in the other two catagory. These are some things that lasted the trials of time, something that we have come to depend on over the years.

Perceived Reliability of Eyewitness Statements

impact of frequency and severity of a crime on perception of eyewitness confidence statements

Blogposts

In contrast to the eyewitness testimonies at the time of trial, recent empirical works suggest that eyewitnesses' confidence statements at the time of initial identification may be a highly reliable indicator of accuracy. During the lineup, eyewitnesses are told to state in words how confident they are about the identification. For this study, a convenient sample of individuals online was asked to interpret how confident eyewitnesses are to observe the impacts of varying factors on the perception of eyewitness confidence.

The study shows that the frequency and severity, as expected, holds a significant impact on the perception of how confident the eyewitness was in the lineup. Somewhat interestingly, the race of lineup did not play a statistically significant role according to the dataset. The race, age, and state/region of the evaluator remain to be explored as a part of future analysis.