Loading Events
  • This event has passed.

Yannick Guyonvarch (CREST): "Finite Confidence Intervals for Ratios of Means"

November 22, 2018 @ 12:30 pm - 1:30 pm

Firms and Markets Seminar
 
Abstract: 

In applied economics, many parameters of interest can be viewed as (combinations of) ratios of means. Two leading examples are (i) average treatment effect estimands and (ii) the average outcome in a population where individuals are allowed to be clustered (e.g. average grade of pupils when the latter are assumed to be clustered at the class level). The construction of confidence intervals for such quantities relies most of the time on asymptotic normality and the delta method. However, when the number of observations is small and/or the denominator of the ratio is close to zero, the relevance of asymptotic confidence intervals can be questioned.
The aim of next week’s talk is threefold:
1. define the notion of finite sample confidence intervals and present classical results from the statistical literature to build such finite sample confidence intervals for means;
 2. present our finite sample confidence intervals for ratios of means and discuss the limitations of what we have found so far;
 3. show some simulations to assess the behavior of our finite sample confidence intervals and how they compare to asymptotic ones.
As for now, the state of the project is rather preliminary so we are looking forward to hearing your comments!