Sensitivity analysis for missing data and measurement error
Speaker(s):
Bio: I am an Associate Professor in Quantitative Criminology with a background in Social Statistics. Most of my research has focused on the analysis of unwarranted disparities in criminal justice decisions, for what I have collaborated with the Crown Prosecution Service, the Sentencing Council for England and Wales, and the Parole Board. More recently I have been working on the problem of measurement error in police statistics, where I have been exploring its prevalence, impact, and adjustment strategies.
Bio: Sara's research interests centre around causal inference. This is the area of statistical methodology concerned with identifying and estimating effects of interventions. She has been involved in a number of grants where causal inference methods have been applied to large data sets in public health and criminology. Most recently, Sara has focused on natural experiments (regression discontinuity and interrupted time series designs), current interests include exploring the role of discrimination in the Criminal Justice System using MoJ data. Sara is also a Bayesian and all her research is embedded in this paradigm.
Abstract:
Missing data and measurement error are two common problems affecting social research. Sometimes we can rely on auxiliary data to adjust for them, but often all we have is just an educated guess, which can still be used to undertake sensitivity analysis.
In this workshop we will explore how to simulate a range of ‘likely’ scenarios based qualitative insights regarding the magnitude and direction of the suspected missing data or measurement error mechanisms. Specifically, we will use multiple imputations and the MICE package in R.
Following a short introduction to the topic, participants will be invited to work on one of two exercises: i) estimating the mitigating effect of ‘showing remorse in court’ when more punitive judges are less likely to complete their questionnaires; or ii) the association between unemployment and crime across Police Force areas when crime reporting rates are themselves affected by the unemployment rate.