Skip to content
Browse files

RI basics, choices

  • Loading branch information...
1 parent 208c675 commit 5dcecfe11d53e981c8d25e7751028da2fca5c4d6 @markmfredrickson committed Mar 29, 2012
Showing with 40 additions and 0 deletions.
  1. +40 −0 presentation/analysis.Rnw
40 presentation/analysis.Rnw
@@ -98,3 +98,43 @@ cat(tmp.str)
These combine to form the \emph{Equal Environments Assumption (EEA)} \citep{Suhay:2009fy}.
+\begin{frame}{Randomization Inference}
+ \begin{itemize}
+ \item Inference based only the random assignment; potential outcomes fixed
+ \item Intuition: If we remove the ``true'' effect, we can shuffle the MZ
+ and DZ labels and see no change.
+ \item Relatively few assumptions on the data; flexible enough to consider very complex
+ interactions between units, including interference \citep{Bowers:2012ys}
+ \end{itemize}
+\begin{frame}{Randomization Inference Algorithm}
+ \begin{itemize}
+ \item Assignment Mechanism (e.g. simple random sample, coin flip)
+ \item Model of Effects ($Y_0 = Y_1 + \tau$, $Y_0 = \beta Y_1$,
+ interference)
+ \item Test Statistic (mean/median difference, rank sum, KS)
+ \item Algorithm:
+ \begin{enumerate}
+ \item Adjust data consistent with the model
+ \item Draw more random assignments
+ \item Compute test statistic
+ \end{enumerate}
+ \item Compute $p$-value of the model by comparing observed test statistic to the distribution
+ of test statistics
+ \item Implemented in development branch of \texttt{RItools}
+ \end{itemize}
+\begin{frame}{Selecting Mechanisms, Models, and Statistics}
+ \begin{itemize}
+ \item Researcher has freedom (burden?) to select the assignment mechanism,
+ model of effects, and test statistic.
+ \item \textbf{Assignment mechanism:} Usually fixed at design time. In this case,
+ will assume a simple random sample within blocks defined by matching.
+ \item \textbf{Test Statistic:} Proper procedure uses power analysis. I just pick
+ mean difference.
+ \item \textbf{Model of effects:} I develop several models that conform to the
+ restrictions imposed by the ideology variable.
+ \end{itemize}

0 comments on commit 5dcecfe

Please sign in to comment.
Something went wrong with that request. Please try again.