day19-OpenDiscussion.Rmd

---
title: Open Discussion of Various Topics
date: '`r format(Sys.Date(), "%B %d, %Y")`'
author: ICPSR 2021 Session 1
bibliography:
 - refs.bib
 - BIB/master.bib
 - BIB/misc.bib
 - BIB/causalinference.bib
fontsize: 10pt
geometry: margin=1in
graphics: yes
biblio-style: authoryear-comp
colorlinks: true
biblatexoptions:
  - natbib=true
output:
  beamer_presentation:
    slide_level: 2
    keep_tex: true
    latex_engine: xelatex
    citation_package: biblatex
    template: icpsr.beamer
    includes:
        in_header:
           - defs-all.sty
---

TSCS/Longtudinal
Interference
Covariance adjustment after matching
Regression sensitivity analysis.

## Today


  1. Agenda: Talk about the topics you mentioned last time and/or any topics
     that you came to class wanting to discuss
  2. Questions arising from the reading or assignments or life?

# But first, review

## What have we done so far this week?

 1. Using discontinuities as natural experiments or cut-points in
    deterministic processes that create mini-experiments.
 2. Creating matched designs or used instruments or discontinuities (or
    randomization itself) to create interpretable comparisons and justifiable
    statistical inferences about causal effects.
 3. Assessed the sensitivity of a design which adjusted well for
    $\bm{x}$ but which could not adjust directly for unmeasured confounders, $\bm{u}$
    1. Rosenbaum's approach focusing on $\Gamma$ (assuming strong relationship with $Y$).
    2. The $H^3$ approach focusing on $\teeW$ and $\pcor$ (noticing that even
    high $\teeW$ might not cause trouble if $\pcor$ is very low).

\begin{center}
\includegraphics[width=.6\textwidth]{xyzudiagram.pdf}
\end{center}

# Topics from last time.

## Topics:

 - Multiple imputation and matching (recall that missing data on covariates is
   different from missing data on outcomes; also could do MH matching on
   multiple PS scores created from multiply imputed data so that you have only
   one single matched set.)
 - Meta-analyses in combination with one-study testing and estimation.
 - General principles for the design of randomized experiments (for example,
   power)
 - Multilevel designs (like cluster randomized experiments, or test items
   within students, etc..)
 - Longitudinal designs (what is the outcome? what is, a priori, a good
   counterfactual?)
 - Experiments on networks
 - Bayesian predictive models
 - Bayesian posteriors with randomization based likelihoods
 - Academic styles and trends (How to respond to "I don't like X" where X can
   be any existing approach to research. What might "I don't like that causal
   inference stuff" actually mean?)
   - More than more way to conceptualize causal relationships.
   - More than one way to use what we observe to reflect on social science
     theory. 


## Summary


## References