LocalControlStrategy (version 1.4)
Local Control Strategy for Robust Analysis of Cross-Sectional
Data
Description
Especially when cross-sectional data are observational, effects of treatment
selection bias and confounding are revealed by using the Nonparametric and Unsupervised
"preprocessing" methods central to Local Control (LC) Strategy. The LC objective is
to estimate the "effect-size distribution" that best quantifies a potentially causal
relationship between a numeric y-Outcome variable and a t-Treatment or e-Exposure
variable. Treatment variables are binary {either 1 = "new" or 0 = "control"}, while
Exposure variables vary continuously over a finite range. LC Strategy starts by
CLUSTERING experimental units (individual patients, US Counties, etc.) on their
X-confounder characteristics. Clusters represent exclusive and exhaustive BLOCKS of
relatively well-matched units. The implicit statistical model for LC is thus simple
one-way ANOVA. Within-Block measures of effect-size are Local Rank Correlations (LRCs)
when Exposure is numeric with (many) more than two levels. Otherwise, Treatment choice
is Nested within BLOCKS, and effect-sizes are LOCAL Treatment Differences (LTDs) between
Within-Cluster y-Outcome Means ["new" minus "control"]. An Instrumental Variable (IV)
method is also provided so that Local Average y-Outcomes (LAOs) within BLOCKS may also
contribute information for effect-size inferences ...assuming that X-Covariates influence
only Treatment choice or Exposure level and otherwise have no direct effects on y-Outcome.
Finally, a "Most-Like-Me" function provides histograms of effect-size distributions to
aid Doctor-Patient or Researcher-Society communications about Heterogeneous Outcomes.