Analysis via Simulation of Interrupted Time Series (ITS) Data

Uses simulation to create prediction intervals for post-policy outcomes in interrupted time series (ITS) designs, following Miratrix (2020) . This package provides methods for fitting ITS models with lagged outcomes and variables to account for temporal dependencies. It then conducts inference via simulation, simulating a set of plausible counterfactual post-policy series to compare to the observed post-policy series. This package also provides methods to visualize such data, and also to incorporate seasonality models and smoothing and aggregation/summarization. This work partially funded by Arnold Ventures in collaboration with MDRC.

The simITS Package

This package is for analyzing ITS designs with parametric simulation and extrapolation

Active code is on github, with clean releases posted to CRAN.

How to learn how to use this package

I would recommend starting by reading the methods overview paper, "Using Simulation to Analyze Interrupted Time Series Designs", on ArXiV https://arxiv.org/abs/2002.05746.

There is also a vignettes in the package that walks through almost everything.

Disclaimer

The post-stratified ITS code is both messier and less developed. There are multiple ways one might adjust for time-varying covariates, and this is an active area of ongoing research. Buyer beware!

Making contributions

You are most welcome to check out a copy from github https://github.com/lmiratrix/simITS and play with the code. All suggestions and improvements most welcome!