Inferring causal impact using Bayesian structural time-series models

By Kay Henning Brodersen, F. Gallusser, J. Koehler, Nicolas Remy, Steven L. Scott. 2015

879 citations recorded by the source corpus.

An important problem in econometrics and marketing is to infer the causal impact that a designed market intervention has exerted on an outcome metric over time. In order to allocate a given budget optimally, for example, an advertiser must determine the incremental contributions that dierent advertising campaigns have made to web searches, product installs, or sales. This paper proposes to infer causal impact on the basis of a diusion-regressi on state-space model that predicts the counterfactual market response that would have occurred had no intervention taken place. In con- trast to classical dierence-in-dier ences schemes, state-space models make it possible to (i) infer the temporal evolution of attributable impact, (ii) incorporate empirical priors on the parameters in a fully Bayesian treatment, and (iii) exibly accommodate multiple sources of variation, including the time-varying inuence of contemporane- ous covariates, i.e., synthetic controls. Using a Markov chain Monte Carlo algorithm for posterior inference, we illustrate the statistical properties of our approach on synthetic data. We then demonstrate its practical utility by evaluating the eect of an online advertising campaign on search-related site visits. We discuss the strengths and limitations of our approach in improving the accuracy of causal at- tribution, power analyses, and principled budget allocation.

Subfields: Marketing, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/bc8af8d2f13018b4e516f4ddcedf067f4184383e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Marketing

In Statistics and Probability

Citations in the corpus

Data Lifecycle Challenges in Production Machine Learning (2018, background): “Time series analysis [10, 27, 18] can help track these statistics.”
Reliable Decision Support using Counterfactual Models (2017, background): “Conversely, Brodersen et al. [2015] estimate the effect that a single discrete intervention has on a discrete time series.”
Chapter Six - Mutation Testing Advances: An Analysis and Survey (2019, background): “In software engineering, counterfactual reasoning plays a role in causal impact analysis [66], allowing software engineers to discover the impact of a particular event, thereby partly obviating the problem ‘correlation is not causation’.”
NBER WORKING PAPER SERIES BALANCING , REGRESSION , DIFFERENCE-IN-DIFFERENCES AND SYNTHETIC CONTROL METHODS : A SYNTHESIS (2016, neutral): “This may motivate timeseries approaches as in Brodersen et al. [2015], von Brzeski et al. [2015]. In other cases the magnitudes may be similar, T0 ≈ N , and the choice between strategies may be more difficult, and a regularization strategy for limiting the number of control units that enter into the”
D’ya Like DAGs? A Survey on Structure Learning and Causal Discovery (2021, background): “Method Keywords & Software causaleffect [233] general causality, R daggity [230] general causality, R dosearch [232] causal effect identification, R Causal Discovery Toolbox [121] causal discovery, Python pcalg [123] causal discovery, R bnlearn [211] causal discovery, R rEDM [180] dynamic modeling a”
Matching on Balanced Nonlinear Representations for Treatment Effects Estimation (2017, background): “Causal questions exist in many areas, such as health care [24, 12], economics [14], political science [17], education [36], digital marketing [6, 43, 5, 15, 44], etc.”
The Effect of Extremist Violence on Hateful Speech Online (2018, methodology): “Following existing techniques for causal inference on time series (Eichler 2012; Brodersen et al. 2015), we measure this effect by comparing the behavior of an observed time series (which we refer to as treated) after an event with a counterfactual time series of its behavior had the event not…”
The Augmented Synthetic Control Method (2018, methodology): “For each DGP we consider SCM augmented with (1) LASSO, (2) a random forest, (3) CausalImpact (Brodersen et al., 2015), (4) matrix completion using MCPanel (Athey et al.”
Using Synthetic Controls: Feasibility, Data Requirements, and Methodological Aspects (2021, methodology): “They include Bai and Ng (2019); Brodersen et al. (2015); Gobillon and Magnac (2016); Gunsilius (2020); Kennedy-Shaffe, de Gruttola, and Lipsitch (2020); Viviano and Bradic (2019); and Xu (2017), among many others. Samartsidis et al. (2019) study the performance of the canonical synthetic control est”
Causal Explanation Under Indeterminism: A Sampling Approach (2016, methodology): “Finally, the causal impact method of Brodersen et al. (2015) computes the effect of an intervention (advertising campaign) on a continuous variable (market response), but requires a speciﬁed alternative to the intervention rather than supporting general counterfactual queries.”
Editorial: Enhancing quantitative theory-testing entrepreneurship research (2019, methodology): “Additional approaches include propensity score matching and new Bayesian estimators, such as the causalImpact time series analysis package for the R language (Brodersen et al., 2015).”
What-If Reasoning with Counterfactual Gaussian Processes (2017, methodology): “The CGP builds on previous ideas in continuous-time causal inference (e.g., Robins 1997, Arjas and Parner 2004, Lok 2008), but is unique in that it can predict the full counterfactual trajectory ; we combined marked point processes (MPP) with GPs to model observational traces , and described three a”
Causal Falling Rule Lists (2015, background): “…segmentation of data into regions of differential treatment effect has been of recent interest in social science, medical, and marketing domains (Brodersen et al. 2015; Sun et al. 2014; Cai et al. 2011; Foster, Taylor, and Ruberg 2011; Taddy et al. 2014), precisely so that the relevant…”