DAGs Primer Resource

Directed Acyclic Graphs (DAGs) or causal graphs are useful for representing assumptions about the causal associations between exposures or treatments and their outcomes, as well as the confounders—observed and unobserved—that make causal inference in observational designs challenging to estimate.

The properties of DAGs and the Clock and Grid method are described in Rose, Cosgrove & Lee (2024) published in the Journal of the Society for Social Work & Research. On this resource page, we provide further details and tutorials about the Clock and Grid method.

What is a DAG?

These graphs consist of directed edges (arrows) connecting nodes (variables) to indicate the causal effects of antecedents on descendants. DAGs are generated largely from subject-area knowledge and theory, combined with logic, common sense (such as about time ordering), and study design considerations (for both retrospective and prospective research designs.) DAGs can represent known statistical phenomena such as mediation & over-control bias, sample selection, measurement error, and also provide credible explanations for numerous statistical paradoxes such as Simpson’s and Berkson’s.

A pivotal aim of DAGs is to identify the backdoor paths between the treatment and the outcome—the potentially numerous ways in which other variables exert confounding influence on the association between the treatment and the outcome, leading to biased causal estimates.

Deconfounding a DAG using the Clock & Grid Method?

Rules are applied to DAGs to identify the sufficient set (or in the case of multiple sufficient sets, a minimum sufficient set) that consists of the confounders that (in the context of regression adjustment) must be controlled for in order to estimate an unbiased causal effect of the treatment on the outcome. This process is referred to as deconfounding.

We developed a structured pencil-and-paper based method for deconfounding called the Clock and Grid method. The Clock part of this method is a structured way of tracing all backdoor confounding paths such that all of these paths are included in a numbered inventory. The Grid part of this method consists of rules applied to the inventory to identify one or more sufficient sets including potentially a single minimum sufficient set. Below, we provide further details on this method, including helpful videos.

  • Introduction to the Clock and Grid Approach

    Here Dr. Rose introduces the Clock and Grid method. The Clock and Grid approach makes it possible to construct an inventory of back-door paths while preserving the history of path already added to the inventory; and to use this inventory to identify the sufficient set or sets needed to deconfound the DAG that can facilitate how to identify a causal effect.

    Watch the Clock and Grid method Video

  • Demonstrating the Clock Rules

    The Clock part of this method is a structured approach to thoroughly tracing out all of the backdoor paths and entering them in an inventory, while preserving the history of paths already traced to help ensure that none are missed. On this page we describe the rules for tracing backdoor paths using the Clock rules, and provide a helpful video that demonstrates this process.

    Go to the Clock Rules

  • Demonstrating the Grid Rules

    The Grid part of this method is a set of rules used to identify the sufficient set or sets, including the minimum setvariables that should be conditioned on in order to estimate a causal effect. On this page we describe the rules, and how the rules can be modified and extended to handle challenging scenarios.

    Go to the Grid Rules

Examples of Deconfounded DAGs

ITFC DAG
The ITFC DAG is the example from the JSSWR article. It is a straightforward application of the grid rules. 

PLC DAG
The PLC DAG is an example that demonstrates a conditional approach to Step III that occurs because there is no row with only a single variable. Instead, the minimum number of variables on a row is a pair (this occurs twice), but the constituent members of each pair are not always grouped together, so they can’t be crossed out in pairs on all of the lines they appear on. Two strategies are used: 1. A conditional approach; 2. Ruling out, a priori, colliders. 

Knowledge of End-of-Life Wishes DAG
The Knowledge of End-of-Life Wishes DAG, developed by Sarah Clem, MSW, is an example that includes multiple sets of pairs in Step III, one of which includes a collider. This collider can be addressed in a couple of different ways, and in fact, in one of the sufficient sets identified, no further action is needed. 

Ready to be a Changemaker?