DAG: when should we use variables marked as "adjusted"? daggity.net allows to define variables as "adjusted". Manual gives the following definition for adjusted variables: for variables that have been adjusted for in a statistical analysis
In the example below, "adjusted" variables will be automatically shown in minimal sufficient adjustment sets when estimating a direct effect.

Confusion
Usually, DAG is the one that tells us which variables should be adjusted. However, when we use "adjusted" variables, we tell to DAG which variables certainly should be adjusted? Is this option for cases when we have certain prior knowledge that this ONE VARIABLE definitely needs to be adjusted, and DAG tells the rest of the variables?
Question
Could you please explain the need for marking variables as "adjusted"?
 A: In dagitty, when you indicate that a variable $A$ is "adjusted", you indicate that you will definitely adjust/control for it in the analysis. Dagitty will then tell you whether and how you can still estimate a causal effect of the variable of interest $E$ from this analysis via adjusting for additional variables or using an instrumental variable.
There might be different reasons for why you want to definitely adjust for a variable $A$:

*

*You control for it because you think it correlates with the outcome $Y$ and might therefore increase the precision (standard error) of your estimates.


*You are interested in effect modification, and want to interact the adjustment variable with the independent variable of interest $E$ (or run analyses in subgroups defined by the adjustment variable $A$).


*You are forced to control for it, because you only observe a subset of the population you are actually interested in, which could be conceptualized as a binary variables $A$ indicating sample selection (so that all your data is conditional on $A = 1$).
Furthermore, in your picture, dagitty is showing you the variables you need to control for when you are interested in the direct effect of the exposure on the outcome. In this graph, this is the effect that is left over if "adjusted" and "observed" are hypothetically fixed. Dagitty here indicates that under the assumptions in the graph, statistical control for "adjusted" and "observed" will give valid estimates of the direct effect of the exposure on the outcome.
