# DAGs and Model Building Flashcards

1
Q

What is a DAG?

A

Directed acylic graph.

2
Q

What does a DAG allow us to do?

A

Draw out pathways between different variables to see which variables we need to account for.

Allow us to think through potential biases in exposure and outcome causal relationships.

3
Q

How is a backdoor represented in a DAG?

A

An arrow that feeds into the exposure. We need to hold it constant or adjust for it.

4
Q

What is a node in a DAG?

A

Letters that represent variables.

5
Q

What is an egde/arc in a DAG?

A

An arrow.

Indicates a hypothesized relationship between two nodes (variables). The direction of the arrow is the hypothesized direction of the relationship.

6
Q

What is a path in a DAG?

A

A path in a directed graph is a sequence of edges having the property that the ending vertex of each edge in the sequence is the same as the starting vertex of the next edge in the sequence. Can be directed or non-directed.

7
Q

What is a directed path in a DAG?

A

Indicates a causal relationship; edge with single arrow.

8
Q

What is a non-directed path in a DAG?

A

Edge without an arrow. Non-causal association between variables.

9
Q

How do we depict a blocked path?

A

We adjust for variables by putting a square around it.

10
Q

What is a front door path?

A

Path leaving exposure going to outcome.

11
Q

What is a backdoor path?

A

Enters exposure then to outcome.

12
Q

What is a collider in a DAG?

A

A variable that is a common effect of two variables:

X->Z
Y->Z
Z is a collider

Do NOT adjust as backdoors are blocked by colliders and otherwise a spurious relationship would exist. Introduces bias if adjusted for.

13
Q

What is the difference between model selection in epi versus stats?

A

In stats, try to find the best fit as most accurate description of data.

In epi, test differences in outcomes while minimizing bias.

14
Q

What is forward selection?

A

Start with minimum key variable and add one at a time based on statistical significance of model.

15
Q

What is backward selection?

A

Start with all possible predictors and drop one at a time based on statistical significance.

16
Q

What is best subset selection?

A

Fit all possible models and choose best fitting model containing specified number of variables.

17
Q

What is stepwise regression?

A

Modifcation of forward selection that incorporates checking statitistical significant of all variables in model as new variables are added.

18
Q

What are the steps for fitting epi models?

A

1) Variable specification
2) Interaction assessment
stratify analysis and see if Breslow-Day significant (tests difference in OR in each strata)
3) Confounding assessment
Pooled MH different from crude by 10%?