Semanlink - Causal discovery

Parents:

Causality

Causal discovery

ChatGPT says: **causal discovery** and **causal inference** are related but distinct concepts in causal analysis. Here’s how they differ:

### **1. Causal Discovery**  
- **Goal:** Identify causal relationships from data, often without prior knowledge of the underlying causal structure.  
- **Approach:** Uses statistical and algorithmic methods (e.g., PC algorithm, LiNGAM, Granger causality) to infer causal graphs or structures.  
- **Assumptions:** Typically assumes conditional independence, structural causal models (SCMs), or other constraints to make causal identification possible.  
- **Example:** Given observational data on sleep, caffeine intake, and productivity, causal discovery aims to determine whether caffeine causes better productivity or if both are influenced by sleep.

### **2. Causal Inference**  
- **Goal:** Estimate the effect of a specific intervention (e.g., treatment, policy change) on an outcome, assuming the causal structure is already known or hypothesized.  
- **Approach:** Uses methods like **instrumental variables (IV), propensity score matching (PSM), difference-in-differences (DiD), and causal Bayesian networks** to estimate causal effects.  
- **Assumptions:** Requires counterfactual reasoning, potential outcomes framework (Rubin Causal Model), or structural equation models.  
- **Example:** If a study wants to measure the effect of a new drug on blood pressure, causal inference helps estimate how much the drug (treatment) influences the outcome (blood pressure reduction).

### **Key Differences**  
| Feature | Causal Discovery | Causal Inference |
|---------|----------------|-----------------|
| **Objective** | Identify causal relationships | Estimate causal effects |
| **Data Type** | Typically observational | Experimental or observational |
| **Assumptions** | Often minimal prior knowledge | Requires known causal structure or strong assumptions |
| **Methods** | Graph-based algorithms, constraint-based, score-based | Potential outcomes, instrumental variables, matching techniques |
| **Example Question** | "What causes obesity?" | "Does sugar intake increase obesity rates?" |

In short:  
- **Causal Discovery** helps build a causal model.  
- **Causal Inference** helps estimate causal effects once a model is assumed or discovered.

Both are crucial in fields like economics, epidemiology, AI, and social sciences.

Related Tags:

Descendants

1 Document (Long List)

Aminata Elisabeth Ndiaye | LinkedIn

Tags:

2025-03-14 About

Properties

sl:creationDate : 2025-03-14
sl:creationTime : 2025-03-14T12:21:46Z
rdf:type : sl:Tag
skos:prefLabel : Causal discovery