What is the icc method

The “ICC method” usually refers to the **Intraclass Correlation Coefficient (ICC)**, a statistical method used to measure **reliability or agreement** when: - You have **multiple measurements** of the same thing (e.g., repeated tests, multiple raters, multiple devices), and - You want to know **how consistent** those measurements are. --- ## What ICC Measures ICC answers questions like: - *“Do different raters give similar scores?”* - *“Are repeated measurements on the same subject stable?”* It quantifies this on a **0–1 scale**: - **< 0.5** → poor reliability - **0.5–0.75** → moderate - **0.75–0.9** → good - **> 0.9** → excellent (Thresholds vary by field, but this is a common rule of thumb.) --- ## How It Works (Conceptually) ICC is based on **variance components**: - Total variance = variance **between subjects** + variance **within subjects** (error / rater differences). - ICC ≈ variance_between / (variance_between + variance_within) So: - If most variance is **between subjects** → high ICC (measurements agree well). - If a lot of variance is **within subjects** (raters disagree, unstable over time) → low ICC. This is usually estimated via an **ANOVA model** (one-way or two-way). --- ## Common ICC Types You’ll often see notations like **ICC(1)**, **ICC(2)**, **ICC(3)**, sometimes with “single” vs “average” and “consistency” vs “absolute agreement”: - **One-way random (ICC(1))** – subjects rated by different raters chosen at random; not all raters rate all subjects. - **Two-way random (ICC(2))** – all subjects rated by all raters, raters considered random sample from a larger population. - **Two-way mixed (ICC(3))** – all subjects rated by all raters, but raters are fixed (you care only about these raters). Plus: - **Single measures** – reliability of **one** rater/measurement. - **Average measures** – reliability of the **mean of several** raters/measurements (usually higher ICC). --- ## Simple Example Suppose 3 analysts each rate 20 companies on a risk score from 1–10. You compute ICC: - If **ICC = 0.85** → their scores are **highly consistent**. - If **ICC = 0.30** → their scores are **not very reliable**; who rates the company matters a lot. --- To fine-tune the explanation: Are you asking about ICC in **statistics/research (reliability)**, in **trading/finance**, or in some **other field**?