What? I can’t hear you! This is what you might say when analyzing your data. All data has signal and noise associated with it. Let’s see the difference between them and understand what we mean by noise.

## Overview: What is noise?

It is accepted that there is variation in every process. The variation has been described as common cause (or random) variation and special cause (or assignable) variation. Random variation is a function of the variation of the process elements: people, materials, equipment, environment, and methods. It is not controllable but is expected and predictable. Special cause variation signals that something is not expected nor random and is an indication of something different happening in the process.

The relationship between signal and noise is often represented as the signal to noise ratio (SNR). This ratio shows the proportion of noise relative to signal. If the noise component is large, then the signal or information you are trying to learn from your data will be harder to detect.

The SNR is the key calculation in ANOVA. The concept is also prevalent in control charts. In both cases, the signal is referred to as between-sample variation, while the noise is described as within-sample variation. While noise can not be eliminated, it can be reduced by taking action on the source(s) of the noise.

## An industry example of noise

When using an Xbar and R control chart, you are primarily concerned with monitoring the change in your process over time. You want the Xbar chart to signal when something unexpected is happening. The calculations of the control limits for the Xbar chart are based on the amount of variation in the samples you collected. This variation is shown on the R chart.

If the variation of the samples is large, the control limits of the Xbar chart will be wide. This makes the Xbar chart less sensitive to signaling change in the process. Therefore, we refer to the R chart as the noise and the Xbar chart as the signal. If the noise of the R chart is large, it will mask the signal of the Xbar chart, and you may miss a change in your process.

### 1. Is noise in my data a bad thing?

Yes. Noise will mask the signal you are hoping your data will provide you. If your data-collection methodology has too much variation, your analysis will be hindered because of the uncertainty, or noise, that exists in your data.

### 2. What are the typical sources of noise in a process?

Noise can come from the natural and random variation in your people, raw materials, work methods, equipment, and work environment. It can also come from poor data collection or errors in recording the data.

### 3. Where did the concept of signal and noise come from?

The terms “signal” and “noise” come from radio engineering. When you turn the dial on your radio searching for your favorite radio station, you may hear two sounds. One will be the music you hear once you find the right station setting. That is the signal since you want to listen to your music. The second sound is the static or hissing you hear while you are turning the dial but before you find the station. That is the noise. You don’t really want to hear the static while driving – just the music.