R in a Nutshell, 2nd Edition

Data Cleaning

Even when data is in the right form, there are often surprises in the data. For example, I used to work with credit data in a financial services company. Valid credit scores (specifically, FICO credit scores) always fall between 340 and 840. However, our data often contained values like 997, 998, and 999. These values did not mean that the customer had really super credit; instead, they had special meanings like “insufficient data” or there might be duplicate records in the data. Again, suppose that you were analyzing data on patients at a hospital. Often, the same doctor might see multiple patients with the same first and last names, so multiple patients may be rolled up into a single record incorrectly. However, sometimes the same patient might see multiple doctors, creating multiple records in the database for the same patient.

Data cleaning doesn’t mean changing the meaning of data. It means identifying problems caused by data collection, processing, and storage processes and modifying the data so that these problems don’t interfere with analysis.