Overall Match Rate: This refers to the amount of records you receive from your data provider with respect to the number you submitted for enhancement. Enhancement is defined as the addition of information to an individual consumer record. For example, if you sent in a list of 1,000 of your customer names and the data provider returned data on 800 of those, you would have an overall match rate of 80%. This applies only to the total amount of records with data provided, not the amount of data appended to each record.
When comparing data providers, many companies find match rates to be an extremely important variable, especially in modeling. Low match rates may mean that the data provider does not have a large enough representation of your customer base to give you the information you need. An overall match rate of 70-80% should be expected as a minimum from most large providers.
Elemental Match Rates: This refers to the number of elements requested for each record versus the total number of elements appended to your file. Not all providers will be able to supply you with all of the elements you require. Conversely, some may have all of the elements present but return very few of them, i.e., there may be several blank fields.
When comparing data providers be sure that you are able to receive all of the data elements you request. A company providing a 100% match rate but returning only half of the elements you desire is probably not going to meet your needs.
It is also important to look at the average number of elements returned per record for the elements provided. A 100% overall match rate with a 50% elemental match rate implies that 1/2 of their database for this element contains blank fields.
Be aware that some companies measure elemental match rates as the ratio of elements appended to matched records. In the 1,000 record example above, they would measure an ordered element with 600 matches for a single element as a 600/800 (80% overall match rate). This would compute as a 75% elemental match rate. The real way to measure this number is 600/1000, or 60%. Don't be fooled by these inflated numbers.
Accuracy: At first glance accuracy would seem to be a simple thing to measure. Just pick 10 records at random then call those people and validate the information, right?
Unfortunately it is not that easy. In order to be statistically accurate your records should be chosen at random, be diverse, and of sufficient size (10 is usually not a sufficient size). In addition it is very important that the data is tested against a valid benchmark. See the data test section later in the paper for a more detailed explanation.
Other Measurements
There are more aspects to data quality than just numbers. How easy the data is to understand, interpret, use and format is often as relevant as data accuracy. It would do you little good to get 100,000 records of 100% accurate data if you were unsure what you actually bought.
Listed below are a few of the other important parameters you should be aware of when designing a data test.
Data interpretation and representation: Is a data dictionary available for the data elements? Is it readable and understandable? Does it provide you with an accurate representation of the data and where it came from (e.g., survey, warranty card, etc.)?
Data quantity: Was the match rate in each element high enough for you to use that element in modeling or list processing? What is the minimum number of returned elements that you will need for your modeling or lists? It is possible to buy data based upon overall match rates and price and not get what you need to fulfill your job.
