Using Giving Variables in Data Mining

In fundraising, we seek charitable gifts. When we model prospects, we hope to find those who will give at a certain level. Because of that, we often debate whether it is valid to use a prospect’s giving history in a major gifts model, since doing so might be what we call, “co-linear”, or comparing oranges to oranges, so to speak.

There are two reasons to use giving variables in a prospect model. First, unlike any other part of a fundraising shop, gift records are audited. That means that they are the cleanest and most consistent of all the data you would use. Second, past behavior often indicates future behavior – past giving can show the pattern of future giving. Using giving history judiciously can improve your major gifts model well.

Let’s take a look at a few variables related to giving that many modelers use to distinguish donors from invested donors.

• Loyalty variables include the number of years given in the past 5 or 10, the number of years giving consecutive up to this year (consecutive years giving), and the number of years giving since the prospect became involved with your organization.
• Recency variables include last gift date and average giving over the last 5 years.
• Frequency variables include number of gifts and pledge payments over a period of time, percentage mix of new pledges vs. outright gifts, and whether a prospect has her gift deducted from a credit card on a monthly basis.
• Method variables include the campaign code, tender type, honorarium/memorial flag, or whether the prospect gave through a corporation rather than from personal funds.
• Movement variables include an up or down score (compare the number of years the prospect has increased giving vs. the number of years that he decreased or maintained giving) and a velocity measurement (giving this year divided by average giving over the previous 3 years).
• Time lapse variables include distance from record creation date to first gift date, distance from first membership date to first charitable gift date, distance from first gift date to highest gift date, and distance from first contact to first major gift date.
• Giving totals variables include dividing the donor’s life giving by the number of years in the database, average giving over the last 3 or 5 years, and percentage of highest gift to total life giving.

 
These are some of the ways that one can look at giving in order to find major gifts prospects. Note that any model that tries to predict giving this year still can not include independent variables that have giving this year in them. Be careful to distinguish the models. Feel free to try out some of these variables as a way of looking at current major gift donors as well.
 
Be sure to tweet us what you find @Staupell.

This blog post appeared in the APRA Upstate New York newsletter in the winter of 2010.