Do non-transactional data help predict whether people will give? This was the debate we unwittingly touched off when talking about SimioAudience, our new co-op that takes such data into account.
Enter Dr. Russell James III, whose Inside the Mind of a Bequest Donor, set the standard for planned giving research. (We’ve written about his work previously, discussing what language works best to get a gift in wills). He recently looked at Health and Retirement Study data to see what best predicted bequest giving. The top ten factors, in descending order of predictive power:
You can see two things going on here:
Transactional data is really important, holding the number one most predictive variable slot, along with numbers three and eight. This is data you want in your model with primary importance.
Other variables make the model better. Demographic data (#2, #5, and #7), wealth data (#4, #6, and #9), and behavioral data (#10) all make a difference even when transactional data is controlled for. Any model that ignores these data is going to be describing only one part of the elephant and giving you less accurate results than you could achieve with a more robust model.
In other words, giving data should be the start of giving models. It should not be the end.
It’s also important to note that this was only across charities giving and thus without charity-specific data. Thus, data that can be asked of the donor (e.g., loyalty to the organization, satisfaction with their donor experience) or appended en masse (e.g., cause connection to the organization) are often highly predictive also.
The data stew is richer when there are more ingredients in it. We miss out on a great deal of the donor, and the donor experience, if we look at donors only through their donations. They have a richer, fully realized life. And the more we can speak to that experience in our copy and learn from it in our data, the better off we will be.