Lapsed or Dormant? Leveraging Statistical Analysis in Fundraising

All three individuals are 10 years from their graduation dates but differ in the frequency and recency of their contributions. Whereas graduates 1 and 2 have both made two contributions to the annual fundraising drives and differ in regard to when their last donations occurred, graduates 2 and 3 both made their last contributions five years ago but differ in how frequently they had made donations.

Based on these donation histories, who is most (and least) likely to contribute in the future? This question can be reframed into a related question: Which of these donors is most likely to be dormant, and which has probably lapsed and will never donate again? Simple summaries (such as recency and frequency) provide some relevant inputs, but they fall far short of offering any specific recommendations. Upon closer inspection, graduate 3 is the most likely to have lapsed, since he donated consistently early on but has gone completely cold since. It is much more difficult to compare the seemingly sporadic pattern of graduate 1 with the “off and on” pattern of graduate 2.

Rather than trying to flag lapsed and dormant donors using casual heuristics, formal statistical analysis can be brought to bear on this question. A recently published paper by Peter S. Fader, Bruce G.S. Hardie and Jen Shang, “Customer-Base Analysis in a Discrete-Time Noncontractual Setting,” examined exactly these kinds of trade-offs using annual donation data from a major public radio station. It showed remarkable accuracy in predicting donation patterns over a five-year forecasting horizon. While the model has some complex math within it, the ultimate implementation is a relatively simple Excel spreadsheet, requiring nothing more than recency and frequency as inputs. Many of the results and implications within the paper run quite contrary to the usual intuition that many nonprofit organizations have about the behavioral propensities (and thus future contributions) of their past donors. For instance, the paper found that recency is far more important than frequency in determining estimates of future donation patterns.