This report uses a previous methodology
Click here to check more details.
Panel Methodology
We removed organizations that did not have a minimum of 25 donations and $5,000 in revenue in each of the previous three years. We removed organizations at either tail of the revenue growth curve. If revenue growth was more than 300% or less than -66% in any of the past three years, organizations were removed. We also excluded individual contributions above $10M.
Each year we analyze the organizations with stable growth and $5K – $25M total transactions in each of the three previous years, via the Growth in Giving database. This means that the YOY comparisons are comparing a different sample of organizations each year.
Weighting
We weighted our data by organization size and NTEE major group to make it reflective of 2018 IRS filers in the $5K-$25M range of contributions. Details regarding estimation methodology for late reported data can be found here.
Late Data Estimation
The top-line metrics in the FEP reports vary quarter over quarter, even when looking at the same time-periods in the past. A large cause of this variance is data that is reported late (also known as data drift), such as Q1 data which trickles in throughout the year. It is important to estimate and account for this variance (also known as “error”) since it can skew our view of the sector’s performance.
The following are the estimation of error to the top-line year-over-year metrics in the FEP report to give more representative and stable reporting on the state of the sector. Without doing so, the FEP report would consistently undershoot the true year-over-year growth of dollars, donors, and retention.
We can break down the top-line estimations of error into two components to model it in more detail. The first is data drift: fundraising data that belongs to a particular time period but is reported later. This is the largest and most important component to capture. The second is methodology error: variability as we tweak our methods. The data in the FEP report is processed to account for outliers and bad data. This preprocessing introduces an element of variance which will always exist, even if we do no preprocessing. If we just used raw data in our analysis, we would have very high methodology error due to noise and outliers.
Request more information about the methodology via email at [email protected].