By Jon Bergdoll
Hi folks. I’m Jon Bergdoll, statistician here at the school, lead analyst for Giving USA, the Philanthropy Outlook, and several other projects and reports the Lilly Family School of Philanthropy has generated over the years. We’re trying something a little new and different here, so let us know if it strikes your fancy.
I’m going to go back to one of our recent reports and highlight a finding from it that I thought of as particularly interesting, and do a bit of a deeper dive into it; why we looked into it, what we found, and why it matters.
For this post, we’re going to be looking at our Women’s Philanthropy Institute (WPI)’s 2020 COVID-19 report and recap penned by the author Tessa Skidmore, and the interaction among age, giving, and COVID spread.
To briefly summarize the report: we fielded a survey early in the pandemic (around May 2020) asking households how the pandemic had impacted their philanthropic activities and in what ways. The main findings of the report, written by Skidmore with analysis by me, were that while a majority of Americans engaged in some philanthropic activity that spring, households were more likely to report their charitable giving had decreased rather than increased.
Giving during the pandemic
Single women seemed to be feeling the brunt of the pandemic more strongly than others—an early piece of evidence pointing to the “she-cession” brought on by COVID and its potential impact on the nonprofit sector.
When we’re looking into potential findings, one thing we always do to check the validity of what we’re seeing is to design a basic demographic-based regression model. This allows us to help isolate the effect of specific variables. For instance, married households typically have a higher household income than single women, so it’s not surprising to see them donating more, and more often.
In this example, to get a better idea about their behavior, excluding the effect of income, we can control for gender and marital status along with income. That way, we’d be looking at whether or not a single woman would be more likely to give than a married couple if they both had the same household income and other demographic characteristics (in fancy economics speak, this is called ceteris parabis, but seeing as I had to Google that to confirm the spelling, this is not a phrase you need unless you want to show off).
Given how often the same demographic controls are used in research, over time, you get a pretty good sense—both from your work and that of others—as to what demographics point in what direction with basic philanthropic variables. And one of the strongest, most common ones is age; older households tend to give more often and give a larger amount. This is why, looking at our regression models for giving for COVID relief, it struck me as so odd that age had a negative effect.
Exploring the variables
Then there were the COVID impact variables. There’s evidence in the disaster giving literature that the more closely tied someone is to a disaster—particularly by geography—the more likely they are to donate in relief.
Given this, I thought it would be relevant to test some variables on COVID’s impact at a state level for these respondents (I ended up using measured cumulative cases per capita when the respondent took the survey). Oddly, though, similarly counterintuitive as age – no effect.
Doubly strange, right? I wondered if these two oddities were interrelated, and so I designed and ran a model that included the interaction of the two. That is, a variable that combines the measurements of age and COVID impact, letting us see if there is a relationship between the two with giving alongside the more direct relationships of the two with giving.
And it popped. When interacted, we see negligible and positive coefficients on age and COVID impact (respectively) with giving for COVID relief, which is closer to what’s expected, and a strongly negative relationship on the interaction effect. In other words, older folks were likely to give about as much, and people in more affected areas were more likely to give more, but older people in more affected areas were likely to give less.
What does this tell us? Well, on its face, nothing beyond what that last sentence says; the variables had the relationship to giving that theory would’ve predicted. Still, they had an interesting wrinkle in their relationship with each other. This wrinkle was causing our initial models to show counterintuitive results when not controlled for, which led to us exploring it in the first place.
Our functional hypothesis as to the why of this has to do with risk: individuals who felt more at risk due to COVID—here captured as older households in heavily affected areas—might have felt more risk to their personal lives and felt less able to give charitably as a result.
If this were the case—and it’s only a hypothesis and a challenging one to test at that—it impacts how we view certain donations.
While disaster donations are typically related to proximity to the disaster, this could be because it’s rarely people in the disaster being surveyed. For these large-scale disasters—yes, the pandemic, climate change, but potentially other encompassing disasters like the California and Australian wildfires of recent years—it could be that proximity to the disaster is only associated with philanthropic action up to the point of personal risk.
Jon Bergdoll is the applied statistician in the Indiana University Lilly Family School of Philanthropy’s research team. He directs and conducts analyses for many of the school’s research projects.