Introduction

In safety net programs, different rules apply to people in different circumstances. Whether someone has stable housing, is a student, or is self-employed makes a big difference in determining eligibility and benefit amounts. Yet the definitions of these circumstances can be tricky to assign to applicants. Even when a definition appears crystal clear, a caseworker must decide whether it applies to a particular instance. This step requires human interpretation, which might be influenced by knowledge of rules, precedents, local practices, and even psychological factors like priming (perhaps, for instance, they’ve been reading the news about whether Uber and Lyft drivers ought to be considered employees).

In the CalFresh program, if a caseworker determines an applicant is self-employed, 40% of their income is automatically deducted as expenses (and more can be deducted if they itemize). That means they are not only more likely to be eligible, they will receive more money for food each month.

Many people who apply for CalFresh are trying to make ends meet by cobbling together gigs, side jobs, or occasional work. Increasingly, they might drive for ridesharing companies, deliver food or goods, or do remote or web-based freelance work. Even when people are self-employed for the purpose of CalFresh (see Figure 1), they may not think of themselves as self-employed.

Though GetCalFresh launched over four years ago(and has grown from a several-county pilot to a statewide service) we are constantly looking for ways to iterate and improve upon the product to help more people put food on the table—and move closer to our north star goal of closing the SNAP participation gap in California. In this post, we’ll describe how we improved our application to help people recognize whether they are self-employed for the purpose of CalFresh and to verify that income.

We began with user-centered research about how self-employed people think and talk about their work. Based on these findings, we first revised the language and examples around self-employment. Second, we created a template to help applicants self-attest about their self-employed work. Finally, we validated that both potential changes would help applicants through two experiments. Overall, we sought to answer two questions:

  1. Can we get more people who are self-employed for the purpose of CalFresh to describe themselves as such?
  2. Can we encourage self-employed applicants to submit a self-attestation with the application?
Figure 1: Considerations for determining self-employment in California counties
Adapted from San Diego’s CalFresh policies here.

Mapping everyday language to CalFresh regulations about self-employment

We started by mapping the language people use to describe their work to the criteria that eligibility workers use to determine if someone is self-employed (Figure 1). We began by interviewing three eligibility workers to learn about how they process applicants with self-employment income, what information they need to know, and what they ask clients to submit.

We next conducted a visual analysis of hundreds of earned income verification documents submitted across California during one month. These included signed letters from the payors, receipts from check deposits, to 1099 tax forms, screenshots from mobile app payment summaries, and handwritten declarations of income. This research had two benefits: 1) it left us with a sense of the most pertinent examples of work that counts as self-employed for GCF users, and 2) we had a sense of the ways they try to verify that work.

We also wanted to improve how we refer to self-employment itself. Many people didn’t consider themselves self-employed if they hadn’t incorporated a business, for example. To explore what meanings people associate with the terms we might use to describe self-employment, we ran an online card sorting exercise.

We asked more than thirty low income participants (some who are familiar with self employment and some who are not) to tell us what a variety of related words mean to them (Figure 2). We asked them 1) “If you had to explain this term to a friend, what would you say?” And 2) “Can you think of any jobs or professions that might fit in this category?”

Figure 2: Self-employment terms tested during online card sorting

We learned that self-employment is too confusing a term, and suggests to people an autonomy they don’t feel. We decided to lead with the words freelance or independent when referring to the CalFresh category of self-employment.

Learning how to guide self-employed applicants in verifying their work

Itemizing self-employment expenses is quite difficult, so counties typically ask individuals to submit a signed letter, sometimes called a self-attestation, declaring how much they make in their self-employment work. That gets them the 40% deduction from their gross income (they would need to itemize to get more).

Typically, people who are self-employed will have to wait until their interview to find out that they need to submit this signed letter. This increases the back and forth between the county and the applicant, which we know makes it more difficult to complete the enrollment process.

We studied rules and practices around self-attestation by talking to social services caseworkers and CBO outreach partners, and by reading regulations and county handbooks. For instance, in one interview, a case worker told us:

“The client just has to write it down: I go over it with them, ‘your name, employer (themself), hours per month, income for the month (after expenses).’ If they just write it down, we'll take that as their self-employment.”

Translating this information into user-friendly and understandable screen designs took several days of writing and editing. An ongoing challenge in our design work is turning what is normally done in a two-way conversation (with a caseworker, or a community assister for example) into a static screen. Our process involved getting a lot of feedback from a variety of people, consulting the regulations, consulting the research that we had done, and a lot of wordsmithing and layout work to make the content as clear as possible.

Testing user-centered designs through experiments

Once we translated the findings from our research into new language and screens for self-employment, we needed to ensure that these changes were beneficial to our users by implementing each as an experiment. One experiment would determine whether we were able to get more self-employed people to describe themselves as such in their application (the “Description Experiment”). The other experiment would determine whether we could encourage more people to submit self-attestations about self-employment income with their application (the “Attestation Experiment”).

Description experiment

Our Description Experiment examined changes to the screens that ask people to indicate whether they are self-employed (Figure 3). We defined success as at least 5% more applicants who have earned income describing themselves as self-employed. Put another way, among applicants with earned income, at least 5% more of them would be starting their interview having already identified themselves as self-employed.

Figure 3: Control and variant screens for self-description experiment

Results from new self-employment language

The new terms and examples we implemented based on our user research led to a 6.4 percentage point increase in applicants with earned income describing themselves as self employed (p<.001). Among applicants in the who saw the existing screens, 19.4% described themselves as self-employed, compared to 25.8% of applicants who saw the new screens. This gap suggests that previously, nearly 1 in 3 people who could be considered self-employed did not indicate this on their application (of course, it’s ultimately up to the eligibility worker to determine who is self-employed).

Figure 4: Applicants who saw the new self-employment screens were one-third more likely to say they are self-employed

Attestation Experiment

Our attestation experiment examined the effects of a guide for submitting a self-attestation about self-employment income. We defined success as at least 10% of users who see the self-attestation guide submitting a self-attestation. In practical terms, that would mean that at least 1 in 10 self-employed applicants has verified their self-employed income before the interview.

Figure 5: Self-attestation guide

22.0% of 2,002 people who identified themselves as self-employed submitted a self-attestation. We more than doubled our success target of 10%.

Importantly, these individuals were not merely substituting a self-attestation for another verification they would have otherwise submitted (see Table 1). Instead, they largely added to the verifications submitted. If we compare use of any job verifications among self-employed applicants, we find that 32.6% of people who saw the self-attestation guide submit at least one type of job-related verification (a self-attestation or other proof of earned income), compared to 28.7% of people who did not see it, a 13.8% relative increase (p<.01). In other words, even considering usage of existing relevant verification options, we see a large increase in people who verify their work at all. Meanwhile, if we compare the total number of earned-income related documents submitted by self-employed people, we see that people who saw the guidance submit an average of 28.0% more than did those who did not see the guidance (p<.001., mean was .62 without the guidance and .79 with the guidance).

Table 1: Verifications among self-employed applicants in the attestation experiment

Monitoring and mitigating potential harms

As with other changes in which we encourage applicants to do something differently, we needed to ensure that the new behavior did not harm their chances of receiving benefits. To ensure users weren’t being harmed by the changes, we did four things:

  1. We reviewed self-attestations and other income documents submitted by people in both variants the day the experiments went live. Based on the work-related documents people were submitting, we didn’t have any doubts about their self-employed status.
  2. A month later, we looked at the handful of cases where people had submitted a self-attestation with the application and then submitted it again. We found no evidence that the original self-attestation was invalid. There was some evidence that some caseworkers wanted to receive this information on a county form, so we will continue to watch for that.
  3. We tracked application completion rates for all groups. There were no meaningful differences between all the groups in terms of application completion (in fact, the only differences were after the decimal point).
  4. We monitored early outcomes to make sure that applicants seeing the new screens in either experiment were not doing worse than applicants seeing the original application. With over 1,000 outcomes per group in the self-attestation experiment, we saw no difference (a 0.7 point higher approval rate in the variant, not significant) between the variant and control groups. Similarly, with over 4,000 outcomes per group in the Description Experiment, we also saw no meaningful difference in outcomes (again, a 0.7 point higher approval rate in the variant, not significant).

Conclusion and recommendations

We found that, among people with job income, using everyday language and better examples increases the proportion who consider themselves self-employed by about one-third. We also found that providing people with a simple template for writing an attestation about their self-employed work led over one in five people to submit a self-attestation with the initial application.
If you’re thinking about designing a service in which you need to determine if people are self-employed:

  • Avoid using the term self-employment alone, use it with either freelance or independent contractor in order to catch self-employed people who work for companies or other individuals.
  • Survey people in the community for examples of low income, typically freelance jobs common to the area where a service is live. Use these examples in the copy.
  • Language changes, as does the work people do. Stay attentive to changes in the terms people use to talk about their work, particularly gig work.

If you’d like to learn more about who uses GetCalFresh and what kinds of work they do, take a look at SNAP Stories, where our users share their stories about why they applied for CalFresh and how it’s helped them. Check out other stories about our safety net work, or get in touch with us at research@getcalfresh.org.

Appendix

Reported statistics are based on two-sided chi-square tests for proportions and Welch’s t-tests for means.

Experiment design: As people went through the GCF application, everyone would have a 50% chance of getting into each experiment (Table A1). That meant about a quarter of applicants would be in both experiments, a quarter would be in the Description Experiment only, a quarter would be in the Attestation Experiment only, and a quarter would be in neither. Putting some people into both experiments would help us learn if there was any interaction between these two related changes. For example, are people who are prompted by different language about self-employment going to react differently to encouragement to submit a verification about that self-employment? As it turned out, we found no meaningful or significant interactions across the two variants.

Table A1: Design of overlapping self-employment experiments
 

Tags:   GetCalFresh User Research Data Science