Association for Public Policy Analysis & Management

Background: This work aims to encourage the saving of the tax refund through an experiment embedding behavioral interventions in a tax filing platform serving almost a million low- and moderate-income households. American households, and LMI households in particular, do not have enough savings for unforeseen drops in income or rise in expenses. Data from the Pew Charitable Trusts (2015a) found that 41% of households do not have enough liquid savings to cover a $2,000 expense; for low income families that rises to 78%. Financial emergencies are also commonplace: 60% of American households report experiencing a financial shock in the past year (Pew Charitable Trusts, 2015b). The tax refund offers a small period of relief from the low level of liquid funds that households have on hand and provides one of the few opportunities LMI households have to save.

Experimental Design: Researchers tested behavioral economics techniques in a randomized controlled trial within the TurboTax Freedom Edition, a free online tax filing system used by approximately one million LMI filers each year. Results are drawn from a dataset of highly accurate administrative tax data. In this experiment, 812,072 LMI tax filers were randomly assigned to one of five experimental conditions including: 1) a control condition (2) a treatment combining pre-commitment to save the refund with a choice architecture intervention that emphasized savings; (3) a tailored choice architecture intervention that promoted different savings options based on savings account ownership, i.e. depositing the refund to a savings account for tax filers with savings accounts, and purchasing bonds for filers who did not have a savings account; and (4) a combination of treatments #2 and #3. As a supplemental analysis, we also tested the impact of using a highly interactive pre-commitment intervention combined with choice architecture promoting savings on a subsample of the population. Overall savings impacts are assessed through an intent-to-treat approach, while the incremental impact of pre-commitment is assessed through a treatment-on-treated approach using 2SLS regression.

Results: The control group deposited their refund into savings accounts at a rate of 9.6% while the most successful treatment group—combining pre-commitment with a choice architecture emphasizing savings —deposited their refund into savings vehicles at a rate of 15.1% (p<.001). This represents a 57% increase over the control. Similarly, the control group deposited an average of $167 of the refund while the most successful treatment group deposited $253 (p<.001). The other experimental conditions increased the rate of savings deposits by 42% (experimental condition 3; p<0.001) and 44% (experimental condition 4; p<0.001). Additionally, we find that filers pre-commit to save their refund at higher rates when the pre-commitment decision has a higher degree of interactivity. Finally, we find evidence that pre-commitment itself is associated with increased savings rates.

Implications: This experiment serves as an example of how small, nearly costless changes to large systems can have an impact on thousands of people. These results can inform the design of similar initiatives in public, private, and nonprofit programs.