**Principal Investigator(s):** Brian Guay Duke University Email: [brian.guay@duke.edu][1] Home page: [https://polisci.duke.edu/people/brian-m-guay][2] David Landy Indiana University Email: [dlandy@indiana.edu][3] Home page: [https://www.davidlandy.net/][4] **Sample size:** 2104 **Field period:** 07/13/2014-11/12/2014 **Abstract**: Numbers in the millions, billions, and trillions range are common in political science and public life, yet many American adults make large and predictable errors when estimating the magnitudes of these numbers. We test the effect of an established numerical training treatment on participants’ support for public spending on government programs. This experiment has two main aims: to determine how the public uses numerical information to come to policy decisions and to examine whether the public would change these decisions if they understood the relative magnitudes of large numbers. First, all participants were asked to complete a number line task that provides a baseline measure of participants’ numeracy, or understanding of the magnitudes of large numbers. Half of participants were randomly assigned to an experimental treatment that seeks to improve participants’ numeracy. Participants then received information about spending on government programs and were asked to assess whether the government should spend more, less, or about the same on these programs. **Hypotheses:** H1: We predict that participants with high levels of numeracy, and those who receive a brief numeracy training exercise, will be more supportive of government programs because they will perceive them as inexpensive relative to less numerate participants. H2: A participant’s political party and ideology will cause the effect of numeracy to matter less on support for spending on government programs. H3: Higher levels of education will predict greater numeracy, which is a relationship found with smaller numbers, but not yet with larger numbers. **Experimental Manipulations:** Numeracy Training Exercise: The standard training approach in psychological studies of number line use, often called “one-trial learning”, is to show a single correct numerical value on a number line. In our training, participants see the correct location of 10 million on a number line and read that 1 million is 1 thousand thousands, that 1 billion is 1 thousand millions, and that 1 trillion is 1 thousand billions. **Key Dependent Variables:** Number Line Task: The number line task is a standard assessment tool in psychological studies of small number comprehension. During the number line task, participants view a horizontal line marked with two numerical endpoints and are instructed to place a target number on the line. In this task, we included 0 and 1 billion as endpoints. Participants were asked to place four randomly ordered numbers (e.g., 280 million) on a number line ranging from 0 to 1 billion. **Spending Preferences:** Participants were given brief numerical information about four government programs in the form of news articles, and were asked whether the federal government should spend more, less, or about the same on these programs. The selected government programs (National Science Foundation funding for climate research, U.S. Customs & Border Patrol agency’s drone program, foreign aid, and military weapons systems) vary on the basis of whether conservatives or liberals traditionally support them, and the spending amounts include both millions compared to billions and billions compared to trillions. **Summary of Findings:** We find that acting linearly (accurately) on the number line task was associated with a shift in aggregate support for maintaining or increasing funding for the government programs. Accuracy on the number line task was moderately correlated with income, education, and gender. Still, when controlling for these variables linear behavior on the number line task predicts support for increased funding. [1]: mailto:brian.guay@duke.edu [2]: https://polisci.duke.edu/people/brian-m-guay [3]: mailto:dlandy@indiana.edu [4]: https://www.davidlandy.net/