2.1. Process for hand-collecting data

Each school’s website is unique in the way they document their course catalogs and graduation requirements, and the inclusion of personal finance content in a course is not always straightforward. Thus, the data construction requires hand-coding. While imperfect due to potential human error, hand-collected data allow researchers to pinpoint the information needed. The first step in the process was obtaining a list of all schools in the country. ^{Footnote 1}

The construction of hand-collected data requires a rigorous process. Each year’s data collection process started with a team of undergraduate research assistants supervised by one full-time research associate. The research associate trained each research assistant on the data collection procedure. Each research assistant chose a small state to start with, and upon completion, went through their data collection and all questions with the research associate (and/or Dr. Urban) to ensure the methodology was consistent. After each state’s completion, the research associate went through the data for an additional spot check. Finally, once all states were completed, Dr. Urban compiled all the state data and identified any oddities (e.g., large changes over time, states that did not match their policies). At this point, the research associate re-evaluated these states for a final quality control exercise.

Research assistants searched all school websites on the master list for a course catalog and graduation requirements. They excluded schools where no course catalog could be found or where online course catalogs were locked at the time of access. Appendix Table A1 compares schools that are and are not in our course catalog data. The excluded data are more disadvantaged on some dimensions: they serve students with higher eligibility for free and reduced-price lunch (FRPL), lower median household incomes, lower median home values, and higher fractions of Black and Hispanic students. However, the missing schools have lower student:teacher ratios and similar expenditures per pupil compared to schools in our dataset. Our sample includes fewer city schools and more rural schools.

If a catalog was not current (and it was still the only one on the website as of January 2023), they assumed the standards had not changed and referenced the previous year for classes offered. If a class including financial literacy topics was discovered in the catalog, researchers then recorded the class name, description (when available), duration (when available), whether it was a standalone course or a class that embedded financial literacy topics into another class, and whether or not it was a graduation requirement. ^{Footnote 2} The course was either labeled “required” or “as a cluster,” meaning it was one of many courses that could be taken to fill a graduation requirement. If a course catalog was found, but a course with financial literacy content was not offered at the school, this was classified as having no offerings. At times, there were course catalogs that did not include financial literacy (or typical classes that included financial literacy) but referenced electives that were not listed. We labeled these schools as “missing,” as a clear distinction could not be made. If a standalone course in personal finance was listed in the course catalog, but there was no further description of requirements, we coded it as an offering but not a requirement.

From this hand-collected dataset, we create school-level financial education course offering standards. We assign each school its “maximum” standard using the following sequencing: standalone requirement, embedded requirement, standalone offering, embedded offering, and no requirements or offerings. We append the data across years, spanning the 2019–2020, 2020–2021, 2021–2022, and 2022–2023 academic years.

2.2. Auxiliary data sources

To determine the correlates of school course offerings, we collect supplementary data. We then merge these data with the hand-collected high school-level data. We employ data from the US Department of Education’s NCES Non-Fiscal School Survey. ^{Footnote 3} Relevant to this study, these school-level data include the following variables: student:teacher ratios, the fraction of students eligible for FRPL, and student demographic characteristics (e.g., percentages by race). NCES data also include indicators for the geographic area in which each school is located. ^{Footnote 4} The data are categorized into four major groupings: city, suburban, town, and rural. In our analysis, we group town and rural into one category, as both represent more remote areas.

2.3. State policies

While state personal finance policies can take many shapes, there are two main policies that exist for financial education in high schools. Most research bundles these two policies together when estimating the causal effects of required financial education on downstream financial outcomes. The policies are as follows:

• eight states require every student to complete a semester of personal finance prior to high school graduation;

• 25 states require every student to complete personal finance content within another required course or content area (e.g., Economics and Social Studies).

The state-specific data on financial education requirements can be found at https://osf.io/ksah9/. Figure 1 shows the evolution of state-level high school personal finance policies by graduation year. No states required the graduating class of 2000 to take a standalone personal finance requirement, and only four states required embedded content for this cohort. Both policies dramatically expanded over the following 23 years. The graduating class of 2007 was the first to have a standalone course requirement, and by the class of 2023, eight states required every student to complete a standalone personal finance class, and 33 states required at least some personal finance content for high school graduation.

Figure 1. States with financial education requirements over time.

Notes: Any requirement represents states require personal finance content within a standalone course or within another class for high school graduation. Standalone course requirement represents states where a full semester standalone personal finance class is required for high school graduation.

The following states require every student to complete a semester of personal finance prior to high school graduation: AL, IA, MO, MS, NC, TN, UT, and VA. According to the course catalog data, nearly every school in these states has standalone personal finance requirements in place. These states are commonly referred to as the “guarantee states,” as all students are guaranteed access to personal finance education.

Additional states require students to complete personal finance content within another course or content area.

3.1. Equity in access

Table 1 next explores which schools are likely to have personal finance required. These models (Equation (1)) account for differences in school poverty levels, measured by the fraction of students receiving FRPL, the fraction of non-White students, and the geographical location of the school. The dependent variable (Req’d) alternatively equals 1 if school s in state j requires a standalone personal finance course or personal finance content at all and 0 otherwise. The independent variables of interest are whether the school has above median students receiving FRPL (AM %FRPL), has above median White students (AM %White), is in an urban area, or is in a rural area. The comparison geography are suburban schools. In some models, we also included state-level fixed effects ( $_j$ ):

(1) $${\rm{Req'}}{{\rm{d}}_{sj}} = {\alpha _0} + {\alpha _1}{\rm{AM\% FRP}}{{\rm{L}}_{sj}} + {\alpha _2}{\rm{AM\% Whit}}{{\rm{e}}_{sj}} + {\alpha _3}{\rm{Urba}}{{\rm{n}}_{sj}} + {\alpha _4}{\rm{Rura}}{{\rm{l}}_{sj}} + \left( {{\gamma _j} + } \right){\varepsilon _{sj}}$$

Table 1. Predicting school financial literacy policies

Notes: Coefficient estimates reported with standard errors in parentheses. * $p \lt 0.10$ , ** $p \lt 0.05$ , *** $p \lt 0.01$ . Each outcome variable equals 1 if the school had a standalone personal finance course requirement (odd columns) and any personal finance content required (even columns). Results for states without policies are depicted in the first four columns, and results for states with embedded requirements are in columns 5 through 8. $ \gt 53$ FRPL equals 1 for schools that have over 53 percent of students receiving free or reduced-price lunch; 53 percent is the median. $ \lt 69$ White equals 1 for schools that have fewer than 69 percent of students that are White, non-Hispanic; this is the median level, and the dummy represents schools with greater fractions of non-White populations. Urban and rural are dummy variables (the excluded group is suburban). Data are from the 2022–2023 academic year. Columns (3), (4), (7), and (8) include state-level fixed effects.

The first four columns include only states that have no personal finance policies. Columns (1) and (2) show that schools with above median students of color are 13 and 14 percentage points less likely to be in a school with a personal finance standalone course requirement or a school where personal finance is required at all. While there is no difference in access based on school poverty levels after accounting for racial/ethnic differences, as well as geography, this is likely because rural students are more likely to have access to either school requirement than urban or suburban schools. Gagnon and Mattingly (Reference Gagnon and Mattingly2016) find that rural schools are less likely to offer advanced placement (AP) courses, which could potentially mean that rural schools are more likely to focus on courses related to life skills than college-level coursework.

We next refine the model to only make within-state comparisons in Columns (3) and (4). This tells us if the differences are more likely to be across states or within states. In these models, the lack of differences across school poverty remains, and the differences across rural and urban schools also remain. While schools with above median non-White populations still have less access, the magnitude of the association falls by a third.

Columns (5)–(8) predict access in states that require personal finance content to be integrated into another class or content area. Columns (5) and (6) are the cross-state comparisons. In states with embedded requirements, students in higher-poverty schools have less access to standalone courses but greater access to any requirements. Columns (7) and (8) show that these differences disappear when we only make within-state comparisons. Schools with above median proportions of non-White students are 3.5 percent less likely to have a standalone personal finance course requirement; this does not hold up to within-state comparisons (Column (7)). While students in rural areas are more likely to have access to standalone personal finance guarantees, the magnitude is less than half that in states with no policies.

Overall, embedded state policies reduce the inequities in access that exist when no state policy exists, though some inequities based on geographical area still remain.

5.1. Implications for research

Many empirical studies in social science – especially economics – rely upon quasi-experimental methods to detect the causal effects of policies. These quasi-experimental methods often hinge on comparing a treatment and control group before and after a policy went into effect, referred to as a difference-in-difference strategy. This method is particularly popular among studies in the United States estimating the effects of required personal finance education state policies on credit, debt, and schooling outcomes. The findings from these studies rest upon several assumptions about the timing of and the actual take-up of policies locally. The local course catalog data can help to inform the relevance of these assumptions and how they may affect the magnitude of effects in causal studies.

The implementation of personal finance coursework is not perfect, which can understate the effectiveness of personal finance education. Because states without personal finance education policies still contain schools that require personal finance coursework locally, researchers may report a dampened effect of the policy on financial outcomes. Recall that in states where personal finance is required within another subject, only 39 percent of students are in schools where personal finance content is required. Since slippage exists in these states, an intent-to-treat (ITT) estimate will accurately state the effect of the policy, while a treatment-on-treated (TOT) estimate will instead produce the effect of financial education on outcomes.

Being explicit about the difference between ITT and TOT estimates remains important for the implications of research. For example, Stoddard and Urban (Reference Stoddard and Urban2020) find that requiring financial education in high school coursework increases the likelihood a first-time four-year student has a grant by 3.3 percentage points. While that is the effect of the policy (the ITT estimate), the education is likely to be 2.5 times more effective (the TOT estimate) once we account for the lack of implementation in some schools – an effect size of 8.25 percentage points. Further, Harvey (Reference Harvey2020) finds that requiring personal finance education in high school reduces the use of payday loans by 3.9 percentage points (the ITT estimate). Scaling the ITT estimate up to account for the lack of universal take-up suggests that education could have reduced payday lending by 9.75 percentage points (the TOT effect). These TOT effects are more easily comparable to other large-scale randomized-controlled trials where schools randomized into the treatment group implemented some level of financial education, such as Frisancho (Reference Frisancho2020, Reference Frisancho2023a, Reference Frisancho2023b). Being clear about the take-up of policies – and which schools are being left behind – remains a critical part of financial education research.

Further, anticipatory effects, such as those in North Carolina, Iowa, and Nebraska, could also dampen causal estimates by shifting some schools to treatment status before the researcher counts them as “treated.” Using a donut approach that omits years between policy announcement and policy execution would help to more cleanly estimate the effects.

An additional potential concern brought to light by recent research on two-way fixed-effects methods is that comparing always-treated to newly treated states or already-treated to newly treated states can bias estimates (Goodman-Bacon, Reference Goodman-Bacon2021). Recent research in Urban (Reference Urban2023) shows that 95 percent of the variation comes from a comparison of newly treated and never-treated states when looking at guarantee states and never-treated states. ^{Footnote 8} Further, early states could have cumulative effects that made them bad comparisons (Callaway and Sant’Anna, Reference Callaway and Sant’Anna2020; Sun and Abraham, Reference Sun and Abraham2020). However, early financial education graduation requirements had fewer resources devoted to them and were likely less effective than more recent graduation requirements. Early states may also be a bad comparison since their personal finance content was in a different form and more likely to be housed within a consumer economics course. For robustness exercises and in an abundance of caution, researchers should drop early implementers or use methods developed by Sun and Abraham (Reference Sun and Abraham2020); Callaway and Sant’Anna (Reference Callaway and Sant’Anna2020) to make sure results remain consistent.