Before the Great Recession, macroeconomists had chronicled a Great Moderation—a reduction in the volatility of the business cycle—achieved by the judicious use of monetary policy (Clarida, Galí, and Gertler 2000; Galí and Gambetti 2009). Now, in the wake of the most severe downturn and the slowest recovery in the postwar period, it seems that such talk was premature.

Given the massive human cost of recessions, it is incumbent upon policy makers to assess the policy tools at their disposal and identify those that are most effective at hastening economic recovery during a downturn. In this document we describe how different groups of workers were affected by the Great Recession, what works in fiscal stimulus, what could be done differently in future recessions, and the fiscal preparedness of states for the next downturn.

There are two sets of policy tools used to foster recovery following recessions: monetary policy and fiscal policy. Monetary policy, consisting of actions taken by the Federal Reserve, is used to keep interest rates low and reduce unemployment during and after a recession. Fiscal policy includes various forms of government spending and tax cuts enacted by Congress. Following a recession, both sets of policy tools can be used to increase demand, thereby raising output and more quickly returning the economy to prerecession conditions.

To be most effective, it is crucial that stimulus be expeditious. In the Great Recession, a portion of the fiscal policy response occurred automatically within preexisting programs. These programs are called “automatic stabilizers” because they provide immediate stimulus during a recession without requiring action from Congress. For this reason, automatic stabilizers like unemployment insurance (UI), the Supplemental Nutrition Assistance Program (SNAP, formerly Food Stamps), and Medicaid—in addition to automatic stabilization associated with tax revenue—are particularly valuable.

As soon as a recession arrives, participation in these programs expands as incomes fall and unemployment rises—and in some cases, participation increases because of automatically reduced eligibility requirements for participants. The result is that additional funds are automatically disbursed (or taxes reduced), immediately providing fiscal stimulus. The United States makes considerable use of automatic stabilizers, which amounted to about 2 percentage points of GDP during the depths of the Great Recession (Congressional Budget Office [CBO] 2016a).

Monetary policy makers were also quick to react. As the Great Recession began and GDP and employment plunged, the Federal Reserve reduced the federal funds rate. (The federal funds rate is the interest rate that banks charge each other on a particular kind of overnight loan.) This conventional monetary policy action aimed to lower borrowing costs for individuals and businesses, thereby encouraging both immediate consumption and investment. However, the ongoing very low federal funds rate—which cannot be lowered much farther—and the severity of the Great Recession prompted an increased focus on fiscal policy as a recession-fighting tool.

Fiscal stimulus that requires congressional action takes longer. In some cases it may take a considerable amount of time to observe that a recession has begun, debate the appropriate legislative response, pass legislation, disburse funds to states and individuals, and finally spend authorized funds. Such lags are perhaps inevitable, but quick delivery of stimulus is preferable for both the economy and the affected workers and families. The American Recovery and Reinvestment Act of 2009 (ARRA) was a major vehicle for such fiscal stimulus, authorizing spending on infrastructure, health care, and education; expanding automatic stabilizers; and making various tax cuts.

With unemployment and poverty spiking, two major programs—UI and Temporary Assistance for Needy Families (TANF)—responded in starkly different ways. UI claims shot up in response to the increasing number of the newly unemployed, buffering eligible workers against earnings losses. Not all of the newly unemployed were eligible for UI, but the program was broadly successful in achieving its mission. By contrast, the caseload of TANF—the successor to Aid to Families with Dependent Children—barely increased as poverty grew. Though TANF plays a role in alleviating deep poverty, it is not currently structured to respond effectively to cyclical poverty variation or to act as an automatic stabilizer.

The fiscal stimulus in ARRA is widely believed to have reduced the severity of the Great Recession (Chodorow-Reich et al. 2012; CBO 2015). By the CBO’s estimate, the fiscal stimulus bill caused GDP to be 0.4 to 2.3 percent higher in 2011 than it otherwise would have been (CBO 2015). But which components of the fiscal stimulus were most useful? Stimulus aimed at low-income or otherwise cash-constrained households tends to be more effective, whereas business tax cuts tend to be less effective (Whalen and Reichling 2015). By some calculations, government spending is typically a more effective stimulus than tax cuts, partially because workers tend to spend only a fraction of stimulus provided through the tax system. Across all stimulus types, it is widely believed that fiscal stimulus is more effective during recessions and less effective during expansions (Auerbach and Gorodnichenko 2012; Fazzari, Morley, and Panovska 2014), likely because downturns are characterized by slack in both labor and capital markets (i.e., available resources are not fully employed), allowing fiscal stimulus to increase total output.

Despite all of the monetary and fiscal actions taken to mitigate the severity of the Great Recession, output has still not returned to either its prerecession trend or to potential output (CBO 2016b). Unemployment has only recently recovered to a level close to the prerecession rate, and other measures of labor market slack, such as the fraction of workers who work part time for economic reasons, remain elevated (Bureau of Labor Statistics [BLS] 2016a). In addition, states are not adequately prepared for the next recession. In most states, rainy-day funds remain insufficient to cope with a large, unexpected decline in tax revenues. These states will either need to cut crucial government services—hurting both state residents and the broader economy—or rely on the federal government for support to maintain their budgets.

A founding principle of The Hamilton Project’s economic strategy is that long-term prosperity is best achieved by fostering economic growth and broad participation in that growth. To that end, The Hamilton Project offers the following nine facts about the Great Recession and the kinds of fiscal stimulus that can help mitigate the severity of future recessions.

Between December 2007 and June 2009, the United States experienced the most severe recession in the postwar period. Recessions are conventionally measured by declines in GDP, which measures overall economic activity and is defined as the value of the economy’s total output of goods and services. Figure 1a compares the depth and duration of the Great Recession with the four most recent recessions, plotting the changes in GDP from the first quarter of the recession to the quarter when GDP was the lowest relative to the beginning of the recession (bold lines) and the economy’s subsequent recoveries (light lines). Not only was the Great Recession (grey) deeper than any recent recession, but it look nearly four years for the economy to regain the prerecession GDP level—twice as long as for the 1981–82 recession. As discussed in The Hamilton Project jobs gap analysis, the labor market was slower to recover following the Great Recession than it was after previous downturns (Kearney and Hershbein 2015), and still remains weaker in May 2016 than it was before the recession (The Hamilton Project n.d.). This continues a pattern in which successive recessions have been followed by more-prolonged job-market recoveries—reflecting long-term structural changes in the economy as well as cyclical factors (Kearney and Hershbein 2015).

During the Great Recession, the nation’s actual GDP contracted by more than 4 percent. How much of a reduction was this relative to the economy’s hypothetical capacity, or “potential GDP”? Potential GDP refers to the economy’s maximum sustainable output, which mirrors GDP under normal conditions and to which actual GDP is expected to return quickly after a downturn. Today—more than six years after the recession ended—economic output remains substantially below potential GDP. As shown in figure 1b, the gap between actual and potential GDP reached its widest point in the third quarter of 2009, when the economy fell short of potential GDP by $1.2 trillion in constant 2015 dollars, or 7 percent of potential GDP. The financial stabilization and fiscal stimulus policies enacted during the recession helped mitigate the severity of the downturn (Blinder and Zandi 2015), but the gap between actual and potential GDP, at $410 billion in constant 2015 dollars, still had not closed by the fourth quarter of 2015.

​The Great Recession sharply reduced employment for many workers in the United States, but these reductions were concentrated among men, younger workers, and workers with lower levels of education. This may be the result of a so-called job ladder that renders low-skilled workers—and young workers—disproportionately sensitive to the business cycle (Barnichon and Zylberberg 2014; Beaudry, Green, and Sand 2013). Figure 2a shows the change in the employed share of the population aged 25 and older by education level and gender, from just before the recession began (December 2007) to December 2009—roughly the trough of the labor market. Employment fell more sharply for workers with low levels of education, though women with less than a high school degree saw only a 3.4 percent loss in employment. Men tended to experience steeper declines than women, within both education and age groups. The employment share fell the farthest for men without a high school degree; note that men tend to suffer more during recessions largely because they tend to be employed in industries that are more cyclical (Hoynes, Miller, and Schaller 2012).

Figure 2b shows the change in the share of the population employed for different age groups, from December 2007 to December 2009. The steep decline captured by the leftmost bars in figure 2b (corresponding to ages 20–24) highlights the impact of the recession on young people. This pattern partly reflects the option that young people possess to acquire education rather than participate in a weak labor market—an alternative less readily available to middle-aged workers. This pattern may also be attributable to the tendency of firms to lay off inexperienced workers before those in mid-career.

Policy makers rely on two sets of tools to foster recovery following a recession. Monetary policy, consisting of actions taken by the Federal Reserve, is used to keep interest rates low and reduce unemployment during and after a recession. Fiscal policy includes various forms of government spending and tax cuts enacted by Congress. Following a recession, both sets of policies can be used to increase demand, thereby raising output and more quickly returning the economy to prerecession conditions.

During the Great Recession, fiscal policy played an important role in the economic recovery (figure 3). The purple line shows the actual path of GDP: note that the economy reached its low point in the second quarter of 2009 and then began growing again. Blinder and Zandi (2015) estimate that, absent the fiscal stimulus, the economy would have continued to contract until the fourth quarter of 2009, to $15.5 trillion in output annually (in constant 2015 dollars), and would not have reached prerecession levels until the second quarter of 2012—a year after it actually did.

As the central banking institution of the United States, the Federal Reserve has two primary responsibilities, often referred to as its dual mandate: (1) to maintain price stability and (2) to help the economy remain near maximum sustainable employment. As shown in figure 4, the unemployment rate rises as the economy contracts: in the past 45 years, as the economy has gone through each recession, unemployment has increased on average by 2.5 percentage points. The Federal Reserve responds to these changes in the labor market by using open-market operations, which consist of the acquisition and sale of securities, to lower the federal funds rate (the interest rate at which institutions lend deposits at the Federal Reserve to other institutions overnight) and recently to expand its balance sheet. When economic conditions worsen, a reduction in the federal funds rate lowers interest rates throughout the economy, encouraging businesses to invest and employ more workers and encouraging consumers to spend more, consequently lowering the unemployment rate.

As shown in figure 4, after the Federal Reserve lowers the federal funds rate, the unemployment rate tends to drop, albeit with a lag. When economic conditions improve, the Federal Reserve raises the federal funds rate again to forestall inflation, which occurs when the economy overheats and prices rise too rapidly.

The unemployment rate remained elevated well after the end of the Great Recession, and the Federal Reserve has consequently kept the federal funds rate close to zero for some time, while also pursuing other expansionary monetary policies to encourage economic activity. If another recession were to occur while the federal funds rate remains near zero, the Federal Reserve would not have its conventional arsenal of tools since the federal funds rate cannot be lowered much farther. It would then have to make do with unconventional tools to help the economy recover. Given that considerable uncertainty remains about the effectiveness and costs of unconventional monetary policies, expansionary fiscal policy may be an attractive option.

Governments may use fiscal policy—additional government spending or tax cuts—to stimulate the economy during a recession. A fiscal multiplier is an estimate of the increased output caused by a given increase in government pending or reduction in taxes. Any multiplier greater than zero implies that additional government spending (or reduced taxes) adds to total output. Fiscal multipliers greater than one indicate an increase in private-sector output along with an increase in output from government spending.

Although there is disagreement among economists over the exact size of various fiscal multipliers (see Auerbach, Gale, and Harris [2010] for a discussion), multipliers are generally believed to be higher during recessions than they are under normal economic conditions when the economy is near its full potential (Auerbach and Gorodnichenko 2012; Fazzari, Morley, and Panovska 2014; see Ramey and Zubairy 2014 for a dissenting view). This is likely because downturns are characterized by slack in both labor and capital markets (i.e., available resources are not fully employed), thereby allowing fiscal stimulus to increase total output. Multipliers are also higher when the spending program or tax cut targets lower-income people, who are more likely to spend the stimulus (Parker et al. 2013; Whalen and Reichling 2015).

Not all spending or tax cuts are created equal, as indicated by the variation in fiscal multipliers shown in figure 5 and figure 6. But during the depths of the recession, each spending multiplier analyzed by Blinder and Zandi (2015) was greater than one, indicating that spending on these programs raised output by more than their costs. Moreover, during the recent recession, every spending multiplier featured here was higher than the tax multipliers described in Fact 6. Note that the multipliers reported here are broadly similar to those estimated by CBO (Whalen and Reichling 2015).

As shown in figure 5, the most stimulative type of spending during the recession was a temporary increase in the SNAP maximum benefit: for every $1 increase in government spending, total output increased by $1.74. Work-share programs and UI benefit extensions were also relatively stimulative. Consistent with economic theory, the programs with the largest multipliers were those directed at low-income or newly unemployed people. More recently, as the economy has improved, the multipliers have diminished. However, the multipliers for SNAP benefits, workshare programs, and UI benefits remain above one, indicating that these programs remain effective as forms of stimulus, generating additional private-sector economic activity.

Tax cuts can also provide useful stimulus, though multiplier estimates are generally lower and more variable than estimates for spending programs. Estimates range from 1.38 for the Child Tax Credit in the midst of the recession to 0.23 in 2015 for accelerated depreciation, which effectively allows firms to postpone tax liabilities.

Although many tax cuts had a fiscal multiplier greater than one during the recession, in 2015 only the multipliers for the Child Tax Credit and refundable lump sum tax rebate were still greater than one. Furthermore, their values were smaller than those for spending multipliers (Fact 5) such as increasing SNAP benefits and extending UI benefits. Spending programs or tax cuts that focus on lower-income people tend to have higher multipliers because those people are more likely than higher-income people to spend what they receive. In addition, policies such as accelerated depreciation and cuts in the corporate tax rate, both of which benefit businesses, are less stimulative than tax cuts for individuals.

Fiscal stimulus may take two forms during a recession. The first is discretionary, when Congress authorizes new tax cuts or government spending to shore up the economy (as it did with the passage of the ARRA in 2009). However, some fiscal stimulus also occurs automatically without congressional action, making for a quicker response to deteriorating economic conditions. For example, as employment and income levels decline during economic downturns, participation in existing safety net programs increases and affected households’ tax bills fall. According to the CBO, three types of outlays constitute the large majority of automatic stabilization in federal spending: UI benefits, Medicaid benefits, and SNAP benefits (Russek and Kowalewski 2015). In addition, there are several sources of revenue that contribute to automatic stabilization: individual income taxes, taxes on corporate profits, Social Security and Medicare payroll taxes, taxes on production and imports, and UI taxes. During expansions, rising incomes generate more tax revenue; during downturns, taxes are automatically lowered as incomes fall. These automatic stabilizers help to moderate the booms and busts of the business cycle, increasing aggregate demand to offset the negative effects of an economic downturn, and decreasing demand once the economy has recovered.

As shown in figure 7, spending on automatic stabilizers varies over the business cycle, expanding promptly during recessions. This stands in contrast to discretionary fiscal policy: by the time that ARRA was authorized in 2009—five quarters after the start of the recession—spending on automatic stabilizers had already grown by 2 percent of GDP.

Poverty and economic hardship typically increase in recessions and decrease in economic expansions. In particular, households with few resources are especially affected by the business cycle. Among poor households, the effect of the Great Recession was particularly severe relative to previous recessions. Bitler and Hoynes (2015) estimate that for a 1 percentage point increase in the unemployment rate, the share of households below 50 percent of the poverty threshold expanded more in the Great Recession than during the two recessions of the early 1980s. (The poverty threshold in 2015 was $24,036 for a family of four with two children [U.S. Census Bureau 2016b].) The safety net plays an important role in mitigating these effects, partly by automatically expanding during economic downturns as eligibility for safety net programs increases. However, there were stark differences among safety net programs in their responsiveness to the Great Recession.

For example, the TANF program—which supports poor families with cash assistance, resources for child care, and work-related services, among others—expanded only slightly during the most-recent recession, and the number of families benefiting from the program has now fallen below its prerecession level, despite the fact that poverty remains elevated. Figure 8a shows the cumulative change in the TANF caseload (purple line) during and after the recession compared with the cumulative change in the number of families in poverty (green line); the dramatic split between the two lines suggests that TANF is failing to reach many needy families. Unlike SNAP and UI, TANF is largely funded through a federal block grant with a fixed value, making it less responsive to changes in need.

Other programs, including SNAP and UI, functioned more effectively as automatic stabilizers during the most-recent recession (Bitler and Hoynes 2010; Di Maggio and Kermani 2016). Note that during the Great Recession, Congress increased spending on SNAP and UI above and beyond the increase that would have occurred automatically. Figure 8b shows the change in the number of UI recipients compared with the change in the number of people unemployed. In contrast to TANF, UI did respond to the economic downturn, although many workers are either ineligible for or do not claim UI, and the program consequently covers only a portion of newly unemployed workers. Despite these limitations, UI was a powerful automatic stabilizer, with the increase in UI recipients amounting to more than 7 million during the economic downturn.

Rainy-day funds—also called budget stabilization funds—are dedicated pools of money set aside by states during good times to help them weather economic downturns. Since states are generally restricted by law from running budget deficits, rainy-day funds help them to balance their budgets when tax revenues fall, without resorting to devastating spending cuts or tax increases at exactly the wrong moment. In each of the previous two recessions, states used their rainy-day funds to avoid more than $20 billion in tax increases and/or cuts to services (McNichol 2014).

Nonetheless, many states still struggle to allocate sufficient savings to rainy-day funds. This can be attributed in part to poor design: 43 states set arbitrary caps or targets on annual contributions to rainy-day funds (The Pew Charitable Trusts 2015), and in many states these caps and targets on contributions are too small. In 2015, six years after the end of the Great Recession, only eight states had accumulated enough in their rainy-day funds to offset a hypothetical one-year loss of 10 percent or more of their annual expenditures. Given that average state taxes dropped 11 percent from fiscal year 2008 to fiscal year 2009, and 21 states experienced losses above 10 percent (The Nelson A. Rockefeller Institute of Government n.d.), many states could struggle to absorb a similar loss in the next recession using rainy-day funds alone.

To be sure, the phenomenon of food insecurity in the United States is not equivalent to the severe malnutrition observed in some developing countries. Nonetheless, it has far-reaching impacts on the health and well-being of an unacceptably large number of Americans adults and children. The common-sense notion that lack of access to food is harmful has been established by rigorous research. Children living in food-insecure households tend to have a lower health-related quality of life (Casey et al. 2005), higher rates of asthma (Mangini et al. 2015), less-nutritious diets (Fram et al. 2015), and behavioral problems that affect school performance (Whitaker, Phillips, and Orzol 2006).

Food insecurity is measured using the U.S. Department of Agriculture (USDA 2015) 18-question U.S. Household Food Security Survey Module, implemented annually in the December Supplement (CPS-FSS) to the U.S. Census Bureau’s Current Population Survey (CPS). This survey tool asks a series of questions about households’ resources available for food and whether adults or children in the household adjusted their food intake—cutting meal size, skipping meals, or going for a day without food—because of lack of money for food. A household is considered to be “food insecure” if, due to a lack of resources, it had difficulty at some time during the year providing enough food for all of its members. The more-severe categorization of “very low food security” status describes those food-insecure households in which members’ food intake was reduced and their normal eating patterns disrupted at some point during the year because of a lack of resources for food. Food insecurity and very low food security are measured at the household level, though questions about adults and children are asked separately. In other words, a child may live in a food-insecure household, but be buffered from the direct effects of food insecurity by the adults in the household. The food insecurity measures are collected annually in December as a supplement to the U.S.Census Bureau’s Current Population Survey (CPS); the results are the Current Population Survey Food Security Supplement (CPS-FSS) data files.

Note that food insecurity is distinct from poverty. While the poverty rate measures the share of families with annual gross income below a particular threshold, the food insecurity rate reflects the resources available to purchase food. In thirty states and the District of Columbia the rate of food insecurity is higher than the rate of poverty. While the rate of food insecurity declines as household income increases, its reach extends farther up the income distribution than many would guess: two-thirds of food-insecure households have annual incomes above the federal poverty level (FPL). And because many households may be food secure one year but not the next, an even larger share of households has had some experience with food insecurity than any single-year snapshot suggests.

There is an important role for the safety net in insuring households against food insecurity. The largest of the federal nutrition assistance programs, the Supplemental Nutrition Assistance Program (SNAP; formerly called the Food Stamp Program), is highly effective, lifting millions of people out of poverty and increasing the resources they have available to purchase food.

In addition, because it is designed to expand and contract according to need, SNAP serves as an important macroeconomic stabilizer. Furthermore, several studies have found that SNAP reduces the likelihood that a household will experience food insecurity or very low food security (Ratcliffe, McKernan, and Zhang 2011; Schmidt, Shore-Sheppard, and Watson 2012; Shaefer and Gutierrez 2013). Moreover, evidence from safety net expansions—such as the temporary benefit increase under the American Recovery and Reinvestment Act of 2009 (ARRA) and a pilot program that provided additional benefits to families of children during the summer months when school meals were not available—shows reductions in rates of food insecurity and very low food security. Recent studies have shown that SNAP improves health outcomes and households’ financial well-being, and even improves the later-life outcomes of individuals who had access to the program as children.

A guiding principle of The Hamilton Project is that long-term prosperity is best achieved by fostering economic growth and broad participation in that growth. This necessitates increasing economic security—which in turn can increase economic growth by enabling people to invest in their education or that of their children, and by helping families get back on their feet quickly after unexpected shocks. In this spirit, The Hamilton Project offers the following 12 facts on food insecurity, SNAP, and other nutrition support programs.

In 2014 more than 15.3 million children—or more than one in five—lived in a food-insecure household (Coleman-Jensen et al. 2015) in the United States. This is a marked increase from the years prior to the Great Recession, when an average of 12.9 million children lived in a food-insecure household.

The USDA defines a household as food insecure if it reports that it had difficulty at some time during the year providing enough food for all of its members due to a lack of resources. This broad measure of food insecurity includes households that report a reduction in the quality, variety, and desirability of diet but little or no reduction in food intake, as well as households that experience very low food security—that is, who report disruptions in eating patterns and reductions in food intake (Coleman-Jensen et al. 2015). This latter measure, which accounted for 5.6 percent of households in 2014, is sometimes used as the preferred metric for hunger experienced in the United States (Chilton and Doar 2015; Cook and Jeng 2009).

Almost one in seven households—and almost one in five households with children—reported difficulty providing enough food in 2014 for all of their members (Coleman-Jensen et al. 2015). Households with children have a higher rate of food insecurity than households overall and households with seniors; this elevated rate persists despite those households’eligibility for additional nutrition support through the subsidized school meals programs and Special Supplemental Nutrition Program for Women, Infants, and Children (WIC).

After the onset of the Great Recession all household types saw sharp increases in rates of food insecurity, with households with children experiencing the largest increase. From 1998 to 2007 an average of 15.7 percent of households with children, 10.8 percent of households overall, and 6 percent of households with seniors were food insecure. The average from 2008 to 2014 was roughly 4 percentage points higher for households overall and for households with children, and about 2 percentage points higher for households with seniors. These changes amount to millions more Americans living in food-insecure households. Despite recent improvements in the economy, food insecurity rates are still higher than they were prior to the Great Recession, potentially reflecting higher rates of poverty and increased costs of other necessities such as housing.

In every state a higher share of children than the equivalent share of adults lived in a food-insecure household (Coleman-Jensen et al. 2015). However, the 15 million children and 29 million adults who lived in food-insecure households in 2014 were unevenly distributed across states. Figure 2 shows the average annual percentage of children living in food-insecure households over the period 2012–14. In half of states plus the District of Columbia, at least 20 percent of children lived in a food-insecure household, and in nine states the share of children who lived in a food-insecure household was at least 25 percent.

Since the Great Recession began in 2007, 41 states have seen the percent of children living in food-insecure households increase; in 28 states, the increase was greater than 4 percentage points (CPS 2005–14). The states with the highest rates of food insecurity are Louisiana, Alabama, and Mississippi, where almost 30 percent of all the children in these states live in a food-insecure household. In only two states—Iowa and Wyoming, as well as the District of Columbia—has the percent of children living in food-insecure households decreased by more than 4 percentage points since 2008.

The vast majority of food-insecure households with children are working households: over 85 percent of households with children reporting food insecurity in 2014 also reported at least one adult who earned an income in 2014. Food-insecure households with children were slightly more likely to be headed by married couples (46.9 percent) than single mothers (44.6 percent). Among every type of household reporting food insecurity—i.e., households headed by married couples, single mothers, or single fathers—at least 75 percent of households had an earner during the year they experienced food insecurity. Nonetheless, the income generated by work over the year was not sufficient to protect those working families from food insecurity.

Employment provides many benefits to households and children, yet working necessarily reduces the amount of time that an earner has available to do other tasks, including shopping for and preparing food. This time constraint may increase the monetary cost of food and, in turn, increase the likelihood that a family experiences food insecurity. A number of studies find that working mothers report using food preparation strategies that take less time and cost more, such as relying more on convenient meals, spending less time preparing food, and preparing fewer family dinners (Bauer et al. 2012; Cawley and Liu 2012).

As any parent of teenagers knows, food consumption increases during the teenage years (Docter and Breuner 2012; Stang and Story 2005). This is reflected in U.S. dietary guidelines, which estimate the calorie needs of teens to be on par with that of their parents, and more than three times the needs of younger children (U.S. Department of Health and Human Services [DHHS] 2015). It is also apparent in measures of food intake and food spending, both of which increase when children enter their teenage years (Anderson and Butcher 2016).

Households with teenage children experience higher rates of food insecurity than do households with only younger children. Among households with teenagers, over 20 percent were food insecure—2 percentage points higher than the rate among households with younger children. The rates were similar whether the teenage child was male or female. Rates of very low food security also increase markedly when a teenager is present—by about 1.5 percentage points, or 25 percent. After holding other factors constant, Anderson et al. (forthcoming) found that adding an additional teenager to a household increases by about 50 percent the probability that children in a family experience very low food security.

SNAP benefits do not vary by child age, and feeding a family of four with two teenage boys according to the USDA’s Thrifty Food Plan would cost $50 per month more than the maximum SNAP benefit available to the family (USDA 2016a). Recent evidence shows that modest funding increases in food support programs can greatly reduce food insecurity, especially among households with teenagers (see Fact 11). School meals programs specify higher calorie recommendations for meals served to middle and high school students (USDA 2012), but high school students are less likely than younger children to participate in school meals.

Annual rates of food insecurity mask the extent of the food insecurity problem. Using the Current Population Survey, we can follow large numbers of households across two consecutive years, allowing us to compare food security status over time. In consecutive years during the post-recession period 2008–14, over 24 percent of households with children experienced food insecurity in one or both years: 9 percent of household experienced food insecurity in consecutive years, and an additional 15 percent of households experienced food insecurity in only one of the two years.

Suring the temporary periods when those families experience food insecurity, there still may be negative impacts on their children. For example, one study found a decline in cognitive development and health status among toddlers who lived in households that were food secure when the child was nine months old but had become food insecure by the child’s second birthday (Hernandez and Jacknowitz 2009).

For many households, food insecurity appears to be a temporary challenge. Around 40 percent more households reported being food insecure in the past two years than reported being food insecure last year. To the extent that food insecurity is a temporary or sudden experience, policies to address it must also be quick to respond.

Food insecurity, often thought to be a characteristic of poverty, is actually dispersed widely over the income distribution, though food insecurity does decline as a household's income increases. Notably, a large share of food-insecure households live above the federal poverty level (FPL) with incomes that are above the reach of SNAP and other food support programs. One-third of food-insecure households have reported incomes between 100 and 200 percent of the FPL and another third have reported incomes above 200 percent of the FPL—a level at which households are typically not eligible for SNAP or subsidized school meals. Many of these families have incomes above the reach of the EITC as well, which phases out near 185 percent of the FPL (Urban-Brookings Tax Policy Center 2016).

Among families with annual incomes below the FPL, more than 35 percent experience food insecurity—much higher than those with incomes reported between two and three times the FPL (20 percent), and those with incomes more than three times the FPL (less than 10 percent).

Very low food security is even more concentrated than food insecurity among the poor. The highest rate of very low food security is reported among the poorest households, or those living below half of the FPL (about $12,000 for a family of four).

In 2012 safety net programs lifted almost 50 million people—including over 10 million children—out of poverty, as defined by the Supplemental Poverty Measure (Sherman and Trisi 2015). The largest reductions in poverty among children can be attributed to SNAP, the Earned Income Tax Credit (EITC), and the child tax credit. After adjusting for the underreporting among social program beneficiaries in the Current Population Survey, it is estimated that SNAP lifts more than 10 million people out of poverty, including around 5 million children. Other programs such as housing assistance, Temporary Assistance for Needy Families (TANF), the National School Lunch Program (NSLP), and Supplemental Security Income (SSI) lift a smaller but still substantial number of children out of poverty.

Recent research has documented strong, long-term impacts from investing in certain antipoverty programs aimed at children. Programs such as the EITC that supplement the earnings of low-income working families have been shown to increase children’s test scores (Dahl and Lochner 2012), thereby raising their expected earnings when they reach adulthood (Chetty, Friedman, and Rockoff 2014). These studies suggest that the benefits of poverty alleviation will accrue over the long run.

A new study by Hoynes, Schanzenbach, and Almond (2016) finds long-term positive effects from consistently providing access to the Food Stamp Program (now called SNAP) during early life. Taking advantage of the relatively long rollout period when the program was originally introduced, the study compares children who lived in different counties within a state and who were born at different times to measure the long-term impacts of access to the program. Access to the Food Stamp Program at early ages—starting before birth in cases where the mother received food stamps during pregnancy, and continuing through age five—leads to a number of positive long-run health and economic outcomes.

As shown in figure 8, access to the Food Stamp Program over this age range has substantial positive impacts on later health, lowering women’s and men’s incidence of metabolic syndrome—a health measure that includes diabetes, high blood pressure, obesity, heart disease, and heart attack—by 0.3 and 0.5 standard deviations, respectively. Women are also 34 percentage points more likely to report excellent or very good health if they had access to food stamps from before birth through age five.

These gains also extend to economic outcomes. Women with access to the Food Stamp Program over the full early life period have much higher economic self-sufficiency—a measure that includes completed education, employment status, earnings, and financial success—than those who did not. Furthermore, access to food stamps increased high school graduation rates by more than 18 percentage points.

Access to food stamps and the NSLP also seems to improve educational outcomes among older children (ages 6 to 18) as well, which may indicate that better nutrition helps students gain more from school, setting them on a path toward greater self-sufficiency (Hinrichs 2010; Hoynes, Schanzenbach, and Almond 2016).

The four largest nutrition programs in the United States are the Supplemental Nutrition Assistance Program (SNAP), the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), the National School Lunch Program (School Lunch), and the School Breakfast Program (School Breakfast). Combined spending on these four programs was $97 billion in 2015, with SNAP accounting for $74 billion (USDA 2016b). Eligibility for each program is means-tested and targeted to specific populations, such as children enrolled in a school that participates in the SBP and the NSLP.

Figure 9 describes the overlap of nutrition program participation and food security status for households with children and annual incomes at or below 185 percent of the FPL. Each row is limited to households that meet the categorical eligibility criteria for the program; that is, it is limited to households with a school-aged child eligible for School Breakfast or School Lunch, or to households with a child under age five eligible for WIC. Households are grouped into four populations: food insecure and reporting no program benefits, food insecure but reporting program benefits, food secure and reporting program benefits, and food secure and reporting no program benefits.

The NSLP has the largest reach among these programs, and is available in nearly all public schools. As the figure illustrates, 64 percent of low-income households report participating in the free or reduced-price school lunch program. Of participants, almost half are food insecure despite participating in the program. Nine percent of low-income families with children are food insecure but report receiving no subsidized school lunch. Participation is similar but slightly more muted for the school breakfast program.

WIC has the lowest participation rate among the top four nutrition programs, and the highest percentage of eligible households reporting both food insecurity and no program participation. This could be due in part to how WIC participation is reported in the CPS, where respondents are asked whether they received WIC in the past month; by contrast, SNAP participation is reported for the past 12 months. Among households with annual incomes at or below 185 percent of the FPL, 44 percent report participating in SNAP. This estimated participation rate may be understated for at least two reasons. First, SNAP eligibility is typically limited to households with monthly incomes at or below 130 percent of the FPL, so a portion of the low-income sample may be ineligible for SNAP. Second, participation in safety net programs appears to be systematically underreported in the CPS (Meyer, Goerge, and Mittag 2014). With these caveats in mind, one-sixth (16 percent) of low-income households were food insecure but reported that they did not receive SNAP.

The fact that many children who receive benefits remain food insecure does not imply that the programs are ineffective. Families most in need of food assistance are most likely to enroll in nutrition programs, but for some the additional resources are still inadequate to end their food insecurity. For example, school meals are available only when school is in session, and food insecurity rates rise for children during the summer when the benefits are not provided (see Fact 11).

More troubling is the share of households that are food insecure and eligible for nutrition programs but do not participate. More work is needed to understand why this nonparticipation occurs, and whether policies can improve take-up among the population most at risk of food insecurity.

Figure 10 shows changes in the rates of very low food security and in the average per person monthly SNAP benefits from year to year. In the immediate wake of the Great Recession, the rate of very low food security increased for households receiving SNAP. Among SNAP households with children, very low food security increased by nearly 1 percentage point in 2007 and 3 points in 2008 over the prior year’s level, while monthly benefits levels went down by $2 and then up by $3.

However, in response to the Great Recession, ARRA temporarily increased maximum SNAP benefit levels by 13.6 percent. Per person average monthly benefit levels increased $25 from 2007 to 2008 and an additional $8 from 2008 to 2009 (in inflation-adjusted 2016 dollars). The ARRA benefit increase coincided with the only substantial decreases in very low food insecurity among households with children in the past eight years. When benefit levels were at their highest and the recession was at its peak, the incidence of very low food security among SNAP-receiving households with children fell to prerecession levels.

After the decline in rates of very low food security in 2009 and 2010 that coincided with a SNAP benefit increase, rates of very low food security increased again when benefit levels decreased in 2011 by $5 per month over the previous year. By 2014 rates of very low food security were 4 percentage points higher than they were in 2006. There were more households with children participating in SNAP who had very low food security in 2014 than there were at the end of the recession in 2009.

Although the school meals programs serve as the front line of defense against food insecurity for children, food insecurity rises when students lose access to those programs during the summer months (Nord and Romig 2006). To address this problem, the USDA created an experimental pilot program called the Summer Electronic Benefits Transfer for Children (Summer EBT) to provide additional benefits during the summer months that can be used at food stores (Collins et al. 2013). In 2012 USDA implemented a demonstration project as a randomized trial to test the impact of a $60 per summer month benefit—or about the combined monthly cost of a student participating in the SBP and the NSLP.

The results of the pilot program show that additional benefits caused a dramatic reduction in the share of households and children that experienced food insecurity or very low food security. In particular, the share of households reporting very low food security over the summer declined by one third. Furthermore, the share of children who directly experienced very low food security also declined by a third. The broader measure of food insecurity was also substantially reduced in children by around 20 percent. The $60 per month Summer EBT benefit also helped parents to provide their children with healthier food options. Compared to the control group that did not receive benefits, Summer EBT recipients consumed a healthier diet, including 13 percent more fruits and vegetables, 30 percent more whole grains, and 10 percent more dairy products.

In a subsequent demonstration project the following summer, USDA found that a smaller benefit of $30 per month had nearly equal impacts on reducing hunger, but was less effective at reducing the broader measure of food insecurity (Collins et al. 2014).

In addition to reducing food insecurity, SNAP participation may also reduce households’ risk of suffering financial hardships. Shaefer and Gutierrez (2013) use variation in state-level policies that affect SNAP access to study the impact of SNAP participation on a variety of outcomes. They find that receiving SNAP reduces the likelihood of food insecurity by 13 percentage points.

SNAP also has spillover impacts on other aspects of families’ financial well-being. Households have more resources available for other essential expenses, such as housing, utilities, and medical bills. Shaefer and Gutierrez (2013) estimate that SNAP participation reduces the risk of falling behind on rent or mortgage payments by 7 percentage points and on utility bills (gas, oil, and electricity) by 15 percentage points. Participants are also less likely to experience medical hardship: SNAP participation decreases the likelihood of forgoing a necessary visit to a doctor or hospital by 9 percentage points.

The Hamilton Project supports policies that foster economic growth and broad-based participation in that growth. For the past decade we have emphasized that evidence-based improvements to public education are essential to the success of our economy and society. As the following fourteen facts demonstrate, the evidence makes an overwhelming case for policy changes in education.

Over the past three decades the gap in earnings between workers with high levels of education and those with less education has grown substantially (see Fact 2). For men, median real earnings increased only among those with college or advanced degrees: their earnings are 29 percent higher than they were in 1980. Men with less than a bachelor’s degree have seen their real earnings stagnate or decline, with an especially large drop among workers with the lowest levels of education. In contrast, women have seen their real median earnings increase across all education levels, with a nearly 40 percent increase among college-educated women. The share of young women with a bachelor’s degree has also increased over the past three decades, almost doubling for women age 25–34 from just 21 percent in 1980 to 39 percent in 2015. The rate of high school completion (including GED attainment) among young women also increased, rising 6 percentage points over this period to 91 percent. The share of young men who completed high school also rose, but by less than the gains among women (see Fact 3).

Over the past 20 years both high- and low-income students have made strong gains in math achievement, with the share of fourth-grade students scoring at or above proficiency in math in 2014 substantially higher than in 1996. In fact, low-income students’ proficiency rates tripled over this period. However, there remain large and persistent gaps in skills between high- and low-income students, as illustrated in figure A. Math proficiency rates among low-income students were still lower in 2015 than high-income students’ rates in 1996. Identifying ways to close these achievement gaps is perhaps the largest challenge facing K–12 education today.

Large differences in skills between high- and low-income students make it especially daunting to address disparities in student achievement, in part because these disparities reflect broader economic challenges such as the large share of children living in poverty and, in particular, the high concentration of poverty at the school level. In 2014 a majority of public school students nationwide attended high-poverty schools—defined as schools in which more than 40 percent of students are low-income (see Fact 6). At the state level, performance among low-income students declines as the share of low-income students in the state increases (see Fact 8). This may partly reflect differences in access to resources: while education spending overall has steadily increased each decade over the past 50 years (see Fact 11), spending is highly uneven across states because most local education budgets are funded by state and local revenues (see Fact 10). In fact, some northeastern states spend about twice as much per student as states in the South and West, even after adjusting for variation in cost of living.

There are several promising interventions that have been shown to improve test scores and high school graduation rates, especially among low-income students living in areas with high concentrations of poverty. The charter school sector serves a small but increasing share of students, and achievement gains among students attending these schools can be quite strong in some cases (see Fact 14). An emerging stream of evidence also shows that interventions aimed at younger children—such as preschool programs and class-size reduction in the early grades—can improve high school graduation rates (see Fact 12) and students’ later economic outcomes.

There are many factors at work in determining educational outcomes; some of these are more easily addressed by policy reforms than others, and not all can be addressed directly within the K–12 education system. To illustrate the payoffs from increasing educational attainment, the challenges faced by our nation’s K–12 schools, and the promise of targeted childhood interventions, The Hamilton Project offers the following fourteen facts on education and economic opportunity.

Skills are an important determinant of workers’ earnings and employment opportunities. No test perfectly measures the skills relevant to economic success, but tests that aim to measure analytical ability often do correlate with outcomes in the labor market. For example, the left panel of figure 1 shows a correlation between scores on the Armed Forces Qualification Test (AFQT) and earnings in middle age. Stronger analytical skills as measured by the AFQT correspond to higher wages, resulting in the upward sloping pattern shown in the figure. A similar relationship holds for measures of a range of analytical skills, such as verbal reasoning and problem solving. As seen in the figure, the relationship between AFQT scores and earnings is different for men and women: a 10-point increase in AFQT is associated with about a 15 percent increase in earnings for men but an 11 percent increase for women. Skills measured as early as kindergarten have a positive impact on adult earnings (Chetty et al. 2011).

Employment outcomes also depend on soft skills—a bundle of character attributes (sometimes referred to as noncognitive skills) that includes motivation, perseverance, tenacity, creativity, self-control, and the ability to work productively in groups (Deming 2015; Heckman, Stixrud, and Urzua 2006). The right panel of figure 1 shows on the horizontal axis a combined index of three scales that measure social skills, self-esteem, and the extent to which a person believes that her own actions, as opposed to forces outside of her control, determine rewards and success in life. The upward sloping pattern indicates that, by this measure, those with higher noncognitive skills also earn more, on average. The relationship between this noncognitive index and earnings is similar for men and women: a 10-point increase in the noncognitive index is associated with about a 7 percent increase in earnings for both men and women. Emerging research has found that both cognitive and soft skills can be taught, but that the teachers who are effective at raising cognitive skills are not necessarily the same teachers who are effective at imparting soft skills (Jackson 2012; Kraft and Grace 2016).

It is important to note, however, that tests of both cognitive and soft skills are correlated with other factors that influence employment outcomes, such as educational attainment, parents’ income, and the neighborhood environment. This makes it difficult to measure precisely the relative importance of these other factors compared to skills in determining a young person’s later economic success. Compounding this difficulty is the fact that skill formation is a cumulative process in which experiences in early childhood and adolescence lay the foundations for future learning. This cumulative feature is partly attributable to perceptions of elf-efficacy: early success at learning not only makes later skill-building easier, but also shows that effort has rewards, which can create a self-reinforcing motivation to continue learning (Heckman 2006). The flip side is that impediments to early learning can have cumulative negative effects on later skill formation. Such obstacles are particularly troubling given that the effects of skills on economic outcomes reach far beyond earnings; early childhood test scores are correlated with high school graduation rates (see Fact 12), college attendance, marriage, home ownership, and retirement savings (Chetty et al. 2011).

The gap in earnings between high-school-educated and college-educated workers has more than doubled in the United States over the past three decades (Autor 2014). Although a large number of developed countries show a similar trend, the United States appears to have the largest skills premium—that is, the gap between the wages of low- and high-skilled workers. In part, the magnitude of this gap reflects the fact that the inflation-adjusted earnings of workers with high levels of education have risen much more over the past 35 years than have the earnings of less-educated workers, as shown in figure 2. In fact, real earnings for men without a bachelor’s degree have actually fallen since 1980—and fallen dramatically for men with no high school diploma, whose earnings are down by more than 20 percent. Gains in earnings for women with a bachelor’s degree or an advanced degree were more than twice as large as those for women with a high school diploma, some college, or an associate degree. However, earnings for women at all education levels tended to increase at a higher rate than they did for similarly educated men. The differences may be partially explained by a lower initial level of earnings for women relative to men.

In recent decades the growing earnings premium for education has contributed substantially to the net growth of earnings inequality. Roughly two-thirds of the overall increase in earnings dispersion between 1980 and 2005 is accounted for by the rising returns to schooling—primarily the growing premium to postsecondary education (Goldin and Katz 2007). To exclude the effect of the very top and bottom of the earnings distribution on growing inequality, one can compare earnings at the 90th percentile of the wage distribution to earnings at the 10th percentile. This 90/10 ratio has also grown substantially in recent decades. Fully 95 percent of its growth between 1984 and 2004 was attributable to rising returns to education (Fortin, Lemieux, and Firpo 2011). In other words, if the education premium had remained constant, there would have been essentially no increase in wage inequality between 90th- and 10th-percentile workers.

Why has the education premium risen in recent decades? Economists continue to study this question, but a number of studies point to the interplay of supply and demand for skills. In The Race between Education and Technology, Harvard economists Claudia Goldin and Lawrence Katz (2008) make the case that the U.S. economy prospered during the twentieth century in large part because educational attainment kept up with the rising demand for skills, which was catalyzed by significant technological change. The labor market’s demand for analytical skills, written communications, and specific technical knowledge increased dramatically. Between 1900 and 1980 Americans kept pace by steadily increasing their level of education, reflecting in large part the country’s commitment to a secondary school system essentially free and open to all (Goldin and Katz 2008). Yet over the past three decades the rise in educational attainment has slowed, even as technological progress—and the corresponding demand for skills—has accelerated. Goldin and Katz (2008) succinctly capture the essence of the story: “In the first half of the century, education raced ahead of technology, but later in the century, technology raced ahead of educational gains” (p. 8).

Despite a slowing pace, educational attainment in the United States has gradually risen over the past 35 years, as shown in figure 3. The share of young women with college degrees almost doubled, from 21 to 39 percent, and young women are now more likely than young men to have a college degree. As shown in Fact 2, this has coincided with large increases in the wage premium for women with college degrees. High school completion rates have also increased over this period. From 1980 to 2015 the share of American men age 25–34 that completed high school inched up from 86 to 90 percent, and rose from 85 to 91 percent among women.

Nonetheless, many researchers worry that skill levels among young people in the United States are far too low. Historically, successive generations have attained more education at a rate of approximately one additional year of schooling every decade; for example, individuals born in 1930 averaged about 11 years of schooling by the time they reached age 30, compared with 13 years for people born in 1950 (Goldin and Katz 2008). But for individuals born around 1950 and later, this pace was roughly cut in half, even though the economic returns to education have increased (see Fact 2). Individuals born around 1985, who were 30 years old in 2015, averaged 14 years of schooling.

There is concern among some researchers that public policies aimed at improving skills and educational attainment are too often evaluated on narrowly defined criteria such as conventional cognitive test scores. For example, in the 1960s the Head Start program was thought to have fallen short of its goals because it did not permanently raise the IQs of its participants. This kind of write-off may be premature in light of evidence from early childhood interventions such as the Perry Preschool Project, which had lasting benefits that were not picked up by conventional measures of cognitive ability (Schweinhart et al. 2005). In particular, by age 10 the IQ scores of the participants were no higher than those of the control group, but the participants did show more motivation to learn, and by age 40 they were more likely to have graduated high school, to be earning higher salaries, to own a home, to be less dependent on welfare, to have had fewer out-of-wedlock births, and to have not been arrested.

On average, more education decreases one’s likelihood of living in poverty. As shown in figure 4, the poorest Americans—with incomes below 100 percent of the federal poverty line, or below $24,000 for a family of four—are typically much less educated as measured by the highest educational attainment within the household. Fully one-fifth never finished high school, and over half have a high school diploma or less. Those with household incomes between 100 and 299 percent of the federal poverty line, or annual incomes between $24,000 and roughly $72,000 for a family of four, are much more educated: roughly four in ten have completed at least an associate degree, and only about one-third of them have a high school diploma or less. Among households earning 300 percent of the federal poverty line or more, over 90 percent have at least some college education. While education is a strong predictor of income, there is still considerable variation in education within income groups. For example, among households living below the poverty line, about 20 percent have at least one member who has completed an associate degree or more.

Shifts in the labor market help to explain the premium for college degrees. Workers lacking college degrees today are much more likely to be employed in low-wage, low-skilled occupations than were such workers in the past (Kearney, Hershbein, and Jácome 2015). For example, the share of working-age men without high school diplomas who were employed as operators and laborers—with median earnings of $25,500 in 2013—fell from 40 percent in 1990 to 34 percent in 2013. Meanwhile, the share working in food, cleaning, and grounds-keeping occupations—whose median earnings were $20,400 in 2013—almost doubled, from 11 to 21 percent.

Increasing one’s education is a good bet to increase one’s earnings. In an empirical simulation, Hershbein, Kearney, and Summers (2015) test what would happen to earnings if one-tenth of the men without a college degree were to obtain one. hey find that the decline in median earnings observed over the past three decades would be nearly erased, and earnings at the 25th percentile of the distribution would be 43 percent higher than they are today, thus helping to reduce inequality in the bottom half of the distribution.

Conditional on attaining a bachelor’s degree, the choice of what to study is also an important predictor of lifetime earnings. For the median graduate, total lifetime earnings range from close to $800,000 to more than $2 million depending on the major, as shown in a Hamilton Project economic analysis (Hershbein and Kearney 2014). Although there is substantial variation in earnings within majors, those that emphasize quantitative skills—such as engineering, computer science, economics, and finance—tend on average to have graduates with the highest lifetime earnings, while majors that train students to work with children or provide counseling services—such as early childhood education, elementary education, and social work—tend to have graduates with lower lifetime earnings. There is compelling evidence that effective teachers in early childhood impart large gains to society through their students’ subsequent gains in earnings and other life outcomes (Chetty et al. 2011). If low pay for teachers in early education dissuades potentially impactful workers from going into the field in the first place, the result could be costly to society in the long run.

Measuring the direct effect of education on economic success is no straightforward task: not only does education improve skills, but students with higher skill levels are more likely to pursue further education. In other words, the education wage premium can be attributed partly to selection (the tendency that students who go on to college already have higher levels of cognitive and soft skills) and partly to treatment (the direct impact of additional years of schooling). As a result, it would be incorrect to infer that giving a college education to someone who would have otherwise obtained just a high school diploma would cause her earnings to increase by the size of the gap between high-school-educated and college-educated workers. However, research that isolates the causal impact of additional schooling does find that additional years of schooling improve outcomes.

To isolate the effects of educational attainment on earnings, Messacar and Oreopoulos (2012) take advantage of changes to state laws governing minimum school-leaving age. As shown in figure 5, one year of additional schooling lowers the probability of a student later being unemployed by 3.6 percentage points, of being on welfare by 5.5 percentage points, and of living below the poverty line by 8.1 percentage points. Among those working at least 25 hours per week, an additional year of compulsory schooling is associated with a 10.7 percent increase in annual earnings. These results may be understated because education earnings gaps tend to increase with age and these results focus only on younger cohorts.

Title I of the Elementary and Secondary Education Act (ESEA) of 1965 provides federal funding to schools either through targeted assistance to low-income students, or through a schoolwide program if at least 40 percent of the school’s students qualify as low income (based on local criteria such as eligibility for the free or reduced-price school lunch program or receipt of means-tested family assistance through the Social Security Act or Medicaid) (§ 1113). The U.S. Department of Education (ED) classifies schools with 40 percent or more low-income students as high-poverty schools (Lippman, Burns, and McArthur 1996, 18). In 2014 a majority of students nationwide—56 percent—attended high-poverty schools. The state with the highest share of students attending high-poverty schools is Mississippi, at nearly 92 percent, and the state with the lowest share is Minnesota, at roughly 17 percent. As shown in figure 6, southern states tend to have the largest shares of students attending high-poverty schools.

School poverty concentrations are a strong predictor of school achievement averages (Kennedy, Jung, and Orland 1986; Murnane and Steele 2007), and concentrations of low-income students lead to a variety of challenges for a state’s public school systems. Research demonstrates that concentrations of low-income students are more expensive to educate across several dimensions (Downes and Pogue 1994; Duncombe and Yinger 2004). For example, schools with high shares of low-income students have higher teacher attrition rates due to difficult working conditions such as larger class sizes and lower-quality facilities (Greenlee and Brown 2009; Murnane and Steele 2007). Schools with high concentrations of poverty also face higher student turnover because low-income students tend to change schools more often (National Center for Education Statistics [NCES] 1996, figures 3.14). Title I of ESEA provides supplemental funding to school districts with large numbers or concentrations of low-income students.

Title I of ESEA is the primary source of federal funding for schools with high concentrations of poverty. The purpose of Title I, particularly its schoolwide program, is to allocate more money to schools with concentrations of poverty. As discussed in Fact 8, low-income students are more likely to struggle in school and require additional support. However, the negative trend in figure 7 shows that states with higher shares of low-income students receive less Title I funding per eligible student. Because Title I is the primary mechanism for the federal government to distribute school funds across states, this negative relationship is counterintuitive and suggests that the funds may not be targeted appropriately in the status quo.

The negative relationship between share of students eligible and funding per eligible student is due in large part to the interaction between Title I’s chronic underfunding and the complex Title I formulas that distort allocations in the absence of full funding. For example, to fully fund the Basic Grants portion of Title I in 2015 would have required $50 billion in allocations; Congress instead appropriated only $6.5 billion. A new Hamilton Project policy proposal by Nora Gordon, "Increasing Targeting, Flexibility, and Transparency in Title I of the Elementary and Secondary Education Act to Help Disadvantaged Students” (2016) discusses the causes of this dilemma and offers potential policy solutions. Gordon explains that a small state minimum leads to states such as Vermont receiving a disproportionate level of Title I funding. Additionally, a provision known as hold harmless, which allows districts to continue to receive allocations based on allocations received in previous years, prevents Title I funding from adapting quickly to structural changes in poverty levels across states and districts.

High poverty concentrations pose a challenge, in part because average test scores among low-income students decline as the share of low-income students increases. The graph demonstrates that, on average, a 10 percentage point increase in the share of low-income students in a state predicts a 2.3 percentage point decrease in the share of low-income students scoring at or above proficient on the National Assessment of Educational Progress (NAEP); a proficient score indicates “solid academic performance,” positioned between the basic and advanced levels (NAEP 2015a). The national average for the share of low-income students scoring proficient or better is 18 percent, compared to 48 percent for the rest of the student population. In states with high concentrations of low-income students, this figure is as low as 7.4 percent.

However, some states with similar shares of low-income students have very different levels of proficiency. For example, Connecticut, Massachusetts, and New Jersey all have a share of low-income students that is just under 40 percent, but there is a difference of nearly 20 percentage points in the share of low-income students scoring at or above proficient. While it is difficult to directly compare school quality across states, Connecticut has a school finance formula that is heavily reliant on local property taxes—leading to more funding for schools in high-value areas—and has been noted as one of the states with the largest achievement gaps between high- and low-income students across a variety of measures (NAEP 2015b; Zimmer and Hodgson 2015).

Average earnings among women with college degrees have increased dramatically over the past 25 years. While the average female college graduate working full-time earned an inflation-adjusted $56,204 in 1993, by 2014 her earnings had grown almost 20 percent to $67,088. Teacher pay, on the other hand, has been stagnant over these years, dropping slightly from $58,048 to $56,689. As shown in figure 9, both college-educated women overall and school teachers in particular saw their wages decline through the Great Recession.

Although not all teachers are female, teaching remains the second-most popular occupation for women (U.S. Department of Labor [DOL] 2014), and three-quarters of public school teachers are female (NCES 2015a). As alternative labor market opportunities for women have expanded, teaching has become a relatively less attractive career path (Murnane and Steele 2007). The average starting salary for an education major in 2014 was $40,267, the only category to see a decline from the previous year (National Association of Colleges and Employers [NACE] 2014). This stands in stark contrast to 1960, when teachers earned on average 13 percent more than their fellow female college graduates, and even in contrast to 1980, when teachers earned 4 percent more (Hurley 2013).

Earlier work found that salary increases during the 1980s did not improve the quality of new teachers entering the profession (Ballou and Podgursky 1997). Since then, though, the accountability movement has realigned incentives so that schools today are more likely to be evaluated by student performance. As argued in Duncan and Murnane (2014), a well-designed accountability system can promote improvements in school practices, such as greater willingness to use resources and to work together in new ways to better promote students’skill development. As a result, salary increases may matter more today than they did in the past.

In a recent poll, 60 percent of undergraduate students reported some interest in a teaching career. When asked what policies might induce them to join the profession, the top responses included increasing pay for every teacher and, in particular, increasing pay for high-performing teachers (Hiler and Hatalsky 2014). The pay and prestige disincentives among the best and brightest to teach may pose an especially large problem for high-poverty schools, where work environment challenges associated with teaching in such schools, including lower parental engagement and leadership instability, further compound the opportunity costs to becoming a teacher (Simon and Johnson 2013). Emphasis on school accountability also increases pressure on teachers to improve student test scores and serves as a disincentive to working at a low-performing school (Murnane and Steele 2007). Among those who chose to teach directly after college, those who scored in the highest quarter on college entrance exams were less likely to be observed in the profession 10 years later than were those who scored in the lowest quarter; 13 percent of teachers reported leaving the profession altogether due to low pay (Alt and Henke 2007).

There is a large degree of variation in instructional spending across states. Instructional spending includes salaries and benefits for teachers, textbooks, supplies, and other purchased services for extra- and cocurricular activities (NCES 2012b). While schools in the Northeast spend $8,000 to $13,000 or more per student on instruction, schools in western and southern states typically spend much less (NCES 2012a). After adjusting for differences in living costs across regions (as proxied by variation in average earnings across states), the spending gaps narrow slightly but remain: northeastern states spend almost twice as much as western and southern states.

States predominantly rely on income and sales taxes for education funding, which can provide wealthier states more resources for education (Oliff and Leachman 2011). As a result of varying resources and preferences for education across states, a student in Vermont may receive as much as $95,000 more in instructional expenditures from kindergarten through 12th grade (adjusted for cost of living) than a student in Utah (NCES 2012a). Ensuring that students have equal access to a quality education across all states is a key rationale for federal involvement in school funding (Rentner 1999; ED 2012).

It is worth noting that spending also varies substantially within states and even within school districts (see Fact 11). The Every Student Succeeds Act (ESSA) of 2015 introduced requirements to report school-level spending, which will bring unprecedented transparency to these spending differences.

Federal, state, and local education spending for public elementary and secondary schools totaled $620 billion in 2012, the most recent year for which data are available. Although the population of students has grown by 40 percent since 1960, per-pupil spending has increased by 274 percent (NCES 2014a). Federal education funds flow to states primarily through grant programs, such as funding for low-income students through Title I of ESEA and special education grants through the Individuals with Disabilities Education Act (IDEA) of 1990. The federal role in funding education grew with the 2001 reauthorization of ESEA, known as the No Child Left Behind Act of 2001, and temporarily increased further with the economic stimulus package in 2009 through additional grants for Title I and IDEA, along with State Fiscal Stabilization Fund grants for education (ED 2009). Despite increases in the federal share of spending, from 6 percent in 1990 to a peak of 13 percent in 2010 and back to 10 percent in 2012, the vast majority of school funding comes from state and local levels.

Local school funding is predominantly raised from property taxes, which make up 81 percent of total school revenue (NCES 2015c). Since wealthy families tend to live in affluent communities, increasing the tax base and revenues for local schools, their children’s per-student spending is typically higher in these districts than in poor districts. As a result, revenues raised for school spending can differ dramatically due to differences in property values and tax rates. In Texas, for example, the Fort Sam Houston school district receives only $265 in local revnue per student, while the neighboring Alamo Heights school district receives $13,007 in local revenue per student (NCES 2012a).

In cases where variations in property values lead to wide disparities in local revenue for educational spending, state and federal funds can serve to offset these differences. Beginning in the 1970s, many states have reformed their school finance systems to address this inequality. Often reacting to mandates from courts that found local finance systems unconstitutional, states have moved away from funding based primarily on property taxes and have implemented state aid formulas that direct more money to low-income and low-tax-base school districts (Lafortune, Rothstein, and Schanzenbach, 2016). Once state and federal spending are included in the example of Texas school districts above, the gaps are reduced dramatically: total revenues per student in Fort Sam Houston are $14,640, compared with $15,607 in Alamo Heights.

On average, a black child entering kindergarten scores 0.66 standard deviations in math (and 0.40 in reading) below her white peers (Fryer and Levitt 2006). By comparison, the income achievement gap is even worse: upon entering kindergarten, the difference in scores between a child from a household in the 10th percentile of the income distribution and a child from a household in the 90th percentile was more than two times greater than the black–white achievement gap (Reardon 2011). Furthermore, the income achievement gap remains as the child advances through school (Reardon 2011).

The persistence of these gaps suggests that intervening early in life may have lasting benefits, and that a number of early-life interventions, as shown by figure 12, may help children to develop the foundational cognitive and emotional skills needed to successfully complete later milestones, such as high school graduation (Murnane 2013).

Three programs focused on preschool-aged children have substantially improved high school graduation rates, especially among students at greater risk of dropping out. Head Start, the federally funded program targeting poor children between the ages of three and five with preschool, health, and social services, increased high school graduation rates among recipients by up to 20 percentage points, depending on the cohort and demographic group studied (Deming 2009; Garces, Thomas, and Currie 2002; and Ludwig and Miller 2007). Head Start particularly benefited black and Hispanic males, increasing their high school graduation rates by 5.5 and 18.1 percentage points, respectively. Furthermore, it is estimated that the Perry Preschool intervention, which provided high-quality preschool to black children, had positive effects on a number of outcomes, including increased earnings and lower rates of crime among participants. Preschool attendees graduated high school at rates 20 percentage points higher than nonattendees (Schweinhart et al. 2005). Targeting somewhat older children, the Tennessee Student/Teacher Achievement Ratio (STAR) experiment investigated the effects of being assigned to a small class size from kindergarten through third grade (Chetty et al. 2011). Black and low-income males experienced particularly sizeable gains in high school graduation rates from assignment to smaller classes (Murnane 2013).

In addition, early-life interventions outside of school settings have been shown to increase high school graduation rates. Recent work on cohorts born in the 1960s and 1970s has shown that access to the Food Stamp Program from the time of a child’s conception through age five increased high school graduation rates by 18 percentage points (Hoynes, Schanzenbach, and Almond 2016). The success of these childhood programs points to the lasting benefits of intervening well before students enter high school.

There is strong evidence that a variety of high school‑age interventions, most of them targeting students from low-income and minority backgrounds, have successfully increased high school graduation rates. Figure 13 graphs the percentage point increase in the high school graduation rate for several high school interventions, with the per-student net cost of the intervention displayed below the bars when applicable. Note that we do not mean to suggest that these are the only effective interventions, but instead that these are the interventions that have been rigorously evaluated.

Many of the interventions in the figure share the theme of expanded school choice (i.e., improving outcomes through the channel of where students go to school). They include school lotteries in Charlotte-Mecklenburg County Schools (which improved high school graduation rates by 14 percentage points), charter high schools in Florida and Chicago (7–15 percentage points), and the District of Columbia’s school voucher program (21 percentage points). Net costs for each of these interventions are hard to estimate, but are generally thought to be modest. New York’s small schools initiative has been successful in raising graduation rates (9 percentage points) particularly among black males (11 percentage points, or from 31 to 42 percent), a group that has historically had some of the lowest graduation rates (Bifulco, Unterman, and Bloom 2014). The additional cost per student per year of the small schools initiative was about $850, and falls to nearly zero when accounting for differences in teacher salaries and student populations (Unterman 2014).

Other interventions have focused on improving outcomes within schools. These interventions tend to be more limited in scope. Double-dose algebra, a program at Chicago Public Schools, requires students whose eighth grade math test scores place them below the national median to enroll in two algebra courses when they begin high school—regular algebra and an algebra support class (Cortes and Goodman 2014). At an estimated cost of less than $1,000 per student, students enrolled in the double-dose classes increased their graduation rates by 2.6 percentage points; however, these effects were only found for the highest-scoring students subject to the intervention, who scored just below the national median (Cortes and Goodman 2014). An intensive mentoring program at 11 high schools, most of them urban, has also improved graduation rates among low-performing youth (Rodríguez-Planas 2012).

An important recent paper documents the impact of court-ordered school finance reforms in the 1970s and 1980s on spending and subsequent student outcomes (Jackson, Johnson, and Persico 2016). A 10 percent increase in per-pupil spending during each year of schooling increases high school graduation rates by 10 percentage points. Although the additional funding was generally unrestricted, the study finds that school spending increases were associated with decreases in student-to-teacher ratios and increased teacher salaries.

Similar to the school choice interventions, it is difficult to extrapolate whether these programs would still be effective if they were more widely adopted both within and across school districts. Murnane (2013) highlights the central challenge facing education reformers: because of their limited scale, modest interventions—such as double-dose algebra—are more straightforward to implement and are easier to rigorously evaluate, but they do not address the underlying challenges facing less-advantaged students. Large-scale interventions, such as New York’s small schools initiative and some charter school models, may show some success at addressing the underlying challenges but are difficult to successfully replicate in other contexts. Improving the methods of evaluation and designing interventions for replicability are crucial to understanding what additional approaches work for improving high school graduation rates.

Charter school enrollment has grown steadily over the past two decades, from about 400,000 students (1 percent of the student population) in 2001 to 2.3 million (4.5 percent of the student population) in 2013 (NCES 2015b). Charter schools differ from traditional public schools because they are subject to fewer regulations and typically receive a fixed funding amount per pupil (though total expenditures per pupil vary due to philanthropic support). Charter schools have greater scope to innovate, employing a variety of methods, such as different curricula, alternative class schedules including longer school days and “Saturday school” to increase instructional time, high-dose tutoring, and other approaches (Angrist et al. 2010; Fryer 2012).

This new approach has yielded some success: many urban charter schools are able to significantly improve test scores in math and English in one year. As shown in figure 14, middle school students attending urban charter schools in Massachusetts and New York have increased their math scores by roughly one-tenth to one-third of a standard deviation relative to their peers who attended traditional public schools. Such gains are substantial: in the case of the Harlem Children’s Zone charter schools, if the same gains were achieved for three years it would be enough to eliminate the black–white achievement gap (Dobbie and Fryer 2011).

However, evaluations of charters at the national level show that, on average, charter schools perform no better than traditional schools (Gleason et al. 2010). In Massachusetts, for example, nonurban charter schools appear to reduce test scores, which may be due to differences in curriculum or student demographics (Angrist, Pathak, and Walters 2012). These results suggest charter school expansion should be considered carefully, factoring in the composition and needs of the district’s student population, as well as the relative success of some charter school models over others.