Experimental Estimates of Impacts of Cost-Earnings Information on Adult Aspirations for Children’s Postsecondary Education

ABSTRACT Economic information may close aspiration disparities for postsecondary education across socioeconomic, ethnic, and partisan divides. In 2017, we estimated impacts of information on such disparities by means of a survey experiment administered to a nationally representative sample of 4,214 adults. A baseline group was asked whether they preferred a 4-year degree, a 2-year degree, or no further education for their oldest child younger than the age of 18 years (or the option they would prefer if they had a child younger than 18 years). Before 3 other randomly selected segments of our sample were asked the same question, they were given either information about (a) both net costs and returns, (b) net costs, or (c) returns to a 2-year and 4-year degree. Information about both costs and returns did not reduce socioeconomic-status disparities but did affect ethnic and partisan divides. The findings suggest that reductions in socioeconomic inequalities in educational opportunity require more than simple changes in the dissemination of information aimed at altering economic cost–benefit calculations. Sustained effort that mitigates deeper-seated cultural and social barriers seems necessary.

In this article, we present results from a survey experiment administered in 2017 to a nationally representative sample of U.S. adults that estimated information effects on their preferences for a 4-year degree, a 2-year degree, or no further education for their oldest child younger than the age of 18 years. If they had no such child, we inquired about preferences they would have for such a child. We treated three randomly selected, similarly sized segments of a sample of 4,214 adults with economic information about (a) both returns and net costs, (b) only net costs, or (c) only returns to 2-year and 4-year degrees before asking them about their higher education preferences for their child. A fourth segment, which received no information before the question was posed, served as the control group.
Two recent studies (Bleemer & Zafar, 2018;Lergetporer, Werner, & Woessmann, 2018) estimated information impacts on adult aspirations for their children's further education. However, neither study provided respondents with a 2-year associate degree response option, even though that degree is a prominent feature of U.S. higher education. In addition, neither study gave respondents information about both net costs and returns simultaneously before ascertaining adult aspirations. Each provided information on either costs only or returns only. The experimental results presented here are the first to be based on a survey administered to a nationally representative sample that provided adults with balanced information on both net costs and returns for both 2-year and 4-year degrees.
Our findings were mixed. Balanced information about both economic costs and returns to college degrees did not narrow the aspiration gap between those from higher-and lower-SES backgrounds. There was suggestive evidence that economic information closed the aspiration gap between Hispanic and White respondents, and information essentially eliminated the gap between Democrats and Republicans. Taken together, the findings suggest that reductions in SES inequalities in educational opportunity require more than simple changes in the dissemination of information aimed at altering economic cost-benefit calculations. Sustained effort that mitigates deeper-seated cultural and social barriers seems necessary.
Theoretical framework and literature review Hossler and Gallagher (1987) identified three phases in the process of pursuing a postsecondary education: (a) predisposition, (b) search, and (c) choice. Our work was focused on investigating the importance of information at the predisposition phase, the time at which aspirations are acquired. These aspirations are influenced by cultural norms, knowledge, expectations, habits (Bourdieu, 1986;Park & Hossler, 2014;Perna, 2006;Perna & Titus, 2005;Serna, 2015), social networks (Coleman, 1988), and economic factors, including perceived credit constraints, debt aversion, and discount rates (Boatman, Evans, & Soliz, 2017;Toutkoushian & Paulsen, 2016). All these factors contribute toward wide SES disparities in college enrollment and completion rates (Cabrera & La Nasa, 2001;Rowan-Kenyon, 2007).
Socioeconomic-status attainment disparities Chetty et al. (2017) found serious gaps by SES among students born during 1980 to 1982. Only half of those born into households with income at the first decile of the distribution enrolled in college by age 28 years, while more than 90% of students whose household incomes were at the 90th percentile enrolled by that age. Similarly, Bailey and Dynarksi (2011) reported that the share of students who enrolled in college among those born in households with an income in the top quartile of the distribution was 51 percentage points higher than the share of students who enrolled in college among those born in households with incomes in the bottom quartile. Among respondents to the General Social Survey born in 1966 to 1977, only 12% of White men and only 17% of White women born of parents without a college background completed college. If both parents had attended college, those percentages were 62% and 64% for men and women, respectively (Buchmann & DiPrete, 2006).
African American and Hispanic high school students were also less likely to enroll in higher education institutions immediately after graduation, though the Hispanic-White divide has been diminishing. According to the U.S. Department of Education, the rate for African Americans, as of 2015, remained at 63%, a level that persisted since the fall of 1998. Meanwhile, the rate for Hispanics grew from 52% to 67% from 1990 to 2015. By comparison, 87% of Asian students and 70% of White students were enrolled in a postsecondary institution in 2015 (Musu-Gillette et al., 2016).
Among students who have postsecondary aspirations, there have been sizeable SES disparities in the types of colleges to which they apply. Compared with higher-income peers, high-achieving students from lowincome backgrounds were more likely to apply to 2-year and nonselective 4-year postsecondary institutions than to selective 4-year institutions (Bowen, Chingos, & McPherson, 2009;Ovink, 2017). This disparity occurred even when the availability of financial aid packages reduced the net costs of attending selective 4-year institutions below the costs incurred at less-selective institutions (Hoxby & Avery, 2013).

Determinants of aspiration gaps
Socioeconomic-status divides have been both wide and deeply rooted in social and economic realities. Based on previous research, we theorize that they can be ameliorated only by intensive interventions that go well beyond the provision of small doses of information administered in the course of a survey. However, minimal amounts of information can eliminate ephemeral divides such as those between political partisans.

Social and cultural influences
Many scholars have explored the ways that SES backgrounds affect educational outcomes (Duncan & Murnane, 2011;Egalite, 2016;Magnuson, Rosenbaum, & Waldfogel, 2008). Children exposed to lower SES environments are at greater risk for traumatic stress and other medical problems that can affect brain development (Nelson & Sheridan, 2011). Better-educated mothers speak more frequently with their infants, use a larger vocabulary when communicating with their toddlers, and are more likely to use parenting practices that respect the autonomy of a growing child (Guryan, Hurst, & Kearney, 2008;Hoff, 2003). Better-educated and higher-income families have access to more enriched schooling environments (Altonji & Mansfield, 2011) and are less likely to live in extremely impoverished communities burdened with high violent crime rates (Burdick-Will et al., 2011). All these factors and other childhood and adolescent experiences create profound disparities in academic preparation by SES, which in turn lead to concomitant differences in college aspirations (Goyette, 2008;Jacob & Linkow, 2011;Kao & Tienda, 1998;Perna, 2006).
Parental desires to maintain their social status also help to explain why children from higher SES families have higher postsecondary attainment rates (Bourdieu, 1986;Sewell, Haller, & Portes, 1969). Expectations of families and peer networks influence students' own ambitions with respect to higher education (Park & Hossler, 2014). Parents are among the first to suggest to their children that they pursue a higher degree and play a key role in facilitating college choice (Hossler, Schmit, & Vesper, 1999;McCarron & Inkelas, 2006). Hossler and Stage (1992) estimated that raising parental aspirations by 1 standard deviation increased student aspirations by about 0.45 standard deviations. Somers, Cofer, and VanderPutten (2002) found that children whose parents expected them to pursue any postsecondary education were 2 percentage points to 10 percentage points more likely to actually do so compared with children whose parents did not have such expectations (see also Perna & Titus, 2005). Jacob and Linkow (2011) provided causal evidence that student expectations were both influenced by SES and had their own independent impact on degree attainment.
The effects of neighborhood quality on educational attainment suggest the importance of the broader cultural milieu and social resources of parents and other adults that may influence students directly or indirectly (Burdick-Will et al., 2011;Stewart, Stewart, & Simons, 2007;Wilson, 1987). Students' educational aspirations are affected by the perceptions of adults besides their parents, such as their teachers (Gershenson & Papageorge, 2018).

Economic influences
Low SES families may additionally perceive financial constraints that dissuade them from pursuing further education. Cunningham and Santiago (2008) suggested that low-SES students perceive credit constraints that prevent them from borrowing to cover college costs, and Caetano, Palacios, and Patrinos (2011) found that respondents in three Latin American countries were averse to the labeling of student loans as debts. However, Boatman et al. (2017) reported that low-income students and adults in the United States were not more risk-averse to borrowing for educational purposes compared with their higher-income counterparts.
Long-term discount rates to returns from higher education investments may vary by SES. Parents who have earned a 4-year college degree are more likely to be aware of its value and therefore are better able to communicate the long-term value of a bachelor's degree to their children (Kao & Tienda, 1998;Perna & Titus, 2005). Meanwhile, many lower-SES families have been less familiar with the necessary information on the net cost of college after financial aid packages are considered, the economic and other returns to different types of degrees, and the range of available postsecondary options (Hoxby & Avery, 2013;Hoxby & Turner, 2015). Students from lower-SES backgrounds have been more likely to borrow money for their further education (Cunningham & Santiago, 2008). So returns need to be ample if they are to cover borrowing costs in addition to direct tuition, boarding costs, and opportunity costs.

Partisan affiliation
By contrast, expectations for a child's further education held by political partisans are likely to be more malleable. Recently, Democratic and Republican leaders have taken contrasting positions on the relative value of a 2-year and 4-year college degree (See Online Appendix A). Given these differences between partisan elites, we expected, in the absence of information, to find contrasting preferences for 2-year and 4-year colleges between Democrats and Republicans. However, partisan differences in college-going preferences have been more likely to reflect the political conversations of the day than to be so strongly rooted in a citizen's partisan identity that they are impervious to information (Zoller, 1992). Although partisan affiliations for some Americans are lifelong and the opinions of partisans on some issues (e.g., abortion, taxes, entitlements, and spending levels) have persisted for decades, political opinions in the United States are noted for their inconstancy across time and their inconsistency from one issue to another (Converse, 1964;Zoller, 1992). Partisan affiliations themselves have been found to be quite malleable (MacKuen, Erickson, & Stimson, 1989). Splitticket voting-ballots simultaneously cast for candidates with different party affiliations-is so widespread that control of legislative and executive branches of government are frequently divided between the two political parties (Fiorina, 1992).

Information interventions
Several interventions have been designed to ameliorate the SES-aspiration connection. The use of mentors and other resources to guide students through the decision-making and application process have resulted in increased enrollments and shifts to more selective, possibly higher-quality institutions (Avery, 2010(Avery, , 2013Avery & Kane, 2004;Bos et al., 2012;Carrell & Sacerdote, 2017;Castleman, Arnold, & Wartman, 2012;Castleman & Goodman, 2018;Castleman, Page, & Schooley, 2014;Oreopolous, Brown, & Lavecchia, 2017;Oreopolous & Ford, 2016). Researchers have also observed increases in enrollment rates from interventions that provide help with completing financial aid and other administrative paperwork (Bettinger, Long, Oreopoulos, & Sanbonmatsu, 2012). Interventions that have provided high-achieving, low-SES students with information about net costs and benefits have encouraged students to apply to more selective institutions (Hoxby & Turner, 2015;Oreopolous & Dunn, 2013). However, Bos et al. (2012) reported no significant impact of a peer-to-peer advising intervention on postsecondary enrollment overall, though they did find a greater utilization of 4-year colleges in California. Avery (2013) identified no overall effects on college enrollment from a 2-year afterschool intervention in St. Paul, MN high schools, though he did find greater enrollment in 4-year institutions rather than 2-year institutions.
The costs of executing many successful information interventions, which are often coupled with intensive mentorship, have precluded investments that could bring them to scale (Carrell & Sacerdote, 2017). But whether less expensive information-based interventions administered to adults can ameliorate deeply entrenched SES disparities in college-going aspirations for children remains an open question (Carrell & Sacerdote, 2017). Bleemer and Zafar (2018) reported that in the United States, small doses of information about the returns to a four-year degree administered by means of a survey significantly closed the aspiration divide, but Lergetporer et al. (2018) found that in Germany, this information had no such impact.

Methods
In the remainder of this article, we shed new light on this topic by presenting the results of a survey experiment that asked adults about their aspirations for their child after we treated randomly selected groups with balanced information on 2-year and 4-year college returns and net costs not provided to a control group.

Data
Our experiment was based on data collected from the Education Next annual survey of American public opinion on education, which has previously been used in scholarly education research (Barrows, Henderson, Peterson, & West, 2016;Chingos, Henderson, & West, 2012). The survey was conducted during May 5, 2017, to June 7, 2017, by the polling firm Knowledge Networks® (KN), a GfK company. Knowledge Networks® maintains a nationally representative, probability-based online panel of 55,000 adults identified through addressbased sampling techniques (for a detailed, empirical assessment of online surveys administered by KN, see Chang & Krosnick, 2009). Sample members without Internet access are provided with a computer and free Internet access. For the 2017 Education Next survey, a subset of 4,214 adults was randomly selected from the full sample of 55,000 adults included in the KN database. Two groups of particular relevance to this study-2,170 parents of children younger than the age of 18 years living in their home and 805 Hispanic respondents-were oversampled. Survey weights, based on demographic and other background information originally collected upon entry into the KN panel, were employed to account for nonresponse and the oversampling of parents and Hispanic respondents to ensure that any estimates based on the Education Next survey sample were nationally representative.

Experimental design
Unweighted and weighted summary statistics for the sample are shown in Table 1. We divided the sample into the following randomly assigned groups: (a) a control group, (b) the net-cost and returns treatment group, (c) the netcost treatment group, and (d) the returns treatment group. All respondents in the control group who had a child younger than the age of 18 years in their household were asked: "Thinking about your oldest child younger than the age of 18 years, would you want your child to go to a community college to earn a 2-year degree, a university to earn a 4-year degree, or neither?" Other adults without school-aged children were asked the same question, except they were asked about a hypothetical child. Before these questions were posed to other respondents assigned to one of the treatment groups, they were given information on either or both the net costs and returns to 2year and 4-year degree programs. See Online Appendix B for exact wording. Estimates of returns and net costs were based on national averages provided by the College Board (2016;Ma, Pender, & Welch, 2016). The cost treatment group was told only the net costs of each postsecondary option, while the returns treatment group was told only the returns to each postsecondary option. Those respondents given balanced information were told both the net costs and returns.
Because respondents were randomly assigned to control and treatment groups, any differences in respondent aspirations across the four groups were almost certainly attributable to differences in the information each group received. This experimental design provides the study with a high degree of internal validity.

Equivalence between treatment and control groups
The experimental design was implemented with fidelity. In Table 2, we present tests for covariate balance across all treatment conditions. In Columns 6 and 7, we report F statistics and p values for joint tests of covariate balance. We consistently failed to reject the hypothesis that means across the four treatment conditions would be equal at the .05 level. Tests for differences between the control group and each of the other three treatment conditions revealed four statistically significant differences at the .05 level, approximately the number that was expected to occur by chance. In subsequent tables, we report results with and without demographic controls.

Empirical strategy
We estimated the effects of information on postsecondary aspirations by comparing the responses of each treatment group to those of the control group. We estimated effect heterogeneity by interacting each treatment condition with various subgroups. This approach is depicted in the following model and was estimated using multinomial logit: In the equation, Y i is a three-level categorical variable indicating whether respondent i desires their child to attend a 4-year university, 2-year community college, or neither. T i represents one of the three information treatment conditions: costs and returns, only costs, or only returns. And G i is an indicator for one of several subgroups that we considered in our tests of effect heterogeneity. Following the theoretical role that SES plays in postsecondary aspirations, we focused on three indicators of SES: the respondent's household income, educational attainment, and ethnicity. We additionally examined effects by political affiliation, given partisan differences in perceptions of higher education among political elites. X i is a vector of variables to control for a respondent's gender, ethnicity, household income, employment status, educational attainment, political affiliation, marital status, number of children younger than 18 years, U.S. Census region of residence, and whether the respondent lived in a metropolitan area. Given that perceptions of both parents and other adults shape children's postsecondary aspirations, it is important to know whether information alters the cultural envelope within which parent opinions are embedded as well as parental opinions themselves (Burdick-Will et al., 2011;Gershenson & Papageorge, 2018;Stewart et al., 2007;Wilson, 1987). We therefore ran the analyses for all respondents, though we did control for parent status in our models.
We employed survey weights and heteroskedastic robust standard errors in all estimations. We alternatively ran sets of linear probability models where we separately considered the three possible outcomes. That is, the dependent variable was one of three binary variables indicating whether the respondent preferred a 4-year university, 2-year community college, or neither instead of the single three-level categorical variable as described in the equation.

Results
Although the multinomial model is econometrically preferred, the presentation of results relies on linear probability models, because results from both sets of econometric models did not differ substantively and the latter was easier to interpret. Results from the full set of multinomial models are available in Online Appendix C. Unless otherwise stated, we discuss results that include controls for the demographic background characteristics given in Table 1 as well as the respondents' marital status, number of children, and employment status.

Parents and other adults
About 79% of U.S. parents preferred their child attends a 4-year university, while nearly 13% of parents desired a 2-year community college and 8% did not wish their child to receive any postsecondary education (Table 3, Panel A). Other adults chose the 4-year option less frequently by 11 percentage points, the 2-year option more frequently by 8 percentage points, and the no-further-education alternative more frequently by an insignificant 3 percentage points. Though almost every adult wanted their child to pursue postsecondary education, they clearly had different conceptions of what postsecondary education entailed. Most had four-year institutions in mind, but many others had two-year institutions in mind.
Despite these differences between parents and other adults, treatment effects for the two groups were similar. As shown in Panel B of Table 3, providing parents with balanced information about both costs and returns lowered the preference for a 4-year university by 6 percentage points, a marginally significant impact. In other words, balanced information had a seemingly perverse impact that actually reduced the amount of further education parents preferred. The impact of the treatment on other adults did not differ significantly from impacts on parents.
The cost of a 4-year degree seemed to be driving the parent results. When netcost information alone was provided (see Panel C), parents were nudged away from a preference for a 4-year degree. The drop was a significant 8 percentage points. That information shifted parental preferences toward the 2-year community college option by about 6 percentage points, with the remainder selecting no further education. The statistically insignificant interaction suggested that cost information affected other adults in much the same way.
Turning to Panel D, we found evidence that parents were insensitive to returns information when presented alone. Parental preference for a 4year university bumped upward by a statistically insignificant 1 percentage point. However, preferences among other adults for a 4-year degree increased by an additional 9 percentage points when only this information was given. Under these circumstances, the opinions of other adults moved away from preferring the 2-year option by 13 percentage points, a significant change. All things considered, treatment with balanced information did not increase the demand for further education on the part of either parents or other adults. When both net costs and returns were shared with respondents, the preferences may have even shifted parents away from selecting the 4-year university option, though this effect was of borderline significance. Parents' aspirations moved toward the 2-year option when they were told about costs, while other adults opted for the 4-year option when they were told about earnings. Importantly, the impact of balanced information was not statistically different for parents and other adults, though we failed to reject the null hypothesis that the overall treatment impact for other adults was different from zero.

Heterogeneous effects
Because effects were similar for parents and other adults when they were provided with balanced information, we followed Bleemer and Zafar (2018) and Lergetporer et al. (2018) and combined these respondents (but controlled for parent status) when estimating impact heterogeneities. This approach enlarged the number of observations in each estimation, thereby enhancing the analytical power to detect differential information impacts across SES and other subgroups. Moreover, both the perceptions of parents and other adults shape student aspirations (Burdick-Will et al., 2011;Gershenson & Papageorge, 2018;Stewart et al., 2007).

Household income
Seventy-nine percent of high-income earners in the control group preferred their child attend a 4-year university (Table 4, Panel A). The percentages for middle-and low-income earners in the control group were 10 percentage points and 26 percentage points lower, respectively. 1 The 26 percentage-point aspiration gap between the high-and low-income groups was nearly half the enrollment differential between the top and bottom quartiles of the income distribution reported for the 1979-1982 birth cohort tracked by the National Longitudinal Survey of Youth (Bailey & Dynarksi, 2011).
Despite the large aspiration disparities by income, the provision of information on both costs and returns to respondents did not reduce the SES gap (Table 4, Panel B). In fact, middle-income respondents, when treated with balanced information, shifted away from the 4-year option, while high-income respondents shifted toward it. Although neither shift was by itself statistically significant, the difference between the two income groups enlarged by 14 percentage points, a significant and seemingly perverse change. Cost information, when given alone, also had a differential impact. Middle-income respondents were 10 percentage points less likely to choose the 4-year option and to shift to the 2-year option. Returns information, given alone, shifted responses of middle-income and low-income respondents away from the 2-year option by 10 percentage points and 15 percentage points, respectively. But this shift did not significantly narrow the SES gap. Most importantly, the differential impact of balanced information on both net costs and returns was not statistically significant for any income group.

Educational attainment
As shown in Table 5, Panel A, 88% of respondents in the control group who held 4-year degrees preferred the same university option for their children, while those with only some or no college experience chose this option less frequently-by 20 percentage points and 28 percentage points, respectively. Despite these large disparities, we found no treatment effects of balanced information about college costs and returns (Panel B) nor was there evidence of any effect of heterogeneity on adult aspirations by educational attainment. None of the effects were statistically distinguishable across subgroups, and overall effects by subgroups were not statistically significant.
Providing net-cost information alone shifted preferences away from the 4year option by a statistically insignificant 5 percentage points for respondents with 4-year degrees (Panel C). Adults with lower levels of educational attainment moved away from the 4-year university option by an additional 4 percentage points when they were given only cost information. Although Note. N = 4,204; 1,036 respondents without a college degree, 1,172 respondents with some college experience, 1,996 respondents with at least a bachelor's degree. Sampling weights included. Heteroskedastic robust standard errors used. **, * , and + indicate the coefficient is statistically distinguishable from 0 at the .01, .05, and .1 levels, respectively. Superscript letters a, b, and c indicate that the treatment effect for respondents who attended some college or did not attend college is statistically distinguishable from 0 at the .01, .05, and .1 levels, respectively.
this differential effect among those with lower educational attainment was not statistically significant, the total effect of 9 percentage points was marginally significant at the .1 level.
In contrast, providing returns information raised preferences for the 4-year option by 6 percentage points among respondents with a 4-year degree (Panel D). This change in educational aspirations stemmed mostly from a shift away from a preference for the 2-year community college option. Respondents with some college experience also increased their preference for the 4-year option by 9 percentage points when provided with returns information-a result that was marginally significant at the .1 level and stemmed from an equally large shift away from the 2-year community college option. Providing returns information did not significantly raise preferences for the 4-year option among respondents who did not attend college. However, providing returns information for this group shifted preferences away from the 2-year community college option by 16 percentage points and surprisingly raised preferences for not pursuing higher education by 8 percentage points (Panel D). Summing up the results for income and education, the two best indicators of SES in our data set, we did not find that the SES aspirations gap closed with treatment that provided respondents with balanced information. The only significant impact of this treatment condition was that it widened the gap between high-and middle-income groups.

Ethnic background
Sixty-eight percent of Whites in the control group wanted their child to pursue a bachelor's degree (Table 6, Panel A). In the models unadjusted for demographic characteristics, this proportion was 10 percentage points less than the percentage of African Americans who had this preference, a large difference but one that was not statistically significant given the small number of observations available for African Americans. Whites were 14 percentage points more likely than Hispanic Americans to prefer the 4-year college degree option, a difference that was statistically significant but not robust to the inclusion of control variables.
For White respondents, treatment effects fell roughly along the same lines as those observed for SES subgroups. When balanced information on both costs and returns was provided, White respondents did not differ significantly from their peers in the control group (Panel B). Information about costs nudged adults away from the 4-year college degree option by 9 percentage points (Panel C), while information on expected returns moved them toward the bachelor's degree option by 6 percentage points (Panel D).
Among Hispanic respondents, a somewhat different pattern of treatment effects was observed. Information about both costs and returns altered their views, as compared with those of Whites, toward selection of the 4-year degree option by 17 percentage points in the model without controls (Panel B). When SES and other demographic controls were introduced, the impact attenuated to 12 percentage points, a difference that fell just short of statistical significance. However, the 16 percentage-point shift away from the 2-year community college option, even after controls had been introduced, did pass the significance threshold. These shifts seemed to have been driven by information on earnings rather than costs, for which we could identify no significant treatment effects. But earnings returns information spurred a shift to the 4-year college option that was 10 percentage points higher among Hispanics than among Whites (Panel D). That shift was large enough to close the Hispanic-White gap that had been observed within the control group. The same pattern was found for results from models that did not include control variables.
The African American sample size was too small to detect anything other than very large treatment effects, and accordingly, we did not observe significant treatment impacts on African American aspirations when only costs and when only returns information were provided. However, balanced information on both costs and benefits reduced the percentage of African Americans wanting a 4-year college degree for their child by nearly 18 percentage points even after SES and other characteristics were controlled, a statistically and substantively significant shift (Table 6, Panel B). We nonetheless urge caution in interpreting this finding because only 59 African American respondents were in this treatment group.

Political affiliation
Turning to results by political affiliation in Table 7, we observed that Republicans in the control group were about 16 percentage points less likely to choose the 4-year college option relative to Democrats. The results cannot be attributed simply to the lower-SES background of Republicans, because they were robust to the inclusion of background characteristics. Treatment altered the gap between Republicans and Democrats. Providing costs and returns information lowered the preference for a 4-year university by 8 percentage points among Democrats, with most shifting to the no-postsecondary education option (Panel B). This information shifted Republican aspirations in the opposite direction by 8 percentage points, effectively eliminating the aspiration gap across political partisans. Costs may be more salient for Democrats: We observed a 12 percentage-point shift away from the 4-year university option among Democrats given only cost information. No change was observed among Republicans when they were given this information. Returns seemed more salient for Republicans, who shifted their aspirations toward the 4-year university option by 12 percentage points after being provided with information on returns alone.

Summary and interpretation
The capacity of information to reduce group differences in college aspirations of adults for their children depends on the structure of the division. If gaps are embedded in enduring social and cultural relationships, small doses of cost-benefit information intended to correct informational asymmetries, such as those provided in a survey, are unlikely to eliminate them. When balanced information was provided, those with less education and lower income remained much more likely to prefer the no-college option or the 2-year community college option compared with those with more education and higher income. Information may have little impact because, as Bleemer and Zafar (2018) found, the facts seem just as well known to those on both sides of the SES divide. Addressing SES inequalities in educational opportunity requires more than simple changes in the distribution of information.
Higher-cost interventions that enhance academic preparation, foster trust, and generate social capital may be crucial for overcoming barriers to pursuing an optimal further education program (Carrell & Sacerdote, 2017). For three reasons, our results differ from those of Bleemer and Zafar (2018), who found that information on returns to further education reduced SES aspiration disparities. First, we offered respondents the option of selecting the pursuit of a 2-year degree, a choice favored by a disproportionate share of lower-SES respondents. Second, ceiling effects may have come into play in the Bleemer and Zafar analysis, as close to 90% of high-SES control-group respondents selected the 4-year college option over the no-further-education alternative. Finally, we treated respondents with balanced information on costs and returns, while they treated them with just costs or earnings information.
If a divide is ephemeral, a modest amount of information can narrow substantial differences. The contrasting college aspirations of those in the United States who identify with one or another of the two major political parties is a case in point. Public opinion on many issues is unstable, inconsistent, and only weakly connected to partisan affiliations. So it is with college aspirations. Whether students prefer to pursue a 2-year or 4-year degree is a matter that divides political elites, but it has yet to become deeply entrenched in partisan identities of the public at large. Under such circumstances preferences can be erased by small doses of information conveyed in a survey.
Information may also close the Hispanic-White gap. Although our results were of borderline significance, the point estimates were large. Two complementary explanations for the surprising findings deserve careful consideration in further research. First, Hispanic Americans may not be well informed about the returns to postsecondary education. Though they may perceive the importance of education for social mobility, they may be relatively less informed than Whites about the costs and benefits of different kinds of further education (Hoxby & Avery, 2013;Ovink, 2017). In the absence of specific information, many Hispanics prefer the 2-year associate degree with the expectation that this amount of additional education will suffice. As newcomers who speak a different language, Hispanic adults may not have access to the same social networks and may not have accumulated as much relevant social capital as White adults. Without those resources, they find it more difficult to obtain critical information about educational opportunities (Perna, 2006). 2 In such cases, small doses of information may alter aspirations, especially-and this is our second point-if the group has lower discount rates. Massey (1990) theorized that those who migrate across international borders are exceptionally likely to take risks, as they hope that the venture will yield large long-term returns. The results from our experiments were consistent with that theory. Cost information, when presented alone, had little impact on Hispanic respondents, but earnings information, presented alone, shifted opinion decisively toward the bachelor's degree option. The net effect of balanced information on both costs and returns was that it closed the Hispanic-White gap. In sum, it appears that cost-returns information about investments in higher education had the most substantial impact when individuals were initially not as well informed and had a low discount rate for returns on long-term investments. Both factors may allow new information to contribute to a more favorable cultural perception of more extended postsecondary education.

Limitations and further research
We acknowledge several limitations to this research. First, those in our treatment conditions were not told the average annual earnings for those who did not pursue higher education. Second, respondents were not asked for estimates of costs and earnings before the experiment was administered. Although the advantage of this limitation was not embarrassing respondents by correcting errors of those who guessed incorrectly, we could not estimate treatment effects for those who were more or less informed. Future research may explore this topic further by first asking respondents to estimate the costs of and returns to a 4-year degree, 2-year degree, and high school diploma. Third, the information on costs and returns to postsecondary education was provided in national averages, not specific averages for a respondent's location and eligibility for specific scholarship programs. A national average masks variation in costs and returns attributable to differences in, for example, college major, child ability, and geography (Altonji & Zimmerman, 2017). Although these factors should affect all arms of a randomized experiment in similar ways, future research may be able to identify heterogeneities in impacts by any one or all of these variables. Fourth, findings for African American respondents were underpowered. Given the preliminary findings that have emerged, future research on this question is urgently required.
We additionally acknowledge that we were asking parents at a specific point in time. We could not ascertain whether results would differ if, on one hand, parents were given information when their child was near college-going age or if, on the other hand, parents were given information many years earlier. Other questions remain unanswered as well, including questions about persistence of information effects, the dosage required for impact, and the connection between information-induced shifts in aspirations and actual enrollments.
It is also unclear whether our observed effects were attributable to respondents (a) updating their previous beliefs about costs and returns to higher education, or (b) reacting to the framing of the information. Two of our three treatment arms were at risk for framing effects. Those who were told only costs information were framed to think only about immediate impacts on their finances, while those only informed about the returns to education were framed to think about the long-term benefits correlated with a bachelor's degree. For this reason, we believe balanced cost and returns information should be provided when estimating impacts on aspirations. But even this balanced treatment may be biased because it nudges respondents to consider only economic matters while excluding broader cultural and social concomitants of further education.
This study of adult aspirations opened only one window into the large world of social divides in access to further education that will not be closed without addressing the social and cultural forces that affect the preparation of students for pursuing an advanced degree (Park & Hossler, 2014;Perna, 2006;Serna, 2015). Nothing in this study should be taken to suggest otherwise. In fact, the inability of small doses of economic information to close gaps by SES in college aspirations reinforces that conclusion. A more holistic and sustained approach that engages students and families in a comprehensive way may prove to be a more fruitful strategy for improving postsecondary aspirations and outcomes.
Notes 1. We refer to estimates without controls to facilitate comparison to income differentials in enrollment reported by Bailey and Dynarksi (2011).
2. Almost half of the U.S. Hispanic population in 2016 was foreign-born, and 72.4% of Hispanics older than age 5 years lived in households where Spanish was the language spoken at home (U.S. Census Bureau, 2016).

Funding
This work was funded in great part by National Science Foundation Grant No. 0732233, "Learning Progressions for Scientific Inquiry: A Model Implementation in the Context of Energy."