Homework’s Influence on Future Financial Success in the Age of ChatGPT

Parents, educators, and policymakers seek guidance on what advice to share with young people to promote lifelong success. Many want to know how best to advise children, students, and other young stakeholders so they can achieve long-term economic security. They also want to know what policies should be instituted inside and outside of classrooms to promote future financial wellbeing.

Opinions regarding the values or actions needed for success, such as grit and hard work, vary. Nevertheless, relatively little research has directly investigated how specific actions translate into successful adulthood. Completing homework is one action over which youth exercise considerable control, but it remains unclear how it impacts their future financial success. This study seeks to determine the importance to future success of completing homework assignments and to assess whether such behavior is directly profitable or provides indirect benefits through the further pursuit of education. While education can enhance financial literacy among young adults, long-term financial security appears to be strongly influenced by habits formed during secondary school. This study seeks to inform adults interested in promoting financial security by highlighting the importance of encouraging and supporting students to complete their homework assignments.

Gill and Schlossman (2004) trace the history of homework since the 1850s. They find little opposition to homework in the 1800s, but by the 1930s opposition had begun growing, as homework was viewed as a threat to health by depriving children of time for outdoor play. In 1957, the launch of the Soviet spacecraft, Sputnik, spurred demands to increase the homework burden. More recent data available from the National Assessment of Educational Progress (NAEP) survey shows that two-thirds of all students do some homework each night. Students of elementary school age reported a steady rise, from 60% who were assigned any homework in the early 1980s to 75% by 2012. Junior and senior high students saw homework rise in the 1980s, fall in the early 1990s, and remain roughly constant since.

The National Household Education Survey asks parents about the amount of homework children complete each week. Data from 2019 shows that, as students age, they devote more hours to homework. The average 9-year-old completed 3.7 hours of homework weekly; the typical 17-year-old completed 7.1 hours. These figures suggest that homework is an important part of the lives of America’s young people.

Understanding homework’s impact on financial security is also becoming increasingly important as Artificial Intelligence (AI) programs become more widespread. Data from the Rand Corporation (Schwartz and Diliberti, 2026) show recent dramatic upticks both in students using AI to complete homework and a belief that it harms critical thinking skills. Other studies suggest learning is decreased when using AI and that it is difficult for instructors to identify the use of AI for completing assignments. Educators and policy makers need to know the future impact of students outsourcing their homework to large language AI models (LLM), like ChatGPT.

This research follows other educational studies that have explored the relative importance of cognitive and noncognitive abilities in influencing economic outcomes. Cognitive ability can be quantified using tests, grades, and IQ measures, while noncognitive factors are defined by family background, personality, and behavior. While studies that relate educational factors to financial success often focus on cognitive ability (Brown & Reynolds, 1975; Zagorsky, 2007; Borghans, Golsteyn, Heckman, & Humphries, 2016), Lleras (2008) demonstrates that social skills and work habits may be important noncognitive traits.

This study expands on earlier work that examines the importance of noncognitive traits. Using two recent national surveys of distinct sample populations, we investigate whether homework completed in high school directly or indirectly builds human capital. The results, consistently across both datasets, show that completing high school homework does not directly explain differences in future earnings but does influence homeownership, and in the dataset with net worth information, it also impacts overall wealth. Importantly, the two datasets suggest that completing homework is associated with subsequent educational attainment, which in turn is a significant factor that influences future earnings.

Human capital theory states that cognitive and non-cognitive abilities influence financial success. This theory proposes that a positive relationship exists between job skills and labor rates through their impact on productivity. Productive capacities beyond inherent skills may include those learned in school or developed through experience. While the theory suffers from some limitations—see Marginson (2019), for example—extensive empirical analysis has documented positive relationships between income and education, relevant experience, and personal attributes. In a meta-analysis based on 140 articles, Ng et al. (2005) report significantly positive correlations between salary and educational attainment, years of work experience, measures of cognitive ability, greater dependability, and a stronger achievement orientation. Welch (1975) and others note that the relationship between education, a key component of human capital, and income may reflect both skills that are learned in school and inherent abilities that are displayed during school progression.

We suggest that human capital may be enhanced by homework completion as a learned skill as evidence of ability, either directly or via educational attainment. Homework completion has been described as a measure of self-motivation and suggested as a way to improve chances for future educational success. Ramdass and Zimmerman (2011) observe that, “during homework completion, students engage in self-regulation by motivating themselves, inhibiting distractions, using strategies to complete homework, managing time, setting goals, self-reflecting on their performance, and delaying gratification.” Satisfactory completion of homework is also linked to students’ confidence in their capacity to succeed, which is another component of human capital. Fuente et al. (2022) argue that finishing assignments enhances students’ sense of agency, and Schunk and DiBenedetto (2016) state that confidence correlates with greater persistence, increased interest in learning, and stronger effort.

Bempechat (2004) observes similarly that “homework plays a critical, long-term role in the development of children’s achievement motivation.” Her comments are, however, directed primarily to future education, stating that “homework is a vital means by which children can receive the training they need to become mature learners.”

Therefore, we propose two hypotheses. First, we suggest that homework completion, as a measure of dependability, self-regulation, and achievement orientation, helps individuals build human capital, in turn generating financial success. We expect to observe a positive relationship between homework completion rates and higher income, wealth, and rates of homeownership. Second, we suggest that homework completion influences financial success via educational attainment. To the extent that completing homework assignments regularly supports educational advancement, better homework habits will indirectly generate greater financial success.

Research on cognitive and noncognitive abilities suggests that learning ability is reflected in lifetime earnings and that knowledge is associated with wealth and risk-taking and reduced financial distress. For example, Zagorsky (2007) finds that significant determinants of income include IQ, net worth, age, ethnicity, and educational attainment. He finds little evidence however, indicating that the effects of IQ on wealth or financial distress are significantly different from zero, implying that the determinants of income and wealth differ.

Although noncognitive abilities are more difficult to quantify than cognitive abilities, they do influence labor market and financial outcomes significantly. For example, Bowles and Gintis (1976, 2002) find that the contribution of schooling to cognitive ability may explain only a small portion of its overall contribution to income. They suggest instead that noncognitive ability directly influences earnings, supporting the positive relationships identified by Jencks (1979) between leadership, study habits, and future earnings. Similarly, Robb and Woodyard (2011) report that financial knowledge and educational attainment are positively related to good financial practice, and Wang and Hanna (1998) report that those achieving higher educational status invest more extensively in higher-earning, risky assets. Jenkins et al. (2025) followed teenage girls who were involved in the juvenile justice system and found an “indirect path via homework completion on emerging adult education and income levels.”

Lleras (2008) studies the implications of specific behaviors rather than personality traits. She finds that students with strong social skills and work habits, including high rates of homework completion, may be more likely to achieve higher earnings and educational attainment. When including education in a model with cognitive and noncognitive abilities, however, she observes that the coefficient for homework is no longer significant. This finding suggests that the effect of homework completion on future income may be explained by educational attainment—the issue we study in detail.

Homework is also a theme in the educational psychology literature, suggesting that successful completion of such tasks enhances self-confidence, in addition to executive skills, which in turn leads to pursuit of advanced achievements. Schunk and DiBenedetto (2016) apply the theory of “self-efficacy” to the field of education, stating “those with high self-efficacy participate more readily, work harder, persist longer, show greater interest in learning, and achieve at higher levels.” Fuente et al. (2022) describe a relationship between homework completion and student agency. “Students observe their progress toward learning goals as they work on their tasks. For example, assignments completed is one of many progress indicators that reinforce students’ sense of capability for performing, and so increase their self-efficacy for further learning.”

The literature indeed supports a strong connection between educational effort and attainment. For example, a systematic literature review by Guo et al. (2025) found doing homework improved the academic performance of students, especially in math. Suh and Suh (2006) find that high school dropouts who spent an additional one hour per week on homework in the eighth grade were 4.5% more likely to complete their degrees. Similarly, Lleras (2008) finds that greater homework completion increases the odds of achieving a higher-level degree. Studying college data, Astin (2006) finds a significantly positive relationship between hours spent studying or doing homework and completing a degree within four years. Good work habits positively affect both future educational outcomes and performance in the labor market (Heckman et al., 2006; Lindqvist & Vestman, 2011), and we therefore believe that each link should be explored separately.

If completing homework boosts cognitive and confidence development, then using AI to outsource out-of-classroom exercises ensures these developments do not happen. Oakley et al. (2025) observe that “emerging research on learning and memory reveals that relying heavily on external aids can hinder deep understanding.” Consider the experiment by Kosmyna et al. (2025), tasking 150 college students to write essays using three different research tools. One third of the group freely used ChatGPT to prepare their theses, another used an internet search engine, and a third was prohibited from using any form of electronic resource. While the essays written using ChatGPT appeared to be the most polished of the group, they were also the most similar, and those students’ recall of their own work was inferior to the “brain-only” group. Tests of brain activity showed inferior “neural connectivity” of the ChatGPT group as well.

It is difficult for educators to tell when a student is doing their homework and when AI is completing the assignment. Ibrahim et al. (2023) use a ChatGPT model to answer questions used in introductory university courses in Computer Science and English Composition. Submitted homework assignments completed by students and ChatGPT were randomly shared with instructors. Graders found the quality of submissions completed by LLMs for the computer course were as strong as students’, and superior for the English course. More troubling was that graders could not distinguish the LLM’s work from the students’.

Research shows many students are now using LLMs to complete their homework. Schwartz and Diliberti (2026) surveyed American students in May 2025 and again in December. Between May and December, the percentage using AI to complete homework increased from 49% of high school students in May to 63% in December. This does not appear to be an issue exclusive to the United States. Data from Croatia by Oreski et al. (2025) show 58% of 5th to 8th grade students reported using AI tools to do their homework.

While the concept of income or earnings is well defined, Attanasio and Williamson (2000) measure wealth in four ways and show that home equity comprises the largest proportion of total wealth. In a separate study comparing wealth estimates between two national surveys, Juster, Smith, and Stafford (1999) find similarly that home equity represents the greatest proportion of total net worth. Wolff (2016) estimates that owner-occupied housing was the most important household asset between 1983 and 2013, and Artigue et al. (2025) state “housing is the main component of wealth for households in the middle of the income or wealth distribution” in their study of prices between 2004 and 2022. As a result, we use net worth, provided in one of the datasets, and homeownership, available in both data sources, as key measures of financial health.

The datasets for this study were obtained from the Cornell National Social Survey (CNSS), collected by Cornell’s Survey Research Institute, and the National Longitudinal Survey of Youth 1997 (NLSY97) conducted by the Bureau of Labor Statistics. The CNSS dataset was extracted from a private-use file and the NLSY97 sample was extracted from a public-use database. CNSS makes data available publicly through the Cornell Institute for Social and Economic Research.

CNSS collected responses from 1,000 randomly selected residents aged 18 years and over within the continental United States between September 11 and December 11, 2017. We restricted the sample for the purposes of this study to the 978 respondents for whom there were no missing answers regarding demographic factors, employment information, educational effort, income, or homeownership. Of these participants, only slightly more of the respondents were women (50.3%). Most of the respondents were Caucasian (82.0%); African Americans (14.7%) comprised the next highest share of representation. Ages ranged from 18 to 95 years with an average of 48 years. While our study using the CNSS dataset included fewer participants than some other studies, several demographic characteristics are similar to those in the nationally representative NLSY79 data used in Zagorsky (2007) and the National Education Longitudinal Study used in Lleras (2008). We used demographic and employment data as controls; a cognitive control variable is unavailable in the CNSS data.

NLSY97 is a longitudinal study that draws from a randomly selected sample of Americans who were between the ages of 12 and 18 years in 1997. This study uses the set of 2,931 individuals who completed both the initial and sixteenth (2015) surveys and for whom there were no missing data pertaining to educational effort and financial status. Unlike the CNSS dataset, the selected NLSY97 data represents a narrow age group ranging between 31 and 33 years of age. The sample consists of slightly more men (52.0%) than women and notably fewer individuals who identify as Caucasian (53.9%) when compared with the CNSS data.

We measured homework-completion rates using one question in CNSS and several questions in NLSY97. In CNSS, participants were asked: “How often did you complete your assigned homework in high school?” Participants who did not complete high school were asked to provide an estimated rate in their final four years of schooling. Possible response options ranged from 1 (never) to 5 (always). NLSY97 adopted a different approach by asking respondents to estimate the number of hours per weekday devoted to homework. In another question, participants were asked whether they spent time on homework during a typical school week, which is coded as a 1 (yes) or 0 (no) binary response.

To analyze CNSS data in this study we used both the raw data and a modified approach that converts responses into binary values corresponding to 1 (often) or 0 (not often). This approach is similar to the binary metric captured in NLSY97. The datasets differ because NYLS97 homework questions were asked while respondents were still in secondary school while CNSS asked participants to recall their homework habits after leaving high school. Confirming statistical results across both datasets should improve the rigor of our qualitative findings. This design could also help validate the CNSS low-cost survey approach.

Both surveys asked respondents to estimate their household income within the preceding year. NLSY97 broke this question into several components that we summed to obtain aggregate estimates. In contrast, CNSS asked participants to provide aggregate estimates directly. The CNSS survey collected 502 numerical responses and assigned 986 participants to nine income ranges. To maximize the sample size, ranges were converted into income estimates by assigning medians of the available 502 values. A final adjustment involved taking the logarithm of each respondent’s annual income and net worth (available only in the NLSY97 dataset). This approach follows Heckman et al. (2006), Lleras (2008), Lindqvist and Vestman (2011) and Meisenberg and Lynn (2011), accounting for skewness in the income distribution and placing greater importance on percentage rather than absolute dollar differences. Both surveys collected data indicating homeownership.

We selected control variables that in previous studies have been found to help explain variation in income and wealth and are available in both datasets. Frequently cited measures for analyzing income include wealth, age, race, employment status, and attained education (Brown & Reynolds, 1975; Heckman et al., 2006; Zagorsky, 2007; Lleras, 2008; Lindqvist & Vestman, 2011). Regarding wealth outcomes, studies suggest including income, age, race, employment status, attained education, and marital status (Grable, 2000; Zagorksy, 2007; McCarthy, 2011). We included an urban versus rural and three regional dummy variables as controls when analyzing homeownership due to the large disparity between the two areas overall and modest regional rate differences (Mazur, 2016, 2022). This results in a loss of 36 of the 2,931 NLSY97 observations.

Both datasets were analyzed using descriptive statistics, correlations, and regression analyses. Regression models were used to relate socioeconomic characteristics and schoolwork habits to log-income. Extending the analysis to wealth, regression models were used to relate homework habits to log-scale net worth after accounting for controls, including both age and age-squared metrics. To study homeownership, logistic regression was used in place of OLS regression, and we excluded an age-squared model because homeownership rates tend to increase steadily with age (U.S. Census Bureau, 2018). Finally, educational attainment was used as the dependent variable in regressions with the homework dummy, gender, age, race, born-in-the-U.S. and residence location as independent variables.

Descriptive statistics listed in Appendix tables A1 and A2 show that at least 80% of the respondents reported completing homework often (CNSS) or spent time doing homework on a typical weekday (NLSY97). Notable differences in the datasets include homeownership rate, which was 63% for CNSS participants and 33% for NLSY97 participants; employment, which was much lower for CNSS (66%) than it was for NLSY97 (92%) participants; and educational attainment, which on average was “some college” for CNSS respondents and “high school degree or equivalent” for NLSY97 participants. The differences in homeownership and employment may be explained by the older average age of CNSS respondents, which is supported by descriptive statistics. Supplementary regressions were estimated using a subsample of the CNSS data to match more closely the greater labor participation rate of the NLSY97 sample.

Correlation analyses were conducted on both datasets, summarized in the Appendix (tables A3 and A4). The three most significant Pearson correlations in the CNSS data, excluding those between geographic region dummy variables, are between homeownership and age (.446), marital status and homeownership (.436), and employment and age (-.367). These findings are expected because homeownership is associated with married couples (Mulder, 2006; Mundra & Oyelere, 2016), older people have had more time to save for buying a home, and older people are more likely to be retired. Estimated variance inflation factor statistics suggest there is little concern regarding multicollinearity.

Homework completion as a scaled variable is positively associated with homeownership and log-income variables, and the relationships are statistically significant in both cases. As a dummy variable, homework is positively related to log-income and homeownership and significantly with homeownership. In addition, there are significantly negative correlations in the CNSS (-.233) and NLSY97 (-.085) datasets between homework completion and gender, suggesting that females complete homework in high school more often than males do. This result supports previous findings indicating that boys are less interested in homework (Xu, 2008) and that girls demonstrate high motivation for learning (Vallerand, Fortier, & Guay, 1997; Holifield and Dunn, 2003).

The three most significant correlations in the NLSY97 sample, excluding those between geographic region dummy variables, are between marital status and homeownership (.348), wealth and homeownership (.319) and employment and income (.303). These relationships, excluding wealth which is unavailable in both samples, align with the CNSS results. The homework dummy variable is positively associated with net worth, income and homeownership, and the relationships are statistically significant. The homework variables have significantly positive correlations with educational attainment in both datasets, suggesting that increased rates of homework completion in high school are associated with additional schooling undertaken by adulthood. Estimated variance inflation factor statistics for the NLSY97 variables also suggest there is little concern regarding multicollinearity.

Regression results using log-income as the dependent variable are presented in table 1. Preliminary estimates showed that homework completion as a scaled variable was not significant in any regression model, so results are reported using the binary indicator.¹ Three versions of the model were estimated for each dataset, varying the treatment of age and educational attainment controls across models.

We find a number of consistencies between the estimated income models using the two datasets. Not surprisingly, coefficients on employment status and homeownership are positive and significant (p < .01). Educational attainment is also a significant explanatory variable across the first three sets of models. On average, a single point increase in educational attainment (for example, gaining a college degree) raises income in adulthood by 25% for NLSY97 participants and 19% for CNSS participants. Several results differ between the estimates using the two datasets. For example, gender is a highly significant determinant (p < .01) in the NLSY97 data, suggesting that males may be associated with higher incomes. This relationship is positive but not statistically significant in the CNSS data. In addition, age is consistently significant (p < .05) in NLSY97, but only in a subset of the CNSS regressions that exclude educational attainment.

With the inclusion of educational attainment in regressions, no meaningful statistical significance on income can be identified for homework completion in either dataset. Removing the education variable from the analysis in regression (3) results in positive coefficients but no statistical significance in the full samples. This finding suggests that at least a portion of the effect of homework completion on later earnings can be observed through educational attainment, supporting our second hypothesis but not the first.

To compare CNSS data with NLSY97 further, we sought to investigate a narrowed CNSS model using only the observations that matched the ages of the NLSY97 sample respondents; however, the remaining observations resulted in a sample size too small (n=42) for the number of variables and analysis method considered. As an alternative, we restricted the sample to working-age participants, ages 25 through 65, and re-estimated regression (3). As illustrated in regression (4), explanatory power increased and the homework effect grew, attaining statistical significance at the 5% p-level.

To better understand the role of education, we estimated an OLS regression of various controls and the binary homework variable on educational attainment. Results are reported in table 2. Homework completion is significant across both datasets at the .01 p-level including all control variables: the coefficients for CNSS and NLSY97 equal 1.0 and 0.7 academic degrees, respectively. This finding aligns with prior research and suggests that, on average, completing homework often in secondary school results in an increase of up to one degree level, which we have seen is associated with higher income and will be shown later with homeownership. These results support the hypothesis that homework completion affects future financial success indirectly, consistent with our second hypothesis.

We can explore the relationship between completing homework and wealth using a measure of household net worth using the NLSY97 database, but this measure is not available in the CNSS. Results for NLSY97 are presented in Appendix table A3. Missing data in 135 records reduces the size of the sample from the original 2,931. Results support the case for completing one’s homework, but only after excluding income and homeownership as control variables (model 2) since they are highly correlated with net worth. In that case, the homework dummy variable is statistically significant at the 10% level. Educational attainment is not a significant influence on this latter case.

Table 3 summarizes the logistic regression results using homeownership as the dependent variable measuring financial success, which is available in both datasets. Many of the coefficients are significant across the samples and models. The similarities observed between the two datasets reinforce the findings reported in Zagorsky (2007) regarding key indicators of wealth accumulation in adulthood. Marital status, race, and income are all factors that increase the likelihood of homeownership to a significant degree (p < .01). Urban residence is a statistically significant control variable in every model, two regional controls are significant for NLSY97, and educational attainment is significantly positive for both NLSY97 and CNSS.

The homework coefficient is positive and statistically significant when a binary indicator is used. The significance increases when education is removed in regression (2) while maintaining overall explanatory power. NLSY97 results suggest that, when controlling educational attainment, the indicator increases the odds of homeownership by 38%; the same estimate using CNSS data is 50%. Removing education magnifies the significance of the homework variable, increasing the odds ratio to 51% for NLSY97 and 74% for CNSS. Restricting the CNSS sample to working-age participants in model (3) yields results comparable to the alternative specifications.

The goal of this study was to assess determinants of income and wealth in adulthood. This paper is only the second study we know of to examine the influence of homework on future financial success, the first to consider homeownership and net worth, and builds upon prior research using two distinct datasets. We believe using two independently constructed samples drawn from different populations increases the rigor of our findings. Homework completion is the noncognitive proxy explored in this study because it captures a significant portion of the learning experience and effort for students. It is also easily observable and an actionable task. In investigating these relationships, we enable policy makers, parents, educators and financial counselors to provide actionable advice to children regarding financial independence.

Except for one regression specification, there is no strong evidence that homework completion rates in high school have a direct influence on future earnings. Our results do, however, show that homework completion is a significantly positive factor explaining net worth and homeownership. Separating educational attainment from the models increases the estimated influence of homework completion on financial outcomes, suggesting that both direct and indirect effects are at play.

Our findings are especially relevant to the rise in Artificial Intelligence. Kosmyna et al. (2025) provide evidence using measures of student brain activity and memory recall indicating lower levels of learning with the use of AI. If we assume their reported 50% decline in performance is equivalent to students completing on average 25% less assigned homework and applying the average of all regression coefficients in our study, we’d expect to see a reduction in income, education attainment, homeownership, and net worth by 2%, 0.22 academic degrees, 12%, and 11%, respectively. Studies on how AI affects learning from homework suggest that overseeing the use of LLMs by students for assignments is critically important.

Regarding policy implications, our message isn’t to assign more homework but instead guide students to complete what has been assigned to them. We can reference socialization theories that consider the influence parents, teachers, and peers have on young people’s development toward self-sufficiency. For example, Shim et al. (2015) examine how socialization shapes financial competency to illustrate the impact parental advice has on “healthy financial behavior development” including self-confidence in making financial decisions. We argue that parents, teachers, and policy makers should support, encourage, and require students to complete their assigned homework, on their own, in high school if they wish for those young people to attain financial security as adults.