The Effect of Education Level on Exercise for U.S. Adults Ages 65 and Up

An econometrics-based study suggesting that greater exercise levels could be linked with longer formal education levels.

The Effect of Education Level on Exercise for U.S. Adults Ages 65 and Up

By Nathaniel Ro

Abstract

In the United States, Americans have not been getting enough exercise. This is problematic because this may indicate foregone healthful benefits from exercise and potential negative health implications which may arise from a lack of exercise. This study focuses on ways we could increase exercise in America. The conclusion of this paper is that higher formal education levels might result in higher levels of exercise.

Introduction

Americans as a whole have not been getting enough exercise. According to 2023 data from America’s Health Rankings, only 30.4% of adults, and only 29.0% of adults ages 65 and up, in the U.S. met the federal physical activity guidelines “in the past 30 days” (America’s Health Rankings, 2023). Therefore, it may be beneficial for us to try to increase these statistics because of the beneficial effects exercise has on health. Exercise can help lengthen an individual’s lifespan (Ruegsegger & Booth, 2018). Exercise can also help a patient fight against chronic diseases (Anderson & Durstine, 2019). Not exercising may shorten an individual’s lifespan and make one more susceptible to chronic diseases (Booth et al., 2012). Given all of the health benefits that may be derived from exercise, it would be propitious for Americans to become more active. To this end, it may be of use to study if and to what extent an adult’s education level may affect his or her exercise level. Would more or less education be associated with an increase in exercise? On one side, greater education could lead to higher-paying (white collar) jobs, and these higher-paying jobs may be less physically demanding than lower-paying (blue collar) jobs. This could translate into greater education leading to a lower exercise level. However, greater education could also indicate a greater knowledge of the benefits of exercise, and this could lead to more exercise. This would mean that education and exercise are positively associated with each other. Previous literature has shown a positive association between education and exercise. Therefore, it may be useful to specifically analyze the potential effect education (as the explanatory variable) could have on exercise (as the explained variable).

This study looks at the potential effect that having a college degree could have on individuals meeting the federal physical activity guidelines for American adults ages 65 and older. This study particularly focuses on adults ages 65+ because by 65, most adults would probably be finished with their formal education. Therefore, studying adults ages 65+ can be useful in reducing the potential risk of reverse causality. In other words, studying adults 65+ can be useful in looking into the question of how someone’s education level from the past can affect his or her exercise practices in the present and not the other way around.

Literature Review

Previous literature suggests a positive correlation between physical activity and education, although the particular causation of education on exercise is uncertain. A study done by Shaw & Spokane (2008) suggests that the rate of physical activity may be “higher for individuals with high levels of education compared to individuals with less education” (p. 779). This study also proposes that higher levels of education may mitigate a decrease in exercise in old age (p. 779). Another study done by Kari et al. (2020) demonstrates that “higher education is positively related to physical activity, but it is not clear whether this relationship constitutes a causal effect” (p. 1). Therefore, although previous literature is ambiguous on if education is the explanatory factor that positively affects exercise rates or if it is the other way around, there does seem to be a positive correlation between education and exercise.

Previous literature may also suggest a positive association between education and overall health (not just exercise). Because exercise can be positively associated with good overall health, better health could be reflective of greater exercise. A study, conducted by Schnohr et al. (2004), suggests that lower levels of education may be linked with “more smokers, more heavy drinkers, a lower level of physical activity, and more obese individuals” (p. 252). Therefore, this study may suggest a potential link between lower levels of education and poor health in general. This study did concede, though, that the “relations between smoking, alcohol consumption, physical activity, obesity, and mortality are similar at all educational levels, and cannot explain the social gradient in mortality” (p. 256). In another study done by Böckerman et al. (2017), “findings indicate that education could be a protective factor against obesity in advanced countries” (p. 1). Therefore, previous literature suggests a positive correlation between positive health characteristics (such as exercise) and education.

This study contributes to previous literature by examining the effect education could have on exercise for U.S. adults ages 65 and older. Since many 65 year olds may be finished with their formal (that is, primary, secondary, and perhaps tertiary) education by the age of 65, this study may be able to better measure the potential effect education could have on exercise than a study that only considered the relationship between the two for youth. State would be this study’s unit of analysis, and data would be represented at the state level with the District of Columbia included.

Empirical Model

The estimated linear regression model below may be used to analyze the effect education may have on exercise.

% Population of adults 65 and older that met the federal physical activity guidelines in the past thirty days = α + β₁Education + β₃Income + β₄Poverty + β₅% of park and wildlife land + β₆% of Population using public transportation + β₇% Black + β₈% Hispanic + β₉% Asian + β₁₀% Other + β₁₁% Unemployment + β₁₂Region + u

Please Note: Other includes American Indian, Alaska Native, Native Hawaiian, Pacific Islander, and Multiple Ethnicities.

% Unemployment represents state-level unemployment rates for August 2025 according to the U.S. Bureau of Labor Statistics.

In the above model, % Population of adults 65 and older that met the federal physical activity guidelines in the past thirty days measures the state-level percentage (%) of the population ages 65 and older that met the federal physical activity guidelines “in the past 30 days” (America’s Health Rankings, 2023). These guidelines specifically call for “150 minutes of moderate or 75 minutes of vigorous aerobic activity and two days of muscle strengthening per week” (America’s Health Rankings, 2023). This data is from the year 2023 AD. The independent variable of interest is Education, which represents the percentage (%) of a state’s elderly (ages 65 and up) population that are college graduates. Data for Education comes from the year 2023 and is at the state level.

In the above regression, the Income variable gives state-level data for the median income in dollars of an entire state’s population (data from 2023). The Poverty variable represents the percentage (%) of a state’s population ages 65 and older that falls below the federal poverty level (data from 2023). The % of park and wildlife land variable measures the state-level percentage (%) of a state’s land that is designated for parks and for wildlife. It appears that the data for this % of park and wildlife land variable came from an article that was published in 2021, although I could not find the specific year(s) for which this data represents. The % of Population using public transportation variable represents estimates of the percentages (%) of state populations that use public transportation (data from 2023). % Black represents the percentage (%) of a state’s population that is black (data is from 2023). % Hispanic measures the percentage (%) of a state’s population that is hispanic (data from 2023). % Asian shows the percentage (%) of a state’s population that is Asian. The % Other variable measures the percentage (%) of a state’s population that is either American Indian, Alaska Native, Native Hawaiian, Pacific Islander, or of multiple ethnicities. The percentage (%) of a state’s population that is white is excluded. The % Unemployment variable measures the percentage (%) of a state’s population that is unemployed (data from 2025). Finally, geographic region is controlled for, and the Northeast is excluded.

Data

The data for the dependent variable % Population of adults 65 and older that meet the federal physical activity guidelines in the past thirty days comes from America’s Health Rankings. Data is measured as the percentage (%) of a state’s elderly (ages 65 and up) population that met the federal physical activity guidelines “in the past 30 days” (America’s Health Rankings, 2023). The numbers indicate these percentages. Data for Pennsylvania and Kentucky was unavailable. The data for this variable came from 2023. Data can be found here: Exercise - Age 65+ in United States.

Data for the independent variable of interest Education also came from America’s Health Rankings. The numbers represent state-level percentages of adults 65 and older that possess a college degree. Data is given as a percentage (%) of a state’s population. This data came from 2023. Data can be found here: College Graduate - Age 65+ in United States.

Data for the Poverty variable also came from America’s Health Rankings. This data represents the percentage of a state’s elderly population ages 65 and up that live below the poverty level. Data for Poverty comes from 2023. Data can be found here: Poverty - Age 65+ in United States.

Data for the % of park and wildlife land variable’s data came from Cliq. This data measures the percentage (%) of a state’s acreage that is designated for parks and wildlife. In describing their data, Cliq said that the “data used in [their] analysis is from the U.S. Department of Agriculture (USDA) and the Bureau of Economic Analysis (BEA). To determine the states with the most parks and wildlife areas, researchers calculated the proportion of state land designated for parks and wildlife. This was calculated by taking the acreage designated for parks and wildlife areas, and dividing it by the state total land area” (Semow, 2021). This data can be found here: States With the Most Parks and Wildlife Areas.

Data for the % of Population using public transportation variable came from HomeArea.com, and the data within HomeArea.com came from the U.S. Census Bureau American Community Survey. This data contains estimates for the year 2023 (HomeArea.com, 2023). This data can be found here: Highest Public Transit Usage States.

Data for ethnicity, specifically for the % Black, % Hispanic, % Asian, and the % Other variables all come from KFF. This data represents estimates of state-level percentages of ethnicities within a state’s population. In describing their data, KFF said that the population “and demographic data on [sic] are based on analysis of the Census Bureau’s American Community Survey (ACS) and may differ from other population estimates published yearly by the Census Bureau. U.S. and state population data [...] are restricted to the civilian, non-institutionalized population for whom ACS collects and reports poverty information” (KFF, 2023). This data can be found here: Population Distribution by Race/Ethnicity.

Data for the % Unemployment variable came from the U.S. Bureau Of Labor Statistics. This data represents the unemployment rates of August 2025. The data is given as a percentage (%) of the labor force. The percentages for the states are based on the unemployment rates of those who have their place of residence in the respective states. This data may have been revised in September 2025 (U.S. Bureau Of Labor Statistics, 2025). This data may be found here: Unemployment Rates for States.

Finally, data for the remaining Income and Region variables came from the previous short paper for this class. It seems that, according to the sample short paper, state-level data for these variables can be found at: https://www.kff.org/state-health-facts/.

Table 1 below contains descriptive statistics.

Table 1: Descriptive Statistics

Table 1 above shows a significant range in the percentage (%) of a state’s 65-and-older population that met the federal physical activity guidelines in the past thirty days. This range goes from Mississippi’s percentage of 16.9% and Vermont’s percentage of 58.5%. The average proportion of a state’s population ages 65 and up that met the federal physical activity guidelines in the past thirty days for all the states is around 28.982%. The proportion of a state’s elderly 65+ population that are college graduates also has a wide range between 20.6% for West Virginia and 49.7% for the District of Columbia. The average proportion (%) of a state’s elderly 65+ population that are college graduates for all the states is approximately 30.973%. The numbers in the table may be approximates due to rounding.

Original Empirical Results:

Regression results below in Table 2 reflect a positive association between education and exercise for a state’s elderly population. According to the table, a 1 percentage point increase in the proportion (%) of a state’s elderly population that have a college degree would relate with an approximately 0.75 percentage point increase in the proportion of the state’s elderly population that met the federal physical activity guidelines (p<0.01). Because p<0.01, the Education variable is statistically significant at the 1% level. Therefore, according to the regression, there seems to be strong evidence that education has a positive effect on exercise levels for the elderly.

Table 2: Regression Results

According to the table above, poverty also seems to have a positive association with exercise (based on its slope estimate). However, the p-value for the Poverty variable is p>0.1. The p-values of the other variables are also p>0.1: therefore, these other variables are not statistically significant at the 10% level. This may suggest that Education is the most significant explanatory variable of the % Population of adults 65 and older that met the federal physical activity guidelines in the past thirty days variable. Finally, the F statistic’s p-value of this regression is 0.0018, so p<0.01. This means that at the 1% level, we can reject the possibility that none of the variables used in this regression are significant in determining the outcome of the exercise variable.

Appendix: Diagnostic Tests

Several errors could enter into this study and disturb its findings. These errors may include heteroskedasticity, multicollinearity, an uneven distribution (not a normal distribution), and the possibility that the data in this study is not linearly distributed. Therefore, to test for these errors, the Breusch-Pagan, VIF, Skewness-Kurtosis, and Ramsey tests may be used. These tests are below. Passing a test means that that test did not find any errors in the data. Please note that the variable names have been changed for simplicity. However, the results would still be the same, regardless of the new names.

Heteroskedasticity Test (Breusch-Pagan Test): Passed

Table 3: Heteroskedasticity Test

In the above regression, the F-statistic is not significant at the 1%, 5%, or 10% levels. Therefore, the null hypothesis of homoskedasticity would fail to be rejected.

Multicollinearity Test (VIF): Passed

Table 4: Multicollinearity Test

In the above test for multicollinearity, the VIF (Variance Inflation Factor) method was used. According to the VIF, a mean VIF that is greater than 10 could be a reason for further investigation. A VIF greater than 10 might indicate that the variable is a linear combination of other independent variables. For this study, none of the variables have a VIF that is greater than 10, and the mean VIF is 4.14: therefore, according to this method, we can reject the likelihood of errors coming from multicollinearity between this study’s variables.

Normality Test (Skewness-Kurtosis): Failed

Table 5: Normality Test

In the two above tables, the normality test fails on both the original Exercise dependent variable and the corrected Logexercise variable. The Logexercise variable is generated by taking the natural logarithm of the Exercise variable. For the top test for the Exercise variable, the Prob>chi2 value is 0.0000, so p<0.01 and also p<0.1. Therefore, the null hypothesis of normality is rejected at both the 1% and 10% levels. In the corrected normality test with the Logexercise variable, the Prob>chi2 value is 0.0395, so p>0.01, but p<0.05 also. Therefore, the null hypothesis would be rejected at the 5% and 10% levels, but it would fail to be rejected at the 1% level. In short, this study failed the normality test. This could indicate that the confidence of the conclusions of this study’s findings might not be as sure as if the normality test passed. This would be because the p-values for the t-tests and F-test might not be valid due to this study failing the normality test. However, other than this, this failure might not pose a grave danger to this study’s conclusions.

Ramsey Test (RESET): Failed

Table 6: Ramsey Test

The Ramsey Regression Equation Specification Error Test (RESET) may be used to test whether or not other non-linear models would be better for explaining the dependent variable of exercise. The null hypothesis for this test is that this study’s model does not have omitted variable bias (OVB). Since the p-value is 0.0004 (p<0.01), this null hypothesis would be rejected, so there might be variables not included in this study’s model that could affect this study’s findings. Therefore, this study fails the Ramsey (RESET) test.

Alternative Regression

Table 7: Alternative Regression

To account for the failed normality test, the above table gives results from an alternative regression accounting for a corrected Logexercise dependent variable (generated as the natural logarithm of the Exercise variable). Overall, the results of this regression and of the original one are very similar. The Education variable’s p-value in this updated regression is 0.014, so p<0.05, and Education is significant at the 5% and 10% levels. However, the original regression’s p-value for the Education variable is 0.007, so p<0.01, and Education, for the original regression, would be significant at the 1%, 5%, and 10% levels. The p-value for this regression’s F-statistic is 0.0003, so p<0.01: therefore, similar to the original regression, the possibility that none of the included explanatory variables are significant for determining the results of the Logexercise dependent variable can be rejected at the 1%, 5%, and 10% levels. Another similarity to the original regression may be that the other explanatory variables besides the Education variable are not significant at the 1%, 5%, and 10% levels. Overall, the levels of confidence are only slightly lower in the above alternative regression than with the original regression. Largely, the results between the original and the alternative regressions are similar.

Conclusions and Policy Implications

Results from this study suggest that education levels are significant determinants of exercise levels. One potential conclusion of this study could be that more education can translate into a more active lifestyle for people. The findings of this study may be consistent with those of previous literature. For example, both this study and another study done by Kari et al. (2020) suggest a positive association between education and exercise. Therefore, to increase exercise levels and the benefits associated with exercise, it may arguably be useful to try to raise education levels. Policy options could be to extend secondary education (high school) beyond twelfth grade. However, this could also introduce negative consequences such as increased tax rates and a decreased labor force in technical industries such as in car maintenance because individuals with higher levels of formal (primary, secondary, and tertiary) education might incorrectly see technical jobs as “less desirable”. Another option could be to change the educational focus from being on public schools to being on private schools and on homeschooling. This could foster greater educational initiative among students as this could lead them to be more responsible for their education. This is a better option than extending high school beyond twelfth grade. For instance, focusing more on homeschooling and private education could free up a lot of tax dollars that are spent on public schools. It may also give families more freedom in directing their children’s education, and this could lead to greater specialization to the betterment of society. Another option could be to reevaluate the current educational agenda. Instead of focusing on school “fillers” that waste time and money, and instead of expending excessive effort on re-teaching students academic material from the previous academic year(s), reevaluating and redefining the educational content of schools well could help students learn more. Therefore, they might be able to attain a higher level of substantive education while in high school than someone else may have upon graduating from college. These measures could be used to increase the education levels of individuals, and this may lead to an increase in exercise and its healthful benefits.

Limitations

Results of this study are not without limitations. For one, the data analyzed in this study came from the U.S.’ state-level 65+ population. Findings could be different for other countries. For example, in extremely hot or extremely cold climates, greater formal education could mean a lower level of exercise because those who would want to pursue more education might choose to do this because of a preference to stay inside and study rather than go outside and work/exercise. Therefore, these findings might not be consistent with those of other nations.

Another limitation may be that this data is representative of adults ages 65 and older. This may introduce error into the data because American society is very different now than it was in the twentieth century. For example, fifty years ago, screens, the internet, and AI were not as prevalent as they are today (if not nonexistant). The increasing prominence of these factors could have an effect on both formal education attainment and on exercise levels, thus introducing potential bias into this study.

Another limitation of this study is that its model failed the Ramsey and normality tests (see below). Failing the Ramsey test means that there is potential for omitted variable bias (OVB) to corrupt this study’s findings because there may be another variable(s) not included in this study’s model that could be related with both education and exercise. Failing the normality test could indicate that the p-values of the t-tests and F-test are inaccurate. Therefore, the level of confidence of this study’s findings might not be as certain as if the normality test passed. However, as demonstrated in the above original and alternative regressions, this difference in the levels of confidence might not be substantially perilous to this study’s findings. Therefore, further study might be primarily concerned with what other variables beside the ones included could affect an individual’s exercise habits.

References

Anderson, E., & Durstine, J. L. (2019). Physical activity, exercise, and Chronic diseases: a Brief Review. Sports Medicine and Health Science, 1(1), 3–10. https://doi.org/10.1016/ j.smhs.2019.08.006

Böckerman, P., Viinikainen, J., Pulkki-Råback, L., Hakulinen, C., Pitkänen, N., Lehtimäki, T., Pehkonen, J., & Raitakari, O. T. (2017). Does higher education protect against obesity? Evidence using Mendelian randomization. Preventive Medicine, 101, 195–198. https:// doi.org/10.1016/j.ypmed.2017.06.015

Booth, F. W., Roberts, C. K., & Laye, M. J. (2012). Lack of exercise is a major cause of chronic diseases. Comprehensive Physiology, 2(2). https://doi.org/10.1002/cphy.c110025

College Graduate - Age 65+ in United States. (2023). Americashealthrankings.org. https://www.americashealthrankings.org/explore/measures/education_sr

Exercise - Age 65+ in United States. (2019). Americashealthrankings.org. https:// www.americashealthrankings.org/explore/measures/exercise_sr

Exercise in United States. (n.d.). America’s Health Rankings. https:// www.americashealthrankings.org/explore/measures/exercise

“Explore Poverty - Ages 65+ in the United States | AHR.” America’s Health Rankings, www.americashealthrankings.org/explore/measures/poverty_sr.

“Highest Public Transit Usage States - HomeArea.com.” Homearea.com, 2023, www.homearea.com/rankings/state/percent_using_public_transportation/.

Kari, J. T., Viinikainen, J., Böckerman, P., Tammelin, T. H., Pitkänen, N., Lehtimäki, T., Pahkala, K., Hirvensalo, M., Raitakari, O. T., & Pehkonen, J. (2020). Education leads to a more physically active lifestyle: Evidence based on Mendelian randomization. Scandinavian Journal of Medicine & Science in Sports, 30(7), 1194–1204. https://doi.org/10.1111/sms.13653

“Population Distribution by Race/Ethnicity | KFF State Health Facts.” KFF, 9 Aug. 2025, www.kff.org/state-health-policy-data/state-indicator/distribution-by-raceethnicity/? currentTimeframe=0&sortModel=%7B%22colId%22:%22Location%22.

Ruegsegger, G. N., & Booth, F. W. (2017). Health Benefits of Exercise. Cold Spring Harbor Perspectives in Medicine, 8(7). https://doi.org/10.1101/cshperspect.a029694

Semow, Thomas. “States with the Most Parks and Wildlife Areas.” CLIQ, 10 May 2021, www.cliqproducts.com/blogs/news/states-with-the-most-parks-and-wildlife-areas? srsltid=AfmBOoodh1LqC8FuNI-DQfikjRefti3LFZBp4oINF0VabCS5CBKaGh52. Accessed 31 Oct. 2025.

Schnohr, C., Højbjerre, L., Riegels, M., Ledet, L., Larsen, T., Schultz-Larsen, K., Petersen, L., Prescott, E., & Grønbæk, M. (2004). Does educational level influence the effects of smoking, alcohol, physical activity, and obesity on mortality? A prospective population study. Scandinavian Journal of Public Health, 32(4), 250–256. https:// doi.org/10.1177/140349480403200403

Shaw, B. A., & Spokane, L. S. (2008). Examining the Association Between Education Level and Physical Activity Changes During Early Old Age. Journal of Aging and Health, 20(7), 767– 787. https://doi.org/10.1177/0898264308321081

“Unemployment Rates for States.” Bls.gov, 2008, www.bls.gov/web/laus/laumstrk.htm#laumstrk.f.p.

The Effect of Education Level on Exercise for Adults Ages 65+.pdf

Download PDF • 540KB