Gregory M. Dempster
Elliott Professor of Economics and Business, Hampden-Sydney College, GDempster@hsc.edu
This research note represents an attempt to identify some of the major determinants of the decision to homeschool using a single, publicly available data set, the Parent and Family Involvement portion of the 2007 National Household Education Survey (NHES) conducted by the National Center for Education Statistics (NCES). Among the important findings of this study are the statistically significant impacts of household income, religious orientation, children’s gender, and self-identified race/ethnicity on the decision to home school a child.
Keywords: homeschooling, demographics, socio-economic factors, national surveys, parental and family involvement in educational choices
Homeschooling is becoming increasingly prevalent in the United States. Estimates of the number of homeschooled children have risen from 1.1 million school-aged children, or 2.2% of the school-age population, in 2003 (Lips & Feinberg, 2008) to approximately 2.04 million in 2011 (Ray, 2011), with a per annum growth rate of around 7%-8% that has been consistent since the late 1990s. Because the education of these children is being handled with little in the way of public, taxpayer-funded resources, the upward trend in homeschooling represents a potentially significant source of reduced public spending. Ray (2011) estimates that up to $16 billion in annual tax savings may result from individual family decisions to homeschool K-12 children.
Although the systematic study of the impacts of homeschooling is in its early stages, the evidence suggests that there might be significant educational benefits as well as resource savings for over-burdened public school systems. Ray (2103) reviews the literature on homeschooled children’s academic achievement and social, emotional, and psychological development, as well as the success of adults who were home educated, finding generally positive outcomes. Many studies also indicate that homeschooled students tend to outperform traditionally schooled students in various aspects of academic achievement (see, e.g., Rudner, 1999; Ray, 2004; Clemente, 2006; and Snyder, 2013). Granted, there are numerous methodological issues involved in disentangling the effects of homeschooling from the impacts of other factors, but there is nonetheless sufficient force behind the growing trend in home education to warrant optimism about its future role in decreasing the burden on public resources.
The decision to homeschool is, of course, a complex one involving many disparate factors. Some frequent reasons given for homeschooling include control of curriculum and learning environment, higher academic achievement, innovative pedagogy, family and social relationship development, and the particular values and beliefs of importance to the family (Ray, 2011). Mackey et al. (2011) finds several demographic factors to be significant in the decision to homeschool, including ethnicity, number of parents, education level of parents, and religious (e.g., evangelical Christian) orientation. Green and Hoover-Dempsey (2007) employed a parental-involvement questionnaire for the parents of 136 homeschooled students and found that these parents “appear to be motivated by an active role construction, strong sense of efficacy for helping the child learn, and positive perceptions of life context” while “beliefs about the values, content, adequacy, and methods of public school education appear to be implicated less strongly in their decisions” (p. 264). A number of other, mostly qualitative, studies have examined the question of parental motivations for homeschooling (see, e.g., Gray, 1993; Gustafson, 1988). However, few quantitative analyses have been developed using mass data sets. The current paper represents an attempt to more closely identify some of the major determinants of the decision to homeschool using a single, publicly available data set, with the hope that this could lead to more detailed analyses of the homeschooling decision employing both quantitative and qualitative data.
The data set comes from the latest available administration of the National Household Education Surveys Program (NHES) from the National Center for Education Statistics (NCES, 2013, available at http://nces.ed.gov/nhes). The NHES conducts telephone surveys of the non-institutionalized, civilian population of the United States in order to provide information on domestic educational issues (NCES, 2011, p. 345). These surveys have a variety of users, from policymakers to researchers, and offer numerous statistics on the condition of education in the United States. Their stated purposes are to “(1) provide reliable estimates of the U.S. population regarding specific education-related topics; and (2) conduct repeated measurements of the same educational phenomena at different points in time” (NCES, 2011, p. 345).
Components of the NHES Program include Adult Education, Before- and After-School Programs and Activities, Civic Involvement, Early Childhood Education and School Readiness, Household Library Use, Parent and Family Involvement in Education, and School Safety and Discipline. The current paper focuses on information gathered from the 2007 administration of the Parent and Family Involvement (PFI-NHES: 2007) survey. This survey collected information on school choice, homeschooling, school characteristics, student experiences, teacher feedback on a child’s performance and behavior in school, family involvement and help with school and schoolwork, and similar subject matter. The focus here is on information concerning the decision to homeschool a child.
NHES Surveys: Applications and Characteristics
The PFI-NHES survey was first administered in 1996 and was repeated in 1999, 2003, 2007, and most recently, in 2012. Isenberg (2006, 2007) first addressed the use of the NHES surveys for assessing the determinants and impacts of the decision to homeschool, referring to the surveys as “the richest data for studying homeschooling” (Isenberg, 2007, p. 393). He produced estimates of the number of homeschooled children and reasons for the homeschooling decision from the 1996, 1999, and 2003 administrations of the survey. These surveys contained open-ended and/or limited response (Yes/No) questions about the reasons for the decision to homeschool. The top three reasons indicated for the 1996 and 1999 administrations of the survey, which employed open-ended questions, were “better education,” “religious reasons” and “poor learning environment at school” (ibid., p. 399). In 2003, the methodology made use of Yes/No questions, but the top three reasons for the decision to homeschool remained similar: “concern about environment” of other schools, “dissatisfaction with instruction” at other schools, and “to provide religious or moral instruction” (ibid.). Because of the sensitivity of these questions to changes in format and construction, it is difficult to provide firm estimates of the relative impacts of different reasons for the decision to homeschool. Furthermore, the lack of reliable income data in these surveys makes it difficult to assess the opportunity cost of homeschooling compared to public and private school alternatives.
The target population for all NHES program surveys is the non-institutionalized civilian population of households in the United States. For the PFI survey, the educational population of interest is school-aged children. NHES samples are selected using random-digit-dialing (RDD) methods (NCES, 2011, p. 349). About 45,000 to 64,000 households are screened for each administration. Individuals within households who meet predetermined criteria, such as parent or guardian of school age children, are then sampled for more detailed or extended interviews (ibid.). Within-household sample designs for the NHES collections are determined by the specific goals of the surveys administered in that year.
For the 2007 administration of the PFI, interviews were conducted with the parents or guardians of sampled children and youth in kindergarten through 12th grade with a maximum age of 20 (NCES, 2011, p. 352). If the household had at least one child ages 3 through 20 enrolled in kindergarten through 12th grade, then exactly one parent from the household was randomly chosen for the PFI survey. NHES program surveys are conducted using computer-assisted telephone interviewing (CATI). Westat has been the contractor on all surveys to date (NCES, 2011, p. 355). Data collection takes place over a 3-4 month period beginning in January. Post-interview editing is conducted throughout data collection and after data collection is completed (ibid.).
The objective of the NHES surveys is to make inferences about the entire target population and subgroups of interest. Since only households with a telephone that can be matched to an address are surveyed, adjustments are made to totals for telephone and non-telephone households from the Current Population Survey (CPS). Weighting consists of two stages: Household-level and Person-level. Household weights take into account factors due to telephones being sampled at different rates. Two factors common to all NHES years include (1) differential sampling rates for Black and Hispanic respondents and (2) adjustments to account for households with more than one telephone number. Additionally, in 2007, weighting adjustments were made for screener non-response. Person-level weights account for factors such as unit non-response.
Isenberg (2007) produces by far the most comprehensive empirical survey of homeschool participation to date, utilizing data from state-level administrative sources as well as well as survey data from Educational Testing Services (ETS), the 1994 Current Population Survey Education Supplement (CPS), and the successive administrations of the NHES surveys in 1996-2005. The limitations of the ETS and CPS data sets are considerable, and state-level data is generally non-standardized and inconsistent (Isenberg, 2007, pp. 390-93). By contrast, the NHES data, while somewhat variable in its approach over the time period studied, is nonetheless fairly rich and, importantly, avoids many of the problems of haphazard data collection and selection bias associated with the other sources. Isenberg (2007) derives a number of reasons for homeschool participation from the 2003 NHES data, ultimately combing them into four broad categories: Educational environment, religious reasons, behavioral or special needs, and others. Estimates from these data indicates that almost half (48%) of homeschool decisions are primarily due to educational preferences, 30% primarily religious in nature, 14% for behavioral or special needs, and the rest for other reasons. However, it is notable that many parents list multiple reasons for the decision to homeschool. The current study attempts to further address the problems of understanding the importance of various preferences to homeschool.
Data are drawn entirely from the PFI-NHES: 2007, available on the eDat database of the National Center for Education Statistics (NCES, 2013). The variable of interest is whether the school-age child identified in the sampling frame is being homeschooled (HOMESCHL). Original variable names from the survey are retained where possible. The functional form of the model is that HOMESCHL should be determined by the following factors:
AGE2006: The age of the child in years
RESPAGE: The respondent’s age (parent or guardian) in years
SEX: The child’s gender (M/F)
HINCOME: The annual household income in dollars (categorical)
MOMSTAT: The mother’s marital status (categorical)
FORELCLS: Participation in religious and/or church-related activities in previous 12 months (Yes/No)
AGE2006 is employed to control for factors specific to the child’s age. There is some evidence that younger children might be more likely to be homeschooled, perhaps because parents feel that their own knowledge of the subjects covered in a grammar school curriculum is more sufficient than of secondary school subject matter. Likewise, the respondent’s (guardian’s) age (RESPAGE) might be important for similar reasons; younger parents and guardians might be more likely to consider themselves “qualified” to homeschool a child than their older contemporaries. Gender effects are controlled for by inclusion of the categorical variable SEX, while income levels are accounted for by controlling for reported household income, HINCOME. Effects of marital status are controlled for with a categorical variable MOMSTAT, a Yes/No variable relating to current marital status only. Finally, there is extensive anecdotal evidence that the homeschool decision is often influenced by family value and belief systems, particularly religious ones. Thus, a categorical Yes/No variable asking whether the child has participated in religious and/or church-related activities in the previous year, FORELCLS, is employed to capture the effects of religious affiliation or orientation.
It is also posited that racial/ethnic identity might play a part in whether a child is homeschooled or not. To gauge this effect, data on the mother’s identification with particular race/ethnic categories was employed. The categories include Hispanic, Black, American Indian, Asian, Pacific Indian, and Other. The baseline is non-Hispanic White. Each of these is surveyed in Yes/No format. The complete functional form is:
HOMESCHL = f (AGE2006, RESPAGE, SEX, HINCOME, FORELCLS, MOMSTAT, MHISPAN, MBLACK, MAMIND, MASIAN, MPACI, MRACEOTH).
Binomial logit under a general linear model is employed as the estimation method due to the dependent variable’s Yes/No characteristic. The statistical program R is used to derive coefficient estimates. The source data on HOMESCHL is coded homeschooling=1, no homeschooling=2, and NA=-1. In order to employ the binomial model these data are transformed to a (0,1) format with homeschooling=1 and all other answers coded as 0. It is assumed that an answer of NA implies that no child is being homeschooled. All other data is used exactly as presented in the NCES database. Race/ethnic variables and the religious participation variable (FORELCLS) are coded as Yes=1, No=2. SEX is coded Male=1 and Female=2. Since MOMSTAT is a non-ordinal categorical variable, it is designated as a factored variable in R, thus turning it into a set of Yes/No variables on different categories of marital status. Other data are treated as ordinal and continuous.
Data analysis indicates that just under 3% of weighted respondents indicated a homeschooled child. This is very much in line with the most recent estimates. Results are presented in Table 1 (see appendix).
Based on the derived coefficient estimates, it appears that household income, child’s gender, religious activity, and particulars about race/ethnicity are all important predictors of the decision to homeschool. Household income (HINCOME) is negatively correlated with homeschooling, presumably because homeschooling is a cheaper alternative to public schooling than entering the child into a private school. In line with economic theory, the overall demand for an ‘inferior’ good with high-quality substitutes will fall as income rises due to the predominance of substitution effects over income effects. Religious activity, proxied here by the FORELCLS variable, indicates a strong correlation with the decision to homeschool; since lower values for FORELCLS correspond to greater religious activity, these activities correlate positively with the dependent variable, indicating a greater likelihood of homeschooling. Thus, parents and guardians who place a high value on a child’s participation in religious activities appear more likely to consider homeschooling as a viable option to public and private education. In fact, consistent with the findings of Isenberg (2007) and others, the impact of religious orientation appears to be among the most important factors in the decision to homeschool, being comparable in magnitude to income and other factors.
The race/ethnicity effects are just as strong but less amenable to easy explanations. Keeping in mind that socioeconomic status is controlled for by income level, minority status in the form of self-identification as Black, Hispanic, or Pacific Indian appears to bear a strong negative correlation to homeschooling (note that higher values for these variables mean non-identification) relative to the non-Hispanic white baseline. In other words, these minority populations appear to be significantly less likely to homeschool. Asian status, on the other hand, has no effect—perhaps because of the relatively low levels of segregation among Asian and, particularly, Asian-American, groups when compared to the others—while American Indian status has a positive and statistically significant relationship to homeschooling. These findings represent an important area for future research in alternative education because of their potential consequences. To the extent that homeschooling might be a viable, long-term solution to situations of systematic underfunding in public schools, a finding that large segments of the population face significant barriers to implementing this solution should be examined carefully.
One possible explanation for race-oriented effects on homeschooling decisions is that individual states that allow homeschooling as an option to comprehensive public schools do so under a set of laws and regulations specific to that state. For example, the Code of Virginia (§22.1-254.1) requires that a parent or guardian who wishes to homeschool their child must: (i) hold a high school diploma; (ii) be a teacher of qualifications prescribed by the Board of Education; (iii) provide a program of study or curriculum which may be delivered through a correspondence course or distance learning program or in any other manner; or (iv) provide evidence that he or she is able to provide an adequate education for the child (Code of Virginia, 2008). Likewise, parents or guardians of homeschooled children must provide evidence of progress to the division superintendent’s office on an annual basis, in the form of
(i) evidence that the child has attained a composite score in or above the fourth stanine on any nationally normed standardized achievement test or (ii) an evaluation or assessment which the division superintendent determines to indicate that the child is achieving an adequate level of educational growth and progress, including but not limited to: (a) an evaluation letter from a person licensed to teach in any state, or a person with a master’s degree or higher in an academic discipline, having knowledge of the child’s academic progress, stating that the child is achieving an adequate level of educational growth and progress; or (b) a report card or transcript from a community college or college, college distance learning program, or home-education correspondence school (Code of Virginia, 2008).
These requirements, while well-intentioned, may present an unintended barrier to entry for parents and guardians of traditionally underprivileged groups. Ray (1999) notes, for example, that the parents of homeschooled children or much more likely to have a four-year post-secondary degree compared to the general public (p. 7); thus, state-level requirements may discourage the participation of certain race/ethnic groups. This would not explain the peculiar relationship of the homeschooling decision to American Indian status, however, which is more likely to be found in the specifics of tribal “sovereign nation” relations, lack of public and private schooling options, or other, yet uncovered, factors.
Finally, a child’s gender (SEX) also has a statistically significant effect on the likelihood of homeschooling, albeit at a lower level of significance than the aforementioned factors. Since the data are coded so that higher values correspond to females, this indicates that females are more likely than males to be homeschooled, even accounting for sampling differences that are adjusted for in the weighting process. Again, there is no easy explanation for this discrepancy. Perhaps concerns about the safety and/or sexual activity of female adolescents are an important motivating factor for some homeschooling decisions. As an alternative hypothesis, Sheffer (1995) suggests that young women, in particular, may be ill-served by formal education in its current form, with its emphasis on conformity, so that home education can provide a means of development unavailable in the formal system. Yet another reason for the discrepancy might be related to observations that boys are more difficult to homeschool because of behavioral characteristics. Further research is necessary to establish the causes and consequences of this high degree of female representation among homeschooled students.
This paper has presented findings from binomial logit estimation of the determinants of homeschooling using data from the 2007 administration of the National Household Education Survey (NHES). Findings indicate that household income, religious activity, race/ethnicity, and a child’s gender are all statistically significant predictors of the decision to homeschool, although there is no direct evidence of causation, only correlation. Given the fact that all of these data come from one well-published data set, the results suggest that some potentially useful quantitative data are being generated from the NHES surveys. More opportunities to exploit such data sets should be explored by quantitative researchers where possible.
Among the important findings of this study are the following. First, an often-posited relationship between religious orientation and the homeschooling decision has been confirmed. Families in which religion plays a major role, as evidenced by religious activity, are more likely to homeschool their children. Second, it is found that low-income families are also more likely to homeschool, presumably for the economic benefits of such a decision. Many such families might have otherwise decided on private schooling as an alternative choice, but find that the resources necessary are lacking. Third, a child’s gender is found to be a statistically significant indicator of the decision to homeschool, with females more likely to be homeschooled than males. Finally, the self-identified race/ethnicity of the parents is also a significant indicator, with Blacks, Hispanics, and Pacific Indians being less likely to homeschool after controlling for income and other factors. Each of these last two effects (gender and race/ethnicity) is particularly worthy of further exploration.
Indeed, given the widely acknowledged benefits from home education for both the welfare and educational attainment of students, in addition to the fiscal benefits of reducing the demand for public resources, these issues of access to homeschooling as a viable alternative to both public and private schooling are likely to become increasingly important in the near future. There is sufficient evidence of widening achievement gaps between black and white K-12 students, as well as between male and female students, to argue for an expansion of available alternatives (see Perie, et al., 2005). Homeschooling can provide a valuable option within this menu of alternatives. Solving problems of access to this option for parents of all students, regardless of race, gender or socioeconomic status, should be at the forefront of objectives for home education proponents in the U. S. and everywhere.
Code of Virginia, Home Instruction Statute (§22.1-254.1). (2008, amended). Available at http://leg1.state.va.us/cgi-bin/legp504.exe?000+cod+22.1-254.1. Last accessed on March 15, 2013.
Clemente, D. (2006). Academic Achievement and College Aptitude in Homeschooled High School Students Compared to their Private-schooled and Public-schooled Counterparts (doctoral dissertation). Regent University, Virginia Beach, VA.
Dotterweich, D., M. McKinney and L. M. Michael. (2013). “Economic Impact of Home and Private Schooling on the Public Education System.” Home School Researcher, 29(3): 1-11.
Gray, S. (1993). “Why Some Parents Choose to Home School.” Home School Researcher, 9(4): 1-12.
Green, C. and K. V. Hoover-Dempsey. (2007). “Why Do Parents Homeschool? A Systematic Examination of Parental Involvement.” Education and Urban Society, 39(2): 264-85.
Isenberg, E. (2006). The Choice of Public, Private, or Home Schools (Occasional Paper No. 132). New York: National Center for the Study of Privatization in Education, Teachers College, Columbia University.
Isenberg, E. (2007). “What Have We Learned About Homeschooling?” Peabody Journal of Education, 82(2-3): 387-409.
Lips, D. and E. Feinberg. (2008). “Homeschooling: A Growing Option.” April 3, 2008.. Heritage Foundation. Available at http://www.heritage.org/research/reports/2008/04/homeschooling-a-growing-option-in-american-education. Last accessed on March 15, 2013.
Mackey, B. W., Reese, K., & Mackey, W. C. (2011). “Demographics of Home Schoolers: A Regional Analysis within the National Parameters.” Education, 132(1), 133-140.
NCES, U.S. Department of Education. (2006). “Homeschooling in the United States: 2003 Statistical Analysis Report.” Available at http://nces.ed.gov/pubs2006/homeschool/. Last accessed on June 30, 2015.
NCES, U.S. Department of Education. (2011). NCES Handbook of Survey Methods. 2nd edition. S. Burns, X. Wang, and A. Henning, eds. Available at http://nces.ed.gov/nhes. Last accessed on March 15, 2013.
NCES, U.S. Department of Education. (2013). “National Household Education Surveys Program.” Available at http://nces.ed.gov/nhes. Last accessed on October 21, 2013.
Perie, M., Moran, R., and Lutkus, A. D. (2005). NAEP 2004 Trends in Academic Progress: Three Decades of Student Performance in Reading and Mathematics. U. S. Department of Education, Institute of Education Sciences, National Center for Education Statistics. Washington, DC: Government Printing Office.
Ray, B. (1999). Home Schooling on the Threshold: A Survey of Research at the Dawn of the New Millennium. National Home Education Research Institute.
Ray, B. (2004). “Homeschoolers on to College: What Research Shows Us.” Journal of College Admission, 185, 5-11.
Ray, B. (2007). “On Blacks Choosing Home-Based Education.” Home School Researcher, 17(4): 9-12.
Ray, B. (2011). “Research Facts on Homeschooling.” June 8, 2011. National Home Education Research Institute. Available at http://www.nheri.org/ResearchFacts.pdf. Last accessed on March 15, 2013.
Rudner, L. (1999). The Scholastic Achievement of Home School Students. College Park, MD: ERIC Clearinghouse on Assessment and Evaluation.
Sande, J. A. (1995). “The Impact of Home Schooling on Math Education.” Home School Researcher, 11(3): 7-15.
Sheffer, S. (1995). A Sense of Self: Listening to Homeschooled Adolescent Girls. Portsmouth, NH: Heinemann Educational Books.
Snyder, M. (2013). “An Evaluative Study of the Academic Achievement of Homeschooled Students Versus Traditionally Schooled Students Attending a Catholic University.” Catholic Education: A Journal of Inquiry & Practice, 16(2), 288-308.
Logistic Regression of the Decision to Homeschool
|Table 1:||Estimate||Std. Error||t value||Pr(>|t|)|
*, **, and *** indicate coefficient estimates significant at the .10, .05, and .01 level, respectively. ¯
 The initial findings of the 2011-12 NCES Parent and Family Involvement survey report 3.4% of families homeschooled at least one of their K-12 aged children during that academic year.
 For more on attainment and resource impacts of homeschooling, see Sande (1995), Ray (1999, 2004), and Dotterweich, et al. (2013).
 An important exception, Isenberg (2007), is discussed below.
 The PFI was administered again in the 2011-12 academic year, and the initial report of findings has been presented as of August 30, 2013 (see NCES, 2013). As of this time, however, the data set has not been made available to the public.
 More on the Isenberg (2007) analysis of the 2003 data is presented later in this section.
 For example, the 1999 NCES survey indicated that 50.4% of homeschooled children were in grades K-5, compared to 49.6% in the next seven grades. See NCES (2006), Table 3.
 Due to the categorical nature of most of the independent variables, detailed comparisons of magnitude in these effects on the homeschool decision are not possible.
 Another explanation, put forward by Ray (2007), is that non-evidence-based ideological or philosophical critiques discourage participation among some race/ethnic groups.
 I thank an anonymous referee for pointing out this possibility.