Homeschooling Choice and Timing: An Examination of Socioeconomic and Policy Influences in Wisconsin

ABSTRACT

Over the past two decades homeschooling has become increasingly popular, but this educational alternative has lacked rigorous empirical evaluation because of data limitations. Since little data are available for individual students, we examine homeschooling participation at the statewide and district level in Wisconsin. The most compelling finding is the large decrease in homeschooling at the upper levels of high school which may distort the evaluation of homeschooling as preparation for college. We also examine district and community factors associated with overall homeschooling participation and find evidence, for example, of the importance of test scores and specific religious preferences to that choice. Specifically, we find that higher district level homeschool participation is associated with lower district grade school test scores, lower expenditure per pupil, and a lower percentage of Catholic individuals living in the surrounding area.

KEYWORDS:

• homeschool
• homeschoolers
• homeschooling

JEL CLASSIFICATION CODES:

• I2
• I21

Acknowledgments

The authors wish to thank the University of Wisconsin–Whitewater Fiscal and Economic Research Center, the University of Wisconsin Extension, and the University of Wisconsin–Whitewater College of Business and Economics for financial support for data collection. We would also like to thank Quinton Baker, Matthew Le Donne, and Justin Kasper for research assistance in collecting this data and other related activities.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. U.S. Department of Education, National Center for Education Statistics, Parent Survey of the National Household Education Surveys Program (NHES), Parent and Family Involvement in Education Survey of the NHES, 2003, 2007, and 2016.

2. For an excellent reviews on the history of homeschooling, the reasons parents choose homeschooling, and more detail on the relevant literature, see Isenberg (2007) and Murphy and colleagues (2017).

3. We do find some evidence to support our argument. We examine several recent years of virtual school enrollment data. In a cohort to cohort comparison we see an increase in enrollment between 10th to 11th grade of approximately 48% in virtual schools, however, we do see a decline of approximately 21% between 11th and 12th grade.

4. It is unclear how homeschooling is reported to colleges—either as information required with the application or simply volunteered by the applicant. Our argument may be weaker in the latter case.

5 The full dataset actually includes 2,549 observations because in 2006/2007 two districts consolidated, leaving only 424 districts, while in 2007/2008 two other districts split apart bringing the number back to 425.

6. The community characteristics describe those who live in the geographic area of the district whether or not they have children in K–12 education.

7. Some districts did not report Free and Reduced Lunch Eligible; we coded these as zero because most of these districts did not participate because they were high-income districts.

8. The Racial Herfindahl index = $\sum _{i=1}^N{\left(\mathrm{P}\mathrm{r}\mathrm{o}\mathrm{p}\mathrm{o}\mathrm{r}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{f}\mathrm{R}\mathrm{a}\mathrm{c}{\mathrm{e}}_i\right)}^2$. The racial Herfindahl will be between zero and one. A higher Herfindahl index will imply less diversity. We can include the proportion of each of five races: Asian, Black, Hispanic, American Indian, and White.

9. The variable “membership” is similar to the total number of students except that it adjusts for part-time attendance (e.g., a 4-year old kindergarten student who attends only in the morning would be considered half a student).

10. In Wisconsin, most districts are “K–12 districts” that offer a full range of grades, but there are 47 “K–8 districts” and nine “UHS districts” (“union high school” districts). These K–8 districts offer only K–8 education and “feed into” UHS districts that are only high school districts. There were only 46 K–8 districts starting with the 2006/2007 school year because the consolidation mentioned previously involved two K–8 districts.

11. “Advanced” is the highest level a student can earn and “minimal” is the lowest level. Advanced performance level means that students were able to perform grade-level skills and use strategy and critical thinking to draw conclusions or apply knowledge. Proficient performance level means that students were able to perform grade-level skills adequately. Basic performance level means students can perform some grade-level skills and can understand below grade-level material. Minimal performance level means students can perform some below grade-level skills. According to WDPI: “The long-term goal is for all students, except students with severe disabilities, to progress to the Proficient and Advanced levels”.

12. The classification of various religious bodies into “evangelical Christian,” “Mainline Protestant,” “Catholic,” and “Other” categories was created by the staff of the American Religion Data Archive (ARDA). These definitions are adapted from Steensland and colleagues (2000).

13. Note, if we include total homeschooled students in the second equation as well, we get similar results; but then we are relying on the functional form of the Probit in the first stage for identification.

14. We do not discuss the results of the selection equation in the results section, but we do include a table of these results in the Appendix as Table A1. Of particular note is the statistical significance of the estimated coefficient on lagged homeschoolers, our variable excluded from the outcome equation.

15. Negative binomial specifications measure approximately what the proportionate change (or percentage change if the coefficient is multiplied by 100) in the dependent variable will be given a one-unit change in the independent variable (if the independent variable is in level and not log form). Technically, it measures how much the difference in logs of the expected counts changes (or the log of the ratio of counts) for a one-unit change in the independent variable: this is approximately the proportion if the proportion is small.

16. We note that this most likely does not reflect the parents’ education level and preference for homeschooling, since our community characteristics already control for the education level of the citizens in the area.

17. However, a counterargument to this explanation is that our model controls for number of students (membership) which will be somewhat correlated with number of schools.