Research proposal.

PredictingCollegePerformanceofHomeschooledver.pdf

Home >Psychology homework help >Research proposal.

Educational Measurement: Issues and Practice Winter 2016, Vol. 35, No. 4, pp. 31–39

Predicting College Performance of Homeschooled Versus Traditional Students

Martin C. Yu, Paul R. Sackett, and Nathan R. Kuncel, University of Minnesota, Twin Cities

The prevalence of homeschooling in the United States is increasing. Yet little is known about how commonly used predictors of postsecondary academic performance (SAT, high school grade point average [HSGPA]) perform for homeschooled students. Postsecondary performance at 140 colleges and universities was analyzed comparing a sample of traditional students matched to a sample of 732 homeschooled students on four demographic variables, HSGPA, and SAT scores. The matched sample was drawn from 824,940 traditional students attending the same institutions as the homeschooled students, which permitted a very precise level of matching. This comparison did not show a difference in first-year college GPA (FGPA) or retention between homeschooled and traditional students. SAT scores predicted FGPA and retention equally well for both groups, but HSGPA was a weaker predictor for the homeschooled group. These results suggest that, among college students, those who were homeschooled perform similarly to traditionally educated students matched on demographics and academic preparedness, but there are practical implications for college admissions in the use of HSGPA versus standardized test scores for homeschooled students.

Keywords: admissions, college, homeschool, predictive bias

T he prevalence of homeschooled students in the UnitedStates has been steadily increasing, and as of the 2011– 2012 school year, this amounts to 3.4% of the K-12 student pop- ulation in the United States (Noel, Stark, & Redford, 2013). Prominent reasons for why parents choose to homeschool their children include a concern about the environment of other schools, a dissatisfaction with academic instruction at other schools, a desire to provide moral or religious instruc- tion, or a desire to provide a nontraditional approach to their child’s education (Noel et al., 2013). Taken together, these reasons suggest that many parents who choose to homeschool their children intend for it to replace and to potentially pro- vide a better education than traditional schooling systems in the United States.

At the same time, because homeschooling in the United States is regulated differently from traditional education, and because homeschooling regulations vary among states (Ruger & Sorens, 2013), we are concerned with whether homeschool- ing would yield particularly meaningful grades and whether it would produce students who would be as effective in col- lege as traditional students. To address this, we examine whether high school grades and standardized test scores for homeschooled students are predictive of college grades and

Martin C. Yu, Doctoral Student, Department of Psychology, Uni- versity of Minnesota, 75 East River Road, Minneapolis, MN 55455; yuxx0407@umn.edu. Paul R. Sackett, Beverly and Richard Fink Distinguished Professor of Psychology and Liberal Arts, Univer- sity of Minnesota; psackett@umn.edu. Nathan R. Kuncel, Marvin D. Dunnette Distinguished Professor of Psychology, University of Minnesota; kunce001@umn.edu.

retention, and we do so by comparing these relationships with those observed for a large matched sample of traditional students to determine: (1) if traditional college admissions information is equally predictive for the two groups and (2) if homeschooling is associated with better college grades or retention when controlling for standardized test scores, high school grades, and socioeconomic status (SES).

Descriptively, homeschooled students appear to differ from the national average in high school performance and other characteristics. Survey studies conducted by Rudner (1999) and Ray (2000, 2010) found that on standardized achieve- ment tests for various K-12 subjects (e.g., reading, math, sci- ences) the mean national percentile for homeschooled stu- dents ranged approximately from the 70th to 90th percentile, compared to the 50th percentile for the overall national aver- age. They also tend to come from families with higher levels of education and income (Ray, 2000, 2010; Rudner, 1999). This suggests that, on average, homeschooled students may be showing better academic performance due to factors other than homeschooling. Therefore, homeschooled students may be a unique group with characteristics not proportionately represented in the general student population, so it is pos- sible that the higher performance of homeschooled students may be explained by other factors, meaning that direct com- parisons between homeschooled students and traditional stu- dents in general may not be appropriate.

For postsecondary academic outcomes, the existing re- search is limited by small sample sizes of homeschooled stu- dents from a small number of postsecondary institutions, but the research does provide some support that homeschooled students perform similarly to other students. Jones and

Gloeckner (2004a) compared 55 homeschooled first-year stu- dents from public colleges or universities in Colorado with a sample of 53 traditional students who attended the same insti- tutions. They found no significant differences in ACT scores, first-year grade point average (GPA), first-year retention (fall to spring semester), and first-year credit hours earned. Cogan (2010) examined academic data from 7,776 students from a private university in the upper Midwest, of which 76 students reported that they were homeschooled. In this study, home- schooled students were found to have significantly higher ACT composite scores and higher first-year and fourth-year GPA, but there was no effect of homeschooling on retention or graduation rates. Interpretation of these findings is difficult, as students do not apply to college at random and are not admitted at random, meaning that the characteristics of col- lege students who were homeschooled may be different from those who were traditionally educated. The impact of home- schooling on students can only be inferred from these studies if assumptions are made about self-selection effects during application and the admissions decision. What can be learned from enrolled students is the extent to which characteristics of homeschooled students are related to academic achieve- ment in college and whether these relationships are similar in magnitude to those observed for students who attended traditional public and private schools.

A question of practical and scientific interest is whether the academic performance of homeschooled students dur- ing high school can effectively predict college performance. The predictive validity of high school GPA and standardized tests (e.g., ACT, SAT) for academic performance in college has been well established in previous research (e.g., Sackett, Kuncel, Arneson, Cooper, & Waters, 2009). To our knowledge, however, it is currently unknown whether the predictive va- lidity of high school GPA and standardized tests for college performance generalizes to homeschooled students, and if homeschooling is a moderator of the relationship between high school performance and college performance. For pur- poses such as college admissions, knowing if there is any predictive bias that depends on whether a student is home- schooled or not will help to better inform admissions deci- sions. For example, is a B high school grade point average for home schooled students associated with the same level of per- formance in college as a B average for traditionally educated students?

In this study, we aim to expand and to improve on the ex- isting research on the academic outcomes of homeschooled students as compared to traditional students, and to address the lack of knowledge as to what predicts performance dur- ing college or university for homeschooled students. Using a nationwide sample of 732 homeschooled students enrolled in college paired with a closely matched sample drawn from 825,672 first-year students at 195 postsecondary institutions across the United States, we first conduct a descriptive analy- sis to provide normative comparisons between homeschooled and traditional students. Then, we examine any moderating effects of homeschooling on the predictive validity of high school GPA, SAT scores, and socioeconomic status for first- year college GPA and retention after first-year.

Previous studies on this topic have compared samples of homeschooled students to some overall sample of tradi- tional students, but we are concerned about whether doing so would actually provide meaningful results. Because the characteristics of homeschooled students enrolled in college

may not be proportionally represented among traditional stu- dents enrolled in college, we believe that, in order to effec- tively isolate the effects of homeschooling from other student characteristics on college performance, the most appropri- ate comparison would be between homeschooled students and a sample of traditional students matched to as many characteristics of the homeschooled students as possible. For example, Jones and Gloeckner (2004a) matched each home- schooled student in their study with a traditional student from the same postsecondary institution. However, match- ing only on postsecondary institution may be insufficient, as there are many other potential variables, such as gender, eth- nicity, and SES, that may potentially be disproportionately represented in samples of college-attending homeschooled versus traditional students, especially when the sample of homeschooled students obtained for a study may be limited in size and not necessarily representative of the population of homeschooled students. Additionally, since homeschooled students have been found to perform better in high school than traditional students (Ray, 2000, 2010; Rudner, 1999), further matching based on measures of high school achieve- ment such as high school GPA and scores on standardized college admissions tests will be necessary to account for prior differences in college preparedness when examining differ- ences in academic outcomes during college. The main chal- lenge with generating a matched sample is that it requires drawing from a large pool of traditional students to be feasi- ble, and the size of our sample provides such an opportunity. Ultimately, analyses using matched samples will provide a more meaningful analysis of the effects of homeschooling on academic performance.

Method Sample

Data for this study were provided by the College Board on 825,672 first-year students from 2009 to 2011 at 195 postsec- ondary institutions across the United States, of which 732 stu- dents at 140 of these postsecondary institutions were home- schooled prior to admission (specifically, at the time they took the SAT).

Measures

SAT scores. SAT scores consisted of scores on the three sections of the SAT: Math, Critical Reading, and Writing. These scores were averaged into a composite SAT score for each student.

High school GPA. Two forms of high school GPA (HSGPA) were provided in the data set. One was self-reported by the student at the time of taking the SAT, and the other was school-reported by a subset of the postsecondary institutions based on their own calculations of each of their student’s high school GPA. Self-reported GPA has been found to be less accurate than GPA that is not self-reported, particularly for students with low GPA (Kuncel, Credé, & Thomas, 2005). Because of that, we conducted analyses using both forms of HSGPA so that any results idiosyncratic to either self- or school-reported high school GPA could be identified. Our conclusions turned out the same no matter which form of GPA was used in the analysis, so only the results from the analyses

using the school-reported HSGPA will be presented for the sake of simplicity.

Socioeconomic status. At the time they took the SAT, stu- dents reported their father’s education, mother’s education, and parental income. The natural logarithm of parental in- come was used in this study. A composite SES score was calculated by equally weighting these three SES variables using a method described by Sackett et al. (2009).

First-year college GPA. First-year college GPA (FGPA) was provided by each postsecondary institution. To account for differences in difficulty, grading policies, or grading scales that may result in similar students obtaining different grades at different colleges or universities, the first-year college GPA of each student was adjusted based on the procedure used by Cullen, Hardison, and Sackett (2004). This procedure ad- justed the FGPA for each student based on the expected difference in FGPA for students with similar SAT scores who attend different postsecondary institutions.

Retention after first year. Each institution reported whether each of their students was retained from first year to second year.

Matching Procedure

From the overall sample of 824,940 traditional students, a subset was matched to the homeschooled students on post- secondary institution, gender, ethnicity, HSGPA, SAT, and SES. For the purposes of the matching procedure only, HS- GPA, SAT, and SES were standardized within institution so that they would all be on the same scale with a mean of 0 and SD = 1. The matching procedure was carried out se- quentially for each homeschooled student. First, the whole sample of traditional students was subset into a smaller sam- ple matched to the postsecondary institution, gender, and ethnicity of the homeschooled student. Next, the absolute differences in standardized HSGPA, SAT, and SES between the homeschooled student and each traditional student in this subset were then averaged to create a matching index. For matching on HSGPA, the school-reported HSGPA was used if available, and the self-reported HSGPA was used only if the school-reported HSGPA was missing. Any variable with data missing for the homeschooled student was excluded from consideration in the matching process. The traditional stu- dent with the closest match to the homeschooled student (i.e., smallest matching index score) was selected into the matched sample of traditional students. To prevent the same student from being selected more than once, the selected stu- dent was then removed from consideration in matching with subsequent homeschooled students. This process was then repeated for each homeschooled student.

This matching process was able to closely match each homeschooled student with a traditional student, produc- ing nearly identical matches on the six matching variables. The distribution of matching index scores in the sample of matched students clustered towards zero (Figure 1) had a median of .03, a mean of .07, and SD = .13.

The homeschooled sample consisted of 732 students from 140 postsecondary institutions, was 48.2% male, and had an ethnic composition of 79.0% White, 4.2% Hispanic, 2.9% Black,

2.7% Asian, 0.3% American Indian, and 2.9% other ethnicity (percentages for ethnicity do not sum to 100% due to non- response). The full sample of traditional students consisted of 824,940 students from 195 postsecondary institutions, was 46.1% male, and had an ethnic composition of 52.0% White, 7.7% Hispanic, 6.7% Black, 8.4% Asian, 0.4% American Indian, and 2.0% other ethnicity. The matched sample of traditional students consisted of 732 students from the same 140 postsec- ondary institutions as the homeschooled students, was 48.2% male, and had an ethnic composition of 83.7% White, 4.5% Hispanic, 3.1% Black, 3.1% Asian, 0.3% American Indian, and 3.0% other ethnicity.

Analyses

Missing data for the sample of homeschooled students and the matched sample of traditional students were imputed by mul- tiple imputation using predictive mean matching (Schenker & Taylor, 1996). This was done separately for the sample of homeschooled students and the matched sample of tradi- tional students. Missing data were not imputed for the full sample of traditional students due to the size of this sample and because our focal analyses involved the homeschooled and matched sample.

Means, standard deviations, and intercorrelations for all study variables were computed separately for homeschooled students and the full and matched samples of traditional students.

Using the matched samples, moderated multiple regression analysis was carried out to determine whether the predictive validities of HSGPA and SAT for postsecondary outcomes of FGPA and retention after first-year (RET) were moderated by whether a student was homeschooled or not. Linear regres- sion was used to fit models where FGPA was the criterion, whereas logistic regression was used to fit models where RET was the criterion (RET was dummy coded as 1 = retained to second year, and 0 = not retained to second year). Models were fit using hierarchical regression analyses, with separate analyses conducted for each criterion variable (FGPA and RET) and each predictor variable (HSGPA and SAT). To do so, FGPA was first regressed onto HSGPA as the focal predic- tor in the initial step. Homeschooling and its interaction with HSGPA were then entered in the second step, followed by SAT entered as a control variable in the third step, and finally SES was entered as a control variable in the fourth step. A similar analysis was carried out using SAT as the focal predictor and HSGPA as a control variable. This was then repeated using RET as the criterion instead of FGPA.

Results Means, standard deviations, and intercorrelations for all study variables are shown in Tables 1–3 for homeschooled students and the full and matched samples of traditional students, respectively. Compared to the full sample of traditional stu- dents, homeschooled students on average had higher HSGPA, SAT, FGPA, and SES, but there was not any difference in re- tention. On the other hand, when compared to the matched sample of traditional students, homeschooled students ap- peared to show no differences in FGPA and RET. There were also no differences in HSGPA, SAT, and SES in the matched samples comparison, reflecting the fact that close matches were obtained on these variables that were used for matching.

FIGURE 1. Histogram of matching index scores from the matched sample of traditional students.

Comparing the homeschooled students to both the full and matched samples of traditional students, the correlations be- tween HSGPA and the two postsecondary outcomes of FGPA and RET were substantially higher for traditional students, whereas there did not appear to be any substantial differ- ences in the correlations between SAT and FGPA or RET. For homeschooled students, SAT had a stronger correlation with FGPA and RET than did HSGPA. SES was more strongly cor- related with SAT for traditional students in the full sample, and this may be explained as being due to a restriction in the range of SES in the sample of homeschooled students and the matched sample of traditional students.

Table 1. Means and Intercorrelations for the Sample of Homeschooled Students

Variable M SD 1 2 3 4 5

1. HSGPA 3.80 .25 2. SAT 603.54 83.52 .47 3. FGPA 1.07 .93 .21 .39 4. RET .88 .32 .04 .13 .24 5. SES .32 .67 .09 .21 .20 .11

Note. n = 732; HSGPA = high school GPA; SAT = SAT composite score; FGPA = first-year college GPA; RET = college retention after first year; SES = socioeconomic status. SAT is the average of scores on the three SAT sections. FGPA and SES are computed from rescaled variables, so their values are not directly interpretable. RET is coded as retained = 1, not retained = 0.

Regression analyses predicting FGPA are shown in Table 4. The step 1 models for either HSGPA or SAT as focal predictors show that both HSGPA and SAT independently predicted FGPA. In the step 2 models, the interaction of homeschooling with HSGPA or SAT were both significant, but the results started to diverge on subsequent steps. After controlling for SAT, and then for SES, the interaction of homeschooling with HSGPA was still significant. On the other hand, after controlling for HSGPA, the interaction of homeschooling with SAT was no longer significant, and this held after SES was controlled for as well. Comparing the regression lines between homeschooled and traditional students, Figure 2 illustrates

Table 3. Means and Intercorrelations for the Matched Sample of Traditional Students

Variable M SD 1 2 3 4 5

1. HSGPA 3.77 .31 2. SAT 599.06 81.52 .51 3. FGPA .99 .93 .51 .49 4. RET .89 .31 .20 .19 .32 5. SES .33 .69 .10 .30 .14 .13

Table 2. Means and Intercorrelations for the Full Sample of Traditional Students

Variable M SD 1 2 3 4 5

1. HSGPA 3.46 .48 439,163 354,727 431,887 439,160 338,726 2. SAT 557.81 87.87 .50 643,817 633,948 643,815 609,149 3. FGPA .64 1.04 .49 .43 811,076 811,071 607,265 4. RET .85 .35 .22 .18 .34 824,935 616,622 5. SES .02 .96 .15 .41 .22 .11 616,624

Note. Numbers on the diagonal indicate the sample size for each variable, and numbers above the diagonal indicate the sample size for each correlation. HSGPA = high school GPA; SAT = SAT composite score; FGPA = first-year college GPA; RET = college retention after first year; SES = socioeconomic status. SAT is the average of scores on the three SAT sections. FGPA and SES are computed from rescaled variables, so their values are not directly interpretable. RET is coded as retained = 1, not retained = 0.

FIGURE 2. Relationship between high school GPA and first-year college GPA by homeschooling status (homeschooled = 1, traditional = 0). The solid line represents traditional students and the dashed line represents homeschooled students.

the large slope difference when HSGPA was the predictor, and Figure 3 illustrates the negligible slope difference when SAT was the predictor.

Regression analyses predicting RET are shown in Table 5. Similar to what was found for FGPA, the step 1 models for

either HSGPA or SAT as focal predictors show that both HS- GPA and SAT independently predicted RET. The step 2 models show that the interaction of homeschooling with HSGPA was significant, but its interaction with SAT was not. This result held for subsequent steps controlling for SAT or HSGPA, and

FIGURE 3. Relationship between SAT composite score and first-year college GPA by homeschooling status (homeschooled = 1, traditional = 0), controlling for high school GPA and SES. The solid line represents traditional students and the dashed line represents homeschooled students.

Table 4. Hierarchical Regressions With HSGPA and SAT as Focal Predictors of First-Year College GPA, Moderated by Homeschooling

Focal Predictor Model Predictor �

Std. Error R2

HSGPA 1 HSGPA 1.219** .080 .137 2 HSGPA 1.531** .103 .151

Home 2.991** .616 Home ×

HSGPA −.778** .162

3 HSGPA .996** .105 .243 Home 3.292** .582 SAT .004** .000 Home ×

HSGPA −.858** .153

4 HSGPA 1.004** .105 .247 Home 3.283** .581 SAT .003** .000 SES .087** .032 Home ×

HSGPA −.855** .153

SAT 1 SAT .005** .000 .196 2 SAT .006** .000 .200

Home .782* .321 Home ×

SAT −.001* .001

3 SAT .004** .000 .229 Home .580 .316 HSGPA .645** .087 Home ×

SAT −.001 .001

4 SAT .004** .000 .232 Home .532 .316 HSGPA .654** .087 SES .085** .032 Home ×

SAT −.001 .001

Note. *p < .05, **p < .001. Home = homeschooled; HSGPA = high school GPA; SAT = SAT composite score; SES = socioeconomic status. Home is coded as homeschooled = 1, traditional = 0.

then controlling for SES. Figure 4 illustrates the large slope difference when HSGPA was the predictor, and Figure 5 il- lustrates the negligible slope difference when SAT was the predictor.

Discussion Previous large-scale studies have found that, on average, homeschooled students are superior to traditional students on academic outcomes, and also differ on other character- istics such as parental education and income (Ray, 2000, 2010; Rudner, 1999). In our comparison of students enrolled in college, we also found that when compared to the over- all sample of traditional students, homeschooled students came from families with higher SES, had obtained better test scores, and earned better grades in high school and col- lege. However, direct comparisons between homeschooled students and traditional students in general may not be ap- propriate given that the characteristics of homeschooled stu- dents may not be proportionately represented in the gen- eral student population. Indeed, when we compared our sample of homeschooled students who were enrolled in col- lege to a more representative group of traditional students matched on postsecondary institution, gender, race, academic

Table 5. Hierarchical Regressions With HSGPA and SAT as Focal Predictors of College Retention After First Year, Moderated by Homeschooling

Focal Predictor Model Predictor �

Std. Error

HSGPA 1 HSGPA 1.143** .249 2 HSGPA 1.161** .323

Home 3.991* 1.977 Home × HSGPA

−1.124** .528

3 HSGPA 1.053** .351 Home 4.891* 2.097 SAT .005** .001 Home × HSGPA

−1.366* .561

4 HSGPA 1.087** .354 Home 4.859* 2.096 SAT .045** .001 SES .370** .118 Home × HSGPA

−1.358** .560

SAT 1 SAT .006** .001 2 SAT .008** .002

Home 1.578 1.189 Home × SAT

−.003 .002

3 SAT .007** .002 Home 1.402 1.192 HSGPA .527 .297 Home × SAT

−.003 .002

4 SAT .006** .002 Home 1.250 1.200 HSGPA .557** .299 SES .364** .118 Home × SAT

−.003 .002

Note. *p < .05, **p < .001. Home = homeschooled; HSGPA = high school GPA; SAT = SAT composite score; SES = socioeconomic status. HS is coded as homeschooled = 1, traditional = 0.

preparedness, and socioeconomic status, performance differ- ences between homeschooled and traditional students effec- tively disappeared. Therefore, while homeschooled students can be as successful as traditional students in college or uni- versity, simply having been homeschooled does not appear to be an advantage or disadvantage among students enrolled in college.

The lack of a difference in first-year retention (RET) in- dicates that homeschooled and traditional students advance from the first to second years of college at the same rate, which is consistent with the findings of previous studies (Co- gan, 2010; Jones & Gloeckner, 2004a). Also, regardless of homeschooling, both SAT and SES were correlated with RET, a result that has been observed in overall student samples (Mattern & Patterson, 2009). Advancement from year to year typically requires passing courses to obtain some minimum number of credits, and students with higher ability are more likely to do well in their coursework. If the cost of education is a factor, students higher in SES would be more likely to proceed through college.

Interestingly, the correlations between HSGPA and both FGPA and RET for homeschooled students are weaker than those for both the full and matched samples of traditional

FIGURE 4. Probability of being retained after first year of college depending on high school GPA by homeschooling status (homeschooled = 1, traditional = 0). The solid line represents traditional students and the dashed line represents homeschooled stu- dents.

students. The same issue was not present for SAT predicting FGPA or RET because the relationships are similar for all three groups. From Figure 2, it is clear that HSGPA performs much worse as a predictor of FGPA for homeschooled students than for traditional students, and from Figure 3 SAT can be seen to predict FGPA similarly for both homeschooled and traditional students. As depicted in Figures 4 and 5, a similar pattern of results can be observed when predicting RET as well. Therefore, in contrast to the high school grades of tradi- tionally educated students, the high school grades of college students who were homeschooled do not appear to be as use- ful for predicting their college grades or their probability of being retained from first to second year of college.

As it stands, the only differences observed in this study between homeschooled and traditional students who were enrolled in college is the differential prediction of first-year college GPA and retention by high school GPA. In sum, while the SAT composite score predicts FGPA and RET equally well for both groups, HSGPA is a drastically worse predictor of FGPA and RET for the homeschooled group. Furthermore, when homeschooling is accounted for, if SAT is used to pre- dict FGPA, adding HSGPA provides a smaller incremental prediction (�R2 = .03) when compared to adding SAT to HS- GPA (�R2 = .09). In other words, the SAT is a substantially more useful predictor of college success for homeschooled students than is HSGPA.

At this point, we can only speculate as to why high school GPA is a worse predictor for homeschooled students. The most apparent explanation is that the SAT is a standardized test whereas high school GPA is not, leading to the possibility that the reliability of high school GPA for homeschooled students may be lower than that for traditional students. Because states

widely vary in their regulation of homeschooling (Ruger & Sorens, 2013), whether HSGPA is a useful predictor or not may depend on how strictly homeschooling is regulated and how carefully grades are assigned. Unfortunately, we could not account for between-state differences in our analyses as there were no indications of where each homeschooled student was homeschooled in the data set used for this study. Other possible reasons include different norms for grading homeschooled students, a difference when teachers have to grade just one or a few students in a homeschool setting versus an entire class of students in a traditional setting, or the postsecondary institution having a different policy for obtaining or calculating the admission GPA of homeschooled students. Further research will be needed to pinpoint exactly how the properties of HSGPA differ between these two types of students.

Regardless of why HSGPA is an inferior predictor of FGPA for homeschooled students compared to traditional students, the predictive bias due to homeschooling was found in the analyses using the matched samples that attempted to iso- late as much as possible the effects of homeschooling from other variables that may explain college performance, so it is likely to be a real effect with practical implications for college admissions. Whereas the SAT is an equally effective predic- tor for both homeschooled and traditional students, HSGPA appears to do a poorer job of differentiating between which homeschooled students are more likely to perform well in their first year of college than it does for traditional students, so it would follow that the use of HSGPA for college admis- sions should be considered differently depending on whether the applicant was homeschooled or not. For example, when evaluating homeschooled students, more emphasis could be

FIGURE 5. Probability of being retained after first year of college depending on SAT composite score by homeschooling status (home- schooled = 1, traditional = 0). The solid line represents traditional students and the dashed line represents homeschooled students.

placed on SAT scores over HSGPA. Obviously, the reality of actually doing so would be dependent on the admissions poli- cies of each institution. Some institutions may be or may have been using such an admissions policy: in a survey of 55 admis- sions officers at U.S. colleges, Jones and Gloeckner (2004b) found that 74.5% of them had an official homeschool admis- sions policy and a high importance was placed on SAT or ACT scores for homeschooled students. That said, if policies are altered to account for differential prediction for home- schooled students, care should be taken so that they are not inadvertently put at a disadvantage in the admissions process.

The strength of this study lies in the analyses using a sam- ple of traditional students matched to the characteristics of the homeschooled group, and the large pool of traditional students that allowed close matches to be found for each homeschooled student. This provided findings that were more meaningfully interpretable in contrast to the analyses with the overall group of traditional students. However, due to the fact that all of the students in our sample have attended a college or university, the generalizability of our findings may be limited to the population of students who at least meet the requirements that deem them acceptable to attend such insti- tutions. Therefore, despite the conclusion that homeschooled students who attend college can be as successful as their tra- ditionally educated counterparts, we currently cannot make any conclusions about the performance of homeschooled stu- dents relative to traditional students for those students who do not meet admissions requirements or who choose not to at- tend a college or university. Further research will be needed to address these other areas of comparison, and we suggest that future research comparing homeschooled students to tradi- tional students should use a sample of traditional students matched to as many characteristics of the homeschooled

students as possible. As demonstrated in this study, differ- ing conclusions can be obtained depending on whether the homeschooled students are compared to a matched sample of traditional students or some overall sample of traditional students. Given that the characteristics of homeschooled stu- dents are not proportional to those of traditional students, we believe that more realistic conclusions can be obtained from matched sample comparisons.

In summary, the results of this study provide positive support that college students who were homeschooled can perform as well as those who received a traditional educa- tion. However, while SAT scores predict college performance equally well for both homeschooled and traditional students, the high school GPA of homeschooled students should be carefully considered because it may not be as predictive of their performance in college compared to the high school GPA of traditional students. Ultimately, this suggests that the high school GPA of homeschooled students should always be con- sidered in conjunction with standardized measures such as the SAT, and supports some implementation of different ad- missions policies for homeschooled and traditional students.

Acknowledgments This research was supported by a grant from the College Board to Paul R. Sackett and Nathan R. Kuncel. Paul R. Sackett serves as a consultant to the College Board. This relationship has been reviewed and managed by the Univer- sity of Minnesota in accordance with its conflict of inter- est policies. This research is derived from data provided by the College Board. Copyright 2009–2011 The College Board. www.collegeboard.com

References Cogan, M. F. (2010). Exploring academic outcomes of homeschooled

students. Journal of College Admission, 208, 18–25. Cullen, M. J., Hardison, C. M., & Sackett, P. R. (2004). Using SAT-grade

and ability-job performance relationships to test predictions derived from stereotype threat theory. Journal of Applied Psychology, 89, 220–230.

Jones, P., & Gloeckner, G. (2004a). First-year college performance: A study of home school graduates and traditional school graduates. Journal of College Admission, 183, 17–20.

Jones, P., & Gloeckner, G. (2004b). A study of admission officers’ per- ceptions of and attitudes toward homeschool students. Journal of College Admission, 185, 12–21.

Kuncel, N. R., Credé, M., & Thomas, L. L. (2005). The validity of self- reported grade point averages, class ranks, and test scores: A meta- analysis and review of the literature. Review of Educational Research, 75, 63–82.

Mattern, K. D., & Patterson, B. F. (2009). Is performance on the SAT related to college retention? (College Board Research Report No. 2009-7). New York, NY: The College Board.

Noel, A., Stark, P., & Redford, J. (2013). Parent and family involve- ment in education, from the National Household Education Surveys

Program of 2012. NCES 2013-028. Washington, DC: National Center for Education Statistics, Institute of Education Services, U.S. De- partment of Education.

Ray, B. D. (2000). Home schooling: The ameliorator of nega- tive influences on learning? Peabody Journal of Education, 75, 71–106.

Ray, B. D. (2010). Academic achievement and demographic traits of homeschool students: A nationwide study. Academic Leadership, 8, 1–31.

Rudner, L. M. (1999). Scholastic achievement and demographic charac- teristics of home school students in 1998. Education Policy Analysis Archives, 7, 1–31.

Ruger, W. P., & Sorens, J. (2013). Freedom in the 50 states: An index of personal and economic freedom. Arlington, VA: The Mercatus Center at George Mason University.

Sackett, P. R., Kuncel, N. R., Arneson, J. J., Cooper, S. R., & Waters, S. D. (2009). Does socioeconomic status explain the relationship be- tween admissions tests and post-secondary academic performance? Psychological Bulletin, 135, 1–22.

Schenker, N., & Taylor, J. M. G. (1996). Partially parametric techniques for multiple imputation. Computational Statistics and Data Analy- sis, 22, 425–446.

Copyright of Educational Measurement: Issues & Practice is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.