Predicting Success for Virtual School Students: Putting Research-Based Models into Practice

Virtual schooling has the potential to offer K-12 students increased access to educational opportunities not available locally, but comparatively high dropout rates continue to be a problem, especially for the underserved students most in need of these opportunities. Creating and using prediction models to identify at-risk virtual learners, long a popular topic in distance education, is assuming increasing urgency in virtual schooling. Though many studies have tested the contributions of various factors to online success, this article emphasizes that prediction models must be developed and used in ways that yield findings to support student success rather than prevent students from enrolling. One such model is offered here. After a description of data collection and statistical processes used to derive the model, procedures are outlined for how to implement it in virtual school settings in ways that increase both the accuracy and utility of predictions. Introduction: A Rationale for Predicting Performance The Quest for a Success Prediction Model: Popular and Problematic The early promise of virtual schooling (school courses offered through distance technologies) was to provide access to high-quality educational opportunities for students who traditionally lack such opportunities: rural, underserved, and at-risk populations (Davis & Roblyer, 2005). However, there are indications that virtual schooling opportunities tend to benefit primarily already-advantaged learners (Roblyer & Marshall, 2003; Roblyer, et. al, 2008). Growing numbers of middle and high school students are taking virtual courses (Watson & Ryan, 2007), but compared to traditional in-person courses, virtual school courses almost always reflect higher student failure and dropout rates (Kozma & Zucker, 2003), a finding consistent with those from postsecondary populations (Bernard et al., 2004). While virtual schools are being founded to provide access to educational opportunities not locally available (Roblyer, Freeman, Mason, & Schneidmiller, 2007; Watson & Ryan, 2007) Hartley and Bendixen (2001) point out that that educational access does not equate to educational opportunity. For example, in at least one large virtual school, minority students tended to enroll less but drop out more (Florida TaxWatch, 2007). Hartley and Bendixen are among those who feel that certain cognitive characteristics (e.g., lack of selfregulation) could predetermine low performance in distance environments. Thus, the desire to identify and, if possible, support at-risk virtual learners in ways that increase chances for their success has generated considerable interest among virtual schools. However, the quest for a prediction model to identify at-risk virtual learners has proven problematic. Studies have hypothesized and identified a variety of student and environmental characteristics that contribute to success, but no one set of characteristics has emerged as dominant and none of the studies that offered a model has offered an efficient way to apply its findings in practice. A recent study reported in Roblyer, Davis, Mills, Marshall, and Pape, (2008) has produced a prediction model that helps explain variations in previous findings and lends itself to practical implementation. While the Roblyer, et al. article emphasized how and why the model was generated, the information reported here focuses on how the model they created could be used in practice to help identify students who may need additional assistance in order to be successful in virtual environments. Background: Studies of Contributors to Persistence in Distance Courses Though it has long been acknowledged that distance courses have the potential to offer educational opportunities of equivalent quality to in-person courses (the so-called "no significant differences phenomenon" reported by Russell (2001) and others), research findings also consistently confirm that failure and dropout rates are higher in distance environments (Bernard & Amundsen 1989; Cyrs 1997; Dille & Mezack 1991). As the problem of low retention rate in distance environments became apparent over the years, a variety of studies emerged to explore the causes (see Table 1 at end of article). Lines of research on characteristics of successful learners took several forms, including: identifying demographic and psychological characteristics that were predictors of success, creating and testing retention models based largely on learner characteristics, and developing instruments to identify at-risk distance learners. Other researchers hypothesized factors other than learner characteristics that were also important contributors to success. Smith and Dillon (1999) and Chyung (2001) felt that the way distance learning delivery systems were designed and configured could explain much of the variance in comparisons of performance in distance and traditional environments. Of particular interest were studies that found that providing better social and emotional support to reduce what Woolcott (1996) referred to as "psychological distance" could reduce attrition. Frankola (2001), Willgin and Johnson (2004), Bocchi, Eastman, and Swift (2004) and Santaovec (2004) all found that factors with most influence on decisions to drop out of distance courses had to do with “issues of isolation, disconnectedness, and technological problems” (Frankola, 2001, p. 53). They believed that, if course environments were designed to increase facilitation, communication, and feelings of connectedness to a learning community, dropout rate would decrease. However, in light of the fact that so many students are successful in the same courses in which others drop out, it seems likely that some students require even more facilitation and monitoring than others in virtual courses. A Rationale for Studying Success Prediction The rationale underlying studies of both learner and learning environment characteristics is that effective strategies are needed to help organizations increase student success and reduce dropout rates in distance courses. Since it makes intuitive sense that a combination of these factors contribute to success, a model is needed that has two essential qualities: (1) it is based on the combined factors that research indicates could contribute to predicting success, and (2) it would itself to efficient measurement and implementation in virtual school settings. Creating and using such a model is assuming increasing urgency in virtual schooling. Recent reports confirm that it has become one of the fastest-growing international trends in education today (National Forum, 2006; Setzer, Lewis, & Green, 2005; Zandberg & Lewis, 2008). States are increasingly looking to online strategies and resources to provide students with courses not available locally and to allow accelerated or remedial alternatives for students who need them. The recent National Center for Education Statistics' report (Zandberg & Lewis, 2008) found that in 2004–05, there were an estimated 506,950 technology-based distance education course enrollments in public school districts. "Ten percent of all public schools nationwide had students enrolled in technology-based distance education courses during 2004–05, an increase from 9 percent in 2002–03" (p. iv). Based on these findings, the report observed that "technology-based distance education has established its presence in the nation’s public schools" (p. ix). Despite anticipated and real benefits of virtual schooling, it is not unusual for virtual schools to report a dropout rate of from 40-70% (Oblender, 2002; State of Colorado, 2006), though some established schools claim a dropout rate from 10-20%. In the case of one program, it was found that virtual students were forced to repeat grades at a rate four times that of students statewide (Rouse, 2005). Some virtual school programs have addressed high dropout and failure rates through front-end means such selecting and admitting students on the basis of identified criteria, instituting required pre-course orientations, and increasing the length of the drop-add period to 28 or more days. Some schools have also increased levels of students monitoring and facilitation. Virtual schools report no data on the success of the latter strategies, but informal reports indicated they have met with at least some success (Pape, Revenaugh, Watson, & Wicks, 2006). As the virtual schooling movement gains momentum and states increase their virtual schooling offerings, virtual school populations will increase in both size and diversity of students. Equal opportunity and equity requirements will make it impossible for most schools to select only certain students to take online courses, so the emphasis will be on strategies to support students in ways that help promote retention and success in virtual courses. However, using such models in typical virtual school settings presents formidable obstacles. Not only must such a model offer valid and reliable predictors of success, procedures to implement it must be efficient and lend themselves to quick identification of and interventions for at-risk students. Its use should identify students for specific kinds of extra assistance, but not emphasize factors that would be difficult to address or take so long to employ that, in essence, it prevents at-risk students from enrolling, rather than promoting their success once they do sign up. Thus, creating prediction models presents challenges from both a theoretical research standpoint, as well as from practical and logistical ones. The next part of this article describes a model created by Roblyer, et al. (2008) that could help meet these challenges. After a description of the data collection and statistical procedures Roblyer, et al. used to derive the model, procedures will be described for how to implement it in virtual school settings. Methodology and Findings from the Roblyer, et al. Success Prediction Study As previously reported