Lessons and directions for person-centered research in strategic processing
While many researchers might see person-centered analyses as one more analytical approach that might be applied to a data set, this kind of thinking vastly undervalues its perspective and misjudges the preparation necessary to ensure meaningful results. As an initial guide for strategic processing researchers interested in this approach, three straightforward steps might support appropriate and meaningful use. The first is framing the proposed research appropriately. This includes the theories, resulting selection of constructs and relevant research questions. It is generally the case that person-centered methods are applied when the theory under consideration is well substantiated, with considerable variable-centered research evidence accumulated. This foundation can then be used to extrapolate and develop appropriate hypotheses. Much like variables arising from factor analyses, which cannot be effectively described by theory, subgroups which are unexplainable are unlikely to be replicable and not useful research outcomes. Robust, well-tested theory and constructs that have established construct validity are as essential for person-centered as they are for variable-centered research and must be confirmed first. And as a second step, the potential shape as well as level (Morin & Marsh, 2015) of the hypotheses the chosen theory can generate for different subgroups. Shape refers to qualitative differences across subgroup profiles, in contrast to quantitative difference only (level) (Morin et al., 2017). Some theories, especially those that are narrow and function across a simple continuum of “a lot” = good, and “scant” = bad are not the best theoretical frames for finding meaningful subgroups. Approaches to learning (Marton, Hounsell, & Entwistle, 1984) are an example of this, as surface and deep approaches have been organized as existing on a continuum (i.e., surface to deep; Kember, Biggs, & Leung, 2004; Kember & Leung, 1998). Another theory that describes a broader range of cognitive processing strategies (Learning Patterns; Vermunt, 1994; Vermunt & Donche, 2017) acknowledges the reality of concurrent strategy use that results in subgroups with profiles that demonstrate differences in shape as well as level (Vermetten, Lodewijks, & Vermunt, 1999). A third point for consideration is sample size. The sample size necessary for the effective recovery of subgroups using LPAs varies depending on a number of criteria ranging from the number of indicators, the true number of classes and the inter-class distance (for an in-depth discussion see Tein, Coxe, & Cham, 2013). Increases in any of these can increase the sample size necessary for obtaining a meaningful set of subgroups.
A Review of Person-centered Research in Strategic Processing
Since the late 1990s several strategic processing researchers have applied personcentered analyses within their research programs. Earlier research classically applied clustering techniques, often focusing on an exploration of how a single theory might be expressed across a sample of students. More recent research has focused on complementing or integrating theoretical perspectives and modelling transitions between subgroups across time.
Beginning with the early person-centered strategic processing studies, applications have aimed to examine the convergence of theoretically related constructs. Clustering research by Prosser, Ramsden, Trigwell, and Martin (2003) brought together aspects of the learning environment (course experience questionnaire; Elphinstone, 1989; Ramsden, 1991), and approaches to learning (Biggs, 1987) are an example of applying clustering to a large sample with two indicators (surface and deep approaches to learning; Marton et al., 1984). This analysis resulted in high quality and low quality learning clusters, i.e., clusters based on level differences. Consistent with this study’s general approach, but applying latent profile analysis, Abar and Loken (2010) also examined longstanding variable-centered theory and empirical evidence in a student population. A collection of indicators from well-established surveys (Midgley et al., 2000; Pintrich, Smith, Garcia, & McKeachie, 1991) with relevance to self-regulation were used to cluster a sample of high school students. Three subgroups were identified as High, Low and Average SRL subgroups: subgroups again based on level differences. Both studies were modest, largely exploratory studies, essentially aiming to see what their respective theories might look like across a population.
In an early longitudinal person-centered study, Alexander and Murphy (1998) brought together knowledge, interest and strategy measures to cluster students’ prepost experience during a psychology course. One of the aims of this approach was to examine changes in subgroup number and profile shape, while also connecting profiled members with achievement in the course. An increase in the number subgroups (three to four) was found, pointing to changes across the courses. Furthermore, the subgroup that demonstrated a convergence of high interest and strategic processing along with moderate levels of domain knowledge presented the higher achievement. Robust theory and their pre-post approach yielded information about how the theory applied to specific population and subgroup differentiation by shape and level. By connecting the clusters to students’ course-end examinations, specific achievement related findings were also made possible.
During the 2010s there has been an explosion of person-centered studies across educational psychology, but scant application to the field of strategic processing. Recent research both within and outside the field of strategic processing will be drawn on here to provide potential direction for future research. There is a growing recognition that if our learning strategy research is to continue to progress, it needs to increase its connections to other aspects of the learning process. Seeking to draw together students’ goals for learning (instrumental; Fryer, 2013; Simons, Dewitte, & Lens, 2004), amotivation (Legault, Green-Demers, & Pelletier, 2006) and their strategic processing (approaches to learning; Trigwell & Ashwin, 2006), an LPA was undertaken with these constructs along with annualized GPA as a covariate. The results yielded subgroups that were differentiated by level and shape, suggesting strong connections between distal goals and depth of processing. The person-centered outcomes also highlighted the reality that individuals pursue multiple goals when learning and that it is the balance of these goals (distal vs. proximal) that marks students applying a deep approach to learning. These results contrasted with previous variable-centered (longitudinal SEM) findings in the same research context (Fryer, Ginns, & Walker, 2014), which only highlighted the role of one type of distal goal and failed to demonstrate the fact that students pursue multiple goals simultaneously and that the balance of these goal pursuits, as much as any single goal, might be playing a role in students’ strategic processing.
Educational psychology research is primarily concerned with the nature of student learning and how best to support students in being more effective in their learning. From a quantitative perspective, person-centered research is the more effective (relative to variable-centered) means of getting at these experiences, both cross-sectionally and longitudinally. One area of concern receiving considerable attention recently is the student experience across the transition from secondary to higher education (Kyndt, Donche, Trigwell, & Lindblom-Ylanne, 2017) - an area of longstanding concern in Japan (Cummings, 1984). As such, an extension to LPA, Latent Profile Transfer Analysis (LPTA) was employed. This analysis integrates autoregressive modelling to test longitudinal subgroup membership (Nylund, Asparoutiov, & Muthen, 2007). LPTA provides profile information at multiple time points and indicates the stability/variability of these subgroups and the transitions of each student between measurement points. LPTA is therefore a means of establishing how students might change categorically over time, by assessing their movement between established subgroups across as many as five consecutive time points (for an extensive review see Nylund, 2007).
LPTA was applied to a sample of students prior to and after their first year at a Japanese university (Fryer, 2017b). The aim was to reveal latent subgroups based on their learning experiences and strategic processing. Analyses suggested three subgroups were present. Consistent with Prosser et al., (2003) and the broad continuum these experiences and processes were distributed on, only level differentiated the subgroups from one another. The transition analyses that examined how and whether students meaningfully changed across the year-long experience revealed a pattern of students’ transition towards subgroups reporting lower quality learning experiences and less overall strategic use. Furthermore, students in the “Low” subgroup presented more dependence on surface strategies between the two time points suggested, a “poor get poorer” downward spiral. This type of analysis provides a unique perspective on students’ longitudinal experiences by indicating categorical change (rather than incremental) and highlighting how learning experiences might play a role in the development of these distinct subgroups over time.
A second longitudinal example from recent research touches on an avenue for strategic processing research which person-centered approaches have a potential to contribute to: theoretical convergence. This is a critical issue for strategies researchers (Coertjens, 2018; Dinsmore & Fryer, 2018; Fryer, 2017a) that few research programs meaningfully address. Person-centered analyses are well positioned to explain how different learning strategies combine and develop over time within different subgroups. Using LPTA, Fryer and Vermunt (2018) examined how the regulation of students’ learning and their strategic processing of learning materials converge across a year of study at university. Consistent with Alexander and Murphy’s early work (Alexander & Murphy, 1998), the longitudinal nature of the data offered an opportunity to look at the stability of subgroups’ profiles across time. Rather than two snapshots provided by clustering students in two separate analyses (or as in an LPA at each time point), LPTA determines the subgroups at both time points simultaneously (reducing Type 1
errors) while also indicating students’ potential movement between subgroups. Results from this analysis presented four subgroups at both times, with the shape of profiles remaining relatively consistent across time. Transition findings, for example, pointed towards the low quality (i.e., low self-regulation and deep approach to learning relative to higher lack of regulation and surface approaches to learning) strategy group as both growing in size and demonstrating the highest level of stability (i.e., lowest level of students transitioning out and in, relative to “remainers”). The least stable subgroup consisted of students reporting the lowest overall strategy use, with over half of them moving on to subgroups reporting more strategy use. While results suggested paired self-regulation and deep approaches, consistent with theory (Vermunt, 1987), it also highlighted that fact many students reported using relatively consistent amounts of all strategies. This kind of analytical approach is therefore well-disposed to demonstrating where theory seems to work (i.e., for some students) and other areas it might not meaningfully apply to. It is also an excellent means of examining at-risk subgroups that might not be getting the support they need to develop the kinds of strategies necessary to be successful.
To provide additional examples of how person-centered analyses can support our understanding of strategic processing it is necessary to draw on research from the related area of motivation and beliefs for learning. Considerable research has investigated the development of strategic processing skills across time (e.g., Alexander & Murphy, 1998; Gordon & Debus, 2002), but variable-centered research relies on mean-level difference comparisons at worst and latent growth analysis at best. LPTA can provide a clearer, categorical outcome that establishes how many students actually developed (or failed to), while indicating where they started. Oga-Baldwin and Fryer (2018) is an example of a study assessing whether a national initiative - supporting students’ intrinsic motivation for learning a foreign language - was actually having any meaningful traction. To demonstrate “traction”, a mean increase, regardless of its effect size, is not sufficient evidence. Through LPTA, this study could establish where students started and where they ended up after two years of elementary school. Results from the study demonstrated a clear pattern of students’ annual movement towards subgroups with more adaptive motivational profiles (represented by five motivations and beliefs for learning - including intrinsic motivation), thereby suggesting “traction” for the national policy.
Person-centered studies that span multiple learning contexts can indicate how students vary in different learning environments (e.g., Alexander, Jetton, & Kulikowich, 1995). These kinds of studies can also provide external validity for theories, expanding their relevance, or, just as importantly, constraining it. Person-centered studies that bridge domains of learning also have the potential to challenge longstanding beliefs about the differences in studying different subjects. In a recent study with secondary students, Oga-Baldwin and Fryer (2018) have applied this research design to testing the age-old premise that students’ motives for studying a new language differ substantially from their reasons for studying their native language. Results from this study demonstrated strong overlap for subgroups - their membership, level and shape - for the same students’ motives in these two subjects. Person-centered analysis is the only quantitative means of making a comparison of multiple theoretical components and challenging assumptions about students’ learning experience in different contexts.
There are a few examples of how variable (longitudinal latent structural equation modeling) and then person-centered (Latent Profile Transition Analyses) research can be undertaken synergistically within a single study (Fryer & Ainley, 2019; Fryer & Bovee, 2018). Fryer and Bovee initially used SEM to demonstrate the critical longitudinal connections between teacher support and prior computer skills for online learning (controlling for prior knowledge). The examination of transition of students (between the three subgroups revealed at each time point), however, revealed the critical role of students’ prior competence and teacher support together for student learning experiences and outcomes, i.e., teacher support was critical but only when students had sufficient prior knowledge. This pair of variable and person-centered approaches can offer a deeper analysis, allowing researchers to step beyond the mean and assess whether there are vulnerable subpopulations in need of additional support.
Person-centered approaches offer researchers a number of choices. Some examples from research in strategic processing and neighboring fields have been presented. It is worth noting that the most common approaches are those taken by early research in this area: examining samples for theoretically consistent groups. The majority of this research employs variance testing (e.g., MANOVAs) and difference testing (e.g., ANOVAs) to assess the validity of subgroups revealed and the impact of group membership on covariates or distal outcomes like achievement.
Mixture Modeling (e.g., Hagenaars & McCutcheon, 2002) incorporates Latent Class Analysis (LCA; Latent Profile Analysis is a type of LCA with continuous indicators) along with a wide range of analytical techniques. Mixture Modeling offers researchers a powerful means of both establishing the validity of subgroups, as well as ascertaining the reasons for and outcomes of membership. For readers familiar with Mplus (Muthen & Muthen, 1998-2015), Morin (2016) has organized a supplementary instructional document which details in stepwise fashion, increasingly sophisticated levels of person-centered tests. These analyses offer researchers the opportunity to integrate multiple group solutions, mixture regression and growth mixture modeling (linear and non-linear) into person-centered analyses. Through the analyses outlined by Morin and colleagues, subgroups can also be tested for similarity across a broad range of factors, e.g., Configural Similarity, Structural Similarity, Dispersion Similarity and Distribution Similarity. While measurement issues like those addressed by these tests are still just on the horizon for person-centered research, they are certain to become standard practice in the near future.