Using Cluster Analysis to Explore Differences in the Reader Self-Perceptions of Adolescent Struggling Readers

The underlying factors that shape students’ motivation for reading include students’ goals, dispositions, and beliefs about themselves (Conradi, Jang, & McKenna, 2013). These factors are hierarchical and comprise an evolving area of inquiry in educational research. Self-perception—to some, synonymous with self-concept—is the foundational construct that relates to how individuals see themselves within a particular context; goals and dispositions are then established upon this foundation (Rahmani, 2011). Self-perception may refer to individual’s view of himself or herself as a student, or more specifically focused on his or her self-perceptions related to a particular skill (e.g., reading). Though self-perception is a “work in progress” (Conradi et al., 2013, p. 155) rather than a fixed point of inquiry, research is beginning to investigate how students see themselves generally and in specific academic areas such as reading. To a large extent, history of reinforcement and past academic achievement largely influence one’s academic self-perception (Bong & Skaalvic, 2003). Conversely, self-perception may influence academic achievement and the ways in which students attempt a given task (Montague & van Garderen, 2003).

The multifaceted construct of reader self-perception incorporates students’ beliefs about their ability to read, the processes they choose in learning to read, and whether they have positive experiences with reading (Baçtug & Celick, 2015; Henk & Melnick, 1995). For the purposes of this study, reader self-perception is broadly defined as the way students see themselves as readers (Henk, Marinak, & Melnick, 2013). Aligned to this view, reader self-perception can best be understood by examining its underlying domains (Henk & Melnick, 1995). Based on Bandura’s (1977) model of self-efficacy, the four domains of reader self-perception are (a) observational comparison (how students see themselves in relation to their peers), (b) social feedback (the way students interpret messages received from their peers and teachers), (c) progress (their view of their growth and learning), and (d) physiological state (how reading makes them feel). Together with reading behavior (Chapman & Tunmer, 2003; Melnick et al., 2009), these four self-perception domains serve as the core of the theoretical model underlying the present study (see Figure 1).

Notably, the present study’s theoretical model also posits that reader self-perception is an important element in understanding the academic progress and outcomes of students who struggle in reading, including the sometimes discrepant relationship between how ASRs perceive their reading ability and their actual performance growth and outcomes. In addition to the reciprocal relationship between both perceived and actual academic progress and reader self-perception, previous research suggests that students’ reader self-perception influences their reading behaviors (Chapman & Tunmer, 2003; Henk & Melnick, 1992). For example, self-perception may influence whether a student seeks out or avoids reading tasks (Henk & Melnick, 1992). Likewise, self-perception may impact the level of persistence and effort applied to comprehending difficult text (Henk & Melnick, 1995). Consequently, in our theoretical model we hypothesize that reader self-perception indirectly affects academic performance in reading and is therefore a major factor in understanding chronic under-performance in reading among adolescents.

The participants for the present study were sixth-grade students (n = 71) enrolled in intensive reading courses in three middle schools within a very large urban school district in the southeastern United States. Pursuant to state and district guidelines, intensive reading courses are aimed to remediate identified deficiencies in reading for ASRs who have previously failed to reach proficiency on the state standardized assessment. The sample included students with specific learning disabilities, other health impairments, and speech and language impairments as well as general education students. Table 1 provides additional demographic information. The study’s sample was largely Hispanic, reflecting a demographic characteristic consistent with the school district. A moderate percentage of the sample was currently enrolled in or exited from the district’s English for Speakers of Other Languages (ESOL) program. It is important to note that students enrolled in ESOL programs were only excluded from eligibility in the present study if they had not yet met proficiency in English according to district ESOL guidelines.

Because ASRs for this study were a subset of a larger intervention study, they adhered to an academic-based inclusionary criteria. ASRs were deemed eligible based on a distinguishable pattern of underachievement on state assessments and enrollment in a sixth-grade intensive reading course at their public middle school. Students were also administered two additional assessments used for screening by the study team. Students were included if they scored under a predetermined threshold on the letter word identification (LWI) and word attack (WA) subtests of the Woodcock-Johnson Tests of Achievement, Third Edition (WJIII; Woodcock, McGrew, & Mather, 2001).

Sample means for the age-based W scores for LWI (M = 491.91, SD = 17.19) and WA (M = 486.07, SD = 13.74) highlight students’ underperformance in reading given that the reference W score—the median scores of same-age peers in the norming sample— on the two subtests are, respectively, 515 and 512 (Jaffe, 2009). A W difference score (the difference between the sample mean and the reference W) on LWI of approximately -23 shows these students generally have limited proficiency in word reading and are likely to find grade-level word reading tasks very difficult (Jaffe, 2009). Similarly, a W difference score of nearly -26 on WA demonstrates limited proficiency in phonic decoding such that these students would likely the find grade-level decoding tasks very difficult (Jaffe, 2009). The rationale for the inclusionary criteria was to ensure that participants had significant and sustained deficits in word level reading and decoding.

Reader Self-Perception Scale (RSPS).Henk and Melnick (1995) developed the RSPS to better understand the construct of reader self-perception. Based on an exploratory factor analysis conducted by the instrument’s authors, the RSPS consists of 33 total items, beginning with one general and 32 subsequent items representing four domains on which reader self-perception is measured (observational comparison, social feedback, progress, and physiological states). The RSPS uses a 5-point Likert-type scale with a variation in number of items per factor. Notably, this results in differences in maximum possible scores which is an important consideration in analysis. Alpha reliabilities were calculated by the instrument authors to analyze the internal consistency for scores on each of the four factors, which were found to be highly reliable (32 items; α ranging from .81-.84). Table 2 provides sample items for the four factors of the RSPS as well as further reliability and descriptive statistics.

Woodcock Johnson III Tests of Achievement. In addition to the two screening assessments (LWI and WA), the passage comprehension (PC) and spelling (SP) subtests of the WJIII Tests of Achievement (Woodcock et al., 2001) were individually administered to students to assess reading achievement at the beginning of the sixth grade. The WJIII is a widely used battery of assessments (Schrank, McGrew, & Woodcock, 2001). Each assessment takes about 5 to 7 minutes to complete and was administered per the standardized administration protocol consistent. To determine students’ proficiency in identifying real words in isolation, students were administered the LWI subtest. The WA subtest of the WJIII requires students to read aloud nonsense words that are phonically correct in order to measure students’ ability to decode. These two subtests were also utilized as screening assessments. The PC subtest assesses reading comprehension by asking students to insert a missing key word that makes contextual sense in the text provided. The fourth subtest, SP, is administered orally and has students spell orally presented words. Reliability of scores on these four subtests is strong at .94, .87, .88, and .90, respectively (Schrank et al., 2001).

In accordance with university institutional review board and district research guidelines, parent consent and student assent were obtained prior to the start of data collection and analysis. Participants were administered the RSPS within the first month of the school year using paper and pencil. Members of the study team administered the assessment to students in small groups. Following administration of sample items to clarify procedures for students, survey items were read to students orally and students recorded their responses using the 5-point Likert-type scale. Student responses to each item were then coded, and total scores for each of the four domains of the RSPS were tabulated electronically. All item responses and domain scores were input into a SPSS database by a study team member. Accuracy of data input was assessed for 20% of the data, and agreement was found to be 99.8%.

Cluster analysis is an exploratory statistical technique that maximizes the differences between groups and similarities within groups (Hair, Black, Babin, Anderson, & Tatham, 2010). Groups, or clusters, are formed around predetermined variables selected by the researcher. Because the main research question aimed to differentiate between groups of ASRs based on their reader self-perception, cluster analysis provided the optimal means to accomplish this task. As detailed in this section, the investigation began first with multiple methods of cluster analysis to identify clusters followed by testing of between- and within-group differences.

Identification of Student Clusters. Specifically, to answer the first research question, summative scores on each of the four domains of the RSPS (social feedback, progress, observational comparison, and physiological states) were used as the grouping variables. However, because the measurement scales of the four domains differed, scores were standardized to z scores during the clustering process. Due to the fact that there was no a priori theory as to the number of clusters that existed within the sample, hierarchical clustering using Ward’s method to maximize significance between clusters was utilized first to determine clusters of students based on standardized domain scores on the RSPS. Both three-cluster and four-cluster solutions were initially explored based on analysis of the resulting dendrogram.

Using the three-cluster solution, K-Means clustering was employed to confirm cluster membership. Cohen’s κ was calculated to determine the level of agreement in cluster assignment between the K-Means and Hierarchical methods. There was moderate and statistically significant agreement between the clustering of the two techniques (κ = .536, p < .001) according to Altman’s (1999) benchmarks. The three cluster solution emerged as the best fit given the limited sample size and the statistically significant differences between groups on each of the four clustering variables. Two-way analysis of variance (ANOVA) indicated that the three clusters were significantly different on progress (F(2,69) = 38.954, p < .001), observational comparison (F(2,69) = 20.451, p < .001), social feedback (F(2,69) = 39.868, p < .001), and physiological states (F(2,69) = 29.075, p < .001). Results from pairwise comparison in post hoc analysis adjusted using the Tukey method indicated that Profile 1 and Profile 3, Profile 1 and Profile 2, and Profile 2 and Profile 3 were significantly different in each of the four analyses, p < .05. Table 3 presents the three clusters (herein referred to as student profiles) with a brief description of each.

Figure 2 shows the mean scores for each profile on the four domains assessed in the RSPS in addition to the sixth grade norm data provided for comparative analysis by the instrument’s developers (i.e., Henk & Mel-nick, 1995). Notably, the norm data provided by the developers were not exclusively drawn from a sample of ASRs, though these data provide an important comparative indicator when examining reader self-perception in the study sample. The ordinal nature of reader self-perception among the study sample is evident in that mean scores across all four domains show the progression from extremely low scores in Profile 3 to the average scores associated with Profile 1. In other words, there were no domains in which the mean score for Profile 1 was not the highest and, conversely, the mean score for Profile 3 was not the lowest. As noted previously, these differences were statistically significant on all four domains.

Profile Comparisons. The second research question asked the extent to which profile groups differed in terms of academic outcomes and demographic variables. Interestingly, although statistically significant differences were found for each domain of reader self-perception, the profiles did not differ significantly on mean academic outcomes. The mean differences between profiles on the four standardized academic measures (WJIII: LWI, WA, PC, and SP) were each tested using two-way ANOVA. Age-based W scores for each of the four subtests were used as they are preferable for statistical analyses and comparison among groups (Jaffe, 2009). Results indicated that there were no statistically significant differences between profiles on LWI (F(2,69) = .331, p = .720), WA (F(2,69) = .401, p = .672), PC (F(2,69) = 1.377, p = .263), or SP (F(2,69) = .588, p = .560). In terms of student demographic variables, no association was found between profile membership and gender (Χ²(2) = 3.171, p = 0.205), disability status (Χ²(2) = .134, p = 0.935), or age (F(2,69) = .987, p = .381).

The results from the present descriptive study provide further insight into the heterogeneous nature of the ASR population with regards to reader self. The three distinct cluster profiles showed that ASRs exhibit differing patterns of reader self-perception across all four domains of the RSPS irrespective of their actual academic outcomes. This is noteworthy given that the academic outcomes for these ASRs are relatively homogeneous based on the inclusion criteria for placement into the intensive reading programs and results from the baseline academic assessments administered. The fact that three distinct cluster profiles were found within such an academically similar group of ASRs shows the heterogeneity of reader self-perception among these students with and without disabilities, and may highlight opportunities for differentiated support given these differences. At the least, this finding warrants further exploration as to why these ASRs, with relatively similar reading abilities, have different perceptions of their reading ability.

Interestingly, a majority of the study’s sample (n = 44) clustered into Profile 2, characterized by a generally low rating across each of the four domains of the RSPS. However, despite low ratings, students in Profile 2 viewed their reader self-perception significantly higher than their peers in Profile 3 and significantly lower that Profile 1. It is encouraging that students in Profile 2 appear to have a more accurate view of their reading abilities than students in Profiles 1 and 3. This is the largest profile and contradicts studies that show students with disabilities have an incorrect perception of their academic abilities and tend to see themselves as academically competent (Meltzer, Katzir-Cohen, Miller, & Roditi, 2001; Meltzer, Miller, Katzir-Cohen, & Roditi, 2000; Meltzer, Roditi, Houser, & Perlman, 1998). The demographic makeup of ASRs in this profile varied. In other words, profile membership was not significantly associated with a particular school, gender, age, or disability status. Similarly, the demographic makeup of Profile 2 did not differ significantly from that of Profiles 1 or 3.

Remarkably, ASRs in Profile 1 rated themselves similar to the normative data of typically achieving students (Henk & Melnick; 1995). This is an important insight that corroborates previous research suggesting that there are large discrepancies between the way some struggling readers perceive their reading ability and their actual performance, an idea some refer to as an inflated self-concept (Kloomok & Cosden, 1994). ASRs both with and without disabilities were clustered in Profile 1, highlighting the fact that inflated self-concept is not specific to students with disabilities. This finding corroborates previous research (i.e., Klassen, 2002) on the extent to which students are able to judge their own ability and/or efficacy. Findings showed the most frequent mis-calibration among students with disabilities is in the form of overestimation. Related to this notion, Pajares (1996) questioned, “How much confidence is too much confidence, when can confidence be characterized as excessive and maladaptive in an academic enterprise, and what factors help create inaccurate self-perceptions?” (p. 565). These important questions highlight the complexity of self-perception; that is, whereas higher self-perception is associated with persistence, effort, and engagement, more is not necessarily better because accuracy of perception is also a relevant factor. Therefore, teachers should be cautioned from relying solely on self-reports of student progress and needs. Some ASRs may be unable to effectively advocate for themselves in terms of needing further assistance and support given their inability to accurately judge that need.

Profile 3 was the smallest group (n = 8) and demonstrated alarmingly low reader self-perception. Sound consideration must be given by both researchers and practitioners to the underlying causes of the negative self-perceptions exhibited by these students. Just as overestimating abilities can be damaging, underestimating abilities can feed negative feelings that contribute to decreased motivation in not only reading, but all other academic areas (Baker & Wigfield, 1999). The persistence of ongoing difficulties in reading are often reflected in self-perceptions that are intimately linked to reading experiences (Chapman & Tunmer, 2003). Students with low reader self-perception may need additional support in the affective domain in order to enhance their academic performance. Although their performance on academic outcomes may not differ significantly from their peers, the socioemotional support required for these ASRs may differ from ASRs in other profiles. Teachers should provide these students with strategies that enhance their opportunities to succeed and to feel successful (Chapman & Tunmer, 2003).

The results of the present study underscore the idea that even if ASRs are similar academically, they can be markedly different in their reader self-perception. It is important not to look at middle schools students through a homogeneous lens, but instead from a flexible lens that recognizes the complexities of adolescence and reading difficulties. Most of the ASRs in this sample have struggled with reading for 3 to 5 years without making appreciable gains. The three profiles demonstrated by these ASRs sheds light on the differences that exist within reader self-perceptions despite the commonness of limited achievement.

There are two limitations in this study that must be acknowledged when considering these findings in light of the broader population of ASRs and the inherent complexity in assessing reader self-perception. First, the study sample was relatively small and restricted to 6th graders in just three schools within one district. Second, though self-report is the most common method of measuring academic self-perception (Bong & Skaalvik, 2003), there are inherent limitations in self-report measures, particularly with student participants. The inclusion of semistructured interviews in addition to the self-report measure would have substantiated and further strengthened the findings. Therefore, whereas the present study provides an important contribution to reader self-perception research, its findings should be interpreted cautiously in light of these limitations.

It is important to further investigate the differences in reader self-perception displayed by older ASRs with additional quantitative and qualitative research endeavors. Future research might explore the extent to which RSPS data are related directly to the reading achievement outcomes of ASRs. Additional work might also investigate the extent to which the reader self-perception of ASRs changes over time by administering the RSPS longitudinally. Understanding the factors that both positively and negatively impact the way ASRs see themselves as readers is another important area for future research. Qualitative data collection and methods of analyses could provide a more nuanced account of these changes and further explain underlying mechanisms that influence reader self-perception among different groups of ASRs.

In examining the complexities underlying the process of reading, both cognitive and affective factors need to be considered as both play a significant role in reading behavior and academic outcomes. Across the three profiles identified in the present study, it is clear that many ASRs are internalizing their negative perceptions of their reading progress, the ways in which others view their reading, their own comparisons to peers, and their feelings toward reading. Whereas one cluster of ASRs showed evidence of an inflated self-concept, another demonstrated extremely low reader self-perception. This suggests that ASRs should be characterized as heterogeneous in terms of their reader self-perception. Ultimately, understanding the differing ways in which these ASRs construe themselves as readers has important implications for the nature of support provided to them in the classroom.

Acknowledgment: Preparation of this manuscript was supported by a grant from the Institute of Education Sciences (R324A140003).