Perceived Stereotypes and Self-Regulation 1
Running head: PERCEIVED STEREOTYPES AND ACHIEVEMENT
The Influence of Perceived Stereotypes on Self-Regulation and Achievement: The Role of Socioeconomic Status
Paper submitted in partial fulfillment of requirements for EDUC 811 Structural Equation Modeling
Faye Huie
June 18, 2008
GRADUATE SCHOOL OF EDUCATION
George Mason University
Fairfax VA
Dimiter Dimitrov, Ph.D., Instructor
Summer 2008
The Influence of Perceived Stereotypes on Self-Regulation and Achievement: The Role of Socioeconomic Status
Introduction
Self-Regulation
Self-regulation refers to the degree to which students are metacognitively, motivationally, and behaviorally active participants in their own learning process (Zimmerman, 1989). According to Chong (2005), being able to effectively self-regulate oneself is positively associated with academic achievement across many cultures. Zimmerman (2000) proposed a model of self-regulation that involves three cyclical phases: forethought, performance, and self-reflection. Zimmerman (2000) suggests that self-regulation is not centered on any one of these phases, but in fact, is the integration and interaction of all the factors. All three phases of Zimmerman’s (2000) cyclic model of self-regulation are important aspects which influence one another. There is strong evidence that suggest that self-regulation is significantly and positively related to student academic achievement (Kornell & Metcalfe, 2006; Zimmerman & Kitsantas, 2005). That is, the higher levels of self-regulation that a student exhibits, the higher his/her GPA becomes. Self-efficacy, metacognitive self-regulation, and attributions will be the constructs used to examine the three phases of self-regulation.
Perceived Stereotypes
The literature surrounding stereotypes generally examines stereotypes in the theoretical framework of stereotype threat, where a certain group conforms to the negative stereotypes that are characteristic of that certain racial or gender group (Steele & Aronson, 1995). For example, in this groundbreaking study, Steele and Aronson (1995) gave the GRE test to both Black and White students. In the first condition, Steel and Aronson (1995) told the students (both black and white) that the test was meant to measure intelligence while in the second condition students were told that the test was not designed to measure intelligence. The findings reveal that in the first condition, White students significantly outperformed Black students while in the second condition, both Black and White students performed at similar levels. This suggests that stereotypes may have a serious influence on the academic performance of students across races/ethnicities. However, stereotype threat specifically explores the negative stereotypes surrounding racial groups. This leaves out a significant portion of the student population. Therefore, it is also important to study the “positive” stereotypes, such as the stereotypes surrounding Asian American students. Perceived stereotypes, then, would be the appropriate term to use and examine for the purposes of this study.
It is also important to discuss the role of socioeconomic status (SES) in the perception of stereotypes. According to Berjot and Drozda-Senkowska, (2007) stereotypes differ across SES levels, where students of a higher SES may perceive and have different stereotypes compared to those students of a higher SES. This interaction needs to be further explored, especially in the context of education in the United States where students within one classroom can vary across race and SES.
Rationale
According to Wing (2006) research has yet to thoroughly examine the role of perceived stereotypes on academic performance. Even less research has examined this phenomenon in addition to self-regulation and socioeconomic status. Considering the previously discussed nature of perceived stereotypes, it is important to understand how ones internalization of stereotypes is influenced by SES and how that relates to their self-regulatory behavior. Understanding these processes may be useful in designing intervention programs and inform teachers as well as students of what may be influencing achievement. Therefore, the following model (Figure 1) is analyzed:
Figure 1.
Model of the relationship between perceived stereotypes and achievement through academic self-regulation.
Research Questions
The following research questions in the framework of the proposed model will be analyzed:
1. What is the relationship between perceived stereotypes, self-regulation, achievement, and socioeconomic status?
2. Does socioeconomic status influence perceived stereotypes, self-regulation, and GPA?
3. Does the model fit well for both high and low SES students?
Method
Participants
A total of 570 freshmen students participated in this study. Student participants were enrolled in a university 100, psychology 100, or biology 103 course. Approximately, 63.5% of the participants were female and the median age of participants was 18.9, ranging from 16-46. The ethnic ranges are as follows: 54% White; 8% Black; 8% Hispanic; 21% Asian; 8% Other/Mixed. Of the sample, 92% was made up of first semester freshman, 6% were second semester, and 2% were sophomores. Transfer students made up 10% of the sample and 1% of the population was part-time students. The percentage of the sample who spoke English as their first language was 74% and 10% of the population were transfer students. The average student worked 11.56 (SD = 6.1) hours a week.
Measures
Academic achievement. Students’ grade point average in the introductory mathematics course at the end of the semester will be collected to examine academic performance.
The Motivated Strategies for Learning Questionnaire (MSLQ). The MSLQ is an 81-item, self-report measure that utilizes a 7-point Likert scale (1 “not at all true of me”, and 7 “very true of me”) to evaluate student motivation and application of learning strategies by college students. The MSLQ is comprised of two scales. The first scale called the Motivation Scale includes 31 items and six subscales. The self-efficacy for learning and performance subscale which includes eight items will be utilized in this study. Sample items include, “I believe I will receive an excellent grade in this class” and “I expect to do well in this class.” The second scale is called the Learning Strategy Scale which is comprised of 50 items and nine subscales. The metacognitive self-regulation subscale which includes 12 items will be utilized in this study. Sample items include, “When reading for this course, I make up questions to help focus my reading” and “If course materials are difficult to understand, I change the way I read the material.”
According to Pintrich et al (1991) the MSLQ is a reliable and valid measure for assessing motivation and self-regulation. The reliability statistic for this data suggest partially consistent responses for the self-efficacy for learning and performance scale (α = .85), the metacognitive self-regulation scale (α = .69).
Sydney Attribution Scale (SAS). The Sydney Attribution Scale (SAS) is a measure that assesses attributions of success and failure to external or internal causes developed by Marsh, Relich, Barnes, and Debus (1984). The SAS includes subscales for math, reading, and general academic subjects. The mathematics subscale will be used in this study. Specifically, the SAS provides a small description of a learning scenario and then asks a total of 11 questions. An example question is “Suppose you did badly in a math test. This is probably because: a) you always do badly in math tests (ability); b) You spent too little time studying (effort); and c) The test was hard for everyone (external),” on a 5-point Likert scale ranging from 1 “false” to 5 “true.” The mathematics SAS has a total of 6 subscales that measures effort, external, and ability attributions on success and failure on a certain task: 1) Success/Ability, 2) Success/Effort, 3) Success/External, 4) Failure/Ability, 5) Failure/Effort, and 6) Failure/External. Marsh et al. (1984) reported reasonable reliabilities across all mathematics subscales ranging from .66 to .86. For the purposes of this study, the MSLQ and the Sydney Attribution Scale will be combined into one scale and the factor structure of both scales combined will be discussed later in the results section. The reliabilities for this sample suggest consistent responses (α = .79).
Perceived stereotypes. A scale was developed to address perceived stereotypes. Example questions include: a) “People believe that my ethnic group is the smartest;” and b) “I believe that an understanding of mathematics comes naturally for my ethnic group.” Three items were designed to measure the level of perceived stereotypes. The items were measured on a 5 point Likert scale ranging from 1 “not at all true” to 5 “very true.” The reliabilities for this sample suggest weak consistency (α = .33).
Procedure
Students were administered surveys in the respective courses at the start of the semester. A research assistant administered the questionnaires to students at the beginning, no later than two weeks into the semester. Participating students were given extra credit by their professor for participating in this research.
Results
The results will be discussed and presented in the following order. First, preliminary analyses using exploratory factor analysis examining the factor structure of the self-regulatory items will be presented. Second, a series of confirmatory factor analyses as well as testing for invariance will be examined. Finally, the MIMIC model as well as the full model will be explored.
Exploratory Factor Analysis
Initially a principal components analysis was employed to identify the number of factors to extract in combination with a parallel analysis and an examination of the scree plot. These analyses were conducted on 20 self-regulation items. A total of four factors with eignenvalues ranging from 5.03 to 1.14 emerged from the principal components analysis. However, a total of 10 factors emerged as a result of the parallel analysis. A comparison of the parallel analysis and the principal components analysis revealed an intersect between the fourth and third factor. An analysis of the scree plot (Figure 2) revealed a break between the third and fourth factor as well, therefore a three factor solution was adopted. The analysis was repeated by extracting three factors using principal components factor analysis which accounted for 47.73% of the variance in the self-regulatory data.
The varimax rotation is typically used when there is reason to believe that the factors are correlated. As previously discussed, the three phases of self-regulation are considered to be interrelated. Therefore, a varimax rotation was conducted to further examine the factor structure. The results revealed a clear factor structure with clean factor loadings. Refer to Table 1 for the item coefficients and factor loadings.
Confirmatory Factor Analysis
A total of four confirmatory factor analyses were run. The first analysis examined only the measurement part of the self-regulatory items. The second analyses examined the model fit when all the factors are correlated with one another in addition to including SES as a predictor variable on all the self-regulation factors. The last two analyses focused on comparing the model fit indices between high and low SES groups to examine the stability of the model across SES groups.
CFA 1: Measurement. The fit of the three factor model was examined with a CFA with a sample of freshmen college students. The results suggest that the three factor model fit the data adequately, χ2 (167) = 594.25, CFI = .87, TLI = .85, RMSEA = .07, 95% CI = .06 - .07, SRMR = .07. All of the paths from the factor to the items were all statistically significant and are listed in Figure 2.
CFA 2: SES. The fit of the three factor model with correlations in addition to SES as a predictor variable was explored. The results suggest that there is an adequate model fit when SES is added into the model, χ2 (184) = 639.05, CFI = .86, TLI = .84, RMSEA = .07, 95% CI = .06 - .07, SRMR = .07. In terms of the correlations the only relationship that was significant was the relationship between self-efficacy and attributions. However, none of the path coefficients between SES and any of the three self-regulatory factors were significant. This suggests that SES does not play a significant role in self-regulation. This is apparent when comparing the model fit indices between the model with SES and the previous model with just the measurement aspect. The fit indices are virtually the same.
CFA 3 and 4: Comparison between low and high SES. To examine the model stability across groups, a CFA with only students classified as low SES was run first followed by another CFA with only students classified as high SES. Doing so allowed researchers to compare the model fit indices between both SES groups to see if the model was generalizable across different populations. In terms of students with low SES, the results reveal an adequate model fit, χ2 (210) = 464.56, CFI = .87, TLI = .85, RMSEA = .07, 95% CI = .06 - .08, SRMR = .08. In terms of students with high SES, the results are as follows: χ2 (167) = 358.74, CFI = .82, TLI = .80, RMSEA = .08, 95% CI = .06 - .08, SRMR = .09. These results suggest that the model is a better fit for students with low SES than students with high SES. Therefore, the generalizability of this model across different SES groups may be questionable and further research is suggested. Refer to Table 2 for a list of the fit indices.
Form and Measurement Invariance Analyses across SES Groups
Form invariance. The validity of the self-regulatory factor model and perceived stereotypes was tested separately for the two SES groups through CFA in the framework of structural equation modeling. Refer to Table 3 for the model fit indices. The results of the baseline model for students with low SES as well as high SES both suggest that form invariance is in place, χ2 (164) = 513.89, CFI = .84, TLI = .81, RMSEA = .08, 95% CI = .07 - .08, SRMR = .09; χ2 (164) = 377.46, CFI = .80, TLI = .77, RMSEA = .08, 95% CI = .07 - .08, SRMR = .09, respectively. All the correlations for both low and high SES were also statistically significant with the exception of the relationship between self-efficacy and metacognitive self-regulation, which was not significant across both SES groups.
Measurement invariance. Since the assumption of form invariance is established, it is now important to examine whether each factor has the same meaning across both SES groups, which is referred to as measurement invariance. Refer to Table 3 for the results of the invariance analyses. To examine measurement invariance, the difference between full and nested models needs to be examined. First, Model 0 is run where all intercepts and slopes are free. The chi square statistic for Model 0 was χ2 (328) = 904.07. Next, Model 1, which is a nested model within Model 0 is run. The chi square statistic for Model 1 was, χ2 (344) = 925.94. To examine if the assumption of invariance of regression slopes are met, delta chi square between Model 0 and Model 1 are examined. The chi square difference between Model 0 and Model 1 is Δχ21-0 (16) = 21.87, which is not significant. Therefore, the assumption of invariance of regression slopes across SES groups is met. To examine if the assumption of invariance of regression intercepts is also in place or not a third model, Model 2, will be analyzed and compared to Model 1. The chi square statistic for Model 2 is χ2 (360) = 946.94 and the chi square difference between Model 1 and Model 2 is Δχ22-1 (16) = 21.01, which is also not significant. Therefore, both invariance of regression slopes and regression intercepts are met.