Development and Validation of a Short Form of the College-Going Self-Efficacy Scale

Abstract

Social Cognitive Career Theory (Lent et al., 1994) is a useful framework for understanding educational attainment and reducing educational inequities. A key construct for middle and high school students is college-going self-efficacy. The College-Going Self-Efficacy Scale (CGSES; Gibbons & Borders, 2010a) has been used to measure secondary students’ confidence in their abilities to attend and persist in post-secondary education, but with 30-items, it may be too lengthy for use with other measures in SCCT-grounded research in school settings. Using two independent samples of rural Appalachian high school students, we develop and validate the College-Going Self-Efficacy Scale-Short Form (CGSES-SF). This 14-item measure retains the full breadth of content from the original CGSES, demonstrates measurement equivalence across gender and prospective college generation status, and demonstrates good reliability and validity in these samples. Suggestions for future use of the CGSES-SF are provided.

Decades of research demonstrate that if we want to predict whether someone pursues a particular task, we must understand their confidence in their ability to be successful at that task. These self-efficacy beliefs are a critical component of Social Cognitive Career Theory (SCCT; Len et al., 1994), arguably the most influential contemporary theory for understanding career and educational pursuits. In the educational domain, college-going self-efficacy has emerged as a critical predictor of high school students’ post-secondary goals and behaviors. Given the role of post-secondary education in ameliorating educational, occupational, and financial inequalities, facilitating theory-based research on post-secondary attainment is more critical than ever (Baum & Payea, 2005). Such research depends on the existence of well-validated measures of important constructs, such as college-going self-efficacy. However, school-based research presents a number of practical limitations, not the least of which is the efficient use of students’ classroom time; for teachers and school administrators to consent to the use of instructional time for research activities, the research must not only be of practical significance to their students and school, but the time spent in research activities must also be as brief as possible. For this reason, measures must not only have excellent psychometric properties, they must also be as brief as possible. In this manuscript, we describe the development and validation of a brief version of a popular college-going self-efficacy scale.

Social Cognitive Career Theory

Social Cognitive Career Theory (SCCT; Lent et al., 1994) is a well-researched theory of career development. SCCT postulates that a combination of self-efficacy, outcome expectations, and goals interact to impact career decisions and actions. Self-efficacy, or the belief people have about their ability to successfully complete a specific task, is seen as the most powerful component of this triad (Lent et al., 2000) and is impacted by perceived barriers and supports as well as background and demographic characteristics. Importantly, self-efficacy is considered task-specific, meaning that someone can have strong beliefs about their ability to engage in one task but less secure beliefs about their ability to engage in a different task (Lent et al., 1994). Therefore, it is vital that researchers identify and measure the specific type of efficacy beliefs they are investigating.

College-Going Self-Efficacy

Given the task-specific nature of self-efficacy (Bandura, 1993), a distinct measure of college-going efficacy beliefs is needed. College-going self-efficacy refers to beliefs in one’s ability to successfully enroll and persist in post-secondary education (Gibbons & Borders, 2010a). Its focus is on beliefs about college prior to college enrollment. This focus is important because although nearly all high school students aspire to enter college (DesJardins et al., 2019), the percentage of those who actually attend is significantly lower, especially for students of color or those from low-income households (NCES, 2018). College-going self-efficacy differs from other education-related self-efficacy beliefs in important ways. For example, academic self-efficacy focuses on beliefs about ability to successfully engage in academic tasks (Uwah et al., 2008), while college self-efficacy highlights beliefs about confidence in completing college-specific tasks (Solberg et al., 1993). While these likely correlate with college-going self-efficacy, they do not focus on the beliefs middle and high school students have about their ability to successfully enter and complete college.

College-going self-efficacy is often measured in conjunction with other social-cognitive variables, including outcome expectations, perceived barriers, perceived supports, interests, and choice intentions. Together, these constructs help researchers understand the career and educational development of students based on SCCT (Lent et al., 1994). As predicted by SCCT, levels of college-going self-efficacy vary as a function of contextual factors, such as access to college-educated role models or financial barriers (e.g., Berbery & O’Brien, 2018; Rosecrance et al., 2019); moreover, students with greater college-going self-efficacy report greater belief in the value of post-secondary education (Glessner et al., 2017; Rosecrance et al., 2019) and greater interest in pursuing post-secondary education (e.g., Gonzalez et al., 2013). College-going self-efficacy has been studied in a range of samples, including prospective first-generation college students (Gibbons & Borders, 2010b), rural Appalachian high school students (Rosecrance et al., 2019), and Latinx youth from immigrant families (e.g., Gonzalez et al., 2017).

Several measures of college-related self-efficacy exist. For example, Solberg et al. (1993) developed the 20-item College Self-Efficacy Inventory (CSEI); however, this measure is designed for use by current college students to assess their self-efficacy for tasks important to success in college (e.g., talking to current professors). Our focus is on self-efficacy for tasks necessary for prospective college students who are considering going to college (e.g., choosing and paying for college). Although there is at least one unpublished measure of college-going self-efficacy that has been used in research (Jones et al., 2011, cited in Berberry & O’Brien, 2018), the most commonly used measure of college-going self-efficacy appears to the College-Going Self-Efficacy Scale (CGSES), developed by Gibbons and Borders (2010a).

The College-Going Self-Efficacy Scale (CGSES; Gibbons & Borders, 2010a) is a 30-item measure originally developed for use with middle school students. The items on the original measure were designed to assess aspects of both self-efficacy for attending college and for persisting in college once enrolled. This was based on a review of the literature on career and educational planning, which identified beliefs about both getting into and staying in college as critical components of college-going self-efficacy (see Gibbons & Borders, 2010a for details). For example, students might be confident in their ability to complete college applications and achieve the grades needed for admission, but lack confidence in their ability to handle the rigors of academic coursework or to pay for extended study. A review of the literature further identified specific aspects of self-efficacy related to these domains of attendance and persistence, and items were written to cover these aspects including finances, academic ability, family, and overall self-efficacy. In addition, items were also included related to decision-making skills for self-efficacy for attending college; for self-efficacy for persisting in college, there were items for life skills such as time management (Gibbons & Borders, 2010a). This breadth of coverage in the original CGSES is a clear strength of the measure.

Although exploratory factor analyses clearly supported a two-factor structure (attendance, persistence) underlying the 30 items, the original authors recommended use of a total score, which virtually all subsequent users have done. Strong evidence supporting the validity and reliability of these total scores has been found in a range of middle and high school samples in the U.S. For example, Cronbach’s alphas are typically above .90. In a number of studies, prospective first-generation college students (who are, on average, less likely to enroll or persist in post-secondary education; e.g., Cahalin et al., 2019) score significantly lower than prospective continuing generation peers (e.g., Gibbons & Borders, 2010b; Rosecrance et al., 2019). Moreover, CGSES scores tend to be higher following interventions designed to promote college-going (e.g., Glessner et al., 2017). The original CGSES has been widely used with diverse students from both middle and high school (e.g., Glessner et al., 2017; Gonzalez et al., 2012, 2014, 2017; Vela et al., 2017), and adapted for use with parents (Gonzalez & Villalba, 2018).

Clearly, the CGSES (Gibbons & Borders, 2010a) has facilitated important research on the career and educational pursuits of diverse students. However, the original development of the measure relied on exploratory factor analyses, and the underlying factor structure has not been confirmed in other samples. Moreover, with 30-items, the measure is not sufficiently brief to allow it to be easily incorporated with other theoretically-important measures in a single study. Especially when conducting school-based research with middle and high school students, in which disruption of class time must be minimized, using measures that are as brief as possible is essential. Our purpose was to develop a psychometrically-sound, brief version of the CGSES for use in future research. Our goal was to reduce the number of items by at least half while preserving the breadth of content domains represented in the original CGSES.

Study 1

Method

The data for this study were collected as part of a five-year, federal grant-funded career education project to promote interest in post-secondary education in general and in science, technology, engineering, math, and medical sciences (STEMM) careers, in particular, among rural Appalachian high school students. All 10th graders at our partner schools participated in a multi-week in-school curriculum designed to (a) increase college and STEMM self-efficacy and outcome expectations, (b) reduce perceived barriers and increase supports for post-secondary education, and (c) increase interest in pursuing education, particularly for STEMM careers, after high school. Gibbons et al. (2019) provide a detailed description of the program.

Participants

Usable data were collected from 1,563 high school students for the initial development of the shortened CGSES. These data were collected as part of the larger grant-funded project. For the purposes of this study, we used data collected from students across three consecutive academic years (Fall 2015, 2016, and 2017). Where we had repeat data from the same student (e.g., the same student participated in 2015 and 2016), one data point was randomly selected. Data from just over half of the sample (57.7%, n = 930) have been included in a previous publication (Rosecrance et al., 2019), with the remaining 43.4% of the sample being new.

Students were enrolled in three high schools in two different counties in East Tennessee, both of which are designated as rural Appalachian by the Appalachian Regional Commission (ARC); both counties were also designated as economically distressed by the ARC (ARC County Economic Status, FY 2017). Approximately 66% of the students in the final sample were in the 10th grade, with the remaining students being in the 11th grade. The students had an average age of 15.4 years. Consistent with the demographics of the region, the majority (88.5%) of the students self-identified as non-Hispanic White, 6.1% identified with two or more racial/ethnic identities, and less than 1% each identified as Latinx, African American/Black, Asian, Pacific Islander, or Native American. Half (50.2%) of the participants in the sample self-identified as female, 45% as male, with the remaining declining to indicate gender. Nearly a quarter (23.2%) of the students reported they were prospective first-generation college students, which we defined as students whose parents have no post-secondary education whatsoever; another 14.2% of students were unsure of their parents’ educational attainment. The remaining 62.6% of the participants reported having at least one parent who had attended at least one semester of any type of post-secondary education. Only data from students who assented to have their data used for research purposes and whose parents had provided informed consent were included.

Instrumentation

College-Going Self-Efficacy

All students in Study 1 completed the 30-item College-Going Self-Efficacy Scale (CGSES; Gibbons & Borders, 2010a). The CGSES was originally developed for use with middle school students to assess students’ beliefs in their abilities (a) to successfully navigate college-preparation tasks necessary to attend college and (b) to successfully persist in college once enrolled. Responses to each item are indicated using a 4-point Likert scale (1 = not at all sure, 2 = somewhat sure, 3 = sure, 4 = very sure), with higher scores indicating higher self-efficacy perceptions. The CGSES has been used with middle and high school students from varying backgrounds, including rural Appalachian high school students (e.g., Rosecrance et al., 2019). The total score, which provides an indication of overall strength of college-going self-efficacy beliefs, is typically used in research and demonstrated good reliability in the current sample (Cronbach’s alpha = .95). As recommended to detect careless responding in survey data (Meade & Craig, 2012), we included two instructed response items embedded within the 30 CGSES items (e.g., “Please select somewhat sure”). Only data from students who correctly answered both instructed response items were included.

College Outcome Expectations

We measured students’ beliefs about the value of pursuing post-secondary education using the 19-item College Outcomes Expectations Scale (COE; Flores et al., 2008), which was developed for use with high school students. A sample item is If I get a college education, then I will do well in life; students indicate their agreement with each statement using a 10-point Likert-type scale ranging from 1 (strongly disagree) to 10 (strongly agree). Item responses are averaged yielding total scores that may range from 1 to 10, with higher scores indicating more positive outcome expectations. The COE has been used successfully with a range of high school student samples, including rural Appalachian students (Rosecrance et al., 2019). Internal consistency in the current sample was good (alpha = .94). We included one instructed response item embedded within the 19 COE items; only data from students who correctly answered the instructed response item were included.

Demographics

All students self-identified their gender and parental educational attainment. Students were also asked to self-identify their age, grade in school, and race/ethnicity.

Procedure

The above measures were included as part of a larger battery of instruments administered biannually to students at schools participating in our grant-funded intervention. Informed consent letters describing the career education program and program evaluation are sent home to parents annually. These letters offer parents the opportunity to decline consent for their student’s program evaluation data to be used for research purposes. Students completed the battery of measures as an electronic survey administered on iPads in intact classrooms as part of their regular school day early in the fall semester; for 10th graders, data collection occurred prior to beginning the in-school career education curriculum. Measures were administered in counterbalanced order, with the demographics questions always administered last. After completion of the measures for program evaluation purposes, students were presented with an information form and given the opportunity to assent or decline assent for their data to be used for research purposes. Students are given a unique code number that allows us to match data from students over time without recording students’ names with their data.

Results

As noted above, when originally developed, the CGSES demonstrated a two-factor structure, corresponding to the attendance and persistence domains. However, in practice, CGSES total scores are almost always used, and the underlying factor structure has not been confirmed. Our first step, therefore, was to run a maximum likelihood (ML) confirmatory factor analysis (CFA) testing the hypothesized two-factor structure. To provide an independent sample with which to cross-validate future analyses, we randomly split the entire sample in half. Using one of these subsamples (n = 782), we tested a two-factor structure in Mplus Version 8 (Muthén & Muthén, 2017) in which the first 14 items loaded on the Attendance factor and the last 16 items loaded on the Persistence factor. We used the latent standardization method of model identification (see Little et al., 2007), in which all factor loadings were freely estimated and the variances of the two factors were fixed to 1.0. Although all items loaded highly on their specified factor (see Table 1), fit indices did not indicate good fit: RMSEA = 0.084 (90% c.i. 0.081, 0.087); CFI = 0.83, TLI = 0.81, chi-square = 13174.35 (435), SRMR = 0.06. We also ran a single factor CFA, but fit indices were similar (e.g., RMSEA = .089, TLI = 0.79). The questionable fit indices of the original factor structure in our sample provide further support for the need for a revised measure, and we proceeded to selecting items for a shortened form.

Table 1.

Standardized factor loadings from 2-factor CFA in Studies 1 and 2.

Original CGSES Item	Factor	Domain	Standardized Loadings
			Study 1	Study 2
2. I can get accepted to a college	Attendance	Abilities	0.77	0.81
6. I can make an educational plan that will prepare me for college	Attendance	Abilities	0.75	0.70
10. I can get good grades in my high school math classes	Attendance	Abilities	0.62	-
11. I can get good grades in my high school science classes	Attendance	Abilities	0.58	-
13. I can know enough about computers to get into college	Attendance	Abilities	0.50	-
12. I can choose the high school classes needed to get into a good college	Attendance	Decision-making	0.73	0.67
4. I can choose a good college	Attendance	Decision-making	0.72	0.70
8. I can choose college courses that best fit my interests	Attendance	Decision-making	0.62	-
7. I can make my family proud with my choices after high school	Attendance	Family	0.61	0.57
3. I can have family support for going to college	Attendance	Family	0.50	-
5. I can get a scholarship or grant for college	Attendance	Finances	0.70	0.70
1. I can find a way to pay for college	Attendance	Finances	0.62	-
9. I can pay for college even if my family cannot help me	Attendance	Finances	0.48	-
14. I can go to college after high school	Attendance	Overall	0.73	0.75
21. I could finish college and receive a college degree	Persistence	Abilities	0.82	0.81
28. I could be smart enough to finish college	Persistence	Abilities	0.78	0.79
25. I could get the education I need for my choice of career	Persistence	Abilities	0.72	-
30.I could do the classwork and homework assignments in college classes	Persistence	Abilities	0.76	-
16. I could get A’s and B’s in college	Persistence	Abilities	0.70	-
22. I could care for my family responsibilities while in college	Persistence	Family	0.62	0.49
17. I could get my family to support my wish of finishing college	Persistence	Family	0.56	-
20. I could get good enough grades to get or keep a scholarship	Persistence	Finances	0.75	0.71
15. I could pay for each year of college	Persistence	Finances	0.55	-
29. I could pick the right things to study in college	Persistence	Life Skills	0.75	0.62
23. I could set my own schedule while in college	Persistence	Life Skills	0.63	0.48
26. I could get a job after I graduate from college	Persistence	Life Skills	0.58	-
18. I could take care of myself at college	Persistence	Life Skills	0.58	-
24. I could make friends at college	Persistence	Life Skills	0.52	-
19. I could fit in at college	Persistence	Overall	0.63	0.60
27. I would like being in college	Persistence	Overall	0.60	-

Our goal in creating a short form of the CGSES was to reduce the number of items by at least half while maintaining content coverage of the five areas (financial issues, academic ability, family-related issues, life skills, and decision-making skills) covered in the two domains (attendance and persistence) on the original measure. To determine which items to retain on the brief version of the CGSES, we used the results of the 2-factor CFA reported above to identify items with the highest loadings from each of the original 10 subscale areas. On the original 30-item measure, three of the 10 areas were represented by five items, two were represented by three items, and five by two or fewer items. In order to maintain content coverage most similar to the original 30-item scale, we retained a second item from the three areas that had the most items on the original subscale. For example, on the original 30-item scale, there were more than twice as many items pertaining to beliefs about students’ academic abilities allowing them to attend college (n = 5) than items pertaining to beliefs about how financial issues would affect students’ abilities to persist in college (n = 2). We therefore retained twice as many of the academic abilities (attendance) items (n = 2) compared to the financial issues (persistence) items (n = 1). Finally, we chose to maintain a second item from the original three Decision-Making (Attendance) items because the loadings were so similar (0.73 and 0.72, compared to 0.62 for the third item); for the other three-item area (Finance: Attendance), the factor loadings more clearly supported retention of only one item (loadings were 0.70. 0.62, and 0.48; see Table 1). This process resulted in identification of 14 of the original 30 items to be retained on the CGSES short form (CGSES-SF) that reflected the same content domains in the same general proportion as covered in the original CGSES: two of the five Abilities (Attendance) items, two of the original three decision-making (Attendance) items, one of the original two Family (attendance) items, one of the original three finances (Attendance items), the original Overall (Attendance) item, two of the original five Abilities (Persistence) items, one of the original two Family (Persistence) items, one of the original two Finances (Persistence) items, two of the original five Life Skills (Persistence) items, and one of the original two Overall (Persistence) items.

Using data from our second subsample (n = 781), we conducted 1- and 2-factor CFAs in Mplus Version 8 (Muthén & Muthén, 2017), again freely estimating all factor loadings and fixing factor variances to 1.0. Fit indices for the 1-factor model were mixed, but generally did not indicate good fit: RMSEA = 0.092, 90% c.i. = 0.085, 0.099; CFI = 0.911, TLI = 0.894, SRMR = 0.044; fit indices for the 2-factor model, however, suggested an acceptable (RMSEA = .083; 90% c.i. .076, .090) to good fit (CFI = 0.929, TLI = 0.915, SRMR = 0.041) to the data. A chi-square difference test comparing the 1-factor (X² = 584.053, df = 77) to the 2-factor (X² = 707.917, df = 90) was significant p < .001, further supporting the 2-factor structure over the 1-factor. All factor loadings in the 2-factor solution were high (all standardized loadings > 0.48), and the correlation between the two factors was also quite high (r = .90).

Tests of Multi-Group Invariance

Having selected the items for the CGSES-SF and confirmed the underlying 2-factor structure in an independent sample, our next step was to conduct tests of multi-group invariance to determine whether the model fits equivalently across gender and prospective first-generation college student status. We used our combined sample from Study 1 (n = 1,563) for these analyses. For both sets of analyses, we followed recommended steps for assessing invariance (see Little et al., 2007; Widaman & Riese, 1997). We first tested a configural model in which all relevant parameters were freely estimated in each group simultaneously. This allowed us to test whether the underlying 2-factor structure provided a good fit to the data in each group (configural invariance) and provided a baseline model to which to compare subsequent models. We next tested for metric or weak factorial invariance (Little et al., 2007, p. 123) by constraining all factor loadings to be equal across groups, fixing the latent means to zero in both groups in addition to fixing the factor variances to zero to aid in model identification. Several authors (e.g., Little et al., 2007) have argued that chi-square difference tests are too strict for determining invariance at this level, so we relied on recommended indices of relative model fit to determine equivalence. We compared fit indices from the constrained model to those from the baseline (configural) model; invariance is supported if the RMSEA point estimate from the constrained model falls within the confidence interval from the unconstrained configural model and if the change in CFI across the two models is less than or equal to 0.01. If metric or weak factorial invariance was supported, we next conducted a stronger test of invariance by constraining the item intercepts to be equivalent across groups (see Little et al., 2007), freeing the latent variances in the second (and third) groups. We again used the same tests of relative fit described by Little et al. (2007), comparing the RMSEA and CFI across the weak and strong models to determine strong factorial invariance, or measurement equivalence. Finally, we tested for equivalence of latent means by constraining them to zero in all groups; here we used traditional chi-square difference tests for nested models to determine equivalence, as recommended for tests of latent constructs (see Little et al., 2007).

As shown in Table 2, the fit indices from both configural models (gender and prospective college generation status) provided a good fit to the data. Constraining the factor loadings to be equal across groups did not result in a marked decline in model fit, supporting weak factorial invariance across both genders and in all three college status groups (continuing generation, prospective first-generation, and those unsure of their parents’ educational attainment). Constraining the item intercepts to be equal across groups also did not result in a marked decline in model fit, supporting strong factorial invariance (i.e., measurement equivalence) across groups. Finally, constraining the latent means to be equal across groups did result in a significant decline in model fit, indicating that the latent means are not equivalent across groups, as would be expected.

Table 2.

Fit indices from tests of multigroup invariance in Study 1.

Groups	Model Tested	RMSEA (90% c.i.)	CFI	TLI	chi-sq (df)
Gender	Configural	0.082 (0.077, 0.087)	0.923	0.915	995.76 (164)
	Weak	0.078 (0.074, 0.083)	0.930	0.923	922.34 (164)
	Strong	0.079 (0.075, 0.084)	0.923	0.921	1010.02 (176)
	Latent means	0.080 (0.076, 0.085)	0.921	0.919	1037.67 (178)
Prospective college student generation status	Configural	0.077 (,0.072, 0.081)	0.932	0.927	1021.45 (252)
	Weak	0.078 (,0.073, 0.083)	0.930	0.924	1052.25 (252)
	Strong	0.075 (,0.071, 0.08)	0.928	0.929	1089.10 (276)
	Latent means	0.076 (,0.071, 0.08)	0.926	0.928	1115.50 (280)

^*Notes: In the weak invariance model, factor loadings were constrained to be equal; in the strong invariance model, loadings and intercepts were constrained to be equal; in the latent means model, the loadings, intercepts, and latent means were all constrained to be equal. Gender groups were young men and young women; Prospective college student generation status groups were prospective first-generation, continuing generation, and those unsure of their parents’ educational attainment.

Construct Validity of the CGSES-SF

Due to the very high correlation between the two latent factors underlying items on the CGSES-SF and consistent with the ways in which the original CGSES has been used, we calculated total CGSES-SF scores by averaging responses to the 14-items in our total sample (n = 1,563). The CGSES-SF total score exhibited strong inter-item reliability (a = .93). The 14-item short form was highly correlated (r = .98) with the original 30-item version. In addition, the correlation between scores on the COE and the original CGSES (r = .54, p < .001, n = 1,537) was not significantly different from the correlation between the COE and the CGSES-SF (r = .53, p < .001, n = 1,450; z = 0.38, p > .35). Finally, scores on both the original CGSES [F (2, 1,560) = 13.93, p < .001, _pη² =.018] and the CGSES-SF [F (2, 1483) = 11.15, p < .001, _pη² =.015] differed significantly based on prospective first-generation college student status. Although the effect size is small, the statistically significant differences on both versions suggest that the scale maintained its utility with diverse groups. Post-hoc Tukey’s tests indicated that on both the original and shortened form, prospective continuing generation students scored higher (M = 3.07 and 3.12, respectively) than both prospective first-generation college students (M = 2.92 and 2.98, respectively, ps < .01) and those unsure of their parents’ educational attainment (M = 2.91 and 2.95, respectively, ps < .01), who did not differ from each other (ps > .78).

Study 1 Discussion

The purpose of this study was to develop and validate the College-Going Self-Efficacy Scale - Short Form (CGSES-SF). We began with a validated instrument, the 30-item CGSES (Gibbons & Borders, 2010a), which has demonstrated sound psychometrics with diverse groups of middle- and high-school students in the United States. One goal was to maintain representation of the item domains but reduce the total items by at least half to increase the scale’s usefulness when measuring multiple social-cognitive variables, particularly for school-based research. Both the original 30-item and the revised 14-item short form include items that assess self-efficacy for attending college related to ability, decision-making, family, finance, overall attendance self-efficacy, as well as self-efficacy for persisting in college related to ability, life-skills, family, finance, and overall persistence self-efficacy. Thus, with fewer than half the items, the CGSES-SF assesses all components of college-going self-efficacy as articulated in the original scale.

A second goal was to ensure that the short form maintained its excellent psychometric properties. There are several sources of evidence that this goal was achieved. First, the 14-item CGSES-SF demonstrated both strong factorial invariance across all groups, indicating that “constructs can be meaningfully compared across groups because their reliable measurement properties have been defined in the same operational manner. This degree of invariance allows comparisons across groups to be made in a meaningful and truly comparable manner because it demonstrates that individuals in separate groups with the same level of the latent construct will have the same expected score on the measured indicators,” (Little et al., 2007, p. 125). We did not find that the latent means were equivalent across groups, which would be expected, based on past research that young women tend to demonstrate significantly higher college-related self-efficacy than young men (e.g., Rosecrance et al., 2019; Gore et al., 2005) and that prospective continuing generation college students demonstrate significantly higher college-going self-efficacy than prospective first-generation college students and those unsure of their parents’ educational attainment (Rosecrance et al., 2019; Gibbons & Borders, 2010). Second, the Cronbach’s alpha for the CGSES-SF total score was .93, nearly identical to the original CGSES. Third, the overall correlation coefficient between the original and SF was extremely high at .98, suggesting that the scales had a nearly overlapping relation. Taken together, these results indicate that the CGSES-SF demonstrates excellent psychometric properties in this sample that are as good or better than those of the original scale (Gibbons & Borders, 2010a).

We also found support for the construct validity of the CGSES-SF. As would be predicted by SCCT, and consistent with past research in other samples (e.g., Dickenson et al., 2017; Ojeda et al., 2011), higher college-going self-efficacy scores were associated with greater college outcomes expectations. In addition, the strength of the association between college outcome expectations and college-going self-efficacy did not differ based on which version of the CGSES was used. These results support the utility of the CGSES-SF within an SCCT-based theoretical framework.

Although these results are promising, the results of this first study are based on data from participants who completed the original 30-item measure. In light of the fact that the CGSES-SF would have slightly different instructions than the original scale, in Study 2 we administered the 14-item CGSES-SF with a new sample of high school students in an attempt to replicate the results from Study 1 and provide additional evidence for the utility of the CGSES-SF.

Study 2

Method

Participants

To validate the brief version of the scale, we obtained usable data from a second sample of 394 tenth graders in Fall 2019. These students were enrolled in four high schools located in two counties in East Tennessee. One of these high schools was also included in Study 1; the remaining three high schools were not represented in Study 1. However, the new county in which these three high schools are located is also designated as rural Appalachian by the ARC; in addition, both counties were designated as economically at-risk at the time of the data collection (ARC, 2019). Nearly one-third (30.5%) of the students in Study 2 identified as prospective first-generation college students and another 11.4% reported they were unsure of their parents’ educational attainment, with the remaining 56.6% reporting at least one parent with some college experience. The sample self-identified as 48.7% female, 45.7% male, 3% non-binary, and 2% declined to answer. Information on race/ethnicity and age was not collected (see Procedure).

Instrumentation

College-Going Self-Efficacy

Students completed the College-Going Self-Efficacy Scale-Short Form (CGSES-SF) developed in Study 1 (see Appendix). Responses to each item are indicated using a 4-point Likert scale (1 = not at all sure, 2 = somewhat sure, 3 = sure, 4 = very sure), with higher scores indicating higher self-efficacy perceptions. The original 30-item CGSES has separate instructions for the attendance and persistence items; for the 14-item CGSES-SF, students were provided with a single set of instructions (see Appendix). As recommended to detect careless responding in survey data (Meade & Craig, 2012), we included one instructed response item embedded within the CGSES-SF items (e.g., “Please select somewhat sure”). Only data from students who correctly answered the instructed response item were included. Cronbach’s alpha in the current sample was good (a = .94).

Program Evaluation Questions

Students also completed 19 program evaluation questions on the final day of the grant-funded in-school career education curriculum. These items asked about students’ overall satisfaction with the career education program and the program staff, as well as perceptions of what they had learned. Three of these evaluation questions that pertained to theoretically- and empirically-supported correlates of college-going self-efficacy were analyzed for this study. One of these questions asked for the extent to which students agreed that the program activities had helped me learn more about options after high school. Students indicated their agreement on a 4-point scale (1 = Strongly Disagree, 2 = Disagree, 3 = Agree, 4 = Strongly Agree). The other two items asked students whether their interest in attending a two-year college or a four-year college was less, the same, more after completing the career education program activities.

Demographics

Participants were asked about basic demographic information. All students self-identified their gender and parental educational attainment. All students were known to be 10th graders who participated in the in-school program.

Procedure

In Study 2, data were collected on the final day of the multi-week career education curriculum. The curriculum is delivered to 10th grade students in intact classrooms by teams of trained graduate and undergraduate students as either eight 45-minute or six 60-minute weekly sessions, depending on school preference. Students were asked to respond to several program evaluation questions followed by the CGSES-brief items using paper and pencil computer-scored bubble sheets. No identifying information was collected from students during this end-of-curriculum evaluation.

Results

We first conducted a CFA using Mplus Version 8 (Muthén & Muthén, 2017) to confirm the two-factor structure of the CGSES-SF and ensure that the short form was still reliable after modifying the instructions. This model produced good fit indices (RMSEA = .076; CFI = .933, TLI = .920) and all item loadings were greater than .40 (see the right column of Table 1). The correlation between both factors was also very high (r = .97). Reliability was measured through Cronbach’s alpha, which was .92, indicating excellent internal consistency. There were significant differences in CGSES-SF scores based on gender [t (377) = −2.76, p < .01] and prospective college-student generation status [F (2, 386) = 15.71, p < .001]. Young men scored significantly lower (M = 2.85, SD = 0.63, n = 185) than young women (M = 3.03, SD = 0.62, n = 194). Prospective continuing-generation students scored higher (M = 3.04, SD = 0.58, n = 224) than prospective first-generation college students (M = 2.80, SD = 0.65, n = 121), both of whom scored higher than students unsure of their parents’ educational attainment (M = 2.52, SD = 0.65, n = 44).

Several of our post-curriculum program evaluation questions are relevant to college-going self-efficacy beliefs and were therefore included in analyses with Sample 2. Specifically, students were asked whether their interest in attending a 2- or 4-year college was less, the same, or more after participating in our career education curriculum. There were no differences in CGSES-SF scores based on responses to the question about interest in attending a 2-year college [F (2, 390) = 1.74, p > .17], perhaps because the majority of participants (n = 261) indicated no change in their interest in attending a 2-year school (44 students reported less interest and 94 students reported more interest). However, there were significant differences in CGSES-SF scores based on responses to the question about interest in attending a four-year college [F (2, 390) = 22.55, p < .001]; students who reported less interest in attending a four-year school after our career education intervention (n = 46) reported significantly lower CGSES-SF scores (M = 2.54) than students whose interest remained unchanged (n = 191, M = 2.80); both groups reported lower CGSES-SF scores than those who reported increased interest in attending a four-year college (n = 162, M = 3.13). Similarly, the more strongly students agreed that our career education curriculum had helped them learn more about options after high school, the higher their CGSES-SF scores (r = .30, p < .001).

Study 2 Discussion

The results of Study 2 largely replicated those of Study 1, adding further evidence to support the psychometric properties and utility of the 14-item CGSES-SF in this population. When administered in its short form with a single set of instructions, Cronbach’s alpha remained high, and the underlying two-factor structure was replicated. We also found significant differences in CGSES-SF scores by gender and prospective college student generation status consistent with past research (e.g., Rosecrance et al., 2019; Gibbons & Borders, 2010a). Finally, we found a positive relation between increased interest in college-going and higher college-going self-efficacy, and between higher college-going self-efficacy and increased instrumental support (knowledge about college).

General Discussion

Taken together, the results of these studies support the utility of the 14-item CGSES-SF in this population of high school students. The CGSES-SF retains the full breadth of content from the original scale in fewer than half the items, demonstrates excellent inter-item reliability and a two-factor structure that is invariant across gender and prospective college-generation status, correlates in expected ways with college-going outcome expectations and educational aspirations, and distinguishes between students whose parents have no college experience and students whose parents do have college experience. By cutting the administration time in at least half, the CGSES-SF is likely to be far preferable to researchers and school staff than the original measure.

In addition, this study has added to the sparse knowledge base about the college-going attitudes of rural Appalachian high school students. Past research in this population (Rosecrance et al., 2019) identified students who are unsure of their parents’ educational attainment as a possible third group that should be considered when exploring differences among students based on parental education. Importantly, in this study, we replicated not only the existence of a sizable minority of students (~11% of our sample) who are unsure of their parents’ education status, but also the significant differences in college-going self-efficacy for these students, who had lower college-going self-efficacy than both prospective continuing- and first-generation college students Regardless of the reason students are unsure of their parents’ educational attainment (parents are not available due to death, separation, etc.; parents have not discussed their education; students have not paid attention), their lack of awareness of parental education level appears to signal a vulnerability to poor self-efficacy for pursuing education beyond high school. Future research that examines educational and career outcomes related to parental education and college-generation status should continue to explore the prevalence of this third group in other populations to better understand their educational and career trajectories. In addition, whereas there is burgeoning attention among practitioners to the strengths and needs of (prospective) first-generation college students, we encourage counselors to be aware of this third group and the possibility that their lack of information about parental education may signal an even greater need for access to information, role modeling, and positive messaging around post-secondary educational and career opportunities.

Limitations and Future Directions

Although we had a robust sample size across two separate samples of high school students, a limitation of this study is that all of the participants came from rural Appalachian high schools in a single state. Our participants were mostly non-Hispanic White and approximately one-third were prospective first-generation college students. It is important for future research to investigate the psychometric properties of the CGSES-SF in other samples, particularly those with more racial / ethnic diversity and from more urban settings. The original CGSES has been used most recently in a number of studies with Latinx youth (e.g., Gonzalez et al., 2017; Gonzalez et al., 2014; Vela et al., 2017); validating the CGSES-SF with similar samples may be especially useful in advancing this research agenda with Latinx youth. In addition, since the original scale was developed for use with middle school students, it is important for researchers to test the utility of the CGSES-SF with diverse samples from this younger population, as well.

Assuming that the CGSES-SF demonstrates good psychometric properties in other samples, it has the potential to facilitate research with diverse youth. As the results of the most recent Pell Institute report (Cahalan et al., 2019) demonstrate, there continue to be wide equity gaps in post-secondary education enrollment and attainment based on family income, race / ethnicity, and parental education level. Clearly, additional research is needed to understand barriers to and supports for the post-secondary aspirations and behaviors of diverse youth in order to ameliorate these equity gaps.

In addition to investigating the psychometric properties of the CGSES-SF in other samples, future research could also explore the stability of college-going self-efficacy over time to determine the test-retest reliability of the CGSES-SF. Although there is some evidence that general self-efficacy is relatively stable over brief (e.g., 4-month) and longer (2 year) periods (Lazić et al., 2018), we did not explore the test-retest stability of the CGSES-SF. SCCT posits that self-efficacy beliefs can be positively impacted through new learning experiences (Lent et al., 1994), so establishing that the overall stability of college-going self-efficacy (without interventions) would be an important step for future research.

Another limitation of this study is that we only included one theoretically-relevant construct, college outcome expectancies, limiting our ability to fully test the extent to which the new CGSES-SF functions as would be expected within the full SCCT theoretical model. The CGSES-SF also needs to be analyzed in relation to a variety of SCCT variables to ensure that the measure truly fits within the social-cognitive model. For example, researchers might explore the college-going beliefs of diverse high school students using a number of social-cognitive constructs, such as perceived barriers and supports, interests, and goal intentions. Results would help strengthen the argument for the use of the short form in place of longer scales that measure college-going beliefs.

The purpose of this study was to shorten a popular, well-validated, measure of college-going self-efficacy. The study results provide initial evidence for the use of the CGSES-SF to measure college-going self-efficacy in at least some samples of high school students. The development of this shortened scale offers a more efficient measure to study this construct, which has the potential to facilitate future research aimed at understanding college-going behaviors and ameliorating equity gaps in educational attainment.

Appendix

College-Going Self-Efficacy Scale, Short Form

Please read each of the following questions and answer them as honestly as possible. There are no right or wrong answers. When answering these questions, remember that college means any type of schooling after high school (community college, four-year university).

Using the scale below, how sure are you about being able to do the following:

• 1 = not at all sure

• 2 = somewhat sure

• 3 = sure

• 4 = very sure

  1. I can get a scholarship or grant for college
  2. I can get accepted to a college.
  3. I can make my family proud with my choices after high school.
  4. I can choose a good college.
  5. I can make an educational plan that will prepare me for college.
  6. I can choose the high school classes needed to get into a good college.
  7. I can go to college after high school.
  8. I could get good enough grades to get or keep a scholarship.
  9. I could finish college and receive a college degree.
  10. I could care for my family responsibilities while in college.
  11. I could fit in at college.
  12. I could pick the right things to study in college.
  13. I could be smart enough to finish college.
  14. I could set my own schedule while in college.