Development of the Higher Education Value Inventory : Factor Structure and Score Reliability

Students bring to college a value system that affects their levels of academic achievement and persistence. The goal of this project was to develop a self-report inventory that measures the value students place on higher education. The Higher Education Value Inventory (HEVI) surveys students’ attitudes and behaviors in five domains: family expectations, scholastic focus, achievement value, general education value, and achievement obstacles. We describe the development of the HEVI and report the results of reliability studies and factor analyses. HEVI scores accounted for 35.9% of the variance in freshman grades. Implications for educational researchers and admissions officers are provided


Introduction
Researchers in higher education seek comprehensive theories of learning and instruction to guide their daily work.Our philosophy has been shaped by Piaget (1928Piaget ( /2009)), Ausubel (1968), and Novak (1977) who emphasized that what the learner already knows and values are among the most important influences on behavior.According to Rokeach (1979) and Schwartz (1992Schwartz ( , 2010)), values are transcendent core beliefs that, when activated, serve as standards that guide our actions, choices, judgments, and justifications.Values mediate students' decision-making as they pursue scholastic activities (Feather, 1982) and are related to motivation in the sense that the value one instills in a behavior functionally determines the strength with which the behavior is pursued (Rheinberg, Vollmeyer, & Rollett, 2000).
Given that students bring to college a value system that can affect their levels of engagement and persistence, students' values have been investigated extensively.Rotter (1954) and Atkinson (1957) proposed that students' expectancies for success and the inherent value they place on that success interactively mediate achievement-related behavior.Unfortunately, investigations of college academic achievement using Rotter's theoretical framework have focused almost exclusively on the expectancy component (i.e., locus of control) of the expectancy-value dichotomy (e.g., Findley & Cooper, 1983;Kalechstein & Nowicki, 1997).This is problematic because even if students are certain they can master certain tasks, they may have little incentive to do so (Eccles & Wigfield, 2002).
Grounded in the seminal work of Atkinson (1957), Eccles et al. (1983) and Eccles, Adler, and Meece (1984) developed a comprehensive model of achievement-related choices, which highlighted the multidimensional nature of task value as well as its importance as a proximal predictor of behavior.They proposed that task value is reflected in students' interest in a task (intrinsic value), its importance to them (attainment value), its utility for them (utility value), and its cost.Using confirmatory factor analysis, Eccles and Wigfield (1995) demonstrated that the three value components could be differentiated in the mathematics domain, supporting the construct validity of their model.

Rationale for the Higher Education Value Inventory (HEVI)
Value, expressed as a verb, refers to the process of appraising the worth of some commodity (Rohan, 2000).Guided by the premise that the value construct should occupy a central position in research on motivation and cognizant of the expectancyvalue framework, the purpose of our study was to develop a self-report inventory that measures the value students place on higher education.Fundamentally, we hypothesized that those students who placed greater value on higher education would evidence greater academic achievement and be more likely to matriculate through an undergraduate curriculum.A goal from the outset, therefore, was to make the HEVI unique in its assessment of the degree to which college students value their education in a general sense.An instrument that facilitated this assessment would be of considerable assistance to admissions officers interested in estimating a student's likelihood of completing an undergraduate degree.
In addition to its potential utility in admissions decisionmaking, the HEVI can provide a means to assess the relationship between higher education value and academic achievement and/or retention.Also, the inventory could facilitate examination of longitudinal changes in students' valuing of higher education over the course of their undergraduate experience.Longitudinal studies may provide insight into why certain students remain motivated to achieve and persist while others become disinterested and drop out (Neuville et al., 2007).Finally, when used alone or in conjunction with expectancy measures, the HEVI might incrementally increase prediction of undergraduate success and retention afforded by high school GPA or standardized achievement scores.Haynes, Richard, and Kubany (1995) and the Standards for Educational and Psychological Testing (American Educational Research Association, American Psychological Association, & National Council on Measurement in Education, 1999) suggest that to maximize validity, researchers should take into consideration their intended population of test takers when designing and/or selecting assessment tools.Although questionnaires have been developed to study perceived value (e.g., Eccles et al., 1983, Lupart, Cannon, & Telfer, 2004), they were unsuitable for our purposes because they were designed for younger students and focused on discrete subject areas, to include math, science, English, and computer usage.Given our focus on entering freshman students and our understanding that value priorities may differ for students of traditional and nontraditional-age (e.g., Faust & Courtenay, 2002;Hermon & Davis, 2004;Jinkens, 2009), our target population includes traditional-age freshman students, who are 18 to 24 years old.

Content Domain Specification, Identification of Facets, and Item Generation
We conceptualized the higher education values domain as encompassing those values that bear directly on a person's motivation for excelling in postsecondary education.The value domain was composed of five facets.Four of the facets were based on the work of Eccles et al. (1983Eccles et al. ( , 1984) ) and were tentatively labeled interest value, utility value, attainment value, and perceived cost.An additional facet, family expectations, was included to tap important family and social influences on an individual's appraisal of worth.
We defined interest value as the enjoyment or satisfaction derived by an individual from participating in school-related activities.Interest value is conceptually similar to intrinsic motivation; consequently, if a task has high interest value, one is intrinsically motivated to perform it (Eccles et al., 1983).We defined utility value as the value an individual places on a task or activity due to its instrumentality in achieving one or more short-and long-range goals.Although a task high in utility value may not be intrinsically interesting, it may still be valued because of its perceived effects on one's personal or professional achievement (Husman & Lens, 1999;Kauffman & Husman, 2004).We defined attainment value as the value a person places on doing well in school.The perceived value of grades has been found to be a statistically significant predictor of GPA even when standardized test scores are controlled (Pollio, Eison, & Milton, 1988;Valencia, 1997).
We defined perceived cost as the subjective estimate of loss suffered by an individual as a result of engaging in schoolrelated activities.If the perceived costs of educational attainment outweigh the perceived benefits, scholastic success may be devalued.For example, a person may devalue education if it is perceived as conflicting with or inhibiting development of identity, interpersonal relationships, immediate financial income, or other incompatible but valued activities.Finally, we defined family expectations as a student's perception of family expectations for his or her academic achievement.The family expectations facet was included as a component of the higher education values domain based on findings that values and achievement-related choices are a function of the value orientations of parents and significant others (Ferry, Fouad, & Smith, 2000;Frome & Eccles, 1998;Jacobs & Harvey, 2005;Reynolds & Burge, 2008).
Drawing from Feather (1988), Schwartz (1992), and Rohan (2000), we recognized that underlying a behavior is a trade-off between competing value priorities.We conceptualized interest value, utility value, attainment value, and family expectations as characteristics that would increase the perceived value of attending college.Perceived cost was conceptualized in terms of factors that would detract from that value.With these relationships in mind, we rationally constructed 76 Likert-type items to reflect each of the five facets.Response options ranged from 0 (Strongly disagree) to 4 (Strongly agree).

Method Participants and Procedure
The initial version of the HEVI was administered to 781 undergraduates at a large state university in the Midwest.Participants completed the informed consent, a demographic questionnaire, and the inventory.Given our educational focus, however, only those inventories completed by freshmen under the age of 25 were analyzed.Given these inclusion criteria, the final sample consisted of 305 freshmen (173 women and 132 men).Participants ranged in age from 18 to 24 years (M = 18.79,SD = .90)and included 277 Whites, 14 Blacks, 6 Asians, 3 Hispanics, and 5 students self-classified as "Other".Completion times averaged about 20 minutes.

Statistical Analyses
Items were evaluated in terms of skewness, kurtosis, and inter-item correlations.Items were rejected if the ratio of skewness and/or kurtosis to their respective standard errors (e.g., skewness index/standard error of skewness) was greater than ±2.00.Items with nonnormal distributions were eliminated, and highly intercorrelated items (r ≥ .80)were examined for redundancy of content.
Based on recommendations by Floyd and Widaman (1995), we subjected data to a principal components analysis (PCA).The number of components to be retained was determined via visual inspection of a scree plot and component eigenvalues greater than 1.Retained components were subjected to varimax rotation (Kaiser, 1958) and decisions regarding item retention were based on structure coefficients reported in the rotated component matrix.Following the recommendations of Comrey (1973) and Stevens (2002), we set the coefficient criterion to .45 to ensure that a constituent item correlated significantly with its respective component and minimally shared 20% of its variance with the component.In order to minimize component overlap, we excluded items with complex coefficients.An item with complex coefficients was defined as an item that loaded ≥.45 on one component and ≥.30 on one or more other components.Excluding these items maximized the unidimensionality and internal consistency of scores on each subscale.Regarding internal consistency, Nunnally and Bernstein (1994) advised that alpha coefficients should reach or exceed .70 during preliminary scale development.Therefore, scales with alpha coefficients of .70 were determined to show an acceptable degree of item interrelatedness.

Results
Seven of the 76 items were removed from the pool based on extreme skewness and/or kurtosis.The remaining 69 items were subjected a PCA (Cattell, 1966).We removed 42 items for the following reasons: initial communality below .20 (n = 8), item failure to obtain a coefficient of .45 or greater on any component (n = 12), or complex items (n = 22).We ran the analysis a second time using the remaining 27 items.Five principal components accounted for 49.29% of variance explained.
After inspecting the items, we assigned content-relevant labels to the components.The five components were named as follows (original theoretically-derived facet labels appear parenthetically): Scholastic Focus (9 items; interest value), General Education Value (8 items; utility value), Achievement Value (5 items; attainment value), Competing Obligations or Obstacles (3 items; personal cost) and Others' Expectations (Family Expectations; 2 items).Although the conceptual similarity between the factor labels and their respective facet may seem obvious, the revised factor labels better reflected the content of the actual items.Regarding internal consistency, coefficient alphas for scores on the Scholastic Focus, General Education Value, Achievement Value, Competing Obligations or Obstacles subscales, and Others' Expectations were α = .84,.77,.77,.63,and .62,respectively.

Study 2: Item Refinement and Final Scale Construction
In preparation for the second item tryout, we generated 26 new items to enhance the measurement of each facet, bringing the total number of items to 53.We then asked six undergraduate students to evaluate the 53 items for meaning and clarity of content.After receiving information about the purpose of the HEVI and descriptions of each facet, students independently rated the wording of each item by way of a dichotomous rating scale (i.e., clearly written items were rated 1, while items requiring revision were rated 0).In the case of a revision, students were asked to provide recommendations.Students also had the option of generating additional items for any or all of the five facets.Students judged 11 of the 53 items to be written unclearly, and those items were revised.Students provided no new item suggestions for inclusion in the pool.
Providing evidence for content validity requires multiple experts to assess the degree to which items represent the facets of interest (Haynes et al., 1995).Therefore, we selected five university professors, who specialized in testing and measurement to participate in an item-sorting task.For the sorting task, we typed each item on an index card.Labels representing each of the five facets were also typed on individual cards, with an extra card labeled "Other."Using descriptions of the five facets as guides, experts attempted to sort each item into the facet that seemed logically most appropriate.If an item did not seem to fit logically into any of the facets, we instructed them to sort it into the "Other" category.Items failing to be assigned to the same facet by at least 80% (or 4 out of 5) of the experts were eliminated from the item pool.Experts were also asked to generate additional items for any or all of the five facets.As a result of the task, we excluded 3 of the 53 items because they were sorted into multiple facets.Experts provided no recommendations for additional items.
After having students and experts review the item pool in its entirety, we administered the HEVI to a large sample of undergraduates and conducted a PCA in a manner similar to that reported in Study 1.In addition, we took a preliminary look at convergent validity by regressing HEVI subscale scores on self-reported GPAs and also tested for gender-related differences in HEVI performance.

Participants
A total of 560 students agreed to participate, but only those inventories completed by freshmen under the age of 25 were analyzed.Given these inclusion criteria, the final sample consisted of 326 freshmen (180 women and 146 men).This produced a participant-to-item ratio of 6.52, which is consistent with published recommendations (Gorsuch, 1997;Guadagnoli & Velicer 1988).Participants ranged in age from 17 to 24 years (M = 18.79,SD = .86)and included 302 Whites, 7 Blacks, 9 Asians, 3 Hispanics, and 6 students self-classified as "Other".Completion times averaged approximately 10 minutes.

Procedure and Statistical Analyses
Items were evaluated for skewness, kurtosis, and redundancy of content, and data were subjected to a PCA.We also examined the correlation between subscale scores and self-reported overall GPA and conducted a standard multiple regression analysis.Scores on the five HEVI subscales served as predictors and overall self-reported GPA was the criterion.The standard multiple regression procedure, where all predictor variables remain in the model, was selected so that the value of the entire set of HEVI predictors could be evaluated.Finally, men's and women's scale scores as well as self-reported GPAs were compared by way of two-tailed independent-sample t-tests.

Item and Principal Component Analyses
The 50 HEVI items were examined initially for normality of response distributions.One item was excluded because of extreme skewness and kurtosis.Because the purpose of this study was to reduce the number of HEVI items to a final set, PCA was considered the most appropriate exploratory method available (Floyd & Widaman, 1995).Subsequent to conducting the PCA, we conducted a common factor analysis using maximum likelihood extraction with varimax rotation to investigate whether the two procedures led to any meaningful differences in structure coefficients and variance accounted for in the respective solutions.
Based on observation of the scree plot, five factors were extracted and subjected to varimax rotation with Kaiser Normalization (Kaiser, 1958).We removed 16 items for the following reasons: initial communality below .20 (n = 3), failure for an item to achieve the coefficient criterion of .45 or greater on any of the five components (n = 8), and complex coefficients (n = 5).To produce an inventory with equal-item subscales, we retained the six items that evidenced the highest coefficients for each of the five components, resulting in a 30-item inventory.
We conducted a PCA again with these 30 items.Total variance explained was 53.12%.One item ("I love school.")failed to reach the .45coefficient criterion for Factor IV; however, the decision was made to retain this item after we found the item contributed positively to the alpha coefficient for the subscale.
We calculated scale scores by summing scores on constituent items.Overall, factor correlations were small, with only one correlation exceeding .30.Given the small intercorrelations and the greater interpretability of orthogonal rotations relative to oblique rotations, our use of varimax rotation was justified (see Nunnally & Bernstein, 1994).However, given that 5 of the 10 factor intercorrelations were statistically significant, we subjected the retained factors to oblique (promax) rotation to see if meaningful differences emerged.The underlying factor structure remained unchanged.Thus, we report results from the PCA for the final 30 items, subscale descriptive statistics, Cronbach alphas, and component correlations in Table 1.

Study 3: Reliability of HEVI Scores
The main purpose of Study 3 was to examine the test-retest reliability of scores on the HEVI.The goal of Study 3 was to administer the HEVI twice over a two-week period to a sample of undergraduate students.

Participants and Procedure
We administered the HEVI to 61 undergraduate students with the intent of having each student complete the HEVI a second time two weeks later.Of the original 61 students, 47 returned for the retest session (20 men and 27 women).Participants ranged in age from 18 to 24 years (M = 19.25,SD = 1.50) and included 42 Whites, 3 Blacks, 1 Asian, 0 Hispanics, and 1 student self-classified as "Other."Completion times averaged less than five minutes per testing session.

Discussion
Value, in its verb form, refers to the process of appraising the worth of some commodity (Rohan, 2000).The HEVI includes 30 items that measure the value college students place on higher education.Across two undergraduate samples, factor analytic results suggested a five-factor solution was the most parsimonious and theoretically-consistent accounting of variance in HEVI scores.The five facets contributing to subjective worth included Family Expectations, Scholastic Focus, Achievement Value, General Education Value, and Achievement Obstacles.An obvious shortcoming of our work was that only about 10% of the students we surveyed were nonwhite, and we focused exclusively on traditional-age freshman students.Given that traditional and nontraditional students may have different mindsets (Faust & Courtenay, 2002;Hermon & Davis, 2004;Jinkens, 2009) and that value orientations may differ as a function of ethnicity (e.g., Schwartz, 1992;Valencia, 1997), future analyses with other student samples will allow us to explore the generalizability of our findings to more diverse samples.
Cumulatively, our results suggest that the scores on the HEVI possess acceptable reliability with samples of traditional-age university freshmen.Using Cronbach alpha, all internal consistency estimates exceeded .70,suggesting acceptable item interrelatedness.Likewise, all retest reliability coefficients exceeded .70.However, scores on the Family Expectations and Achievement Obstacles subscales were more stable over a two-week period than were scores on the other scales.Item content on the Family Expectations and Achievement Obstacles subscales is interpersonal in nature, and quantifies students' perceptions of others' expectations as well as their behavioral activities that involve significant others.Scores on these scales are likely influenced by long-standing learning histories, particularly where family members are involved, and are less likely to fluctuate over short time intervals.By contrast, the Achievement Value, General Education Value, and Scholastic Focus scales include items that may be more situationally-determined and context-relevant.Scores on these scales may be sensitive to fluctuations in students' recent experiences and the demands of their academic schedules.In the second item tryout, we examined the relationship between HEVI scores and overall GPA.As hypothesized, higher scores on Family Expectations, Scholastic Focus, Achievement Value, and General Education Value were related to higher GPAs.Achievement Obstacles were inversely related to GPA because higher scores on this scale reflect fewer perceived obstacles to achievement.Therefore, students who valued earning high grades, valued becoming well educated across curricular boundaries, and believed that outside obstacles would not impede their scholastic success had higher GPAs than students who held contrasting beliefs.Likewise, HEVI total scores were positively associated with overall GPAs.These findings lend support for the construct validity of scores on the inventory.
When the five HEVI subscales were regressed on self-reported GPA, 35.9% of the variance in overall GPA was accounted for.An obvious limitation of these analyses is that the self-reported current GPAs of second-semester freshmen served as the criterion.Such reports are subject to memory biases and errors.In addition, because of sample restrictions, students' GPAs necessarily reflected only one semester's worth of course grades.Future criterion-related studies of the HEVI should use grade point averages documented by the university that include grades from the entire year.
Another limitation is that the studies did not directly assess the degree to which HEVI scores incrementally improved prediction of freshman academic performance relative to other standardized instruments of academic proficiency (e.g., SAT and ACT scores).Measures of achievement-related psychological variables, such as control expectancies and academic self-concept, have been shown to enhance predictive accuracy over and above scores on traditional measures, particularly in minority students and students with lower pre-admission criteria (Cone & Owens, 1991;Gerardi, 1990).Given the ambiguous predictive validity of traditional academic measures, particularly for lower-performing, disadvantaged, or minority students (e.g., Farver, Sedlacek, & Brooks, 1975;Kanoy, Wester, & Latta, 1989;Sedlacek, 2005), future studies will examine the degree to which HEVI scores can incrementally improve accuracy in prediction of academic achievement.
We also found evidence that women placed more value on making studying a priority, earning high grades, and becoming well educated across curricular boundaries than men.Given these differences, and the hypothesized relations between HEVI scores and GPA, we were not surprised to learn that women also had higher self-reported GPAs than men.Gender-related differences, particularly in achievement-related choices in math, pervade the task value literature (e.g., Eccles, 1985;Lupart et al., 2004).Eccles et al. (1983Eccles et al. ( , 1984) ) proposed that task value is a function of a person's needs, goals, and self-perceptions.Therefore, one possibility is that scores on the HEVI may be affected by gender-role socialization and personal goals.As examples, men may exhibit single-minded devotion to a single goal, such as career success (Eccles, Barber, & Jozefowicz, 1999), whereas women may focus on career training as well as general intellectual improvement (Green & Hill, 2003;Schab, 1974).As such, women, who have multiple roles and multiple goals (Eccles et al., 1999), may place a higher value on demonstrating competence in several academic domains simultaneously than men.

Conclusion
According to Rokeach (1973), "the value concept, more than any other, should occupy a central position... able to unify the apparently diverse interests of all the sciences concerned with human behavior" (p.3).Although Rokeach made this statement nearly 40 years ago, empirical investigations of the values construct are relatively limited at the university level and ample opportunities for inspection and discovery remain.We presented preliminary evidence for the factorial validity, concurrent validity, and reliability of scores on the HEVI, an inventory designed to measure the value college students place on higher education.At the present time, however, we have not collected data regarding the discriminant validity of HEVI scores nor have we examined the relationship between HEVI scores and attrition.Future investigations will aim to provide further evidence of the construct validity of HEVI scores and will assess factor invariance across gender and culturally diverse groups.

Table 1 .
Structure coefficients for the PCA with varimax rotation of the 30 final HEVI items.
N ote: Lowercase "r" indicates reverse-scored item.Primary structure coefficients are in boldface.