Expectation Formation for All? Group Differences in Student Response to Signals about Academic Performance

ABSTRACT

Theories of adolescent expectation formation hypothesize that socially and academically disadvantaged students are less responsive to signals about their academic performance when forming educational expectations. To test this hypothesis, I examine variation in student responses to new information about their academic performance by a rich set of background characteristics. I find little support for the hypothesis. In contrast, low-SES students who are high performing or come from supportive home environments appear to be the most responsive to new information about their academic performance. I discuss the implications of this finding for research on expectation formation among adolescents.

KEYWORDS:

• Expectations
• Education
• Inequality
• Students

Acknowledgments

I thank Mads Meier Jæger, Anders Holm, Richard Breen, Volker Stocké, Paul Bingley, Peter Skov, and five anonymous reviewers for their invaluable comments and suggestions.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. The importance of performance information cannot be overstated. As Nicholls (1978, 1984, 1990) has demonstrated, ability is defined among adolescents and adults in terms of individual capacity and is viewed as the most salient attributed cause of failure and success in schools.

2. Given its public nature, performance information may also lead the student’s significant others to change expectations they hold for the student, and in this way mediate the effects that demonstrated performances have on expectation formation processes (Haller and Portes 1973; Otto and Haller 1979). I return to this point in the discussion section. Moreover, sociologists distinguish between expectations and aspirations. Whereas expectations refer to a realistic future orientation, aspirations refer to an idealistic orientation (Rehberg 1967). As my study centers on how academic performance affects adolescents’ realistic appraisals of their future opportunity over time, I focus on the concept of expectations.

3. In this context, what defines an educational expectation as realistic is the degree to which the student’s appraisal of his or her chances of completing future levels of schooling is congruent with the academic situation facing him or her (Hoelter 1982). Of course, there are other ways of characterizing expectations as realistic. Because expectations also derive from the perceived costs and benefits of future schooling, incorrect or uncertain beliefs about these components can also lead to expectations unaligned with the future realities that students face (Avery and Kane 2004; Grodsky and Jones 2006). Moreover, the correspondence between educational and occupational plans is another aspect of having realistic expectations in that it reflects the student’s understanding of how much education is needed for obtaining aspired-to positions (Goyette 2008; Schneider and Stevenson 1999). Although these definitions may lead to related perspectives on the realism of students’ educational plans, I do not pursue them further here.

4. Many factors other than learning about specific academic abilities influence the college dropout decision (Tinto 1987). For example, students drop out because they lack a feeling of belonging or because they have a hard time interacting with educators (Jensen and Jetten 2015).

5. As the attrition caused by including control variables concerns only 11 individuals, it does not affect the analyses reported here.

6. Although other longitudinal surveys such as the National Educational Longitudinal Study of 1988 or the Educational Longitudinal Study of 2002 would provide an entry into studying the question I pose in this article, they do not measure previously demonstrated abilities and the quality of the students’ home environments as they grew up, nor do they provide highly reliable measures of family background. Thus, although my analytical CYA sample is selected, the possibility of using inverse probability weights––as I have information on all children born to NLSY79 mothers––to check the robustness of my results means that the CYA sample provides a good entry into testing whether expectation formation processes differ according to student group differences.

7. Given the discontinuous nature of failing, I replicated the analysis I report here assigning the value 0 to grades E or F. As the results are unaffected by this modification, I report the results using the rescaled version of the GPA measure described in the main text.

8. According to a meta-analysis (Kuncel, Credé, and Thomas 2005), the reliability of self-reported high school GPA is high (0.82 correlation with transcript-based GPA) and its predictive validity is substantial. Moreover, measurement error appears to vary as a function of minority status and academic abilities. However, such systematic misreporting has little impact on the panel data models I estimate in this article, as they control for time-invariant, individual-specific unmeasured variables that affect misreporting. However, as I explain in the main text, any residual random measurement error does lead to an attenuation of estimates, meaning that the estimates I report are lower bounds to the true estimates.

9. When including age as a main effect in the first-difference models I use in my analysis, the variable drops from the model because the difference in age from year to year is two for all students.

10. Because the model holds individual-specific factors constant, it automatically controls for variables usually associated with expectation-formation processes, including family background, aptitudes, preferences, and habitual dispositions.

11. Because the main effect of age drops out in a model of first differences, I only include the squared term.

12. Note that Table 5 does not include main effects of the background covariates. In a model of first differences, these effects are captured by the individual-specific unobserved component and therefore drop out of the model.

13. I also tested for possible nonlinearities breaking down the estimates of changes in GPA on changes in expectation on the respondents’ GPA reported in the initial wave. I found no significant differences in the estimates, suggesting that changes in GPA convey similar information irrespective of one’s position in the overall GPA distribution (thereby also ruling out any potential threshold effects). In a final set of analyses, I included a squared term of changing GPA, but this term was also insignificant, suggesting that the two change variables are linearly related.

14. The reported effects are the predicted effects, that is, those predicted on the basis of the marginal predictions of the model. These effects are also known as average marginal effects (Wooldridge 2010).

15. High-performing, low-SES students who come from supportive home environments constitute about 14 percent of the analytical sample, meaning that the large effect for this group pertains to about one seventh of all students.

Additional information

Funding

The research leading to the results presented in this article has received funding from (a) the European Research Council under the European Union’s Seventh Framework Programme ([FP/2007-2013]/ERC grant 312906) and (b) the Danish Council for Strategic Research (Grant No. DSF-09–065167).

Notes on contributors

Kristian Bernt Karlson

Kristian Bernt Karlson is Associate Professor of Sociology at the University of Copenhagen. His research interests lie within the areas of educational stratification, social mobility, and quantitative methods. Recent papers appear in Sociological Science, Social Science Research, and Sociological Methods & Research.