Interest in school and educational success in Portugal

ABSTRACT

A large body of work in educational economics displays the tenuous relationship between school inputs and cognitive achievement. Among others, the inability to establish a strong link has been attributed to the difficulty of controlling for attributes such as ability, motivation, and interest. Against this background, and inspired by work in economics and educational psychology, the authors examine whether a child's interest in school has a bearing on educational success. The panel data estimates, based on Portuguese data, show that after controlling for socioeconomic background, school inputs, and time-invariant unobservable traits, children with high levels of interest are 6–10 percentage points less likely to fail as compared with children with low and medium levels of interest.

KEYWORDS:

• Educational outcomes
• interest in school
• noncognitive skills
• Portugal
• school success

Acknowledgments

The authors thank Michael Grimm, Mansoob Murshed, Remco Oostendorp, Antonio Di Paolo, and Erik Plug for comments. Useful comments from participants at the First Lisbon Research Workshop on Economics and Econometrics of Education, a seminar at the Centre for European Economic Research (ZEW), Mannheim, and the Second International Workshop on Applied Economics of Education, Catanzaro, Italy, are also gratefully acknowledged.

Notes

1. Educational expenditure as a percent of GDP rose from less than 1 percent in the mid-70s to about 3.5 percent in the mid-80s and to about 4.5 percent in the mid-90s. This followed systematic improvements since the late 1960s (Goulart & Bedi, 2017).

2. EU15 refers to the 15 initial member states of the European Union. The New Member States refers to recent entrants from Eastern Europe. The “early school leavers rate” is defined as the share of the population aged 18–24 with less than upper secondary education and who are no longer in education or training. The figure for Portugal is 41.1 percent, it is 18.1 percent for the EU15 and 7.5 percent for NMS (see OECD, 2003 for details).

3. The three constructs, motivation, interest and self-perceptions of ability are closely linked and in section II we discuss motivation and interest, which are the focus of this paper, in some more detail. To add to the terminological complexity, the terms self-belief, self-concept and self-perceptions are terms used by different authors to communicate similar ideas.

4. The Big Five factors are often condensed in the form of an acronym – OCEAN. These are Openness to experience, Conscientiousness, Extraversion, Agreeableness, Neuroticism. An alternative is the Gigantic Three dimensions of personality which includes Neuroticism, Extraversion and Psychoticism.

5. Our examination is similar in spirit to the work by Borghans, Meijers, and ter Weel (2008) who used an experimental approach to examine the correlation/marginal effect of various personality traits on cognitive test scores of adults. Their work was based on 128 students at Maastricht University and showed that performance-motivation increases the probability of giving a correct answer by 7 to 10 percentage points. Other papers include Wolfe and Johnson's (1995) work on a sample of 201 psychology students which revealed that self-control/self-discipline explains 9 percent of the variance in GPA as opposed to SAT scores which explained 5 percent. Chamorro-Premuzic and Furnham (2003) worked with two samples of (N = 70–75) British University students and found that the Big Five and the Gigantic Three set of personality traits explained 10 to 17 percent of the variation in test scores. Duckworth and Seligman's (2005) work with samples of 140 to 164 eighth-grade students demonstrated that there was a stronger correlation between self-discipline and final GPA as compared to the correlation between IQ and GPA. Similar to the work of educational psychologists, reviewed in section II, typical papers in this area tend to use small samples, rely on cross-section data and do not control for parental or family background characteristics.

6. According to Holmlund and Silva (2014) most remedial interventions tend to focus on improving educational outcomes by focusing on measures such as additional instruction time, smaller class sizes and the like, however, there are few programmes directed at enhancing non-cognitive attributes as a way of enhancing educational outcomes. Heckman (2000) reviews some of these programmes such as the ‘Big Brothers Big Sisters’ community and school-based mentoring programmes which focus on pairing children ages 6 through 18 with role models in one-to-one relationships as a way to improve student motivation and awareness of education.

7. For instance, it may be readily argued that educational interest and motivation are a consequence of educational success and do not cause educational success.

8. For example, Fuller and Clarke (1994) report that only 9 out of 26 primary-school studies find a significant impact of class size on achievement in developing countries. Harbison and Hanushek (1992) examine the effect of teacher-pupil ratios and find that in 16 out of 30 papers with statistically significant effects, eight studies yield positive effects and eight studies yield negative effects. Hanushek (2003) provides an updated discussion which displays a similar pattern.

9. Extrinsic motivation refers to a situation where activities ‘are performed not out of interest but because they are believed to be instrumental to some separable consequence’ (Deci, Vallerand, Pelletier, & Ryan, 1991:328). The intrinsic-extrinsic dichotomy may suggest that intrinsic motivation is immutable. However, this is unlikely and as argued by Deci et al. (1991), external incentives may drive actions initially but it is possible that over time internalization occurs and leads to a breakdown of the dichotomy. Ryan and Stiller (1991) also argue against this dichotomy and Connell and Wellborn (1991) do not draw a distinction between interest and motivation.

10. Satisfaction with school includes items such as “The school and I are like: Good friends; Friends; Distant relatives; Strangers; Enemies”, “I like school very much: True or False” and “Most of the time I do not want to go to school: False or True”. Commitment to class work includes, “Work in class is just busy work and a waste of time: Always to Never”, “In class, I often count the minutes till it ends: False or True” and “The things I get to work on in most of my classes are: Great stuff – really interesting to me to Trash – a total loss to me”.

11. The question used to capture interest in school is, ‘How is or was your relation with school?’ The options are (i) Good, very interested (ii) Fair (iii) Not attractive (iv) Bad. Since very few children opted for options (iii) and (iv), we reclassified the information into three categories, High level of interest (good, very interested), medium level of interest (fair) and a low level of interest (not attractive, bad). In the 1998 survey only children were asked to respond to this question while in the 2001 survey parents and children responded to this question.

12. The school success specification is estimated on the basis of information for all the children in the sample and is not a select sample of children who are still enrolled in school. The school enrolment rate in both years for which we have data is 97 to 98 percent and information on grade repetition is available for all children regardless of whether they are currently enrolled in school or not. School enrolment and regular school attendance are almost universal and hence the appropriate concern is the educational performance of children.

13. Estimates based on (3) and (4) will only be consistent if the assumption of strict exogeneity of the regressors, conditional on the unobserved fixed effect is satisfied. This is unlikely to hold in the current case as interest and achievement are likely to be simultaneously determined. In the current application, as long as the error terms are not correlated, it is probably reasonable to assume that past values of the interest variable are unlikely to be correlated with the contemporaneous error term, that is With this assumption it should be possible to obtain consistent estimates using I_t-2 as an instrument for or I_it-1. However, we only have information for two periods. Alternatively, one could obtain consistent estimates if we could find additional suitable instruments for the formation of interest. Such instruments are not readily available and in the absence of these we focus on a specification, that is, (4) where we have been able to control for time-invariant unobservables and use a lagged value of interest to mitigate feedback effects.

14. Suppose I_p and I_c represent observed measures of child interest as reported by parents and children, respectively. These two measures are designed to measure the unobserved trait child interest (I). Assuming classical measurement error (), the reliability of any observable measure may be defined as the ratio of the true variance in I divided by the total variance in an observed measure. That is Var(I)/Var(I_p) or Var(I)/Var(I_c). In this case the reliability ratio is the same as the correlation between I_p and I_c. In our data the rank (Pearson) correlation between the two measures is 0.40 (0.42) and in 64 percent of the cases, parent and child responses coincide. Additional discussions on measuring unobserved traits and classical measurement error are available in Ashenfelter and Krueger (1994) and Mueller and Plug (2006).

15. It may seem more appealing to have information at the level of the school, however, an advantage of regional level information is that it is less likely to be susceptible to household choice of school. Additionally, although we do include a range of school inputs, based on data collected in 2000, Carneiro (2008) finds that there is little variation in school inputs across Portugal. He reports that, hours of schooling per year, number of computers per students in a school, student-teacher ratio and the proportion of teachers with a degree in pedagogy does not differ across paternal schooling levels.

16. In 1998, of the 1,733 children - 1,730 were in school and 3 were not enrolled. In 2001 we were able to identify the same 1,733 children but an additional 40 dropped out of school between 1998 and 2001 and for 8 children information was missing for some of the variables. This left us with a total of 1,733 – 43 – 8 = 1,682 observations that may be used for the panel data regressions. We did not include children who dropped out between 1998 and 2001 as effectively we only have one observation for their level of interest and educational success. Children who dropped out of school had a lower success rate compared to those who were still in school (44.2 versus 75.4 percent) and also had a lower level of interest (a high level of school interest for 20.9 versus 55.7 percent). This is likely to lead to an underestimate of the link between educational success and interest. However, since sample attrition is quite low (2.4 percent) it is unlikely to have a substantial effect on the estimates.

17. Regression estimates (available on request) based on the smaller and larger data sets are not very different. Although, as may be expected estimates based on the smaller data set are less precise.

18. There are some discrepancies, as 16 children (0.92 percent) who indicated that they had failed in 1998 are found in the category of never failed in 2001. We exclude these children from the regression analysis.

19. That is, say using the estimates in Table 15-column 3, we have 0.036 + 0.026 = 0.062 with a standard error of 0.021.

20. In addition to the effect of motivation, the estimates reveal that increases in the qualification of teachers (decrease in the percentage of teachers with less schooling) and larger schools are associated with an increase in educational success. A 5 percentage point reduction in temporary contracts is associated with a 3 percentage point increase in educational success and a larger number of pupils per school translates into higher educational success. For example, an increase in average school size by 50 students is associated with a 4 percentage point increase in educational success. The link between achievement and school size is interesting as in 2010–2011 the Portuguese schooling system underwent a major change. Following demographic and locational changes in the population, around 700, mainly primary schools were closed and 10,000 students moved into larger schools. Although, the mechanism is not clear, our results suggest that such a move may be associated with an increase in educational achievement.

21. In cases, as in the present context where some of the regressors include variables with repeated values within groups, ignoring intra-group error correlation may lead to incorrect statistical inference (see Moulton, 1986). We ran a series of regressions, replicating the estimates presented in columns 4, 5 and 6 of Table 15 but now allowing for intra-municipality error correlation. These estimates (available on request) display a slight increase in standard errors but it does not alter the conclusions drawn on the basis of Table 15.

22. The correlation between the two measures is 0.42 and 64 percent of the parental and child responses coincide. We may thus consider a reliability ratio of between 0.42 and 0.64. Adopting a conservative approach we provide estimates pinning the reliability at 0.6.

23. Martins (2011) finds that the program is associated with a 10 percentage point reduction in grade retention.