Blaming the Environment?
The results of the multinominal logit regression are presented in Table I. It is interesting to observe that, once controlled for the interaction with motivation, gender does not have a significant effect on belonging to one or another group (the estimate of 0.8889 is not significantly different from 1).2 This is not to say that motivation is not related to gender – indeed it appears it is. Therefore, it seems that differences in drop out are partly driven by motivational differences among boys and girls. This adds nuance to previous work by Stearns and Glennie (2006), Goldschmidt and Wang (1999), and Rumberger (1995, 2001). It also impacts the interpretation of findings by Dekkers and Driesen (1997) that within the group of students with a low socio-economic status, dropout is more likely for boys in The Netherlands. We can make a similar observation for Moroccan students, truants and parents' work (finding values in Table I not statistically significantly different from 1). However, students with Turkish or Surinam/Antilles roots drop out more easily than those of Dutch origin (i.e. both 1.4282 and 1.5603 are statistically above 1).
Table 1. Multinominal logit regression
|A. Students' Characteristics|| |
|A1. Students' Exogenous characteristics|| |
|Gender (male = ref)||0.8889|
|Homeland mother (NE = ref)|| |
|Surinam / Antilles||1.4282*|
|Ability * ability||1.0014***|
|A2. Students' Ideas and Behaviour|
|“I like school”||0.8533***|
|“Opinion about math teacher”||0.7479***|
|“Teachers do their best at school”||0.8247***|
|Number of retentions||0.7261***|
|“Teachers are pleased with my results”||0.8034***|
|“I pay attention during explanations”||0.8773**|
|Start homework after||1.1144***|
|B. Environment Characteristics|
|B1. Exogenous characteristics parents|
|Work (worker = ref)|| |
|B2. Interest in schooling and parents' aspirations|
|Attended parents' evening||0.3874***|
|“A high degree is important”||0.9115*|
|“Affiliation with homework”||1.1964***|
|“Talking about school at home”||0.9124*|
Concerning the influence of parents' work, students whose parents work in one-man/one-woman businesses or are self-employed do not drop out more frequently (estimated values, respectively 1.1619 and 1.0859 not significantly different from 1). Similarly, the dropout rates of children of lower, middle, and high level employees (estimated values of 0.9391, 0.7935, and 0.8821) do not significantly differ from those of children from workers' families. We proxy students' abilities by the average of the cito-standarised test scores for math, languages and information (cito-test is a national and standardised test at the end of primary school which is used as an advisory tool for the school track). Increasing abilities of students initially reduce (estimate of 0.8945) the probability of dropping out. This corresponds with previous literature (Alexander et al., 2001; Lee & Burkam, 2003). However, as revealed by the estimated coefficient for the squared term (1.0014 statistically higher than unity), this impact fades out for higher ability students. As a last determinant, students who are older at the start of the cohort drop out far more often than younger students (estimate of 1.3221). Lee and Burkam 2003) reported a similar result.
Variables related to student's ideas, beliefs, motivations, and attitudes towards school all have a significant effect on dropout. More precisely, students who like school have a favourable opinion of their teachers in general and of their maths teacher in particular, have a normal school career in terms of no retentions, have teachers who are pleased with their results, and pay attention during classes are less likely to drop out (all estimates, respectively 0.8533, 0.8247, 0.7479, 0.7261, 0.8034, and 0.8737, are statistically significantly lower than 1). Also, the longer it takes for students to start their homework, the higher the dropout probability (estimate of 1.1144 significantly higher than 1). This confirms the results of previous studies (Fredricks et al., 2004, Dekkers & Claassen, 2001, Entwistle et al., 1997; 2005) that students displaying ‘anti-school behaviour’ are at a greater risk of dropping out of secondary school. Dekkers and Claassen (2001, p. 346) found that most dropout students expressed their dissatisfaction by paying limited attention during classes and not doing their homework.
Parents' opinion and background are also important. Children of parents who are more educated, who attend parents' evenings, talk about school at home, and are more convinced that a high degree is important are much less likely to drop out (estimates respectively equal to 0.8572, 0.3874, 0.9124, and 0.9115, all significantly below 1). De Graaf et al. (2000) show that children of highly educated parents receive cultural resources such as knowledge, tastes and preferences that favour their educational career.
Conversely, extensive parental control and a strong affiliation with homework are associated with a higher probability of student drop out (the estimates, 1.1521 and 1.1964, well above 1). However, students should do their homework themselves (although, it is likely that extensive parental control is a consequence of weak performance at school and not a cause for drop out per se). The location of the school significantly affects the dropout decision, as students drop out more frequently in more urban areas. In The Netherlands, this seems plausible, as urban schools have higher shares of students at risk of dropping out (e.g. students with a minority and/or low socioeconomic status). With respect to the influence of parent's occupation, our results confirm the trend observed by De Graaf and Ganzeboom 1993) that occupational status is becoming less important (they examined the period 1891–1960). In this 1993 wave of data, it does not significantly influence the dropout probability (the estimates of 1.1619, 1.0859, 0.9391, 0.7935, and 0.8821 are not statistically different from 1).3 Nevertheless, literature shows that effective programmes to reduce dropout in high schools often involve collaboration between the school and the parents (Hale, 2000; Gandara et al., 1998).
As a robustness check and a test of the remark by DesJardins et al. (1999), we estimated the multinomial logit model with unobserved heterogeneity. To model a multinominal logit with unobserved heterogeneity (also known as variance of the disturbances), we added unobserved individual effects αi to the probability in equation (1) (and hence in the derived formulations (2) and (3) (Haan & Uhlendorff, 2006). The results are available. Note that the estimations do not change dramatically. Variables that have a significant influence on the decision in the traditional multinomial logit do still have a significant influence when allowing for unobserved heterogeneity. Also, the direction of the effects is largely similar. One important difference with previous estimations is that gender seems to have a significant impact on student dropout. Females are less likely to drop out of school than children of middle employees.
In sum, with regard to students' decision to drop out of school, the estimation results in Table I reveal that motivation and ability are crucial. In this sense, our results follow the conclusions of Dekkers and Driessen (1997) and Dekkers and Claassen (2001). This motivation should come from both the students and their parents, although the latter should not exaggerate: too strict control on homework inhibits the students so that. Hence, they have a higher probability of dropping out.