2.5 Results
2.5.5 Heterogeneity
2.5.5.1 Gender
There has been increasing interest in gender difference in terms of attitudes
towards technology or computers (Ardies et al.,2015(7); Potvin and Hasni,2014
etc(99).). Relative to boys, girls might have more negative attitudes and may be
less actively engaged in technology-related activities. This study also demonstrates gender differences in a way. Boys are more likely to have their own computer as their average propensity score is approximately 13% higher than that of girls. Gender differences also exist in selected behaviours: reading, school and home ICT usage. On balance, boys tend to be more enthusiastic about computer-related
activities than girls, as can be seen in the upper panel of Table 2.6.
The lower panel of Table 2.6 shows greater discrepancy insofar as the treatment effects of boys is almost twice as much as that of girls - this might be explained by some gender-specific behaviours. After controlling for these behaviours, the gender gap in treatment effects is narrowed. Boys are relatively more influenced by personal computers which increase their university participation by around seven percentage points on average; the same estimate is four percentage points for girls. However, such differences in the effect of a personal computer on university attendance are not statistically significant, partially owing to larger standard errors in subgroup analysis with fewer observations.
When computer-related behaviours are broken down further, it is observed that boys do not devote more time to schoolwork using a computer at home, and they play computer games much more than girls. Whereas girls, on average, have better reading habits and doing schoolwork. However, such constructive computer usage on schoolwork might be offset by their greater interest in online-chatting and music, or browsing probably. The results of this study are consistent with some literature showing no evidence of greater treatment effects of ICT for girls (Malamudet al.,2012(84); Faberet al.,2015(44)).
The gender gap is also often discussed when it comes to education attainment and participation. Some explanations include gender socialization and innate dif- ferent interests and skills (Schoon and Ecceles 2014)(106)). But empirical evidence based on similar UK cohorts suggests that the gender gap in HE participation could be substantially reduced by including the prior academic attainment into account (Crawford and Greaves,2015(26)). We may conjecture that girls’ decisions over higher education are not additionally affected by these ICTs but basically corre- spond to their academic performance in secondary school. For boys, the positive
2.5 Results
effect is still sensitive to the inclusion of behavioural variables that also relate to learning habits. A laptop might not necessarily switch some of their entertainment habits to more learning-oriented ones. Instead, it might promote their aspiration for the university where ICT can be more widely and freely used.
2.5.5.2 Family Background
It is of interest to investigate potential heterogeneity in the ICT effectiveness by family background. Parents play important roles in home ICT investment and may have additional influence on computer usage as a result of their different ed- ucational levels or working experience. As shown in Table 2.7, there appears to be a larger impact of personal computers on university participation for students with less educated parents, which is similar to the findings by Fiorni (2010)(52). Regarding family social-economic class, the ATT is around five percentage points and statistically significant for the groups of parents who hold intermediate oc- cupations in sales, clerical, service and auxiliary. In the absence of significant difference in the propensity score of having own computer, our results suggest fewer impacts from the advantaged background but moderate positive impacts from other groups. The likely explanation is that the individual-specific computer purchase is more affected by students’ every-day usage and preference instead of parental discipline.
2.5.5.3 Propensity Score Stratification
In general, the propensity score in our sample ranges from 0.3 to 0.8 and is mostly clustered around 0.5. It is worth recalling that observations within each stratum might have specific characteristics that are ambiguously reflected by the average treatment effects. Table 2.8 presents different matching results within
different stratum that is divided to ensure the mean of covariates does not dif- fer within each stratum. In the main specification, it seems that the estimated treatment effect is largely driven by the groups of people with higher propensity score over 0.6. The ATT reaches up to 0.045 in the group with relatively highest propensity score, compared to the average treatment effect 0.030 in the baseline specification. Figure 2.5 graphically plots the varying treatment effect using local
polynomial regression 13(Fan and Gijbels 1996 (50)). We can observe the mono-
tonically increasing treatment effect in the main specification but not for the other specification with behavioural variables. For the high-tendency group above 0.6, the treatment effect becomes smaller and statistically insignificant. These results show suggestive evidence of an inverted-U shape of the treatment effect that is usu- ally higher between the 50 and 75 percentile than the other two extremes, which further highlights the importance of behavioural controls. It seems that personal ICT investment does not enhance the university attendance for the most computer desirers who might need them more out of entertainment purpose than e-learning. Apart from this discrepant trend, the estimates for the average population with propensity score between 0.5 and 0.6 are similar and consistently positive.