National Bureau of Economic Research (NBER), April 2011, Working Paper 16939. Available From:Link to Source
This article evaluates the impact of a scholarship programme on changes in political and social attitudes among adolescent girls in Kenya. Some theories of change suggest that education promotes democratisation and empowerment because people are more willing to challenge authority, whereas others posit that education promotes obedience to authority and reinforces power structures.
The authors use a randomised design to test these theories. They compare outcomes among girls eligible for the scholarship in 69 schools in rural western Kenya (34 treatment and 35 control schools). The top 15 percent of girls in treatment schools who were enrolled in 6th grade in 2001 and 2002 were awarded school fees and a cash grant to cover 2 years of primary schooling. All schools were similar on observed characteristics, and no other interventions were conducted in the schools during this time.
From 2005 to 2007, the authors conducted a follow-up survey of test scores, family traits and social and political attitudes. The girls surveyed included both scholarship winners and non-winners, and the 80 percent respondent rate resulted in a final sample of 1,385 girls aged 17–21 years. Specification checks using control variables confirmed the sample was equally balanced among treatment and control schools.
The authors estimate overall programme impacts using ordinary least squares regression. To control for the effect that winning the scholarship may have had on outcomes, the authors construct an estimate for girls in the treatment schools who had a significantly lower chance of winning the scholarship. Using the follow-up survey’s test score as an endogenous variable in a first-stage model that they then use to estimate an instrumental variable, two-stage least squares model, the authors estimate that persistent human capital gains were unrelated to winning the scholarship. The instrumental variable, two-stage least squares model is also used as a robustness check alongside the ordinary least squares model.
Although girls were more politically knowledgeable and willing to challenge authority, this did not lead to more democratic attitudes. Girls in the treatment schools spent significantly more time reading newspapers and were 19.4 percentage points more likely to identify the most influential English-language newspaper as their favourite. The likelihood that they could name five main political figures increased by a significant mean of 0.220 standard deviations over the control group.
The attitudes of girls in the treatment schools toward household authority changed: the likelihood that parents chose their daughter’s spouse decreased by 2.4 percentage points, and girls’ acceptance of the legitimacy of domestic violence decreased by 25 percent. They were also significantly less likely to think they should show more respect for political authority, and were 5.4 and 6.1 percentage points less likely to feel positive about the government and the economy, respectively.
However, there was no evidence that the scholarship had an impact on seven measures of democratic attitudes, based on responses about how government and society should be organised. Treated girls were 4.4 percentage points less likely to say that ethnic identity was not very important to them, although this outcome was reported only for 2005–2006 survey respondents. There was no significant change in their level of perceived political efficacy, or in their interest in participating in political and community affairs, both of which were quite low. Girls in the treatment schools were also significantly more likely to think that the use of violence to achieve political change was sometimes justified.
The robustness checks comparing the two models show similar results, with the main difference occurring in terms of respondents’ satisfaction with authority. The authors explain this finding by suggesting that the ability to do well in school is tied to acceptance of authority, which may lead to omitted variable bias in ordinary least squares estimates.