Measuring Non Cognitive Skills using portable EEG devices

This project study the Chilean program ``Mining’s Rockstars’’ - ``Rockstars para la Mineria’’, that aims to foster life skills, in particular self confidence and creativity, following principles that could be summarized as: ``learning by failing, gaming, doing and rethinking’’\footnote{For details see the Appendix.}. The intervention evaluated here was a randomized control trial that targeted young students that were in technical –vocational- high school for one academic semester (around four months). The randomization was done by the NGO together with the researcher of this study. We randomized at school level, since administrative restrictions.

2015-03-16

2015-11-13

I explore the effects of student’s participation on cognitive, life skills and creativity by using commonly-use psychometric tests. In particular, it was measured the Big Five Inventory Test, Locus Control Test, Grit Test, and Torrance’s Creative.
I also took a raven test as a placebo test, since shouldn't be affected by the treatment.

I measured a proxy for emotional state, in the valence dimension, and emotional responsiveness to positive and negative stimulus.

The constructed outcomes will be the emotional state and emotional responsiveness.
In addition, I am planning to construct emotionally-corrected scores of the aforementioned psychometric test (Grit, Locus Control, etc.). The idea is to correct the scores by the emotional state at baseline.

In particular, I use low-cost portable electroencephalogram (EEG) headsets to estimate subject’s emotional state following methodologies from emotion detection theory (See for example Takahashi, 2004; Ramirez and Vamvakousis, 2012; Verma and Tiwary, 2014). In fact, emotions will be modeled using the arousal-valence model developed by Lang (1995), in which arousal is related to activation and deactivation and valence is related to gratification and displeasure. I'll have particular focus on valence dimension, since it is theoretically unambiguously related with perception, having a positive relation with many psychometric test scores.

This study consists of eight randomly selected schools. The randomization considered only "technical" schools, i.e. vocational schools, as well as schools that belong to the 4th Region of Coquimbo in Chile, since they were eligible to receive the program.
The study will measure the program's impacts through experiment and surveys of students in a random sample of four schools from the eight randomly selected schools participating in the study. The randomization was done by both the NGO that run the program an the main researcher of this project.

The data analysis will follow the methodology developed by Ramirez and Vamvakousis (2012) to measure emotions from EEG signals. Evidence has shown that the prefrontal cortex coordinates consciousness and regulates emotions (Bear et al. 2007). In fact, to measure emotions it would be consider the EEG signal using electrodes AF3, AF4, F3, and F4 in the 10-20 standard classification, which are located on the prefrontal lobe (For the exact location of each electrode in the brain, see Figure 10 20 SYSTEM in the appendix.). By applying a Fourier transformation, it is possible to separate the EEG signal into different frequencies, from 1 to 80Hz. The frequencies of interest in measuring emotions are both alpha (8-12Hz) and beta (12-30Hz) waves (Bos, 2011). Alpha waves are predominant in relaxed states and brain inactivation, while beta frequencies are associated with alertness and excited state of mind. Consequently, the beta-alpha ratio is an effective indicator of the level of arousal. On the other hand, previous neurophysiological evidence had shown that emotional valence has different representation in right and left brain's cortical hemisphere (see Ramirez and Vamvakousis, 2012). Right frontal inactivation is correlated with positive emotion, while left frontal inactivation is related with negative emotion. Since inactivation could be measured as the inverse of arousal level, it is possible to estimate valence level by subtracting the inverse of arousal on the left hemisphere with the the same index in the right hemisphere.

Randomization done in office by a computer.

The unit of randomization was the school.

Yes

8 schools.

200 students for each round.

4 schools control, 4 schools treatment.

Grit Test, Big Five Personality Inventory, Study parameters: Study parameters: alpha = 0.0500 power = 0.8000 N = 300 N per group = 150 m1 = 3.2000 sd = 0.6000 Estimated effect size and experimental-group mean: delta = 0.1947 m2 = 3.3947 i.e., the minimum detectable effect size is 1/3 standard deviations. Locus Control Test (which uses a different scale) will have: Study parameters: alpha = 0.0500 power = 0.8000 N = 300 N per group = 150 m1 = 65.0000 sd = 11.0000 Estimated effect size and experimental-group mean: delta = 3.5700 m2 = 68.5700 i.e., the minimum detectable effect size is 1/3 standard deviations.