The impact of a gamified financial education course to mitigate the effects of myopic behavior. Evidence from a Randomized Controlled Trial.

Schools are assigned to the following four experimental conditions:
- Baseline group: Students do not receive any treatment.
- Control group: Students receive a regular form of the financial education course material (without components related to myopia).
- Treatment group 1: In addition to the content received by the control group, students receive materials about estimation of risks as well.
- Treatment group 2: In addition to the content received by the treatment group 1, students receive materials about short-term preference as well.

2021-01-25

2021-04-02

The main outcomes are:
(1) Financial literacy by a test based on twelve questions. The questions measure financial knowledge, attitude and behavior.
(2) Myopic behavior by a test based on eleven questions. The questions measure the (under)estimation of risks and the degree of short-term preference.
(3) The financial knowledge obtained through a test of the main-part of the course material (the one common to the control and treatment groups) based on six questions. The questions are related to pensions, insurance and investment products.

For the outcome variable of financial literacy, we will follow a methodology close to the approach of Atkinson & Messy (2011), Maldonado et al. (2019) and OECD (2013) to both elaborate the questions and the score of the outcome variable. Answers that show good financial literacy in terms of financial knowledge, attitude and behavior will get a score of 1 (in some questions more than one option could be interpreted as good financial literacy), and 0 if related to a low financial literacy degree. In the end, to estimate the outcome variable, the score will be summed as a ratio of the total questions answered.

Part of the questions for myopic behavior were inspired in Jacobs & Matthews (2012), while others were new to reflect other components of myopia. The myopic score was estimated in a similar way to the financial literacy one, with answers indicating a high myopic behavior getting a score of 1.

The outcome variable related to the knowledge of the basic course material will be evaluated by a simple multiple-choice test with only one right answer in each question, with the total score being normalized to one.

References:
Atkinson, A., & Messy, F. A. (2011). Assessing financial literacy in 12 countries: an OECD/INFE international pilot exercise. Journal of Pension Economics & Finance, 10(4), 657-665.
Jacobs, A. M., & Matthews, J. S. (2012). Why do citizens discount the future? Public opinion and the timing of policy consequences. British Journal of Political Science, 903-935.
Maldonado, J. E., De Witte, K., & Declercq, K. (2019). The effects of parental involvement in homework. Two randomised controlled trials in financial education. FEB Research Report Department of Economics DPS19. 14.
OECD (2013), PISA 2012 Assessment and Analytical Framework: Mathematics, Reading, Science, Problem Solving and Financial Literacy, OECD Publishing.

Schools that registered for participation were randomized to the aforementioned four experimental conditions right-after completion of the pre-test. We assessed the level of financial literacy, myopic behavior and knowledge of the course material content of all students before as well as after followed the course. Students assigned to the control and baseline groups completed the same tests at the same time as students in the treatment groups. The students in the baseline group complete the tests even though they will not receive any intervention between the tests. A second post-test will also be applied after a waiting period.

The experiment design will be as follow:
- The students in the baseline group, will complete the pre-test in the beginning of the experiment and a first post-test after four weeks. Between the two tests no intervention will be performed, but after the end of the experiment, schools in the baseline condition can receive the course material if they wish so (for ethical reasons, not leaving any schools behind).
- For the case of students in the control condition, they will also complete in the beginning of the experiment the pre-test. The schools then can schedule a date to apply the course material (between January 25th and April 2nd) and students will answer the post-test immediately after the course. The intervention consists in a gamified financial education course material aimed at teaching the concepts and financial products related to insurance, investing and pensions. After three months of application, students will then answer a second post-test.
- Students in the first treatment group will follow the same schedule as students in the control condition. They will also receive the content of the game received by the control group, but in addition will have an introduction about the bias of myopia and have specific gamified course work related to the estimation of risks, since many people that experiment myopic behavior underestimate risks. This addition will be related to how to interpret different degrees of risks (from negligible to high) in a context (e.g. insurable risks like protection against ski accident in a trip), besides how risks can grow over time.
- Finally, students in the second treatment group will follow the exact same schedule and receive the same content of students in the control and first treatment group. In addition, they will also receive a gamified course material about time-preference, since a higher degree of short-term preference is associated with myopic behavior. This will be done using context (e.g. decision to save extra money for retirement) and in simulation of decisions that can have an impact in the future.

Schools will be randomly assigned to the different experimental conditions by a random number generator in STATA after completion of pre-tests.

The treatment will be randomized at school level. All students and teachers in the same school will be assigned to the same experimental group. In this way, all teachers in the same school received the same teaching material and instructions in order to minimize the possibility of spill-over effects and contamination of the different experimental group intra-cluster.

Yes

Given previous randomized controlled trial studies with Flemish schools in Belgium (e.g. Iterbeke et al., 2020; Maldonado et al., 2019), our work will aim to reach similar number of participating schools (e.g. 60).

References:
Iterbeke, K., De Witte, K., Declercq, K., & Schelfhout, W. (2020). The effect of ability matching and differentiated instruction in financial literacy education. Evidence from two randomised control trials. Economics of Education Review, 78, 101949.
Maldonado, J. E., De Witte, K., & Declercq, K. (2019). The effects of parental involvement in homework. Two randomised controlled trials in financial education. FEB Research Report Department of Economics DPS19. 14.

With an average number of participating students per school of 45, we plan to have around 2700 students.

Baseline group = 450 pupils, 10 schools
Control group = 630 pupils, 14 schools
Treatment group 1 = 810 pupils, 18 schools
Treatment group 2 = 810 pupils, 18 schools
Average number of schools per condition = 15
Average number of pupils per condition = 675
Average number of pupils per school = 45

The computation is based on List et al. (2011) and accounts for intra-cluster correlation in the calculation of the minimal detectable effect size. In our planned experimental setting, we expect 15 schools in each experimental condition, with on average 45 students by school. Details of the calculation: According to List et al. (2011), in a clustered design, the minimum number of observations in each experimental group can be computed as follows: n=2(t_(α/2)+t_β)²(σ/δ)²(1+(m-1)ρ) This implies that the minimum detectable effect size is equal to: δ=σ/√(n/(2(t_(α/2)+t_β)²(1+(m-1)ρ))) Or the minimum detectable effect size expressed as a fraction of a standard deviation is equal to: δ/σ=1/√(n/(2(t_(α/2)+t_β)²(1+(m-1)ρ))) Following the standards of the literature and the rules of thumb from List et al. (2011), we can set a significance level of 0.05 and power to 0.80, which would lead us to t_(α/2)=1.96 and t_β=0.84 (from normal tables). Replacing in the formula (with the planned sample size and assuming intra-correlation of cluster equal to 0.1) we get the minimal detectable effect size of: δ/σ=1/√(675/(2(1.645+0.84)²(1+(45-1)0.1)))=0.31 Reference List, J., Sadoff, S. and Wagner, M. (2011), So you want to run an experiment, now what? Some simple rules of thumb for optimal experimental design, Experimental Economics 14, 439-457