Experimental Design
We will conduct a pilot study and a randomized controlled trial (RCT) with university students recruited online and offline from universities in China. Using the RCT method of impact evaluation, we will be able to ensure that the intervention and control groups have similar characteristics at baseline. By assuring the similarity of characteristics (such as family income, age, sex, etc,) between the treatment and control groups at baseline, we can confidently attribute any significant differences in outcomes between control and intervention groups to the program.
For the pilot study, which we will implement prior to the full-scale RCT, a smaller sample (n=112) will be randomly allocated into two experimental arms (in-person MBP and a pure control). Using STATA 16 software (https://www.stata.com/), with 80% power, and a significance level of 0.05, we determined that a sample size of 51 students per arm was required to detect a difference of 0.5 standard deviations in outcome measures. Our recruitment of 56 students per arm (112 in total) allows for 10% attrition.
For the full-scale RCT, participants (n=290) will be randomly assigned to one of three groups: online 8-week MBP training, in-person 8-week MBP training, and a control group. The 290-student sample size is based on the statistical power calculations for the randomized controlled trial. Using STATA 16 software (https://www.stata.com/), for 80% power and a significance level of 0.05, we calculated a sample size of 78 students per arm was required to to detect a difference of 0.4 standard deviations in outcome measures. Our recruitment of 94 students per treatment arm allows for 20% attrition.
With two treatment arms (in-person and online), we will have 188 students for the treatment group. Due to the multi-arm experimental design, we will increase the size of the control group by 30% to ensure an efficient sample for the comparison across different groups. This means we will have 102 students in the control group. With a total of 290 students, we will have efficient power to detect the effect size of 0.4 standard deviations. The effect size that the trial is designed to detect is based on other mindfulness trials: Breedvelt et al. (2019) reviewed 24 studies with experimental designs and found an overall effect size of 0.6 SD. Thus, as a conservative estimate, we use 0.4 SD to ensure sufficient statistical power.