Experimental Design
We will randomly assign SMEs into two treatment arms and one pure control. The two treatment arms are orthogonal to enable us unpack the effect of each additional intervention. A number of dependent variables are of interest in this study. These outcome variables include digital technology adoption or change in business practices, and business outcomes or performance (sales, cost, profit, productivity, and employment). We will first examine whether randomization across treatments achieved balance in pre-treatment characteristics. In this regard, we will examine the differences in means of baseline variables for the two treatment groups and the control group. To attenuate any concern that baseline imbalance can drive our results, we will include the full set of baseline characteristics as controls in our main regressions.
In this study, we will focus on intent-to-treat (ITT) estimates, largely because we do not expect everybody who receives the digital training will adhere to the latter.
The total sample size is 1,200 firms. Assuming a sample size of 800 firms with 400 firms in each treatment (33.3% in each treatment arm) which is compared with 400 firms (33.3%) in the control group, we are powered to detect an effect size of at least 0.09 standard deviations on intensive margin (degree of adoption) with power of 0.8 and significance level of 0.05. This effect size is reasonable based on results found in other interventions that perform training for small and medium enterprises.