The method for random treatment allocation was Coarsened Exact Matching (CEM). The purpose of this method is to maximize the likelihood of identifying a statistical significant effect through a randomized control trial. To do this, the method forms groups of similar clinics and randomises treatment status within the group.
The following variables were selected to stratify clinics and form groups with the CEM:
1. A social lag index. This index is developed by the National Council of Social Policy Evaluation (CONEVAL) and updated every 5 years. Based on this index, CONEVAL classifies localities in five different categories: very low, low, medium, high and very high social lag.
2. Type of clinic. Clinics were segmented in blocks according to the type of locality where they are located [rural or urban] and the size of the clinic [small, medium or large]. The size is determined by the number of nuclei available. One nuclei is a set of two or three medical personnel attending in the clinic. Five blocks were defined: (i) small, rural clinics, (ii) medium and large rural clinics, (iii) small, urban clinics, (iv) medium, urban clinics, and (v) large, urban clinics.
3. Federal Mexican entity (State). Five states were selected in the sample: Chiapas, Estado de Mexico, Guanajuato, Hidalgo and Puebla. These states were selected based on their population, geographical, demographical and economic heterogeneity. Looking at different states creates differential ease of access to clinics, exposure to technology, traditions, cultures and languages.
4. Level of education. Using the Mexican census, we create two categories based on the percentage of female population above 15 years of age with incomplete secondary schooling.
The following stages were followed to allocate each clinic to either of the three treatment groups or to the control:
In the first stage, all the stratifying variables were used to form groups. Sets of six clinics were formed whenever a group had more than six clinics in it. If the number of clinics in the group is not a multiple of six, clinics were randomly selected to form sets of six clinics and those that were not allocated to a set of six moved on to the next stage. In the second stage, we drop level of education as a stratifying variable and repeat the procedure described above. The third and fourth stages repeat the previous steps by first dropping social lag and then state. This means that the third stage matched clinics based on state and type of clinic, and lastly, the forth state only matched clinics based on their type. Finally, the fifth stage collected the clinics that had not been allocated and randomly chose for each clinic their allocation.
This method was executed with the 655 clinics sample, out of which 329 were allocated to the control group, 107 to treatment arm one, 111 to treatment arm two and 108 to treatment arm three.