**2. Materials and Methods**

We began applying the RP method in 2016, the fourth year of running the World Food System Summer School courses. We conducted the RP method in the 2016 course in Switzerland, in the 2017 course in South Africa, and in the 2018 course in Côte d'Ivoire. Although each course followed a similar concept and framework, each context, cohort of participants and program was unique (Table 1). We developed a protocol for conducting the RP method with the participants, and it was administered by the same faculty member each time to ensure consistent execution.


**Table 1.** Overview of the three summer school cohorts.

For each summer school, we collected data at two points in time—pre-course and post-course. This translated to conducting the RP method on the very first and very last day of the course. The pre-course RPs were constructed just after the participants arrived, to ensure that we were collecting their baseline knowledge before any course inputs. Following the initial welcome and logistics overview, the participants were each provided with an empty sheet of paper (DIN A4) and a single colored felt pen. They were given 15 min to work individually and in silence to draw a picture of the food system, as they understood it at that exact point in time. They were specifically told to rely on images as much as possible, keeping text to a minimum and to use only one color felt pen. Participants were requested to add their first name, date and course to the back of the drawing.

At the end of the 15 min, the participants were broken into small groups together with a facilitator from the faculty team. During the next 20–30 min, the group went around one by one, explaining their picture and what was drawn. The facilitator made notes to support the interpretation of each of the pictures, and participants had the opportunity to discuss with one another questions or insights that arose. In this sense, the activity itself also provided the first content input, through being able to discuss and learn from the drawings of other participants.

Following the group discussions, the lead faculty collected all the pictures for safe keeping until the end of the course. At this point, a scan was made of each picture to capture the baseline picture. On the final day of the course, after the o fficial program was finished, the same process was repeated (post-course RP). This time, the participants were given back their picture they had drawn on the first day of the course and instructed to take another color felt pen and add anything they had learned during the course. They were also allocated 15 min for this process.

The timing of the RP activity was important to make sure that the data represented the baseline at the beginning of the course and status at the end of the course. This created the challenge that any participant who arrived late or left early due to travel restrictions did not draw both pictures. In total, we collected 132 RPs from the three summer school cohorts. Among these were 56 pairs of completed pre- and post-course pictures. The remainder of the pictures were either pre- or post-course pictures of participants who arrived late to the courses or left early. We also collected auxiliary information about all summer school participants, namely gender, the highest degree obtained, and the field of study (natural or social sciences).

The RPs were anonymized before coding and analysis. For coding of the RPs, a set of six food system categories was defined, i.e., value chain, outcomes, actors, system elements, boundaries, and special topics, based on the key elements of food systems [15]. These six categories were further divided into a total of 38 sub-categories (see Table 3). The categories were linked to one of the knowledge-related learning objectives of the course, namely to understand food systems and their outcomes and challenges. The pictures were also scanned for emergen<sup>t</sup> categories that were not included in this initial list. The (sub-)category definition process was conducted by two of the course faculty. Presence or absence of each sub-category in all RPs was coded as 1 or 0, respectively. All pictures were coded by the same person, a second person was consulted in cases of doubt. In cases where interpreting the drawn elements presented in the RPs was not possible without making many assumptions, therefore making coding arbitrary, the picture was not included in the analysis. Thus, after coding, a total of 51 pairs of RPs were analyzed.

For an initial analysis of the coded data, we computed a presence/absence (1/0) variable for each of the six categories, based on the presence or absence of sub-categories in the respective category. We summed up the number of categories represented in each picture and calculated the di fference in the number of categories in the post-compared to the pre-course RPs for each pair (i.e., for each participant). Additionally, the average of this presence/absence variable represents the proportion of pictures that included drawn elements of the respective category. The number of categories in pre- and post-course pictures, the di fference in categories represented in the RPs as well as the proportion of pictures containing elements of the di fferent categories were used to broadly assess in which elements of the food system knowledge gain was highest. Similar to the categories, we also summed up the total number of sub-categories represented in each picture and calculated the di fference in the number of sub-categories in the post-compared to the pre-course picture for each pair (i.e., participant). This di fference was used as a proxy for overall food system knowledge gained in the course.

To identify sub-categories with a particularly low or high knowledge gain, we calculated the proportion of pictures containing each sub-category for the pre- and post-course pictures. This then allowed us to identify areas that could be given more emphasis in future courses to improve the food systems understanding of participants even further. Statistics were run using the R software version 3.6.0 Patched [16]. A generalized linear model (adjusted to the Poisson distribution) was used to test for di fferences between pre- and post-course pictures. To analyze the full dataset, we used a generalized linear model with course cohort, gender, highest degree, and field of studies as explaining variables.
