Description of the data and file structure

As a first step to identify relevant datasets, participants of the SESYNC working group Can Enhancing Diversity Help Scale Up Agriculture's Benefits to People and the Environment? were identified that had studied farm sites with varying levels of diversification in different parts of the world.
Our goal was to encompass a wide range of farming systems across geographic latitudes, from subsistence to commercial farming, while also considering the geophysical landscape context of agricultural diversification outcomes.

As a second step to obtain a sufficient representation of various geographic regions as well as types of agricultural diversification strategies, the group identified additional data contributors with relevant datasets (done through a broader solicitation request for data) as potential collaborators.

We had three inclusion criteria for datasets. That is, datasets should have recorded at least: 1) one agricultural diversification practice, while also including control study sites without or with less diversification, 2) one environmental outcome variable, and 3) one social outcome variable. We compiled 24 datasets that cover 11 countries from 5 world regions: Brazil (5); Malawi (3); Costa Rica (2); Ethiopia (2); Germany (2); Bolivia (1); Canada (1); Colombia (1); USA (5); Ghana (1); and Indonesia (1).

The .xlsx file contains the raw data.

The first sheet of the .xlsx file includes two of the social variables: Food security and human well-being.
The second sheet includes the third social variable: Yield.
The third sheet includes one of the environmental variables: Biodiversity
The fourth sheet includes the second environmental variable: Ecosystem Services
The fifth sheet includes the third environmental variable: Reduced externalities. Also, this sheet includes all data on agricultural diversification practices.

The csv-file includes the constructed variables: six compound outcome variables (including different configurations of human well-being used for the robustness checks), diversification variables as well as landscape variables.

We combined the 24 datasets by standardizing within each dataset by computing z-scores for each data point (field and/or farm) across the subcategories of social and environmental outcome variables. Subsequently, we averaged across computed scores for each of the six aggregated outcome variables. For example, if food security had been measured through two variables, such as the number of hungry months and a dietary diversity score, we first standardized data for each of these two variables within the dataset using z-scores, and then averaged the z-scores to obtain the final value for the aggregate food security outcome variable for that data point within that study.

We used the same standardization procedure whether data were entered by data contributors as a binary variable or continuous variable, by transforming binary data entries to 0/1 values before computing z-scores. For example, one indicator used to assess human well-being was men's'/women's' participation in networks. Participation was recorded by some data contributors as a binary variable of participation/no participation whereas other data contributors recorded the actual number of networks that men/women engaged in. For all binary data entries, we transformed these.

For some variables such as the number of hungry months as an indicator for food security, the direction was reversed with higher values indicating lower food security. We transformed these variables by multiplying by -1 before computing z-scores. We note that the directionality of variables was defined individually by each data contributor thereby accounting for the context in which it was measured. The variables with reversed directionality were:

• Food Security: Number of hungry months
• HFIAS (The Household food insecurity access scale)
• Food security: Other food insecurity metrics
• Human well-being: Mental health
• Ecosystem services: Fruit damage
• Ecosystem services: Pest and disease damage
• Ecosystem services: Partial nitrogen mass balance

Code/Software

All data processing and statistical analyses were performed using R software. The R script is available.