We conducted two rounds of one day long group model building workshops – between October 2018 and April 2019 (n=323) and between November 2019 and July 2020 (n=325) respectively – in 40 schools of Badakhshan, Ghazni and Takhar provinces of Afghanistan and 59 schools of Punjab and Sindh provinces of Pakistan. Group model building is a set of Community Based System Dynamics (CBSD) methods that provides a structured process and forum for stakeholders to identify and prioritize issues through the language of systems. We conducted four distinct activities: (i) focus group discussions (FGDs) to define a shared vision of inclusive and equitable quality education; (ii) variable elicitation to identify multiple factors perceived to influence inclusive and equitable quality education; (iii) connections circle to establish perceived connections between identified factors and identify causal feedback loops and, (iv) action ideas elicitation for participants to agree on interventions to affect the system. Trained facilitators assisted school stakeholders in the process. Visual causal maps were elaborated by children, parents, teachers and school management committees. During each workshop, several iterations of the model with ongoing discussion of connections between factors were elaborated before a final visualization was agreed upon by the group. The building of the CLD and its review allows researchers to gauge each group of participants’ consensus vision of the system. After the conclusion of the workshops, facilitators reproduced the final Causal Loop Diagram models using Vensim® PLE software. We used techniques from multivariate analysis of ecological communities to compare identify features commonly perceived to be central to a successfully inclusive and equitable education system. These methods, which have been previously applied beyond ecology, allowed us to compare CLDs by i) quantifying distance/similarity measures between them based on number of shared components, ii) distinguishing clusters of CLDs that are more similar to one another, and iii) aggregating across clusters to identify the common components shared by all. These multivariate analyses also allowed us to test which school, teacher or student characteristics correlate with differences among CLDs. To do this, we ordered CLDs built by each group using non-metric multi-dimensional scaling on Euclidian distance matrices of causal feedback loops expressed by each CLD using the R package vegan. We then projected the CLDs into an ordination space in which their distance from one another was defined by the difference of their causal links. Thus, CLDs that shared a great number of causal links were plotted close together; those that differed in causal links were plotted farther apart. Finally, we tested how well the resulting distribution of CLDs in the ordination space was explained by five sets of characteristics of the groups who built them. First, we considered role in the educational system (children, teachers, school management committee members, parents). Second, the sex balance of the model-building group was considered, which we quantified as the percentage of participants identifying as male, or, for the sake of characterizing groups, a binning of this variable into models built by groups >80% female, >80% male, or mixed. Third, geography was examined at three nested levels: the 99 schools, eight administrative areas (within Afghanistan the provinces of Badakhshan, Takhar and Ghazni and the distinct district of Jaighori within Ghazni; within Pakistan the districts of Bahawalnagar, Rahim Yar Khan and Vehari in Punjab province and Gothki in Sindh province), and two countries (Afghanistan and Pakistan). Fourth, wealth was measured by two alternative indices with scores calculated using polychoric principal component analysis [57]. One characterized common durable goods: radios, mobile phones, TV sets, pressure cookers, lamps, refrigerators, generator/solar panels, sewing machines, bicycles, motorbikes/auto-rickshaws, cars, houses. The other characterized rural wealth: animals, i.e. camels, cows and buffalos, sheeps and goats and poultry. Each index was divided into three categories (lowest 20%/ middle 20-80%/ highest 20%). Fifth, time period, which captured the difference between workshops conducted before and workshops conducted after an intervention composed of an intensive two week inclusive education training and the implementations of three to five action ideas identified and selected by participants. Action ideas could include construction of infrastructure (such as classrooms, playground, restrooms, boundary walls, library), classroom material (chairs, tables, books) or procedures (regular parent-teacher meetings, classroom regulations, teacher and students’ attitudes).Disadvantaged children in Low Income Countries (LICs) particularly children with disabilities are increasingly accessing schools, but not learning effectively due to social exclusion within the classroom and poor teaching methods, that perpetuate inequality. In order to unpack the equity-quality nexus, education research and policy need to move beyond considerations of service delivery assessed through achieving benchmarks on basic cognitive skills and completing school cycles, traditionally considered valid proxies for quality of education; a view challenged by experts advocating for the importance of non-cognitive or psychosocial skills in assessing quality education. Promoting processes of accountability is recognised as a way of promoting quality and equity in education. We argue that the role that parents and community members can play in improving the quality of education through innovative social accountability mechanisms has not been sufficiently explored in LICs. Building upon previous research, we will develop, implement and evaluate a social accountability intervention - combined with inclusive education training - engaging parents, teachers and children. We will assess the intervention's impact on basic cognitive but also psychosocial skills of learners as well as parent's expectations and engagement, and teachers' confidence with regards to inclusion of children with disabilities. Two partners, Swedish Committee for Afghanistan (SCA) and the National Rural Support Programme in Pakistan (NRSP) run Community Based Schools (CBS) in remote areas. They require knowledge as to the factors that can promote or hinder a meaningful learning experience. We hypothesize that strengthening existing social horizontal accountability mechanisms through a Community Based System Dynamics (CBSD) - a participatory approach promoting local ownership in the process of deciphering and changing complex systems from the feedback perspective of system dynamics - can enhance inclusion and the learning experience of disadvantaged children. Developing an effective role for school management committee (SMCs) members requires careful consideration of context and community dynamics, which CBSD upholds. We will carry out Group Model Building (GMB) sessions with SMCs - a specific CBSD method- to identify insights about social accountability reforms. SMCs supported by SCA/NRSP will implement changes based on these insights in intervention CBSs. A randomised control trial (RCT) will evaluate this intervention. Qualitative methods will validate psychosocial assessment tools and explore stakeholders' perception of education. In stage 1, investigators will (i) decipher existing mechanisms of accountability and monitoring being used in CBSs (ii) train teachers, 4 NRSP and SCA teams of 3 facilitators and 2 coordinators each on inclusive education and (iii) train SCA/NRSP teams on GMB to facilitate sessions in 80 randomly selected intervention schools. In stage 2, SMCs members (principals, staff, teachers and parents) will participate in separate GMB sessions to design a relevant school social accountability system and identify leverage points on which to focus the intervention. In stage 3, each school, with the support from SCA/NRSP teams and the investigators will implement the intervention. In stage 4 the investigators will measure the impact of the social accountability intervention on learning outcomes using a cluster RCT with two waves of interviews: baseline survey in all 160 schools from staff, teachers, students in classes 3 to 5 and parents in year 1; end line survey will take place at the beginning of year 4. In depth interviews and FGDs, games and audio-visual activities with children will take place in years 2 and 3. In stage 5, the investigators will develop capacity in countries (universities, education NGOs) and disseminate findings to a broad audience in various formats through academic, policy makers and practitioners' networks.

We conducted workshops between October 2018 and April 2019 (n=323) and between November 2019 and July 2020 (n=325) respectively – in 40 schools of Badakhshan, Ghazni and Takhar provinces of Afghanistan and 59 schools of Punjab and Sindh provinces of Pakistan, and established causal loop diagrams.