The Margin Muse

*Monitoring and evaluation consultant visits revealing interesting organizational dynamics. Local NGO staff present polished success stories when evaluators from Delhi arrive, but informal conversations reveal different picture. Program adaptations happening on ground but not documented in official reports because they deviate from donor-approved design. Real innovation getting lost because reporting systems can't capture emergent learning. Need evaluation frameworks that reward adaptation, not just compliance.

FEATURED

NOTE#

Development Economics Field Notes

*Effective altruism workshop in Bangalore highlighting cultural assumptions embedded in utilitarian frameworks. Participants struggling with idea of maximizing aggregate welfare when local ethics emphasize relational obligations and community harmony. "Greatest good for greatest number" conflicts with duty-based moral systems. EA movement needs cross-cultural dialogue about moral foundations, not just better impact measurement. Western philosophy isn't universal foundation for doing good.

*Digital literacy program for women's groups showing interesting technology adoption patterns. Older women learning smartphones faster than expected when training focuses on family communication rather than abstract digital skills. WhatsApp groups for coordinating SHG activities becoming platform for wider knowledge sharing. Technology adoption accelerates when aligned with existing social practices rather than imposing new behaviors. Start with familiar use cases, expand from there.

*Randomized controlled trial in rural Jharkhand facing unexpected contamination effects. Control villages learning about intervention through marriage networks and seasonal migration patterns. Randomization assumes neat boundaries between treatment and control that don't exist in practice. Social networks don't respect research design constraints. Need better methods for handling spillover effects in highly connected communities. Pure control groups might be impossible in small-world settings.

*Capacity building workshop on participatory research methods revealing power dynamics within research teams. Senior researchers talking about "giving voice" to communities while junior local researchers doing actual listening and translation work. Participatory rhetoric masking hierarchical research practices. Most important insights coming from research assistants who spend months in field, but they're excluded from analysis and writing phases. Need to decolonize research practice, not just research topics.

*Social emotional learning program in Haryana showing unexpected results. Kids from lower castes responding better to SEL interventions than upper caste children. Working theory: marginalized children already develop emotional regulation skills as survival mechanism - formal SEL just gives them vocabulary and frameworks they intuitively understand. Privileged kids less practiced at emotional self-management. SEL isn't neutral - it interacts with existing social hierarchies in complex ways.

*Art-based education pilot in government schools facing resistance not from teachers but from parents. "Why is my child drawing when they should be learning math?" Art seen as luxury, not learning tool. But follow-up assessments show kids in art program scoring better on creative problem-solving tasks, which correlates with better performance in math word problems. Need to rebrand creative education as "analytical thinking" rather than "art" to get community buy-in.

*ImpactMojo dashboard development hitting classic user experience problems. Program managers want simple red/green indicators, but development work is inherently messy. Binary success/failure metrics miss most interesting insights. Building toggle between "board presentation view" and "researcher view" - same data, different framings. Lesson: impact visualization needs multiple audiences, not one-size-fits-all approach.

*Open source data stack for NGOs keeps hitting the same bottleneck: institutional knowledge locked in Excel sheets on individual laptops. Technical solution exists, but organizational behavior change harder. Convinced one organization to migrate by starting with their donor reporting requirements - everyone understands that pain point. Sometimes need to find the burning platform to drive adoption of better systems.

*Climate resilience study in Sundarbans revealing interesting gender dynamics. Women more accurate at predicting weather patterns and flood timing than men, but their knowledge rarely incorporated into formal early warning systems. Traditional ecological knowledge coded as "women's intuition" rather than valuable data. Development programs keep reinventing weather monitoring when local expertise already exists - just not valued by formal institutions.

*Animal welfare evaluation in dairy cooperatives challenging utilitarian assumptions. Higher animal welfare standards increase costs, reduce profits for small farmers. But also reduce veterinary expenses, improve milk quality, create premium market access. Classic cost-benefit analysis misses temporal dimension - upfront costs, delayed benefits. Effective altruism framework needs better modeling of transition pathways, not just equilibrium outcomes.

*Forest rights implementation varying dramatically based on local political economy. In areas with strong leftist movements, communities more successful at claiming rights under Forest Rights Act. In areas dominated by traditional elites, same legal framework produces different outcomes. Institutions don't work in vacuum - they interact with existing power structures. Policy evaluation needs to control for political context, not just demographic variables.

*Caste census discussions revealing measurement challenges familiar from development economics. How do you categorize fluid, contextual social identities? Same family might claim different caste status for different government programs. Enumeration process itself changes what's being measured - people start thinking strategically about classification. Census as intervention, not just measurement tool.

*Professional development workshop on machine learning for development hitting classic garbage-in-garbage-out problems. Beautiful algorithms, terrible data quality. Participants excited about fancy techniques but haven't solved basic data collection problems. Machine learning can't fix fundamental attribution issues in impact evaluation. Need to teach data foundations before advanced methods.

*Econometrics training focusing too much on internal validity, not enough on external validity. Perfect randomization in one village doesn't tell us about scalability across different contexts. Field experience shows implementation varying dramatically across sites even within same program. Need ethnographic methods alongside experimental ones. Mixed methods isn't compromise - it's necessity.

*Inequality measurement missing crucial dimensions visible from field perspective. Wealth indices capture material possessions but miss social capital, political access, cultural legitimacy. Upper caste landless farmers have different opportunities than lower caste farmers with same land holdings. Standard economic measures of inequality missing most important forms of stratification in Indian context.

*Public health program evaluation caught in classic attribution trap. Maternal mortality declining, but multiple interventions happening simultaneously - new hospital, ASHA worker training, cash transfer program, better roads. Rigorous evaluation would require isolating each component, but real world doesn't work that way. Sometimes need to evaluate packages of interventions, not individual components.

*Data operations for development sector incredibly primitive compared to tech industry. NGOs spending weeks on Excel gymnastics that could be automated in hours. But resistance isn't just technical - it's about control and job security. Data democratization threatens existing hierarchies. Change management as important as technical architecture.

*Education evaluation missing classroom realities. Teacher training program shows improvement in test scores, but field observation shows teachers teaching to the test, abandoning other subjects. Gaming the metrics while defeating program purpose. Need measures of educational breadth, not just performance on specific assessments. Goodhart's law in full effect - when measure becomes target, it ceases to be good measure.

*Climate adaptation project in Odisha cyclone-prone areas showing interesting migration patterns. Families sending one member to city not as distress migration but as risk diversification strategy. Urban earnings provide insurance against climate shocks to rural livelihoods. Migration as adaptation, not failure of rural development. Policy framework needs to facilitate circular migration rather than trying to prevent it.

*Gender program targeting women's economic empowerment inadvertently increasing domestic violence in some households. Economic empowerment challenging existing power dynamics, creating backlash. Need safety measures alongside economic interventions. Women's empowerment isn't just about access to resources - it's about renegotiating household and community power relations.

*Forest conservation program using payments for ecosystem services showing elite capture problems. Large landowners claiming payments for forest conservation they were doing anyway. Small farmers can't access program due to documentation requirements. Environmental programs replicating existing inequalities unless explicitly designed to address them. Conservation and social justice can't be separated.

*Effective altruism cost-effectiveness calculations breaking down when applied to complex social interventions. Easy to calculate cost per malaria net, harder to value institutional capacity building or social norm change. Long-term transformative change doesn't fit neatly into QALY frameworks. Need better methods for valuing systems change alongside direct service delivery.

*Open source technology adoption in development sector following classic diffusion patterns but with twist - donor requirements often drive adoption before organizational readiness. Organizations implementing complex systems to satisfy funder mandates, then struggling with maintenance and capacity. Technology adoption needs to match organizational development stage, not just donor preferences.

*Social emotional learning research showing measurement challenges similar to other "soft skills" work. How do you measure empathy, emotional regulation, social awareness without culturally biased instruments? Western-developed SEL assessments missing important cultural dimensions of emotional intelligence. Need locally-adapted measures, not just translated versions of US instruments.

*Participatory evaluation workshop revealing power dynamics invisible in standard evaluation frameworks. Community members giving different feedback when NGO staff present vs absent. "Beneficiary" voices change based on who's listening. Participation isn't neutral - it's shaped by existing hierarchies. Need to design evaluation processes that account for these dynamics, not pretend they don't exist.

*Impact measurement for arts education programs struggling with attribution challenges. How do you isolate effect of drama therapy on trauma recovery when kids are also receiving counseling, family support, improved nutrition? Arts interventions work holistically but evaluation frameworks demand component-wise analysis. Maybe need to evaluate arts programs as part of integrated support systems rather than standalone interventions.

*Building open data infrastructure for small NGOs revealing interesting capacity building needs. Organizations want dashboards and automation but lack basic data hygiene practices. Garbage in, garbage out applies to fancy visualization tools too. Need to sequence capacity building - data collection standards before analysis tools, analysis skills before machine learning workshops.

*Climate resilience evaluation in coastal Andhra Pradesh showing interesting intergenerational dynamics. Older fishermen's traditional knowledge about weather patterns more accurate than meteorological forecasts for local conditions. But younger generation dismissing traditional knowledge as superstition. Climate adaptation programs need to bridge traditional and scientific knowledge systems, not replace one with other.

*Animal welfare measurement in poultry sector revealing cultural and economic tensions. Consumers say they care about animal welfare in surveys but price sensitivity dominates purchasing decisions. Middle-class urban consumers willing to pay premium for "cage-free" but rural consumers focus on affordability. Effective altruism interventions need to account for economic constraints, not just moral preferences.

*Forest rights documentation process creating interesting legal anthropology insights. Same forest land claimed by different communities using different legal frameworks - traditional use rights vs formal titles vs environmental regulations. Legal pluralism in action. Forest Rights Act trying to reconcile different property regimes but implementation varying based on local power dynamics.

*Caste dynamics in education programs more complex than simple discrimination narratives. Dalit children sometimes performing better in mixed-caste classes due to peer effects, sometimes worse due to social exclusion. Context matters enormously - teacher attitudes, community norms, school leadership. Anti-discrimination policies need locally-adapted implementation strategies.

*Gender programming evaluation missing male engagement dimensions. Women's empowerment programs focusing exclusively on women while ignoring need to work with men and boys. Unaddressed masculinity norms undermining program effectiveness. Gender transformation requires working with entire community, not just target demographic.

*Data governance for development sector lagging behind data collection capabilities. Organizations collecting sensitive information about vulnerable populations but lacking proper data protection protocols. GDPR-style frameworks needed for development sector but adapted for low-resource contexts. Data ethics isn't luxury - it's necessity for maintaining community trust.

*Professional development in international development sector showing interesting geographic biases. Training opportunities concentrated in major cities, leaving field-based practitioners underserved. Remote learning helps but bandwidth and device limitations real constraints. Need to decentralize capacity building, not just program implementation.

*Econometric methods training overemphasizing technique sophistication at expense of contextual understanding. Students learning advanced panel data methods but can't identify when sample selection bias might be problem. Field experience suggests methodological pluralism more valuable than technique purism. Need more mixed-methods training, less method wars.

*Inequality research showing interesting patterns when incorporating social mobility measures. Static wealth measurements miss dynamic aspects of stratification. Families may have similar asset levels but very different trajectory expectations based on caste, gender, geographic location. Social mobility aspirations affect current behavior in ways standard economic models don't capture.

*Traditional surveys miss so much. Spent two hours at a roadside chai stall in Rajasthan - learned more about informal credit networks than any structured questionnaire would capture. The chai wallah knows everyone's business, tracks who's borrowing from whom, notices when remittances arrive. Next time: budget formal time for "informal" data collection. The margins are where the real economy lives.

*Annual Status of Education Report data goes back to 2005 - publicly available, district-level learning outcomes. But buried in appendices: school infrastructure data nobody talks about. Single-teacher schools, lack of toilets, mid-day meal disruptions. Could link this to long-term earnings data from IHDS panels. Why isn't anyone doing this longitudinal infrastructure-outcome analysis?

*Textbook says credit constraints bind investment. Reality: farmers in Sitapur district have access to formal credit but don't use it for inputs. Why? Trust issues with bank officers, paperwork hassles, timing mismatches with seasonal needs. It's not about access - it's about usability. Need better measures of "effective" financial inclusion beyond account ownership.

*Key players not obvious from org charts: Retired IAS officer now consulting for multiple NGOs, has real influence on state water policy. PhD student at IISc whose father is in irrigation department - knows technical and political sides. Local journalists who've covered droughts for decades. Academic collaborations need these "bridge" people who speak multiple languages of expertise.

*Women's self-help groups in Tamil Nadu villages incredibly sophisticated about financial management - detailed tracking, peer monitoring systems. But smartphone adoption for agriculture apps much slower than among male farmers. Not about capability - about social norms around technology as "male domain." Intervention design needs to account for gendered spaces of learning.

*Still using British-era revenue maps from 1920s in some areas. Land fragmentation patterns today reflect colonial administrative decisions. Current microfinance repayment issues trace back to historical jotedar-cultivator relationships. Development economists treat history as "background" but it's actively shaping present outcomes. Need more papers linking historical data to contemporary puzzles.

*Learned the hard way: village elections happened during baseline survey period. Completely changed household participation patterns. Political affiliations suddenly mattered for research participation. Sample selection bias through the roof. Lesson: political calendars are as important as agricultural calendars for research timing. Local partners should flag these windows.

*Aadhaar rollout creating natural experiments everywhere. Administrative data linkages getting easier. But privacy concerns rising - need to strike research agreements now before access gets restricted. JAM trinity (Jan Dhan, Aadhaar, Mobile) changing how we can measure financial inclusion in real-time. Window for baseline studies closing fast.

*Everyone studying climate-agriculture links. But what about climate-education? Extreme heat days affect school attendance. Flooding disrupts learning continuity. Air pollution and cognitive development. ASER data has geographic coordinates - could match with weather station data. Surprised this isn't a bigger research area given climate urgency.

*Same development challenge, different institutional responses. Rural employment guarantee works differently in India (MGNREGA) vs Bangladesh (Employment Generation Program). India's federal structure allows state-level variation - natural experiment. Bangladesh's centralized approach more uniform implementation but less local adaptation. Need systematic comparison of federal vs unitary approaches to social protection.

*June-September: Don't schedule rural surveys - farming season, people unavailable. October-November: Post-harvest, good time for income/consumption surveys. December-February: Wedding season affects spending patterns. March-May: Pre-monsoon anxiety affects behavior. Research timing isn't just about weather - it's about emotional and economic cycles.

*Best paper used administrative data on teacher transfers to identify impact of corruption on education outcomes. Methodological insight: corruption creates quasi-random variation in teacher quality if you can identify clean vs corrupt transfers. Could apply similar logic to health worker postings, agricultural extension officer assignments.

*Science, Technology & Innovation Policy fellowship specifically wants India-focused development research. Perfect fit for technology adoption studies. Deadline usually October. Need to start thinking about mobile money, digital agriculture platforms, telemedicine adoption. Make sure to emphasize innovation ecosystem angle - they love "startup India" connections.

*Standard asset indices miss crucial items for rural South Asia. Include: quality of roof (tin vs thatch vs concrete), number of rooms, separate kitchen, type of toilet facility, irrigation access, livestock ownership, mobile phone types (smartphone vs basic), two-wheeler vs four-wheeler. Rural-urban asset indices need different weights - don't just transplant urban measures.

*Economic Research Unit has amazing long-term datasets from West Bengal villages. Professor Banerjee's successors still maintaining panels from 1970s. Could be goldmine for intergenerational studies. Need to reach out through proper channels - institutional MOUs take time but worth it for data access. Their economic anthropology approach complements standard econometrics nicely.

*Poor Economics experimental approach: start with specific behavior puzzle, design intervention, measure mechanisms not just outcomes. Key insight from their work - don't just ask "does it work?" but "why does it work and for whom?" Always include heterogeneity analysis. Survey modules should capture psychological/social mechanisms, not just economic outcomes.

*Women's self-help groups work at household level but scale up to state-level policy influence. Tracked one federation in Andhra Pradesh - household savings discipline leads to collective bargaining power with government schemes. Micro-level behavior change enables macro-level institutional change. Need more papers tracing these scaling mechanisms.

*Spent morning with ASHA worker in Jharkhand village. She tracks every pregnancy, immunization, birth in her catchment area but gets paid per task completed, not for preventive counseling. Result: she focuses on deliverables that pay rather than education that prevents problems. Perverse incentives embedded in well-meaning programs. Wonder how many development interventions fail because we measure outputs instead of outcomes.

*Reading old Planning Commission documents from 1960s. Same problems, different decade. Rural credit access, agricultural productivity, urban migration patterns - identical language to current policy papers. Either we've learned nothing in 60 years or these are genuinely persistent structural challenges. Probably both. Historical perspective should be required reading for every development economist.

*Fascinating conversation with microfinance borrower in Kerala. She takes loan ostensibly for tailoring business but actually uses money for daughter's college fees. Repays from husband's remittances from Gulf. Loan officer knows but doesn't care as long as repayment happens. Fungibility of money makes program evaluation tricky - measuring impact on business income misses actual welfare improvements.

*Administrative data from Pradhan Mantri Awas Yojana shows interesting spatial patterns. House construction completion rates vary dramatically within same district based on distance from block office. Last-mile bureaucracy matters more than state-level policy design. Could use GIS data to measure "administrative distance" as development constraint. Physical infrastructure and bureaucratic infrastructure are both binding.

*Village in Odisha has three different SHG programs running simultaneously - government scheme, two different NGOs. Women belong to multiple groups, taking loans from each. Coordination failure at policy level creating opportunities for over-borrowing at household level. Need better mapping of overlapping programs before designing new interventions. The development sector's left hand doesn't know what the right hand is doing.

*Child malnutrition data from NFHS-5 shows puzzling patterns. Districts with better healthcare infrastructure don't always have better nutrition outcomes. Talked to pediatrician in Madhya Pradesh - she says it's about maternal education and food practices, not just access to treatment. Supply-side investments won't work without demand-side understanding. Culture eats policy for breakfast.

*Interesting natural experiment emerging from Goods and Services Tax implementation. Different tax rates for agricultural inputs vs processed foods creating weird substitution effects. Farmers switching from fertilizer to organic inputs not for environmental reasons but for tax arbitrage. Unintended consequences of complex tax structures. Could track this using agricultural input sales data.

*Technology adoption patterns in Punjab wheat farming don't follow standard diffusion models. Early adopters are often medium-sized farmers, not largest ones. Large farmers have too much to lose if new technology fails. Smallest farmers can't afford initial investment. Sweet spot for innovation adoption might be middle of wealth distribution, not top. Standard models assume monotonic relationship between wealth and risk-taking.

*Gender bias in data collection: male enumerators get different responses than female enumerators for questions about domestic violence, women's mobility, household decision-making. But training materials rarely address this. Our baseline survey shows 15% difference in reported domestic violence rates based on enumerator gender. Need to budget for gender-matched interviews, not just assume training eliminates bias.

*Talking to municipal corporation official in Tamil Nadu about urban MGNREGA implementation. Rural program being adapted for urban slums but nobody thought about how different social networks function in cities. Rural MGNREGA works through village hierarchies and kinship networks. Urban poor have weaker community ties, higher mobility. Same program, completely different social infrastructure.

*Following up on families from 2015 drought study in Maharashtra. Found kids who missed school during crisis never fully caught up academically. But standard education surveys would show they're "enrolled" and "attending" now. Duration effects of temporary shocks get lost in cross-sectional data. Need longitudinal studies to capture these scarring effects. Development isn't just about levels, it's about trajectories.

*Sanitation program evaluation missing crucial behavioral insights. Toilets get built but usage patterns vary by caste, gender, age within same household. Teenage girls use them consistently, elderly women prefer traditional practices. Infrastructure provision without social change creates white elephants. Should be measuring sustained behavior change over 2-3 years, not just construction completion.

*Remittances from Gulf countries to Kerala follow predictable seasonal patterns tied to Eid, Onam festivals, school fee cycles. But recent analysis shows growing volatility - oil price shocks creating employment uncertainty for migrant workers. Household consumption smoothing strategies breaking down when remittance income becomes unpredictable. External economic shocks reaching rural Indian households through migration channels.

*Land fragmentation in Bangladesh showing different patterns than India despite similar demographic pressures. Islamic inheritance law creates different subdivision patterns than Hindu succession practices. Same economic forces, different institutional responses. Could use this variation to identify causal effects of land titling systems on agricultural productivity. Religion as instrumental variable for land institutions.

*Mobile phone tower data from telecom companies could revolutionize development research. Real-time population movement, economic activity proxies, social network mapping. But accessing this data requires navigating privacy laws, corporate partnerships, technical capacity building. Academic institutions need better infrastructure for big data partnerships. Current grant cycles too short for these complex data acquisition processes.

*Noticed interesting correlation in IHDS panel data - households that experienced major health shocks more likely to invest in children's education, not less. Contrary to expected liquidity constraint story. Talking to families, seems like health crises create urgency about human capital as insurance against future shocks. "We realized money can disappear but education stays with you." Need to model education investment as risk management strategy, not just consumption.