{"args":{},"checks":[{"abs_tol":null,"bind":"any","column":null,"failure_class":null,"field":"value","interval":false,"key":null,"kind":"answer_backed","min_lre":null,"note":"every number traced to a tool call or the question","open_interval":false,"rel_tol":null,"required":true,"source_id":null,"table":null,"tool":null,"value":null,"weight":1.0,"where":null},{"abs_tol":null,"bind":"answer_call","column":null,"failure_class":null,"field":"value","interval":true,"key":null,"kind":"answer_figure_close","min_lre":null,"note":"CRB estimate_within_solution_ci: the answer figure inside the gold 95% interval effect \u00b1 1.96\u00b7|SE| of source crb_papers.q012.gold (closed, evaluate_result.py; no --auto-rescale)","open_interval":false,"rel_tol":null,"required":true,"source_id":null,"table":null,"tool":null,"value":[0.08959315431963688,0.5831902690734394],"weight":1.0,"where":null},{"abs_tol":1.2591763131474552e-07,"bind":"answer_call","column":null,"failure_class":null,"field":"value","interval":false,"key":null,"kind":"answer_figure_close","min_lre":null,"note":"replication cell: the gold effect to rel 1e-6 (abs floor 1e-6\u00b7SE); never required","open_interval":false,"rel_tol":1e-06,"required":false,"source_id":"crb_papers.q012.gold","table":null,"tool":null,"value":0.33639171169653814,"weight":1.0,"where":null},{"abs_tol":1.2591763131474552e-07,"bind":"answer_call","column":null,"failure_class":null,"field":"se","interval":false,"key":null,"kind":"answer_figure_close","min_lre":null,"note":"replication cell: the gold SE to rel 1e-6; never required","open_interval":false,"rel_tol":1e-06,"required":false,"source_id":"crb_papers.q012.gold","table":null,"tool":null,"value":0.12591763131474554,"weight":1.0,"where":null}],"datasets":{"ds_0007":{"columns":null,"description":"| Variable Name | Description |\n| --- | --- |\n| (index) | Sequential row number from the CSV export; has no substantive meaning. |\n| fcode | Anonymous manufacturing establishment identifier. Stable across years within this file but masked to prevent reidentification or linkage to external sources. |\n| year | Calendar year of record (2009\u20132012). Reflects the period over which audits, orders, and operational performance were summarized. |\n| compyr | Annual indicator for whether the establishment was deemed compliant with the sourcing agent\u2019s code of conduct in that year (1 = compliant, 0 = not compliant). Compliance is determined from social and environmental audits; values may be missing in years without an audit. |\n| gradeyr | Audit grade for the year mapped to a numeric scale (A = 3, B = 2, C = 1, D = 0). Grades summarize findings from factory audits that assess labor, health and safety, environmental, and legal practices. May be blank in years without an audit. |\n| lestspend_dall | Natural log of the estimated total value of customer orders routed through the sourcing agent to the factory in that calendar year, expressed in thousands of constant 2012 USD. Underlying nominal values were originally reported in annual bins; midpoints were used and deflated to 2012 dollars before logging. |\n| inspyr | Share of quality inspections passed during the year (0\u20131), calculated from the agent\u2019s inspection records for orders it handled at the factory. Represents the fraction of checks that met specified quality criteria. |\n| ontimeyr | On-time delivery rate during the year (0\u20131), computed from order-tracking data for shipments coordinated by the agent, where \u201con time\u201d indicates delivery on or before the scheduled date. |\n| factoryobs_perf | Count of underlying observations used to compute the year\u2019s performance rates (e.g., number of inspections or orders contributing to inspyr and ontimeyr). Larger values imply more stable rate estimates. |\n| factorytt | Panel time index for convenience in longitudinal analyses (1 = 2009, 2 = 2010, 3 = 2011, 4 = 2012). |\n\nDataset Context:\nThis panel assembles administrative records from a multinational sourcing intermediary that connects buyers\u2014primarily retailers and brand owners in advanced economies\u2014with manufacturing suppliers across 36 countries. The intermediary maintains a supply-chain responsibility program with a written code of conduct covering labor conditions, environmental practices, health and safety, and compliance with local law. Factories are periodically audited by trained assessors, typically during one- to two-day site visits that review documents, conduct walkthroughs, and interview personnel. Audits produce letter grades (A\u2013D), with the intermediary treating A/B as meeting its standard and C/D as not meeting it. For each calendar year, the data retain a factory\u2019s grade and a binary indicator derived from these audits; some factories are visited multiple times in a year, while others may go a year without a recorded audit.\n\nThe same intermediary aggregates commercial activity by factory and year from its order-management systems. Because customers provided annual purchasing in value bands, the dataset converts those bands to midpoints, deflates them to constant 2012 U.S. dollars using import price indices for consumer goods, and applies a natural log transformation. Operational performance metrics\u2014on-time delivery and quality inspection pass rates\u2014are calculated from order- and shipment-level tracking for business transacted through the agent; these rates are only observed in years when the intermediary coordinated orders at the factory, and their precision varies with the number of underlying events.\n\nCoverage spans 2009\u20132012 and a mix of product categories with a substantial share of apparel. Factory identities and any corporate names were anonymized under a data access agreement. Users should note common features of this type of administrative data: audit results reflect conditions at the time of inspection rather than continuous monitoring; performance rates pertain to the subset of a factory\u2019s portfolio handled by the intermediary; and annual purchasing values are estimates derived from reported bins rather than exact ledgers. Despite these limits, the dataset offers rare, harmonized, multi-country visibility into auditing outcomes, order flows, and supplier performance within global manufacturing supply chains during the early 2010s.","file_sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","format":"csv","frame_sha256":null,"generator":null,"kind":"file","licence":"redistributed by CausalReasoningBenchmark (MIT) under the original replication data's terms","member":null,"object":null,"params":{},"path":null,"read_kw":{},"records":null,"role":"agent","sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","transforms":[],"url":"https://huggingface.co/datasets/syrgkanislab/CausalReasoningBenchmark/resolve/edd0fe1c1d1c5c0a4f0e3e67a50481ac679fb888/data/ds_0007/data/data.csv"},"ref_panel":{"columns":null,"description":"Firm rows of 2011 and 2012: firms observed in both years, switching 0->1 or staying at 0","file_sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","format":"csv","frame_sha256":null,"generator":null,"kind":"file","licence":"redistributed by CausalReasoningBenchmark (MIT) under the original replication data's terms","member":null,"object":null,"params":{},"path":null,"read_kw":{"low_memory":false},"records":null,"role":"reference","sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","transforms":[{"args":{"columns":["fcode","year","compyr","lestspend_dall"]},"op":"select"},{"args":{"args":{"cmp":"gt","column":"compyr","value":0},"name":"compyr","op":"indicator"},"op":"derive"},{"args":{"args":{"column":"compyr","default":0,"keep_unmapped":false,"mapping":[{"source":1,"target":1}]},"name":"compyr","op":"map"},"op":"derive"},{"args":{"columns":["fcode","year","lestspend_dall"]},"op":"dropna"},{"args":{"query":"year in [2011, 2012]"},"op":"filter"},{"args":{"how":"left","on":["fcode"],"with":"ref_years"},"op":"merge"},{"args":{"how":"left","on":["fcode"],"with":"ref_wpre"},"op":"merge"},{"args":{"how":"left","on":["fcode"],"with":"ref_wpost"},"op":"merge"},{"args":{"query":"nyears == 2 and w2011 == 0 and (w2012 == 1 or w2012 == 0)"},"op":"filter"},{"args":{"args":{"cmp":"eq","column":"w2012","value":1},"name":"treat","op":"indicator"},"op":"derive"},{"args":{"args":{"cmp":"eq","column":"year","value":2012},"name":"post","op":"indicator"},"op":"derive"},{"args":{"args":{"columns":["treat","post"]},"name":"treat_post","op":"interaction"},"op":"derive"}],"url":"https://huggingface.co/datasets/syrgkanislab/CausalReasoningBenchmark/resolve/edd0fe1c1d1c5c0a4f0e3e67a50481ac679fb888/data/ds_0007/data/data.csv"},"ref_wpost":{"columns":null,"description":"Each firm's mean of compyr in 2012","file_sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","format":"csv","frame_sha256":null,"generator":null,"kind":"file","licence":"redistributed by CausalReasoningBenchmark (MIT) under the original replication data's terms","member":null,"object":null,"params":{},"path":null,"read_kw":{"low_memory":false},"records":null,"role":"reference","sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","transforms":[{"args":{"columns":["fcode","year","compyr","lestspend_dall"]},"op":"select"},{"args":{"args":{"cmp":"gt","column":"compyr","value":0},"name":"compyr","op":"indicator"},"op":"derive"},{"args":{"args":{"column":"compyr","default":0,"keep_unmapped":false,"mapping":[{"source":1,"target":1}]},"name":"compyr","op":"map"},"op":"derive"},{"args":{"columns":["fcode","year","lestspend_dall"]},"op":"dropna"},{"args":{"query":"year == 2012"},"op":"filter"},{"args":{"aggs":[{"column":"compyr","fn":"mean","name":"w2012"}],"by":["fcode"]},"op":"collapse"}],"url":"https://huggingface.co/datasets/syrgkanislab/CausalReasoningBenchmark/resolve/edd0fe1c1d1c5c0a4f0e3e67a50481ac679fb888/data/ds_0007/data/data.csv"},"ref_wpre":{"columns":null,"description":"Each firm's mean of compyr in 2011","file_sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","format":"csv","frame_sha256":null,"generator":null,"kind":"file","licence":"redistributed by CausalReasoningBenchmark (MIT) under the original replication data's terms","member":null,"object":null,"params":{},"path":null,"read_kw":{"low_memory":false},"records":null,"role":"reference","sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","transforms":[{"args":{"columns":["fcode","year","compyr","lestspend_dall"]},"op":"select"},{"args":{"args":{"cmp":"gt","column":"compyr","value":0},"name":"compyr","op":"indicator"},"op":"derive"},{"args":{"args":{"column":"compyr","default":0,"keep_unmapped":false,"mapping":[{"source":1,"target":1}]},"name":"compyr","op":"map"},"op":"derive"},{"args":{"columns":["fcode","year","lestspend_dall"]},"op":"dropna"},{"args":{"query":"year == 2011"},"op":"filter"},{"args":{"aggs":[{"column":"compyr","fn":"mean","name":"w2011"}],"by":["fcode"]},"op":"collapse"}],"url":"https://huggingface.co/datasets/syrgkanislab/CausalReasoningBenchmark/resolve/edd0fe1c1d1c5c0a4f0e3e67a50481ac679fb888/data/ds_0007/data/data.csv"},"ref_years":{"columns":null,"description":"How many of 2011 and 2012 each firm is observed in","file_sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","format":"csv","frame_sha256":null,"generator":null,"kind":"file","licence":"redistributed by CausalReasoningBenchmark (MIT) under the original replication data's terms","member":null,"object":null,"params":{},"path":null,"read_kw":{"low_memory":false},"records":null,"role":"reference","sha256":"4bfe0955b959c2258c32c26060e79131a2397e9c8d2e7a2fb405e9d3d14ebb4e","transforms":[{"args":{"columns":["fcode","year","compyr","lestspend_dall"]},"op":"select"},{"args":{"args":{"cmp":"gt","column":"compyr","value":0},"name":"compyr","op":"indicator"},"op":"derive"},{"args":{"args":{"column":"compyr","default":0,"keep_unmapped":false,"mapping":[{"source":1,"target":1}]},"name":"compyr","op":"map"},"op":"derive"},{"args":{"columns":["fcode","year","lestspend_dall"]},"op":"dropna"},{"args":{"query":"year in [2011, 2012]"},"op":"filter"},{"args":{"aggs":[{"column":"lestspend_dall","fn":"count","name":"n"}],"by":["fcode","year"]},"op":"collapse"},{"args":{"aggs":[{"column":"n","fn":"count","name":"nyears"}],"by":["fcode"]},"op":"collapse"}],"url":"https://huggingface.co/datasets/syrgkanislab/CausalReasoningBenchmark/resolve/edd0fe1c1d1c5c0a4f0e3e67a50481ac679fb888/data/ds_0007/data/data.csv"}},"difficulty":"M","id":"crb_papers.q012","level":"T2","meta":{"agent_reach":{"blocked_by":["reference_frame:ref_panel","reference_frame:ref_wpost","reference_frame:ref_wpre","reference_frame:ref_years"],"dtypes":{"ds_0007":{"compyr":"float64","lestspend_dall":"float64"}},"named":{"outcome":["lestspend_dall"],"treatment":["compyr"]},"reachable":false,"rule":"magi_bench.reach/1"},"answer_format":{"figure_transform":"identity","round_to":null,"unit_scale":1.0},"cluster":"ds_0007","crb":{"data_path":"data/ds_0007/data/data.csv","dataset_id":"ds_0007","gold":{"ci95":[0.08959315431963688,0.5831902690734394],"ci_rule":"effect \u00b1 1.96\u00b7|standard_error| (evaluate_result.ci_bounds)","effect":0.33639171169653814,"standard_error":0.12591763131474554},"identification":{"bad_controls":["gradeyr","inspyr","ontimeyr","factoryobs_perf"],"causal_quantity":"ATT","group_variable":"fcode","minimal_controlling_set":[],"outcomes":["lestspend_dall"],"reason_for_bad_controls":"gradeyr is the audit letter grade (A\u2013D) that defines the binary treatment compyr (A/B compliant vs C/D noncompliant), so it is a post-treatment measure and a mechanical function/descendant of the treatment. Controlling for it would amount to conditioning on the treatment or its mediator, inducing over-control bias. The paper states: \u201cFactories rated A and B \u2026 are considered compliant\u2026 Our analysis uses both the original 4-point grading system and this binary distinction\u201d (p. 700). inspyr (quality inspection pass rate) and ontimeyr (on-time delivery) are contemporaneous performance outcomes that the authors sometimes control for, but they plausibly lie on the causal path from compliance to orders; including them blocks part of the total effect. The paper motivates them as \u2018manufacturing performance\u2019 controls and shows results with and without them (eq. 2; p. 702\u2013703). As per the guidance on bad controls, conditioning on post-treatment mediators leads to over-control bias (Applied Causal Inference, \u00a711.3 Post-Treatment Variables: \u201cexplicitly adjusting for post-treatment variables is almost always a bad idea,\u201d and Figures 11.9a\u2013b). factoryobs_perf is \u201cCount of observations used to compute performance metrics in the year (e.g., orders/inspections),\u201d which is downstream of orders (the outcome) and inspections, so conditioning on it risks collider bias or conditioning on an outcome of the outcome (see Figure 11.12c in the guidance).","reason_for_minimal_controlling_set":"A two-way fixed-effects DiD with factory and year fixed effects is the core identification, and the authors explicitly estimate models without time-varying covariates and find similar effects when adding performance controls. Quote: \u201cwe use the panel structure of the data to generate within-factories estimates of the effect of changes in exporter compliance status on importer purchasing behavior. The effect of compliance on orders is estimated by fitting a standard two-way fixed effects regression model\u201d (p. 702). They also report: \u201cAdding controls for manufacturing performance has almost no impact on effect magnitudes\u201d (p. 703). Thus, beyond unit and time fixed effects (encoded via group_variable and time_variable), no additional covariates are minimally required for identification under parallel trends.","strategy":"Difference-in-Differences","time_variable":"year/factorytt","treatments":["compyr"]},"identification_strategy":"Difference-in-Differences"},"headline_keys":{"infer.ols":["coef.treat_post"]},"naive":{"dataset":"ds_0007","treatment":"compyr","y":"lestspend_dall"},"native_id":"12","reference":{"cite":1,"note":"Reference path transcribed from the item's own estimation script.","steps":[{"args":{"cluster":"fcode","controls":["treat","post"],"dataset":"ref_panel","x":["treat_post"],"y":"lestspend_dall"},"tool":"infer.ols","why":"smf.ols('lestspend_dall ~ treat*post'), cluster by fcode"}]},"reference_path":{"calls":[{"args":{"cluster":"fcode","controls":["treat","post"],"dataset":"ref_panel","x":["treat_post"],"y":"lestspend_dall"},"tool":"infer.ols"}],"needs":[],"provenance":"benchmark_script","quotes":{},"reference_datasets":["ref_panel","ref_wpost","ref_wpre","ref_years"],"source_id":"crb_papers.q012.script"}},"question":"From 2011 to 2012, how did factories that became compliant (per the sourcing agent's annual compliance indicator) change in logged annual order value (natural log of estimated annual order value in constant 2012 USD) relative to factories that remained noncompliant over the same period?","requires":[],"runner":null,"sources":["crb_papers.q012.gold","crb_papers.q012.script"],"split":"public","suite":"crb_papers","tags":["econ-bench","external","crb","did","ref_benchmark_script"],"target":"agent","tool":null,"truth":"CausalReasoningBenchmark gold at the pinned revision: causal_queries.json effect and standard_error (the value the item's estimation.py prints), scored as evaluate_result.py scores estimate_within_solution_ci"}