1{2 "validation_errors": [],3 "doc_id": "jof_12492",4 "ontology_version": "1.0",5 "prompt_version": "1.0",6 "model": "claude-opus-5-5",7 "effort": "medium",8 "annotated_at": "2026-11-03T22:57:01+00:00",9 "source_sha256": "90490da86b17d09e74b3f92843e603754efcafb27d151e2026d42411ac64daeb",10 "parser": "pymupdf-jof-heuristics-1.3",11 "payload_chars": 27298,12 "usage": {13 "input": 9001,14 "output": 2564,15 "cache_read": 3968,16 "cache_write": 017 },18 "cost_usd": 0.088078,19 "request_id": "req_011CggpZDKXBkaGYQ6qHe2x2",20 "annotation": {21 "genre": "empirical",22 "subfield": "corporate_finance",23 "subfield_secondary": "financial_intermediation_banking; small business credit / government loan guarantee evaluation",24 "primary_method": "instrumental_variables",25 "identification": {26 "estimand": "Average effect on the treated of SBA loan amount (per $1 million) on firm employment growth, measured as post-loan minus pre-loan average employment, for growing small firms",27 "strategy": "Longitudinal matching (exact matching plus kernel propensity scores on employment history) combined with treatment-control group fixed-effects regressions. IVs built from the lagged local presence or share of branches of Preferred Lender Program banks, with PLP activity defined outside the local market.",28 "key_assumptions": [29 "Bank entry into a county is not based on potential growth of SBA borrowers there",30 "A bank's choice to become a PLP lender is not based on local SBA-recipient growth potential",31 "For the branch-share IV, PLP banks do not gain local market share based on SBA-recipient growth potential",32 "Small businesses rely on local lenders",33 "For matching/OLS: no correlated time-varying unobservables"34 ],35 "threats_addressed": [36 "Selection into loan receipt (positive or negative)",37 "Pre-loan growth trends of recipients",38 "Correlation of SBA availability with local financial development and growth",39 "Firm exit/survivorship",40 "Displacement of competing firms (nontradables)",41 "Time-varying idiosyncratic demand, productivity or cost shocks"42 ],43 "causal_language_strength": 3,44 "uses_causal_claim": true45 },46 "abstract_moves": {47 "question": {48 "present": true,49 "sentence_index": 050 },51 "gap": {52 "present": false,53 "sentence_index": -154 },55 "method": {56 "present": true,57 "sentence_index": 158 },59 "results": {60 "present": true,61 "sentence_index": 262 },63 "magnitude": {64 "present": true,65 "sentence_index": 266 },67 "contribution": {68 "present": false,69 "sentence_index": -170 }71 },72 "introduction": {73 "moves": [74 {75 "paragraph_index": 0,76 "move": "motivation"77 },78 {79 "paragraph_index": 5,80 "move": "motivation"81 },82 {83 "paragraph_index": 5,84 "move": "literature_gap"85 },86 {87 "paragraph_index": 5,88 "move": "tension_or_puzzle"89 },90 {91 "paragraph_index": 6,92 "move": "contribution_statement"93 },94 {95 "paragraph_index": 6,96 "move": "preview_of_approach"97 },98 {99 "paragraph_index": 7,100 "move": "preview_of_approach"101 },102 {103 "paragraph_index": 8,104 "move": "identification_defense"105 },106 {107 "paragraph_index": 9,108 "move": "related_literature_positioning"109 },110 {111 "paragraph_index": 10,112 "move": "related_literature_positioning"113 },114 {115 "paragraph_index": 11,116 "move": "other"117 }118 ],119 "contribution_paragraph_index": 6,120 "contribution_types": [121 "new_evidence",122 "new_data_or_measure",123 "new_identification"124 ],125 "roadmap_present": false,126 "literature_positioning_location": "inside_introduction"127 },128 "section_roles": [129 {130 "idx": 0,131 "role": "introduction"132 },133 {134 "idx": 1,135 "role": "other"136 },137 {138 "idx": 2,139 "role": "data"140 },141 {142 "idx": 3,143 "role": "results"144 },145 {146 "idx": 4,147 "role": "results"148 },149 {150 "idx": 5,151 "role": "results"152 },153 {154 "idx": 6,155 "role": "conclusion"156 }157 ],158 "results_presentation": {159 "economic_significance_discussed": true,160 "statistical_significance_language": "unclear",161 "magnitude_statements_in_intro": false162 },163 "robustness": {164 "location": "mixed",165 "types": [166 "alternative_measure",167 "alternative_specification",168 "subsample",169 "other"170 ]171 },172 "tables": {173 "convention_notes": "Table notes define the dependent variable as the change in average employment, t+1..t+3 versus t-3..t-1. They state that loan amount is in millions of 2010 dollars, list the county-level controls, and distinguish OLS from IV and scaled from unscaled specifications. Summary-statistics tables come first, then OLS, IV, alternative-IV and heterogeneity tables.",174 "units_stated_in_notes": true,175 "standard_errors_description": "unclear"176 },177 "figures": {178 "uncertainty_shown": "unclear"179 },180 "conclusion": {181 "bounded_claims": true,182 "policy_implications": true,183 "open_questions": false,184 "restates_contribution": true185 },186 "hedging": {187 "level": "moderate",188 "typical_devices": [189 "explicit caveats paragraph on non-random allocation and external validity",190 "'arguably exogenous' framing of instruments",191 "'if these estimates are correct' conditional",192 "lower-bound interpretation",193 "'preliminary investigation' and 'rough estimates' qualifiers",194 "'suggesting' when linking results to credit constraints"195 ]196 },197 "claim_evidence": [198 {199 "claim": "Each $1 million of SBA loans raises recipient employment by about 3–3.5 jobs over three post-loan years",200 "evidence_type": "table",201 "evidence_ref": "Tables IV, VI, VII"202 },203 {204 "claim": "Effects are larger for younger and larger firms",205 "evidence_type": "table",206 "evidence_ref": "Table VIII"207 },208 {209 "claim": "No systematic difference between 7(a) and 504 programs or over the business cycle",210 "evidence_type": "table",211 "evidence_ref": "Tables IX, X"212 },213 {214 "claim": "Effects are smaller where local conventional credit is more plentiful",215 "evidence_type": "table",216 "evidence_ref": "Table XI"217 },218 {219 "claim": "Displacement of competing firms is around 10% of the total effect in nontradables",220 "evidence_type": "table",221 "evidence_ref": "Table XII"222 },223 {224 "claim": "Results are similar when exited firms are assigned zero employment",225 "evidence_type": "table",226 "evidence_ref": "Table XIII"227 },228 {229 "claim": "Taxpayer cost per job is roughly $21,000–$25,000, low relative to other government programs",230 "evidence_type": "prior_literature",231 "evidence_ref": "Section V.B; Neumark (2013) comparison in conclusion"232 },233 {234 "claim": "PLP lenders account for a large share of SBA loans, especially larger loans",235 "evidence_type": "figure",236 "evidence_ref": "Figure 1"237 },238 {239 "claim": "Only 3.2% of SBA loans occur in counties with no PLP branch in the prior year",240 "evidence_type": "descriptive_statistic",241 "evidence_ref": "Introduction"242 }243 ],244 "structural_summary": "The paper is a policy-evaluation empirical study. The introduction opens with a historical epigraph and the size of the SBA program, then notes that the employment effect is theoretically ambiguous. It previews linked administrative-census data and the matching method, and motivates instruments from lender institutional features while stating the identifying assumptions. Literature positioning follows, then an explicit caveats paragraph on external validity. The body describes the program institutions and the data, presents matching/OLS results, then the IV approach, and finally heterogeneity, displacement, exit and cost-per-job extensions. The conclusion restates the estimates, bounds them as a lower bound, and draws a cost-effectiveness policy implication.",245 "writing_profile_keywords": [246 "program evaluation",247 "linked administrative data",248 "matching estimator",249 "instrumental variables",250 "explicit caveats",251 "cost per job",252 "heterogeneous effects",253 "policy relevance"254 ],255 "confidence": 0.8,256 "notes": "The introduction-paragraph indices are fragmented: paragraphs 0–4 are pieces of the epigraph, and paragraph 5 contains the opening motivation, theoretical ambiguity and empirical difficulty. Section IV is labeled 'data' by the role rule but is actually IV methods and results. Inference reporting conventions (stars or SEs) are not visible in the excerpts. Robustness checks (alternative IVs, exit zeros, controls) appear within the results and extensions sections, and an internet appendix exists."257 },258 "deterministic_checks": {259 "roadmap_regex": false,260 "contribution_regex_in_intro": true,261 "magnitude_in_abstract": false,262 "abstract_n_sentences": 6,263 "intro_n_paragraphs": 12,264 "intro_n_words": 1457,265 "causal_words_in_abstract": 0,266 "hedge_words_in_abstract": 0,267 "stars_mentioned": false,268 "clustered_se_mentioned": false,269 "internet_appendix_mentioned": false,270 "section_roles_rule": {271 "0": "introduction",272 "1": "other",273 "2": "data",274 "3": "results",275 "4": "data",276 "5": "robustness",277 "6": "conclusion"278 }279 },280 "schema_valid": true,281 "code_version": "feb6c9e-dirty"282}