Dr. QuantMaster MCP Server

import { Tool } from "@modelcontextprotocol/sdk/types.js"; import { searchKnowledgeBase } from "../db/embeddings.js"; import { erf, calculatePower, normalCDF } from "../utils/math.js"; // ============================================================ // Dr. QuantMaster MCP Tools (45개) // ============================================================ export const tools: Tool[] = [ // === CATEGORY 1: Knowledge Search (5 tools) === { name: "search_stats_knowledge", description: "통계/계량경제학 지식베이스 RAG 검색. 방법론, 가정, 해석 가이드 제공", inputSchema: { type: "object", properties: { query: { type: "string", description: "검색 쿼리" }, category: { type: "string", enum: ["foundations", "regression", "econometrics", "advanced", "meta", "all"], description: "검색 카테고리" }, n_results: { type: "number", description: "결과 수 (기본: 5)" }, }, required: ["query"], }, }, { name: "get_method_guide", description: "특정 통계 방법의 상세 가이드 (가정, 절차, 해석, 보고)", inputSchema: { type: "object", properties: { method: { type: "string", description: "방법론 (예: ols, panel_fe, did, sem, meta)" }, aspect: { type: "string", enum: ["assumptions", "procedure", "interpretation", "reporting", "all"], description: "가이드 측면" }, }, required: ["method"], }, }, { name: "suggest_method", description: "연구질문과 데이터 특성에 맞는 통계 방법 추천", inputSchema: { type: "object", properties: { research_question: { type: "string", description: "연구 질문" }, dv_type: { type: "string", enum: ["continuous", "binary", "ordinal", "count", "time_to_event", "proportion"], description: "종속변수 유형" }, data_structure: { type: "string", enum: ["cross_sectional", "panel", "time_series", "clustered", "multilevel"], description: "데이터 구조" }, causal_design: { type: "string", enum: ["none", "experimental", "quasi_experimental", "observational"], description: "인과추론 설계" }, }, required: ["research_question", "dv_type"], }, }, { name: "compare_methods", description: "여러 통계 방법 비교 (장단점, 적용조건)", inputSchema: { type: "object", properties: { methods: { type: "array", items: { type: "string" }, description: "비교할 방법들" }, criteria: { type: "array", items: { type: "string" }, description: "비교 기준 (예: assumptions, efficiency, robustness)" }, }, required: ["methods"], }, }, { name: "get_formula", description: "통계 수식 및 LaTeX 표기법 제공", inputSchema: { type: "object", properties: { concept: { type: "string", description: "개념 (예: ols_estimator, did_att, hausman_test)" }, format: { type: "string", enum: ["latex", "text", "both"], description: "출력 형식" }, }, required: ["concept"], }, }, // === CATEGORY 2: Sample Size & Power (5 tools) === { name: "calc_sample_size", description: "필요 표본크기 계산 (t-test, ANOVA, regression, proportion)", inputSchema: { type: "object", properties: { test_type: { type: "string", enum: ["t_test_two", "t_test_paired", "anova", "regression", "proportion", "chi_square"], description: "검정 유형" }, effect_size: { type: "number", description: "효과크기 (Cohen's d, f, f²)" }, alpha: { type: "number", description: "유의수준 (기본: 0.05)" }, power: { type: "number", description: "검정력 (기본: 0.80)" }, groups: { type: "number", description: "집단 수 (ANOVA)" }, predictors: { type: "number", description: "예측변수 수 (회귀)" }, }, required: ["test_type", "effect_size"], }, }, { name: "calc_power", description: "주어진 표본크기에서 검정력 계산", inputSchema: { type: "object", properties: { test_type: { type: "string", description: "검정 유형" }, n: { type: "number", description: "표본크기" }, effect_size: { type: "number", description: "효과크기" }, alpha: { type: "number", description: "유의수준" }, }, required: ["test_type", "n", "effect_size"], }, }, { name: "calc_effect_size", description: "효과크기 계산 및 해석 (Cohen's d, η², f², OR, RR)", inputSchema: { type: "object", properties: { type: { type: "string", enum: ["cohens_d", "eta_squared", "f_squared", "odds_ratio", "correlation"], description: "효과크기 유형" }, values: { type: "object", description: "계산에 필요한 값들" }, }, required: ["type", "values"], }, }, { name: "mde_calculator", description: "최소탐지효과크기(MDE) 계산 - A/B 테스트, 실험설계", inputSchema: { type: "object", properties: { n_per_group: { type: "number", description: "그룹당 표본크기" }, baseline: { type: "number", description: "기준값 (비율 또는 평균)" }, alpha: { type: "number", description: "유의수준" }, power: { type: "number", description: "검정력" }, test_type: { type: "string", enum: ["proportion", "mean"], description: "검정 유형" }, }, required: ["n_per_group", "baseline"], }, }, { name: "power_curve", description: "검정력 곡선 데이터 생성 (시각화용)", inputSchema: { type: "object", properties: { test_type: { type: "string", description: "검정 유형" }, n_range: { type: "array", items: { type: "number" }, description: "[최소, 최대] 표본크기" }, effect_sizes: { type: "array", items: { type: "number" }, description: "효과크기 배열" }, alpha: { type: "number", description: "유의수준" }, }, required: ["test_type", "n_range", "effect_sizes"], }, }, // === CATEGORY 3: Diagnostics & Assumptions (5 tools) === { name: "check_assumptions", description: "통계 가정 점검 가이드 (정규성, 등분산, 독립성 등)", inputSchema: { type: "object", properties: { method: { type: "string", description: "분석 방법" }, data_description: { type: "string", description: "데이터 설명" }, }, required: ["method"], }, }, { name: "diagnose_regression", description: "회귀분석 진단 가이드 (다중공선성, 이분산, 자기상관 등)", inputSchema: { type: "object", properties: { issues: { type: "array", items: { type: "string" }, description: "의심되는 문제 (multicollinearity, heteroscedasticity, autocorrelation, outliers)" }, model_type: { type: "string", description: "모형 유형" }, }, required: ["issues"], }, }, { name: "test_selection", description: "적절한 통계 검정 선택 가이드", inputSchema: { type: "object", properties: { purpose: { type: "string", enum: ["comparison", "relationship", "prediction", "model_fit"], description: "분석 목적" }, variable_types: { type: "object", description: "변수 유형 정보" }, sample_characteristics: { type: "object", description: "표본 특성" }, }, required: ["purpose", "variable_types"], }, }, { name: "interpret_diagnostics", description: "진단 결과 해석 (VIF, DW, BP test 등)", inputSchema: { type: "object", properties: { test_name: { type: "string", description: "진단 검정명" }, statistic: { type: "number", description: "검정 통계량" }, p_value: { type: "number", description: "p-value" }, context: { type: "string", description: "분석 맥락" }, }, required: ["test_name", "statistic"], }, }, { name: "robustness_guide", description: "강건성 검정 가이드 (대안모형, 민감도분석)", inputSchema: { type: "object", properties: { main_analysis: { type: "string", description: "주 분석 방법" }, concerns: { type: "array", items: { type: "string" }, description: "우려사항" }, }, required: ["main_analysis"], }, }, // === CATEGORY 4: Causal Inference (6 tools) === { name: "causal_design_guide", description: "인과추론 설계 가이드 (DID, RDD, IV, PSM, Synth)", inputSchema: { type: "object", properties: { design: { type: "string", enum: ["did", "rdd", "iv", "psm", "synthetic_control", "event_study"], description: "인과추론 설계" }, aspect: { type: "string", enum: ["identification", "assumptions", "implementation", "threats", "all"], description: "가이드 측면" }, }, required: ["design"], }, }, { name: "parallel_trends_check", description: "DID 평행추세 검정 가이드", inputSchema: { type: "object", properties: { pre_periods: { type: "number", description: "사전 기간 수" }, approach: { type: "string", enum: ["visual", "statistical", "placebo"], description: "검정 방법" }, }, required: ["pre_periods"], }, }, { name: "iv_strength_check", description: "도구변수 강도/유효성 검정 가이드", inputSchema: { type: "object", properties: { n_instruments: { type: "number", description: "도구변수 수" }, first_stage_f: { type: "number", description: "1단계 F-통계량" }, overid_test: { type: "boolean", description: "과대식별 검정 필요 여부" }, }, required: ["n_instruments"], }, }, { name: "psm_guide", description: "성향점수매칭 가이드 (추정, 매칭, 균형검정)", inputSchema: { type: "object", properties: { matching_method: { type: "string", enum: ["nearest", "caliper", "kernel", "full"], description: "매칭 방법" }, balance_check: { type: "boolean", description: "균형검정 포함" }, }, required: ["matching_method"], }, }, { name: "rdd_bandwidth", description: "RDD 대역폭 선택 가이드", inputSchema: { type: "object", properties: { design: { type: "string", enum: ["sharp", "fuzzy"], description: "RDD 유형" }, method: { type: "string", enum: ["ik", "cct", "cv"], description: "대역폭 선택 방법" }, }, required: ["design"], }, }, { name: "event_study_guide", description: "사건연구(Event Study) 설계 가이드", inputSchema: { type: "object", properties: { event_type: { type: "string", description: "사건 유형" }, staggered: { type: "boolean", description: "시차적 처치 여부" }, estimator: { type: "string", enum: ["twfe", "cs", "sa", "did_m"], description: "추정량" }, }, required: ["event_type"], }, }, // === CATEGORY 5: Code Generation (8 tools) === { name: "generate_r_code", description: "R 코드 생성 (분석, 진단, 시각화)", inputSchema: { type: "object", properties: { analysis_type: { type: "string", description: "분석 유형" }, variables: { type: "object", description: "변수 정보" }, options: { type: "object", description: "추가 옵션" }, }, required: ["analysis_type"], }, }, { name: "generate_stata_code", description: "Stata 코드 생성 (분석, 진단, 시각화)", inputSchema: { type: "object", properties: { analysis_type: { type: "string", description: "분석 유형" }, variables: { type: "object", description: "변수 정보" }, options: { type: "object", description: "추가 옵션" }, }, required: ["analysis_type"], }, }, { name: "generate_python_code", description: "Python 코드 생성 (statsmodels, sklearn)", inputSchema: { type: "object", properties: { analysis_type: { type: "string", description: "분석 유형" }, library: { type: "string", enum: ["statsmodels", "sklearn", "linearmodels"], description: "라이브러리" }, variables: { type: "object", description: "변수 정보" }, }, required: ["analysis_type"], }, }, { name: "code_template", description: "분석 전체 워크플로우 코드 템플릿", inputSchema: { type: "object", properties: { workflow: { type: "string", enum: ["regression_full", "panel_full", "did_full", "meta_full", "sem_full"], description: "워크플로우 유형" }, language: { type: "string", enum: ["r", "stata", "python"], description: "언어" }, }, required: ["workflow", "language"], }, }, { name: "visualization_code", description: "시각화 코드 생성 (ggplot2, matplotlib, Stata graphs)", inputSchema: { type: "object", properties: { chart_type: { type: "string", enum: ["coefficient_plot", "marginal_effects", "residual_plot", "power_curve", "forest_plot"], description: "차트 유형" }, language: { type: "string", enum: ["r", "stata", "python"], description: "언어" }, customization: { type: "object", description: "커스터마이징 옵션" }, }, required: ["chart_type", "language"], }, }, { name: "table_code", description: "결과표 생성 코드 (stargazer, esttab, pandas)", inputSchema: { type: "object", properties: { table_type: { type: "string", enum: ["descriptive", "regression", "correlation", "balance"], description: "표 유형" }, format: { type: "string", enum: ["latex", "html", "word", "markdown"], description: "출력 형식" }, language: { type: "string", enum: ["r", "stata", "python"], description: "언어" }, }, required: ["table_type", "format", "language"], }, }, { name: "debug_code", description: "통계 코드 디버깅 도움", inputSchema: { type: "object", properties: { code: { type: "string", description: "문제 코드" }, error_message: { type: "string", description: "에러 메시지" }, language: { type: "string", enum: ["r", "stata", "python"], description: "언어" }, }, required: ["code", "error_message", "language"], }, }, { name: "optimize_code", description: "대용량 데이터용 코드 최적화 제안", inputSchema: { type: "object", properties: { code: { type: "string", description: "최적화할 코드" }, data_size: { type: "string", description: "데이터 크기 (행 수, GB)" }, language: { type: "string", enum: ["r", "stata", "python"], description: "언어" }, }, required: ["code", "language"], }, }, // === CATEGORY 6: Results Interpretation (5 tools) === { name: "interpret_coefficient", description: "회귀계수 해석 가이드 (OLS, Logit, 상호작용 등)", inputSchema: { type: "object", properties: { model_type: { type: "string", description: "모형 유형" }, coefficient: { type: "number", description: "계수값" }, se: { type: "number", description: "표준오차" }, variable_type: { type: "string", enum: ["continuous", "binary", "categorical", "interaction"], description: "변수 유형" }, transformation: { type: "string", enum: ["none", "log", "standardized"], description: "변환" }, }, required: ["model_type", "coefficient"], }, }, { name: "interpret_model_fit", description: "모형적합도 해석 (R², AIC, BIC, Pseudo-R²)", inputSchema: { type: "object", properties: { metrics: { type: "object", description: "적합도 지표들" }, model_type: { type: "string", description: "모형 유형" }, comparison: { type: "boolean", description: "모형 비교 여부" }, }, required: ["metrics", "model_type"], }, }, { name: "marginal_effects_guide", description: "한계효과 계산 및 해석 가이드", inputSchema: { type: "object", properties: { model_type: { type: "string", enum: ["logit", "probit", "poisson", "tobit"], description: "모형" }, effect_type: { type: "string", enum: ["ame", "mem", "mer"], description: "한계효과 유형" }, interaction: { type: "boolean", description: "상호작용항 포함" }, }, required: ["model_type"], }, }, { name: "interpret_test", description: "통계 검정 결과 해석", inputSchema: { type: "object", properties: { test_name: { type: "string", description: "검정명" }, statistic: { type: "number", description: "검정통계량" }, p_value: { type: "number", description: "p-value" }, df: { type: "number", description: "자유도" }, context: { type: "string", description: "맥락" }, }, required: ["test_name", "p_value"], }, }, { name: "write_results_section", description: "결과 섹션 작성 템플릿 생성", inputSchema: { type: "object", properties: { analysis_type: { type: "string", description: "분석 유형" }, results: { type: "object", description: "주요 결과" }, style: { type: "string", enum: ["apa", "asa", "econometrica"], description: "스타일" }, }, required: ["analysis_type", "results"], }, }, // === CATEGORY 7: Meta-Analysis (4 tools) === { name: "meta_effect_size", description: "메타분석용 효과크기 계산 및 변환", inputSchema: { type: "object", properties: { input_type: { type: "string", enum: ["means_sd", "t_value", "f_value", "correlation", "odds_ratio", "proportion"], description: "입력 유형" }, values: { type: "object", description: "입력값" }, target_metric: { type: "string", description: "변환 목표 지표" }, }, required: ["input_type", "values"], }, }, { name: "meta_heterogeneity", description: "메타분석 이질성 해석 (Q, I², τ²)", inputSchema: { type: "object", properties: { q_stat: { type: "number", description: "Q 통계량" }, i_squared: { type: "number", description: "I² 값" }, tau_squared: { type: "number", description: "τ² 값" }, k: { type: "number", description: "연구 수" }, }, required: ["i_squared", "k"], }, }, { name: "publication_bias", description: "출판편향 검정 가이드", inputSchema: { type: "object", properties: { methods: { type: "array", items: { type: "string" }, description: "검정 방법 (funnel_plot, egger, trim_fill, selection_model)" }, k: { type: "number", description: "연구 수" }, }, required: ["methods"], }, }, { name: "meta_code", description: "메타분석 코드 생성 (metafor, meta, metan)", inputSchema: { type: "object", properties: { analysis: { type: "string", enum: ["basic_ma", "meta_regression", "subgroup", "sensitivity"], description: "분석 유형" }, effect_measure: { type: "string", description: "효과크기 지표" }, language: { type: "string", enum: ["r", "stata"], description: "언어" }, }, required: ["analysis", "language"], }, }, // === CATEGORY 8: Reporting & Publication (5 tools) === { name: "journal_guide", description: "저널별 통계 보고 가이드", inputSchema: { type: "object", properties: { journal: { type: "string", enum: ["econometrica", "aer", "jfe", "ms", "smj", "amj", "jf", "rfs"], description: "목표 저널" }, method: { type: "string", description: "사용 방법론" }, }, required: ["journal"], }, }, { name: "apa_reporting", description: "APA 스타일 통계 보고 템플릿", inputSchema: { type: "object", properties: { test_type: { type: "string", description: "검정 유형" }, results: { type: "object", description: "결과 값들" }, }, required: ["test_type", "results"], }, }, { name: "prereg_template", description: "사전등록 템플릿 생성 (OSF, AsPredicted)", inputSchema: { type: "object", properties: { platform: { type: "string", enum: ["osf", "aspredicted"], description: "플랫폼" }, study_type: { type: "string", enum: ["experiment", "observational", "replication"], description: "연구 유형" }, hypotheses: { type: "array", items: { type: "string" }, description: "가설 목록" }, }, required: ["platform", "study_type"], }, }, { name: "replication_package", description: "재현성 패키지 구조 가이드", inputSchema: { type: "object", properties: { journal: { type: "string", description: "목표 저널" }, components: { type: "array", items: { type: "string" }, description: "포함 요소 (code, data, readme, codebook)" }, }, required: ["components"], }, }, { name: "reviewer_response", description: "리뷰어 통계 지적 대응 가이드", inputSchema: { type: "object", properties: { critique_type: { type: "string", enum: ["endogeneity", "selection", "measurement", "robustness", "sample_size", "identification"], description: "지적 유형" }, current_method: { type: "string", description: "현재 사용 방법" }, data_constraints: { type: "string", description: "데이터 제약" }, }, required: ["critique_type", "current_method"], }, }, // === CATEGORY 9: Advanced Methods (5 tools) === { name: "sem_guide", description: "구조방정식모형 가이드 (측정모형, 구조모형)", inputSchema: { type: "object", properties: { model_type: { type: "string", enum: ["cfa", "path", "full_sem", "multigroup"], description: "모형 유형" }, fit_indices: { type: "object", description: "적합도 지수" }, }, required: ["model_type"], }, }, { name: "mlm_guide", description: "다층모형 가이드 (ICC, 랜덤효과, 교차수준 상호작용)", inputSchema: { type: "object", properties: { levels: { type: "number", description: "수준 수 (2 or 3)" }, random_effects: { type: "array", items: { type: "string" }, description: "랜덤효과" }, cross_level: { type: "boolean", description: "교차수준 상호작용" }, }, required: ["levels"], }, }, { name: "bayesian_guide", description: "베이지안 분석 가이드 (사전분포, 수렴진단, 해석)", inputSchema: { type: "object", properties: { analysis_type: { type: "string", description: "분석 유형" }, prior_type: { type: "string", enum: ["uninformative", "weakly_informative", "informative"], description: "사전분포" }, convergence: { type: "boolean", description: "수렴진단 포함" }, }, required: ["analysis_type"], }, }, { name: "ml_for_causal", description: "인과추론용 ML 가이드 (Double ML, Causal Forest)", inputSchema: { type: "object", properties: { method: { type: "string", enum: ["double_ml", "causal_forest", "lasso_iv", "honest_tree"], description: "방법" }, target: { type: "string", enum: ["ate", "att", "cate", "late"], description: "추정 대상" }, }, required: ["method"], }, }, { name: "timeseries_guide", description: "시계열 분석 가이드 (ARIMA, VAR, 공적분)", inputSchema: { type: "object", properties: { analysis: { type: "string", enum: ["arima", "var", "vecm", "garch", "state_space"], description: "분석 유형" }, stationarity: { type: "boolean", description: "정상성 검정 포함" }, }, required: ["analysis"], }, }, // === CATEGORY 10: File Operations (2 tools) === { name: "write_analysis_file", description: "분석 코드/결과를 파일로 저장", inputSchema: { type: "object", properties: { content: { type: "string", description: "파일 내용" }, filename: { type: "string", description: "파일명" }, directory: { type: "string", description: "저장 디렉토리" }, encoding: { type: "string", description: "인코딩 (기본: utf-8)" }, }, required: ["content", "filename"], }, }, { name: "create_project_structure", description: "연구 프로젝트 폴더 구조 생성", inputSchema: { type: "object", properties: { project_name: { type: "string", description: "프로젝트명" }, base_path: { type: "string", description: "기본 경로" }, template: { type: "string", enum: ["basic", "replication", "full"], description: "템플릿" }, }, required: ["project_name", "base_path"], }, }, ]; // ============================================================ // Tool Handler Implementation // ============================================================ export async function handleToolCall( name: string, args: Record<string, unknown> ): Promise<unknown> { switch (name) { // Knowledge Search case "search_stats_knowledge": return await handleSearchKnowledge(args); case "get_method_guide": return await handleGetMethodGuide(args); case "suggest_method": return await handleSuggestMethod(args); case "compare_methods": return handleCompareMethods(args); case "get_formula": return handleGetFormula(args); // Sample Size & Power case "calc_sample_size": return handleCalcSampleSize(args); case "calc_power": return handleCalcPower(args); case "calc_effect_size": return handleCalcEffectSize(args); case "mde_calculator": return handleMdeCalculator(args); case "power_curve": return handlePowerCurve(args); // Diagnostics case "check_assumptions": return handleCheckAssumptions(args); case "diagnose_regression": return handleDiagnoseRegression(args); case "test_selection": return handleTestSelection(args); case "interpret_diagnostics": return handleInterpretDiagnostics(args); case "robustness_guide": return handleRobustnessGuide(args); // Causal Inference case "causal_design_guide": return handleCausalDesignGuide(args); case "parallel_trends_check": return handleParallelTrendsCheck(args); case "iv_strength_check": return handleIvStrengthCheck(args); case "psm_guide": return handlePsmGuide(args); case "rdd_bandwidth": return handleRddBandwidth(args); case "event_study_guide": return handleEventStudyGuide(args); // Code Generation case "generate_r_code": return handleGenerateRCode(args); case "generate_stata_code": return handleGenerateStataCode(args); case "generate_python_code": return handleGeneratePythonCode(args); case "code_template": return handleCodeTemplate(args); case "visualization_code": return handleVisualizationCode(args); case "table_code": return handleTableCode(args); case "debug_code": return handleDebugCode(args); case "optimize_code": return handleOptimizeCode(args); // Results Interpretation case "interpret_coefficient": return handleInterpretCoefficient(args); case "interpret_model_fit": return handleInterpretModelFit(args); case "marginal_effects_guide": return handleMarginalEffectsGuide(args); case "interpret_test": return handleInterpretTest(args); case "write_results_section": return handleWriteResultsSection(args); // Meta-Analysis case "meta_effect_size": return handleMetaEffectSize(args); case "meta_heterogeneity": return handleMetaHeterogeneity(args); case "publication_bias": return handlePublicationBias(args); case "meta_code": return handleMetaCode(args); // Reporting case "journal_guide": return handleJournalGuide(args); case "apa_reporting": return handleApaReporting(args); case "prereg_template": return handlePreregTemplate(args); case "replication_package": return handleReplicationPackage(args); case "reviewer_response": return handleReviewerResponse(args); // Advanced Methods case "sem_guide": return handleSemGuide(args); case "mlm_guide": return handleMlmGuide(args); case "bayesian_guide": return handleBayesianGuide(args); case "ml_for_causal": return handleMlForCausal(args); case "timeseries_guide": return handleTimeseriesGuide(args); // File Operations case "write_analysis_file": return handleWriteAnalysisFile(args); case "create_project_structure": return handleCreateProjectStructure(args); default: throw new Error(`Unknown tool: ${name}`); } } // ============================================================ // Handler Implementations (Core Examples) // ============================================================ async function handleSearchKnowledge(args: Record<string, unknown>) { const query = args.query as string; const category = (args.category as string) || "all"; const nResults = (args.n_results as number) || 5; const results = await searchKnowledgeBase(query, category as any, nResults); return { query, category, results_count: results.length, results: results.map(r => ({ content: r.content.substring(0, 500), source: r.metadata.source || "knowledge_base", relevance: r.distance ? (1 - r.distance).toFixed(3) : "N/A" })) }; } async function handleGetMethodGuide(args: Record<string, unknown>) { const method = args.method as string; const aspect = (args.aspect as string) || "all"; const methodGuides: Record<string, any> = { ols: { name: "Ordinary Least Squares (OLS)", assumptions: [ "1. Linearity: E(Y|X) = Xβ", "2. Random Sampling: i.i.d. observations", "3. No Perfect Multicollinearity: rank(X) = k", "4. Zero Conditional Mean: E(ε|X) = 0", "5. Homoscedasticity: Var(ε|X) = σ²", "6. Normality (for inference): ε ~ N(0, σ²)" ], procedure: [ "1. Specify the model: Y = β₀ + β₁X₁ + ... + βₖXₖ + ε", "2. Check data (missing, outliers, distributions)", "3. Estimate: β̂ = (X'X)⁻¹X'Y", "4. Diagnostics: residual plots, VIF, normality tests", "5. Interpret coefficients and significance", "6. Report with robust standard errors if needed" ], interpretation: { coefficient: "1단위 X 증가 → β 단위 Y 변화 (다른 변수 통제)", r_squared: "설명된 분산 비율 (0-1)", f_test: "전체 모형 유의성", t_test: "개별 계수 유의성" }, reporting: "β = X.XX, SE = X.XX, t = X.XX, p < .05, 95% CI [X.XX, X.XX]" }, panel_fe: { name: "Panel Fixed Effects", assumptions: [ "1. Strict Exogeneity: E(εᵢₜ|Xᵢ₁,...,Xᵢₜ,αᵢ) = 0", "2. Time-invariant unobserved heterogeneity: αᵢ", "3. No perfect multicollinearity within entities", "4. Sufficient within-variation in X" ], procedure: [ "1. Demean: ỹᵢₜ = yᵢₜ - ȳᵢ, x̃ᵢₜ = xᵢₜ - x̄ᵢ", "2. Estimate: ỹᵢₜ = x̃ᵢₜβ + (εᵢₜ - ε̄ᵢ)", "3. Cluster standard errors at entity level", "4. Hausman test vs Random Effects" ], interpretation: { coefficient: "개체 내 X 변화 → Y 변화 (개체 고정효과 통제)", limitation: "시간불변 변수 효과 추정 불가" }, reporting: "β = X.XX (cluster SE), entity FE: Yes, time FE: Yes/No" }, did: { name: "Difference-in-Differences", assumptions: [ "1. Parallel Trends: 처치 없었다면 동일한 추세", "2. No Anticipation: 사전 처치 효과 없음", "3. SUTVA: 처치 간 간섭 없음", "4. No Composition Change: 그룹 구성 안정" ], procedure: [ "1. Basic 2x2: Y = β₀ + β₁Post + β₂Treat + β₃(Post×Treat) + ε", "2. Pre-trend test: 사전 기간 평행추세 확인", "3. Event study: 시간별 처치효과 추정", "4. Staggered DID: CS, SA, or did_imputation if staggered" ], interpretation: { att: "β₃ = E[Y(1)-Y(0)|Treat=1] = 처치집단 처치효과", event_study: "각 시점별 처치효과 (동적 효과)" }, reporting: "ATT = X.XX (SE), Pre-trend p = X.XX, N = X,XXX" } }; const guide = methodGuides[method.toLowerCase()] || { message: `'${method}' 가이드가 Knowledge Base에서 검색됩니다.`, suggestion: "search_stats_knowledge 도구로 검색해보세요." }; if (aspect !== "all" && guide[aspect]) { return { method, aspect, guide: guide[aspect] }; } return { method, guide }; } function handleSuggestMethod(args: Record<string, unknown>) { const dvType = args.dv_type as string; const dataStructure = (args.data_structure as string) || "cross_sectional"; const causalDesign = (args.causal_design as string) || "none"; const suggestions: Record<string, any> = { continuous: { cross_sectional: { observational: ["OLS Regression", "Robust Regression"], quasi_experimental: ["Propensity Score Matching", "IV/2SLS"], experimental: ["t-test", "ANOVA", "OLS with controls"] }, panel: { observational: ["Panel FE/RE", "Correlated Random Effects"], quasi_experimental: ["DID", "Synthetic Control", "Event Study"] }, time_series: { observational: ["ARIMA", "VAR", "VECM"], quasi_experimental: ["Interrupted Time Series", "Event Study"] } }, binary: { cross_sectional: { observational: ["Logistic Regression", "Probit"], quasi_experimental: ["PSM + Logit", "IV Probit"] }, panel: { observational: ["Conditional Logit (FE)", "Random Effects Logit"], quasi_experimental: ["DID Logit", "Linear Probability Model + FE"] } }, count: { cross_sectional: { observational: ["Poisson", "Negative Binomial", "Zero-Inflated"] }, panel: { observational: ["FE Poisson", "FE Negative Binomial"] } } }; const methodList = suggestions[dvType]?.[dataStructure]?.[causalDesign] || suggestions[dvType]?.[dataStructure]?.["observational"] || ["검색 필요: search_stats_knowledge 사용"]; return { research_question: args.research_question, dv_type: dvType, data_structure: dataStructure, causal_design: causalDesign, recommended_methods: methodList, considerations: [ "데이터 크기와 분포 확인 필요", "가정 검토 (내생성, 선택편의 등)", "robustness check 계획 수립" ] }; } function handleCompareMethods(args: Record<string, unknown>) { const methods = args.methods as string[]; const comparison: Record<string, any> = { fe_vs_re: { methods: ["Fixed Effects", "Random Effects"], key_difference: "FE는 개체 고정효과와 X 상관 허용, RE는 비상관 가정", decision: "Hausman test: H₀ 기각 → FE, H₀ 수용 → RE (효율성)", tradeoff: "FE: 일관성 | RE: 효율성 (시간불변 변수 추정 가능)" }, did_vs_scm: { methods: ["DID", "Synthetic Control"], key_difference: "DID는 다수 처치/통제, SCM은 단일 처치 단위에 적합", decision: "처치 단위 수: 1 → SCM, 다수 → DID", tradeoff: "DID: 통계 검정 용이 | SCM: 평행추세 가정 완화" }, psm_vs_iv: { methods: ["PSM", "IV/2SLS"], key_difference: "PSM은 관찰변수 통제, IV는 비관찰 내생성 해결", decision: "Selection on observables → PSM, Selection on unobservables → IV", tradeoff: "PSM: 도구변수 불필요 | IV: 비관찰 교란 통제" } }; const methodKey = methods.sort().join("_vs_"); const result = comparison[methodKey] || { methods, message: "상세 비교를 위해 search_stats_knowledge 도구 사용을 권장합니다." }; return result; } function handleGetFormula(args: Record<string, unknown>) { const concept = args.concept as string; const format = (args.format as string) || "both"; const formulas: Record<string, any> = { ols_estimator: { latex: "\\hat{\\beta} = (X'X)^{-1}X'Y", text: "β̂ = (X'X)⁻¹X'Y" }, did_att: { latex: "\\text{ATT} = (\\bar{Y}_{T,post} - \\bar{Y}_{T,pre}) - (\\bar{Y}_{C,post} - \\bar{Y}_{C,pre})", text: "ATT = (Ȳ_T,post - Ȳ_T,pre) - (Ȳ_C,post - Ȳ_C,pre)" }, hausman_test: { latex: "H = (\\hat{\\beta}_{FE} - \\hat{\\beta}_{RE})'[Var(\\hat{\\beta}_{FE}) - Var(\\hat{\\beta}_{RE})]^{-1}(\\hat{\\beta}_{FE} - \\hat{\\beta}_{RE})", text: "H = (β̂_FE - β̂_RE)'[Var(β̂_FE) - Var(β̂_RE)]⁻¹(β̂_FE - β̂_RE) ~ χ²(k)" }, cohens_d: { latex: "d = \\frac{\\bar{X}_1 - \\bar{X}_2}{S_{pooled}}", text: "d = (X̄₁ - X̄₂) / S_pooled" }, r_squared: { latex: "R^2 = 1 - \\frac{SS_{res}}{SS_{tot}} = 1 - \\frac{\\sum(y_i - \\hat{y}_i)^2}{\\sum(y_i - \\bar{y})^2}", text: "R² = 1 - (SS_res / SS_tot)" } }; const formula = formulas[concept.toLowerCase()]; if (!formula) { return { concept, error: "Formula not found. Try: ols_estimator, did_att, hausman_test, cohens_d, r_squared" }; } if (format === "latex") return { concept, latex: formula.latex }; if (format === "text") return { concept, text: formula.text }; return { concept, ...formula }; } function handleCalcSampleSize(args: Record<string, unknown>) { const testType = args.test_type as string; const effectSize = args.effect_size as number; const alpha = (args.alpha as number) || 0.05; const power = (args.power as number) || 0.80; // Simplified sample size formulas const zAlpha = alpha === 0.05 ? 1.96 : (alpha === 0.01 ? 2.576 : 1.645); const zBeta = power === 0.80 ? 0.84 : (power === 0.90 ? 1.28 : 0.52); let n: number; let formula: string; switch (testType) { case "t_test_two": n = Math.ceil(2 * ((zAlpha + zBeta) / effectSize) ** 2); formula = "n per group = 2 × ((z_α + z_β) / d)²"; break; case "t_test_paired": n = Math.ceil(((zAlpha + zBeta) / effectSize) ** 2); formula = "n = ((z_α + z_β) / d)²"; break; case "anova": const groups = (args.groups as number) || 3; n = Math.ceil((groups * (zAlpha + zBeta) ** 2) / (effectSize ** 2)); formula = "n per group = k × (z_α + z_β)² / f²"; break; case "regression": const predictors = (args.predictors as number) || 5; n = Math.ceil((8 / (effectSize ** 2)) + predictors); formula = "n ≈ 8/f² + k (Green, 1991 rule)"; break; default: n = Math.ceil(2 * ((zAlpha + zBeta) / effectSize) ** 2); formula = "Generic formula used"; } return { test_type: testType, effect_size: effectSize, alpha, power, required_n: n, formula, interpretation: `${power * 100}% 검정력으로 효과크기 ${effectSize} 탐지에 필요한 최소 표본: ${n}`, note: "실제 계산은 G*Power 또는 pwr 패키지 사용 권장" }; } function handleCalcPower(args: Record<string, unknown>) { const n = args.n as number; const effectSize = args.effect_size as number; const alpha = (args.alpha as number) || 0.05; // Simplified power calculation using helper function const power = calculatePower(n, effectSize, alpha); return { n, effect_size: effectSize, alpha, estimated_power: Math.min(power, 0.999).toFixed(3), interpretation: power >= 0.80 ? "적절한 검정력 (≥80%)" : "낮은 검정력 - 표본 증가 권장", r_code: `pwr.t.test(n = ${n}, d = ${effectSize}, sig.level = ${alpha}, type = "two.sample")` }; } function handleCalcEffectSize(args: Record<string, unknown>) { const type = args.type as string; const values = args.values as Record<string, number>; let result: any = { type }; switch (type) { case "cohens_d": if (values.mean1 && values.mean2 && values.sd_pooled) { result.effect_size = ((values.mean1 - values.mean2) / values.sd_pooled).toFixed(3); } else if (values.t && values.n1 && values.n2) { result.effect_size = (values.t * Math.sqrt((values.n1 + values.n2) / (values.n1 * values.n2))).toFixed(3); } result.interpretation = { small: "0.2", medium: "0.5", large: "0.8" }; break; case "eta_squared": if (values.ss_effect && values.ss_total) { result.effect_size = (values.ss_effect / values.ss_total).toFixed(3); } result.interpretation = { small: "0.01", medium: "0.06", large: "0.14" }; break; case "f_squared": if (values.r_squared) { result.effect_size = (values.r_squared / (1 - values.r_squared)).toFixed(3); } result.interpretation = { small: "0.02", medium: "0.15", large: "0.35" }; break; case "odds_ratio": if (values.or) { result.effect_size = values.or; result.cohens_d_approx = (Math.log(values.or) * Math.sqrt(3) / Math.PI).toFixed(3); } break; } return result; } function handleMdeCalculator(args: Record<string, unknown>) { const nPerGroup = args.n_per_group as number; const baseline = args.baseline as number; const alpha = (args.alpha as number) || 0.05; const power = (args.power as number) || 0.80; const testType = (args.test_type as string) || "proportion"; const zAlpha = 1.96; const zBeta = 0.84; let mde: number; if (testType === "proportion") { const se = Math.sqrt(2 * baseline * (1 - baseline) / nPerGroup); mde = (zAlpha + zBeta) * se; } else { mde = (zAlpha + zBeta) / Math.sqrt(nPerGroup / 2); } return { n_per_group: nPerGroup, baseline, alpha, power, mde: mde.toFixed(4), mde_percentage: (mde / baseline * 100).toFixed(2) + "%", interpretation: `현재 표본으로 baseline 대비 ${(mde / baseline * 100).toFixed(1)}% 차이 탐지 가능` }; } function handlePowerCurve(args: Record<string, unknown>) { const nRange = args.n_range as number[]; const effectSizes = args.effect_sizes as number[]; const alpha = (args.alpha as number) || 0.05; const curveData: any[] = []; const nMin = nRange[0]; const nMax = nRange[1]; const step = Math.ceil((nMax - nMin) / 20); for (let n = nMin; n <= nMax; n += step) { for (const d of effectSizes) { const power = calculatePower(n, d, alpha); curveData.push({ n, effect_size: d, power: Math.min(power, 0.999).toFixed(3) }); } } return { alpha, n_range: nRange, effect_sizes: effectSizes, curve_data: curveData, r_code: `library(pwr)\npower_curve <- pwr.t.test(n = seq(${nMin}, ${nMax}, by=${step}), d = c(${effectSizes.join(", ")}), sig.level = ${alpha}, type = "two.sample")` }; } // Placeholder implementations for remaining handlers function handleCheckAssumptions(args: Record<string, unknown>) { const method = args.method as string; return { method, assumptions_checklist: { ols: ["Linearity", "Normality of residuals", "Homoscedasticity", "Independence", "No multicollinearity"], logit: ["Binary DV", "Independence", "No multicollinearity", "Linearity in logit"], panel_fe: ["Strict exogeneity", "Time-invariant heterogeneity", "Within-variation"] }[method] || ["Method-specific assumptions - use search_stats_knowledge"], diagnostic_tests: ["검정 목록은 diagnose_regression 도구 사용"] }; } function handleDiagnoseRegression(args: Record<string, unknown>) { const issues = args.issues as string[]; const diagnostics: Record<string, any> = { multicollinearity: { tests: ["VIF (>10 문제)", "Condition Number (>30 문제)", "Correlation Matrix"], r_code: "car::vif(model)", stata_code: "vif", solution: "변수 제거, PCA, Ridge regression" }, heteroscedasticity: { tests: ["Breusch-Pagan test", "White test", "Residual plot"], r_code: "lmtest::bptest(model)", stata_code: "hettest", solution: "Robust SE, WLS, Log transformation" }, autocorrelation: { tests: ["Durbin-Watson", "Breusch-Godfrey", "ACF plot"], r_code: "lmtest::dwtest(model)", stata_code: "dwstat", solution: "HAC SE, GLS, ARIMA errors" } }; return { issues, diagnostics: issues.map(issue => diagnostics[issue] || { issue, message: "진단 정보 검색 필요" }) }; } function handleTestSelection(args: Record<string, unknown>) { return { args, suggestion: "Variable types에 따른 검정 추천 - 상세 구현 필요" }; } function handleInterpretDiagnostics(args: Record<string, unknown>) { const testName = args.test_name as string; const pValue = args.p_value as number; const interpretations: Record<string, string> = { "hausman": pValue < 0.05 ? "FE 사용 권장 (RE 불일치)" : "RE 사용 가능 (더 효율적)", "breusch_pagan": pValue < 0.05 ? "이분산 존재 - Robust SE 사용" : "등분산 가정 충족", "durbin_watson": "DW≈2: 자기상관 없음, DW<1.5 또는 >2.5: 자기상관 의심" }; return { test: testName, p_value: pValue, interpretation: interpretations[testName.toLowerCase()] || "검정 결과 해석 필요" }; } function handleRobustnessGuide(args: Record<string, unknown>) { return { main_analysis: args.main_analysis, robustness_checks: [ "Alternative specifications (different controls)", "Different estimation methods", "Sample restrictions (outliers, subgroups)", "Alternative measures of key variables", "Placebo tests" ] }; } function handleCausalDesignGuide(args: Record<string, unknown>) { const design = args.design as string; const guides: Record<string, any> = { did: { identification: "Parallel Trends Assumption", key_tests: ["Pre-trend test", "Placebo test", "Event study"], threats: ["Anticipation", "Spillover", "Composition change"], stata: "did_imputation, csdid, reghdfe", r: "did, fixest::feols" }, rdd: { identification: "Local randomization at cutoff", key_tests: ["McCrary density test", "Covariate balance at cutoff"], threats: ["Manipulation", "Compound treatment"], stata: "rdrobust", r: "rdrobust, rdd" }, iv: { identification: "Exclusion restriction + Relevance", key_tests: ["First-stage F (>10)", "Overidentification (Sargan/Hansen)"], threats: ["Weak instruments", "Invalid exclusion"], stata: "ivreg2, ivregress", r: "ivreg, fixest::feols" } }; return { design, guide: guides[design] || { message: "Design guide not found" } }; } function handleParallelTrendsCheck(args: Record<string, unknown>) { return { pre_periods: args.pre_periods, approaches: { visual: "사전 기간 처치/통제 그룹 추세 비교 그래프", statistical: "사전 기간 시간×처치 상호작용항 유의성 검정", placebo: "가짜 처치 시점에서 효과 없음 확인" }, r_code: "feols(y ~ i(time, treat, ref = -1) | id + time, data)", stata_code: "reghdfe y i.time#1.treat, absorb(id time) cluster(id)" }; } function handleIvStrengthCheck(args: Record<string, unknown>) { const firstStageF = args.first_stage_f as number; return { first_stage_f: firstStageF, interpretation: firstStageF >= 10 ? "Strong instrument" : "Weak instrument concern", additional_tests: [ "Stock-Yogo critical values", "Anderson-Rubin confidence sets (robust to weak IV)", "LIML estimator (less biased with weak IV)" ] }; } function handlePsmGuide(args: Record<string, unknown>) { return { matching_method: args.matching_method, steps: [ "1. Propensity score estimation (logit/probit)", "2. Check common support", "3. Perform matching", "4. Balance check (standardized differences <0.1)", "5. Estimate treatment effect on matched sample" ], balance_metrics: ["Standardized mean difference", "Variance ratio", "Overlap plots"] }; } function handleRddBandwidth(args: Record<string, unknown>) { return { design: args.design, methods: { ik: "Imbens-Kalyanaraman (2012) - 기본 선택", cct: "Calonico-Cattaneo-Titiunik (2014) - bias correction", cv: "Cross-validation - 데이터 기반" }, stata: "rdrobust y x, c(0) kernel(triangular)", r: "rdrobust(Y, X, c = 0)" }; } function handleEventStudyGuide(args: Record<string, unknown>) { return { event_type: args.event_type, staggered: args.staggered, estimators: { twfe: "Two-way FE (동질적 처치효과 시)", cs: "Callaway-Sant'Anna (이질적 처치효과)", sa: "Sun-Abraham (이질적 처치효과)", did_m: "de Chaisemartin-D'Haultfoeuille" }, recommendation: args.staggered ? "CS 또는 SA 추정량 권장" : "TWFE 사용 가능" }; } function handleGenerateRCode(args: Record<string, unknown>) { const analysisType = args.analysis_type as string; const templates: Record<string, string> = { ols: ` # OLS Regression library(tidyverse) library(fixest) library(modelsummary) # Estimate model <- lm(y ~ x1 + x2 + x3, data = df) # Robust SE model_robust <- feols(y ~ x1 + x2 + x3, data = df, vcov = "hetero") # Diagnostics car::vif(model) # Multicollinearity lmtest::bptest(model) # Heteroscedasticity # Results table modelsummary(list("OLS" = model, "Robust" = model_robust)) `, panel_fe: ` # Panel Fixed Effects library(fixest) library(modelsummary) # Entity FE model_fe <- feols(y ~ x1 + x2 | id, data = panel_df, vcov = ~id) # Entity + Time FE model_twfe <- feols(y ~ x1 + x2 | id + year, data = panel_df, vcov = ~id) # Results modelsummary(list("Entity FE" = model_fe, "TWFE" = model_twfe)) `, did: ` # Difference-in-Differences library(fixest) library(did) # Basic DID did_model <- feols(y ~ treat:post | id + time, data = df, vcov = ~id) # Event Study es_model <- feols(y ~ i(time, treat, ref = -1) | id + time, data = df, vcov = ~id) iplot(es_model) # Callaway-Sant'Anna (staggered) cs_did <- att_gt(yname = "y", tname = "time", idname = "id", gname = "first_treat", data = df) aggte(cs_did, type = "dynamic") |> ggdid() ` }; return { analysis_type: analysisType, r_code: templates[analysisType] || "# Analysis template not found\n# Use search_stats_knowledge for guidance" }; } function handleGenerateStataCode(args: Record<string, unknown>) { const analysisType = args.analysis_type as string; const templates: Record<string, string> = { ols: ` * OLS Regression reg y x1 x2 x3 * Robust SE reg y x1 x2 x3, robust * Diagnostics vif hettest ovtest `, panel_fe: ` * Panel Setup xtset id time * Fixed Effects xtreg y x1 x2, fe cluster(id) * Two-way FE (reghdfe) reghdfe y x1 x2, absorb(id time) cluster(id) * Hausman test xtreg y x1 x2, fe estimates store fe xtreg y x1 x2, re hausman fe `, did: ` * Basic DID gen treat_post = treat * post reg y treat post treat_post, cluster(id) * reghdfe reghdfe y treat_post, absorb(id time) cluster(id) * Event Study forvalues t = -5/5 { gen treat_t\`t' = treat * (time == event_time + \`t') } reghdfe y treat_t*, absorb(id time) cluster(id) coefplot, keep(treat_t*) ` }; return { analysis_type: analysisType, stata_code: templates[analysisType] || "* Analysis template not found" }; } function handleGeneratePythonCode(args: Record<string, unknown>) { const analysisType = args.analysis_type as string; const templates: Record<string, string> = { ols: ` import pandas as pd import statsmodels.api as sm import statsmodels.formula.api as smf # OLS model = smf.ols('y ~ x1 + x2 + x3', data=df).fit() print(model.summary()) # Robust SE model_robust = smf.ols('y ~ x1 + x2 + x3', data=df).fit(cov_type='HC3') # VIF from statsmodels.stats.outliers_influence import variance_inflation_factor X = df[['x1', 'x2', 'x3']] vif = pd.DataFrame({ 'Variable': X.columns, 'VIF': [variance_inflation_factor(X.values, i) for i in range(X.shape[1])] }) `, panel_fe: ` import pandas as pd from linearmodels import PanelOLS # Set index df = df.set_index(['id', 'time']) # Entity FE model_fe = PanelOLS(df['y'], df[['x1', 'x2']], entity_effects=True) result_fe = model_fe.fit(cov_type='clustered', cluster_entity=True) # Two-way FE model_twfe = PanelOLS(df['y'], df[['x1', 'x2']], entity_effects=True, time_effects=True) result_twfe = model_twfe.fit(cov_type='clustered', cluster_entity=True) ` }; return { analysis_type: analysisType, python_code: templates[analysisType] || "# Analysis template not found" }; } function handleCodeTemplate(args: Record<string, unknown>) { return { workflow: args.workflow, language: args.language, template: "Full workflow template - use specific code generation tools for detailed code" }; } function handleVisualizationCode(args: Record<string, unknown>) { const chartType = args.chart_type as string; const language = args.language as string; const templates: Record<string, Record<string, string>> = { coefficient_plot: { r: `library(ggplot2)\nmodelsummary::modelplot(model, coef_omit = "Intercept")`, stata: `coefplot, drop(_cons) xline(0)`, python: `import matplotlib.pyplot as plt\ncoefs = model.params[1:]\nerrs = model.bse[1:]\nplt.errorbar(coefs.index, coefs, yerr=1.96*errs, fmt='o')` }, forest_plot: { r: `library(metafor)\nforest(meta_result)`, stata: `metan effect se, forestplot`, python: `# Use forestplot package or matplotlib` } }; return { chart_type: chartType, language, code: templates[chartType]?.[language] || "# Visualization code template" }; } function handleTableCode(args: Record<string, unknown>) { return { table_type: args.table_type, format: args.format, language: args.language, code: "Table generation code - use modelsummary (R), esttab (Stata), or stargazer" }; } function handleDebugCode(args: Record<string, unknown>) { return { error: args.error_message, language: args.language, suggestions: [ "Check variable names and data types", "Verify data structure (missing values, duplicates)", "Check package/library installation", "Review function syntax and arguments" ] }; } function handleOptimizeCode(args: Record<string, unknown>) { return { language: args.language, data_size: args.data_size, recommendations: { r: ["Use data.table instead of data.frame", "Use fst for file I/O", "Consider collapse package"], stata: ["Use gtools (gisid, gegen)", "Use ftools (reghdfe)", "Compress data"], python: ["Use pandas with chunking", "Consider dask or polars", "Use parquet format"] } }; } function handleInterpretCoefficient(args: Record<string, unknown>) { const modelType = args.model_type as string; const coefficient = args.coefficient as number; const variableType = args.variable_type as string; let interpretation = ""; if (modelType === "ols" || modelType === "linear") { interpretation = `X가 1단위 증가하면 Y가 ${coefficient.toFixed(3)} 단위 변화 (다른 변수 통제)`; } else if (modelType === "logit" || modelType === "logistic") { const or = Math.exp(coefficient); interpretation = `Odds Ratio = ${or.toFixed(3)}. X가 1단위 증가하면 odds가 ${((or - 1) * 100).toFixed(1)}% 변화`; } return { model_type: modelType, coefficient, variable_type: variableType, interpretation, caution: "인과적 해석은 연구설계에 따라 결정" }; } function handleInterpretModelFit(args: Record<string, unknown>) { const metrics = args.metrics as Record<string, number>; return { metrics, interpretation: { r_squared: metrics.r_squared ? `설명력: ${(metrics.r_squared * 100).toFixed(1)}%` : null, adj_r_squared: metrics.adj_r_squared ? `조정된 설명력: ${(metrics.adj_r_squared * 100).toFixed(1)}%` : null, aic: metrics.aic ? "낮을수록 좋음 (모형 비교용)" : null, bic: metrics.bic ? "낮을수록 좋음 (복잡성 페널티 더 큼)" : null } }; } function handleMarginalEffectsGuide(args: Record<string, unknown>) { return { model_type: args.model_type, effect_types: { ame: "Average Marginal Effect - 평균적 한계효과", mem: "Marginal Effect at Mean - 평균값에서 한계효과", mer: "Marginal Effect at Representative values" }, r_code: "margins::margins(model)", stata_code: "margins, dydx(*)" }; } function handleInterpretTest(args: Record<string, unknown>) { const pValue = args.p_value as number; return { test_name: args.test_name, p_value: pValue, statistic: args.statistic, conclusion: pValue < 0.05 ? "통계적으로 유의함 (p < .05)" : "통계적으로 유의하지 않음 (p ≥ .05)", effect_size_note: "p-value 외에 효과크기도 보고 필요" }; } function handleWriteResultsSection(args: Record<string, unknown>) { return { analysis_type: args.analysis_type, style: args.style, template: "Results section template based on analysis type and style guide" }; } function handleMetaEffectSize(args: Record<string, unknown>) { return { input_type: args.input_type, values: args.values, r_code: "escalc(measure = 'SMD', m1i, sd1i, n1i, m2i, sd2i, n2i, data)" }; } function handleMetaHeterogeneity(args: Record<string, unknown>) { const iSquared = args.i_squared as number; return { i_squared: iSquared, interpretation: iSquared < 25 ? "Low heterogeneity" : iSquared < 75 ? "Moderate heterogeneity" : "High heterogeneity", implications: iSquared > 50 ? "Random effects model 권장, 이질성 원인 탐색" : "Fixed effect 가능" }; } function handlePublicationBias(args: Record<string, unknown>) { return { methods: args.methods, tests: { funnel_plot: "Visual asymmetry check", egger: "Regression test for asymmetry", trim_fill: "Impute missing studies", selection_model: "Model publication process" } }; } function handleMetaCode(args: Record<string, unknown>) { return { analysis: args.analysis, language: args.language, r_code: "library(metafor)\nres <- rma(yi, vi, data = dat)\nforest(res)", stata_code: "metan effect se, random" }; } function handleJournalGuide(args: Record<string, unknown>) { const journal = args.journal as string; const guides: Record<string, any> = { econometrica: { style: "Rigorous mathematical notation", expectations: ["Novel theoretical contribution", "Clean identification", "Formal proofs"], stats_reporting: "Full estimation details, robustness in appendix" }, aer: { style: "Clear causal identification", expectations: ["Policy relevance", "Clean natural experiment or RCT", "Replication data"], stats_reporting: "Main results in text, extensive robustness" }, jfe: { style: "Empirical finance focus", expectations: ["Financial data", "Asset pricing or corporate finance", "Economic significance"], stats_reporting: "t-statistics in parentheses, clustering at firm level" } }; return { journal, guide: guides[journal] || { message: "Journal guide not found" } }; } function handleApaReporting(args: Record<string, unknown>) { return { test_type: args.test_type, template: "t(df) = X.XX, p < .05, d = X.XX, 95% CI [X.XX, X.XX]" }; } function handlePreregTemplate(args: Record<string, unknown>) { return { platform: args.platform, sections: [ "1. Study Information", "2. Design Plan", "3. Sampling Plan", "4. Variables", "5. Analysis Plan", "6. Other" ] }; } function handleReplicationPackage(args: Record<string, unknown>) { return { components: args.components, structure: { code: "/code - Analysis scripts", data: "/data - Raw and processed data", output: "/output - Tables and figures", docs: "/docs - README, codebook" } }; } function handleReviewerResponse(args: Record<string, unknown>) { const critiqueType = args.critique_type as string; const responses: Record<string, any> = { endogeneity: { options: ["IV/2SLS", "Control function", "Selection model", "Bounds analysis"], template: "We address endogeneity concerns by [method]. Results remain robust..." }, selection: { options: ["Heckman selection", "PSM", "Bounds", "Placebo tests"], template: "Selection concerns are addressed through [method]..." } }; return { critique_type: critiqueType, response_guide: responses[critiqueType] || { message: "Response template not found" } }; } function handleSemGuide(args: Record<string, unknown>) { return { model_type: args.model_type, fit_indices: { acceptable: "CFI > .90, TLI > .90, RMSEA < .08, SRMR < .08", good: "CFI > .95, TLI > .95, RMSEA < .06, SRMR < .05" }, r_code: "library(lavaan)\nfit <- sem(model, data = df)\nsummary(fit, fit.measures = TRUE)" }; } function handleMlmGuide(args: Record<string, unknown>) { return { levels: args.levels, key_concepts: { icc: "Intraclass Correlation - 집단간 분산 비율", random_slope: "기울기의 집단간 변이 허용", cross_level: "수준간 상호작용" }, r_code: "library(lme4)\nfit <- lmer(y ~ x1 + (1 + x1 | group), data = df)" }; } function handleBayesianGuide(args: Record<string, unknown>) { return { analysis_type: args.analysis_type, prior_type: args.prior_type, convergence_checks: ["Rhat < 1.01", "ESS > 400", "Trace plots", "Posterior predictive checks"], r_code: "library(brms)\nfit <- brm(y ~ x1 + x2, data = df, family = gaussian())" }; } function handleMlForCausal(args: Record<string, unknown>) { return { method: args.method, description: { double_ml: "Debiased/Double ML - nuisance parameter estimation with ML", causal_forest: "Heterogeneous treatment effects via random forests" }, r_code: "library(DoubleML)\n# or library(grf) for causal forest" }; } function handleTimeseriesGuide(args: Record<string, unknown>) { return { analysis: args.analysis, workflow: [ "1. Stationarity test (ADF, KPSS)", "2. Determine order (ACF, PACF, Information criteria)", "3. Estimate model", "4. Diagnostics (Ljung-Box, residual ACF)", "5. Forecasting / Impulse response" ] }; } async function handleWriteAnalysisFile(args: Record<string, unknown>) { const content = args.content as string; const filename = args.filename as string; const directory = (args.directory as string) || "."; // Note: Actual file writing would require fs module return { status: "File writing capability - requires file system access", filename, directory, content_preview: content.substring(0, 200) + "..." }; } async function handleCreateProjectStructure(args: Record<string, unknown>) { const projectName = args.project_name as string; const template = (args.template as string) || "basic"; const structures: Record<string, string[]> = { basic: ["/code", "/data", "/output", "/docs"], replication: ["/code/analysis", "/code/data_prep", "/data/raw", "/data/processed", "/output/tables", "/output/figures", "/docs"], full: ["/code/00_master", "/code/01_data", "/code/02_analysis", "/code/03_robustness", "/data/raw", "/data/processed", "/data/temp", "/output/tables", "/output/figures", "/output/logs", "/docs", "/paper"] }; return { project_name: projectName, template, structure: structures[template], readme_template: "# Project Name\n\n## Structure\n\n## Replication Instructions\n\n## Data Sources\n\n## Contact" }; }