Skip to content

📖 《把 AI 当助手——经济实证研究的诚信、方法与工具》双语网页版:每章中文部在前,English Edition 在后,内容与最新合订 PDF 一致。

第6章 · 数据来源全景 ​

学习目标 ​

完成本章后,你应当能够:

  1. 识别数据来源:说出中国官方四件套 + 国际三剑客 + 金融数据组的覆盖范围与获取方式

  2. 评估数据质量:用四维框架(口径一致性 / 频率匹配 / 覆盖完整性 / 获取成本)评估候选数据源

  3. 合规获取数据:遵守爬虫三原则,用学校正版渠道获取商业数据库

  4. 产出数据源矩阵:完成 data_source_matrix.xlsx,为你的研究选定数据源


§6.1 官方门户四件套 ​

中国官方统计门户 ​

表 6.1 中国官方统计门户。

L5|L5|L5|L5|L5

数据源 & 网址 & 核心数据 & 适用方向 & 获取方式
国家统计局 & stats.gov.cn / data.stats.gov.cn & 国民经济核算 / 价格指数 / 人口 / 居民收入 / 农产品产量 & 三方向通用(宏观与农业底座) & 网页下载 / API
中国人民银行 & pbc.gov.cn(调查统计司) & 金融统计数据报告 / 社会融资规模 / 货币政策执行报告 / 利率 / LPR & 【金】货币金融主源 & 报告 PDF 下载
国家金融监督管理总局 & nfra.gov.cn & 银行业总资产总负债(月度)/ 各地区保费收入 / 行业运行 & 【金】【险】监管口径主源 & 统计报表下载
农业农村部 & moa.gov.cn & 种植业 / 畜牧业 / 农业农村经济信息 & 【农】【险】农业底座 & 数据频道下载\

延伸门户(按需使用) ​

  • 海关总署:进出口贸易数据

  • 证监会:资本市场数据(上市公司公告、IPO)

  • 中国债券信息网 chinabond.com.cn:债券估值、收益率曲线

跟做练习 ​

从国家统计局下载“农村居民人均可支配收入”2010-2025 年年度时间序列:

  1. 打开 data.stats.gov.cn → 年度数据

  2. 选择“人民生活” → “居民人均可支配收入”

  3. 筛选“农村” + 时间范围 2010-2025

  4. 下载 Excel → 存入 data/raw/ 文件夹

  5. 记录元数据(S6.1):来源 URL、下载日期、文件大小


§6.2 国际三剑客(本机 MCP 实装) ​

三大国际组织数据源 ​

表 6.2 三大国际组织数据源。

L5|L5|L5|L5|L5

数据源 & MCP 工具 & 数据库 & 覆盖范围 & 适用场景
IMF & imf-mcp-server(5 个工具) & IFS 国际金融统计 / WEO 世界经济展望 & 全球 190+ 国家宏观金融指标 & 跨国金融稳定分析、资本充足率对比
OECD & oecd-mcp(9 个工具) & OECD 数据流 + 指标 & 38 个成员国 + 伙伴国 & 农业政策对比、乡村发展指标
FAO & (6 个工具) & FAOSTAT 69 个数据域 & 200+ 国家/地区 & 农业产量、粮食安全、土地利用、化肥\

延伸:World Bank ​

World Bank WDI(世界发展指标)和 GFDD(全球金融发展数据库)可通过网页直查:

  • WDI: data.worldbank.org(200+ 国家、1600+ 指标)

  • GFDD: 金融包容性指标(银行账户覆盖率、ATM 密度等)

跟做示例:用 imf-mcp 查询中国银行业资本充足率 ​

工具:imf_query_dataset
参数:
  dataset_code: "IFS"(国际金融统计)
  query: { country: "CN", indicator: "capital_adequacy_ratio", period: "2015-2025" }

你需要做的:

  1. 用 imf-mcp-server 获取中国银行业资本充足率 2015-2025 时间序列

  2. 记录数据来源(S6.1):“IMF International Financial Statistics, accessed via imf-mcp-server, 2026-09-21”

  3. 存入 data/raw/imf_capital_adequacy_CHN.csv

[提示] MCP 工具的优势:直接在命令行/AI 助手中调用,无需打开网页手动下载。但数据出处的引用规范不能因为获取方式便利就省略。


§6.3 金融数据组(专业数据源) ​

按获取方式分三层 ​

第一层:商业终端(学校订阅 / 机构付费) ​

表 6.3 按获取方式分三层。

L4|L4|L4|L4

数据源 & 覆盖范围 & 获取方式 & 适用场景
Wind & A 股 / 债券 / 基金 / 宏观 / 行业,最全 & 学校图书馆 VPN 或终端 & 论文标准数据源(审稿人认可度最高)
iFinD 同花顺 & 全品类金融数据(7 个子服务) & 学校/机构订阅 & 替代 Wind 的选择
Choice 东方财富 & 股票 / 基金 / 宏观 & 学校订阅 & 数据覆盖面广
Bloomberg & 全球市场 & 机构终端 & 国际金融方向\

第二层:学术数据库(学校订阅) ​

表 6.4 数据源 / 覆盖 / 特点速查表。

L3|L3|L3

数据源 & 覆盖 & 特点
CSMAR 国泰安 & 上市公司财务 / 公司治理 / 资本市场 & 面板数据最完整,论文引用率高
CNRDS 中国研究数据服务平台 & 绿色经济 / 创新 / 社会责任等特色数据 & 有独特的非财务指标
CFPS 中国家庭追踪调查 & 家庭收入/消费/资产/教育/健康 & 北京大学主持,微观面板
CHFS 中国家庭金融调查 & 家庭资产负债表/金融资产配置 & 西南财大主持,金融风险偏好\

第三层:开源免费(人人可用) ​

表 6.5 数据源 / 本机 MCP / 覆盖 / 注意事项速查表。

L4|L4|L4|L4

数据源 & 本机 MCP & 覆盖 & 注意事项
Tushare & tushareMcp(254 个工具) & A 股/基金/期货/宏观/港股/美股 & 需注册 token + 积分制
china-stock-mcp & china-stock-mcp(30 个工具) & A 股实时/历史/财务/资金流 & 免费替代
AKShare & Python 库 & A 股/港股/期货/宏观 & pip install akshare
yfinance & Python 库 & 全球股票市场 & Yahoo Finance 接口\

特色数据(专题方向必知) ​

表 6.6 数据 / 来源 / 适用速查表。

L3|L3|L3

数据 & 来源 & 适用
北京大学数字普惠金融指数 & 郭峰等(2020),蚂蚁金服合作 & 数字金融 / 普惠金融方向
中国家庭金融调查(CHFS) & 西南财经大学 & 家庭金融 / 风险偏好
中国农业统计年鉴 & 农业农村部 & 农业生产 / 保险精算\

[提示] 选择策略:硕士论文首选 Wind/CSMAR(审稿人熟悉);如果学校没订阅,Tushare + AKShare 可以覆盖 80% 的需求。关键是注明来源(S6.1)。

图 6.1 数据来源四象限全景——权威性 × 获取难度二维矩阵,帮你快速定位研究数据的入口。


§6.4 数据源选择决策树 ​

四步选择法 ​

面对一个研究问题,如何从众多数据源中选出最合适的一个?遵循以下四步:

  1. 明确变量定义:你的核心变量是什么?(如"农户收入"是年度可支配收入还是月度现金流?)

  2. 确定数据频率:需要年度/季度/月度/日度数据?面板还是截面?

  3. 评估覆盖范围:需要全国/省份/县域/村级/个体层面数据?

  4. 权衡获取成本:学校订阅了哪些数据库?开源数据能否满足?

决策树示例 ​

你的研究问题是什么?
├─ 宏观金融稳定分析
│  ├─ 跨国比较 → IMF IFS / WEO
│  ─ 中国时间序列 → 人民银行 / 国家统计局
├─ 微观家庭金融
│  ├─ 全国代表性样本 → CFPS / CHFS
│  └─ 特定地区 → 自编问卷
├─ 上市公司研究
│  ├─ 财务数据 → CSMAR / Wind
│  └─ 创新/绿色 → CNRDS
└─ 农业/农村发展
   ├─ 宏观统计 → 国家统计局 / 农业农村部
   └─ 微观调查 → 自编问卷 / CHFS

案例:数字普惠金融对农户收入的影响 ​

假设你的研究问题是"数字普惠金融发展对农户收入的影响",需要以下数据:

表 6.7 案例:数字普惠金融对农户收入的影响。

L3|L3|L3

变量 & 候选数据源 & 选择理由
数字普惠金融指数 & 北大数字金融研究中心 & 唯一省级/县级/村级数字金融指数
农户收入 & CFPS / 国家统计局 & CFPS 有家庭微观数据;统计局只有省级
控制变量(教育/年龄等) & CFPS & 微观调查包含人口统计学变量\

最终选择:CFPS 微观面板 + 北大数字金融指数(县级)匹配。

[提示] 数据源选择没有"标准答案",只有"最适合你的研究设计"。关键是在论文中说明选择理由,让审稿人看到你的决策过程。


§6.5 爬虫合规三原则 + 四维质量评估 ​

网络数据采集合规三原则 ​

表 6.8 网络数据采集合规三原则。

L4|L4|L4|L4

# & 原则 & 含义 & 违反后果
1 & 遵守 robots.txt & 网站明确禁止爬取的路径不碰 & 法律风险(反不正当竞争法)
2 & 控制请求频率 & 不给服务器造成负担(建议 ≥ 1 秒/次) & IP 被封禁;影响他人正常使用
3 & 尊重版权与隐私 & 不爬付费内容;不爬个人隐私数据 & 学术不端 + 法律风险\

工具提示:本机 chrome-devtools MCP 可以抓取公开统计表格(如金融监管总局的地区保费表),但必须遵守上述三原则。

四维数据质量评估框架 ​

表 6.9 四维数据质量评估框架。

L4|L4|L4|L4

维度 & 含义 & 检查方法 & 对接第 5 章
口径一致性 & 同一指标不同年份的定义是否变化 & 查统计年鉴注释/官方口径说明 & → 内容效度(测的是不是同一个东西?)
频率匹配 & 数据频率是否满足研究需要 & 年度 vs 季度 vs 月度 vs 日度 & → 测量精度(粗粒度 = 信息损失)
覆盖完整性 & 是否有缺失地区/年份/行业 & 面板平衡性检验 & → 样本流失(第 4 章内部效度威胁之一)
获取成本 & 时间/金钱/技术门槛 & 评估 ROI(投入产出比) & → 可行性(设计阶段的现实约束)\

口径一致性的典型案例 ​

案例:国家统计局“农村居民人均可支配收入”

  • 2013 年之前:使用“纯收入”口径(扣除生产经营费用但不扣除社保缴费)

  • 2013 年及之后:统一为“可支配收入”口径(扣除个税 + 社保缴费 + 转移支出)

  • 影响:2013 年前后不可直接比较!如果你的研究跨越 2013 年,必须做口径调整或在论文中说明

△  这是农经/农村金融方向最常踩的坑之一。很多论文直接把 2005-2025 年的收入数据拉成面板跑回归,完全没有注意到口径变化——审稿人一眼就看出来。


AI 辅助研究栏 ​

1. 用 MCP 工具快速获取数据概览 ​

场景:你想知道某个金融指标是否可获取、覆盖范围如何。

# 示例:用 tushareMcp 查看可用数据
工具:stock_basic
参数:{ exchange: "SSE", list_status: "L" }
-> 返回所有上交所上市公司列表

# 示例:用 faostat-mcp 查看农业数据域
工具:faostat_list_domains
-> 返回 69 个可查询域(产量/贸易/价格/化肥/土地…)

2. 自动化数据清洗建议 ​

拿到原始数据后,贴给 AI 问:

  • “这份数据有哪些常见问题?(缺失值、异常值、格式不一致)”

  • “用什么 Stata 命令做初步清洗?”

  • “这个面板数据是否平衡?用 xtset + xtdescribe 检查”

3. 数据源推荐 ​

把你的研究问题告诉 AI:

我的研究问题是"数字普惠金融对农户收入的影响"。
请推荐 3-5 个可能的数据源,并说明各自的优缺点。

△  AI 推荐的数据源你必须自己验证是否存在、是否能获取、是否覆盖你的变量。AI 可能“幻觉”出不存在的数据库。


🛡 AI 诚信栏|S6.1 数据出处与版本留痕 ​

必须做到:

  • ✅ 每个数据集记录:来源 URL / 下载日期 / 版本号 / 文件大小

  • ✅ data/raw/ 文件夹存放原始下载文件,永不修改

  • ✅ 论文中注明:“数据来源于 XX(访问日期 YYYY-MM-DD)”

  • ✅ MCP 工具获取的数据同样注明:“IMF IFS, accessed via imf-mcp-server, 2026-09-21”

数据留痕模板:

# 数据获取日志

| 日期 | 数据源 | URL/工具 | 文件名 | 大小 | 备注 |
|---|---|---|---|---|---|
| 2026-09-21 | 国家统计局 | data.stats.gov.cn | rural_income_2010_2025.xlsx | 45KB | 年度数据 |
| 2026-09-22 | IMF IFS | imf-mcp-server | imf_car_CHN.csv | 12KB | 资本充足率 |

S6.2 爬虫合规 ​

绝对禁止:

  • ❌ 爬取需要登录的付费数据库(Wind/CSMAR/Bloomberg)→ 用学校正版渠道

  • ❌ 爬取个人隐私数据(身份证号/手机号/银行账号)

  • ❌ 绕过验证码、反爬机制

  • ❌ 批量转售或二次发布原始数据

正确做法:

  • ✅ 只抓公开页面的统计表格

  • ✅ 请求频率 ≥ 1 秒/次

  • ✅ 引用规范与手动下载完全一致



本章产出物 ​

data_source_matrix.xlsx(数据源对比矩阵) ​

用 Excel 建立你的数据源评估矩阵:

表 6.10 data_source_matrix.xlsx(数据源对比矩阵)。

L4|L4|L4|L4

研究变量 & 候选数据源 1 & 候选数据源 2 & 候选数据源 3
(如:农户收入) & CFPS & 国家统计局 & 自编问卷
口径一致性(1-5) & 4 & 3(2013 口径变化) & 5(自己设计)
频率匹配(1-5) & 5(两年一次面板) & 4(年度) & 5(按需)
覆盖完整性(1-5) & 4(25 省) & 5(全国) & 3(仅样本县)
获取成本(1-5) & 3(需申请) & 5(免费下载) & 2(需入户)
综合得分 & 16 & 17 & 15
最终选择 & & ✅ &\

[提示] 评分规则:5 = 最好,1 = 最差。最终选择不仅看分数,还要看你的研究设计需要(如需面板数据,截面数据得分再高也不行)。



English Edition ​

Chapter 6 · A Panorama of Data Sources ​

Data is the raw material of empirical research. This chapter organizes the data sources commonly used in financial research—especially rural finance research—into four blocks: the four official portals, the international big three, financial data sources in three tiers, and the decision-tree selection method. All of these have been tested on this very machine (the MCP tools actually run). After finishing this chapter you will have a data_source_matrix.xlsx that provides a documented, defensible answer to the question “Where does your data come from?”

Learning Objectives ​

After completing this chapter, you should be able to:

  1. Identify data sources: state the coverage and access routes of China’s official big four, the international big three, and the financial data group

  2. Assess data quality: evaluate candidate data sources with the four-dimension framework (definitional consistency / frequency match / coverage completeness / acquisition cost)

  3. Acquire data compliantly: follow the three web-scraping principles and use the university’s licensed channels for commercial databases

  4. Produce a data-source matrix: complete data_source_matrix.xlsx and settle on the data sources for your research


6.1 The Four Official Portals ​

China’s Official Statistical Portals ​

Table 6.1 China’s official statistical portals.

L5|L5|L5|L5|L5

Source & URL & Core data & Research fit & Access
National Bureau of Statistics & stats.gov.cn / data.stats.gov.cn & National accounts / price indices / population / household income / farm output & Universal for all three tracks (macro and agriculture base) & Web download / API
People’s Bank of China & pbc.gov.cn (Survey and Statistics Dept.) & Financial statistics reports / aggregate social financing / monetary policy reports / interest rates / LPR & 【Fin】primary source for monetary finance & Report PDF downloads
National Financial Regulatory Administration & nfra.gov.cn & Banking assets and liabilities (monthly) / regional premium income / industry performance & 【Fin】【Ins】primary source of regulatory data & Statistical reports
Ministry of Agriculture and Rural Affairs & moa.gov.cn & Crop farming / animal husbandry / agricultural-rural economy & 【Agr】【Ins】agriculture base & Data channel downloads\

Extended Portals (Use as Needed) ​

  • General Administration of Customs: import–export trade data

  • China Securities Regulatory Commission: capital market data (listed-company announcements, IPOs)

  • ChinaBond chinabond.com.cn: bond valuations, yield curves

Hands-On Exercise ​

Download the 2010–2025 annual series of “per capita disposable income of rural residents” from the NBS:

  1. Open data.stats.gov.cn → Annual data

  2. Choose “People’s livelihood” → “Per capita disposable income”

  3. Filter “rural” + time span 2010–2025

  4. Download the Excel → save it in the data/raw/ folder

  5. Record the metadata (S6.1): source URL, download date, file size


6.2 The International Big Three (Implemented as Local MCP Tools) ​

Data Sources of the Three International Organizations ​

Table 6.2 Data sources of the three international organizations.

L5|L5|L5|L5|L5

Source & MCP tool & Database & Coverage & Use cases
IMF & imf-mcp-server (5 tools) & IFS International Financial Statistics / WEO World Economic Outlook & Macro-finance indicators for 190+ countries & Cross-country financial stability, capital adequacy comparisons
OECD & oecd-mcp (9 tools) & OECD dataflows + indicators & 38 member countries + partners & Agricultural policy comparison, rural development indicators
FAO & (6 tools) & FAOSTAT, 69 data domains & 200+ countries/regions & Farm output, food security, land use, fertilizers\

Extension: World Bank ​

World Bank WDI (World Development Indicators) and GFDD (Global Financial Development Database) can be queried directly on the web:

  • WDI: data.worldbank.org (200+ countries, 1,600+ indicators)

  • GFDD: financial inclusion indicators (bank account coverage, ATM density, etc.)

Hands-On Example: Querying Chinese Banks’ Capital Adequacy via imf-mcp ​

Tool: imf_query_dataset
Arguments:
  dataset_code: "IFS" (International Financial Statistics)
  query: { country: "CN", indicator: "capital_adequacy_ratio", period: "2015-2025" }

What you need to do:

  1. Use imf-mcp-server to fetch the 2015–2025 capital adequacy series of Chinese banks

  2. Record the source (S6.1): “IMF International Financial Statistics, accessed via imf-mcp-server, 2026-09-21”

  3. Save as data/raw/imf_capital_adequacy_CHN.csv

[Tip] The advantage of MCP tools: call them straight from the command line / AI assistant without opening a browser to download manually. But the citation norms for data provenance must never be skipped just because access has become easy.


6.3 The Financial Data Group (Professional Sources) ​

Three Tiers by Access Route ​

Tier 1: Commercial Terminals (University / Institutional Subscription) ​

Table 6.3 The three tiers by access route.

L4|L4|L4|L4

Source & Coverage & Access & Use cases
Wind & A-shares / bonds / funds / macro / industries, the most complete & University library VPN or terminal & The standard source for theses (highest reviewer recognition)
iFinD (Hexin Flush) & All financial data categories (7 sub-services) & University / institutional subscription & The alternative to Wind
Choice (Eastmoney) & Stocks / funds / macro & University subscription & Broad data coverage
Bloomberg & Global markets & Institutional terminal & International finance track\

Tier 2: Academic Databases (University Subscription) ​

Table 6.4 Sources, coverage, and features.

L3|L3|L3

Source & Coverage & Features
CSMAR (GTARS) & Listed firms’ financials / governance / capital markets & The most complete panel data; heavily cited in papers
CNRDS & Green economy / innovation / CSR and other distinctive data & Unique non-financial indicators
CFPS (China Family Panel Studies) & Household income / consumption / assets / education / health & Run by Peking University; micro panel
CHFS (China Household Finance Survey) & Household balance sheets / financial asset allocation & Run by Southwestern University of Finance and Economics; risk preferences\

Tier 3: Open Source (Available to Everyone) ​

Table 6.5 Sources, local MCP tools, coverage, and caveats.

L4|L4|L4|L4

Source & Local MCP & Coverage & Caveats
Tushare & tushareMcp (254 tools) & A-shares / funds / futures / macro / HK stocks / US stocks & Requires a registered token + points system
china-stock-mcp & china-stock-mcp (30 tools) & A-share real-time / historical / financials / fund flows & Free alternative
AKShare & Python library & A-shares / HK stocks / futures / macro & pip install akshare
yfinance & Python library & Global equity markets & Yahoo Finance interface\

Distinctive Data (Must-Know for Specialized Tracks) ​

Table 6.6 Data, source, and research fit.

L3|L3|L3

Data & Source & Research fit
Peking University Digital Financial Inclusion Index & Guo Feng et al. (2020), with Ant Financial & Digital finance / inclusive finance
China Household Finance Survey (CHFS) & Southwestern University of Finance and Economics & Household finance / risk preferences
China Agricultural Statistical Yearbook & Ministry of Agriculture and Rural Affairs & Agricultural production / insurance actuarial work\

[Tip] Selection strategy: for a master’s thesis prefer Wind/CSMAR (familiar to reviewers); if your university lacks a subscription, Tushare + AKShare can cover 80% of needs. The key is to cite the source (S6.1).

Figure 6.1 A four-quadrant panorama of data sources—an authority × access-difficulty matrix that helps you locate the entry point for your research data.


6.4 The Data-Source Decision Tree ​

The Four-Step Selection Method ​

Given a research question, how do you pick the most suitable source from many candidates? Follow these four steps:

  1. Define your variables: What exactly are your core variables? (For example, is “farm household income” annual disposable income or monthly cash flow?)

  2. Fix the data frequency: Do you need annual / quarterly / monthly / daily data? Panel or cross-section?

  3. Assess coverage: Do you need national / provincial / county / village / individual level data?

  4. Weigh acquisition cost: Which databases does your university subscribe to? Can open sources suffice?

A Decision-Tree Example ​

What is your research question?
|-- Macro financial stability
|   |-- Cross-country comparison -> IMF IFS / WEO
|   '-- China time series      -> PBoC / NBS
|-- Micro household finance
|   |-- Nationally representative -> CFPS / CHFS
|   '-- Specific region           -> Own survey
|-- Listed-company research
|   |-- Financial data        -> CSMAR / Wind
|   '-- Innovation/green      -> CNRDS
'-- Agriculture / rural development
    |-- Macro statistics      -> NBS / MOA
    '-- Micro survey          -> Own survey / CHFS

Case Study: Digital Financial Inclusion and Farm Household Income ​

Suppose your research question is “the impact of digital financial inclusion on farm household income.” You need the following data:

Table 6.7 Case study: digital financial inclusion and farm household income.

L3|L3|L3

Variable & Candidate sources & Rationale
Digital financial inclusion index & Peking University Digital Finance Research Center & The only province/county/village-level digital finance index
Farm household income & CFPS / NBS & CFPS has household micro data; the NBS offers province level only
Controls (education / age, etc.) & CFPS & Micro surveys include demographic variables\

Final choice: CFPS micro panel matched with the PKU digital finance index (county level).

[Tip] There is no “standard answer” in data-source selection—only “the best fit for your research design.” The key is to state your selection rationale in the paper so reviewers can see your decision process.


6.5 Three Web-Scraping Compliance Principles + Four-Dimension Quality Assessment ​

Three Compliance Principles for Web Data Collection ​

Table 6.8 Three compliance principles for web data collection.

L4|L4|L4|L4

# & Principle & Meaning & Consequences of violation
1 & Obey robots.txt & Never touch paths the website explicitly forbids & Legal risk (Anti-Unfair Competition Law)
2 & Control request frequency & Do not burden the server (recommend ≥ 1 request/second) & IP bans; harming other users
3 & Respect copyright and privacy & No scraping paywalled content; no scraping personal data & Academic misconduct + legal risk\

Tool note: the local chrome-devtools MCP can scrape public statistical tables (e.g., the NFRA regional premium table), but you must observe the three principles above.

The Four-Dimension Data Quality Framework ​

Table 6.9 The four-dimension data quality framework.

L4|L4|L4|L4

Dimension & Meaning & How to check & Link to Chapter 5
Definitional consistency & Has the definition of the same indicator changed across years? & Check yearbook notes / official definitions & → Content validity (is it measuring the same thing?)
Frequency match & Does the data frequency meet the research need? & Annual vs quarterly vs monthly vs daily & → Measurement precision (coarse granularity = information loss)
Coverage completeness & Any missing regions / years / industries? & Panel balance tests & → Sample attrition (one of Chapter 4’s internal validity threats)
Acquisition cost & Time / money / technical threshold & Evaluate the ROI & → Feasibility (a real constraint at the design stage)\

A Classic Case of Definitional Inconsistency ​

Case: the NBS’s “per capita disposable income of rural residents”

  • Before 2013: the “net income” definition (deducting production costs but not social insurance contributions)

  • From 2013 on: unified to the “disposable income” definition (deducting income tax + social insurance + transfer expenditures)

  • Impact: the pre- and post-2013 figures are not directly comparable! If your study spans 2013, you must adjust the definition or state so in the paper

△  This is one of the most common traps in agricultural economics / rural finance. Many papers stack 2005–2025 income data into a panel and run regressions without noticing the definitional change—reviewers spot it at a glance.


AI-Assisted Research Corner ​

1. Getting a Quick Data Overview with MCP Tools ​

Scenario: you want to know whether a financial indicator is available and how well it is covered.

# Example: list available data with tushareMcp
Tool: stock_basic
Arguments: { exchange: "SSE", list_status: "L" }
-> Returns the list of all SSE-listed companies

# Example: explore agricultural data domains with faostat-mcp
Tool: faostat_list_domains
-> Returns 69 queryable domains (output / trade / prices / fertilizer / land ...)

2. Suggestions for Automated Data Cleaning ​

After obtaining raw data, paste it to the AI and ask:

  • “What common problems does this dataset have? (missing values, outliers, inconsistent formats)”

  • “Which Stata commands should I use for preliminary cleaning?”

  • “Is this panel balanced? Check with xtset + xtdescribe”

3. Data-Source Recommendations ​

Tell the AI your research question:

My research question is "the impact of digital financial
inclusion on farm household income."
Please recommend 3-5 possible data sources with their
respective pros and cons.

△  You must personally verify every data source the AI recommends: does it exist, can you access it, does it cover your variables? AI can “hallucinate” databases that do not exist.


🛡 AI Integrity|S6.1 Data Provenance and Version Records ​

You must:

  • ✅ For every dataset record: source URL / download date / version / file size

  • ✅ Keep original downloads in the data/raw/ folder and never modify them

  • ✅ State in the paper: “Data from XX (accessed YYYY-MM-DD)”

  • ✅ For MCP-acquired data likewise note: “IMF IFS, accessed via imf-mcp-server, 2026-09-21”

Data provenance log template:

# Data acquisition log

| Date       | Source | URL/Tool          | File                        | Size | Note        |
|------------|--------|-------------------|-----------------------------|------|-------------|
| 2026-09-21 | NBS    | data.stats.gov.cn | rural_income_2010_2025.xlsx | 45KB | Annual data |
| 2026-09-22 | IMFIFS | imf-mcp-server    | imf_car_CHN.csv             | 12KB | Cap. adequacy |

S6.2 Web-Scraping Compliance ​

Absolutely forbidden:

  • ❌ Scraping subscription databases behind logins (Wind/CSMAR/Bloomberg) → use the university’s licensed channels

  • ❌ Scraping personal private data (ID numbers / phone numbers / bank accounts)

  • ❌ Bypassing CAPTCHAs or anti-scraping mechanisms

  • ❌ Reselling or republishing raw data in bulk

The right way:

  • ✅ Scrape only statistical tables on public pages

  • ✅ Request frequency ≥ 1 second per request

  • ✅ Cite exactly as you would for a manual download



Chapter Deliverables ​

data_source_matrix.xlsx (Data-Source Comparison Matrix) ​

Build your data-source evaluation matrix in Excel:

Table 6.10 data_source_matrix.xlsx (data-source comparison matrix).

L4|L4|L4|L4

Research variable & Candidate source 1 & Candidate source 2 & Candidate source 3
(e.g., farm household income) & CFPS & NBS & Own survey
Definitional consistency (1–5) & 4 & 3 (2013 definitional change) & 5 (self-designed)
Frequency match (1–5) & 5 (biennial panel) & 4 (annual) & 5 (on demand)
Coverage completeness (1–5) & 4 (25 provinces) & 5 (national) & 3 (sample counties only)
Acquisition cost (1–5) & 3 (application needed) & 5 (free download) & 2 (field visits)
Composite score & 16 & 17 & 15
Final choice & & ✅ &\

[Tip] Scoring rule: 5 = best, 1 = worst. The final choice is not just about scores—it must also fit your research design (e.g., if you need panel data, no cross-section data qualifies no matter how highly it scores).


本站教学资源仅供四川农业大学选课学生学习使用,版权归原作者所有,未经授权不得转载、转发或用于商业用途