加权抽样

使用 Python 分析调查数据

EbunOluwa Andrew

Data Scientist

什么是加权抽样？

不等概率
允许在数据收集阶段纠正问题

桌上的天平，砝码，衡器

加权抽样常用变量

人口特征，如性别、年龄、地区、教育
考虑参与与未参与研究的差异

家庭，不同年龄人群线描风格

基于单元的加权

¹ https://www.geopoll.com/blog/weighting-survey-data-raking-cell-weighting/

基于单元的加权

¹ https://www.geopoll.com/blog/weighting-survey-data-raking-cell-weighting/

基于单元的加权

¹ https://www.geopoll.com/blog/weighting-survey-data-raking-cell-weighting/

基于单元的加权

¹ https://www.geopoll.com/blog/weighting-survey-data-raking-cell-weighting/

用 Python 分析青少年消费模式

yp_survey

| Gender | Age | Entertainment |
|--------|-----|---------------|
| female |  19 | Agree         |
| female |  20 | Agree         |
| male   |  19 | Agree         |
| female |  19 | Disagree      |
| female |  24 | Disagree      |
| male   |  18 | Agree         |
| male   |  18 | Agree         |
| male   |  20 | Disagree      |

用 Python 分析青少年消费模式

import pandas as pd
yp_crosstab = pd.crosstab(
  yp_survey['Gender'],
  yp_survey['Entertainment'])
yp_crosstab

yp_crosstab.plot.barh()

|        | Agree | Disagree |
|--------|-------|----------|
| female |    81 |       84 |
| male   |    84 |       46 |

用 Python 分析青少年消费模式

survey = yp_survey.groupby(['Gender', 'Entertainment'])['Age'].count().reset_index()

survey.columns = ['Gender', 'Entertainment', 'Respondents']
survey

| Gender | Entertainment | Respondents |
|--------|---------------|-------------|
| female | Agree         |          81 |
| female | Disagree      |          84 |
| male   | Agree         |          84 |
| male   | Disagree      |          46 |

用 Python 分析青少年消费模式

survey['% total respondents'] = survey.Respondents * 100./survey.Respondents.sum()

survey['% of population'] = [35, 25, 20, 20]

survey['Weight'] = survey['% of population']/survey['% total respondents']

survey['Weighted Respondents'] = survey.Weight * survey.Respondents

用 Python 分析青少年消费模式

survey[['Gender','Entertainment',
        'Respondents','Weighted Respondents']].set_index(
  ['Gender','Entertainment']).plot.barh()

Passons à la pratique !

使用 Python 分析调查数据