Stratified Random Sampling

Analyzing Survey Data in Python

EbunOluwa Andrew

Data Scientist

What is stratified random sampling

Stratified sampling better reflects population
Technique in which population is divided into discrete units (strata) based on similar attributes
Involves re-sampling the sample data so proportions match the population

Demographic change as a large group of people as a changing diversity in a population

Why use stratified random sampling?

Minimizes selection bias
Increases certain population group's representativeness
Examples:
- Estimating income for varying populations
- Estimating polling elections
- Estimating life expectancy

Hand held tally counter counting headcount of people

When not to use stratified random sampling

Subgroups should not overlap
- Subjects that fall into multiple groups -> misrepresentation
Example of overlap in survey question
- How long have you worked at your current job?
  - 1-2 years
  - 2-4 years

thinking lady

Onsite work survey results at firm ABC

| employee_id | gender | onsite_work |
|-------------|--------|-------------|
|    fffe6838 | Male   | Yes         |
|   fffe12184 | Female | Yes         |
|    fffe9404 | Female | Yes         |
|   fffe17578 | Male   | Yes         |
|   fffe22257 | Female | Yes         |
|    fffe6217 | Male   | Yes         |
|    fffe7828 | Female | Yes         |
|   fffe18192 | Male   | Yes         |
|    fffe2839 | Female | Yes         |
|   fffe16173 | Female | Yes         |

Check proportions on population

survey.gender.value_counts(normalize=True)

Female    0.556
Male      0.444
Name: gender, dtype: float64

Plotting proportions on population

import pandas as pd
import matplotlib.pyplot as plt

survey.gender.value_counts().plot.pie()

Survey female to male ratio

Stratified sampling example

strat_sample = 
survey.groupby(
  'gender', group_keys = False).apply(
  lambda x: x.sample(frac = 0.1))

| employee_id | gender | onsite_work |
|-------------|--------|-------------|
|    fffe4934 | Female | Yes         |
|    fffe3958 | Female | Yes         |
|      fffe18 | Female | Yes         |
|     fffe283 | Female | Yes         |
|   fffe20382 | Female | Yes         |
|    fffe8721 | Male   | Yes         |
|    fffe5955 | Male   | Yes         |
|    fffe7081 | Male   | Yes         |
|     fffe353 | Male   | Yes         |
|    fffe1765 | Male   | Yes         |

Check proportions on sample

Original population

survey.gender.value_counts(normalize=True)

Stratified sample

strat_sample.gender.value_counts(
  normalize=True))

Female    0.556
Male      0.444
Name: gender, dtype: float64

Female    0.56
Male      0.44
Name: gender, dtype: float64

Let's practice!

Analyzing Survey Data in Python