From ESG Questions to Python Data Analysis

30 min

From ESG Questions to Data Questions

Learning objective: define what you want to answer first, then decide which data and which method you need.

Data Science begins with a question, not with Python. If the question is vague, even well-written code only produces numbers without a direction. So the first step in this course is to rewrite a broad concern as a specific question that data can actually answer.

• Does it name a time range? "Recently" should become a specific span, such as 2021 to 2025.
• Does it name a comparison? Against your own past results, or against the other sample companies?
• Does it state the unit clearly? "Emissions" should say whether it means tCO2e or an intensity metric.
• Does it state the boundary of the data? For example, this dataset only runs through 2025 and cannot say what happens after that.

Key takeaway: Define the question before writing code. Analysis quality begins with problem framing.

30 min

Understanding ESG Datasets and Metrics

Learning objective: understand row, column, observation, variable, dimension, measure, metric, and unit.

This course uses DS101_ESG_Company_Data_Clean.csv. It is synthetic educational data: four sample companies (Alpha Tech Co., Harbor Foods Co., Evergreen Components Co., and Metro Services Co.), each with one row per year from 2020 to 2025, for 24 rows in total. The company names and values were designed for teaching and do not represent any real company.

Scope 1+2 emissions = scope1_tco2e + scope2_tco2e
Emissions intensity = Scope 1+2 emissions / revenue_m_twd
Unit = tCO2e per NT$ million revenue

• When the concern is total emissions responsibility, regulatory reporting, or a net-zero target, absolute emissions is usually the primary figure.
• When comparing companies of different sizes, or the same company across years of changing operations, you usually need the intensity metric too.
• Calling a drop in absolute emissions "progress" on its own can overlook a simultaneous drop in revenue.
• Calling a drop in intensity "progress" on its own can overlook absolute emissions that are still rising.

Key takeaway: A number is only meaningful together with its definition and unit.

35 min

Notebook and Essential Python

Learning objective: run basic Python and pandas code in a notebook, without needing to memorize syntax.

# Use = to store a value in a variable (assignment)
company = "Alpha Tech"   # text (a string) goes in quotes
year = 2025              # numbers do not
threshold = 1000

• Assignment: = stores a value under a name you can use later
• Strings and numbers: text goes in quotes, such as "Technology"; numbers do not, such as 2025
• Function calls: a name followed by parentheses, such as pd.read_csv(...)
• Dot notation: it is enough to recognize df.head() as "do head on df"
• Comments: text after # is not run; use it to explain what a line does
• Reading error messages: start from the last line, which usually names the exact variable or column that caused the problem

import pandas as pd

# The file must be in the same folder as the notebook
df = pd.read_csv("DS101_ESG_Company_Data_Clean.csv")
df.head()

1. 1Explain line by line
   Ask AI to explain what each line of code does, not just to summarize the whole block.
2. 2List the columns used
   Ask AI which columns the code uses, then compare them with df.columns or the data dictionary.
3. 3State the expected output
   Ask AI how many rows and columns, or what kind of result, the code should produce, then check it against the real output.
4. 4No hallucinated columns
   Tell AI to use only columns that exist in the data. If you see a column name you do not recognize, check the data dictionary before using it.

• Have all the names this cell uses already been created in an earlier cell?
• If you reopened an older notebook, did you run Restart & Run All before trusting the output?
• When you paste someone else's code, do the column names match your own data, rather than the columns in their example?

Key takeaway: You are not being tested on memorizing Python; you are learning to understand and verify an analysis workflow.

40 min

Inspect, Select, and Filter with pandas

Learning objective: use pandas to answer "what does the data look like, which columns does it have, and which rows meet a condition?"

df.head()
df.shape
df.columns
df.info()
df.describe()

df[["company_name", "year", "scope1_tco2e", "scope2_tco2e"]]

df[df["year"] == 2025]

df[(df["year"] >= 2023) & (df["sector"] == "Technology")]

df.sort_values("scope1_tco2e", ascending=False)

• Row count: how many rows are left, and is that what you expected?
• Selected columns: are these the columns you meant, with nothing missing or extra?
• Filter condition: does the condition say what you meant (year, sector, greater than or greater than or equal to)?
• Unexpected empty result: if you get 0 rows, check spelling, capitalization, and data types before concluding that "there is no data"
• Surprising maximum or minimum: confirm the unit and definition first, then decide whether it needs verification

• df["sector"].unique(): lists the distinct values in a column, useful for spotting inconsistent spelling
• df["sector"].value_counts(): counts how often each value appears, good for a quick look at the category breakdown
• df["sector"].nunique(): the number of distinct values, when you do not need to see every one of them
• df.isna().sum(): how many missing values each column has; check this before filtering so you do not undercount rows

Key takeaway: Every analysis step should state what went in, what was done, and what remains.

45 min

Guided Lab: Your First ESG Data Analysis

Learning objective: with teacher guidance, complete a small reproducible analysis of the clean dataset and sort your observations into fact, interpretation, and needs verification.

• Go to the DS 101 Practice page
• PreviewDownload the clean dataset (CSV, synthetic data)
• Download the Guided Lab starter notebook
• PreviewDownload the data dictionary (Chinese)
• PreviewDownload the data dictionary (English)

1. 1Load the data
   Load DS101_ESG_Company_Data_Clean.csv.
2. 2Inspect the data
   Check the shape, columns, data types, and first rows.
3. 3Confirm the units
   Use the data dictionary to confirm the units of revenue_m_twd, scope1_tco2e, and scope2_tco2e.
4. 4Create the Scope 1+2 column
   Create scope12_tco2e = scope1_tco2e + scope2_tco2e.
5. 5Create the intensity column
   Create scope12_intensity = scope12_tco2e / revenue_m_twd, in tCO2e per NT$ million revenue.
6. 6Filter the years
   Keep only 2021 to 2025, and check the row count.
7. 7Sort one year
   Pick one year, sort by emissions intensity, and note that this is a single-year comparison.
8. 8Organize annual values by company
   Group by company and year so the year dimension is kept, rather than collapsing every year into one number.
9. 9Record three observations
   Write down, in your own words, three things you see in the output.
10. 10Label the evidence
    Label each observation as Fact from data, Interpretation, or Needs verification.
11. 11Write one limitation
    Write one sentence stating what this analysis cannot tell you.

import pandas as pd

df = pd.read_csv("DS101_ESG_Company_Data_Clean.csv")
df.shape
df.info()
df.head()

# Derived column: Scope 1+2 emissions (tCO2e)
df["scope12_tco2e"] = df["scope1_tco2e"] + df["scope2_tco2e"]

# Derived column: emissions intensity (tCO2e per NT$ million revenue)
df["scope12_intensity"] = df["scope12_tco2e"] / df["revenue_m_twd"]

# Step 6: keep only 2021 to 2025
df_lab = df[(df["year"] >= 2021) & (df["year"] <= 2025)]
df_lab.shape  # expect 20 rows: 4 companies x 5 years

# Step 7: pick one year and sort by emissions intensity
df_lab[df_lab["year"] == 2025].sort_values("scope12_intensity", ascending=False)[["company_name", "year", "scope12_intensity"]]

# Step 8: group by company and year, keeping the year dimension
annual = (
    df_lab.groupby(["company_name", "year"])[["scope12_tco2e", "revenue_m_twd"]]
    .sum()
    .reset_index()
)
# Recalculate intensity from the summed numerator and denominator; never sum intensities
annual["scope12_intensity"] = annual["scope12_tco2e"] / annual["revenue_m_twd"]
annual

• A notebook that runs again from top to bottom
• Two derived columns: scope12_tco2e and scope12_intensity
• At least three clearly labelled evidence statements (fact, interpretation, or needs verification)
• At least one limitation statement
• No causal claim that the dataset cannot support

Key takeaway: A good analysis produces not only numbers but also keeps the formula, units, evidence, and limitations.