R & Python for Development

Back to Labs

How the live code works. The first time you Run R or Python, your browser downloads the language engine once (R ≈ 7 MB, Python ≈ 10 MB; later modules fetch a package or two more). Give the first click 10–30 seconds; after that it's quick. A shared dataset (SURVEY_CSV) is loaded for you when the engine starts, so every code cell can use it.

Module 1 of 7

Your first line of code

Two languages run most development-sector data work: R (loved by statisticians and evaluators) and Python (loved by data scientists). You don't have to choose — this course teaches both, side by side. Click Run, then switch the tab from R to Python and Run again.

editable

print("Hello from R")

A first calculation

Store five people's ages and compute the average. In R a list of numbers is c(...) and the mean is mean(...); in Python it's a list [...] and sum(...) / len(...).

editable

ages <- c(19, 22, 25, 31, 45)
mean(ages)

That's 80% of analysis: data goes into a variable, a function does something to it, you read the result.

Module 2 of 7

Data in, data out

Real analysis starts with a dataset — rows (people, households) and columns (variables). We've loaded a small illustrative household-survey sample into SURVEY_CSV for you. Illustrative data

Read it into a data frame — a spreadsheet-like table. In R, read.csv(text = SURVEY_CSV); in Python, pandas. Then peek at the first rows and the shape.

first Python run loads pandas

df <- read.csv(text = SURVEY_CSV)
head(df)
cat("rows:", nrow(df), " cols:", ncol(df), "\n")

Look at one column

Columns are accessed by name: df$pc_exp in R, df["pc_exp"] in Python. Let's summarise monthly per-capita expenditure.

editable

df <- read.csv(text = SURVEY_CSV)
summary(df$pc_exp)

In the field you'll read NFHS, ASER, PLFS or your own survey the same way — from a .csv file with read.csv("mydata.csv") or pd.read_csv("mydata.csv").

Module 3 of 7

Wrangling — filter and select

Most data work is reshaping: keep some rows, keep some columns. Say we want only rural households. In R (base) we index with a condition; in Python we use pandas.

editable

df <- read.csv(text = SURVEY_CSV)
rural <- df[df$area == "Rural", ]
rural[, c("district", "gender", "caste", "pc_exp")]

Add a new column

Flag "low expenditure" households (below ₹1,500/month). Assign a new column and count how many.

editable

df <- read.csv(text = SURVEY_CSV)
df$low_exp <- df$pc_exp < 1500
table(df$low_exp)

Once you're comfortable, R's tidyverse (dplyr) and Python's pandas chaining make this even cleaner: df |> filter(area=="Rural") / df.query('area=="Rural"'). Same idea, nicer grammar. (Our Code Convert Pro tool translates between them.)

Module 4 of 7

Summaries & disaggregation

An average hides as much as it reveals. The real story is in the disaggregation — the same number split by gender, caste, or geography. (This is exactly the lens of the Data Feminism Lab.)

Mean expenditure by gender, then by gender × caste:

editable

df <- read.csv(text = SURVEY_CSV)
aggregate(pc_exp ~ gender, df, mean)
aggregate(pc_exp ~ gender + caste, df, mean)

Read the gaps. A single "average expenditure" number would erase the gap between, say, General-caste and SC/ST households, and between men and women within each group. Disaggregation is where inequality becomes visible.

Module 5 of 7

Your first chart

A picture makes a gap obvious. We'll draw a bar chart of mean expenditure by caste. In Python we use matplotlib and call show() (a helper we've provided) to display it; in R we use base barplot().

First run of this module downloads the plotting library (a few MB more). The chart appears below the code.

draws an image

df <- read.csv(text = SURVEY_CSV)
m <- aggregate(pc_exp ~ caste, df, mean)
barplot(m$pc_exp, names.arg = m$caste,
        col = "#0EA5E9", main = "Mean expenditure by caste",
        ylab = "Rupees / month")

The production-grade tools are R's ggplot2 and Python's matplotlib/seaborn. For design principles, see the Data Visualization 101.

Module 6 of 7

Regression & the idea of impact

A regression asks: as one thing changes, how does another move? Here — does an extra year of education go with higher expenditure? In R, lm(y ~ x); in Python, numpy.polyfit for the same slope.

editable

df <- read.csv(text = SURVEY_CSV)
model <- lm(pc_exp ~ years_edu, data = df)
summary(model)$coefficients

From correlation to impact

That slope is a correlation, not proof that education causes higher spending. Impact evaluation isolates cause — e.g. difference-in-differences compares the change in a treated group to the change in a comparison group. A tiny illustration: Illustrative

editable

# before/after means for treated (T) and comparison (C) groups
T_before <- 1000; T_after <- 1600
C_before <- 1000; C_after <- 1200
did <- (T_after - T_before) - (C_after - C_before)
cat("Difference-in-differences estimate:", did, "\n")

Go deeper in the Causal Inference for Development course and Impact Evaluation 101. A great open text with R/Python/Stata code: Martin Huber, Causal Analysis (MIT Press).

Module 7 of 7

Reproducibility & sharing

Analysis you can't re-run is analysis nobody can trust. The habit: put your steps in a script, comment it, and share the script + data so anyone can reproduce your numbers. Here's the whole pipeline from this course in one runnable cell.

the full pipeline

# --- A tiny reproducible analysis ---
df <- read.csv(text = SURVEY_CSV)   # 1. read data
df$low_exp <- df$pc_exp < 1500       # 2. derive a variable
by_grp <- aggregate(pc_exp ~ gender + caste, df, mean)  # 3. summarise
print(by_grp)
cat("Share low-expenditure:", round(mean(df$low_exp), 2), "\n")  # 4. headline

Comment every step (the # lines) so future-you understands it.
Never edit data by hand — every change should be a line of code you can re-run.
Share the script and data (a folder, a GitHub repo, an .Rproj / notebook). Reproducibility is the difference between a claim and evidence.

You can now code with data

From "hello" to reading a survey, wrangling it, disaggregating, charting, and running a regression — in both R and Python, entirely in your browser.

RPythontidyverse / pandasWebR + PyodideNo install

Where to go next

→

Your first line of code

A first calculation

Data in, data out

Look at one column

Wrangling — filter and select

Add a new column

Summaries & disaggregation

Your first chart

Regression & the idea of impact

From correlation to impact

Reproducibility & sharing

You can now code with data

Where to go next

Econometrics 101

Data Feminism Lab

Code Convert Pro

Sampling Toolkit