Common mistake 1 with India - district map

Treating district boundaries as fixed. They are not, and a series that spans a split is not a series about one place - it is two places sharing a name.

Common mistake 2 with India - district map

Reading a district choropleth as if area meant importance. Rajasthan’s desert districts are enormous and thinly populated; Mumbai is a dot with more people than most states. Shading a rate mitigates this, shading a total does not, and a bubble overlay is the usual answer.

Common mistake 3 with India - district map

Inferring a within-state relationship from the map and then testing it on the same data without accounting for spatial dependence. Neighbouring districts are not independent observations, which is what Spatial autocorrelation is for.

Common mistake 4 with India - district map

Mapping a count rather than a rate. A district map of totals is a map of population, and it will look the same whatever the variable.

EconStatsCraft › India - district map

India - district map

Shades India by district - 641 of them, at ten times the resolution of the state map. The level at which most Indian administrative data is published and most within-state variation lives: a state average hides the fact that its richest district is often several times its poorest.

What it needs

District column - District names as the census writes them. Common variants match.
What to shade by

The intuition

A district map is a state map with more regions and less data behind each one. Finer geography looks more precise and is usually less reliable: the same survey spread over 700 districts instead of 30 states gives each unit a twentieth of the sample, so the extremes on the map are mostly the districts where the fewest people were asked.

When to use it

Any district-level Indian data, which is most of what the census, NREGA, the NFHS and the district statistical abstracts publish. Also the honest first look before any within-state claim, because a state average hides the thing you usually want: the richest district of a large state is often several times its poorest.

Where the estimator comes from

For a district with n_d observations the sampling variance of its mean is \sigma^2/n_d, so the SPREAD of estimates across districts is inflated by sampling noise, not just by real variation. The map's extremes are therefore biased outward, and systematically so: the smallest n_d produce the most extreme \bar{v}_d in both directions. Empirical Bayes shrinkage is the standard repair, pulling each district towards the overall mean by a factor that depends on its own sample size - exactly the credibility weight Z=n/(n+k) used in insurance.

What it assumes

That your column names districts as the census does. District names are far less standardised than state names - the same place appears as Bangalore and Bengaluru, Gurgaon and Gurugram, Allahabad and Prayagraj - and the map names every row it could not match rather than dropping it quietly. Read that list.
That the district set matches the boundary file, which is the pre-2019 configuration. Districts have been created and split repeatedly since: Telangana went from 10 to 33, Jammu and Kashmir was reorganised, and several states have added districts. Recent data will carry names this file has never heard of, and the unmatched list is where you will see that.
One value per district. A district panel needs its time column named so the map knows which year it is drawing.
The district boundaries match the vintage of your data. Indian districts have been split repeatedly, so a dataset collected before a split cannot be mapped onto current boundaries without deciding how to allocate the parts.

Statistical properties

The most extreme districts on any raw map are usually the smallest samples. This is the small-numbers problem, and it makes rare-event maps - rare cancers, rare crimes - nearly useless untreated.
Shrinkage towards the state or national mean is the standard fix and is the same arithmetic as actuarial credibility: weight each district's own estimate by n/(n+k).
Show the sample size somewhere. A companion map of n, or crosshatching for thin districts, keeps the reader from over-reading the tail.
District boundaries change frequently. India has gone from 640 districts at the 2011 census to well over 750, so a map joining recent data to 2011 geometry silently drops or mismatches units.
At district scale a name join will fail on spelling, transliteration and renaming. Join on a stable code, and always report how many units failed to match.
More regions means more multiple comparisons. Some district will look extreme at any threshold, and pointing at it after the fact is not a finding.

How to read the output

At 641 units the eye reads PATTERN, not individual districts. Look for north-south and coastal-inland gradients, and for clusters that cross state borders - the last of those is the interesting one, because it is what a state-level analysis cannot see at all.
Individual districts are too small to read off the map; use the table underneath for the extremes. At this resolution one outlier district can own a whole colour class if you cut by equal interval, so quantiles are the safer default here.
Grey means unmatched. With 641 units a handful of grey districts is normal, while a grey REGION is a naming mismatch worth fixing rather than accepting.
Small districts are visually invisible however extreme their values, and large sparsely populated ones dominate the picture. Area is geography, not importance.

A worked reading

A district map of a rare outcome will always put its darkest colours on the least populous districts, because one event in a district of 5,000 is a rate of 20 per 100,000 while the same event in a district of 500,000 is 0.2. The map appears to show a geographic pattern and is showing a population pattern. Shrinkage fixes it arithmetically, and the diagnostic is simple: plot each district's rate against its denominator, and if the extremes all sit at the left of that plot the map is displaying sample size.

Symptom → diagnosis → what to do

If you see this	It means	Do this
A few districts far darker than the rest	Small samples, most likely	Map the sample size beside it
Salt-and-pepper pattern	Noise dominating	Shrink towards the regional mean
Districts missing from the map	Boundary vintage or a failed name join	Join on codes and report unmatched counts
A very smooth map	Already shrunk, or a modelled surface	Check what was mapped
A striking cluster	Multiple comparisons across hundreds of units	Test it, do not just point at it
Rates from tiny denominators	One event in a small district gives a huge rate	Suppress below a minimum denominator

Common mistakes

Treating district boundaries as fixed. They are not, and a series that spans a split is not a series about one place - it is two places sharing a name.
Reading a district choropleth as if area meant importance. Rajasthan’s desert districts are enormous and thinly populated; Mumbai is a dot with more people than most states. Shading a rate mitigates this, shading a total does not, and a bubble overlay is the usual answer.
Inferring a within-state relationship from the map and then testing it on the same data without accounting for spatial dependence. Neighbouring districts are not independent observations, which is what Spatial autocorrelation is for.
Mapping a count rather than a rate. A district map of totals is a map of population, and it will look the same whatever the variable.

Where to read this properly

Efron and Morris (1975), "Data Analysis Using Stein's Estimator", JASA 70(350).
Gelman and Price (1999), "All Maps of Parameter Estimates are Misleading", Statistics in Medicine 18(23).
Openshaw (1984), The Modifiable Areal Unit Problem.

All analyses

Data & variables

Each example is simulated from a known model, so you can check whether the estimator recovers the truth - and nothing here is anyone's private data.

Filter variables

Drag a variable into a slot on the right - or just click it.

Choose an example dataset above to begin.

Choose a transformation - it opens in the data editor, with a live preview.

Apply to

Multiply by

Columns ctrl-click for several

Group by ctrl-click for several

Within groups

Statistic

Bands

How many

Cut points

Keep rows where

Value

and

Fill with

Value

Also mark which rows were filled

Rules

Base period = 100

Which part

Percent at each tail

Threshold

Survey weight optional

Direction

Names column

Values column

Largest first

Delete them, not just count them

Read this column as

Tidy the text

New name optional

EconStatsCraft rigorous, plain-spoken econometrics

Econometrics that explains itself.

Regression, panel data, time series and the whole battery of tests. Every result is followed by a verdict you can defend.

Load a dataset to see it here.

India - district map

What it needs

The intuition

When to use it

Where the estimator comes from

What it assumes

Statistical properties

How to read the output

A worked reading

Symptom → diagnosis → what to do

Common mistakes

Where to read this properly

See also

Related Maps

All analyses

Charts

Maps

Explore

Regression

Time series - one series

Time series - several series

Panel data

Multivariate statistics

Marketing research

Causal

Finance & risk

Actuarial

Tests & breaks

Machine learning

Econometrics that explains itself.