How to read the data without getting it wrong
A map of self-reported bribes is a map of two things at once: where bribes happen, and where people report them. Keeping those apart is most of the skill in reading it.
These are accounts, not findings
Every record is an unverified first-hand account. Nothing here has been adjudicated, and no individual is named. The right description is “people report that”, never “it is established that”.
A high count is not the same as a high rate
A city with more reports may have more corruption, more people, more internet, or one person who told their neighbours about this site. Without a denominator — how many people used that service at all — counts across places are not comparable. Compare shapes instead: which services recur, what the amounts cluster around, how often refusing worked.
What the floor hides, and why
Any slice with fewer than five reports is suppressed. That is a privacy rule, not a data-quality one: a single report from a small town, broken down far enough, stops being anonymous. So an empty cell means “too few to show safely”, not “none here”. Reading blank as zero is the most common mistake made with this dataset.
Amounts are what was asked, not what is normal
The amount demanded is what one person was told to pay on one day. Medians across a large enough group are meaningful; a single large figure is an anecdote. The site only shows a median where the group already clears the floor.
The refusal rate is the most useful number
How often people refused, and whether they still got the service, says something the count cannot: whether the demand had teeth. It is also the number most likely to change behaviour, because it answers the question everyone actually asks — what happens if I say no?
Citing it
Say what it is: an open, anonymised dataset of self-reported bribe demands, unverified, with small slices suppressed. Link the page you took the figure from and give the date, because the figures move as reports arrive. The full export is available for anyone who wants to check the working.
Every report makes the next comparison a little more honest.
Report a bribe