The Statistics That Matter in Local Stories

Most local data stories need about six statistical ideas. None of them requires more than division. Every one of them gets botched in print weekly, usually by someone who knew better and was in a hurry.

Mean or median

The mean is what most people call the average: add everything, divide by the count. The median is the middle value after sorting. They agree when data is balanced. They split apart when a few values are extreme.

Use the median for incomes, home prices, rents, salaries and anything else with a long tail on one side. That is why the Census Bureau reports median household income. Use the mean when the total matters, such as average cost per mile when you will multiply back up to a budget. When in doubt, report the median and say so.

Percent versus percentage points

When something measured in percent changes, you can describe the change in percentage points or in percent, and those are different numbers.

Measure	Before	After	Change in percentage points	Percent change
County unemployment rate	3.0%	4.5%	up 1.5 points	up 50%
Levy approval vote	48%	52%	up 4 points	up 8.3%
Graduation rate	90%	85%	down 5 points	down 5.6%

"Unemployment rose 1.5 percent" is wrong, and the mistake is not small. It suggests a change of 0.045 points, from 3.0 to about 3.045. Write "rose 1.5 percentage points." If you want the relative change, write "the rate rose by half" and show both numbers.

Rates: per 1,000 and per 100,000

Bigger places have more of everything. Raw counts mostly measure population. To compare, divide by population and multiply by a round base.

Choose the base so the result is a readable number. Per 100,000 is standard for crime, deaths and disease. Per 1,000 suits school data: 37 suspensions among 1,460 students is 25.3 per 1,000; 12 suspensions among 310 students is 38.7 per 1,000. Use the same base and the same year of population for every place you compare.

And pick the right denominator. Crashes per resident ignores the interstate running through a county with few residents. Crashes per million vehicle miles traveled, which state transportation departments publish, may be fairer. Name the denominator in the copy.

Inflation: compare real dollars

A dollar in 2015 bought more than a dollar in 2024. Comparing budgets across years without adjusting is comparing in two currencies. The usual tool is the BLS Consumer Price Index for All Urban Consumers (CPI-U), U.S. city average. Its annual averages: 237.017 for 2015 and 313.689 for 2024. The BLS online inflation calculator uses this same series.

Construction costs have their own price indexes, and they do not always move with the CPI. If a county engineer says paving costs rose faster than general inflation, ask which index they used and check it.

Base effects and small numbers

Percent change on a small base produces dramatic, meaningless numbers. Robberies in a town of 1,800 went from 2 to 6. That is a 200 percent increase, and a jump from 111 to 333 per 100,000. It is also four robberies. One person could be responsible for all of them.

Federal health data builds the warning in. CDC WONDER marks death rates as unreliable when the count is under 20, and suppresses counts from 1 to 9. Borrow that habit:

• Give the counts when they are small. "Six robberies, up from two."
• Combine years. Three-year or five-year totals smooth out noise.
• Check the base year. A change measured from 2020, when much of life shut down, can look like a boom that is only a return to normal.

The average of averages

Averaging averages gives every group the same weight regardless of size. It is wrong whenever the groups differ in size.

The same trap hides in county averages of city per capita incomes, average tax rates across districts, and averages of percentages. In a spreadsheet, =SUMPRODUCT(B2:B4,C2:C4)/SUM(B2:B4) gives the weighted average in one step.