Why the median tells you more about your internet speed than the mean ever will.
Dave Noel · 9 min read
If you want your internet to sound faster than it is, report the average. Most people won't notice.
Say ten households on the same street run a speed test. Nine of them get 50 Mbps. One just got gigabit fiber and gets 1,000 Mbps.
One household moves the average, nobody else gets faster
Download speed of ten households on one street (a made-up example), Mbps
Nine households at 50 MbpsOne gigabit household
Mean = (9 × 50 + 1,000) ÷ 10 = 145 Mbps. Median = the middle value = 50 Mbps.
Which number describes that street? Nine out of ten neighbors would laugh at 145. One household nearly tripled the average, and nobody else's internet got a bit faster.
That's how averages behave when data is lopsided, and internet speeds are very lopsided. Most connections sit in a crowd near the middle, and a smaller group of very fast ones stretches way out to the right. Statisticians call that a right-skewed distribution. I call it the reason I don't trust a headline average speed.
The math is fine. The question is wrong.
The average is correct math. It just answers a different question than most people are asking. Total speed divided by everyone is the mean. What a typical person gets is the median. When someone says "the average customer," they almost always mean the second one.
Where this bites in real life:
Coverage reports. A county can look great because of one city, while the rural half of the county lives at the slow end.
Before and after. Upgrade a handful of power users and the average jumps, even if almost everyone else sees nothing.
Customer conversations. Signal strength (RSSI or RSRP) does this all the time. A handful of readings taken right under a tower can pull the average up, and a few from deep inside a building can drag it down. What we need to show is the most common experience for the user, not a few extreme readings at either end.
Most serious speed reporting leans on medians for exactly this reason.
What I do instead
Lead with the median. It's the closest thing to "what a normal person gets."
Show the spread. A median plus the 10th and 90th percentiles tells you what the slow end, the middle and the fast end look like.
Plot it when it matters. One histogram ends most arguments.
If you like this kind of thing, David Spiegelhalter's The Art of Statistics is a great read on how summaries can mislead.
So next time someone hands you an average speed, ask for the median. And if they ask why, scream "averages lie!" at the top of your lungs, or just send them this.
Update: I ran it on Texas
I promised I'd try this on real data, so here it is! I took Ookla's open data fixed broadband tiles for Q4 2025 and worked out the mean and the median download speed for every Texas county with at least 2,000 speed tests. That's 89 counties.
One thing to know first. Ookla's open data comes as roughly 600 m squares (tiles), and each tile gives you the average of its tests plus a count, not the individual tests. So the means here are exact, and the medians are estimated from the tile averages, weighted by how many tests each tile has. Close, though not perfect.
Statewide, the average plays it straight
Here's the twist I didn't expect. For Texas as a whole, the average actually undersells things a little: a mean of 392 Mbps against a median of 401 Mbps. So much for my headline!
Go county by county, though, and the street from the top of this note shows up again. In 58 of the 89 counties, the average beats the median. The biggest gaps are all smaller, more rural counties:
County
Mean (Mbps)
Median (Mbps)
Gap
Palo Pinto
262
162
+100
Austin (the county west of Houston, not the city!)
205
118
+87
Reeves*
137
62
+75
Hill
200
132
+68
Fannin
175
111
+64
A few fast connections pull the average up while the typical test stays slow. The big metros sit almost right on the line, because they have lots of tests and lots of fast connections, so the two numbers agree.
*Reeves comes with an asterisk: one very busy tile holds a big share of its tests, so its numbers lean on that one spot.
The average flatters most Texas counties
Fixed download speed by county, Q4 2025, Mbps. Dots above the dashed line are counties where the mean beats the median.
Average beats median (58)Median beats average (31)
Source: Speedtest® by Ookla® open data, fixed tiles, Q4 2025. 89 counties with 2,000+ tests. *One busy tile holds a big share of Reeves County's tests.
Does money buy speed?
While I had every county lined up, I couldn't resist one more question. Do richer counties get faster internet? I pulled median household income for each county from the Census (American Community Survey, 5-year estimates for 2024) and put it next to the median fixed download speed.
The answer is: barely. The rank correlation is 0.18, where 0 means no relationship at all and 1 means income lines them up perfectly.
The counties that stand out are more fun than the number:
Cameron and Hidalgo, down in the Rio Grande Valley, have some of the lowest incomes in the group ($52,601 and $54,338) and some of the fastest medians (449 and 416 Mbps).
The slowest counties are middle income. Reeves (62 Mbps), Fannin (111) and Austin County (118) all sit between $64k and $76k.
At the other end, Kendall has one of the highest incomes ($114,962) and a median of 197 Mbps, while Fort Bend, at almost the same income ($114,041), gets 443.
This compares counties, not households, so it shows a pattern and can't tell you a cause. A county's median income and its median speed test can line up (or not) without any one family's paycheck deciding their speed.
Income barely predicts speed across Texas counties
Median household income vs median fixed download speed, Q4 2025. Rank correlation 0.18.
Counties called out aboveOther counties
Sources: Speedtest® by Ookla® open data, fixed tiles, Q4 2025; Census ACS 2024 5-year. 89 counties with 2,000+ tests.
Phones tell a slightly different story
Everything above is fixed broadband: home and office connections. Ookla publishes mobile (cellular) tiles as a separate dataset, so I asked the same question there for the same quarter. Mobile has far fewer tests per county, so I lowered the bar to 500 tests, which leaves 45 counties.
On mobile, the rank correlation with income is 0.28. A little stronger than fixed, still weak. And it wobbles: count only counties with 1,000+ mobile tests and it drops to 0.18, and when I ran Q4 2024 it came out at 0.41. I wouldn't lean on any one of those numbers.
The standouts flip, though, and I love that part:
Cameron and Hidalgo go from near the top on fixed to near the bottom on mobile (143 and 145 Mbps). Cameron's 306 Mbps drop from fixed to mobile is the biggest in the state.
Rockwall and Wichita are the only two counties where mobile beats fixed, by about 49 Mbps each.
One more thing jumped out. In Collin County, a single tile holds 90% of the county's 87,218 mobile tests that quarter, from just 68 devices. That one tile is 29% of all the Texas mobile tests I pulled for the quarter. Collin's mobile median is really just that one tile's average, so I drew it hollow on the chart. (A similar busy tile turned up in the Dallas-Fort Worth numbers for my Beryl story, and I've updated that one too.)
Fixed or mobile, income barely predicts speed
Median household income vs median download speed, Texas counties, Q4 2025. Same axes on both sides.
Counties called out aboveOther countiesOne tile holds most testsMiddle county
Sources: Speedtest® by Ookla® open data, Q4 2025; Census ACS 2024 5-year. Hollow means one tile holds over half the county's tests.
What I took from it
The note above still holds: in most Texas counties, the average makes the internet look faster than the typical test. But the statewide number was honest, which is a good reminder to check before you shout "averages lie!" at anyone.
And once you're looking at medians county by county, the interesting stuff is in the exceptions. Some low-income counties have great fixed broadband while some rich ones are stuck in the slow lane. Neither shows up in a single headline average.
Sources: Speedtest® by Ookla® Global Fixed and Mobile Network Performance Maps, fixed and mobile tiles, Q4 2025, CC BY-NC-SA 4.0. US Census Bureau, American Community Survey 5-year estimates (2024), table B19013, median household income. County boundaries from the US Census Bureau via plotly/datasets.