Commute Math: Measuring the Drive You'll Actually Make

Ask a map app for the drive between two addresses at ten on a Tuesday morning and it hands back a tidy number — fourteen minutes, say — with a green line and total confidence. Ask the same app for the same route the way you would actually drive it, leaving at 7:45 on a weekday, and the number can climb past twenty-five. Nothing about the first number was wrong. It just answered a question nobody asked: how long does this take when almost nobody else is on the road.

That gap is the whole subject of this piece. Not "is the commute bad," which is a feeling, but how to pull an actual number for the trip you would actually make, at the hour you would actually make it, more than once, from sources that will hold still long enough to compare two addresses fairly.

Why the default number is a snapshot, not a forecast

Open a map app cold and search a drive, and most of them quietly default to "leave now." That's a real-time or near-real-time read of the route — useful for the drive you're about to take, useless for judging a commute six weeks from now on a lease you haven't signed. The traffic on the route at 2pm on a Sunday, when a lot of house-hunting research actually happens, tells you almost nothing about the traffic on that same route at 8am on a Wednesday.

Google's own developer documentation for the Directions API draws the line precisely. A driving request that carries a departure_time comes back with a duration_in_traffic field, which the documentation describes as "the predicted time in traffic based on historical averages" blended with live conditions. Leave departure_time out and the same page is equally explicit about what you get instead: "choice of route and duration are based on road network and average time-independent traffic conditions." That is not a traffic-free number, and it is not tonight's number either. It is an average with the clock taken out of it.

You don't need to touch the API to get the benefit. The same predictive engine sits behind the ordinary "Depart at" / "Arrive by" option in the consumer app and on the website. Set directions to a place, open the More menu at the top of the route, pick a future date and time, and the estimate is recalculated for the traffic expected then rather than for whatever is on the road at the moment you happened to be looking.

Setting a real departure time instead of trusting the default

The mechanics take under a minute and they're worth doing for every candidate commute, not just the one you like best.

1. Enter both addresses — the actual ones, not the neighborhood centroid or the zip code.
2. Choose driving mode, then open the option to set a time rather than leaving it on "now."
3. Set "Depart at" for the real morning you'd actually leave — 7:45, not 8:00 rounded — on a representative weekday, and separately set "Arrive by" if the constraint that matters is a start time rather than a leave time. The two aren't symmetric: arriving by a set time works backward through predicted congestion to a suggested departure, while departing at a set time works forward to a predicted arrival.
4. Do the same for the return leg at the actual end-of-day time, since evening congestion patterns rarely mirror the morning ones on the same road.
5. Write down the predicted duration next to the free-flow "leave now" duration for the same route. The gap between them is a rough measure of how congested that specific corridor gets — a route where the two numbers barely differ is a genuinely different commute from one where they differ by fifteen minutes, even if the free-flow numbers looked identical.

Two limits are printed in Google's own help page for the feature, and both bite on a real commute: it is "only available if you have one destination," and it is "only available for Driving and Transit." A school drop-off followed by the office will not go into one predictive query, and a bike or walking leg will not go in at all. Build them as separate depart-at queries and add the pieces up yourself.

Running it as a week, not a guess

A single query, even a good one, is still one data point pulled from a historical average. Treat the first reading as a starting figure and confirm it across the actual days you'd be driving, not just the one you happened to check.

The table below is a worked example rather than a measurement — illustrative numbers, not a route anyone drove — showing the shape the exercise produces: one route, five weekday mornings, departure fixed at 7:45.

The free-flow number never moves, because it's describing an empty road that doesn't exist at 7:45. The predicted number does move, and the spread — 26 to 34 minutes in this example — is the actual answer to "how long is this commute," not any single cell in that column. Your own five cells will not look like these. The point is that five of them exist, and that the worst one is a morning you will actually have. A commute worth signing a lease over is the average of a realistic range, plus a look at the worst day, not the best-case number a listing agent will happily quote you.

If you can, replace a few of those historical predictions with a real drive or a real transit ride before you commit — the predictive model is built from patterns, and patterns miss the specific: a school with a staggered pickup line two blocks off your route, a single-lane bridge under repair, a traffic signal that's newer than the historical data feeding the model.

What the Census Bureau's number is actually measuring

It's tempting to reach for American Community Survey commuting tables here, since the Census Bureau collects exactly this topic — travel time to work — for every state, metro, county and tract in the country. It's worth understanding precisely what that number is before using it, because it answers a different question than the one above.

The ACS commuting questions ask each respondent to report the total number of minutes it usually took them to get from home to work, during the survey's reference week — a self-reported estimate of a typical trip, not a measured one. The Census Bureau aggregates those self-reports into a mean travel time for whatever geography you're viewing, published in subject table S0801 (Commuting Characteristics by Sex) and detailed table B08303 (Travel Time to Work), both available at data.census.gov. A companion table, B08302 (Time of Departure to Go to Work), breaks the same population into fourteen departure bands — half-hour wide from 5:00 a.m. through 8:59 a.m., coarser on either side, with everything from noon to 3:59 p.m. dumped in one bucket. That resolution is the point. It is a direct read on whether a town's workers pile into one crunch half-hour or spread across three.

What that gives you: a real, defensible sense of what an entire tract's or metro's workers experience on average, useful for comparing two towns' general commuting burden the way the Census tables covering who lives in a tract let you compare their demographics. What it can't give you: a number for your specific address to your specific employer's address, because nobody was ever asked about that exact pair of points. Like the rest of the American Community Survey, the geography you're viewing matters more than the number appears to — as does the vintage, since the standard local release remains the five-year file, whose center of gravity sits a couple of years behind the release date. Pull S0801 and B08303 for the same geography, at the same vintage, before you compare two candidate towns on this axis, or you're comparing a 2022-centered estimate for one place against a different vintage for the other without knowing it.

The traffic count that tells you whether a road is already full

A depart-at query tells you how long the drive takes. It doesn't tell you why, and it won't tell you whether the road has room to get worse — a new subdivision two exits up, a distribution center under construction, a lane closure that becomes permanent. For that, the number to pull is Annual Average Daily Traffic, or AADT: the estimated total number of vehicles that cross a given point on a road in both directions over a full day, averaged across the year.

Nearly every state department of transportation publishes an interactive AADT map, so the fastest way in is to search "[your state] DOT AADT map" or "[your state] DOT traffic counts."

Read the method note on whichever one you land on before you read the number off it. California states its own plainly in the description attached to its data: AADT is "the total volume for the year divided by 365 days," the count year runs 1 October to 30 September, and "very few locations in California are actually counted continuously" — most segments are short samples adjusted afterward for seasonal and weekly variation, published through the department's traffic census program. Other states work the same way. What you are reading is an estimate with a method behind it, not a tally. The federal dataset underneath all of them is the Federal Highway Administration's Highway Performance Monitoring System, which every state reports into and which FHWA republishes as public geospatial data by state, segment by segment, with the explicit caution that it's built for planning, not turn-by-turn navigation.

Some states go further and publish the peak-hour volume alongside the daily average, which is the more useful number for a commute question — it tells you what share of a road's daily traffic is crammed into its worst hour. California's public traffic-volume service is a working example, queried 19 August 2026 and unchanged when re-checked the next day. Its records are keyed to named count locations rather than to addresses, which is worth knowing before you go looking for your own street. At the location Caltrans labels LOS ANGELES, FIRST STREET on US-101, post mile 0.907, the ahead segment reports an AADT of 114,000 vehicles against a peak hour of 8,100 — about 7% of the whole day inside that single hour. Move out to EAST SHINGLE SPRINGS on US-50 in El Dorado County, post mile 10.295, and the next location east at GREENSTONE ROAD, post mile 12.19, and the same two fields give an AADT between 43,500 and 49,500 with a peak hour of 3,950 to 4,050.

That is a similar 8–9% share on a road carrying well under half the daily volume. Two roads can post nearly the same share of traffic in their peak hour and still be entirely different commutes, because one of them is moving two and a half times the vehicles to begin with. The share alone won't tell you that, and a raw AADT map won't either unless the peak-hour field is broken out separately. Not every state publishes it, so check what your state's map actually hands you before assuming it works like this one.

None of this replaces the depart-at reading. It explains it. A route with a high AADT and a fat peak-hour share is the kind of road where next year's traffic count is more likely to be worse than better, which matters if the plan is to live somewhere for longer than one lease term.

The last mile the map never counts

Every method above measures curb to curb, and a commute rarely starts or ends at a curb. Add the pieces a map query silently drops:

• Parking search time, if the destination doesn't have an assigned spot — five minutes on a good day, considerably more in a building with known crunch hours.
• The walk from the car or the transit stop to the actual desk, which runs seven or eight minutes on a large campus, in a hospital, or inside a government complex where the visitor entrance is not the one you use.

Transit adds a third piece the driving numbers never contain: transfer and wait time. Schedule-based trip planners handle that better than driving directions do, but every minute of it assumes the posted schedule is the real one, which is a claim worth testing on the day you visit rather than on the app.

And any stop the commute actually includes — a daycare drop-off, a park-and-ride leg — gets priced as its own segment with its own depart-at query. Folding it into one estimate for the whole chain is how a twenty-five-minute commute quietly becomes forty.

All of it gets added to the depart-at range, never to the free-flow number, and what comes out is closer to door-to-desk than to address-to-address. It's also worth checking this against the same rental verification habits that apply to any address you haven't stood on yet: a leasing office quoting "twenty minutes downtown" is describing the free-flow number, on the route they'd choose, at an hour they didn't specify.

Putting a range on the line, not a guess

Bring this back to a single sheet and it looks like the transportation line in the seven-number cost frame this site runs on elsewhere — except here the unit is minutes and trips instead of dollars. Three columns cover it: the depart-at range across a real week for the actual commute hours, the AADT and peak-hour share for the specific road if the state publishes one, and the last-mile add-on for parking, walking or transfer. Mark each cell the way any of these sheets should be marked — a number you measured yourself outranks one you read off a map that was guessing at a future Tuesday, and a map's guess outranks a listing agent's adjective.

What you end up with isn't a single number you can defend to a spreadsheet, and it shouldn't be. It's a range, built from a source you can name for each piece of it, which is the only kind of commute estimate worth signing a lease against.

Sources, read 19 August 2026, with the Caltrans figures re-queried 20 August 2026: Google Maps Platform Directions API — traffic and departure time, for how departure_time and duration_in_traffic are calculated; Google Maps Help — setting a depart or arrive time, for the consumer-app mechanics; Census Bureau — why we ask about commuting and data.census.gov for ACS tables S0801, B08303 and B08302; FHWA Highway Performance Monitoring System — public geospatial data; California Department of Transportation public Traffic Volumes (AADT) feature service and its traffic census program page, queried directly for US-101 at post mile 0.907 in Los Angeles County and US-50 at post miles 10.295 and 12.19 in El Dorado County.

No transportation-planning or real estate license sits behind any of this. What sits behind it is a documented API, a help page, four Census table numbers and one state's count file — all of which will answer to your two addresses as readily as they answered here.