The hour
Why the evening band rises again
Sittings beginning after five run forty-seven minutes, longer than the early afternoon. Twenty-three observations from four dwellings, and why that is too few.
The last band of the day recovers. Sittings beginning between five and seven run a median of forty-seven minutes, against thirty-one in the early afternoon.
It is the second largest rise anywhere in the record and it rests on twenty-three sittings from four of the nine dwellings.
What the band contains
sittings 23
homes contributing 4
median 47 min
endings 7 of 23
bare 16 (70%)
Seventy per cent bare is the highest share of any band. Seven endings in twenty-three sittings is the lowest rate. These household notes describe the setting, not the whole workday, so decision-makers can look through the available controls and then interpret any system data beside outcomes and direct employee feedback.
Whatever is happening after five in these four dwellings, the building is largely leaving people alone. For broader research beyond this small household record, consult work published by UC Berkeley.
Why twenty-three is not enough
Four dwellings out of nine, and the four are not a random four.
They are the households that logged work after five at all, which means they are the households whose working days extend into the evening. A person still at a desk at half past five is, by selection, a person whose day is not structured around a five o'clock finish.
That is a confound that cannot be removed. The evening band is not a measurement of what evenings are like; it is a measurement of what evenings are like for people who work in them.
The bands before five describe nine dwellings. The band after five describes four self-selected ones, and the two are presented on the same chart.
This is the weakest figure on the site and it is printed at the same size as the others, which is a defect of the drawing rather than of the data.
What the four have in common
Homes B, D, E and H.
Three of those four are the three longest-median dwellings in the record. All four have a working place that is either private or uncontested — a sofa in a flat with nobody else in it, a bedroom desk, a desk, a study.
So the evening band is substantially a measurement of the dwellings that were already producing the longest stretches, measured at the hour when the rest of the household has stopped needing anything.
Both of those push the same way, and together they are enough to explain the rise without anything else being true.
The one testable part
If the rise were real and general, the dwellings that contribute to it should show the rise within themselves — their own evening median should exceed their own afternoon median.
They do, all four:
| home | 13–15 | 17–19 |
|---|---|---|
| B | 46 | 58 |
| D | 61 | 72 |
| E | 33 | 41 |
| H | 48 | 55 |
Four for four, with gaps of eight to twelve minutes.
That is a within-dwelling comparison and it removes the selection problem between homes. It does not remove the selection problem within a day: a person who chooses to work at six is choosing a time, and may be choosing it precisely because the conditions are good.
Three readings
The house empties. Other people stop needing things, which removes wanted endings. The band has one in twenty-three, against nine in forty-one at midday.
The sun stops mattering. One light ending in the band, against thirteen in the two midday bands. By six the thing that drove a quarter of all endings has ceased to be a factor.
The work is different. Evening work may be tidying-up work: long, low-demand, uninterrupted. The record cannot see this and it is as good an explanation as the other two.
What this page is for
Chiefly to mark a figure as weak.
A chart with six bars invites the reader to treat the six as comparable, and the sixth is not comparable with the other five. It comes from fewer than half the dwellings, from a selected part of the day, and it rests on a tenth of the record.
Anybody reading the hour chart should read the first five bands as a finding and the sixth as a question.
What the band would need to be worth something
Three things, none of which this record has.
All nine dwellings contributing. Five homes logged nothing after five, and their absence is not random: they are the homes whose working day ends earlier, which is probably correlated with everything else about how their day is shaped.
More than twenty-three observations. A median on twenty-three entries moves several minutes if two of them are reclassified.
Some way of separating choice from condition. A person working at six has chosen to, and may have chosen because the conditions are good. The record cannot distinguish a dwelling that is better in the evening from a person who works in the evening when the dwelling is better.
Why it is drawn at all
Leaving the sixth bar off the chart would have been a cleaner presentation and a selection.
The band exists in the data. Twenty-three sittings were logged after five and they have a median. Omitting them because they are inconvenient to interpret would be exactly the kind of tidying this record is supposed to avoid, and a reader who is told which bar is weak is better served than one who is shown five bars and not told a sixth was dropped.