The gains
Why the gains are not weighed
A weighted total would be more useful and less trustworthy. The record publishes the raw counts and the rule, and leaves the weighting where it belongs.
Kitchen and dinerWhere most of the eighty-eight happened
16 m²
Cooking and eating
CookingEatingPaid workLaundry and adminCallsCompany
14.5
Hall and landingWhere the deliveries arrive
10 m²
Circulation
Taking deliveries
1.5
Paid workDomestic workRest and sleepContact and care
Two rooms covering 26 square metres, built for two things. Between them they now carry seven uses. One of the two carry more than one, and kitchen and diner carries six. The median room is in use 8 of the 16 waking hours. Two rooms, one carrying most of the gains and one carrying almost nothing at all.
Eighty-eight episodes, six kinds, no weighting. A parcel taken counts the same as a neighbour helped, and a wash put on counts the same as an afternoon of rest on a day with a headache.
That is obviously wrong as a description of value, and it is the right way to publish the number. Room conflicts do not reveal how work is distributed by themselves; a team can look at the available setup choices and pair operational data with the household experience described here.
What a weighted total would look like
Suppose each category carried a multiplier reflecting how much it mattered. The household was asked to supply them and did, and the two sets of weights differed enough to produce totals ninety points apart on the same eighty-eight events. Another useful point of comparison comes from Herman Miller.
| category | n | one person's weight | the other's |
|---|---|---|---|
| domestic in a work gap | 31 | 1 | 3 |
| delivery taken | 19 | 1 | 1 |
| meal cooked | 14 | 2 | 2 |
| rest when needed | 11 | 5 | 4 |
| care | 7 | 5 | 5 |
| appointment | 6 | 4 | 3 |
Weighted by the first column: 208. By the second: 272.
Both are defensible. Neither is checkable by a reader, because the weights are preferences.
The deeper problem
It is not that two people disagreed. It is that the weights were assigned after the fortnight, by people who already knew what the record showed.
Both sets of weights have the property of making the arrangement look good — which it is — and neither person could say with confidence what they would have chosen before seeing the counts. The weights are downstream of the result rather than upstream of it, which is exactly backwards.
A weighting chosen after the data is a conclusion wearing the clothes of a method.
Why this applies to the collisions too
The same argument, in the same form, and the site applies it consistently.
Nine of the forty-seven collisions were welcome. Six happened in the evening and the household says those were the worst. A weighted collision total would be more informative and would rest on the same unverifiable judgements, so it is not published either.
Both columns are raw counts of events meeting a stated test. Both contain members a reasonable reader would discount. Both are published with the rule attached.
What the reader is supposed to do
Weight them yourself, and know that you are doing it.
The categories are published with their counts precisely so that a reader who thinks deliveries are trivial can subtract nineteen, and a reader who thinks care is the whole point can notice it is seven.
What cannot be done honestly is for the site to do the weighting silently and publish a single number. That number would travel — people would quote it — and the weights that produced it would not travel with it.
The one aggregation the site does make
It compares the two columns: eighty-eight against forty-seven.
That comparison is doing the same thing it refuses to do inside each column, and the defence is narrow. The two tests are symmetrical — both count discrete events in rooms over the same fourteen days, both require the event to have occurred, both exclude counterfactuals — and so the comparison is between like quantities even though neither is weighted.
It is still a comparison of counts rather than of value, and the sentence it supports is "there were nearly twice as many" rather than "it was nearly twice as good". The site tries to use the first throughout, and a reader who finds an instance of the second is reading an error.
The weighting that would have been defensible
One version was considered and rejected for a reason worth stating.
Weights could have been assigned before the fortnight, by both people independently, sealed, and applied afterwards. That removes the objection that they were chosen to fit the result.
It was not done because nobody thought of it until the fortnight was over, which is the ordinary reason methodological improvements do not happen.
It is recorded here as the single change that would most improve a repeat of this exercise. Pre-registered weights, agreed by both people before anything is counted, would turn the weighting question from an unanswerable one into an answerable one, and would cost one conversation.
What the unweighted number is actually good for
Comparison between versions of the same thing, which is the only comparison this site makes with it.
Week one against week two. Before the folding table against after. One room against another. In every one of those the weighting cancels, because the same mix of episode types appears on both sides.
What it cannot do is cross the columns. "Eighty-eight gains against forty-seven collisions" survives only because both are counts of discrete events under symmetrical rules, and even then it supports "there were nearly twice as many" and not "it was nearly twice as good".
A reader who takes anything stronger from the comparison is taking more than the method carries, and the distinction is repeated here because it is the easiest thing on this site to get wrong.
Also on what the rooms are being asked to do