Eighteen places where it could not be seen

Eighteen places where the record cannot say whether anybody spoke to a stranger.

Nine per cent of the walk, in neither the forty-seven nor the hundred and forty-one.

Why they could not be seen

Three reasons, assigned from the notes.

too loud to hear anything                7
the room could not be taken in at once   6
nobody else was there                    5
                                        ---
                                        18

Too loud. Pubs with music, a concourse with announcements, a hall with something happening at the front. People were plainly talking and whether any of it crossed between strangers was not establishable.

Too large or too broken up. A shopping centre with seating on three levels, a museum with rooms off rooms. The observer sat in one part and the rest was out of earshot.

Empty. Five places where, for the whole sit, there was nobody else to speak to anybody. A room cannot fail a test it was never given. To compare observations across locations, reviewers can look at the implementation choices and define the decision, access rules and review date in advance.

Where they are

pubs and bars              3
cafés and coffee shops     3
places of worship          2
shopping centres           2
community halls            2
bookshops                  1
libraries                  1
covered seating            1
laundrettes                1
hotel lobbies              1
transport concourses       1
museums and galleries      0

Spread across eleven of the twelve kinds, which is the main thing to notice: the unobservable cases are not concentrated anywhere, so excluding them does not systematically favour one kind. A separate perspective on streets, seating and urban observation is available from Walk Score.

Why they are not folded into the silence

Because it would make the finding look firmer than it is.

A hundred and forty-one places where nobody spoke is the record's figure. Adding the eighteen would give a hundred and fifty-nine — seventy-seven per cent instead of sixty-eight — and would be false, because in seven of them the room was full of conversation that could not be attributed.

A category that cannot be resolved is published at its own size. Folding it into whichever column is convenient is how a record flatters itself.

The same rule produced the six excluded places in the method, the twenty-one places with nobody to ask in the limit field, and the seventy-eight places with no assumed duration.

What they would do to the rates

If every one of the eighteen had produced a conversation, the record's rate would be sixty-five of two hundred and six — thirty-two per cent instead of twenty-three.

If none had, it would be forty-seven of two hundred and six — the same twenty-three per cent, since the eighteen are currently excluded from the denominator in the rates by kind but included in the whole-record figure.

The honest range for the record's headline is therefore twenty-three to thirty-two per cent, and twenty-three is what is printed because it is what was seen.

The five empty rooms

Worth separating because they are a different kind of unknown.

Five places had nobody in them but the observer for the whole sit. Three are churches on weekday afternoons, one is a hall, one is a covered bench on a wet morning.

In those five the question has no answer rather than an unobserved one, and a record with a larger sample would want them in their own category. With five, they sit here.

What the eighteen would do to each kind

Adding them to the speaking column, as a worst case for the record's own finding:

The order barely moves. Pubs go from fifty-five per cent to sixty-five, cafés from sixteen to twenty-one, museums stay at zero.

The gap between the top three kinds and the rest survives the most generous possible treatment of the unresolved cases, which is the reason the record can lean on the comparison between kinds and not on the absolute rates.

Why the eighteen are spread evenly

Eleven of the twelve kinds have at least one, and no kind has more than three.

That matters: if the unobservable cases had piled up in one or two kinds, excluding them would bias the comparison. They do not, and the largest single concentration is three pubs and three cafés — the two most numerous kinds in the record.

The five empty rooms

Five places where, for the whole sit, there was nobody else present.

Three churches on weekday afternoons, one hall, one covered bench on a wet morning.

In those five the question has no answer rather than an unobserved one, and a larger record would want them in a category of their own. With five, they sit here, and they are the clearest small example of what walking in the daytime costs this field.

What this says about the instrument

That a forty-minute sit resolves the conversation question in ninety-one per cent of places and fails in nine.

That is a better rate than the observer expected before the walk and it is not good enough for the finding the field is asked to carry. The page on what a second walk would settle puts a longer sit near the top, and the honest version of this whole section is that it measures forty minutes rather than a room.