Chapter 10. Counts Compared: Definitions, Scopes and Provisional Totals

A count written during an epidemic is an act as much as a record. Someone defines a case, fixes a population and a period, decides who reports and by what rule a report becomes a case, and writes a total on a date. This chapter reads two such counts, both taken in the twenty-first century from a single institution's situation reporting: the World Health Organization's account of an Ebola Emergency Committee meeting in August 2014, and the outbreak-end total for Middle East Respiratory Syndrome in Korea in 2015. Both totals are qualified inside their own records. Both are then set beside the count cell of the ledger, where the codings of all eighty-nine cases ask a narrower question: does the case's own text give its count as settled, as qualified, as provisional, or not at all. What follows reads the two records first, then the ledger, then what a count would need in order to be read at the hour it was taken.

10.1Counting during an outbreak

A count of the share of cases who die, taken while an outbreak is running, carries three biases that do more than limit precision in this study's reading: they make the direction of movement ambiguous. A total that rises may show more disease, or a wider net of case-finding, or a change in the definition; a total that falls may show fewer cases, or a narrower definition, or a lag in reporting between the moment cases occurred and the moment they entered a table.

Deaths lag behind cases, so a ratio of deaths to cases taken early in an outbreak understates the share who die. Mild cases go uncounted, so the same ratio overstates it. Chapter 5's last section reads counts taken while the thing counted was still moving; this section applies those three biases to that situation, and the result is not merely a wider error bar but an ambiguity about direction. A count written at the hour of a decision carries no sign of its own of whether it moved because the disease moved, because the net of case-finding widened or narrowed, or because the rule that turns a report into a case was rewritten.

The two records of this chapter show what that ambiguity looks like when a count is dated. The Ebola record dates three things in one document: the Emergency Committee meeting on 6–7 August, the tallies reported on 5–6 August, and the situation report of 8 August 2014, which is the case's date in the ledger. What the record leaves undated is the closing moment of the tally itself: 68 new cases and 29 deaths are a figure over two days, and the record gives no hour at which they stop accumulating. It also leaves open whether the same case definition held at the two ends of any later comparison; the record states only what the definition included, test-confirmed cases together with probable and suspected ones. The MERS record dates the notification, 20 May 2015, and gives an outbreak-end total whose composition is stated: 185 cases in Korea and 1 in China, with 38 deaths beside them. What the record leaves undated is the period of those deaths beyond their pairing with the outbreak-end total, and it fixes no closing date for the total beyond its being the end of the outbreak.

What the act was, in each case, is therefore a total written at the hour of a decision, from a single institution's situation reporting, in which the definition of a case sits inside the number rather than beside it. What was known when the act was taken is, on the record's own evidence, only what the report carried — the definition's reach in the first case, the sum of two countries' figures in the second. Nothing in either record shows what any decider knew beyond the page in front of them, and this study does not supply it.

This ambiguity runs both ways at once, which is why a single total cannot by itself decide the direction of any measure taken beside it. In frame terms, which are Skeptron's recurring fifth-question condition and not this chapter's own finding, two totals taken at two moments are comparable only where the definition, the population, the unit and the closing moment of the count are the same at both moments. Where they are not, a comparison across two countries or two weeks compares the rules as much as the disease. A count taken at the hour of a decision is therefore provisional by construction: its later revision is not, by itself, evidence of anything the deciders did wrong. The two records of this chapter both carry that structure in their own words, and 10.2 and 10.3 quote them. In each case the qualification travels with the number — probable and suspected cases inside the definition in the first, two countries' figures summed in the second — and a reader who drops the qualification is no longer reading the record.

What a later reader can tell from these two records is narrow: which moments the documents separate, which definition applied at the hour of writing, and how each total was composed. What the reader cannot tell is whether a change in any total taken under these rules shows a change in the disease; that question needs the same definition, the same population and the same closing moment at both ends, and the records do not supply all three. The ledger's own count cell, where two codings of one model run twice asked whether each case's own text gives its count as settled, as qualified, as provisional or not at all, agreed on 78 of the 89 cases; the agreed values are set out in 10.4.

10.2Ebola in August 2014

The case is fixed in the ledger at 8 August 2014. In this study's rendering of the record, the setting is a page of situation reporting in which two periods sit side by side: an Emergency Committee that met on 6–7 August, and the tallies reported for 5–6 August, which stood at 68 new cases and 29 deaths.

This is a synthetic scene, built from the record alone and adding nothing to it. The place is the page: two dated lines set beside each other, one for the meeting and one for the reports. The roles the record names are the Emergency Committee and the World Health Organization's situation reporting; it names no one who drew up the tally, and the scene names none. The eye runs from one dated line to the other and stops at the number, and nothing on the page shows which of the 68 arrived by test and which were counted as probable or suspected. What is waited for, and what the page holds the answer to, is the definition that decides which reports count as cases.

A reader holding that page waits on one thing the page itself supplies: the definition that decides which of those 68 count as cases. The record's opening line keeps the meeting and the tallies apart:

The scenario goes on to state the definition's reach, and the record repeats it in its answer to the question on concrete facts:

Four moments, therefore, are distinguishable inside one document: the period the cases occurred in, the period they were tested in, the period they were reported in, and the period the committee met in. What the record dates is the meeting, the reporting window and the report; what it leaves undated is each case's own hour of occurrence and of testing, and any moment at which the tally closes. Skeptron's record of the boundaries of judgment answers the question whether all newly reported cases may be read as same-day confirmed infections, or as the final scale of the outbreak, with a refusal, and it names what it is separating:

Skeptron adds that it does not convert observed tallies into an individual's medical judgment. A later reader can therefore keep four moments apart: when the cases occurred, when they were tested, when they entered the tally, and when the committee met. What the record does not supply is a time at which the 68 and the 29 close. They are a count over two days under a definition that includes probable and suspected cases, and they are not, on the record's own terms, the size of the outbreak at any moment.

What the act was, on the record's own terms, is a published count: the situation report of 8 August, which carries the committee's meeting dates in one place and two days of tallies in another. What was known when it was taken is what the report declares about its own number — that probable and suspected cases were inside the total — and nothing further. The count cell reads qualified-by-definition-scope-or-denominator, the bearer cell reads not-named, and the revision cell reads not-named: apart from the studio's recurring fifth-question sentences, the case's own text names no later finding that would change how this count should be read.

The record's adopted source is a World Health Organization statement of 8 August 2014. No source text goes with this section, so nothing here paraphrases the statement; everything quoted here is Skeptron's record of what the statement contains, read in the English translation made once for this study. The absence matters for the reading in one direction only: a later reader of this study cannot check the statement's wording against the record's summary, and the study says so rather than filling the gap.

On the ledger's series axes the case reads thinly. The decider cell is unknown: the record names an Emergency Committee meeting and a situation report, and the codings found no decider stated or implied in the case's own text. The horizon cell is unknown, the reversal cell is unknown, and the human's place cell is unsettled. The gap between event and record is named. The count cell reads qualified-by-definition-scope-or-denominator: the definition that includes probable and suspected cases is part of the number. The bearer cell reads not-named, and the ledger's waits-and-pays cell reads unknown: the record does not say who bore the cost while the tally accumulated between the two days and the meeting. What would change the reading is named in none of the case's own sentences beyond the generic condition of its fifth question.

Two works bear on the surrounding history, each by its own reading note, and neither of them defines the 68. Suerie Moon and colleagues, in Will Ebola change the game? Ten essential reforms before the next pandemic: the report of the Harvard-LSHTM Independent Panel on the Global Response to Ebola (2015), report that the World Health Organization declared a public health emergency of international concern in August 2014, months after the outbreak was reported, and that governments and agencies were slow and poorly coordinated; their panel proposes ten reforms, among them incentives for early reporting, a dedicated centre for outbreak response, a transparent and accountable process for declaring emergencies, and stronger health systems. By Richards's note, local communities in Sierra Leone learned how Ebola spread and changed burial and the care of the sick, while the international response came late. That has a consequence for this chapter's count: the 68 accumulated over that period, and the committee's page records neither what communities changed nor anyone who bore the wait while the tally grew, so the note marks where the record's silence sits rather than filling it. The two notes stand at different distances from the count. Moon's panel speaks to the speed and coordination of the response and to the rules for declaring an emergency; Richards's note speaks to what communities did while the disease was running, at the very moment the tally accumulated. The record itself carries neither: it names no bearer of the wait and no one who drew the tally up. A later reader can place both works beside the count and cannot, from them, fill the record's silence about who waited while the 68 were reported.

What the record carries, and what a later reader can therefore test, is narrower: a date for the meeting, a date for the tally, a date for the report, and a case definition that includes probable and suspected cases. Whether the same definition was applied at the two ends of any later comparison is not in the record. This study reports that absence; it does not judge the decision to hold the meeting or the decision to publish the tally.

10.3MERS in Korea

The case is fixed in the ledger at 20 May 2015. Chapter 2 quotes the sentences of this record that give the date of notification and the date the regional emergency operations centre began operations; this chapter needs the sentence that gives the composition of the total instead, and it is the record's answer to the question on concrete facts:

Before that sentence is read on the ledger's axes, this study gives a scene. Synthetic scene. What follows is this study's rendering of the record and of the reading note beside it; it adds no event, number, person or outcome beyond what those two give. A man who had returned from the Middle East had visited several hospitals before his case was confirmed on 20 May 2015. At each of those hospitals people were at work in ordinary conditions — clinicians attending patients, wards receiving new admissions, administrators keeping records — and none of those people is named in the record or in the note. Confirmation reached the health office on a date the record fixes; what the confirmation consisted of, in the sense of a test or a set of criteria, the record does not state. What is waited for, in the reading, is twofold: the notification that begins the response, whose date Chapter 2 quotes, and a total that only arrives at the end of the outbreak, whose composition the sentence above fixes across two countries. In between those two hours sits a period of hospital transmission that the Korean agency's own report, read below by its note, describes as occurring overwhelmingly inside hospitals.

The total joins two countries. Any total read as the Korean outbreak is therefore a scope question as much as an epidemiological one, and the record does not settle which reading it intends: 185 cases were Korea's, one was China's, and the outbreak-end figure is the sum. The number of deaths given with it is 38. What the record dates is the notification, the activation of the response, and — through the outline of its own answer — the fact that the tally is an outbreak-end one; what it leaves undated is the period those 38 deaths cover beyond their pairing with the total, and the closing moment of the count itself, which is fixed only as the end of the outbreak rather than as an hour or a date.

What the act was, then: a total presented as the outbreak-end count, summing confirmed cases from two countries and pairing them with a death figure. What was known when it was taken, on the record's own evidence: that the cases counted were confirmed, that 185 of them fell inside Korea and 1 inside China, that 38 deaths were recorded beside them, and that transmission paths were a matter of estimation rather than of confirmed case counts. The record's answer to the question on boundaries of judgment refuses two readings at once, and names what it is separating:

So four moments are held apart: notification, activation of the response, the final tallies, and the scope of each country; and two kinds of statement are held apart inside the epidemiology itself, confirmed infections and estimated transmission paths. A later reader can therefore say which hour each of those belongs to, and can refuse the reading that dates all 186 cases back to 20 May or places all of them inside Korea. What the reader cannot say is how many cases the outbreak produced under any other definition, or over what period the 38 deaths accumulated, because the record supplies neither.

The record's fifth question asks what must be observed to keep or withdraw a comparison, and its answer says, in substance, that the result must be recalculated in the same unit and scope, and that if the population, the assets, the exchange or the observation end date differs, the result is not compared as the same measurement result. Skeptron labels that answer a comparative hypothesis, not a prediction that a future result will repeat the past. Chapter 2 sets out the studio's frame field, and the sentences above are that condition in the form the studio repeats across its history cases; they are the method of the record form here, not something this case shows about MERS.

The adopted source for this record is a World Health Organization page on the 2015 outbreak, and no source text goes with this section. The Korean agency's own report is cited by its reading note only. By that note, the report of the Korea Centers for Disease Control and Prevention, Middle East Respiratory Syndrome Coronavirus Outbreak in the Republic of Korea, 2015 (2015), records that the first case was a man returned from the Middle East, confirmed on 20 May 2015 after he had visited several hospitals; that transmission occurred overwhelmingly within hospitals, with a few patients infecting many others; and that the report reviews the response and the lessons drawn for hospital infection control. The note describes a report about transmission inside hospitals and about what the response taught; it does not give a total across countries, and this study does not add one to the record. What the study can put beside the record from the note is therefore narrow: that the first case preceded a body of hospital transmission, and that the agency's own writing about the outbreak was directed at hospital infection control.

The two documents sit at different distances from the total. The agency's report is close to the event and to its own confirmation of the first case; the record's adopted source is the later summary that fixes an outbreak-end count across two countries. A reader holding both can see that the confirmation date and the outbreak-end total are answers to different questions — when the first case was confirmed, and how many cases the outbreak produced under a definition that sums two countries. The record says so itself: it separates the notification from the final tallies, and it will not let a reader date all 186 cases back to 20 May or place all of them inside Korea.

On the ledger's axes, the case reads as follows. The decider cell is unknown; the horizon cell is unsettled; the reversal cell is unknown, and the human's place cell is unsettled. The act cell reads yes, and the act kind reads notice-declaration-or-certification. Act before proof reads yes: the notification of 20 May preceded the outbreak-end total by the structure of the record, which holds the two apart by date. The gap between event and record is named. The count cell reads qualified-by-definition-scope-or-denominator, because the total's scope — two countries, confirmed cases — is inside the number; among the twenty-three cases read closely this study counts five in that value, and this is one of them. The bearer cell reads not-named, and the ledger's waits-and-pays cell reads unknown: the record does not say who bore the cost between the first case and the outbreak-end total. The revision cell reads named for this case: the case's own answer to the fifth question gives the comparison condition as the boundary of the count, and this study reports it as that answer rather than as a prediction of any future result. The reasons cell reads stated, since the record carries the answer above; the ledger's series cells give no condition for revisiting this case's judgment in its own words.

What a later reader can tell from the record is therefore: the date of the first notification, the structure of the response's activation, the composition of the outbreak-end total, the death figure paired with it, and the boundary the record draws between confirmed infections and estimated transmission paths. What the reader cannot tell is the period the 38 deaths cover, the closing date of the count beyond the phrase "outbreak-end", the size of the outbreak under any other case definition, or who bore the cost while those figures accumulated. What would change how this record should be read is the subject of Chapter 11.

10.4Counts across the ledger

The count cell asks one thing of each record: does the case's own text give its count as settled, as qualified, as provisional, or not at all. It does not ask whether a count is correct or well founded. The two codings of one model run twice agreed on this cell in 78 of the eighty-nine cases, or 88 per cent; in the other 11 they differed and the ledger reports the cell as unsettled. The agreed values stand as follows.

The two columns are not the same field. For the eighty-nine the ledger marks each value as stated or implied where a coder found it in the case's words or inferred it from the setting; for the twenty-three read closely it gives plain values, and its unsettled count is the number of cases on which the codings differed. A majority of the agreed cases carry no count at all in the case's own text, the three no-count rows taking 48, 14 and 3 of them. Where a count does appear, thirteen agreed cases carry one: seven qualified by definition, scope or denominator, five stated as settled, one called provisional or revised.

Among the twenty-three read closely the ledger gives ten cases with no count, five where the codings divided, five whose count is qualified, two stated as settled and one called provisional or revised. The five qualified counts are Ragusa's month apart, Jenner's Inquiry, Pasteur's treatment of Joseph Meister, the Ebola committee record and the MERS record; the two settled are the Dongui Bogam and the Broad Street pump; the provisional one is Life Esidimeni. A settled cell records what a case's own text does with its number, not whether the number holds.

The second hypothesis of this study asks, of cases that give a count, whether the case's own text qualifies the number rather than giving it as settled. The verdict rule calls a result supporting if 60 per cent or more of the cases a hypothesis applies to support it, against if 40 per cent or less, treats the band between as unclear, and also returns unclear when unclear cases outnumber the decided ones. On this hypothesis the ledger gives the following.

The share counts eight of thirteen decided cases with a count. Five of the thirteen state their count as settled and so go against the hypothesis; the other eight support it. On the strict reading, 25 unclear cases outnumber the thirteen decided ones and the registered rule forces a verdict of unclear; on the lenient reading fourteen unclear cases move to not applicable, the thirteen stand unchanged, and the share of 62 per cent sits a little above the line registered before coding, giving a verdict of supports. Both verdicts rest on the same thirteen cases. Thirteen decided cases are too few to say what counts in general do, and the share is a fact about these cases rather than a property of outbreak counting.

Group by group the verdicts divide, and the group table gives three of its four counts because it breaks out no not-applicable figure. For the seventeen medicine-history cases the hypothesis is supported, 6 for, 2 against, 4 unclear, a share of 75 per cent on both readings. For the forty-seven disaster-history cases it is unclear, 1 for, 2 against, 13 unclear on the strict reading and 1 for, 2 against, 4 unclear on the lenient reading, each at 33 per cent. For the twenty-two industrial-history cases it is unclear again, 1 for, 1 against, 8 unclear on the strict reading and 1 for, 1 against, 3 unclear on the lenient reading, each at 50 per cent. Where qualified counts sit in this ledger is what the groups show: in the medicine cases, which carry the most institutional reporting, and not in the disaster cases, where most records carry no count at all.

Two further facts from the ledger bear on what a count can carry. Every case lists what is missing — 264 items across the eighty-nine — while none carries a condition for revisiting its judgment; that field reads 0 of 89. A count in this ledger is ordinary company: it may carry a qualification inside its own text, its record may list what is absent, and no case sets a condition under which the count would be re-examined. The ledger's own count of cases whose record shows a decider reads 0 of 89, so for any count here a later reader learns who counted only where a source outside the record supplies it.

Set beside Ebola and MERS are Graunt's bills of mortality of 1665, Nightingale at Scutari, the army's mortality reprint of 1858 and Life Esidimeni, for one thing only: what each record dates, what it leaves undated, and what its count cell reads. On Graunt the codings divided on the count cell, so the ledger does not give the value; the record of the 1665 edition is an analysis of counts drawn from the weekly bills that parish searchers reported, and no side is chosen here. The Scutari record and the reprint of 1858 read no-count: the 1858 report carries tables and diagrams, but the case's own text gives no count figure, and figures the record does not carry are not imported. Life Esidimeni is the one case whose count cell reads provisional-or-revised: an ombud's announcement of 1 February 2017 giving a death count called provisional, with the ledger's close reading naming the bearer of the cost and marking a revision in the case. Of these six cases four are coded as counts or published reports, the MERS record as a notice or certification and the Scutari record as other; the act cell reads yes in all six.

On the ledger's reading of act before proof, the MERS record reads yes and the Ebola case unsettled, where the codings differed; for Graunt it reads not-said, for Scutari yes, for the reprint of 1858 unsettled, and for Life Esidimeni yes.

What the two twenty-first-century records date is the same in shape and different in content. The Ebola record dates the Emergency Committee meeting on 6–7 August, the tallies reported on 5–6 August and the situation report of 8 August 2014, the case's date in the ledger; it leaves undated the moment at which the tally of 68 new cases and 29 deaths closes, and leaves open whether the same case definition held at the two ends of any later comparison. The MERS record dates the notification of 20 May 2015 and gives an outbreak-end total whose composition is stated — 185 cases in Korea and 1 in China, with 38 deaths beside them; it fixes no closing date for the total beyond its being the end of the outbreak, and states no period for the deaths beyond their pairing with that total. In both records the act was a total written at the hour of a decision, and what was known when it was taken is only what the page before the reader carried: the reach of a definition in one case, the sum of two countries' figures in the other. A later reader can tell from either record which moments the document separates, which definition applied at the hour of writing, and how the total was composed. The reader cannot tell whether a change in any total taken under these rules shows a change in the disease; that question needs the same definition, the same population and the same closing moment at both ends, and neither record supplies all three.

The comparison this ledger permits is narrow. Where a modern institutional record carries a count here, the qualification travels with it: probable and suspected cases inside the definition in the Ebola record, two countries' figures summed in the MERS record, provisional status written into the announcement in Life Esidimeni. Where a case carries no count in its own text, the question of settlement does not arise, and in Graunt's case the codings divided, a fact about the ledger rather than about the bills of mortality. The hypothesis on qualified counts rests on eight of thirteen decided cases, the group verdicts on shares between 33 and 75 per cent, and no case in the ledger carries a condition for revisiting its judgment.

10.5What a count would need to be read in its own time

Both records of this chapter carry, in their own fifth question, Skeptron's generic condition for keeping or withdrawing a comparison: fix the denominator, the unit, the equipment and the end point, and observe a result recalculated in the same unit and scope; if the population, the assets, the exchange or the observation end date changes, the result is not compared as the same measurement result. Skeptron labels that a comparative hypothesis, not a prediction. It is the closest thing either record gives to a rule for what a count would need in order to be read at the hour it was taken. Two limits stand on it. It is written to decide whether two totals, taken at two moments, may be compared as one measurement result; it does not say which moment inside a single total a reader should take as that total's own date. And it appears in the fifth question of every history case in the ledger, which is why the series' own coding counts conditions for revisiting the judgment in 0 of the 89 cases: the condition is present in every record and case-specific in none.

The list of what a count would need therefore has to be built from what the two records do and do not carry. For a count to be readable in its own time, a later reader needs, from the record: the definition of a case, including whether probable and suspected cases are inside it; the population and the period over which the total is taken; the unit of the count; the moment at which the count closes; who counted and by what rule; the difference between when cases occurred, were tested, were reported and were published; and the denominator against which any rate is computed. Set against that list, the two records are unevenly complete.

On dates the records are lopsided. The Ebola record dates three things in one document: the Emergency Committee meeting on 6–7 August, the tallies reported on 5–6 August, and the situation report of 8 August 2014, which is the case's date in the ledger. The MERS record dates the notification of 20 May 2015 and the activation of the response; the total it gives is dated only by the phrase "outbreak-end". Neither record fixes a closing hour or a closing date for its own number, and neither gives a period for its death figure beyond its pairing with the total. What the records do instead of closing the count is to separate its moments:

Each of those sentences instructs a reader; none of them is an hour at which the tally stops.

What the act was, and what was known when it was taken. The Ebola act is a published count, the situation report of 8 August, whose composition is declared inside the number:

The MERS act is a total presented as the outbreak-end count, summing two countries' figures:

In each case what was known when the act was taken, on the record's own evidence, is only what the page declares about itself: the reach of the case definition in the first, the sum of two countries' confirmed figures in the second. Neither record shows what any decider knew beyond that page, and this study supplies nothing.

What a later reader can and cannot tell. From these two records a later reader can tell which moments the documents separate, which definition applied at the hour of writing, and how each total was composed. The reader cannot tell whether a later total taken under a different definition shows a change in the disease, because the records do not supply the same definition, the same population and the same closing moment at both ends of such a comparison. The ledger's cells say the same in its own vocabulary. Both cases read yes on the act and both read qualified-by-definition-scope-or-denominator on the count; both read named on the gap between event and record; both read not-named on the bearer, and both read unknown in the series' waits-and-pays cell. The two part on one cell: the Ebola case's own text names no later finding that would change how its count should be read, while the MERS case names the comparison condition as a boundary of the count.

What the reading notes and the source texts add is narrow. Lipsitch and colleagues' note describes three biases that make a live total's direction ambiguous; it gives no denominator, no closing date and no bearer, so it cannot fill the records' silence. Moon and colleagues' note, Richards's note and the Korean agency's report stand beside the counts as histories of the response and of hospital transmission; none of them states how the tallies were assembled. No source text goes with either case, because both originals are World Health Organization statements that this study may not reproduce; a reader of this study therefore cannot check the statements' wording against the records' summaries, and this study does not paraphrase them.

On the ledger as a whole these two counts are a small part of a shape 10.4 gives in numbers: the codings agreed on the count cell in 78 of the 89 cases, most of which carry no count at all, and among the twenty-three cases read closely five carry a qualified count, two a settled one and one a provisional one. Across the four parts of this study's question — a public act, a named gap between event and record, a named bearer, a named revision — the ledger's numbers find one record among the thirty-six that carries all four, and one among the twenty-three read closely; the section 'Across the families' reports the same test where every further reader agrees as 0 of 21 and where a majority of readers agrees as 2 of 28. These two records carry the act and the gap, and not the bearer or, in the Ebola case, the revision.

Two counts, then, from 2014 and 2015, both qualified inside their own records, both silent on who bore the wait, both dated by their documents rather than by their numbers. The condition every record's fifth question carries fixes the unit and the scope for a comparison between two totals; it fixes no hour for a count, names no bearer, and sets no case-specific rule for reading the count differently. That is what these records allow a later reader to say about the hour the counts were written. What would make any of these readings change is the subject of the next chapter.