Starting in Summer 2020, I wrote some R Notebooks showing how to grab public data yourself to see that the COVID-19 deaths are not likely being hyped by the media. Initially, they focused on New York City and Florida because they had been the hardest hit; however, in Fall 2020, I started tracking South Carolina deaths. Periodically, I go back and update this notebook so the data reasonably up-to-date. Here is my last update covering all of 2020. Obviously, a lot more people have died since 2021 started, but that will be a different story involving the waning original virus and the (now) surge of the delta variant. This notebook only covers 2020. That’s enough to provide very compelling evidence that we did not overcount COVID-19 deaths.

Everything I do, I keep in the open. All statistical analysis, all sources of data, all claims, backed up by primary sources (not media accounts). It can all be verified by anyone who decides to do so. I do not rely on a single source of data (e.g., CDC mortality results of COVID deaths), but cross validate different data across different sources.

In this write up, I provide data-driven evidence for:

Source Data

The Center for Disease Control (CDC) and the National Center for Health Statistics (NCHS) provide their data publicly:

For each of these data sets, I did the following:

  1. Went the the link provided above;
  2. Read the dataset description, including what each of the fields are;
  3. Clicked Export then CSV and saved the file on my local hard drive.

You will need to install the statistical packaged R to do this yourself, and you’ll need to install the dplyr and the ggplot2 packages in R. Here’s an example showing the (filtered) first few lines of the weekly death counts for this year:

suppressPackageStartupMessages(library(dplyr))

# Build a Date Converter from a string formatted "mm/dd/YYYY" to a proper date datatype in R
setAs("character", "rpwDateConvert", function(from) as.Date(from, format="%m/%d/%Y") )

# Read the data provided by the CDC/NHCS
WeeklyDeathsThisYear <- read.csv('Weekly_Provisional_Counts_of_Deaths_by_State_and_Select_Causes__2020-2021.csv', 
                                 colClasses=c("Week.Ending.Date"="rpwDateConvert"),  # Convert to a date format
                                 header=T) # Read the first line and use it as a header

# Include only 2020 deaths, and grab the max MMR week index value
maxWeekIDX.2020 = max(filter(WeeklyDeathsThisYear, MMWR.Year==2020)$MMWR.Week)

# Filter so we have just the New York rows and the year, week, and overal death counts
SCDeathsThisYear <- subset(filter(WeeklyDeathsThisYear,
                                  Jurisdiction.of.Occurrence == 'South Carolina',  # Just SC
                                  MMWR.Year == 2020),                       # Just 2020
                              # vvv Just the following fields vvv
                           select = c(Jurisdiction.of.Occurrence,MMWR.Year,MMWR.Week,Week.Ending.Date,All.Cause))

# Correct the fact that the Week.Ending.Date isn't always correctly populated in the CDC data
SCDeathsThisYear <- mutate(SCDeathsThisYear,
                          Week=as.Date("2020-01-01") + 7*(MMWR.Week-1))

# Show the top few rows of this subset
head(SCDeathsThisYear)

I chose to plot the individual points, as well as a line plot through those points. Note, this code below assumes you have run the code above first (so that you have the SCDeathsThisYear data frame).

library(ggplot2)

myPlot <- ggplot(SCDeathsThisYear, aes(x=Week, y=All.Cause)) +
             geom_point(size=2, color="firebrick") +
             geom_line(size=1, color="firebrick") +
             xlab("Week") +
             ylab("Total Reported Deaths in SC (each Week)") +
             theme(text=element_text(family="Times", size=16))
print(myPlot)

Comparing Overall Deaths

Let’s compare overall death counts (from all causes) across several years. I’ll need to merge some datasets:

library(dplyr)

# Build a Date Converter from a string formatted "dd/mm/YYYY" to a proper date datatype in R
setAs("character", "rpwDateConvert", function(from) as.Date(from, format="%d/%m/%Y") )

# Read the data provided by the CDC/NHCS for 2019-2020, then read the 2014-2018 data
WeeklyDeaths.2020.2021 <- read.csv('Weekly_Provisional_Counts_of_Deaths_by_State_and_Select_Causes__2020-2021.csv', 
                                   colClasses=c("Week.Ending.Date"="rpwDateConvert"),  # Convert to a date format
                                   header=T) # Read the first line and use it as a header
WeeklyDeaths.2014.2019 <- read.csv('Weekly_Counts_of_Deaths_by_State_and_Select_Causes__2014-2019.csv', 
                                   colClasses=c("Week.Ending.Date"="rpwDateConvert"),  # Convert to a date format
                                   header=T) # Read the first line and use it as a header

# Filter so we have just the New York rows and the year, week, and overal death counts
SCDeaths.A <- subset(filter(WeeklyDeaths.2020.2021,
                            Jurisdiction.of.Occurrence == 'South Carolina', # Just SC
                            MMWR.Year < 2021),  # Just 2020
                     select = c(Jurisdiction.of.Occurrence,MMWR.Year,MMWR.Week,Week.Ending.Date,All.Cause))

# Put this is a mergable form because the column names differ a bit between the datasets
SCDeaths.AA <- data.frame(Year= SCDeaths.A$MMWR.Year,
                          Week = SCDeaths.A$Week.Ending.Date,
                          WeekIDX = SCDeaths.A$MMWR.Week,
                          TotalDeaths = SCDeaths.A$All.Cause)

# Filter so we have just the New York rows and the year, week, and overal death counts
SCDeaths.B <- subset(filter(WeeklyDeaths.2014.2019,
                            Jurisdiction.of.Occurrence == 'South Carolina'),  # Just SC
                     select = c(Jurisdiction.of.Occurrence,MMWR.Year,MMWR.Week,Week.Ending.Date,All..Cause))

# Put this is a mergable form because the column names differ a bit between the datasets
SCDeaths.BB <- data.frame(Year= SCDeaths.B$MMWR.Year,
                          Week = SCDeaths.B$Week.Ending.Date,
                          WeekIDX = SCDeaths.B$MMWR.Week,
                          TotalDeaths = SCDeaths.B$All..Cause)


# Combine the two datasets so we have all years from 2014 through 2020
SCDeaths <- mutate(rbind(SCDeaths.AA, SCDeaths.BB),
                   ThisYear=(Year==2020))   # We'll use this field later to highlight 2020

That looks like a lot, but the whole point is to get that last dataset, SCDeaths, which contains the total deaths (from all causes) for various weeks across the year for all years from 2014 to 2020. As before, you have to run the code above before the next chunk of code will work.

library(ggplot2)

myPlot <- ggplot(SCDeaths, aes(x=WeekIDX, y=TotalDeaths, group=Year, color=ThisYear)) +
             geom_line(size=1) +
             scale_color_manual(values=c("darkgray", "firebrick"), labels=c("Other Years", "2020"), name="") +
             xlab("Week") +
             ylab("Total Reported Deaths in SC (each Week)") +
              theme(text=element_text(family="Times", size=16))
print(myPlot)

Overall death count is higher than normal since the COVID outbreak. Of course, this is not as bad as New York City or Florida.

Okay, not so bad, then! We’re no New York City or Florida, thankfully!

Still, We’re Not Likely Overcounting COVID Deaths

Let’s first look at adjusted cumulative death counts. That is, let’s total all deaths that occurred up until this point in the year, then subtract off the average count for that year. That is, how do this year’s deaths differ from the expected number of deaths?

# First get only the week indexes for any year that can be compared to 2020
#  Subtract a couple weeks because of the lagging indicator and to be consistent with the COVID dates
SCDeaths.abridged <- filter(SCDeaths, 
                            Year < 2020)           # All years *other than* 2020

# Now let's find the average death counts for all years *other* than 2020 across each week
AggDeathCounts <- summarize(group_by(SCDeaths.abridged, WeekIDX), AvgDeathCount.pre2020 = mean(TotalDeaths))

# Now let's accumulate them:
totalPre2020Deaths <- sum(AggDeathCounts$AvgDeathCount.pre2020)

# Now we'll accumulate deaths for 2020:
SCDeaths.2020 <- filter(SCDeaths, Year==2020)
total2020Deaths <- sum(SCDeaths.2020$TotalDeaths)

# Here's the difference:
cat('How many *more* deaths than typical so far (on average) in all of 2020?  ', total2020Deaths - totalPre2020Deaths, '\n')
## How many *more* deaths than typical so far (on average) in all of 2020?   12713

We can use those same datasets to compute the number of people the CDC said died of COVID (including multiple cause deaths) during all of 2020. It’s a lot less.

suppressPackageStartupMessages(library(dplyr))

# Build a Date Converter from a string formatted "mm/dd/YYYY" to a proper date datatype in R
setAs("character", "rpwDateConvert", function(from) as.Date(from, format="%m/%d/%Y") )

# Read the data provided by the CDC/NHCS
WeeklyDeathsThisYear <- read.csv('Weekly_Provisional_Counts_of_Deaths_by_State_and_Select_Causes__2020-2021.csv', 
                                 colClasses=c("Week.Ending.Date"="rpwDateConvert"),  # Convert to a date format
                                 header=T) # Read the first line and use it as a header

# Grab the counts for SC
SCCovidDeaths <- filter(WeeklyDeathsThisYear, 
                        Jurisdiction.of.Occurrence == 'South Carolina', 
                        MMWR.Year==2020)$COVID.19..U071..Multiple.Cause.of.Death.

# What's the biggest number they have!
totalSCDeaths = sum(na.omit(SCCovidDeaths))
cat('How many deaths has the CDC attributed to COVID through end of 2020?', totalSCDeaths, '\n')
## How many deaths has the CDC attributed to COVID through end of 2020? 5798

Note that I am counting what the CDC codes as U07.1 deaths – deaths where COVID-19 is the direct cause, not people being hit by a bus and happen to test positive. Go read the CDC source material – they are clear about the coding and attribution. FYI: As of July 2021, the CDC has atttributed nearly 10K deaths in SC (overall) to COVID.

Regardless, the COVID-19 attributed death count is significantly lower than the overall number of deaths above the expected count we have experienced this year. Based on the data there is absolutely no reason to believe the CDC overcounted COVID-19 deaths in South Carolina.

The only way you can buy that is if you believe doctors are systemically falsifying death records (not the cause of death, but the fact itself that someone died) across the entire state. That is, you’d have to believe doctors are falsifying death certificates (a felony). If that’s your view, you’ll need to provide evidence. I address the fact more specific claim that doctors are miscoding deaths as an unlikely explanation in a separate section, below.

Here’s a Nature Article that provides even more evidence that there are many more people dying as a result of Coronavirus (for multiple reasons) than we are counting.

It’s Probably Not Suicides

One counter-argument I hear is that the suicide rate is up because of the pandemic, the “hype”, increased unemployment, and the social distancing requirements. This is a reasonable hypothesis. While we have no idea how much the rate of suicide has increased (and wont for probably a year), it’s clear that it there is some evidence of the ideation of suicde and substance abuse as having increased (CDC Report). Moreover, there are almost certainly more suicides this year; it’s been a horrible year for all of us (some much more than others).

But the numbers don’t fly. First of all, suicide rates across the US, and in SC in particular have been on a steady and disturbing rise for a number of years now, so it’s not clear how much of whatever increases will happen this year will be due to the pandemic. But leaving that aside, in the entire year of 2019, there were 819 suicides in SC (see here). While in 2020 there are almost 7K deaths above the expected. That would be over a 850% increase in suicides. If the CDC is overcounting COVID deaths, it’s even more extreme. That’s an absurd claim, and you’d have to provide evidence of that. In fact, early estimates are that the number of suicides in SC in 2020 remained about the same as 2019 at 811 (see here).

Keep in mind that while some (other) causes of death may have gone up during the pandemic (e.g., drug overdonse, domestic violence, etc.), others likely went down (e.g., traffic fatalities, flu) because of social distancing methods. So seeing a massive bump in unexpected deaths is an extremely compelling piece of corroborating evidence that people are dying of the Corona virus, and quite likely in higher numbers than the CDC counted.

Underlying Conditions & Multiple Causes

There was a report released not too long ago by the CDC that suggested that up to 94% of deaths coded as Coronavirus deaths included cases with some other underlying condition. I read on social media a number of people interpreting this to mean that the death count is not real because these people are really dying from multiple causes.

But first of all, that’s not what the report said. Go read the report for yourself (not your favorite media’s article about the report). It’s point is that there are many underlying conditions that make us vulnerable to Coronavirus, not just age, so we should all be more careful. In other words, the report says almost the polar opposite of what was concluded by these media accounts.

Putting that aside, most deaths involve multiple underlying conditions, if the argument is that you cannot count a death as a “COVID death” if there are any comorbidities, then no disease is particularly deadly. The comorbidity of breast cancer can be as high as 75%, as just the first random example that I looked up. Using comorbidity to eschew COVID deaths is just grasping at straws.

People are Dying of Pneumonia Because of COVID

I also heard this argument: “More died in 2020 from pneumonia than COVID!” At first, this argument baffled me. True, if you look at the very CDC datasets I quote here you’ll see higher numbers in the P&I mortality data (pneumonia & influenza), but that’s because pneumonia is a frequent complication of COVID.

People also seem to be conflating this with the “underlying conditions” argument I discredited, above. That is, they have the causal direction exactly backwards: They have decided that people who would have died of pneumonia anyway are were marked as a COVID death because they happen to have COVID. That’s just not how that works. They caught pneumonia because they had COVID. In 2018 and 2019, a little over 40K people died of pneumonia in the US, in 2020 it was nearly 2/3ds of a million.

I was going to pull data and show the pattern over time, which is very, very clear. But I don’t have to. The NIH/CDC have a great dashboard showing this:

https://gis.cdc.gov/grasp/fluview/mortality.html

This is a particularly dishonest argument in my view. It’s like saying that car accidents aren’t as bad as we think they are because more people die per year from blunt-force impacts. I suppose I should be happy that people are looking at the CDC data, but it frustrates me that they stop there and are happy to draw erroneous conclusions. The fact that so many people are dying of pneumonia is evidence of the mortality of COVID, not evidence that COVID is not deadly.

It’s Unlikely That Medical Professionals Misprepresented Numbers for Money

Another narrative was that because medicare reimburses hospitals differently for COVID-19 deaths, doctors and other medical professionals have an incentive to misrepresent causes of death. As I’ve already said, the CDC codings allow for multiple causes, and U07.1 codings are not people who get hit by a bus and happen to test positive. So if the numbers were misrepresented, it must have happened when the cause of death is coded, not because of the coding system. This idea that we systematically screwed up the numbers because we weren’t differentiating between dying “with” Coronavirus vs. “from” Coronavirus is a total myth. You can go read the metadata I linked to above; the CDC is clear on the coding, and it’s a direct-cause coding.

Though there were anecdotal accounts of medical professionals mis-coding deaths for Coronavirus, there’s no study demonstrating this happened in any wide-spread and systematic way. Moreover, it doesn’t make logical sense. Hospitals receive money from the federal government and insurance companies for many things, and we’ve never decided that they whole-sale crooks in the past over other diseases and causes of death. Moreover, doctors aren’t seeing any increased money in their wallets, so what would possibly motivate them to risk legal liability by falsifying a cause of death? Even if a handful of medical professionals made such a choice, what reason is their to believe this is a wide-spread, systemic issue? Anecdotes are meaningless; data is telling.

But ignoring all that, let’s look again at numbers. The case-fatality rate of the US for COVID-19 was roughly 2.7% at its peak. That is, if you caught Coronavirus, you had roughly a 2.7% chance of dying when it was at its mostly deadly point. This is the number President Trump touts as a good number when he suggested we were the “best in the world”. We were not. To be sure, there were many countries with much worse case-fatality ratios, though there were (and still are) also many with much better. Take a look at this Johns Hopkins Study for the current numbers. At the peak, the world a case-fatality ratio of about 2.76% – so we were basically on par. For context: the majority of countries that report were between 2% and 4%.

If doctors in the US are inflating the deaths, then our real case-fatality ratio is much lower than reported – and much lower than other countries! Consider for example, that medical professionals are doubling the death rate … that there have been only about 300K deaths so far due to Coronavirus (which would still make it three times as deadly as flu, by the way), then we’d have had a 1.35% case-fatality rate. If you believe the doctor-fabrication speculation, this number becomes even more extreme.

But why? What have we done in the US that would make us twice as effective as other countries? There’s just no reason to believe that. Again, I need evidence to accept that claim. It’s simply highly improbable that we are systemically defrauding the count and still getting the case-fatality ratio we are getting – the case-fatality rate should be much higher were that the case.

And there’s certainly no reason to believe this conspiracy of medical professionals to defraud the public about Coronavirus is a world-wide conspiracy. The healthcare systems in other countries work very differently that the US, so the “motivation” argument completely breaks down.

It’s Not Just Old People

I heard a lot of people eschewing the deadliness of the virus by observing that the virus really only affects older people or people who were likely to have issues anyway because of underlying medical conditions. But that argument doesn’t make sense to me. First, we’re not talking about people who “might have died anyway”. Over 7K more people died in SC in 2020 than should have based on the last six years of data. These are exactly people who should not have died (by definition).

Let’s put aside the fact that older people are people, and I don’t understand why somehow they should not be counted. On its face, this argument that it’s mostly older people dying doesn’t make sense. First, most of us interact with people of all ages. Though we may not die, we can carry and spread the disease, which makes those around us less safe. Moreover, passing the disease to others means they will pass the disease on … and even if you carefully limit your radius of people to those under retirement age, you have no idea how the others with whom you interact limit their interactions. If the observation that older people are the most affected doesn’t affect our behaviors at all, then the observation is pointless. If the observation does affect our behavior, then we are keeping the vector of the pathogen active and costing unnecessary lives – meaning the fact the virus is very dangerous. So I’m missing the basic logic of this argument entirely, either way.

Further, the data is not quite as black-and-white as people suggest. True, most deaths are of elderly people, but 31% of COVID-19 coded deaths in SC in 2020 were for people under 65. Since I’m 50, I have to say that 65 doesn’t seem that old.

We’ll need yet another data set:

suppressPackageStartupMessages(library(dplyr))

# Build a Date Converter from a string formatted "mm/dd/YYYY" to a proper date datatype in R
setAs("character", "rpwDateConvert", function(from) as.Date(from, format="%m/%d/%Y") )

# Read the data provided by the CDC/NHCS
rawDeathsByAge <- read.csv('Provisional_COVID-19_Deaths_by_Sex_and_Age.csv',header=T)  # Convert to a date format

SCDeathsByAge <- filter(rawDeathsByAge,
                        State == "South Carolina",     # Just SC
                        Year == 2020,                  # Just 2020
                        Group == "By Year")            # Just annual aggreations

# Get the total deaths in SC of C19 from this table
TotalC19Deaths <- sum(na.omit(filter(SCDeathsByAge, Sex == "All Sexes", Age.Group != "All Ages")$COVID.19.Deaths))

# Use the total deaths to get the Percentage of deaths by age group
Over65Deaths <- sum(as.numeric(na.omit(filter(SCDeathsByAge, Sex == "All Sexes", Age.Group == "65-74 years")$COVID.19.Deaths))) + 
                sum(as.numeric(na.omit(filter(SCDeathsByAge, Sex == "All Sexes",  Age.Group == "75-84 years")$COVID.19.Deaths))) +
                sum(as.numeric(na.omit(filter(SCDeathsByAge, Sex == "All Sexes",  Age.Group == "85 years and over")$COVID.19.Deaths))) 

cat(paste("Percentage under 65 that have died of COVID-19 in SC: ", 
          round(100*(1.0 - Over65Deaths / as.numeric(gsub(",", "", TotalC19Deaths))),2)))
## Percentage under 65 that have died of COVID-19 in SC:  31.22

I am not saying those of us under 65 should have been quaking in our boots, but even if I weren’t concerned for other people who are at more risk (which I am), it’s still worth pointing out that even for those of us under retirement age, COVID was a lot more deadly than the chance of homicide in SC in 2020. SLED reported 694 homicides in 2020, the CDC reported over 2K COVID-19 deaths for those under 65. That under-65 COVID death number is more than double the (total) number of suicides in SC in 2020, and much larger than the adjusted number of deaths for diabetes. We take all those things seriously. Why would we not take COVID-19 seriously?

Moreover, across the country this is a lot of people … almost 75K. Almost double the number that died by car accident last year and almost double suicide rate across the US. In 2019, about 38K died of the flu overall. That means more than twice the number of people under 65 died in 2020 by COVID-19 than all flu deaths in the entire year of 2019 (including the elderly) . COVID-19 is a lot more deadly than the flu, even for those of us who are middle-aged.

suppressPackageStartupMessages(library(dplyr))

# Build a Date Converter from a string formatted "mm/dd/YYYY" to a proper date datatype in R
setAs("character", "rpwDateConvert", function(from) as.Date(from, format="%m/%d/%Y") )

# Read the data provided by the CDC/NHCS
rawDeathsByAge <- read.csv('Provisional_COVID-19_Deaths_by_Sex_and_Age.csv',header=T)  # Convert to a date format

USDeathsByAge <- filter(rawDeathsByAge,
                        State == "United States",      # All of the US
                        Year == 2020,                  # Just 2020
                        Group == "By Year")            # Just annual aggreations

# Get the total deaths in SC of C19 from this table
TotalC19Deaths <- filter(USDeathsByAge, Sex == "All Sexes", Age.Group == "All Ages")$COVID.19.Deaths

# Use the total deaths to get the Percentage of deaths by age group
Over65Deaths <- sum(as.numeric(gsub(",","",filter(USDeathsByAge, Age.Group == "65-74 years", Sex != "All Sexes")$COVID.19.Deaths))) +
                sum(as.numeric(gsub(",","",filter(USDeathsByAge, Age.Group == "75-84 years", Sex != "All Sexes")$COVID.19.Deaths))) +
                sum(as.numeric(gsub(",","",filter(USDeathsByAge, Age.Group == "85 years and over", Sex != "All Sexes")$COVID.19.Deaths)))

cat(paste("Total COVID-19 Coded Deaths in the US of people *under* 65: ", 
          as.numeric(gsub(",","",TotalC19Deaths)) - as.numeric(gsub(",","",Over65Deaths))))
## Total COVID-19 Coded Deaths in the US of people *under* 65:  74019

Occam’s Razor

Occam’s Razer: “All other things being equal, the simplest explanation is probably correct.”

All other things are not equal. We have a great deal of compelling statistical and medical evidence that the CDC numbers are a lower bound on the number of deaths from Coranavirus, and we have very little evidence beyond anecdotes and speculation that they are systemically and significantly overcounting COVID deaths. Even if you were somehow willing to completely ignore this data and decide arbitrarily that all things are equal, the alternative explanations for the opposing view simply make no sense. They rely on massive, wide-spread (world-wide) conspiracies of many levels of professionals (e.g., research epidemiologists, medical doctors, coroners, government health officials, etc.) and extremely implausible scenarios (e.g, a spike in suicides of many orders of magnitude). Add to all of this that we all want the virus to be less deadly, and you have a recipe for people to buy into unnecessarily complicated speculations in order to arrange things to say what they want them to say.

No. The simplest explanation is that the medical and research professionals who study this topic for a living and are significantly more educated about it than any of us, are correct. Moreover, the data supports that.

My analyses for NYC and Florida come up with the same, exact results.

Here’s a Scientific American article providing even more evidence that we are not overcounting COVID-19 deaths: (SciAm Oct2020).

Not “Fear Mongering”, Not about Lockdown, Not about Politics

This is not political; it’s basic science and data analysis. I hear a lot of people claiming we’re “fear mongering” for political purposes. I’ll be honest: I don’t understand this position at all. I want to know the truth, whatever that truth is. The data and the experts tell me a pretty compelling story, and I am convinced by that empirical evidence that COVID-19 deaths are not being hyped. I’m not even using the media or politicians to arrive at these conclusions. What difference does politics make? I’m either right or I’m not, irrespective.

Knowing the truth is different from feeling an emotion, and trying to explain to others that truth is not the same as spreading an emotion. Feel how you want to feel: If you’re scared, so be it. If you’re not, so be it. But if you tell me we’re overcounting COVID-19 deaths, I’m going to argue with you because the data just doesn’t support that claim. Combat my analysis with contradictory evidence, but if you tell me I’m “fear mongering”, be prepared for me to ignore your complaint. What’s true is true, it doesn’t matter how anyone feels about it.

When I engage in arguments on these points, a lot of people end up getting upset about the lockdown. What to do about the pandemic is a whole different discussion. First and foremost, before we have any hope of coming to any kind of agreement of how to proceed, we need to agree on what is happening. Whether the lockdowns were good ideas or not doesn’t change whether or not people are or are not dying of COVID-19. So, from my perspective, this is a distraction from the primary point: What, exactly, is happening?

Here’s what: It’s not hype. Half a million people in the US have died of COVID-19 in 2020 (at least). It was the third biggest killer in the US. It was not sensationalized. If anything, the United States did not take it seriously enough. And they still aren’t (delta variant is on the rise again, faster even than COVID-19 to start with).