The Rapid Growth in Deaths from Covid-19: The Role of Politics

Deaths from Covid-19 have been growing at an extremely rapid rate.  The chart above shows what those rates have been in the month of March, averaged over seven day periods to smooth out day-to-day fluctuations.  The figures are for the daily rate of growth over the seven day period ending on the date indicated.  The curves start in the first period when there were at least 10 cases, which was on March 3 for the US as a whole.  Hence the first growth rate shown is for the one week period of March 3 to 10.  As I will discuss below, the chart has not only the growth rates for the US as a whole but also for the set of states that Trump won in 2016 and for the set that Clinton won.  They show an obvious pattern.

The data come from the set assembled by The New York Times, based on a compilation of state and local reports.  The Times updates these figures daily, and has made them available through the GitHub site.  And it provides a summary report on these figures, with a map, at least daily.

I emphasize that the figures are of daily growth rates, even though they are calculated over one week periods.  And they are huge.  For the US as a whole, that rate was just over 28% a day for the seven day period ending March 30.  It is difficult to get one’s head around such a rapid rate of growth, but a few figures can be illustrative.  In the New York Times database, 3,066 Americans had died of Covid-19 as of March 30.  If the 28% rate of growth were maintained, then the entire population of the US (330 million) would be dead by May 16.  For many reasons, that will not happen.  The entire population would have been infected well before (if there was nothing to limit the spread) and it is fatal for perhaps 1% of those infected.  And the 99% infected who do not die develop an immunity, where once they recover they cannot spread the virus to others.  For this reason as well, 100% of those not previously exposed will not catch the virus.  Rather, it will be some lower share, as the spread becomes less and less likely as an increasing share of the population develops an immunity.  This is also the reason why mass vaccination programs are effective in stopping the spread of a virus (including to those not able to receive a vaccination, such as very young children or those with compromised immune systems).

So that 28% daily rate of growth has to come down, preferably by policy rather than by running out of people to infect.  And there has been a small reduction in the last two days (the seven day periods ending March 29 and March 30), with the rate falling modestly to 28% from a 30% rate that had ruled since the seven day period ending March 22.  But it has much farther to go to get to zero.

The recent modest dip might be an initial sign that the social distancing measures that began to be put in place around parts of the nation by March 16 are having a positive effect (and where many individuals, including myself, started social distancing some time before).  It is believed that it takes about 4 to 7 days after being infected before one shows any symptoms, and then, in those cases where the symptoms are severe and require hospitalization (about 20% of the total), another several days to two weeks before it becomes critical for those where it will prove fatal.  Hence one might be starting to see the impacts of the policies about now.

But the social distancing measures implemented varied widely across the US.  They were strict and early in some locales, and advisory only and relatively late in other locales.  Sadly, Trump injected a political element into this.  Trump belittled the seriousness of Covid-19 until well into March, even calling Covid-19 a “hoax” conjured up by the Democrats while insisting the virus soon would go away.  And even since mid-March Trump has been inconsistent, saying on some days that it needs to be taken seriously and on others that it was not a big deal.  Fox News and radio hosts of the extreme right such as Rush Limbaugh also belittled the seriousness of the virus.

It is therefore understandable that Trump supporters and those who follow such outlets for what they consider the news, have not shown as much of a willingness to implement the social distancing measures that are at this point the only way to reduce the spread of the virus.  And it shows in the death figures.  The red curve in the chart at the top of this post shows the daily growth rates of fatalities from this virus in those states that voted for Trump in the 2016 election.  While the spread of the virus in these states, many of which are relatively rural, started later than in the states that voted for Clinton, their fatalities from the virus have since grown at a substantially faster pace.

The pace of growth in the states that voted for Clinton has also been heavily influenced by the rapid spread of the virus in New York.  As of March 30, more than half (57%) of the fatalities in the Clinton states was due to the fatalities in New York alone.  And New York is a special case.  With its dense population in New York City, where a high proportion use a crowded subway system or buses to commute to work, with the work then often in tall office buildings requiring long rides in what are often crowded elevators, it should not be surprising that a virus that goes person to person could spread rapidly.

Excluding New York, the rate of increase in the other states that voted for Clinton (the curve in green in the chart above) is more modest.  The rates are also then even more substantially lower than those in the Trump-voting states.

But any of these growth rates are still incredibly high, and must be brought down to zero quickly.  That will require clear, sustained, and scientifically sound policy, from the top.  But Trump has not been providing this.

A Very Faint First Sign to be Hopeful on Covid-19: Except Not Yet for the US

Source:  New York Times, “Coronavirus Deaths by U.S. State and Country Over Time:  Daily Tracking”, downloaded March 25, 2020, with deaths reported as of 8:20am on March 25.

As in any epidemic where disease spreads person to person, the number of deaths from the Covid-19 coronavirus has exploded at exponential rates.  An important question is how long it will continue to grow like this.  It cannot continue forever, as one cannot infect more than 100% of the population, and most such diseases are turned around well before that.  But where it will turn around, with a leveling off in the cumulative number of cases, depends not only on the characteristics of the disease (how easily it spreads) but also on policy.  Since there is not yet a vaccine nor a treatment that will always work, the spread of the disease and the number of deaths from it depends on the effectiveness of social distancing measures, so there is less person to person contact and therefore less spread of the virus.  So far, this has been the only effective means to reduce the number of those catching the virus.

And there now appears to be some early evidence that such social distancing measures have helped.  The chart at the top of this post is an excellent graphic prepared by the New York Times, updated daily, which presents on a semi-log scale the cumulative number of deaths from Covid-19 by country, plotted against the number of days since that country’s 25th death.  The Financial Times also presents (and updates daily) a very similar chart, although it presents the results for each country in terms of the number of days since the 10th death.  Both of these news sources are making available this material, and all of their coronovirus coverage, free to anyone, including non-subscribers.  I very much encourage everyone to examine these postings, as there is a good deal of interesting further material (including charts on the number of deaths by national sub-regions, as well as the cumulative number of confirmed cases).  The New York Times charts are also interactive, where they present (for any individual country or region chosen) the rate of growth over the most recent 7 days, i.e. how fast it is growing now.

By presenting the numbers on a semi-log scale (where the vertical axis is logarithmic, while the horizontal axis is in regular linear terms), a path that is a straight line will indicate a constant rate of growth, and the slope of that line will indicate what that rate of growth is (where the steeper the line the higher the rate of growth).  [If you are not familiar with this, review your high school algebra textbook or read through a web post such as this one.]  The faint, gray, straight lines on the chart then show what the total number of deaths would be (for some number of days since the 25th death) if the number of deaths doubled each day, or doubled every 2 days, or doubled every 3 days, or every week, or every month.

There are several interesting findings one can draw from this:

a)  What I found most interesting, and the reason I titled this post as a “First Sign to be Hopeful”, is that for most of the countries (with the US a notable exception), the paths start out quite steep, with doubling times of between every day and every 2 days, but that they then begin to bend over to the right.  That is, they shift over time to a flatter slope, meaning a slowdown in the rate of growth in deaths.  Mainland China, which was hit first, is now (as I write this) at an almost completely flat slope, which means close to a zero rate of growth.  The curve for Italy has also bent over so that it now hits the gray line for doubling every 3 days.  But that does not mean deaths are doubling every 3 days in Italy right now.  Rather, cases were growing at a faster rate in Italy earlier (doubling at about every 2 days at first), and have decelerated to the point where the cumulative number of cases in Italy are now where they would be had they doubled every 3 days throughout.  But the slope now is a good deal less than what it was in the early days.  The New York Times interactive chart indicates that at the pace of the last 7 days, the number of deaths in Italy is now growing at a rate of doubling every 5 days.  And one sees that flattening out in a number of other cases as well, including hard-hit Iran and Spain.

b)  Japan and to a lesser degree South Korea are exceptions in that their recent rates of growth have not fallen.  But both are also exceptions in that their rates of growth, while steady (the path lines are close to straight), have also been a good deal less than that of other countries.  Their doubling times over the last 7 days (as I write this) are both low at 11 days for South Korea and 12 days for Japan.

c)  The US is also a notable exception.  The pace of growth was relatively low for the US for the first 10 days (from when the 25th death was recorded in the US).  Unlike any other country, the pace then accelerated in the US to a doubling time of every 3 days.  Or more precisely, the number of deaths in the US grew from 60 as of day 10 (from when the 25th death was recorded) to 728 deaths on day 19 (March 24).  That is a 32.0% rate of growth per day, or an increase of 2.3 times every 3 days.

Why was the pace of growth relatively modest in the US at first, and then picked up?  That is not clear from these aggregate figures, but might be because the US is a large country, where there have been several centers of outbreak and those centers are relatively distant from each other.  The earliest center was the State of Washington, and the second (and to a more limited degree at first) in California.  Then New York was hit, followed now by major centers in Michigan, Illinois, Florida, Louisiana, and Georgia.  Adding up these varied impacts by locale across the country as a whole (and where doubling times may also vary by locale, especially for New York) may explain the US curve that starts relatively slow, but then accelerates.

Furthermore, with the Trump administration unwilling or unable to provide direction and management at the national level, each state and locality has been left to enact measures on their own and at their own pace.  These have been primarily social distancing measures, but with major differences in how strict they have been structured.  And most of the measures have been enacted reactively, to local cases being confirmed, rather than preemptively.

What is perhaps most disturbing of all for those of us in the US is that there is no indication as yet of the aggregate US curve beginning to bend over to the right.  It has been close to a straight upward line for the last 9 days, growing at a rate of 32% a day.  The only encouraging sign is that the curve for Washington State alone (shown at the New York Times posting) does bend over to the right, similar to what is seen in other countries.  Washington was hit first with the virus, with the initial deaths there, and early on had a doubling time of every 3 days.  But it was then the first state to put in place relatively strict social distancing measures, and the pace of doubling has now dropped to every 9 days (based on deaths over the most recent week).

Overall, the US was late to enacting social distancing measures, with most only put in place over the last week to week and a half.  Their impact on the number of deaths from the virus will then only be seen two or three weeks later, due to the lag from when one catches the virus to when they start to show symptoms (about one week), to when their cases become serious and lead, for some, to death (a further week or two).

Watching whether the US curve starts to bend to the right soon, in the next week or two, will certainly be of interest.

The Ineffectiveness of Travel Bans for Addressing the COVID-19 Pandemic

A)  Introduction

The US is sinking into what looks likely to be its biggest public health crisis in over a century (i.e. since the Spanish Flu pandemic of 1918/19).  But President Trump continues to insist that he is not to be blamed for its mismanagement.  Rather, he insists that he should be commended for instituting the travel ban on China in early February, that “everyone” was opposed to him doing it but he decided to impose anyway, and that it turned out to be a “great success”.

None of this is true.

What was in fact done?  On January 31, the Trump administration announced that he would impose a ban on travelers from China entering the US, with this going into effect the evening of February 2.  It would not apply to returning US citizens. But there were other countries doing the same at that point, or even earlier (not many, but some).  Australia, for example, set a ban on travelers from China which went into effect on February 1, and New Zealand set a ban effective February 2.  Furthermore, numerous airlines were already suspending their flights from China.  American Airlines had implemented a suspension on all its flights to the US from China effective on January 31.  Delta and United Airlines had already announced that they would also be suspending their flights, and Delta did so on February 2 and United on February 5.  Air Canada had already suspended its flights on January 30, and numerous European airlines suspended theirs starting January 29 (Lufthansa, Swiss International, Austrian, British Airways), January 30 (KLM, Air France), and January 31 (SAS, Iberia).

And it is not correct for Trump to claim that “everyone” was opposed to such a travel ban.  I read the news closely, and I cannot recall any politician, nor any widely expressed public sentiment, arguing against the ban (although I acknowledge that there may well have been some – just not enough to be significant).  Infectious disease experts did say that such a ban would not do much good at that point, as the disease was certainly already in the US and would spread.  Keep in mind that any such disease starts with only one case, of a newly mutated virus that some animal carries (scientists believe it originated in bats, and then passed to some other animal species before jumping to some person).  It then expands person to person from that one case.  A travel ban, by itself, will not stop a spread if there are cases already here.

What a travel ban can do is buy some time.  It can postpone a major spread of the disease by a few weeks.  That can be of value if the ban is implemented very early and if those weeks are then spent to address aggressively the spread of the disease.  This includes rapid testing of all those individuals that may have been exposed to the virus, the isolation of all the cases thus found, and the quarantining of all those who may have been exposed but have not shown symptoms at that point.

But none of this was done in the US.  And as the experts noted, such travel bans will be harmful if they lull policymakers into a false sense of security, with an excuse then to delay taking urgent measures in the false belief that the country is now protected.  It is clear that Trump himself believed this, or at least acted (or rather did not act) consistent with such a belief.

If such a travel ban might buy time, how much time might that be?  This blog post will present some calculations of scenarios of what to expect.  I should stress that I am not an epidemiologist, and the scenarios discussed here are in no way a forecast of what specifically might have happened.  Epidemiologists are looking at that now, using far more sophisticated models (and with far greater knowledge than I have), but are still in an early stage as many of the characteristics of the disease are not yet known with any degree of certainty.

But what matters most is the basic mathematics of pandemics or epidemics (I will use the terms interchangeably here – a pandemic is simply an epidemic of greater range or coverage).  An infectious disease will expand at an exponential rate early on and is subject to a ceiling on those it can infect (no more than 100% of the population, and normally less).  And it is that basic mathematics of the process which shows why travel bans will be futile, and at best will simply delay by only a very short time the spread of a virus such as that which causes COVID-19.

The basic result is summarized in the chart at the top of this post and will be discussed in the next section below.  With plausible parameters, a complete and total travel ban applied to all travelers (including US citizens) might have delayed the spread of the disease by perhaps 2 1/2 weeks.  That is not much.

Far more effective would be policies to reduce the pace at which the disease spreads.  Such policies include “social distancing”, where activities involving crowds are canceled or avoided, and one encourages everyone to wash their hands frequently, stay away from others to the extent they can, and so on.  The chart shows (in the curve in orange) what that might achieve for a plausible parameter.  Its impact is far greater than that of a travel ban.

Slowing down the pace at which the virus spreads is also supremely important, as otherwise our health system could easily become swamped with an overwhelming number of cases requiring care all at one time.  As will be discussed and illustrated in section C below, a travel ban does not help with this at all.  But social distancing will, and quite remarkably so.  It could reduce the peak load on our health system (for the parameters examined here) by 75%.  That could literally mean that thousands of lives could be saved.

What was done during February, after the travel ban had been put into effect?  Sadly, not much.  There was no significant effort to identify and then isolate cases, and quarantine those exposed to those cases.  The development of a rapid COVID-19 specific test was also delayed as the initial version of the test turned out to be flawed.  While other nations around the world developed and quickly applied tests of their own, the US only tested (through other means) a small number of possible cases of individuals meeting highly restrictive criteria (such as recent travel in China).  And with only highly limited testing being done, the reported number of confirmed cases in the US was low.  But you can only confirm cases if you test, and if you do not test you will have no confirmations.

President Trump, and his administration, has yet to acknowledge its responsibility in this fiasco.  Trump has instead insisted that cases in the US are exceptionally low because, and only because, of the travel ban on China.  The numbers suggest otherwise.

B)  The Simple Mathematics of an Epidemic, and the Impact of a Travel Ban

One can model what an epidemic might look like (in terms of how fast it will spread) with some simple mathematics.  While this is far from the sophisticated models epidemiologists have for such processes, a simple model will suffice for an examination of the issue of what a travel ban might do.

The basic characteristic of an epidemic is that it will grow at an exponential rate to start with, but since it is subject to a ceiling (it cannot infect more than 100% of a population, and normally will tail off well before this point), the expansion will eventually have to level off.  A simple model with such characteristics is a logistic model, which was first proposed by a Belgian mathematician, Pierre Verhulst, in 1838.

The key parameter, called the “basic reproduction number” (and often designated as R0) is the number of new people who will, on average, be infected by a person who has been infected.  If that number is 2.0, then (to start) two new people will be infected by each person that has been infected, and the number of people who are infected at any given time will double in each period (to start).  If that number is 1.0, then (again, to start) one person will on average be infected by each person that has been infected, and the number of people who are infected in any given period will be constant (and the number who have cumulatively been infected will grow linearly over time).  And if the number is less than 1.0, then the number of new cases of infection will decline in each period, eventually going to zero (with the cumulative total climbing week to week as long as there are any new cases, but at a diminishing rate and eventually leveling off).

The basic reproduction number depends both on the characteristics of the disease, and on the degree of interpersonal contact in the society.  For the disease itself, some are more easily transmissible than others.  Measles, for example, spreads extremely easily.  Ebola (fortunately, as it has a high fatality rate) spread only if you had direct contact with bodily fluids, and hence did not spread easily.

But the spread also depends on what society is doing.  When people are in close direct contact, for example in crowds at concerts or in church or in a crowded subway car, more will be infected than if people are well separated.  Hence policy matters, and we will examine below the impact of measures that would reduce the degree of such close contact.

A key question for the virus that causes COVID-19 is how transmissible it is.  A number of scholars have hurriedly examined this, mostly using data from the initial spread in Wuhan, China, but have come up with a fairly wide range of possible figures.  The parameter is inherently hard to measure as data on the total number of people coming down with the virus week to week are simply not available, with the published figures possibly underestimates.  But a careful study published in The Lancet on March 11 estimated a figure of 2.35 in Wuhan before travel restrictions were imposed, falling to 1.05 after the rather draconian travel and quarantine measures went into effect.  An early study by a group of Chinese researchers published in the New England Journal of Medicine on January 29 (and summarized in an editorial co-authored by Dr. Anthony Fauci and others in the New England Journal of Medicine on February 28) arrived at an estimate of 2.2.  An estimate in a study published on February 22 and based on the spread of the virus in the cruise ship Diamond Princess came to a figure of 2.28.  And a review published on February 13 that examined as many other published studies as they could find up to that point (a total of 12, some of which might not have been of high quality) found a median estimate of 2.79, a mean of 3.28, and a range of 1.4 to 6.49.

I used an R0 of 2.3 for the calculations here.  It might be a bit on the low side, and if it were higher then the impact of a full travel ban (the main issue I am examining here) would be even less.  I am erring on the conservative side.  I am also, for these scenarios, looking at what the impact would be if that number remains unchanged over time.  That is, the scenarios examine what the impact would be if nothing is done to reduce the R0 by social distancing measures, either from policy (i.e. school closures) or simply by individuals being more careful and avoiding crowds or places where they could pick up an infection.  I stress again that these are scenarios of what would happen under specific circumstances, not forecasts of what will happen.

Assumptions are required for several other factors as well.  For simplicity, I am taking a discrete form of the logistic model, with calculations of week to week changes.  It is assumed that there will be a one week incubation period of a person who has been infected, and that that person can then infect others in their second week of infection.  After that, they can no longer infect others.  These assumptions are broadly consistent with what appear to be some of the basic parameters of the disease (based on material from a good summary article published in The Lancet on March 9), where the authors state that the mean time it takes for a newly infected person to pass the disease on to others is estimated to be 4.4 to 7.5 days.  So roughly one week after someone catches the virus they, on average, pass it on to others.

To examine the impact of a travel ban, I included as part of the model that a certain number of infected people would arrive from abroad each week, and that they would then add to those who could infect others domestically in the next week.  That is, those who would (domestically) be infected each week depends on the number who had been infected domestically in the prior week plus those infected who had arrived from abroad in the prior week.  To start, in period zero, I assumed there were 100 cases already active in the country domestically, and that 100 cases arrived from abroad.  I also assumed that the cases arriving from abroad, if nothing were done, would increase exponentially week to week (reflecting that the number of cases abroad are also growing) until they reached 10,000 per week (given that there are only so many who fly back and forth, even in normal times), after which the number was kept at 10,000 per week.

Finally, I set the ceiling on the population that might be infected by the virus at roughly one-third of the US population.  This model is too simple to forecast what that ceiling might be, so I used estimates made by others of the share of the US population that might in the end be infected if nothing is done.  But this ceiling is primarily just a scaling variable.  The results would not be impacted much by a different ceiling, within a reasonable range.  What matters is that, for the scaling used here, one starts with 100, caps those coming from abroad at 10,000, and has an overall domestic ceiling of over 100 million.

The scenario then looked at what would happen with a complete and total ban on anyone coming from abroad.  This would be far more extreme than any actual travel ban would be, as it would exclude returning American citizens and not just foreigners, plus it would cover travel from all countries in the world.  This was far more comprehensive than simply a ban on non-citizen arrivals from China.  But the aim was to be as generous as possible in calculating what the impact of a travel ban would be.

The chart at the top of this post shows what that impact might be.  It would not be much.  Even under such an extreme ban on travel, the path of the epidemic would be delayed by only about 2 1/2 weeks.  With other values assumed for the basic reproduction number R0 within a reasonable range, that time delay might be as short as 2 weeks or as long as 3 1/2 weeks.  None of these are large.  A travel ban would, at best, buy some time, but not much time.

But that extra time was not used in any case.  A travel ban in the very early stages of an epidemic can play a role if it is early enough (and February 2 was not early enough), and with then a major effort mounted to test all possible cases for the virus, with those testing positive isolated and those who had come into contact with such cases (or possible cases) quarantined.  None of this was done.

More modestly, what could have been done would be immediately to have increased social distancing, so that the infection rate (the R0) would be reduced.  The chart at the top of this post shows (in the curve in orange) what the impact would be had such measures been undertaken instead of the travel ban, and were sufficient to reduce the R0 to 1.5 from the 2.3 assumed in the other scenarios.  That is, the curve shows the impact where, on average, each infected person then infects a further 1.5 people instead of 2.3 people.  And again, to be clear, the curve assumes no foreign travel restrictions were imposed.

The spread of the disease is then slowed significantly.  Furthermore, the total number infected rises just to 75 million, or one-third less than come down with the disease in the base scenarios (with or without a travel ban).  The lower total number infected following from a lower R0 is an outcome of the random processes assumed in the logistic function, where as you approach the ceiling on the number who might be infected (the population), there is an increased likelihood that one will encounter only people who have already been infected and hence are now immune.  When one encounters fewer people (an R0 of 1.5 rather than 2.3), the likelihood goes up that all of the people encountered will be immune, and hence the number who will be infected in those later periods falls below 1.0.  The further spread of the disease then dies out.  It is for this reason as well that the curves for the case where R0 equals 2.3 level off at the odd number of 113.5 million.  I assumed a potential population of 120 million, and the logistic curve will level off below this.

Another scenario examined was one where the total travel ban was not implemented in week zero but rather in week six.  This would be similar to a delayed travel ban, such as that Trump recently imposed for travelers from Europe.  In the simple model, by week six the number of infected travelers coming in from abroad has reached its assumed peak of 10,000 per week.  I assumed that this was instead brought to zero in week six and then remained at zero.  The impact was trivial.  A plot of the new curve sits basically on top of the old (no travel ban) curve.  I therefore did not include it here as it simply looks almost exactly the same as the curve with no travel ban imposed.

C)  Impact on Cases to be Treated

As many have stressed, what matters is not only the total number of people being infected but also the number of new cases of infection each week.  Since about 20% of those coming down with the disease will likely need hospital treatment (based on current estimates), the burden on our hospital system will depend on how rapidly the number of new cases increase.  There are only a limited number of hospital beds, a far more limited number of the ventilators (about 160,000) that many of those who come down with this respiratory illness will require, and an even more limited number of beds in intensive care units (only 46,500, with perhaps a similar number that could be added in a crisis).  Furthermore, the patients that will need these ventilators and ICU beds may need to use them for two or three weeks.  This is far longer than would be the typical use of such hospital facilities for other disease treatments where they are required.

Hence, as numerous news reports have flagged in recent days, we need to “flatten the curve”.  That is, there is a critical need to reduce the peak load on such hospital facilities, with the need instead spread out over time.  A travel ban does not do this:

The peak loads on our hospital facilities would be almost exactly the same, with or without a total travel ban.  The peak is just shifted by 2 1/2 weeks.  In contrast, policies that by social distancing and other such measures reduces the basic reproduction number to 1.5 would have quite a marked effect, reducing the peak load by almost 75%.  That could directly translate into possibly thousands of lives that might be saved.  A travel ban does not help.

D)  Conclusion

The US is facing a major public health crisis.  Yet the response has been terribly mismanaged by the Trump administration.  Direction starts at the top, but Trump has repeatedly asserted that there is no major problem and that the disease will soon go away.  Even as late as March 10 (less than one week from when I am writing this), Trump said in remarks to the press at the White House that “And it will go away.  Just stay clam.  It will go away.”  He also continued to assert in those remarks that the ban on travel from China that he put in place, which he insisted others would not have done, had “made a big difference”.

But as shown above, imposing a travel ban, and one far more sweeping than the one Trump imposed on non-American travelers from China, will not have a major effect on the path of an epidemic such as the one we are facing.  This follows from the mathematics of compound growth as a disease spreads person to person.  At best it will buy some time, but plausible estimates are that it would amount to only a few weeks at best.  And those extra few weeks will only help if one makes use of that time to aggressively attack the disease.  That was not done.

Furthermore, a travel ban will not change the basic pattern of the epidemic.  It will merely shift it.  The peak loads on a stretched hospital system will remain the same.  Far more effective would be an early and sustained effort to promote social distancing.  This will not only reduce the total number getting the infection, but will also spread the infections out significantly over time.  Even a relatively modest reduction in the pace at which the disease spreads will have a major impact on those peak loads.  And reducing those peak loads on the hospital system can make a major difference in the number of deaths, reducing them by quite possibly thousands.

Why, then, the travel bans?  Probably because it may lead some to believe you are being serious and decisive, even macho, with such a clear-cut (albeit ineffective) measure.  Plus this makes it look like foreigners are to blame.  All this is appealing to someone like Trump.  And as he has repeatedly done throughout his term in office, he discounts the evidence-based advice of scientists with expertise in a field.  He thinks he knows better.

Sadly, Trump is accepting no responsibility for this fiasco.  On March 13, when asked specifically whether he accepts any responsibility for the delay of more than a month in rolling out the extensive testing that is critical early in an epidemic to identify and quickly isolate those infected, Trump replied “No.  I don’t take responsibility at all.”

The Democratic Primaries Thus Far: Bernie Sanders’ Vote Numbers

A.  Introduction

One of the main arguments Bernie Sanders has made for why he should be the nominee of the Democratic Party to run against Trump is that he would spur a much higher turnout, especially of young voters who would not otherwise go to the polls (with those young voters favoring him).  But this has not turned out to be the case in the Democratic primaries held thus far.  While turnout has gone up substantially, Sanders has not been receiving an exceptionally high share of that increased turnout.  And even Sanders has now acknowledged that a higher number of younger voters that he argued would go to the polls to vote for him have not materialized.

So what has been going on?  To summarize what will be discussed in more detail below, in the primaries held thus far the share of the votes going to Sanders has gone down compared to what he received in the same primary states in the 2016 elections.  But the share going to Sanders and Elizabeth Warren combined has been similar (indeed almost identical overall) to what Sanders received in 2016, when it was essentially only him running against Hillary Clinton.  Similarly, the share going to Joe Biden, Amy Klobuchar, Pete Buttigieg, and Michael Bloomberg has been similar to the share that had gone to Clinton.  This very much looks like a case of Democratic Party primary voters with a separation between those who hold the more extreme liberal views of Sanders and Warren, and those with the more moderate views of Biden, Klobuchar, Buttigieg, and Bloomberg (although it is not really correct to view them as moderates – the positions they hold are all well to the left of the positions that were held by Obama when he served as president).  Primary turnout has gone up, but with similar shares as before of voters in those two channels in that increased turnout.

Pundit commentary, at least until recently, has not focused on this.  Rather, in the Democratic primaries and caucuses held in February before South Carolina (i.e. following the contests in Iowa, New Hampshire, and especially Nevada), all attention was on Sanders winning the vote count (modestly in Iowa and New Hampshire, more significantly in Nevada).  It was not on what the outcomes might be telling us on the broader issue of who will, in the end, amass the delegates needed ultimately to win the Democratic nomination.  Sanders was deemed the “front-runner”.

And then all were surprised when the vote in the South Carolina primary appeared to be so different.  However, if a comparison had been made to the results of the 2016 primary in that state one would have seen important similarities.

This has now become more clear with the results from the Super Tuesday primaries.  Turnout (in all but one of the states) has gone up, and sometimes quite substantially.  The Democratic base is clearly energized.  But the higher turnout was not of voters disproportionately supporting Sanders.  Indeed, the share voting for Sanders has gone down compared to the share that voted for him in 2016.  Rather, across the states with primaries held thus far, the share going now to Sanders and Warren together is very close to what Sanders had received before, and the share going to Biden, et. al., was similarly close to what Clinton had received before.  Thus the higher turnout was composed of similar shares of voters in the two groups.

There were of course differences in several of the individual states.  For the analysis here I looked at the ten states who held primaries and not caucuses (vote counts in caucuses are different, with far lower participation), did so in both 2016 and 2020, and held their primaries in each of those years on Super Tuesday (March 1 in 2016, March 3 in 2020) or before.  Thus this excluded states like Colorado and Minnesota (which held caucuses in 2016), or had primaries (or caucuses) after Super Tuesday in 2016.  The most important, and largest, state thus excluded is California, which held its primary on June 7 in 2016.  I will discuss separately the special case of California.

The overall results for those ten states are summarized in the chart at the top of this post.  But rather than discuss that one first, it is perhaps better to examine the cases in a few of the states individually, before looking at the overall totals across the ten states.  The vote numbers are all as reported in the New York Times, at this post for 2016, or at this post for 2020.  The 2020 results are all as shown as of about 2:00 pm on Wednesday, March 4.  At that point, almost all were either complete (with 100% of precincts reporting) or close to it (with 99% or more in two cases, one at 97.0%, one at 93.8%, and one at 93.4%).  There will be some differences, but small, as they get to 100% of precincts reporting, and as mail-in ballots are fully counted (rules vary by state).  However, these will likely not affect the shares to any significant degree, which are the focus of the analysis here.  And while it will not change the shares, I did scale up to 100% the figures for the cases where fewer than 100% of the precincts had reported, in order to estimate what the total votes (and hence change in turnout) will be and to add up the figures consistently across the states.

B.  Individual States

The South Carolina primary, which was critical for Biden, shows well what the pattern has been.  The key results are summarized in this chart:

Sanders received only 26% of the vote in this primary in 2016, losing badly to Clinton who received 73% of the vote.  And that share of Sanders went down to 20% this year, even though there was a 46% increase in turnout.  But Sanders plus Warren together received 27% of the vote, almost the same as what Sanders received in 2016.  Despite an increase in turnout of close to half, the share going to the extreme liberal candidates remained about the same – not more, not less.

One saw the same in Virginia:

Here turnout rose by close to 70%.  And the Sanders share fell again, from 35% in 2016 to 23% in 2020.  But Sanders and Warren together received 34%, very close to what Sanders had received before.  Despite the far higher turnout, the shares were close to unchanged (taking Sanders and Warren together).

As noted above, there were a total of ten states where one could make such a comparison.  I won’t go through them all, and there were individual exceptions.  One noteworthy case was that of New Hampshire, the state with the first primary (Iowa is a caucus):

Bernie Sanders did exceptionally well in that primary in 2016, receiving 60% of the vote, against Hillary Clinton’s 38% (with other candidates receiving the rest).  Sanders won again in 2020, but this time with only 25.7% of the vote (with Pete Buttigieg in second place at 24.4%).  But while the pundits focused on Sanders winning that primary again, I did not see mentioned that despite an increase in turnout (of a not insignificant 18%), the absolute number of votes Sanders received fell in half (falling from 151,584 in 2016, to just 76,234 in 2020).  And even if one adds in the votes that Warren received, the total still came only to 103,711, with a share of 35%.

There were two other states where Sanders and Warren together did significantly worse than Sanders alone in 2020.  One was in Sanders’ home state of Vermont, where Sanders received 86% of the vote in 2016 while Sanders and Warren together received just 63% in 2020 (despite a 17% increase in turnout).  The other was Oklahoma, where Sanders received 52% of the vote in 2016 while Sanders and Warren together received just 39% in 2020 (and is the one state where turnout fell – by 7%).

These states were offset by Texas, where Sanders received 33% of the vote in 2016 (and 30% in 2020), but where Sanders and Warren together received 41% (with turnout rising 47%).  In the other states, the shares of Sanders in 2016 and Sanders plus Warren together in 2020 were pretty much the same.  Especially similar was the case of Massachusetts (the home state of Warren):  Sanders received 48.7% of the vote in 2016, while Sanders plus Warren received 48.3% in 2020.

California is also a special case, but an important one.  In 2016, the California primary was held on June 7, close to the end of the primary season.  Close to 5.1 million voted in the Democratic primary in that year, and Sanders won 45.7% of the vote.  As I write this (in the evening of Friday, March 6, and based on what is shown on the New York Times website), California has posted results for only 89% of the precincts.  Why this is less than 100% three days after the primary is not clear to me.  California also accepts mail-in ballots that were mailed on election day or before, and the state allows up to a month for these to come in.

But based on what has been reported as of now, Sanders plus Warren together received 45.9% of the votes, almost exactly the same as the 45.7% Sanders received in 2016.  But there was a big change in turnout, likely tied to the different election date.  While 5.1 million voted in 2016, the total votes recorded as of today is just 3.3 million.  While this will go up as all the mail-in ballots are counted (and as full reports are provided on all of the precincts), it will certainly not go up to anywhere close to the 5.1 million of 2016.

C.  The Ten States as a Whole

The chart at the top of this post reflects the figures added up across all of the ten states.  And one finds that as with most of the states (where the few exceptions basically offset each other), the share of the vote Sanders and Warren together received in 2020 (38%) was very close to what Sanders alone received in 2016 (39%).  The share of Sanders alone went down, with this offset almost exactly by the share Warren received.  And this was despite a substantial increase in turnout – of 34% across the ten states as a group.

In terms of what has been called the “more moderate” wing, the share across the ten states of those voting for Clinton in 2016 was 59%.  The share going to Biden plus Klobuchar plus Buttigieg plus Bloomberg in 2020 was 58%.  Again almost the same.

With turnout up by a third, the Democratic primary electorate appears to be energized.  There are real concerns about Trump, and what he has done to our country.  But the higher turnout is not because Sanders is pulling in a large number of new voters who will vote for him and him only.  Rather, the split in the new voters between those voting for Sanders or Warren on one side, or for Biden, Klobuchar, Buttigieg, or Bloomberg on the other side, is very close to the split between Sanders and Clinton voters in 2016.

With the withdrawal in the past week of all of the major remaining candidates other than Sanders and Biden, we will now see whether this pattern continues.  It is now basically a two-person race, and the results should be clear to all.

An Update on the Different Employment Estimates from the Survey of Establishments and the Survey of Households, And the Resulting Job Growth Under Trump vs. Obama

A.  Revisions in the Jobs Numbers

The pace of job growth in 2019 was slower than had originally been estimated.  While such revisions to the initial job growth estimates are not unusual (there is a regular annual process that adjusts them based on more complete data), the result for 2019 was that they now estimate there were 0.5 million fewer net new jobs than had been thought before.  Along with other revisions in the estimates going further back, the result is that the pace of job growth under Trump has slowed down by even more than had been thought earlier.  While this is not surprising (unemployment is low), it does point up even more strongly that Trump is simply wrong in his assertions that the pace of job growth during his term in office is “historic”, “unthinkable” (by anyone other than himself), and far faster than before.  See, for example, Trump’s remarks in January 2020 at the Davos meetings.  It was not true before the revisions – it is even less true now.

There were earlier indications that the jobs figures would be revised downwards.  A post on this blog in May 2019 discussed an inconsistency pointing to this that had developed in two estimates of employment growth in the US.  Both come from the Bureau of Labor Statistics (BLS), with one based on the Bureau’s monthly survey of households (the CPS, for Current Population Survey) and the other based on its monthly survey of business establishments (the CES, for Current Establishment Statistics survey).  Figures from these two surveys are released each month in the BLS Employment Situation report, which provides updated estimates on the unemployment rate, net job growth, and other such closely watched numbers.

Both the CPS and the CES provide estimates on employment growth, but they arrive at those estimates from two different sources.  And while there are some small differences in how “employment” is defined in the two (as discussed in that earlier blog post, where the impact of those differing definitions was examined), the two series over time will move together.  However, in the two years leading up to April 2019 the two series drifted significantly apart.  The CPS survey (of households) indicated a slower pace of net job creation than the CES survey (of establishments) did.

With the release of the January 2020 estimates on February 7, we now have updated figures.  And they indicate that job growth has indeed been slower than what the earlier CES figures had indicated.  The chart at the top of this post shows the differences, where all the figures are defined in terms of the change in jobs relative to their level in April 2017.  The curves (with the circles or squares) ending in April 2019 reproduce the chart from the earlier post (with the labor force figures removed, for less clutter), with the estimates on jobs as known at that point.  The curves (with no circles or squares) that end in January 2020 then show the more recent, updated, estimates.

The curves in blue, of the changes in employment as estimated from the CPS survey of households, show some revisions, but generally small and with no strong trend.  While there is a much greater degree of month to month volatility in the figures from the household survey, the revised figures basically follow what had been estimated before.  As was discussed in the earlier blog post, the CPS survey of households uses an effectively far smaller sample size for its employment estimates than the CES survey of business establishments has.  The CPS surveys a sample of 60,000 households each month, and a household will normally have only one or two members employed.  The CES survey, in contrast, surveys 145,000 businesses, covering almost 700,000 different worksites, and each worksite can have dozens if not hundreds of employees.

The employment estimates from the CES survey, shown in the curves in black on the chart, therefore show far less month to month fluctuation, due to the lesser degree of statistical noise.  But the new versus old estimates began to drift apart from each other around June 2018, with the discrepancy then continuing to widen steadily over time.  And the new estimates of employment based on the CES survey (the curve in black) now follows much more closely to the trend in the estimates of employment from the CPS survey (the curve in blue).  They came especially close to each other in the figures for October 2019, but have drifted apart by some since then (although not nearly as apart as what we saw in April 2019).

The changes are significant.  For April 2019, for example, the earlier estimates from the CES were that there were 151.1 million employed in the US (employed as defined in the CES).  The new estimate is that there were only 150.5 million employed in that month, a difference of about 600,000.  When looking at job growth, i.e. changes in the number employed over time, that difference is significant.

B.  Job Growth Under Trump Compared to Under Obama 

The updated estimates provide a clearer picture of how the job market has progressed in recent years.  But it is not as Trump often boasts.

With the publication of the January 2020 estimates, we now have figures on job growth for exactly three years into Trump’s presidential term.  These figures can be compared to the growth seen in the final three years of Obama’s presidency:

This presentation of the CES monthly employment growth figures is not original with me.  A number of news sources have presented something similar (although I have constructed the chart here from the original source BLS numbers).  But it makes the point well.

As one can see, there is a substantial degree of month to month volatility, even in these CES figures.  They are estimates of the month to month changes in total employment, and during Trumps’s presidential term thus far have varied from a high of over 400,000 in one month (February 2018) to a low of zero in another (February 2019).  But the average over the 36 months of Trump’s term in office thus far has been a monthly growth of 182,200.

This is well below the pace of employment growth during Obama’s last 36 months in office.  The average then was 224,400 net new jobs per month.  Trump’s repeated assertions that job creation is now faster is simply not true.

Nor was it true even with the earlier job growth estimates.  It is just even less true now:

Net Employment Growth

As Earlier Estimated

As Revised

Last 36 Months of Obama

Total

8,128,000

8,079,000

Per Month

225,800

224,400

First 36 Months of Trump

Total

6,913,000

6,559,000

Per Month

192,000

182,200

Difference in Job Growth

Total

1,215,000

1,520,000

Per Month

33,800

42,200

Under the earlier estimates, job growth had been an average of 225,800 per month over the last 36 months of Obama’s presidency, and 192,000 per month over the first 36 months of Trump’s term.  The difference was 33,800 more jobs per month under Obama compared to the period under Trump.  The difference as estimated now is 42,200 more.

And while these differences in the monthly averages may not appear to be much, over time they accumulate to a quite substantial difference.  The total growth in employment over the last 36 months of Obama’s presidency was 8,079,000.  Over Trump’s first 36 months it was slower, at a total of 6,559,000.  The difference is a not insubstantial 1.5 million jobs.  And it is higher than the 1.2 million job difference in the earlier estimates.

So Trump’s claims are simply not true.  That is important.  Trump is once again making assertions without bothering with whether or not they follow the facts.  But having said that, I would also note that this slowdown in the pace of job growth should not be at all surprising.  The unemployment rate has been low, it cannot go much if any lower, and hence an increase in the number employed can only come either from regular population growth or from an increase in the share of that population choosing to participate in the labor force.  The adult population grew by 150,600 per month during Trump’s 36 months in office, and the labor force by 137,800 per month.  This accounted for most of the 182,200 net new employment over the period.  The rest was from the reduction in the unemployment rate, from an already low rate of 4.7% when Trump took office, to the 3.6% now.  But the unemployment rate cannot go much lower.  Hence one should not be surprised that employment growth has slowed.

Still, it should not be a big request to expect honesty from a president.