Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Coronavirus New York state update: switching from Johns Hopkins to NYT data

April 19th, 2020, 9:35pm by Sam Wang

Over the last few days, our doubling-time tracker has showed steady progress toward longer times in nearly every state – except New York. We think we’ve identified a source of inaccuracy: uneven updating at the Johns Hopkins site. They’re excellent, but their data isn’t intended for visual display. So we’re switching to the New York Times feed.

Here’s our revised graph based on NYT data:

Further thoughts after the jump.

The Hopkins data is itself just a collection scraped from many sources. On the other hand, the NYT is doing more traditional on-the-ground reporting of cases and tallying the data in-house. In addition, the NYT is making its own graphs, which means that as they apply their editorial judgement on what to count, they can see anomalies in the data as they develop.

Most states aren’t affected by this switch – with the prominent exception of New York.

The Johns Hopkins graph:

The Hopkins data shows a rapid increase in deaths late last week that contradicted most other reports. They appear to have since revised that number downward. My best guess is that it was a consequence of the decision to start including probable covid-19-related death counts (as opposed to only patients who tested positive), since there was some reporting on that earlier in the week. Nevertheless, it does seem the Times has more people attending to their data pipeline, and probably more attention to the accuracy of USA-specific data. That’s not to say we don’t trust the JHU data, it just may take them longer to correct errors. Since the NYT is currently showing more deaths, that may be a better measure to capture everything that’s happening.

Tags: Health

20 Comments so far ↓

  • Ruhrfisch

    Thanks for the doubling figures, which are both interesting and helpful. Do you worry about the lag in number of deaths in the New York Times website, compared to other sources? So at 1:12 PM on Tuesday April 21, NYT has 37,818 deaths in the US, while USA Today reports Johns Hopkins site shows over 42,000 deaths nationwide (at least 11% more).

    For Pennsylvania, NYT shows 1,366 deaths ( but the PA Dept of Health shows 1,564 (14.5% more,

    I know it is good to use a single (aggregate) source of data, but deaths are already a lagging indicator, and the NYT has a substantial lag on top of that. How would this affect doubling times? Thanks again

  • Partha Neogy

    I noted that the doubling time for deaths in the US on April 22nd jumped to 24.6 days (compared to 8.8 days the previous week). This is both encouraging and expected when we approach a peak in the daily death rates. As we go past the peak should we expect, within data variations, an essentially infinite doubling time?

  • ArcticStones

    Here in Maine, the number of active cases has been increasing only marginally in recent weeks. Only four states have fewer active cases. And in the last few days it has actually decreased!

    From 440 on Sunday, to 426 Monday, to 409 yesterday.

    So grateful Mainers replaced Governor Paul LePlague with the highly-capable Janet Mills and a Democratic trifecta! And Dr. Nirav Shah, the Director of the Maine CDC has been doing a phenomenal job.

  • Jack Rems

    The headline U.S. doubling says 20.1, the chart on the right says 12.8. The NY and NJ numbers match.

  • Partha Neogy

    Is there sufficient indications of a trend in the US doubling time data to guess when we get to the downward slope of the daily mortality rates?

  • bks

    Since 1 April the rate of new infections has been strictly linear as has the rate of death. 100,000 new cases each 3-4 days and about 1100 deaths each week.

  • Partha Neogy

    The doubling times (for the US and the regions) appear to show oscillations with peaks occurring every 7 days or so. Any significance to that number?

    • PJC

      Yes. States reported differently depending on the DOW. For example, many don’t report on Sunday. Data for Saturdays, Sundays, and Mondays can often show some kind of weekend effect that look like pre and post weekend oscillations in bar charts.

      Waves of aggregated state data can then combine to show weekly waves in regional and national data.

  • Partha Neogy

    @ bks

    Thanks. Interesting.

    • Jay Sheckley

      Agree this is of interest.
      Seems U.S. is unaware of these vital stats

  • Jay Sheckley

    Every day I wonder what you as a statistician see as the possible actual death tolls.

  • Sean

    I’m in West Virginia and the 105 days time to double is certainly reassuring. I’m near Charleston and from what I see personally most people are taking it seriously around here, even the die-hard Trump supporters. I think the social distancing combined with our geography and low population has helped produce this number. I just hope we continue to improve and the rest of the world as well.

  • Jack Rems

    WY seems to be missing from the 50-state doubling tracker.

  • Steve Bannister

    Hi Sam, the updates for seem to have stopped on 6/11. Is this intentional? My interest is that I have been downloading your files for my own analysis.

Leave a Comment