ASA Connect

 View Only

  • 1.  COVID-19 Scattergram in Wall Street Journal

    Posted 04-27-2020 11:15
    Yesterday's Wall Street Journal contained an op-ed piece entitled "Do Lockdowns Save Many Lives? In Most Places, the Data Say No: The speed with which officials shuttered the economy appears not to be a factor in Covid deaths" (T.J. Rodgers, April 26, 2020 3:55 pm ET). The op-ed included a scattergram.
    The author explained that
    "To normalize for an unambiguous comparison of deaths between states at the midpoint of an epidemic, we counted deaths per million population for a fixed 21-day period, measured from when the death rate first hit 1 per million-e.g.,‒three deaths in Iowa or 19 in New York state. A state's 'days to shutdown' was the time after a state crossed the 1 per million threshold until it ordered businesses shut down.
    We ran a simple one-variable correlation of deaths per million and days to shutdown, which ranged from minus-10 days (some states shut down before any sign of Covid-19) to 35 days for South Dakota, one of seven states with limited or no shutdown. The correlation coefficient was 5.5%-so low that the engineers I used to employ would have summarized it as "no correlation" and moved on to find the real cause of the problem. (The trendline sloped downward-states that delayed more tended to have lower death rates-but that's also a meaningless result due to the low correlation coefficient.)
    "No conclusions can be drawn about the states that sheltered quickly, because their death rates ran the full gamut, from 20 per million in Oregon to 360 in New York. This wide variation means that other variables-like population density or subway use-were more important. Our correlation coefficient for per-capita death rates vs. the population density was 44%. That suggests New York City might have benefited from its shutdown-but blindly copying New York's policies in places with low Covid-19 death rates, such as my native Wisconsin, doesn't make sense."
    He then proceeded to talk about "common-sense guidelines" in Sweden.
    We don't usually see scattergrams (and r^2) in the news. I have not thought through this one, but it might prompt some reactions.

    ------------------------------
    David Kaye
    Penn State Law
    ------------------------------


  • 2.  RE: COVID-19 Scattergram in Wall Street Journal

    Posted 04-28-2020 08:57
    He analyzes a dataset and concludes, "The correlation coefficient was 5.5% ..." Aside from confusing correlations with percentages, he misfits a linear regression line to his data and honors what Tversky and Kahneman called "the law of small numbers." I was going to reanalyze his dataset, but couldn't locate it online. I'm fearfully waiting for some politician to cite this as "scientific evidence." What have we come to when the WSJ publishes nonsense without, apparently, having a statistician review a statistical analysis?


    ------------------------------
    Leland Wilkinson
    H2O
    ------------------------------

    Original Message


  • 3.  RE: COVID-19 Scattergram in Wall Street Journal

    Posted 04-28-2020 11:14
    The graph also does not take into account the lockdowns that happened early because death rates were high (NY being a prime example).  The primary driver of this correlation is likely due to the institution of lockdowns once it was clear that COVID19 was spreading rapidly in the population.

    ------------------------------
    Jean O'Malley
    Biostatistician
    OCHIN, Inc
    ------------------------------

    Original Message


  • 4.  RE: COVID-19 Scattergram in Wall Street Journal

    Posted 04-29-2020 10:07
    KInd of like that old joke of a statistician who studied fire department records and concluded that the more fire engines you sent to a blaze, the more damage they caused.

    ------------------------------
    Stephen Simon, blog.pmean.com
    Independent Statistical Consultant
    P. Mean Consulting
    ------------------------------

    Original Message


  • 5.  RE: COVID-19 Scattergram in Wall Street Journal

    Posted 04-28-2020 11:19
    The relation is nonlinear and usual correlation coefficient fails for nonlinear relations as argued
    in my paper
    Vinod (2017), "Generalized correlation and kernel causality with applications in development
    economics," Communications in Statistics - Simulation and Computation, 46, 4513{
    4534, available online: 29 Dec 2015, URL https://doi.org/10.1080/03610918.2015.
    1122048.
    Also, one needs to consider partial correlation here not just bivariate correlations.
    The population density is an important control variable they did not consider.  My R
    package generalCorr  has routines to do nonlinear correlations and partial correlations.

    I would like to get a hold on Wall Street J data to show what I mean more explicitly.  Anyone has
    the data? Please contact vinod@fordham.edu

    ------------------------------
    Hrishikesh Vinod
    ------------------------------

    Original Message


  • 6.  RE: COVID-19 Scattergram in Wall Street Journal

    Posted 04-28-2020 12:41

    I doubt that the group collected data on many variables, but I did write the Wall Street Journal as follows:

    I wonder if the Journal will make available the data points for the diagram in the op-ed "Do Lockdowns Save Many Lives? In Most Places, the Data Say No" (T.J. Rodgers, April 26, 2020 3:55 pm ET)? On a discussion board of the American Statistical Association, a number of statisticians have expressed interest in obtaining the data set (that is, the exact values for points that are plotted). In the sciences, it is normal for researchers to make their data available to other experts for replication and reanalysis, and that seems particularly appropriate when questions of life and death are involved.
    Thanks for considering this request,

    If you want to ask whether the authors will provide the full data set they created before running the simple regression, the email address I found for the editorial and opinion department of the WSJ is edit.features@wsj.com. The diagram only stated "source: covidtracking.com covid19.healthdata.org". (Letters to the editor for possible publication would go to wsj.ltrs@wsj.com (Editor, The Wall Street Journal,1211 Avenue of the Americas, New York, NY 10036).)



    ------------------------------
    David Kaye
    ------------------------------

    Original Message


  • 7.  RE: COVID-19 Scattergram in Wall Street Journal

    Posted 04-30-2020 15:15
    Regarding the raw data: Although I couldn't find contact information for the author, T.J. Rodgers, I did find on Twitter one of the data analysts he mentions as a collaborator: Yinon Weiss. He has his direct-messaging turned off, but he might respond to a direct tweet. I'm happy to do that if others don't feel comfortable doing so.

    Alternatively, in the past I've had good success with WebPlotDigitizer to extract data from a published scatterplot. 

    Regina.

    ------------------------------
    Regina Nuzzo
    Senior Advisor for Statistics Communication and Media Innovation
    American Statistical Association
    ------------------------------

    Original Message


  • 8.  RE: COVID-19 Scattergram in Wall Street Journal

    Posted 04-29-2020 13:15
    I find it weird that a publication like the Wall Street Journal wouldn't use economics!

    One thing all of these types of articles seem to forget is the simple risk-reward models set up by the federal govt and updated by those "liberals" Reagan and Bush. (I know they were Republicans. No need to write about that.) In the mid 80's, there was an update to the risk-reward calculations to say that a human life is worth $3,000,000. Now that is up to $10,000,000. So, each life saved is $10,000,000. 

    Suppose that the lock downs saves 100,000 lives. The net gain would be ($10,000,000)(100,000) = $1,000,000,000,000 

    If 1,000,000 lives were saved then the benefit is $10,000,000,000,000 or about half a years worth of GDP. 

    That's all assuming that people are not smart enough to stay home on their own. And that people would be spending at the same rate as they were before. 

    Meaning that if a lock down saves enough lives, it is a NET POSITIVE for the economy. But, the longer it goes, the less it is worth it. Basic economics. That is what they should be reporting.

    ------------------------------
    Andrew Ekstrom

    Statistician, Chemist, HPC Abuser;-)
    ------------------------------

    Original Message


Related Content