Skip to content

Latest commit

 

History

History
164 lines (144 loc) · 9.34 KB

NEWS.md

File metadata and controls

164 lines (144 loc) · 9.34 KB

Changelog

  • 2020-04-12 - Georgia started separating non-state residents from their patient counts. Neither The New York Times nor Georgia adjusted the counts prior to the change. Without an adjustment, it destroyed the coherence of the data, so I’ve replaced them with South Carolina in the daily growth rate chart.

  • 2020-05-07 - The New York Times decided to combine “probable” and “confirmed” positive cases and deaths in their datesets so that the data would remain consistent across all states. I think it provides a more accurate description of what’s going, so I’ve decided to keep their data source and discard Indiana State Department of Health’s (ISDH). The ramification is that many charts’ data will be a day late. Hopefully ISDH will opensource their probable cases and probable deaths, so the charts can return to being up-to-date.

  • 2020-05-11 - The calculation of the rate for the Positive Test Rate chart was changed from using total counts to a rolling calculation over a 3 day windows. The interpretation has also changed based on the Johns Hopkins article.

  • 2020-05-12 - Replaced the chart that compares daily growth rates and doubling times of states with similar population densities with a social distancing chart that uses Google Maps data. The daily growth rates chart ceased being interesting as doubling times and rates have pretty much plateaued. Also, I don’t want to present too many charts at once as it might create some information overload. If a second or third wave happens, then this chart might return.

  • 2020-05-26 - Changed wording of the Hospitalizations - Ventilators - ICU Beds chart title. I misinterpreted the description of the hospitalizations data. I thought I was calculating the daily number of people being admitted to the hospital for COVID-19 when it was the change in the present count of people hospitalized for COVID-19. It’s still relevant because the governor’s speech used present count and not daily admittance as a guideline anyways. Apologees though.

  • 2020-05-29 - County chart - Instead of using all the historic data to calculate average daily growth rates, I’m switching to only using data over the past 2 weeks which will make the estimates more sensitive to outbreaks if they happen.

  • 2020-06-11 - Replaced the Apple Mobility chart with an OpenTables chart. Driving levels seem to have returned to pre-COVID levels, and with reopening stages well underway, it’ll be more useful to monitor how industries, like restaurants, hard-hit by COVID are recovering. The Apple Mobility chart will likely return if a second wave emerges though.

  • 2020-06-13 - Changed the range and scale of the axes on the daily-positives-cumulative-policy chart. The data trend was no longer exponential so the log scaling isn’t really needed and it was bunching up the data points. At this point in the pandemic, what happens after the reopening stages is salient, so I’ve started the data at the point a couple weeks before stage 2.

  • 2020-06-21 - Switched to the rt.live model for Re. I haven’t been confortable with estimate I’ve been using for awhile. Their estimate along with their uncertainty range look more reasonable than the one I was using before. They haven’t published their code yet, so I haven’t been able to dig into it too much. But the faq gives some hints to what they’re doing, and it seems sound.

  • 2020-07-08 - Daily positive rate - During a briefing I watched with Gov. Holcomb and Dr. Box, it was pretty clear that they pay close attention to positivity rates and use a seven day window to calculate theirs, so I’m switching my calculation window to 7 days. A 5% percent target was also emphasized by Dr. Box, so also changing my shaded region to a range from 0% - 5%.

  • 2020-08-17 - Daily positive rate - I calculate the rate for all the data I have, but I don’t display the most recent two I’ve calculated, because there isn’t enough data to give a reasonable rate. I’ve decided remove another daily rate. So the last two rates that I calculate won’t be displayed; for the same reason. The lack of data makes these rates misleading, and some people don’t read the chart descriptions.

  • 2020-09-18 - Daily positive rate, Median Age Weekly Cases and Tests, COVIDcast MSA Positivity - For about a month, ISDH’s dashboard has shown and Dr. Box has been referencing a much lower positivity rate, but haven’t included the testing data used to calculate that rate at Indiana Data Hub. Not sure when COVID_TESTS_ADMINISTRATED [sic] was added to the county-level dataset, but it’s there now and I’ll be using that field to calculate my positivity rate.

    The previous field I used for test counts only counted tests for unique individuals (i.e. no repeat tests of individuals were included). If a person was tested in March and then again in July, each test should be recorded for that month in order for the positivity rate to more accurately reflect the state of testing and viral spread. This new field includes these repeat tests and substantially increases the test counts which therefore lowers the positivity rate as a result.

    COVID_TESTS_ADMINISTRATED is an improvement, but it’s not ideal and will give a deflated positivity rate. This new field unfortunately includes daily repeated tests when the count of individuals tested per day is what is actually needed. These COVID-19 diagnostic tests are not perfectly accurate, so false positive and false negative testing results occur. To increase the probability of getting a “true” test result, multiple tests for the same person on the same day may occur. But given that these PCR diagnostic tests are pretty expensive and information I’ve received from people who have been tested, I’m not inclined to think multiple daily tests of the same individual occur that often. With cheaper tests now developed, maybe this becomes a larger issue in the future though. Some states are currently providing a count of individuals tested per day which is named, “test encounters.” A more detailed discussion of this test count is discussed in a post over at The COVID Tracking Project. Hopefully, ISDH will start using test encounters in the near future.

  • 2020-12-20 - Hospitals page, Local Hospital Capacity - The number of available beds used in the calculations is fluid, because only the number of fully staffed beds is reported. Therefore, if a hospital suffers from personnel shortages, the number of available beds reported by that hospital will decrease.

    When any of the counts for any of the data fields is below four, the true count is obscured by the CDC. This is due to the concern that this data combined with other publicly available data might be used violate patient privacy. Anytime that 80% of a hospital’s data used in a calculation was missing and/or obscured, I decided not to perform the calculation for that hospital (empty table cell). If some of the data (but less than 80%) were obsured, then I substituted a “2” for the obscured value and proceded with the calculation.

  • 2020-12-25 - Hospitals page, State Mortality Rate, Staffing, and Admissions - The hospitals deaths and admissions were scraped from the Regenstrief dashboard, but I did check the robots.txt file and followed the rules therein. It’s my first time scraping a Tableau dashboard, so I’m not a hundred percent about the data. It checks out visually with the dashboard sometimes and in some places, but not others. I only collect it twice a day, so maybe the data is being updated before or after I’m collecting. The tableau worksheets and variable names from where I’m getting the data all look right, so I think I’m getting the right stuff.

    The HHS staffing data only includes reporting hospitals, but for the time period I’m looking at and foreseeable future I don’t anticipate much non-response bias. The last time I looked only around 3 hospitals weren’t reporting their staff shortage numbers. That seemed to be the case at least over the last couple months. And looking at the “Local Hospital Capacity” table there seems to be good response there as well. So this isn’t something I’ll be monitoring too closely.

    I chose 14 day rolling window for the mortality rate calculation because the average hospital length of stay for all age groups didn’t exceed that length on the Regenstrief dashboard. So, averaging over a 2-week period should be representative. For the other two calculations, a seven day window seems be a good number to smooth out reporting irregularities, and I think that’s the only concern I have with those samples.