Skip to main content

The Seasonal/Geographic Pattern, Part 2

By March 8, 2021Commentary

DD has done a variety of work attempting to tease out a clear relationship between cases and certain meteorological or geographic factors.  He looked at temperature, humidity, hours of daylight, latitude, altitude and some other factors.  The obvious problem is that if each of these has an effect on transmission–via the virus, the human host or the act of transmission–trying to figure out which ones dominate at any point or the relative contribution of each is very difficult.  And geographic and meteorological factors can be small in area, a state doesn’t have one set of factors uniformly.  Furthermore, there are obvious relationships between factors.  Latitude may just be a substitute for hours of daylight or sunlight intensity or temperature.  Temperature and humidity often are related.

Notwithstanding these difficulties, it is apparent that there are broad latitudinal patterns, with case waves in the northern hemisphere generally moving north to south, at least in the temperate latitudes.  You can see this in charts that people other than DD have put together.  How much this is due to specific meteorological variables is less clear, but latitude, going north from the equator, is generally correlated with less sunshine, less intense sunshine, and lower temperatures, perhaps lower humidity.

Hospitalizations may be better to use to detect patterns than cases, given wide variation in testing strategies, etc. that can affect case reporting.  Here are DD’s conclusions looking at hospitalizations versus some weather variables.  While there appears to be some effect and the effect is greater with a combination of factors, in general the association is not overly compelling and latitude, which to some extent may reflect multiple variables, tends to have the strongest associations.     It is probably a matter of how the meteorological variables interact with factors like population density, population age and health structure, and behavior patterns.

  1.  The later in the year the rise in hospitalizations started the colder the daily temperature was. However, the R-squared value was only 0.0835, meaning that temperature only accounted for 8.35% in the variation in the date of the start of the rise in hospitalizations.
  2. It can be seen that when the rise in hospitalizations started at warmer temperatures that this was weakly correlated to hours of daylight (11.86% correlation). Attempting to correlate hours of daylight to date of start of the rise in hospitalizations simply yields a straight-line correlation so is not beneficial.
  3.  Latitude is only weakly correlated with the start date of the rise in hospitalizations (16.64% correlation).
  4.  The most common average daily temperature 12 days before the start of the rise in hospitalizations was 65 to 75 degrees (29 states).
  5.  The most common hours of daylight 12 days before the start of the rise in hospitalizations was 12.5 to 13.5 hours (23 states).
  6.  The mean average temperature 12 days before the start of the rise in hospitalizations was 65.6 degrees F with a standard deviation of 7.6 degrees F, and mean hours of daylight was 12.0 hours with a standard deviation of 0.9 hours.
  7. A linear regression fit was also computed between latitude, average temperature, and hours or daylight 12 days before the start of the rise in hospitalizations. The equation Hours of Daylight = 1.69 + 0.129 * Latitude + 0.080* Average Temperature was computed and found to have an R-squared value of 44.0%. These variables can be considered to be weakly correlated. Since the r-squared is well below 100% other variables, or different formulations of temperature, latitude, and daylight, affect the start date of the rise in hospitalizations for each state. It could be considered surprising that only 3 variables could correlate at roughly 50% for the start of the rise in hospitalizations, especially considering the relatively crude assumptions used to obtain temperatures, cases, and latitude.

Here are similar conclusions relating to cases.

  1. Unsurprisingly, the later in the year the rise in cases started the colder the daily temperature was. However, the R-squared value was only 0.2064, meaning that temperature only accounted for 20.64% in the variation in the date of the start of the rise in cases.
  2. It can be seen that when the rise in cases started at warmer temperatures that this was weakly correlated to hours of daylight (27.53% correlation). Attempting to correlate hours of daylight to date of start of the rise in cases simply yields a straight-line correlation so is not beneficial.
  3. Latitude is only weakly correlated with the start date of the rise in cases (18.64% correlation).
  4. The most common average daily temperature 7 days before the start of the rise in cases was 65 to 75 degrees (24 states).
  5. The most common hours of daylight 7 days before the start of the rise in cases was 11.5 to 12.5 hours (23 states).
  6. The mean average temperature 7 days before the start of the rise in cases was 64.8 degrees F with a standard deviation of 7.7 degrees F, and mean hours of daylight was 11.8 hours with a standard deviation of 0.9 hours.
  7. A linear regression fit was also computed between latitude, average temperature, and hours or daylight 7 days before the start of the rise in cases. The equation Hours of Daylight = 3.03 + 0.099 * Latitude + 0.076* Average Temperature was computed and found to have an R-squared value of 54.5%. These variables can be considered to be moderately correlated. Since the r-squared is well below 100% other variables, or different formulations of temperature, latitude, and daylight, affect the start date of the rise in cases for each state. It could be considered surprising that only 3 variables could correlate at roughly 50% for the start of the rise in cases, especially considering the relatively crude assumptions used to obtain temperatures, cases, and latitude.

Join the discussion 3 Comments

  • Narcissis Pettit says:

    How does the seasonal progression of Covid compare to influenza? I ask because you’ve previously noted influenza deaths are way down, suggesting Covid may be taking the blame for flu deaths. One way to disprove that would be to show Covid has a markedly different seasonal spread pattern. Does it?

  • Kevin Roche says:

    the pattern appears quite similar. I don’t think CV-19 is being blamed for what are really flu deaths, but it is substituting for those, CV-19 appears to be able to “out-compete” flu and exclude it from infecting susceptible patients who are infected by CV-19 but otherwise might get flu

  • The Dark Lord says:

    this sort of number crunching is a form of intellectual masturbation … its makes the user feel good but contributes nothing to our knowledge of the issue … start with the virus and what causes it to infect people then work from there … they don’t clearly know how it spreads yet … they don’t clearly know what kills it … until they know those 2 things this sort of data analysis is just blind groping for a theory …

Leave a comment