a thoughtful web.
Share good ideas and conversation.   Login or Take a Tour!
comment by nil
nil  ·  11 days ago  ·  link  ·    ·  parent  ·  post: Body Count

¯\_(ツ)_/¯

> summary(linear.model)

``    Call:``    lm(formula = cases ~ day)``

``    Residuals:        Min       1Q   Median       3Q      Max ``    -1580.59  -812.36   -30.43   502.35  2388.39 ``

``    Coefficients:                Estimate Std. Error t value Pr(>|t|)        (Intercept) -2859.08     355.63  -8.039 4.45e-09     day           495.69      18.25  27.159  < 2e-16     ---``    Signif. codes:  0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1``

``    Residual standard error: 998.3 on 31 degrees of freedom    Multiple R-squared:  0.9597,	Adjusted R-squared:  0.9584 ``    F-statistic: 737.6 on 1 and 31 DF,  p-value: < 2.2e-16``

``    Call:``    lm(formula = cases ~ day + day2)``

``    Residuals:        Min       1Q   Median       3Q      Max ``    -1009.28  -347.69    71.42   314.88  1089.37 ``

``    Coefficients:                Estimate Std. Error t value Pr(>|t|)        (Intercept) -791.709    275.600  -2.873 0.007404      day          141.285     37.373   3.780 0.000696     day2          10.424      1.066   9.775 7.75e-11     ---``    Signif. codes:  0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1``

``    Residual standard error: 496.1 on 30 degrees of freedom    Multiple R-squared:  0.9904,	Adjusted R-squared:  0.9897 ``    F-statistic:  1541 on 2 and 30 DF,  p-value: < 2.2e-16``

> summary(exponential.model)

``    Call:``    lm(formula = log(cases) ~ day)``

``    Residuals:        Min      1Q  Median      3Q     Max ``    -1.6131 -0.5552  0.1115  0.5793  0.8511 ``

``    Coefficients:                Estimate Std. Error t value Pr(>|t|)        (Intercept)  4.65189    0.25168   18.48  < 2e-16     day          0.18010    0.01292   13.94 6.72e-15     ---``    Signif. codes:  0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05 ‘.’ 0.1 ‘ ’ 1``

``    Residual standard error: 0.7065 on 31 degrees of freedom    Multiple R-squared:  0.8625,	Adjusted R-squared:  0.858 ``    F-statistic: 194.4 on 1 and 31 DF,  p-value: 6.724e-15``

Looks like the Mexican Government has got some 'splainin' to do. Either that or Ben CHunt and Redditeurs are talking out of their ass once again and this is nowhere close to Matterhorn-esque truth. In fact, the exponential model actually fits the Coronavirus values better (albeit with fewer data points) (N=19, Exponential Fit: R=0.926, Quadratic Fit: R=0.9939).

All epidemics take the form of an exponential function, not a quadratic function.

That's what I call,

in Epsilon Theory-speak, a cartoon – an abstraction of an abstraction in service to the creation of Common Knowledge.

References:

[3] X-scale is Day 1 to 33. Y-scale is number of cases with the first (X,Y)-point being "World" confirmed cases on April 24, 2009.

[4] The variable day2 is equal to day^2.

kleinbl00  ·  10 days ago  ·  link  ·

So you're going to throw up a bunch of matlab? LaTEX? Mathematica? And say look everyone else is an idiot?

R0 is a power function. Period. It is defined as a power function. That's not a "cartoon" that's a definition. It is a coefficient of exponential growth. And while the argument was made in the article that certainly a curve could be fit using a power function, the point was that the numbers being generated clearly weren't.

As to your model, you picked a tough one to get mathy on:

For exam-ple, among individual states in India, the reproductive number for 2009 H1N1 ranged from 1.03to 1.75; likewise, estimates in Peru spanned from 1.2 to 2.2 depending on the specific region studied. Even close geographic neighbors had disparate R0 estimates; China estimated a mean R0 of 1.68, whereas Japan initially approximated a mean of 2.3,which was later reduced to 1.21 to1.35. Correspondingly, in Canada the mean estimate was 1.31 whereas public health officials in the United States initially esti-mated between 2.2 and 2.3, which was subsequently refined to1.7 to 1.8 with additional data collection. On the other hand,not all subsequent estimates of were downwardly biased. Fraser et al. were among the first to estimate the R0 in Mexico, proposing a basic reproductive number of 1.4 to 1.6. Just several months later, another team estimated the R0 was between 2.3 and 2.9.

So even in an article where the argument R0 is tough to estimate and really only useful to discuss indigenous spread where it is measured, you didn't get within a mile of anyone's published estimates. And for the record? One of the hallmarks of the 2009 Mexican Swine Flu outbreak was uncertainty and doubt about the numbers. So much so that I opted to drive from LA to Seattle because I had been told they were going to close the airports.

Looks like the Mexican Government has got some 'splainin' to do.

Indeed. I have a pretty good idea what point you're trying to make? But I can't say that you've made it.

nil  ·  9 days ago  ·  link  ·

For the record, that was more work than the analysis on r/dataisbeautiful. It was more work than Ben Hunt did (with the exception of writing his life story and essay). I'm not sure where the ambiguity is. I said like 3 people were wrong, not everyone. Which they are.

It's completely insane to run a quadratic fit in Excel and claim that is evidence of anything. It's even worse to see someone with an alleged background in statistics doing it because they're either being intellectually dishonest or they just don't give a fuck. His credentials are not only irrelevant to the discussion, they aren't in the same lane. He studied art for 7 years and taught a stats class? I know people in high school, maybe I'll ask them if they think this is a bombshell. I'll ask my brother, he's like halfway through a community college diploma in tech. Let's see what he says. creeedeennnnnttiiiialaaaaaallllsssss bro. I wrote an essay on why we need FreeMarket(tm).

"2 + 2 = 5" and we haven't redefined addition in this vector space (Old White Guy, Ph.D in Math, Harvard 1904). Damn dude he's in his lane and being totally right! Seems like a joke but I just got back from a talk of a dude with a math degree claiming he had proved the brain is a computer based on some gradient model. It's not proof, it's a contention.

Yes, R0 is a power law. But how is this supposed to work? There's an exponential spread, for how many days? What does it mean by "until measures are taken to bring the disease under control?" Isn't that already happening? People who live thousands of miles away are wearing face-masks on a daily basis. The Chinese grocery around here is in threat of going out of business because people are that paranoid.

Now, using an exponential fit the entire planet would be infected in less than 100 days. So at some point between day 1 and day 100 this thing is going to stop exhibiting exponential behaviour. So is the R^2 value of the exponential fit supposed to move from 0.99 to 0.95 after day 30? What behaviour are we exactly looking for here? I didn't even go as far as to do an analysis of residuals or any of the other BS we could be looking at with least-squares regressions. Not to mention accumulating the number of cases over time and running a regression on that is flaky to begin with.

The R0 for measles is estimated to between 12-18. Under those circumstances any measles outbreak would infect the entire planet in like a week. Does that mean every measles outbreak was fabricated? Maybe if I have unlimited free time I'll run an exponential fit on a measles outbreak in the DR Congo to prove they're fabricating numbers too. By your argument what should be expected is a very rapid exponential rise in cases (remember 18 people from 1 grows insanely fast) followed by a levelling off.

That livescience article you linked mentions the fact R0 isn't exactly known for nCov. And this article by them suggests the WHO estimates are anywhere from 1.4 to 2.5. Other organizations have higher estimates, but this isn't an exact science.

Now because I have to satiate everyone's impulses because this is U.S. vs. China basically - yes it's possible they're making shit up. Yes, people that work for the government in China serve bullshit on a platter. Probably even likely. But shit like this isn't proof, nor is it even a worthy contention. It's garbage. It's worse than what China does, because they don't actively finesse the uneducated when they drop outright bullshit like the president can walk through walls or whatever is hot in communist countries. That Lancet study is actually worth discussing rather than some borderline racist bullshit from Reddit.

kleinbl00  ·  9 days ago  ·  link  ·

Let's review:

Three days ago, the argument was that China was lying about the numbers.

Two days ago, your argument was that your analysis proves that nobody knows anything about numbers and nobody should accuse anyone of lying about numbers.

Today, your argument is that your analysis is more work than anyone else's numbers (never mind that it's spurious) and one, two, three, four, five, six paragraphs about how everyone's numbers are bullshit. "borderline racist bullshit" in fact.

The argument at the outset was that China was pushing a false narrative. Three days later, China has relieved party heads and has changed their narrative.