2004 approaches 1999 record for most Category 4 hurricanes.

[NNadir]

This year we've had four hurricanes reach category 4, Charlie, Frances, Ivan and Karl. Two Ivan and Frances reached Category 5.

Alex was a strong category 3.

The season isn't over of course.

The record holder was 1999, when Bret, Cindy, Floyd, Gert and Lenny were all Category 4 or greater. http://www.disasterrelief.org/Disasters/991129hseasonreview/

It seems pretty clear to me at least that the occurance of two years of extreme hurricanes within 5 years is hardly a statistical quirk. (The link above says that 1995-1999 produced the most hurricanes on record in a five year period.)

Of course we should expect that new records every two years for the rest of our lifetimes would not convince some people.

[Boomer]

1. Spot on

So we've known about this upswing in hurricane activity for at least 5 years, but coastal development has not slowed one whit.

[phantom power]

2. I predict that insurance is what finally slows the development

Somebody posted an article about how many people in Florida were going to discover that the insurance companies had already raised their deductibles to something like 2% the value of the home, after the previous large hurricane strike.

I bet that deductible is going to double after this year.

[One_Life_To_Give]

3. Depends on how you present the data

Of course we should expect that new records every two years for the rest of our lifetimes would not convince some people.

How the data is presented will have a big effect on what people choose to beleive. Tell people that Hummers must be outlawed because we have a record number of hurricans and you won't find many listeners other than those who already hated Hummers to start with.

[Donkeyboy75]

4. And, IIRC, an increase in hurricanes and cyclones was to be an

effect of global warming. Funny.

I don't have access to my statistics textbook (and it's been ten years since I took it), but I wouldn't be sure that two records in five years is statistically significant. But lumped together with all the other global warming evidence...

[phantom power]

5. depends a lot on what underlying distribution you assume

Or, you could use some nonparametric test. We could run a Kendall's Tau test, to see if there's any positive correlation between either intensity, or number, of hurricanes with respect to passage of time.

[Donkeyboy75]

6. I don't remember doing those,

so that's probably out of my league. I was thinking more of a simple p test. Ah, nevermind. Don't remember my stats well enough right now...

[phantom power]

7. Kendall's Tau is maybe a bit obscure, but it's great for

determining whether some variable Y is increasing or decreasing with respect to some other variable X, if you don't want to make any assumptions about what (if any) function Y might obey. So, for example, you don't have to assume that Y is a linear function of X, or quadratic, etc.

I like nonparametrics, because assumptions of linearity, or any other function, are too often delusional.

I imagine Kendall's Tau would be especially nice for studying phenomena like global warming, since these phenomena are governed by such complicated dynamics. You could simply ask questions of increases or decreases, and not worry about whether or not you polluted your conclusions by assuming too much about specific functions.

[NNadir]

8. My contention that the hurricanes and global warming is more empirical,

because I would assume, although I'm unfamiliar with this particular test, of the incidence of major hurricanes, that a reasonable sample size doesn't exist.

I'm certainly not familiar with Kendall's Tau, and so a reference would be appreciated.

However, statistical tests aside, the hurricanes are not taking place in a vacuum. For over 140 years scientists, including the great chemist Arrhenius at the turn of the 19th century, have been predicting what is now happening. An excellent test, albeit not a strictly statistical one, of a theory is the accuracy of it's predictions. Moreover, everybody knows the absorbance spectrum of carbon dioxide, nitrogen dioxide, carbon tetrafluoride, etc, etc...

A mere catalog of the data is disturbing.

http://www.wmo.ch/web/wcp/wcdmp/home.html

I would imagine that a statistical test involving "pooling" of data (perhaps say data on the intensity of droughts, floods, extreme temperature data etc) would set some kind of pretty convincing number on the probability that the entire set of observations were likely to have resulted by chance fluctuations.

One should consider though, even if one were to attach a probability of 0.50 that these fluctuations are related to chance fluctuations in say, the solar flux (a favorite dodge for Bush apologists), this is still an enormous risk when one considers the stakes involved. We must act and act quickly.

[Donkeyboy75]

9. Don't get me wrong, I'm not discrediting global warming,

and mounds of data are on our side showing that an effect is indeed there. I just like to be extra careful when attributing anything to global warming...one false step hurts the cause. And since plenty of data shows an effect, there is no need to draw tenuous conclusions.

"I would imagine that a statistical test involving "pooling" of data (perhaps say data on the intensity of droughts, floods, extreme temperature data etc) would set some kind of pretty convincing number on the probability that the entire set of observations were likely to have resulted by chance fluctuations."

I wholeheartedly agree, as I alluded to in post 4.

[phantom power]

10. I can't find a good ref on the web, so here's some code I wrote

data1 and data2 are two vectors of numbers, that you wish to
check for correlation.  The return value is a triple: {tau, z,
prob}, where tau is the "raw" tau value, z is the
value mapped into "error-function" space, and prob
is the confidence value.  I forget whether prob is probability
of null hypothesis (no correlation), or if it's (1-prob).

KendallsTau = Compile[{{data1, _Real, 1}, {data2, _Real, 1}},
Block [{concordant, discordant, extra1, extra2, n,
    tau, tauVar, z, tauProb, t1, t2, t3, t4, d1, d2, pd, j,
k},

  n = Length[data1];

  If [n < 1, Return[{0., 0., 1.}];];

  concordant = discordant = 0;
  extra1 = extra2 = 0;
  For [j = 1, j < n, ++j,
    For [k = j + 1, k <= n, ++k,
        d1 = data1[[j]] - data1[[k]];
        d2 = data2[[j]] - data2[[k]];
        pd = d1 d2;
        Which [
          pd > 0, ++concordant,
          pd < 0, ++discordant,
          d1 != 0, ++extra1,
          d2 != 0, ++extra2
          ];
        ];
    ];
  t1 = concordant + discordant + extra1 // N;
  t2 = concordant + discordant + extra2 // N;
  t3 = N[concordant - discordant];
  t4 = Sqrt[t1] Sqrt[t2];

  tau = Which [
    (t4 <= 0.),   0.,
    True,         t3 / t4
  ];

  If [n >= 2,
    tauVar = (4.n + 10.)/(9.n(n - 1));
    z = tau/Sqrt[tauVar];
    tauProb = 1. - Erf[Abs[z]/N[Sqrt[2]]];

    , (* Otherwise, variance is effectively infinite: *)

    z = 0.;
    tauProb = 1.;
  ];

  {tau, z, tauProb}
]]