Statistics

Female-Named Hurricanes Deadlier Than Males. Implicit Sexism Kills!

This is Wilma. And boy is she unhappy.

This is Wilma. And boy is she unhappy.

“Say, Elbert. What’s the name of that there hurricane that’s fixin’ to pounce upon us?”

“Lolita, I think.”

“What? Lolita!? I ain’t evacuatin’ nowheres! Not for no fee-male ‘cane. Gimme that there beer and turn on the TV.”

Poor Earl and Elbert! Blown away by “implicit sexism.”

If only Earl read Discover or the Washington Post, which tells us of a “new groundbreaking study.” Groundbreaking!

It’s the PEER-REVIEWED paper “Female hurricanes are deadlier than male hurricanes” in the Proceedings of the National Academy of Sciences—the National Academy of Sciences!—by Kiju Jung, Sharon Shavitt, Madhu Viswanathan, and Joseph Hilbe. They say, “Laboratory experiments indicate [female-named hurricanes are deadlier] because hurricane names lead to gender-based expectations about severity and this, in turn, guides respondents’ preparedness to take protective action.”

Look out! “This finding indicates an unfortunate and unintended consequence of the gendered naming of hurricanes, with important implications for policymakers, media practitioners, and the general public concerning hurricane communication and preparedness.”

Everybody remembers how the media downplayed female-named Hurricane Sandy, right?

Listen. This paper has no near rival in sheer awfulness, as evidence by its opening sentence: “Estimates suggest that hurricanes kill more than 200 people in the United States annually, and severe hurricanes can cause fatalities in the thousands.”

If they checked the National Weather Services’s numbers, they’d’ve learned that in 2006 there were 0 fatalities, in 2007 just 1, in 2008 only 12, in 2009 a mere 2, in 2010 another 0, in 2011 a meager 9, in 2012 a paltry 3, in 2013 a whole 1. Update These are USA land deaths. See the comment from me to Dan Jones below.

Since 1940, only 4 years out of 74 had hurricanes which killed more than 200, the number they claim is the average. The 10-year average is 108 yearly deaths but the 30-year average is 47. Hurricane Katrina killed 1,106 in 2005 and bumped the 10-year average.

The authors used hurricanes from 1950 to 2012. Who remembers that up until 1979 hurricanes only had female names? Skip it. This is science, not history. They had “raters” “rate” the degree of femininity of hurricane names, from 1 to 11. Hurricane Flossy (1956) got a score of 7, but 1971’s Ginger beat her with a 10. Numbers are what make it science!

Here come the stats: “A series of negative binomial regression analyses were performed to investigate effects of perceived masculinity-femininity of hurricane names, minimum pressure, normalized damage, and the interactions among them on the number of deaths caused by the hurricanes”.

Guess what? Right! The regressions spat out wee p-values! Negative binomial regression! Your average bad paper relies on everyday ordinary regression. But this is negative binomial. Hoo Ah! does that sound impressive.

Wee p-values prove sexism kills, sisters and brothers. Statistics don’t lie. So what if hurricane names are assigned in advance before the season begins and before anybody has any idea of what may come? And so what if nobody in all of history can be found poo-pooing a hurricane because it had a girl’s name? Implicit sexism kills in the same way that splicing in a single frame of a photo of popcorn into a movie convinces people to buy it subliminally. I despair.

The authors must have conducted actual interviews with real people who recalled thinking about real hurricane names, and how they acted on the femininity of those names, right?

Wrong. They ran six “experiments” to generate fictional data instead.

First experiment asked 346 non-house owners (college kiddies) to ponder boy and girl hurricane names and predict the “intensity” of these from 1 to 7. “Arthur” had a mean fictional made-up imaginary pretend fantasy nothing-to-do-with-real-hurricanes “intensity” of 4.246—not 4.245, nor even 4.247, but 4.246—while Dolly had 4.014. Another wee p-value. What more proof do you need, you misogynist.

Another experiment scrapped 142 volunteers from the Internet and asked them to rate their “evacuation intentions” and fake hurricane “risk”, from 1 to 7 of course. Hurricane Christina had a mean 2.343 fictional made-up you-get-the-idea evacuation score, while Christopher had 2.939. More wee p-values, you sexist.

There were other experiments, but all were equally asinine and had zero bearing on any real-life decisions people make with real storms. The authors’ conclusion that the greater deaths seen under female-named hurricanes is “because feminine- vs. masculine-named hurricanes are perceived as less risky and thus motivate less preparedness” is smellier than that which is ejected out of a cow on a forced diet of wet crabgrass.

Listen sisters and brothers, there is no point being nice about this. This paper is dumb. The idea is dumb. The experiments are dumb. The analysis is dumb. The statistical errors are dumb. The media which reported it lovingly was dumb.

The theory is so preposterous that only an academic could believe it. That it even saw the light of day is a measure of how politicized Science has become, how willing the so-called intelligentsia is to accept any evidence, no matter how farcical, as long as it bolsters their prejudices.

Update Shavitt is professor of marketing at the University of Illinois, Kiju Jung is a PhD student in advertising same place, Madhu Viswanathan is a business prof there, too, and Joseph Hilbe is the statistician at Arizona State. The paper prearranged to have Susan T. Fiske, psychology prof at Princeton specializing in stereotyping, prejudice, and discrimination, as editor.

Update But wait—more research is needed! We still need to discover whether Femnadoes are more destructive than Sharknadoes!

Update Another take at AW’s place. The best comment, “So if we start calling them ‘Butch’ and ‘Vlarg, the Destroyer of Worlds’, it’ll save lives?”

Update There is nothing wrong with using the negative binomial. I was being sarcastic. No model at all should have been run. The picture alone proved the theory was folly (James below made one, though I think he mistakenly classified some early hurricanes as “male”).

Update I’ve run into some folks who say the theory is plausible and that the study, such as it is, might suffer from “low power.” I say the theory is nonsensical because nobody ever thought about a hurricane’s name before deciding whether to run away from it. The authors present ZERO direct evidence anybody ever did. Do we not remember the media frenzy over (female) Sandy? “Run away or die tonight!” However, and quite technically, the theory is contingent, meaning it has a vanishingly small probability of being true. Just as do the theories, “Hurricanes starting with the letter P kill more than the letter R” or “Hurricanes with three syllables kill more than with one or two.” These have low power, too. Just as will any of the infinite theories you can think up which would explain the data.

——————————————————–

Thanks to Ant O’Fearghail ‏(@aofarre) who provided the name Femnado. And thanks for Al Perrella for finding the Discover connection.

Categories: Statistics

44 replies »

  1. One needs only to do what a data analyst should, which is plot the raw data before doing anything else to examine it, to notice that the historical conventions are dominant. See this helpful image of the data to notice just how dominant the time and cutoff effect is. And we don’t even need a model!

    Using their same raw data, an OLS fit returns very un-wee p-values for their gender controls (and very wee p-values for storm strength measures). I suspect that they switched to a negative binomial because of that very reason. It took me 5 minutes to learn all of the above.

    Also, a negative binomial simulates “the number of successes in a sequence of independent and identically distributed Bernoulli trials before a specified (non-random) number of failures (denoted r) occurs”. I have no idea how that even represents how a hurricane kills people.

  2. There are two further problems.

    1. The authors labelled 3 out of 38 pre-1979 hurricane names as “male” even though they were exclusively female.

    2. The authors excluded Audrey and Katrina, the two deadliest hurricanes. Although this does reduce the mean impact of female hurricanes, it reduces the standard deviation by more, thus inflating the significance of the difference between males and females.

  3. I’m confused. I’ve read criticisms of this paper, but the numbers you cite here seem awfully low – but as I’m no expert on weather stuff they could well be true. However, you say that in 2008, for example, only 12 people were killed by hurricanes. Yet just one hurricane from 2008, Ike, killed 112 people (with another 23 still missing), according to Wiki (http://en.wikipedia.org/wiki/Hurricane_Ike – I know, I know, not an infallible source). Maybe Wiki’s numbers are off, but this is just one hurricane for the 2008 season (I’m not going through every other possible instance). What’s up with this discrepancy?

  4. What about a bias because of restricting the data to the Western North Atlantic? Did their “experiments” test the Western North Pacific and the South China Sea Names (as of 2012) for typhoons such as:

    Damrey
    Haikui
    Kirogi
    Kai-Tak
    Tembin
    Bolaven
    Sanba
    Jelawat
    Ewiniar
    Maliksi
    Gaemi
    Prapiroon
    Maria
    Son-Tinh
    Bopha
    Wukong
    Sonamu
    Shanshan
    Yagi
    Leepi
    Bebinca
    Rumbia
    Soulik
    Cimaron
    Jebi
    Mangkhut
    Utor
    Trami

    I thought not.

  5. The negative binomial is often used in such cases, as opposed to a Poisson distribution, as it explicitly accounts for over dispersion that is commonly found in count data. Whether this was a good choice here or not, we do not know. While I wouldn’t even approach this study with any sense of credibility, the NB choice here is leagues better than an assumption of normality.

    That said, the analysis of their survey data is much more problematic.

  6. Dan Jones,

    Excellent questions. For instance, this paper:

    http://www.aoml.noaa.gov/hrd/hurdat/mwr_pdf/2009.pdf

    says 6 died in 2009. But this summary:

    http://www.nws.noaa.gov/om/hazstats/resources/weather_fatalities.pdf

    says only 2 did.

    Both are government sources. The earlier report counts people who died “well offshore”, and I’m not sure whether the other one does.

    It’s even worse in 2008. The same first source:

    http://www.aoml.noaa.gov/hrd/hurdat/mwr_pdf/2008.pdf

    gives 750 deaths. Our PNAS authors say just 137, while the second source gives 12.

    The discrepancy seems to be how the deaths were counted. The second source (and the one I cited in the paper) are USA-land deaths. The first source linked in this comment are all North Atlantic deaths, which counts, for instance, Hurricane Hanna which killed 500+ in Haiti.

    Say, did our PNAS authors account for the multiple ways of counting? And the uncertainty which comes with it?

    No, sir. They did not.

    Update Let’s don’t forget the main focus: the theory that people act differently from hearing a female name for a hurricane than if they had heard a male name. That people scoff at danger when hearing Katrina and that more people would have been saved had (say) Karl hit New Orleans. Nonsense.

  7. For people who want to see the data that the study used, here is a link to the publicly available supplement to the original study.

    They justified the exclusion of the really big storms due to those being incredible outliers. Katrina was large, and the evacuation was very urgent and strongly suggested. Even the authors admitted that it should probably be excluded from the study.

    It turns out people use negative binomials to fit count data. I’m not a social scientist, so I was unfamiliar with that! You learn something new every day. Anyway, a model is a model is a model, so I went ahead and fit a couple on the data.

    You don’t get wee p-values for gender if you do a categorical representation of the gender of the names. You do get that, though, if you use their rating of femininity. The fun part is, if you use the negative binomial after 1978, the femininity has no impact whatsoever. I got a p-value of .97, which might be large enough to make a frequentist go so far as to accept the null!

  8. For more numbers on it, they recorded the following data:

    Female hurricane deaths: 1473
    Male hurricane deaths: 427

    However, when you only look at deaths from 1979 on, you get:

    Female hurricane deaths: 459
    Male hurricane deaths: 413

    (They took out Katrina to “remove bias”, and they listed 3 hurricanes pre-1979 King, Able, and Iones male names)

    Also, over 1/3 of the post-1979 female storm deaths were in just one hurricane though – Sandy. USing data from 1950-1979 provides a data bias of over 1000 extra deaths in female named storms – over half the deaths listed!

  9. Maybe I’m missing something, but I don’t see how there isn’t a big problem with including storms before 1979 when they killed more and had exclusively female names. It seems the inclusion is designed to produce the desired result.

  10. Dan, the wikipedia article is inaccurate/misleading – if you follow the link to the cited NOAA report (http://www.nhc.noaa.gov/pdf/TCR-AL092008_Ike_3May10.pdf) the 112 number includes deaths in Cuba, Haiti, and the Dominican Republic. But the NOAA report does say that 20 people died in US Gulf Coast states. I don’t know why the NWS numbers don’t line up with the NOAA report – maybe they break out hurricanes and floods differently?

  11. Ha! Thanks for the data link James. This should provide some entertainment in classes 🙂 .

    Your results are more what I would expect. And why was their index the MFI and not the FMI? A bit sexist, eh 😉

    The NB is commonly used in biological count data, particularly entomology. Experience has shown it behaves better, although it, and other non-normal distributions, need to be applied very carefully. Stroup (2014) has a very good article on this. (https://www.agronomy.org/publications/aj/articles/0/0/agronj2013.0342)

  12. The “solution” to this “problem” is really quite simple:

    Aubry
    Blair
    Cody
    Drew
    Emery
    Fran
    Gale
    Hunter
    Jamie
    Kelly
    etc.

  13. First experiment asked 346 non-house owners (college kiddies)

    If I recall correctly these were college kiddies in Illinois. Given the large distance between Illinois and the ocean, it’s quite likely that few of these students had ever experienced hurricanes and have little inherent ‘gut feeling’ about what hurricane levels even mean. I grew up in Illinois, went to school in Cham-bana, and my experience with hurricanes consists of ‘watching the news’. Tornadoes? Seen those; have spent time in basements and tornado shelter. Earthquakes? Experienced those when I lived in El Salvador (which we left when I was 6 yo.) But hurricanes? I have no freakin’ idea.

    On the other hand: if you ask someone to guess the intensity of a hurricane based on nothing other than a name, maybe it’s true that people will guess Hurricane “Snidely Whiplash” is fiercer than Hurricane “Nell”. Heck, if we gave them dog names, I’d probably guess Hurricane “Fang” is deadlier than hurricane “Fluffy”. Hurricane “Gandhi” vs “Vlad the Impaler”? The latter would be worse. So, it’s quite possible soft sounding names would be assessed as being less deadly if name is the only thing you have to go on.

    Fortunately, when an actual hurricane forms, the weather service also starts reporting things like windspeed, landfall and other factors.

  14. An additional historical factor not considered: FEMA came into existence in 1979.

  15. Briggs,

    The classification came from their raw data, I did not classify it.

    In response to your update, the raw data that they supplied had a name “King” in 1950 and “Able” in 1952. Those they classified as male. In 1955 the name “Ione” received a femininity score of 5.944444 (so many 4’s), but was placed (by them) into the gender category of male.

  16. Briggs and Kyle, thanks for the replies – and yes, I realise that it’s easy to get lost in details while forgetting the broader, sensational(ist) claims.

  17. Throwing in to say that the negative binomial is a perfectly reasonable distribution to fit to this data — number of deaths is a count, and count data should most typically be modeled as a function of a Poisson or a negative binomial distribution.

    I’d agree that the history is probably the chief effect, though. That raw data plot is pretty clear-cut.

  18. I wonder what effect popular culture might have on perception of the female hurricane names of Buffy (Vampire Slayer), Xena (Warrior Princess), Hermione (Granger), Veronica (Mars) — all tough characters.

  19. What about storms named Chris, Tracy or Kelly. Names that could be either masculine or feminine. I suspect people would just get their sexism confused and completely seize up thus increasing the death toll even more. Wait, I wrote the word masculine before feminine above. Dang implicit sexism strikes again.

  20. Well if you examine the data closely, global warming has caused deaths from male named hurricanes to sky rocket!

  21. M. Jones,

    despite the variation due to counting land deaths, wet deaths and non-US deaths the greater variation is likely due to what is considered a death caused by a hurricane. It has been a few years since I read it, but if I recall correctly, there was a report breaking down of deaths attributed to Ike (using a number that would fit the 112) and the largest number of deaths were due to Carbon Monoxide poisoning. If we limit ourselves to deaths caused directly by a hurricane we will have nice low numbers, if we expand to deaths related to hurricanes (person breaks nose in car accident while fleeing from hurricane 2 days before hurricane makes landfall and dies 3 years later due a reaction to anesthesia while having a broken nose reset is technically a hurricane related death) then the numbers are much larger.

    might be the one I read. notice what is considered a hurricane related death, and how the numbers are compiled. It seems to me that reported hurricane related deaths might easily correlate directly with how seriously a hurricane warning is taken, the exact opposite conclusion given by the paper (people who have heart attacks in evacuation centers are counted, people whos’ bodies are found 4 months later are not).

  22. I am sure they excluded divorced men who have lost their home and everything they owned (from the divorce, not a hurricane) from the study. If they are searching for their preconceived name related sexism, they should have gone both ways in their preconceptions.

  23. As I was looking at their spreadsheet in the SI, I realized that Sandy in 2012 was in a “male” name slot, so the powers that be who determine the names consider it a man’s name. As Brad pointed out, Sandy accounts for 1/3 of the fatalities in the “non-sexist” post-1979 era.

    For those of us with gray hair, Sandy is at least as likely to be a man’s name (remember Sandy Koufax, anyone?). But they only surveyed college students, who overwhelmingly considered a feminine name. Another confounding factor?

    Yes, it’s a detail, compared to the ridiculousness of asking people how severe a hurricane would be just by its name. But still, it points to another facet of the ridiculousness of the study.

  24. More bad science, plenty of it about in all disciplines and this case demonstrates once again that peer review followed by publication is only the start of the process. What counts is post publication peer review, which in this case is damning. Waste of paper, as PNAS is still printed.

  25. The bias lies with the authors naming these hurricanes — which were probably males. 😀

  26. It looks like we are going to have redo our naming schemes:
    Hurricane: Atlantic 2014 #1
    Same for blizzards: Montana 2014 #1

    That way there’s no sexism. (Now, we just have the problem with #1 sounding more important than #5……..)

  27. There’s more devastation in months that contain the letter “r”. Or vice-versa. Or something like that. Or anything else if you wish.

  28. Soooo….how many did Hurricane Little Lord Fauntelroy kill? Or Hurricane Maleficent? What about Hurricane Bubbles and Hurricane Rosa Klebb? Hurricane Thor and Hurricane Amaterasu? Hurricanes THX-1138, Proteus 4, WOTAN, AL-76 and HAL9000? Why, they’ve hardly scratched the surface of what effects different naming conventions might have! Waaay more research is needed (and the attendant funding, of course.)

  29. They did check the fatalities were all mysogynistic males, right? I bet those guys were rushing to throw themselves at “Hurricane Kate Beckinsale in a leather catsuit”, right?

  30. Richard Tol, it looks like the early 1950’s used the Able Baker Charlie alphabet every year. 1953 was the start of female names. I think they picked up a false-positive male name in 1955, “Ione”.

  31. Had they only read Kipling, they would not have had to do any studying…

    “The Female of the Species” (…is more deadly than the male.)

    As with “The Gods of the Copybook Headings”, and many others, he was very prescient.

  32. Two questions:

    1. Was hurricane Katrina excluded from the beginning due to the mentioned criteria? And was it really because the hurricane was an outlier in these characteristics, or because it would immediately raise questions how people underestimating a hurricane due to a female name can, e.g., cause levees to fail due to neglectful maintenance?

    (Although, to be frank, at this point, it would not surprise me to see in a discussion — academic or otherwise — a few people arguing that levees failure was because the ones responsible were ‘deadbeat’ male workers leaving the female canals after they had “finished”.)

    2. Did they state in their study that they knew about the naming convention pre-1979? I hope so, not understanding the data is bad enough, but knowing and not mentioned it is … something else.

  33. I really think the most important point is completely being missed. You can’t just throw away outliers because they seem large relative to your completely empirical estimate of the distribution.
    In fact, what Katrina and other large storms tell us, is that the distribution for deaths from hurricanes is reasonably fat tailed. So using negative binomial, or Poisson, as models, are both completely unreasonable as they are thin tailed.
    In fat tailed distributions, fluctuations are much more significant, so the sample mean will vary much more. Hence the results are not statistically significant.

  34. Nir Friedman,

    In proper statistics, we massage data (sometimes by throwing out outliers) to fit the perfect form of data which, of course, is the normal distribution. The normal distribution exists in nature as a cause of observations, everyone knows that, and any data that doesn’t fit it is just a red herring.

    Sarcasm, of course.

  35. Some other issues:

    1. Sources/Relevance to policy makers:

    Regarding the “no near rival in sheer awfulness, as evidence by its opening sentence”, sometimes I wish the source would be connected to the text like a comment box, easy to see:

    > “Estimates suggest that hurricanes kill more than 200 people in the United States annually, and severe hurricanes can cause fatalities in the thousands (1).”

    (1.) Tracey M (2006) She Was No Lady: A Personal Journey of Recovery Through Hurricane Katrina (iUniverse, Lincoln, NE).

    Uh … whut?

    2. Model and Reality:

    > “In other words, our model suggests that changing a severe hurricane’s name from Charley (MFI = 2.889, 14.87 deaths) to Eloise (MFI = 8.944, 41.45 deaths) could nearly triple its death toll.”

    It’s meant differently, but … whut?

    3. Discrimination

    If one would buy the study, it’s nice to know that women show negative sexual discrimination to female names too. Usually, at least half the participants were female and there were no gender effects.

    4. Similarity to control group

    According to the paper, hurricanes with female names lead to the same estimates regarding danger as unnamed hurricanes. Doesn’t this mean it’s not that women are taken lightly, but rather that a hurricane associated with a male name is taken as more dangerous? It’s the fear of men/maleness that makes the difference from the normal level of dangerousness (if you buy the paper). So if it were implicit sexism, wouldn’t it be implicit sexism to see male (names) as more dangerous, rather than female (names) more lightly?

    5. Participants and association of hurricanes with female names

    > “The similarity in the perceived riskiness of the female-named and unnamed hurricanes may reflect the influence of the historical female-only naming convention. Even in the absence of an assigned name, storms may be more associated with female than male names and, therefore, with milder qualities.”

    In some of the studies, they were students not older than 25, in others Mechanical Turk participants up to 80 or so years. Are the students (and likely part of the MT participants, the range does not tell you much) really used to hurricane names being female? They weren’t even born when the name change happened. Somehow I don’t find the explanation convincing — like 4. I think if at all it would point to a bias associating men with danger, not women with less danger. Wonder whether the participants would refer to the unnamed hurricane with “it”, “she”, or “he”. Personally, I’d go with it if it does not have a name. But that’s an empirical question.

    6. Even split

    > “Although our findings do not definitively establish the processes involved, the phenomenon we identified could be viewed as a hazardous form of implicit sexism. Indeed, in an additional dataset, when asked explicitly whether a male-named or female-named hurricane would be riskier and more dangerous, responses were evenly split between female- and male-named hurricanes (Materials and Methods).”

    I’m missing something here. Evenly split responses are a hazardous form of implicit sexism?

    There are other issues, but … I might be wrong (and please, if I have misunderstood something, correct me), it seems to me that the conclusion supported the study and not the other way around. And I get the impression that the authors — like perhaps many other scientists — are in a severe need of funding. Is it the style of that journal that articles sound a bit like grant proposals?

  36. A moron in a white coat masquerading for a scientist : “So Mr Smith you relax now. I will give you a name and you will tell me the intensity of a hurricane that would have this name. Ready ?”

    Mr Smith (frantically checking if there is an exit in the room) : “Errr … No velocities ? No energy ? Just a name ?”

    A moron (smiling) : “Right Mr Smith. Just a name. This is groundbreaking science.”

    Mr Smith (slowly backing and stottering) : “But, but, but … I have no clue what the intensity would be. Nobody has. None whatsoever !”

    A moron (the smile is widening) : “Just guess Mr Smith. On a scale from 1 to 7. Be very accurate, your health depends on it. Groundbreaking science, remember ?”

    Mr Smith (secretly dialling 911 in his pocket) : “Whatever you want. Just please don’t hurt me.”

    A moron (positively glowing) : “No need for that Mr Smith if you are reasonable. And you are a reasonable man, aren’t you ? Here we go . SANDY !”

  37. I propose naming next hurricane Monica Lewinsky and count how many people get blown away by her. This might settle the issue.

    Alternatively, we can call it Barack Obama and check if disrespect kills more Republicans, either way this theory requires a properly designed experiment.

Leave a Reply

Your email address will not be published. Required fields are marked *