Sexism and Hurricanes; or, the Dangers of Bad Science
June 2, 2014 6:00 PM Subscribe
Hurricanes with feminine names kill more people because people are sexist. Or it could be bad (social) science.
Well, given that they didn't use men's names at all until 1979, and names are reused until a given storm has a significant impact on the US, I'd call it "selection bias."
Note that "Katrina" was used for two North Atlantic storms, a South Atlantic storm, and three Pacific storms *before* the 2005 storm that ravaged New Orleans.
Outlier? Nope. Katrina was named seven times before it was retired. Give enough chances, any name will be a bingo.
Andrew only happened twice -- a tropical storm in 1986 and the devastating Cat 5 that hit Florida in 1992.
Amazing what happens when you're not even in the game until 1979.
posted by eriko at 6:10 PM on June 2, 2014 [2 favorites]
Note that "Katrina" was used for two North Atlantic storms, a South Atlantic storm, and three Pacific storms *before* the 2005 storm that ravaged New Orleans.
Outlier? Nope. Katrina was named seven times before it was retired. Give enough chances, any name will be a bingo.
Andrew only happened twice -- a tropical storm in 1986 and the devastating Cat 5 that hit Florida in 1992.
Amazing what happens when you're not even in the game until 1979.
posted by eriko at 6:10 PM on June 2, 2014 [2 favorites]
I find it hard to believe PNAS would let through such a dumb statistical error, but maybe.
To encourage evacuation, we should give hurricanes really scary names like Murderdeath, Axkiller, Heartslicer, or Drownquick.
posted by miyabo at 6:15 PM on June 2, 2014 [24 favorites]
To encourage evacuation, we should give hurricanes really scary names like Murderdeath, Axkiller, Heartslicer, or Drownquick.
posted by miyabo at 6:15 PM on June 2, 2014 [24 favorites]
Drownquick is the greatest metal band of all time.
posted by NoMich at 6:16 PM on June 2, 2014 [11 favorites]
posted by NoMich at 6:16 PM on June 2, 2014 [11 favorites]
To encourage evacuation, we should give hurricanes really scary names like Murderdeath, Axkiller, Heartslicer, or Drownquick.
Socialism?
posted by elephantday at 6:16 PM on June 2, 2014 [5 favorites]
Socialism?
posted by elephantday at 6:16 PM on June 2, 2014 [5 favorites]
Even though these results are questionable at best, I think the 21st century thing to do would be to just name the storms <year> <letter> and be done with it.
posted by ob1quixote at 6:17 PM on June 2, 2014 [2 favorites]
posted by ob1quixote at 6:17 PM on June 2, 2014 [2 favorites]
Yeah, contrasting this to the reddit thread is an interesting exercise in why-I-hate-the-internet:
Interesting to note that Katrina, certainly the most infamous in recent history, was not especially deadly by historical standards, but it was extremely expensive and, perhaps most importantly, had a large impact on American business interests. Mitch, on the other hand, was less expensive and more deadly, but primarily affected the Yucatan. I think that explains a lot about hurricane coverage in the media.
posted by deathpanels at 6:19 PM on June 2, 2014 [4 favorites]
Because people take female identifying hurricanes less seriously. #YesAllWomenOh, please kill me now.
Interesting to note that Katrina, certainly the most infamous in recent history, was not especially deadly by historical standards, but it was extremely expensive and, perhaps most importantly, had a large impact on American business interests. Mitch, on the other hand, was less expensive and more deadly, but primarily affected the Yucatan. I think that explains a lot about hurricane coverage in the media.
posted by deathpanels at 6:19 PM on June 2, 2014 [4 favorites]
Socialism?
I agree, we should name them after delicious frozen treats.
posted by Ice Cream Socialist at 6:20 PM on June 2, 2014 [11 favorites]
I agree, we should name them after delicious frozen treats.
posted by Ice Cream Socialist at 6:20 PM on June 2, 2014 [11 favorites]
The marketing professor that did the study responds to Ed Yong here (PDF warning).
posted by codswallop at 6:21 PM on June 2, 2014 [1 favorite]
posted by codswallop at 6:21 PM on June 2, 2014 [1 favorite]
What about if the hurricane's name is an indeterminate gender like, oh, "Sandy"?
posted by EmpressCallipygos at 6:26 PM on June 2, 2014
posted by EmpressCallipygos at 6:26 PM on June 2, 2014
miyabo: “To encourage evacuation, we should give hurricanes really scary names like Murderdeath, Axkiller, Heartslicer, or Drownquick.”“Weather Bureau” [Caution: TV-MA]—Metalocalypse
posted by ob1quixote at 6:27 PM on June 2, 2014 [6 favorites]
I think it was an interesting preliminary study -- yes, it would be nice to see the stats from just the past 30 years when hurricanes weren't all female, but the implications are clear. (I am not sure where they get the "femininity" of names, though; to me Fern is at least as feminine as Camille.)
The other part of the study, where they asked people what they would do for upcoming hurricanes by name, is also interesting in a preliminary study sort of way. I think, at least; I don't have access to the actual paper.
posted by jeather at 6:27 PM on June 2, 2014
The other part of the study, where they asked people what they would do for upcoming hurricanes by name, is also interesting in a preliminary study sort of way. I think, at least; I don't have access to the actual paper.
posted by jeather at 6:27 PM on June 2, 2014
Where they are getting the gender from: First, Jung’s team asked nine people to rate the name of US hurricanes on a scale of 1 (very masculine) to 11 (very feminine). (from the second link in the post)
posted by the agents of KAOS at 6:30 PM on June 2, 2014
posted by the agents of KAOS at 6:30 PM on June 2, 2014
Oh, I missed that part. Nine people?
posted by jeather at 6:32 PM on June 2, 2014 [1 favorite]
posted by jeather at 6:32 PM on June 2, 2014 [1 favorite]
Lazo is right to point out that the students of the Unversity of Illinois may not represent the general public as a whole (I imagine that "our test subjects were all undergraduate students" is a not-uncommon flaw in social science studies).
That said, I think it's overkill to suggest that the study would only have been valid if it had restricted itself to investigating attitudes in hurricane-prone areas. It would be an interesting result even if it was a general attitude.
posted by evidenceofabsence at 6:34 PM on June 2, 2014
That said, I think it's overkill to suggest that the study would only have been valid if it had restricted itself to investigating attitudes in hurricane-prone areas. It would be an interesting result even if it was a general attitude.
posted by evidenceofabsence at 6:34 PM on June 2, 2014
Their "hunches" are particularly weird to me because the tradition behind giving storms female names is rooted in another stereotype of women's character -- storms are capricious, irrational, dangerous in vengeance, etc.
posted by desuetude at 6:39 PM on June 2, 2014
posted by desuetude at 6:39 PM on June 2, 2014
I think it's sexist in itself that only female names is were used for 30 years.
posted by bleep at 6:43 PM on June 2, 2014 [1 favorite]
posted by bleep at 6:43 PM on June 2, 2014 [1 favorite]
(compare two female names: Fern, which is less feminine to Camille, a rather
feminine name).
Er sorry, what? That parenthetical is from the authors' response and I have to say I feel less confident in their original paper thanks to that response. Fern is a 'plant and flower' name and thus *very* traditionally feminine and Camille is apparently gender ambiguous.
My immediate thought is that that particular result from the study might be due to WEIRD raring its head, though you'd like to think that the authors controlled for that--it's not clear to me whether they used the college students or the Mechanical Turk people for the name gendering portion of the study so I can't say for sure that it's WEIRD, and for whatever reason we don't have PNAS pre-embargo so I can't check the study itself to see the way I normally would.
WEIRD is particularly a problem in a super diverse city like Houston, which is emphatically not overwhelmingly composed of rich nor Western nor educated people, and yet Houston is a major US city particularly likely to get hit by a hurricane and thus behaviors of people in that city are rather more relevant than Illinois college students. Again, I'm hoping that the authors have already controlled for that (I confess to being a little worried that they are marketing people and not science people, though).
posted by librarylis at 6:56 PM on June 2, 2014
feminine name).
Er sorry, what? That parenthetical is from the authors' response and I have to say I feel less confident in their original paper thanks to that response. Fern is a 'plant and flower' name and thus *very* traditionally feminine and Camille is apparently gender ambiguous.
My immediate thought is that that particular result from the study might be due to WEIRD raring its head, though you'd like to think that the authors controlled for that--it's not clear to me whether they used the college students or the Mechanical Turk people for the name gendering portion of the study so I can't say for sure that it's WEIRD, and for whatever reason we don't have PNAS pre-embargo so I can't check the study itself to see the way I normally would.
WEIRD is particularly a problem in a super diverse city like Houston, which is emphatically not overwhelmingly composed of rich nor Western nor educated people, and yet Houston is a major US city particularly likely to get hit by a hurricane and thus behaviors of people in that city are rather more relevant than Illinois college students. Again, I'm hoping that the authors have already controlled for that (I confess to being a little worried that they are marketing people and not science people, though).
posted by librarylis at 6:56 PM on June 2, 2014
What I love about Metafilter is that this didn't get posted and accumulate several hundred comments before somebody pointed out the simple business about how nearly 30 years worth of the data came before they ever used male names.
And what I really love about Metafilter is that whenever a social science article is discussed, someone invariably starts off the thread with a comment about how these stupid researchers who spent years investigating this question did not think of this obvious explanation for their findings.
In other words, you didn't read the article and this 'simple business' was not only addressed but incorporated into how they constructed their statistical analysis.
posted by MisantropicPainforest at 7:09 PM on June 2, 2014 [2 favorites]
And what I really love about Metafilter is that whenever a social science article is discussed, someone invariably starts off the thread with a comment about how these stupid researchers who spent years investigating this question did not think of this obvious explanation for their findings.
In other words, you didn't read the article and this 'simple business' was not only addressed but incorporated into how they constructed their statistical analysis.
posted by MisantropicPainforest at 7:09 PM on June 2, 2014 [2 favorites]
OK, I did download and read the paper this afternoon. The sum total of "not-interviewing-undergraduates" evidence is binning hurricanes by femininity of name (as judged by *nine* undergraduates), and noting that for weak hurricanes the death rate is roughly flat (sinking with increasing femininity, actually, but probably not significant), and largely increasing for "strong" hurricanes. No attempt to correlate that by pressure at landfall, or minimum pressure, or accumulated cyclone energy...no attempt I could discern to show that strength was evenly distributed across the femininity variable...and of course including the 1950-1979 all female name era, with poorer forecasts, poorer monitoring and worse communication to public...hell I'm just a physicist but I'm surprised the hurricane folks aren't ripping this a bigger new one than they are.
posted by stevis23 at 7:22 PM on June 2, 2014 [5 favorites]
posted by stevis23 at 7:22 PM on June 2, 2014 [5 favorites]
In addition to there being nearly three decades worth of all female named storms, hurricane tracking technology has gotten way, way more accurate since then. The first weather satellite was launched in 1960, the GOES (Geostationary Operational Environmental Satellite) system (which we still use today!) started in 1975. Dynamic statistical models for storm forecasting wasn't introduced until the late 70s. On top of that, weather warnings have been more easily spread since the inception of things like PBS's A.M. Weather (1978) and The Weather Channel (1982). So, yeah, of course female named storms look deadlier. There are more of them, and there were more of them before accurate forecasts and the modern weather media.
I have HUGE BEEFS (beeves?) with the way the weather channel does a lot of things, but I can't begrudge them for their hurricane season coverage. It keeps a lot of people informed.
posted by troika at 7:23 PM on June 2, 2014 [1 favorite]
I have HUGE BEEFS (beeves?) with the way the weather channel does a lot of things, but I can't begrudge them for their hurricane season coverage. It keeps a lot of people informed.
posted by troika at 7:23 PM on June 2, 2014 [1 favorite]
In other words, you didn't read the article and this 'simple business' was not only addressed but incorporated into how they constructed their statistical analysis.
Yes, and if it's true that their reported results weigh in at p=0.073, they're spectacular liars depending on the innumeracy of reporters to carry the story.
posted by CBrachyrhynchos at 7:42 PM on June 2, 2014
Yes, and if it's true that their reported results weigh in at p=0.073, they're spectacular liars depending on the innumeracy of reporters to carry the story.
posted by CBrachyrhynchos at 7:42 PM on June 2, 2014
It's spectacular bullshit.
1. Pose the hypothesis.
2. Split your data to account for a historical change.
3. Crunch your numbers.
4. Fail to reject the null.
5. Report your failure to reject the null as a positive result, and blame the lack of power on #2.
posted by CBrachyrhynchos at 7:45 PM on June 2, 2014 [1 favorite]
1. Pose the hypothesis.
2. Split your data to account for a historical change.
3. Crunch your numbers.
4. Fail to reject the null.
5. Report your failure to reject the null as a positive result, and blame the lack of power on #2.
posted by CBrachyrhynchos at 7:45 PM on June 2, 2014 [1 favorite]
Yes, and if it's true that their reported results weigh in at p=0.073, they're spectacular liars depending on the innumeracy of reporters to carry the story.
That is the result of only one of the tests they did, which as they say, getting a low p value with such a small n is going to have a very high standard. Their other experiments, which have much more data, have very low p values. For example, one result of a two way ANOVA they did had a p value of less than 0.001
posted by MisantropicPainforest at 7:45 PM on June 2, 2014
That is the result of only one of the tests they did, which as they say, getting a low p value with such a small n is going to have a very high standard. Their other experiments, which have much more data, have very low p values. For example, one result of a two way ANOVA they did had a p value of less than 0.001
posted by MisantropicPainforest at 7:45 PM on June 2, 2014
Their other p values are, 0.0001, 0.029, 0.012, 0.037, 0.024, 0.001
posted by MisantropicPainforest at 7:49 PM on June 2, 2014
posted by MisantropicPainforest at 7:49 PM on June 2, 2014
peer review fail
posted by Jacqueline at 8:05 PM on June 2, 2014
posted by Jacqueline at 8:05 PM on June 2, 2014
I always thought cyclones and hurricanes could have much better names.
Hurricane Fuckplank
Cyclone Legsakimbo
Tropical Storm Shitbuckets
posted by turbid dahlia at 8:50 PM on June 2, 2014 [3 favorites]
Hurricane Fuckplank
Cyclone Legsakimbo
Tropical Storm Shitbuckets
posted by turbid dahlia at 8:50 PM on June 2, 2014 [3 favorites]
I can't access the paper, but this has all the marking of a bogus result. Tons of popular hype, and as far as I can tell no external validation of the effect size -- is it really plausible to think that you could cut hurricane related deaths by 65 percent just by changing a name from Eloise to Charley? That seems like a ridiculously large number to pin on a psychological effect that wouldn't even apply to the whole population. It implies that all hurricane related deaths are preventable, it's just that people aren't afraid enough. And it implies that hurricane preparation is mostly a function of how a name makes people feel.
But why bother checking the size of the actual effect? Everyone knows all that matters are the p values. Get p under 0.05, and bam, you have a real scientific result. No need to actually think through the implications.
BTW, the p values in the table in the link just show that people on Mechanical Turk expressed a somewhat stronger hypothetical reaction to male names than to female names. Actually connecting this to real world behavior, and moreover to the actual fatality rates for different hurricanes, is an entirely different matter.
posted by leopard at 9:03 PM on June 2, 2014 [1 favorite]
But why bother checking the size of the actual effect? Everyone knows all that matters are the p values. Get p under 0.05, and bam, you have a real scientific result. No need to actually think through the implications.
BTW, the p values in the table in the link just show that people on Mechanical Turk expressed a somewhat stronger hypothetical reaction to male names than to female names. Actually connecting this to real world behavior, and moreover to the actual fatality rates for different hurricanes, is an entirely different matter.
posted by leopard at 9:03 PM on June 2, 2014 [1 favorite]
I'm doubtful that most midwesterners would have any idea what to do in a hurricane. Let me know when they start naming tornadoes.
posted by desjardins at 9:34 PM on June 2, 2014
posted by desjardins at 9:34 PM on June 2, 2014
They should name hurricanes like Pacific Rim named Jaegers.
posted by Apocryphon at 9:55 PM on June 2, 2014
posted by Apocryphon at 9:55 PM on June 2, 2014
Demon names y'all! These names will make people evacuate IN THEIR PANTS.
Abraxas
Beelzebub
Crocell
Demogorgon
Eisheth
Furcas
Gorgon
Haagenti
Ifrit
Jinn
Krampus
Lucifuge Rofocale
Moloch
Naamah
Ördög
Pazuzu
Rusalka
Sthenno
Tannin
Ukobach
Uvall
Vepar
Wendigo
Xaphan
Yeqon
Ziminiar
posted by a humble nudibranch at 10:07 PM on June 2, 2014 [9 favorites]
Abraxas
Beelzebub
Crocell
Demogorgon
Eisheth
Furcas
Gorgon
Haagenti
Ifrit
Jinn
Krampus
Lucifuge Rofocale
Moloch
Naamah
Ördög
Pazuzu
Rusalka
Sthenno
Tannin
Ukobach
Uvall
Vepar
Wendigo
Xaphan
Yeqon
Ziminiar
posted by a humble nudibranch at 10:07 PM on June 2, 2014 [9 favorites]
One day I'm going to get over the acronym PNAS, but today is not that day.
posted by 2bucksplus at 10:26 PM on June 2, 2014 [7 favorites]
posted by 2bucksplus at 10:26 PM on June 2, 2014 [7 favorites]
Thank you, a humble nudibranch, for naming my next 26 pets.
(Few, if any, will be nudibranches)
posted by evidenceofabsence at 12:29 AM on June 3, 2014 [2 favorites]
(Few, if any, will be nudibranches)
posted by evidenceofabsence at 12:29 AM on June 3, 2014 [2 favorites]
This totally backs up earlier studies showing that among hurricanes named for food, Hurricane Snickerdoodle had far more fatalities than Hurricane Brussels Sprouts.
posted by taz at 2:46 AM on June 3, 2014 [1 favorite]
posted by taz at 2:46 AM on June 3, 2014 [1 favorite]
Demon names y'all! These names will make people evacuate IN THEIR PANTS.
brb planning hurricane-themed D&D campaign
posted by zombieflanders at 4:42 AM on June 3, 2014
brb planning hurricane-themed D&D campaign
posted by zombieflanders at 4:42 AM on June 3, 2014
Regardless of whether or not the science is busted, as one of the people on my Tumblr feed put it, the notion that sexists are eliminated by natural selection because they don't take female-named hurricanes seriously is deeply comforting.
posted by Halloween Jack at 4:43 AM on June 3, 2014 [1 favorite]
posted by Halloween Jack at 4:43 AM on June 3, 2014 [1 favorite]
I think the problem here isn't sexism so much as Amero-centrism. Many of the world's worst (not US worst) hurricanes are not "Hurricanes" at all but "Typhoons" and "Cyclones". But if were going to stick to just the ones called "Hurricanes" then the list would contain Mitch, Hugo, Ike, and Andrew, which, while damage was nothing-to-middlin' in the US, it was horrible in the Carribean and Latin America.
posted by The 10th Regiment of Foot at 4:57 AM on June 3, 2014 [1 favorite]
posted by The 10th Regiment of Foot at 4:57 AM on June 3, 2014 [1 favorite]
It's partly sexism, partly just word association (as female names are generally associated with non-aggressiveness, despite when that goes against reality). As for word association, and going purely gender neutral, rate your intimidation from 1 to 10 when encountering a knife wielding person of indeterminable gender nick-named:
- Bubbles
- Throat-ripper
If the study has any merit, it might be a good idea to explore abandonment of the current naming system as a whole if by doing so more lives could be saved. I'd love to see them take the next step and begin comparing common names with alternative naming conventions (Hurricane 32, Hurricane Theta, etc). Heck, even the demon names idea, while it would be certainly controversial, could be immensely effective.
posted by samsara at 5:22 AM on June 3, 2014
- Bubbles
- Throat-ripper
If the study has any merit, it might be a good idea to explore abandonment of the current naming system as a whole if by doing so more lives could be saved. I'd love to see them take the next step and begin comparing common names with alternative naming conventions (Hurricane 32, Hurricane Theta, etc). Heck, even the demon names idea, while it would be certainly controversial, could be immensely effective.
posted by samsara at 5:22 AM on June 3, 2014
...no attempt I could discern to show that strength was evenly distributed across the femininity variable..
You should have looked at the chart showing the distribution across the masculine/feminine variable. You can find it on page 2 of the pdf.
posted by MisantropicPainforest at 5:55 AM on June 3, 2014
You should have looked at the chart showing the distribution across the masculine/feminine variable. You can find it on page 2 of the pdf.
posted by MisantropicPainforest at 5:55 AM on June 3, 2014
Their other p values are, 0.0001, 0.029, 0.012, 0.037, 0.024, 0.001
Those values are pulled directly from this table which has not a thing to say about whether the named hurricanes are, in fact, deadlier.
I can't find a non-paywall access to the article in question, but their response doesn't inspire confidence in the claim of their article title. They admit their sample isn't significant after the adoption of male names, but then go on to say that there is a difference for "severe" storms using an unexplained benchmark of $1.65 billion damage. Correlating "damage" (which may or may not mean mortality there) with "femininity" strikes me as odd. I've not seen that typically done in research on gender differences because that is a matter of cultural fashion to varying degrees. And if that was the significant correlation found, why not put it in the title?
posted by CBrachyrhynchos at 6:43 AM on June 3, 2014 [1 favorite]
Those values are pulled directly from this table which has not a thing to say about whether the named hurricanes are, in fact, deadlier.
I can't find a non-paywall access to the article in question, but their response doesn't inspire confidence in the claim of their article title. They admit their sample isn't significant after the adoption of male names, but then go on to say that there is a difference for "severe" storms using an unexplained benchmark of $1.65 billion damage. Correlating "damage" (which may or may not mean mortality there) with "femininity" strikes me as odd. I've not seen that typically done in research on gender differences because that is a matter of cultural fashion to varying degrees. And if that was the significant correlation found, why not put it in the title?
posted by CBrachyrhynchos at 6:43 AM on June 3, 2014 [1 favorite]
Those values are pulled directly from this table which has not a thing to say about whether the named hurricanes are, in fact, deadlier.
Which isn't a valid criticism because that is not even close to what they were testing when they generated those p-values.
posted by MisantropicPainforest at 7:01 AM on June 3, 2014
Which isn't a valid criticism because that is not even close to what they were testing when they generated those p-values.
posted by MisantropicPainforest at 7:01 AM on June 3, 2014
Slate's Eric Holthaus has another good debunking.
He also suggests that we not use human names. I don't think we should stop, personally. The study is full of shit, why should we advocate for something based on a flawed premise?
Plus, this year, if there are enough storms, there'll be a Hurricane [my name] and you guys, I've been dreaming of this since I was a kid, tracking Hurricane Andrew on a map from the phone book. It's been a sexennial let down, we've only had that many named storms once since 1950, but I've got a good feeling about 2014.
posted by troika at 7:15 AM on June 3, 2014 [1 favorite]
He also suggests that we not use human names. I don't think we should stop, personally. The study is full of shit, why should we advocate for something based on a flawed premise?
Plus, this year, if there are enough storms, there'll be a Hurricane [my name] and you guys, I've been dreaming of this since I was a kid, tracking Hurricane Andrew on a map from the phone book. It's been a sexennial let down, we've only had that many named storms once since 1950, but I've got a good feeling about 2014.
posted by troika at 7:15 AM on June 3, 2014 [1 favorite]
It's gotten to a point at which I can basically just read the headlines about such studies and, with fair accuracy, mentally throw the conclusions away and move on.
With respect to this story, I saw the headline this morning, immediately thought yeah, no...then couldn't resist (both out of bit of real curiosity and a large dose of the morbid kind) reading some of the details of the study. Ugh. Then I was irked that I hadn't just gone with my initial judgment and noped right out with nary a glance back.
Seriously...WTF is it with this deluge of ridiculous social scientific studies we seem constantly subjected to? It's kinda not funny anymore
posted by Fists O'Fury at 7:16 AM on June 3, 2014 [2 favorites]
With respect to this story, I saw the headline this morning, immediately thought yeah, no...then couldn't resist (both out of bit of real curiosity and a large dose of the morbid kind) reading some of the details of the study. Ugh. Then I was irked that I hadn't just gone with my initial judgment and noped right out with nary a glance back.
Seriously...WTF is it with this deluge of ridiculous social scientific studies we seem constantly subjected to? It's kinda not funny anymore
posted by Fists O'Fury at 7:16 AM on June 3, 2014 [2 favorites]
I'll try to run an actual sensitivity analysis later, but the total number of hurricanes is rather small (under 100) and the number of severe hurricanes is their data set is tiny --
-7 category IV or V hurricanes; 3 male
-8 hurricanes with at least three times the average number of fatalities (21 is the average); 2 male
-10 hurricanes with normalized damage at least 1 standard deviation above average; 4 male
-15 hurricanes in any of the 3 above categories, 5 male
So I'm impressed that they were ambitious enough to not only test the impact of male versus female names but to also do this on a 1-9 scale. That totally seems like the the right level of granularity given the number of meaningful data points that they have. I mean, it's not like hurricanes have highly unpredictable consequences once they make landfall, you basically know exactly what the human impact will be as long as you have a couple of variables measured.
As to the p-values regarding the Mechanical Turk participants, who cares? So on a scale from 1-7 people respond 1/2 point more cautiously to male names and seem to treat female-named hurricanes just the same as unnamed hurricanes. How on earth would this translate to real-life action? How would this lead to dramatic reductions in deaths?
posted by leopard at 7:19 AM on June 3, 2014
-7 category IV or V hurricanes; 3 male
-8 hurricanes with at least three times the average number of fatalities (21 is the average); 2 male
-10 hurricanes with normalized damage at least 1 standard deviation above average; 4 male
-15 hurricanes in any of the 3 above categories, 5 male
So I'm impressed that they were ambitious enough to not only test the impact of male versus female names but to also do this on a 1-9 scale. That totally seems like the the right level of granularity given the number of meaningful data points that they have. I mean, it's not like hurricanes have highly unpredictable consequences once they make landfall, you basically know exactly what the human impact will be as long as you have a couple of variables measured.
As to the p-values regarding the Mechanical Turk participants, who cares? So on a scale from 1-7 people respond 1/2 point more cautiously to male names and seem to treat female-named hurricanes just the same as unnamed hurricanes. How on earth would this translate to real-life action? How would this lead to dramatic reductions in deaths?
posted by leopard at 7:19 AM on June 3, 2014
You should have looked at the chart showing the distribution across the masculine/feminine variable. You can find it on page 2 of the pdf.
I am looking at that chart right now (Fig. 1). I looked at it at my desk yesterday afternoon, and had it well in mind when I made my comment.
It is predicted fatality counts vs. MFI (the 1-11 masculine/femininity rating.) It is not showing that Normalized damage (NDAM) was evenly distributed across MFI. There is no reason to expect that it would be uneven, but there are sufficiently few hurricanes, striking sufficiently small areas near the most dangerous winds and surge, that the potential lives in danger might not be equal between masculine and feminine names in the survey period.
What I want to see but didn't: NDAM vs. MFI. This should look like random noise, right? The atmosphere cares not for what we call a disturbance in it. But at this point I have no idea whether an uneven distribution might be affecting the plot in Fig. 1 or not.
NDAM is defined in a reference I haven't tracked down. By name, it sounds like a reasonable proxy for threat to lives, but I'd like to see that confirmed.
Fortunately, they linked to their data set on this. Give me a few minutes to work on something.
posted by stevis23 at 7:24 AM on June 3, 2014 [1 favorite]
I am looking at that chart right now (Fig. 1). I looked at it at my desk yesterday afternoon, and had it well in mind when I made my comment.
It is predicted fatality counts vs. MFI (the 1-11 masculine/femininity rating.) It is not showing that Normalized damage (NDAM) was evenly distributed across MFI. There is no reason to expect that it would be uneven, but there are sufficiently few hurricanes, striking sufficiently small areas near the most dangerous winds and surge, that the potential lives in danger might not be equal between masculine and feminine names in the survey period.
What I want to see but didn't: NDAM vs. MFI. This should look like random noise, right? The atmosphere cares not for what we call a disturbance in it. But at this point I have no idea whether an uneven distribution might be affecting the plot in Fig. 1 or not.
NDAM is defined in a reference I haven't tracked down. By name, it sounds like a reasonable proxy for threat to lives, but I'd like to see that confirmed.
Fortunately, they linked to their data set on this. Give me a few minutes to work on something.
posted by stevis23 at 7:24 AM on June 3, 2014 [1 favorite]
I've got a good feeling about 2014.
Sorry but NOAA apparently isn't feelin' your vibe.*
*It's this guy's fault.
posted by The 10th Regiment of Foot at 7:33 AM on June 3, 2014
Sorry but NOAA apparently isn't feelin' your vibe.*
*It's this guy's fault.
posted by The 10th Regiment of Foot at 7:33 AM on June 3, 2014
I want better social science and I also want better social science criticism. The link in the FPP was pretty poor criticism as were many glib comments. Surprisingly, the Slate article that troika linked to is quite good and shows that their broader conclusion, which is not the thrust of their research (despite the headliney title) fails some pretty simple robustness checks.
posted by MisantropicPainforest at 7:35 AM on June 3, 2014
posted by MisantropicPainforest at 7:35 AM on June 3, 2014
That totally seems like the the right level of granularity given the number of meaningful data points that they have.
It actually is because they aren't using the 1-9 scale to measure hurricanes, the are doing it to measure the perceived gender of names.
So on a scale from 1-7 people respond 1/2 point more cautiously to male names and seem to treat female-named hurricanes just the same as unnamed hurricanes.
Which is statistically significant and suggests that people view the impact of hurricanes through the lens of gender.
How on earth would this translate to real-life action? How would this lead to dramatic reductions in deaths?
Those are excellent questions and worth exploring in future research! Its also not what the study was trying to find out.
posted by MisantropicPainforest at 7:38 AM on June 3, 2014
It actually is because they aren't using the 1-9 scale to measure hurricanes, the are doing it to measure the perceived gender of names.
So on a scale from 1-7 people respond 1/2 point more cautiously to male names and seem to treat female-named hurricanes just the same as unnamed hurricanes.
Which is statistically significant and suggests that people view the impact of hurricanes through the lens of gender.
How on earth would this translate to real-life action? How would this lead to dramatic reductions in deaths?
Those are excellent questions and worth exploring in future research! Its also not what the study was trying to find out.
posted by MisantropicPainforest at 7:38 AM on June 3, 2014
Of course that's not what the study was trying to find out. This is a trivial bunch of experiments showing that people will answer a questionnaire about hurricanes differently based on the gender of the hurricane's name.
But funny thing, what is the name of the paper? "Female hurricanes are deadlier than male hurricanes." That seems a bit more provocative. Shockingly, the press coverage seems more interested in the point made in the title than in the Mechanical Turk experiments.
And from the paper: "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." Of course if the researchers were actually interested in preventing hurricane deaths they never would have written this, because this is a laughable statement on its face. But again, if you want to publish a paper and stir up some press, you just need some p-values under 0.05. No one is going to be using their negative binomial model to predict hurricane deaths, trust me.
posted by leopard at 7:51 AM on June 3, 2014 [2 favorites]
But funny thing, what is the name of the paper? "Female hurricanes are deadlier than male hurricanes." That seems a bit more provocative. Shockingly, the press coverage seems more interested in the point made in the title than in the Mechanical Turk experiments.
And from the paper: "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." Of course if the researchers were actually interested in preventing hurricane deaths they never would have written this, because this is a laughable statement on its face. But again, if you want to publish a paper and stir up some press, you just need some p-values under 0.05. No one is going to be using their negative binomial model to predict hurricane deaths, trust me.
posted by leopard at 7:51 AM on June 3, 2014 [2 favorites]
OK, here's NDAM vs. MFI as a .png on my Google Drive
Notes:
*I have no idea what "Low" and "High" were for the authors; they don't say. So I don't break it out.
*Should you throw out the top three points (1 "M", 2 "F") as outliers? I don't know. You've only got 92 data points to work with already.
*Anything in the paper's Fig. 1 between bins MFI 4 and MFI 6 should probably be ignored. There's little data there.
34 of the 92 points were from the female-only era; 3 were before that (Easy, King, and Abel), leaving 55 which should alternate. Here's the "modern" ones only, zoomed in to NDAM 30000 or less.
If there's an effect, it's not jumping out at me; it looks pretty even when male names are in play but the inclusion of pre-1979 data might be introducing some skew. That inclusion was problematic for other reasons, as previously discussed.
I can't believe they weren't asked to put this info in in the peer review process. They may have done it, but extraordinary claims demand extraordinary evidence and I'm not impressed with PNAS for letting them publish the archival study section without this (and other) demonstrations.
Presented without comment: "Easy" from 1950 scored a 6.7 on the MFI.
posted by stevis23 at 7:52 AM on June 3, 2014 [2 favorites]
Notes:
*I have no idea what "Low" and "High" were for the authors; they don't say. So I don't break it out.
*Should you throw out the top three points (1 "M", 2 "F") as outliers? I don't know. You've only got 92 data points to work with already.
*Anything in the paper's Fig. 1 between bins MFI 4 and MFI 6 should probably be ignored. There's little data there.
34 of the 92 points were from the female-only era; 3 were before that (Easy, King, and Abel), leaving 55 which should alternate. Here's the "modern" ones only, zoomed in to NDAM 30000 or less.
If there's an effect, it's not jumping out at me; it looks pretty even when male names are in play but the inclusion of pre-1979 data might be introducing some skew. That inclusion was problematic for other reasons, as previously discussed.
I can't believe they weren't asked to put this info in in the peer review process. They may have done it, but extraordinary claims demand extraordinary evidence and I'm not impressed with PNAS for letting them publish the archival study section without this (and other) demonstrations.
Presented without comment: "Easy" from 1950 scored a 6.7 on the MFI.
posted by stevis23 at 7:52 AM on June 3, 2014 [2 favorites]
It actually is because they aren't using the 1-9 scale to measure hurricanes, the are doing it to measure the perceived gender of names.
The chart you were referring to earlier predicts deaths by 1-9 scale (actually 1-11, I misspoke) for high- and low-severity hurricanes separately. So in their view the scale is predictive of real-life hurricane fatalities, and you expect more deaths for something rated 8 than something rated 7.
posted by leopard at 7:57 AM on June 3, 2014
The chart you were referring to earlier predicts deaths by 1-9 scale (actually 1-11, I misspoke) for high- and low-severity hurricanes separately. So in their view the scale is predictive of real-life hurricane fatalities, and you expect more deaths for something rated 8 than something rated 7.
posted by leopard at 7:57 AM on June 3, 2014
"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."
Did they account for the fact that Eloise made landfall and had high fatalities among Spanish- and French-speaking populations in Hispaniola and Cuba?
posted by CBrachyrhynchos at 8:32 AM on June 3, 2014
Did they account for the fact that Eloise made landfall and had high fatalities among Spanish- and French-speaking populations in Hispaniola and Cuba?
posted by CBrachyrhynchos at 8:32 AM on June 3, 2014
The Slate takedown is great. I hate the fact that this "study" has been so publicly making the rounds.
posted by Sticherbeast at 9:41 AM on June 3, 2014 [4 favorites]
posted by Sticherbeast at 9:41 AM on June 3, 2014 [4 favorites]
NoMich: "Drownquick is the greatest metal band of all time."
This was my thought. Not for drownquick, but give them norwegian death metal band names. HATEFUL PUTRESCENCE, GORGOROTH, SPLATERED ENTRAILS.
GRINDCORE HURRICANES SAVE LIVES PEOPLE!
posted by symbioid at 10:10 AM on June 3, 2014 [1 favorite]
This was my thought. Not for drownquick, but give them norwegian death metal band names. HATEFUL PUTRESCENCE, GORGOROTH, SPLATERED ENTRAILS.
GRINDCORE HURRICANES SAVE LIVES PEOPLE!
posted by symbioid at 10:10 AM on June 3, 2014 [1 favorite]
Also - name them Olga and Helga and not things like Sarah or Elizabeth and see how it rates.
posted by symbioid at 10:11 AM on June 3, 2014
posted by symbioid at 10:11 AM on June 3, 2014
stevis23, the supplementary material Table S1 shows the correlation between MFI and NDAM (pretty close to 0, not statistically significant). So the paper does have the basics.
The real problem in the paper is that the researchers "win" by publishing a bunch of regression statistics with statistically significant p-values. There's no relationship between MFI and deaths until you get down to adding the interactions between MFI and damage and MFI and pressure. And of course the modeled effect size is completely unrealistic. But there are *s and **s and ***s in the regression table.
posted by leopard at 1:06 PM on June 3, 2014
The real problem in the paper is that the researchers "win" by publishing a bunch of regression statistics with statistically significant p-values. There's no relationship between MFI and deaths until you get down to adding the interactions between MFI and damage and MFI and pressure. And of course the modeled effect size is completely unrealistic. But there are *s and **s and ***s in the regression table.
posted by leopard at 1:06 PM on June 3, 2014
This was my thought. Not for drownquick, but give them norwegian death metal band names. HATEFUL PUTRESCENCE, GORGOROTH, SPLATERED ENTRAILS.
I like this, if only for the possibilities for retroactive prophesy, since death metal already heralded the coming of the destroyer of worlds, Grumpy Cat.
posted by CBrachyrhynchos at 1:25 PM on June 3, 2014
I like this, if only for the possibilities for retroactive prophesy, since death metal already heralded the coming of the destroyer of worlds, Grumpy Cat.
posted by CBrachyrhynchos at 1:25 PM on June 3, 2014
leopard, thanks I missed that trying to do the five-minute version. Still I think my plot illustrates a glaring weakness in the middle of Fig. 1...the numbers there are based on a strikingly small number of incidents in an already small data set.
Does anybody know about this NDAM, by the way? Does it normalize for, how shall I say it, opportunity? That is, the fact that there's a lot more built up in the path of hurricanes in later years than earlier ones, or that not every eye comes ashore in a similarly populated area.
posted by stevis23 at 2:01 PM on June 3, 2014
Does anybody know about this NDAM, by the way? Does it normalize for, how shall I say it, opportunity? That is, the fact that there's a lot more built up in the path of hurricanes in later years than earlier ones, or that not every eye comes ashore in a similarly populated area.
posted by stevis23 at 2:01 PM on June 3, 2014
I wouldn't worry about NDAM.
Look at it this way.
NDAM is correlated with fatalities.
MFI is uncorrelated with fatalities.
MFI is uncorrelated with NDAM.
If you control for NDAM, MFI is still uncorrelated with fatalities.
All of this is intuitive. The magic happens when the authors add in an interaction term MFI * NDAM.
So in general, MFI tells you nothing about the fatality rate. But for high severity storms, female names lead to more deaths (and for low severity storms, female names lead to fewer deaths).
Gee, this couldn't possibly be a statistical fluke linked to the rather small number of high severity storms, could it? Nah look, the p value is below the critical value, this is real science.
posted by leopard at 2:40 PM on June 3, 2014
Look at it this way.
NDAM is correlated with fatalities.
MFI is uncorrelated with fatalities.
MFI is uncorrelated with NDAM.
If you control for NDAM, MFI is still uncorrelated with fatalities.
All of this is intuitive. The magic happens when the authors add in an interaction term MFI * NDAM.
So in general, MFI tells you nothing about the fatality rate. But for high severity storms, female names lead to more deaths (and for low severity storms, female names lead to fewer deaths).
Gee, this couldn't possibly be a statistical fluke linked to the rather small number of high severity storms, could it? Nah look, the p value is below the critical value, this is real science.
posted by leopard at 2:40 PM on June 3, 2014
« Older Hackers disclose how Russia employs professional... | That's no oomn! Newer »
This thread has been archived and is closed to new comments
posted by Sequence at 6:05 PM on June 2, 2014 [11 favorites]