"But Lazo thinks that neither the archival analysis nor the psychological experiments support the team’s conclusions. For a start, they analysed hurricane data from 1950, but hurricanes all had female names at first. They only started getting male names on alternate years in 1979. This matters because hurricanes have also, on average, been getting less deadly over time. 'It could be that more people die in female-named hurricanes, simply because more people died in hurricanes on average before they started getting male names,' says Lazo.
Jung’s team tried to address this problem by separately analysing the data for hurricanes before and after 1979. They claim that the findings 'directionally replicated those in the full dataset' but that’s a bit of a fudge. The fact is they couldn’t find a significant link between the femininity of a hurricane’s name and the damage it caused for either the pre-1979 set or the post-1979 one (and a 'marginally significant interaction' of p=0.073 doesn’t really count). The team argues that splitting the data meant there weren’t enough hurricanes in each subset to provide enough statistical power. But that only means we can’t rule out a connection between gender and damage; we can’t soundly confirm one either."
And BTW they just excluded Katrina as an outlier. While not necessarily a totally unwarranted decision, it makes already marginal conclusions that much more so.
Isn't this conclusion more significant because Katrina is excluded?
Edit: I meant the layman's sense of "significant" here. I just don't see how excluding Katrina "makes already marginal conclusions that much more so". I read your "but" statement as just saying that they haven't necessarily proven causality, but I don't see what that has to do with my statement.
Well, the word "significant" is a loaded one in this discussion -- in this context, it typically means "statistically significant," which has a very narrow (and sometimes questionable[1]) meaning.
You're right that leaving Katrina in the dataset (assuming the same techniques were used -- there are other ways to deal with outliers other than dropping them) would bias the result further towards indicating that female-named hurricanes are more dangerous. But the persistence of a significant finding in that regard in the absence of that data point does not prove a causal link, much less the specific one the author suggests.
It was intended more as a general point about picking and choosing of data. You're of course correct that including Katrina would buttress the female hurricane deadliness theory but would also open up the study to charges that it was influenced by a single outlier.
Whoa now, Lazo wants to say the experiment isn't useful because it uses undergrads who may not represent the actual demographic of interest? Does he want to abandon human-subject experiments altogether?
Abandoning the inexpensive-but-heavily-skewed "grab some of the local population" (of our university town) (in the Western world) wouldn't be a bad idea for psychology, no.
You do realize that feminists are the ones who lobbied to go from all female names to half male, right? Meaning that if this dubious research is accurate, they saved lives.
> Meaning that if this dubious research is accurate, they saved lives.
That doesn't necessarily follow. It could be that people have on average a fixed amount of "preparedness" per unit time that they will spend on various events as they see fit. Switching from all female names to half female half male might have simply caused more of this preparedness to be spent on the half named after males, but the total amount of preparedness would be unchanged.
I suppose that's possible, but it's the kind of claim that needs to be proven rather than disproven (just like the claim in the article, which is light-years from proven).
Then it will save even more lives for every hurricane to be named after a metal band (e.g.: Destructinator.)
Then we can apply suffixes, like "the Terrible" (Destructinator the Terrible).
Then if that doesn't work well enough to change behavior, we can simply lie about the risks (Destructinator the Terrible will kill you unless you take shelter in Canada with 90% probability.)
Hopefully this drumbeat of psychological warfare will cause the coastal residents to move hundreds of miles from the coasts and there will be no more hurricane deaths again.
Then we can start naming tornados after biker bars.
Hah, your comment highlighted the ridiculousness of the article, but there's also some truth to it. Imagine you were considering moving to Florida, but they had recently been ravaged by "The Destructinator of Death" and the "Plague of Extreme Sorrows." It would probably give you more pause than thinking about Hurricanes Andrew and Katrina.
Oh, that awful society which never takes women seriously; it's so bad that people even die more from Camille, Katrina and Fifi! Let us raise consciousness of the endless woe to all women!
It's feminism's fault that this guy couldn't think of a coherent way to blame feminism for this. Ever notice how "mean" is an anagram for "a men" (an abbreviation for "all men")? After all, they don't call it the "womean". Therefore, statistics have a misandrist bias. #notallmean
I can't see the full-text to check the stats, but I'd wager this claim is heavily dependent on single point Fifi: a category 2 hurricane that was amongst the deadliest.
Edit: Ah, in fact, there is no significant result that corresponds to the title at all. Statistically, they have no right to claim the headline is true. Pure link-bait :/
I call BS on this one. First, the implicit assumption seems to be that hurricanes have approximately the same strength and it's people's expectations that is the dominant factor in the death tool, i.e. (from the paper's abstract): "Feminine-named hurricanes (vs. masculine-named hurricanes) cause significantly more deaths, apparently because they lead to lower perceived risk and consequently less preparedness." Now I'm no meteorologist, but it seems like this may or may not be true. Since the male and female names alternate this may be pointing to an underlying mechanism that modulates the strength of consecutive storms' strength.
The study included some questions where participants rated the perceived danger from some made-up storm names. People said they expected less danger from a given storm description if it had a more feminine name.
Can this effect be reasonably prevented? See http://xkcd.com/915/ : some hurricane names will always be more threatening than other hurricane names, and will induce consequently more preparation. But the only thing to which people compare the names of hurricanes is the names of other hurricanes (and maybe other named storms), so there must always be a least threatening hurricane name, etc. Even if we took the "nuclear" route and named hurricanes after, for example, historical tyrants (Hurricane Genghis, Hurricane Adolf, etc) this could happen nonetheless.
This seems to suggest that hurricanes should not be named in advance. If all we know is that a hurricane of a particular strength will hit tomorrow, we won't hang around due to gender prejudice. Then when we find out next week that the hurricane is named Daenerys we'll realize we shouldn't have worried about it.
This is an excellent example of reasoning that Dawkins refers to as 'PETWAC' (Population of Events That Would happen to Appear Coincidental).
Consider three possible scenarios:
1) female hurricanes, as a set, are deadlier than male hurricanes
2) male hurricanes, as a set, are deadlier than female hurricanes
3) neither male nor female hurricanes are deadlier
Surprisingly, if you assume a uniform distribution of deadliness independent of the names, you actually find that it's more likely that either 1 or 2 is true than that 3 is true; 3 being true would require that an arbitrary partitioning of the hurricanes would have resulted in equal deadliness, but if the partition is arbitrary, it's more likely that it creates two unequal groups. However, a quick appeal to symmetry shows that either 1 or 2 being true are equally newsworthy.
It's like the news getting excited that a coin-toss was heads... "Wow, look at how heads that coin toss was!"... When in reality, the odds of heads OR tails were exceptionally high and the newsworthy item would be the coin landing on edge.
Given that this was published in PNAS and edited by someone from Princeton, it's safe to assume (unfortunately it's paywalled so I can only assume) that the authors know how to perform a significance test.
Exactly. What's at play is a combination of publication bias and (possibly) a slightly-too-narrow null hypothesis. This kind of silliness can squeeze through without any need for the null hypothesis to be entirely absent, which is the aspect of fixermark's post that I object to. I'd bet good money at 5:1 odds that the null hypothesis the paper used was more sophisticated than his uniform distribution let alone the delta distribution he accuses them of using.
"But Lazo thinks that neither the archival analysis nor the psychological experiments support the team’s conclusions. For a start, they analysed hurricane data from 1950, but hurricanes all had female names at first. They only started getting male names on alternate years in 1979. This matters because hurricanes have also, on average, been getting less deadly over time. “It could be that more people die in female-named hurricanes, simply because more people died in hurricanes on average before they started getting male names,” says Lazo.
Jung’s team tried to address this problem by separately analysing the data for hurricanes before and after 1979. They claim that the findings “directionally replicated those in the full dataset” but that’s a bit of a fudge. The fact is they couldn’t find a significant link between the femininity of a hurricane’s name and the damage it caused for either the pre-1979 set or the post-1979 one (and a “marginally significant interaction” of p=0.073 doesn’t really count [this is a link in the actual article - http://mchankins.wordpress.com/2013/04/21/still-not-signific... ]). The team argues that splitting the data meant there weren’t enough hurricanes in each subset to provide enough statistical power. But that only means we can’t rule out a connection between gender and damage; we can’t soundly confirm one either."
Which is more likely everyone involved in publishing this study is an idiot or the difference is non trivial. IMO, this is much more likely that fixermark is suffering from http://en.wikipedia.org/wiki/Dunning–Kruger_effect, but I have also not read the study.
Edit: That came off as overly harsh my point was the chance of 3 is not close to zero.
Nobody has to be an idiot for this study to get published; it's all about incentives.
My point is that if the data showed that male hurricanes were deadlier than female hurricanes, this team could very will still have published a very similar paper reaching very similar but gender-reversed conclusions based upon the same laboratory experiments.
That still definitely leaves open the question of how possible (3) was. Since I'm not a peer (i.e. the study is paywalled and I don't have access), I can't peer-review their figures, so I'm left critiquing from the outside based on hypothesis and conjecture. If anyone has access to the raw numbers, I'd be happy to follow their research.
If you flip a coin a thousand times and it always comes up heads, it's fairly safe to assume it will come up heads the thousandth and one time. I know we should be skeptical at everything, but I'm going to assume the journal and authors aren't completely incompetent to make a mistake like that.
Of course there is prior probability to factor in, and it doesn't seem very likely a priori that this would be the case. So the effect size would have to be fairly large or have a large number of examples to convince me of this.
After a thousand times it's pretty unlikely it's a fair coin. If it came up heads a million times it's pretty much impossible. Even very small prior probabilities will become dominant with enough evidence.
If you're given a fair coin, then it's impossible for it to also be a biased coin.
If you're given a coin, and you don't know whether it's fair or biased, then you can do an inference to give an estimated probability of how biased the coin is.
Those are the two situations. This is a problem about perfect versus imperfect information, not about probabilities.
Nowhere was it stated the coin was 100% guaranteed to be fair. You are missing the point of my original example.
My point was you can estimate the probability of a hypothesis by looking at the data. E.g. is it a fair or biased coin, and see how many times it came up heads or tails. Or are female named hurricanes more likely to kill than male ones or not?
>Nowhere was it stated the coin was 100% guaranteed to be fair.
I agree completely.
>You are missing the point of my original example.
No I wasn't. I was just stating the possible scenarios (of which yours was one) so that others reading could glean the point a little more easily. Perhaps I failed in this regard, but I certainly understood what you were trying to convey.
>My point was you can estimate the probability of a hypothesis by looking at the data. E.g. is it a fair or biased coin, and see how many times it came up heads or tails. Or are female named hurricanes more likely to kill than male ones or not?
One side has been found to go it's way NOT by chance. This is how peer review / science works.
That said there are issues, this is not one of them.
If you can read the article they also performed 6 other experiments where they for instance show people a weather map, tell them the name (M/F) and ask if they should evacuate.
But I have no idea how they justify including years where male names weren't allocated.
That's... not what statistical significance means. All statistical significance means is that there's a less than five percent chance of getting that result at random, if the null hypothesis were true. Now, granted, most studies reported on find p-values well below the .05 threshold. But still, if you go out and count the number of studies published in a year, one percent of them is still a LOT of wrong studies. And remember that, due to publication bias, it's a biased sample of the population of studies. So for every 100 studies that are examining a hypothesis that is in reality false, you'll get 4-5 that get a statistically significant finding instead. And it's those few studies that end up getting published.
This is also not true. For starters many studies don't use a p value of .05.
People seem to have read one xkcd comic and think now all studies are false.
There are issues, especially were the people doing the studies want a particular outcome for financial gain.
But if there are issues with this study talk about them. Were are the other 19 studies they have hidden? Where have they not followed proper statistical practice?
How does the other 6 theoretical experiments on the same paper which some also have significance come into it?
Surprisingly, if you assume a uniform distribution of deadliness independent of the names, you actually find that it's more likely that either 1 or 2 is true than that 3 is true
Right, but if you are a statistician what you should care about is the "distance" of either 1 or 2 from 3. It would not be surprising that one group is deadlier than the other, but it should be surprising if female hurricanes are "quite a bit" deadlier than male hurricanes (or vice versa), assuming this uniform distribution that you mentioned.
Would you be surprised if you won the lottery? Yes. I imagine you'd also be ecstatic. However, the likelihood of your numbers being chosen had equal probability (assuming the numbers are drawn uniformly at random) with any other set of numbers anybody else picked. If someone else had won the lottery instead of you, they would also be surprised.
Female hurricanes being 'quite a bit deadlier' than male hurricanes may well be surprising but it has equal probability with the event that male hurricanes are 'quite a bit deadlier' than female hurricanes.
What you're describing is a rare event. Rare events happen, just rarely.
Female hurricanes being 'quite a bit deadlier' than male hurricanes may well be surprising but it has equal probability with the event that male hurricanes are 'quite a bit deadlier' than female hurricanes.
I definitely agree with this.
What you're describing is a rare event. Rare events happen, just rarely.
I'll restate this hurricane problem. Suppose we're flipping a fair coin. We should expect to see roughly equal numbers of heads and tails. If I flip the same coin a thousand times and see 523 heads, that seems "reasonable". It's certainly possible to get 873 heads, but I might start to wonder if the coin is really fair not because it's very unlikely to get 873 heads with a fair coin (and it is very unlikely) but because it's more plausible that the coin is biased. That is, it's more likely that a coin that gets heads with some probability near 0.873 produces 873 heads in a thousand flips than it is for fair coin to produce 873 heads in a thousand flips.
In the context of hurricanes, if we assign genders randomly we expect close to equal deadliness between male and female hurricanes. If there have been a thousand hurricanes, and female hurricanes cause 87.3% of all hurricane deaths, well... I might start wondering if giving hurricanes female names somehow makes a hurricane more deadly.
Of course, this is all in the context of a uniform distribution of names on hurricanes.
>Suppose we're flipping a fair coin. We should expect to see roughly equal numbers of heads and tails. If I flip the same coin a thousand times and see 523 heads, that seems "reasonable". It's certainly possible to get 873 heads, but I might start to wonder if the coin is really fair not because it's very unlikely to get 873 heads with a fair coin (and it is very unlikely) but because it's more plausible that the coin is biased.
The first sentence said the coin was fair. Given that the coin is fair, flipping 873 heads in a thousand tosses is perfectly fine, it just has a smaller probability than flipping 523 heads (as you have already clearly stated).
Believing that the coin is biased is nothing more than human scepticism because we were told the coin was fair. Given that the coin is biased, you are right to state that the probability is higher. Probability is about working with given information, not about being suspicious about given information. If you had a reason to believe the coin wasn't fair, it being fair would not be part of the information given. The point I'm making here is that it is easy to mislead someone. You told me the coin was fair. If you then cheat and flip a biased coin there is no way for me tell if the coin is biased because it is my belief the coin is fair.
Prior information doesn't change after you look at the data. The distribution taking into account observations is different, and it is called the posterior distribution.
You have essentially taken my statement:
What you're describing is a rare event. Rare events happen, just rarely.
and repeated exactly my argument back to me.
Given that the hurricane gender is assigned uniformly at random, it is fine to observe that female hurricanes are more deadly. Quite a bit more deadly is nothing more than random chance. Remember, the prior was that they were assigned uniformly at random.
It could be the case that people assign names to hurricanes based on behaviour, path, time of year, or whatever other characteristic you like (and perhaps this is done subconsciously). In this case, the prior belief is that they are no longer assigned uniformly at random, but a human assigns them genders correlated with some physical quantity. Now, is it surprising that female hurricanes are more deadly? Perhaps not.
If any of this were true, we wouldn't have statistics departments at universities. Of course you can draw conclusions from data. Of course we should be skeptical of an assumption if the data shows that an incredibly unlikely sequence of events happened in light of that assumption.
By the way, hurricanes are assigned genders alternately. The genders go back and forth.
If any of this were true, we wouldn't have statistics departments at universities.
I'm not sure what statement you're making here.
Of course you can draw conclusions from data.
I never once said you could not make conclusions from data.
The data I had was that a) a fair coin was flipped; and b) out of 1000 flips, 873 of them were heads. Sure, this event has a small probability, but so what? Does it mean the coin was biased? Of course not. You told me that the coin was fair. If you didn't give me a) then I could certainly try to tell the probability of how biased the coin was doing an inference based on the data. But that wasn't part of the parent's question or background information. You go with what you know, not what you're sceptical about. If you win the lottery (which has a tiny probability), does that mean the lottery was biased towards those numbers?
Of course we should be skeptical of an assumption if the data shows that an incredibly unlikely sequence of events happened in light of that assumption.
Let me pose the question to you. You are given a fair two-sided coin. You flip the coin 1000 times. 875 of the flips turn up heads. Is the coin biased?
The statement I'm making is that you can simply brush off every scientific study by saying "yeah, that's just random chance, don't worry about it". The way that you're looking at this is completely backwards. If the data shows something incredibly unlikely happened, that means we should re-evaluate our assumptions.
> Let me pose the question to you. You are given a fair two-sided coin. You flip the coin 1000 times. 875 of the flips turn up heads. Is the coin biased?
>The statement I'm making is that you can simply brush off every scientific study by saying "yeah, that's just random chance, don't worry about it".
That is not what I was doing, and that is not true.
Scientific studies involving statistics and probability, when done properly, are sound. Quantifying uncertainty is a big deal, and when you can do it you can make very solid conclusions. Are there perturbations due to random chance? Sure. That doesn't mean you can always brush them off.
>If the data shows something incredibly unlikely happened, that means we should re-evaluate our assumptions.
Absolutely agree with you here. My point was it is crucial that you document what you believe about the problem at hand. If you believe something is not true, it is not usually assumed to be true.
>> Let me pose the question to you. You are given a fair two-sided coin. You flip the coin 1000 times. 875 of the flips turn up heads. Is the coin biased?
>Probably.
You are incorrect. The event that occurred was just very unlikely. I can tell you that with complete certainty because you were given a fair coin and you knew that information.
If, on the other hand, you didn't know whether or not the coin was fair, then you would be correct (but that wasn't the case).
If I understand this correctly, hurricanes should be named with the most threatening names possible. So no more Sandy or even Edward, but Cerberus, Smaug, and Grendel.
65 comments
[ 2.4 ms ] story [ 199 ms ] threadhttp://www.pnas.org/content/suppl/2014/05/30/1402786111.DCSu...
http://phenomena.nationalgeographic.com/2014/06/02/why-have-...
"But Lazo thinks that neither the archival analysis nor the psychological experiments support the team’s conclusions. For a start, they analysed hurricane data from 1950, but hurricanes all had female names at first. They only started getting male names on alternate years in 1979. This matters because hurricanes have also, on average, been getting less deadly over time. 'It could be that more people die in female-named hurricanes, simply because more people died in hurricanes on average before they started getting male names,' says Lazo.
Jung’s team tried to address this problem by separately analysing the data for hurricanes before and after 1979. They claim that the findings 'directionally replicated those in the full dataset' but that’s a bit of a fudge. The fact is they couldn’t find a significant link between the femininity of a hurricane’s name and the damage it caused for either the pre-1979 set or the post-1979 one (and a 'marginally significant interaction' of p=0.073 doesn’t really count). The team argues that splitting the data meant there weren’t enough hurricanes in each subset to provide enough statistical power. But that only means we can’t rule out a connection between gender and damage; we can’t soundly confirm one either."
Edit: I meant the layman's sense of "significant" here. I just don't see how excluding Katrina "makes already marginal conclusions that much more so". I read your "but" statement as just saying that they haven't necessarily proven causality, but I don't see what that has to do with my statement.
You're right that leaving Katrina in the dataset (assuming the same techniques were used -- there are other ways to deal with outliers other than dropping them) would bias the result further towards indicating that female-named hurricanes are more dangerous. But the persistence of a significant finding in that regard in the absence of that data point does not prove a causal link, much less the specific one the author suggests.
[1] For some criticisms of how statistical significance is currently being used, read this: http://www.deirdremccloskey.com/docs/jsm.pdf
http://www.scientificamerican.com/podcast/episode/psychology...
That doesn't necessarily follow. It could be that people have on average a fixed amount of "preparedness" per unit time that they will spend on various events as they see fit. Switching from all female names to half female half male might have simply caused more of this preparedness to be spent on the half named after males, but the total amount of preparedness would be unchanged.
Then we can apply suffixes, like "the Terrible" (Destructinator the Terrible).
Then if that doesn't work well enough to change behavior, we can simply lie about the risks (Destructinator the Terrible will kill you unless you take shelter in Canada with 90% probability.)
Hopefully this drumbeat of psychological warfare will cause the coastal residents to move hundreds of miles from the coasts and there will be no more hurricane deaths again.
Then we can start naming tornados after biker bars.
Edit: Ah, in fact, there is no significant result that corresponds to the title at all. Statistically, they have no right to claim the headline is true. Pure link-bait :/
Second, up to 1978 hurricanes were given only female names (http://www.nhc.noaa.gov/aboutnames_history.shtml). Where does that leave this theory?
There's a reason they didn't lead with THAT headline, but rather the inflammatory and bogus one.
Consider three possible scenarios: 1) female hurricanes, as a set, are deadlier than male hurricanes 2) male hurricanes, as a set, are deadlier than female hurricanes 3) neither male nor female hurricanes are deadlier
Surprisingly, if you assume a uniform distribution of deadliness independent of the names, you actually find that it's more likely that either 1 or 2 is true than that 3 is true; 3 being true would require that an arbitrary partitioning of the hurricanes would have resulted in equal deadliness, but if the partition is arbitrary, it's more likely that it creates two unequal groups. However, a quick appeal to symmetry shows that either 1 or 2 being true are equally newsworthy.
It's like the news getting excited that a coin-toss was heads... "Wow, look at how heads that coin toss was!"... When in reality, the odds of heads OR tails were exceptionally high and the newsworthy item would be the coin landing on edge.
Only laboratory questioning showed a significant psychological bias (and there an unnamed hurricane was rated less risky than either).
What am I missing?
http://phenomena.nationalgeographic.com/2014/06/02/why-have-...
"But Lazo thinks that neither the archival analysis nor the psychological experiments support the team’s conclusions. For a start, they analysed hurricane data from 1950, but hurricanes all had female names at first. They only started getting male names on alternate years in 1979. This matters because hurricanes have also, on average, been getting less deadly over time. “It could be that more people die in female-named hurricanes, simply because more people died in hurricanes on average before they started getting male names,” says Lazo.
Jung’s team tried to address this problem by separately analysing the data for hurricanes before and after 1979. They claim that the findings “directionally replicated those in the full dataset” but that’s a bit of a fudge. The fact is they couldn’t find a significant link between the femininity of a hurricane’s name and the damage it caused for either the pre-1979 set or the post-1979 one (and a “marginally significant interaction” of p=0.073 doesn’t really count [this is a link in the actual article - http://mchankins.wordpress.com/2013/04/21/still-not-signific... ]). The team argues that splitting the data meant there weren’t enough hurricanes in each subset to provide enough statistical power. But that only means we can’t rule out a connection between gender and damage; we can’t soundly confirm one either."
Edit: That came off as overly harsh my point was the chance of 3 is not close to zero.
My point is that if the data showed that male hurricanes were deadlier than female hurricanes, this team could very will still have published a very similar paper reaching very similar but gender-reversed conclusions based upon the same laboratory experiments.
That still definitely leaves open the question of how possible (3) was. Since I'm not a peer (i.e. the study is paywalled and I don't have access), I can't peer-review their figures, so I'm left critiquing from the outside based on hypothesis and conjecture. If anyone has access to the raw numbers, I'd be happy to follow their research.
Of course there is prior probability to factor in, and it doesn't seem very likely a priori that this would be the case. So the effect size would have to be fairly large or have a large number of examples to convince me of this.
Unless you have reason to believe that it's a fair coin.
If you're given a coin, and you don't know whether it's fair or biased, then you can do an inference to give an estimated probability of how biased the coin is.
Those are the two situations. This is a problem about perfect versus imperfect information, not about probabilities.
My point was you can estimate the probability of a hypothesis by looking at the data. E.g. is it a fair or biased coin, and see how many times it came up heads or tails. Or are female named hurricanes more likely to kill than male ones or not?
I agree completely.
>You are missing the point of my original example.
No I wasn't. I was just stating the possible scenarios (of which yours was one) so that others reading could glean the point a little more easily. Perhaps I failed in this regard, but I certainly understood what you were trying to convey.
>My point was you can estimate the probability of a hypothesis by looking at the data. E.g. is it a fair or biased coin, and see how many times it came up heads or tails. Or are female named hurricanes more likely to kill than male ones or not?
Yes, I completely agree.
This experiment has been found statistical true.
One side has been found to go it's way NOT by chance. This is how peer review / science works.
That said there are issues, this is not one of them.
If you can read the article they also performed 6 other experiments where they for instance show people a weather map, tell them the name (M/F) and ask if they should evacuate.
But I have no idea how they justify including years where male names weren't allocated.
People seem to have read one xkcd comic and think now all studies are false.
There are issues, especially were the people doing the studies want a particular outcome for financial gain.
But if there are issues with this study talk about them. Were are the other 19 studies they have hidden? Where have they not followed proper statistical practice?
How does the other 6 theoretical experiments on the same paper which some also have significance come into it?
Right, but if you are a statistician what you should care about is the "distance" of either 1 or 2 from 3. It would not be surprising that one group is deadlier than the other, but it should be surprising if female hurricanes are "quite a bit" deadlier than male hurricanes (or vice versa), assuming this uniform distribution that you mentioned.
Female hurricanes being 'quite a bit deadlier' than male hurricanes may well be surprising but it has equal probability with the event that male hurricanes are 'quite a bit deadlier' than female hurricanes.
What you're describing is a rare event. Rare events happen, just rarely.
I definitely agree with this.
What you're describing is a rare event. Rare events happen, just rarely.
I'll restate this hurricane problem. Suppose we're flipping a fair coin. We should expect to see roughly equal numbers of heads and tails. If I flip the same coin a thousand times and see 523 heads, that seems "reasonable". It's certainly possible to get 873 heads, but I might start to wonder if the coin is really fair not because it's very unlikely to get 873 heads with a fair coin (and it is very unlikely) but because it's more plausible that the coin is biased. That is, it's more likely that a coin that gets heads with some probability near 0.873 produces 873 heads in a thousand flips than it is for fair coin to produce 873 heads in a thousand flips.
In the context of hurricanes, if we assign genders randomly we expect close to equal deadliness between male and female hurricanes. If there have been a thousand hurricanes, and female hurricanes cause 87.3% of all hurricane deaths, well... I might start wondering if giving hurricanes female names somehow makes a hurricane more deadly.
Of course, this is all in the context of a uniform distribution of names on hurricanes.
The first sentence said the coin was fair. Given that the coin is fair, flipping 873 heads in a thousand tosses is perfectly fine, it just has a smaller probability than flipping 523 heads (as you have already clearly stated).
Believing that the coin is biased is nothing more than human scepticism because we were told the coin was fair. Given that the coin is biased, you are right to state that the probability is higher. Probability is about working with given information, not about being suspicious about given information. If you had a reason to believe the coin wasn't fair, it being fair would not be part of the information given. The point I'm making here is that it is easy to mislead someone. You told me the coin was fair. If you then cheat and flip a biased coin there is no way for me tell if the coin is biased because it is my belief the coin is fair.
Prior information doesn't change after you look at the data. The distribution taking into account observations is different, and it is called the posterior distribution.
You have essentially taken my statement:
What you're describing is a rare event. Rare events happen, just rarely.
and repeated exactly my argument back to me.
Given that the hurricane gender is assigned uniformly at random, it is fine to observe that female hurricanes are more deadly. Quite a bit more deadly is nothing more than random chance. Remember, the prior was that they were assigned uniformly at random.
It could be the case that people assign names to hurricanes based on behaviour, path, time of year, or whatever other characteristic you like (and perhaps this is done subconsciously). In this case, the prior belief is that they are no longer assigned uniformly at random, but a human assigns them genders correlated with some physical quantity. Now, is it surprising that female hurricanes are more deadly? Perhaps not.
By the way, hurricanes are assigned genders alternately. The genders go back and forth.
I'm not sure what statement you're making here.
Of course you can draw conclusions from data.
I never once said you could not make conclusions from data.
The data I had was that a) a fair coin was flipped; and b) out of 1000 flips, 873 of them were heads. Sure, this event has a small probability, but so what? Does it mean the coin was biased? Of course not. You told me that the coin was fair. If you didn't give me a) then I could certainly try to tell the probability of how biased the coin was doing an inference based on the data. But that wasn't part of the parent's question or background information. You go with what you know, not what you're sceptical about. If you win the lottery (which has a tiny probability), does that mean the lottery was biased towards those numbers?
Of course we should be skeptical of an assumption if the data shows that an incredibly unlikely sequence of events happened in light of that assumption.
Let me pose the question to you. You are given a fair two-sided coin. You flip the coin 1000 times. 875 of the flips turn up heads. Is the coin biased?
The statement I'm making is that you can simply brush off every scientific study by saying "yeah, that's just random chance, don't worry about it". The way that you're looking at this is completely backwards. If the data shows something incredibly unlikely happened, that means we should re-evaluate our assumptions.
> Let me pose the question to you. You are given a fair two-sided coin. You flip the coin 1000 times. 875 of the flips turn up heads. Is the coin biased?
Probably.
That is not what I was doing, and that is not true.
Scientific studies involving statistics and probability, when done properly, are sound. Quantifying uncertainty is a big deal, and when you can do it you can make very solid conclusions. Are there perturbations due to random chance? Sure. That doesn't mean you can always brush them off.
>If the data shows something incredibly unlikely happened, that means we should re-evaluate our assumptions.
Absolutely agree with you here. My point was it is crucial that you document what you believe about the problem at hand. If you believe something is not true, it is not usually assumed to be true.
>> Let me pose the question to you. You are given a fair two-sided coin. You flip the coin 1000 times. 875 of the flips turn up heads. Is the coin biased?
>Probably.
You are incorrect. The event that occurred was just very unlikely. I can tell you that with complete certainty because you were given a fair coin and you knew that information.
If, on the other hand, you didn't know whether or not the coin was fair, then you would be correct (but that wasn't the case).
/sorry