Slashdot Mirror
10-Year Cell Phone / Cancer Study Is Inconclusive
crimeandpunishment writes "A major international (retrospective) study into cell phones and cancer, which took 10 years and surveyed almost 13,000 people, is finally complete — and it's inconclusive. The lead researcher said, 'There are indications of a possible increase. We're not sure that it is correct. It could be due to bias, but the indications are sufficiently strong ... to be concerned.' The study, conducted by the World Health Organization and partially funded by the cellphone industry, looked at the possible link between cell phone use and two types of brain cancer. It will be published this week."
At least from this we know that cell phone radiation isn't causing some massive epidemic of brain cancer, and the affects, if there are any, are relatively small. That's not the biggest comfort you could have, but it's something (considering most of us are not going to give up our cell phones anyway).
Qxe4
Yeah, because surveying all those people would be ABSOLUTELY FREE and take NO TIME. Also, it's totally necessary to check everyone. Sampling and statistics don't exist.
How silly.
To get statistical significance, you don't need to sample the entire population. Beyond a certain number for a certain confidence level, you don't get very much more.
I'm a minority race. Save your vitriol for white people.
Not really. Sampling can give accurate results even when sampling a small percentage of the total population. If U.S. political polls select a sample size of between a few hundred and a thousand out of 300 million with only 3% error, it sounds reasonable that 13,000 would be a good sample size of a population 20 times that, giving the same margin of error.
Also remember that, assuming the sample is chosen well (it is a good cross-section of the population and not confined to one specific subgroup), the benefits of adding additional samples drops off. It is essentially logarithmic: at first, adding samples is a huge benefit: after a certain point, the incremental gain from one additional sample is only a tiny fraction of the first samples.
24 beers in a case, 24 hours in a day. Coincidence? I think not!
So people who are convinced cellphones cause cancer are going to take their "possible increase" and declare scientists just definitively said cellphones cause cancer.
On the other hand, cellphone companies may try to take "we're not sure that it is correct" and declare no link to cancer.
My webcomic
And even if there is some correlation, people need to put it in perspective.
The last time I talked to a flat-earth-er about their fear of cell phones causing cancer, they had a drink in one hand and a cigarette in the other.
Now that, Alanis Morrissette, is irony.
--- "We've always been at war with Eastasia."
I have a problem with "medical surveys" in that they a prone to make correlation-causation errors. This seems to be a measurable problem that can be tested in the lab. Why don't people do this instead. Put a lab monkey next to an active mobile phone and keep them there for several years. After that, dissect the monkey for any signs of cancer. If there is, then alert the public. You then look into how it happened, i.e the biochemical interactions that caused it. Just "surveying" people introduces biases, other factors like diet and lifestyle and also crackpots.
Have they done this study against other types of radio frequencies like cordless land-line phones? What about emergency services workers that carry radios on their hips until needed...are they being checked for hip-cancer? Doesn't Nike or some other shoe maker have a device that fits inside a shoe so people can listen to FM whilst jogging? Watch out for heel-cancer! The point being, why are cell-phones being singled out as possible culprits where then are so many other devices out there that use radio technology?
I think the media has way too much control over what is allowed to scare us into taking action. It seems that our efforts could be better directed toward something that actually makes sense. Let Mythbusters handle this type of shit.
Loading...
If U.S. political polls select a sample size of between a few hundred and a thousand out of 300 million with only 3%..."
I'm not so sure those percentages are accurate. You'll often see different polls differ by much more than that (far more often than 5% of the time or whatever the confidence level is).
I have a suspicion that the math works out with a lot of "if a1 through aN are true, then..." and then no one going to the trouble of working out how likely each of those is to actually be true because they're hard to measure.
Certainly actual elections tend to fall well outside the +/- 3% accuracy claimed by many of the election-day pollsters.
Can you be Even More Awesome?!
No, you get a smoother, more natural bass and just generally a warmer...uh, sorry, wrong thread!
The article in USA Today has a nice little gem in it: "The authors acknowledged possible inaccuracies in the survey from the fact that participants were asked to remember how much and on which ear they used their mobiles over the past decade. Results for some groups showed cellphone use actually appeared to lessen the risk of developing cancers, something the researchers described as "implausible."" Now, I don't know why, but something about this statement seems kind of important.
The only thing necessary for evil to triumph is for it to be pitted against a slightly greater evil
The principle is correct, but you're failing to take into account the probability of an the respective events. Given that winning 60% of the vote is considered a landslide, you can think of asking someone whether they're voting Republican or Democrat as a coin flip with a small bias in one way or the other. Because the race is so close, a few extra republicans or democrats in your sample won't produce a huge error in your estimate.
On the other hand, a brain tumor can be thought of as a rare event. If the true incidence rate of brain cancer is five occurrences per thousand people over ten years, and your sample of 1,000 people has six incidences, you have a sample error of 20%. It's because of this that a small variation in the numbers can produce a large error. Therefore if you want to accurately assess the rate of cancer, you need a much bigger sample size.
Slashdot: news for Apple. Stuff that Apple.
It really seems silly when, in America at least, age-adjusted rates of brain cancer have fallen or held steady since the 1990s. From the National Cancer Institute:
It would seem to me that falling cancer rates are no reason for assuming that widespread cellphone use has been a health concern.
If U.S. political polls select a sample size of between a few hundred and a thousand out of 300 million with only 3%..."
I'm not so sure those percentages are accurate.
They look accurate to me. From me undergrad stats classes, I seem to recall that to get 5% confidence level out of population of 10k, one needed a sample of around 850. For populations of 1000k, the sample size only went up by a few tens (perhaps to 900). Sampling is not linear, and it drops off the higher you go - IIRC (and I think I do), their is very little difference in the sample size for a population of 100k as there is for twenty times that number.
I'm a minority race. Save your vitriol for white people.
Certainly actual elections tend to fall well outside the +/- 3% accuracy claimed by many of the election-day pollsters.
Because for many of those pollsters accuracy isn't main goal; swaying people, untill the last minute, to vote for the "winners" is.
One that hath name thou can not otter
The uncertainty in the study is due to the low precision of their data- they asked people to try and remember how much they were typically using their cellphones. Surveying more people isn't going to get people to provide more precise data.
Also, unless the needed data is already available somewhere, gathering more data costs more money. As someone else mentioned in a sibling post, there are diminishing returns when increasing your sample size. Eventually the cost of the data will exceed the benefit to the certainty of your results.
My webcomic
Actually, it kind of does. If you have a null hypothesis "there is no link between cellphone use and brain cancer" then an inconclusive result would fail to disprove the null hypothesis and therefore affirm it. This is based on choosing a null hypothesis that is based on the sensible default position, which in this study is fine as long as you're the kind of person who is willing/capable of understanding that we are constantly bathed in all sorts of EM radiation of which cellphones only play a small part and that the default position from a conventional understanding of physics is that they're likely to be harmless.
It's also based on the idea that, for a risk factor for cancer(s) significant enough to be worth worrying about, we would expect to see an obvious and conclusive result. For instance, when testing the null hypothesis "there is no like between smoking and lung cancer", the observed data would overwhelmingly reject the null hypothesis. The reality is that there's all sorts of things that people think cause cancer, and many of them may do (e.g. drinking hot drinks regularly is linked with oral cancer) but most of the risk factors aren't significant to be worth worrying about.
I'm not so sure those percentages are accurate. You'll often see different polls differ by much more than that (far more often than 5% of the time or whatever the confidence level is).
Election polling is just especially difficult, since what counts is if you actually vote and who you vote for, neither of which have been determined at the time of the poll and could change. Election polling isn't simply an opinion poll, but is obviously supposed to reflect the population of people who will actually vote on election day. The polls have differing models of selecting "likely voters", and will thus have numbers that differ more than the margin of error for any single poll. In other words, taking the margin of error for a single poll and comparing it among multiple polls is invalid, since the differing polls used different means of sample selection.
Certainly actual elections tend to fall well outside the +/- 3% accuracy claimed by many of the election-day pollsters.
I guess I haven't found that to be true if you mean "tend to" is more than 50% of the time. Sure, you're going to find some that are outside of the 3% error bars, but you'd also expect that to happen, statistically speaking.
AccountKiller
While people in large numbers are essentially predictable (and therefore boring, which is why statistics - for the most part - works), those theorems are strictly valid only for true random variables. As GP pointed out, the differences between different polls sometimes like far outside the error bounds set by the poll itself. Kinda makes the error bound meaningless since it has been repudiated by empirical means. As always, observations reign supreme and if there's a conflict with theory, it is usually a case of unjustified assumptions - in this case, taking the approximate equivalence between mathematical random variables and real world people to be exact.
Also, you are right about more not being any better. At some point, you are just adding more and more precision to an inaccurate answer. It's like a calculator fetish - getting predictions to the 18th decimal point using a flawed model and wondering why they don't match reality.
Science isn't inconclusive. There is statistically significant, or not. In this case, not.
Test another hypothesis or test again if data looks fishy.
I mean that if cell phones cause cancer, you would expect the rate of cancer to raise along with the use of cell phones. Instead, cancer rates have fallen or stayed the same for 20 years.
To get statistical significance, you don't need to sample the entire population. Beyond a certain number for a certain confidence level, you don't get very much more.
Exactly right.
There was no statistical significance, which means that the cancers (or absence there of) were distributed over cell phone users and non-users (controls) with no preference for either group.
Normally this would be the end of it.
But by the way the reporter worded it (Inconclusive) and (to a lesser extent) the way the Researcher phrased it, indicates a clear predilection toward finding a positive correlation, which they could not do.
The takeaway is not that the study "inconclusive". The scientific takeaway is that there is yet again no evidence of correlation between cancer and cell usage.
Its over. The absence of evidence destroys this theory. Time to move on.
Sig Battery depleted. Reverting to safe mode.
Here are some additional details for those of you so inclined.
Consider a simple binary choice question. This is easily modelled by the binomial distribution which has well understood distributions. (Other distrbutions may be relevant but the principles remain pretty constant across them all.) The standard deviation is given by sqrt[np(1-p)] where n is the sample size and p is the probability of the observation you are interested in (the mean is np so in what follows I will be dividing by n to talk about percentages if you are taking notes). For example, are you male? If the true p is, say, 75% then you need a sample size of approximately 833 to get a 95% confidence interval (2 s.d.) of +/- 3%.
You might also note that the closer the true p is to 50%, the larger the sample size needed. If the true p is 50% you need a sample size of approximately 1100 for the same confidence interval. Furthermore, if you want to get it within 1%, the sample size goes up dramatically - to 10,000.
The population size is pretty much irrelevant. The population matters for ensuring that your sampling is truly random, but political pollsters can use the same sample sizes in Australia (pop ~20 million) as in the US (pop ~300 million) for similar accuracy. (Sampling bias is the reason that political polls can be out by so much - if you call households during work hours you are going to get a very different sample of people than if you call at dinner time.)
+1 Insightful, came here to say the same thing.
DNA doesn't break until you get into the UV-light range of electromagnetic waves, cell phone frequencies are orders of magnitude away from being able to do it....but don't let the pesky facts get in the way of anecdotes and scaremongering.
No sig today...
Absolutely, the sample size is inversely relative to how close the differential result is to the 'noise floor'.
In this respect, your first example is slightly flawed. As the expected determinant gets closer to the noise floor (ie. if the margin for a Republican or Democrat victory is going to be 0.01%, or 50.01% vs 49.99%), then a much greater sample size is needed to maintain confidence in the resultant prediction.
As you say, 60% is a landslide. So if that is the expected result, then a few percent error either way isn't going to change your final determination of the winner.
It could still be a general drop in cancer rates, but a specific rise in the rates for people who use cellphones (in certain conditions, given that pretty much everyone uses them these days?). Looking at simple numbers like that is inconclusive
The fact that most people don't understand statistics doesn't mean stats are bullshit. It just means people are dumb.