A 2001 Danish study of 7,000 women showed a 50 percent higher risk of breast cancer if they’d worked nights at least six months.
link to column.
I’ve seen various ‘risks of’ for various diseases. What does it mean? Obviously in the above quote there weren’t 50% more women with breast cancer.
That’s exactly what it means, but they mean 50% more of the breast cancer population, not of the whole population.
E.g. if 100 of 10000 women who didn’t work nights got breast cancer, they’re seeing 150 of 10000 who do work nights getting it (50% of 100).
They’re not seeing 5100 women getting breast cancer when they work nights (50% of 10000).
I am not buying that. It doesn’t say 50% more women got cancer, that is easy enough to say.
It says their risk went up.
There were probably some various controlled trials, and statistics and what not, and some sort of factor analysis that assigns the risk of getting cancer to each of several and sundry risks so identified. There is some handwaving statistical magic to make sure each of the risks is orthogonal to the other (IOW there are no correlations to each other), or the reason why that isn’t done is explained in the paper.
In the end, it means that the lifetime chance of getting breast cancer is 50% higher for women who have worked nights than have not. It says nothing whatsoever about the various outcomes of the trial’s raw data.
Could be, but given a properly designed and controlled trial, those two statements should be equivalent. Sure, the numbers aren’t going to be exact, but they should be closer than can be attributed to chance given some desired confidence level. I highly doubt that the incidence discovered was exactly 50%, for example.
I think we’re quibbling here: regardless of the methodology of the trial, the takeaway is that given otherwise identical groups, you’d expect 1.5 x as many women in the “works nights” group to actually get breast cancer as in the “not works nights” group. The “50% higher risk of” is just word choice.
For extra credit, try interpreting this study. It’s not the same as the OP’s, although it’s about the same topic: It took me a good ten minutes to even determine what they were claiming as results in any sort of numeric sense (if I’m right, and I’m not sure I am, they got a much lower percentage).
No, you are confusing the analysis with the raw data. If you want to test a new hypothesis, have at it, but that is not the correct answer to the OP.
We haven’t read the paper, but that is how Cecil put it. Doesn’t matter if it was 1, 12, or 45%.
Whether I conclude that or not from the paper is not what the sentence in the OP means. It is specifically talking about the risk to an individual, not the projected empirical results in an entire population.
You are arguing with a stats major about this, who recently did the math on a dissertation with a similar type study? 
For extra credit, try interpreting this study. It’s not the same as the OP’s, although it’s about the same topic: It took me a good ten minutes to even determine what they were claiming as results in any sort of numeric sense (if I’m right, and I’m not sure I am, they got a much lower percentage).
[/QUOTE]
Yes, that is a related study. The statistics are de riguer. Nothing fancy at all. The results are right in the abstract, especially the last sentence.
And yes, in that paper, they are using risk in the sense I am using it, not in the sense you suggest.
In this case it means that:
http://www.jstor.org/pss/3703682
Meaning that the ratio of nightshift workers in the cancer group compared with the non-cancer group indicated a 50% increase in risk. I’ll leave it to not_alice to explain how that’s different from 50% more women with breast cancer, except in the trivial “fewer women work the night shift”-sense, as that’s how I’d naïvely read understand it.
I agree with TimeWinder; if I would read a claim as in the OP, I would always expect the basis to be that in the one group they find 50% more cases of cancer. I’m not an MD (or PhD in medicine) and haven’t read the paper, so who knows…but in the Social Sciences, this is how things usually get reported.
If I start with 1%, and the rate increases by 100%, now it’s 2%. That’s all it means. You can have huge %'s when you say “increased rate of” or “increased risk of,” when the actual number is very small.
“Increased rate of” and “increased risk of” are not the same thing. AND, saying “increased rate of” is not the same as saying “the rate increased by…”
Powers &8^]
The OP was correct in saying that the study obviously didn’t find 50% more women with breast cancer. for one thing, the raw data needs various corrections, and analyses, as I described earlier, and as the related paper later linked described.
But it is clear that the conclusion is talking about the risk to an individual who matches a pretty specific set of criteria. IN this case, female, worked nights for so long in such and such an environment during such and such ages and so on.
It is a common error among the population to misinterpret statistical language as we see in this thread. I attribute that to “numerical illiteracy”, and to the educational system for not doing a better job.
But then, probably most people don’t really know what a “40% chance of rain” means either…I bet a lot of people think it will rain 40% of the day, or maybe even 40% as hard as it usually rains. It means neither.
I was going to let this go until you called me numerically illiterate. I assure you I am not. You’re overthinking this.
We don’t have the study data the OP referred to. In the absence of other information, the statement that “a study found a 50% increased risk of cancer in nurses that worked the night shift,” it is perfectly correct to assume that that means “Given 2N people selected representatively from the population the study tested, with N night shift workers and N not, we expect to find 150% of the number of cancer cases in the day workers among the night workers.”
Yes, any given individual might not have a 50% increased risk, but in aggregate they do, which is how I interpret the rather vague question the OP was asking. (All this assuming that the study’s conclusions are correct, but I don’t think that’s what we’re arguing).
It’s also true that the original study may have mis-sampled it’s population and corrected for it via later analysis, but most clinical trials try very hard to sample representatively of the population they’re trying to study. If they did mis-sample, then it’s possible the number of night-shift cancer cases IN THE ORIGINAL STUDY would be greater or less than 150% of the day-shift cancer cases (and of course there will be some variation by chance). It’s possible this is what the OP was asking, but it doesn’t seem like a very interesting question: all that analysis and such is done in order to ALLOW us to make the statement above (about what the actual effect is in the representative population). What they actually saw in the non-represensentative data (assuming there was any, which I still doubt) is just a statistical starting-point to getting to the reported conclusion.
In simpler terms: If we were to redo the study, with a sample representative of the studied population, we would, in fact, expect to find 150% of the per-capita cancer cases among the night shift workers compared to the non-night-shift ones.
I don’t know if you are numerically illiterate or not. In that case, I was talking about the general population. 
No it isn’t. It says nothing about whether the night shift works have already developed cancer by the time the data is taken, nothing at all. It is talking about the lifetime risk of individuals.
No, in fact, each individual so assigned that risk by statistical inference does have the 50% increased risk.
People with other risks are assigned other risks and classified as being in other groups.
I agree we can just work with the data as presented - we are debating the form of the text, not the accuracy of the data.
That being said, you are interpreting it incorrectly.
Perhaps it will be clearer to you if you work through an actual example of such a study, do all the statistics that lead to such a conclusion, and see if it really matches what you are claiming here.
I am not saying you couldn’t find some such methods, only that in a legitimately analyzed data set, you would not reach a conclusion about “risk” from merely descriptive statistics such as you are relying on to make your point. The actual analysis is more complex than that.
There are statistical techniques to account for that.
Of course. The OP (and you it seems) are simply reading the conclusion incorrectly. Perhaps because you don’t understand the underlying statistics involved in reaching the conclusion, which, as I said is not your fault, but the fault of the educational system.
Maybe, maybe not, but that is not what the OP was asking, nor was that what was written in the paper. It spoke of lifetime risks to individuals, not the number of cancers in the population.
Look at it this way: Smoking for 10 years between the ages of 15 and 30 probably increases your lifetime risk of certain cancers (and other illnesses) by some percentage, let’s call it x%. But that doesn’t mean you are going to expect to see those illnesses increased by x% in the population of smokers by age 30.
In such a study,you would use the same language as in the OP: The risk of smoking 10 years between 15 and 30 increases the risk of cancer by y%.
This is clearly discussing the risk to an individual, isn’t it? It certainly is not discussing the prevalence of cancers at the time the data was taken (age 30), right?
Also, your bit about the population extrapolation does not hold either, because in some cases, the cancers may be fatal, and people simply won’t live as long to be counted later. Sure, you could take that into account, but if that was what was done, the study would use extensive language to present itself as a population study and explain about the missing data due to early death. That is not done in any of the studies on night work in the thread here apparently.
So why would you draw the correspondingly opposite conclusions about night work?