|
|
Thread Tools | Display Modes |
Ads |
#212
|
|||
|
|||
"Just zis Guy, you know?" writes:
On Sat, 28 May 2005 01:14:39 GMT, (Bill Z.) wrote in message : You said "This "confounding factor" red herring you guys are now touting is just that." They said: it exists. "They" did not say it existed in what you posted. Actually they did. Especially the quote from Spaite, which regards it as a most significant finding. They did not say it existed in what you posted. Do you understand "what you posted" means? You know, the quote you actually provided? Are you now saying you quote one thing and then proceed to talk about something else? All they really said is that you have to compare similar populations or similar accidents (which is kind of basic.) How does that square with tables 3 and 4 of the 1989 Seattle study, to name but one? Now you really are babbling (unless you really are incredibly dumb, I suspect you are either trolling or simply replying without reading what I actually say.) It is not like there is some subtle effect where wearing a helmet interacts with your brain in some way that changes the types of crashes you get into (that could be a "confounding factor" if it actually happened, but it doesn't.) So you say. The helmet researchers say otherwise. What you say they say and what they really say are two different things. I've seen you misquote or misrepresent things before. -- My real name backwards: nemuaZ lliB |
#213
|
|||
|
|||
On Sat, 28 May 2005 23:54:30 GMT, (Bill Z.)
wrote in message : You said "This "confounding factor" red herring you guys are now touting is just that." They said: it exists. "They" did not say it existed in what you posted. Actually they did. Especially the quote from Spaite, which regards it as a most significant finding. They did not say it existed in what you posted. Do you understand "what you posted" means? You know, the quote you actually provided? Are you now saying you quote one thing and then proceed to talk about something else? Yes, Bill, I understand what "what I posted" means. Moreover, I actually understand what I posted, which evidently you do not. "We conclude that helmet nonuse is strongly associated with severe injuries in this study population. This is true even when the patients without major head injuries are analyzed as a group" - A prospective analysis of injury severity among helmeted and non helmeted bicyclists involved in collisions with motor vehicles, Spaite DW, Murphy M, Criss EA, Valenzuela TD, Meislin HW, 1991. Journal of Trauma: 1991 Nov;31(11):1510-6 So: there is a difference between the accident involvement profiles of the two groups - in other words the confounding "red herring" exists. Have you read the paper in question? It is quite clear in context. "We cannot completely rule out the possibility that more cautious cyclists may have chosen to wear helmets and also had less severe accidents" - A case control study of the effectiveness of bicycle safety helmets, Thompson RS, Rivara FP, Thompson DC. 1989. New England Journal of Medicine: 1989 v320 n21 p1361-7 So: a group with a significant pro-helmet bias nonetheless acknowledge that the confounding "red herring" exists. Have you read the paper in question? It is very evident in their discussion. "[...] studies exclude bicyclists who had a head impact but no injury and include bicyclists who did not have a head impact but who suffered a non-head injury. Hence, these studies can either over-estimate or under-estimate the protective effect of helmets. [those which adjust for factors such as age, sex, riding conditions, speed, road surface and collision with motor vehicles may partly compensate for the inadequate design, [...] In this way they are more desirable than studies of Design 3 which make no effort to adjust for these potentially confounding factors." - Bicycle helmets - a review of their effectiveness: a critical review of the literature, Towner E, Dowswell T, Burkes M, Dickinson H, Towner J, Hayes M. 2002. Department for Transport: Road Safety Research Report 30 So: a reiew of various studies found that it was essential for the confounding "red herring" to be adequately controlled for, and notes that some studies fail even to try. Have you read the paper in question? It is even clearer when read in context. It seems to me that you are simply in denial. The existence of the confounding factor is documented in every recent paper on the subject that I can think of, and many of the older ones. All they really said is that you have to compare similar populations or similar accidents (which is kind of basic.) How does that square with tables 3 and 4 of the 1989 Seattle study, to name but one? Now you really are babbling (unless you really are incredibly dumb, I suspect you are either trolling or simply replying without reading what I actually say.) No, Bill, I read what you said. You said (rightly) that it is important to compare similar populations. How does that square, in your opinion, with the data in tables 3 and 4 of the 1989 Seattle study? It is not like there is some subtle effect where wearing a helmet interacts with your brain in some way that changes the types of crashes you get into (that could be a "confounding factor" if it actually happened, but it doesn't.) So you say. The helmet researchers say otherwise. What you say they say and what they really say are two different things. I've seen you misquote or misrepresent things before. So you say. Perhaps you could cite some studies which you say deny the existence of confounding? I'm having trouble thinking of any which explicitly deny it, and not many which ignore its existence, but I've only got a few hundred of them in my library. Guy -- May contain traces of irony. Contents liable to settle after posting. http://www.chapmancentral.co.uk 85% of helmet statistics are made up, 69% of them at CHS, Puget Sound |
#214
|
|||
|
|||
"Just zis Guy, you know?" writes:
On Sat, 28 May 2005 23:54:30 GMT, (Bill Z.) wrote in message : You said "This "confounding factor" red herring you guys are now touting is just that." They said: it exists. "They" did not say it existed in what you posted. Actually they did. Especially the quote from Spaite, which regards it as a most significant finding. They did not say it existed in what you posted. Do you understand "what you posted" means? You know, the quote you actually provided? Are you now saying you quote one thing and then proceed to talk about something else? Yes, Bill, I understand what "what I posted" means. Moreover, I actually understand what I posted, which evidently you do not. You may understand the sentence "what I posted" means, but what you said in "message 1" had nothing to do with your claim in "message 2" and further attempts to pretend otherwise are silly. "We conclude that helmet nonuse is strongly associated with severe injuries in this study population. This is true even when the patients without major head injuries are analyzed as a group" - A prospective analysis of injury severity among helmeted and non helmeted bicyclists involved in collisions with motor vehicles, Spaite DW, Murphy M, Criss EA, Valenzuela TD, Meislin HW, 1991. Journal of Trauma: 1991 Nov;31(11):1510-6 So: there is a difference between the accident involvement profiles of the two groups - in other words the confounding "red herring" exists. Have you read the paper in question? It is quite clear in context. Yawn. Keeping all other variables constant is "how to to research 101". It is quite different from the "confounding factor" red herring you brought up by citing medical research on heart disease, where there are complex interactions between various proteins, enzymes, etc. BTW, previously you anti-helmet people argued the exact opposite - that wearing helmets made people take higher risks (look up threads on "risk compensation".) Also, the date on the study you quote is 1991 - with the accidents being studies occuring earlier. Gee. You pick a time period where helmet was predominant mostly amoung "serious" cyclists and not more casual cyclists and you find that the casual cyclists make more serious mistakes and have worse crashes as a result? And you think this is surprising? Give us a break, Guy ... you are ranting about trivial nonsense and making a fool of yourself. -- My real name backwards: nemuaZ lliB |
#215
|
|||
|
|||
On Mon, 30 May 2005 19:14:16 GMT, (Bill Z.)
wrote in message : Yes, Bill, I understand what "what I posted" means. Moreover, I actually understand what I posted, which evidently you do not. You may understand the sentence "what I posted" means, but what you said in "message 1" had nothing to do with your claim in "message 2" and further attempts to pretend otherwise are silly. So you say. And yet you advance no credible reason to justify your refusal to believe what is written in the reports themselves, to whit: that the groups of helmeted and unhelmeted riders differed in more ways than just their helmet wearing rates. "We conclude that helmet nonuse is strongly associated with severe injuries in this study population. This is true even when the patients without major head injuries are analyzed as a group" - A prospective analysis of injury severity among helmeted and non helmeted bicyclists involved in collisions with motor vehicles, Spaite DW, Murphy M, Criss EA, Valenzuela TD, Meislin HW, 1991. Journal of Trauma: 1991 Nov;31(11):1510-6 So: there is a difference between the accident involvement profiles of the two groups - in other words the confounding "red herring" exists. Have you read the paper in question? It is quite clear in context. Yawn. Keeping all other variables constant is "how to to research 101". It is quite different from the "confounding factor" red herring you brought up by citing medical research on heart disease, where there are complex interactions between various proteins, enzymes, etc. Ah, so you /don't/ understand. Well why didn't you say? Oh but of course you don't: you flushed the posts where I explained it. So, to repeat the explanation I posted in terms you should be able to understand: Two interventions exist, I1 and I2, which are supposed to affect two health effects H1 and H2. Case-control studies S1 predict that intervention I1 will lead to reductions in adverse health effect H1 in the ratio R1. Case-control studies S2 predict that intervention I2 will lead to reductions in adverse health effect H2 in the ratio R2. Which is which? Time-series data T1 show that the rates of H1 do not track changes in intervention I1. Time-series data T2 show that the rates of H2 do not track changes in intervention I2. Which is which? In analysing the difference in benefit between T1 and S1, some researchers noted that voluntary uptake of I1 was strongly socio-economically stratified, and that those selecting for I1 were inherently less likely to exhibit H1 in the first place. In analysing the difference in benefit between T2 and S2, some researchers noted that voluntary uptake of I2 was strongly socio-economically stratified, and that those selecting for I2 were inherently less likely to exhibit H2 in the first place. Which is which? One of the interventions is combined HRT, the other cycle helmets. One of the health effects is CHD, the other head injuries. The two ratios R both vary considerably between studies. Using your skill and judgment, define which set of statements refers to HRT/CHD and which to helmets/HI. You have a 50% chance of being right. Oh, wait, one of them might be multivitamins reducing cancer. Or was it cannabis increasing schizophrenia? No, wait, maybe it was early use of antibiotics causing asthma in later life. Or was it MMR and autism? So many studies, all so similar. Finally, I1 (or is it I2?) is amenable to randomised controlled trials. RCT show that the ratio R1 (or was it R2?) predicted by case control studies is wrong as to both magnitude and direction. Lessons are drawn as to the advisability of believing case-control studies when they are contradicted by the other evidence. BTW, previously you anti-helmet people argued the exact opposite - that wearing helmets made people take higher risks (look up threads on "risk compensation".) Anti-helmet? Who? Where? And yes, the risk compensation effect is also documented. No wonder the small-scale studies have such absurd confidence intervals and fail to agree on the scale of the supposed benefit! Also, the date on the study you quote is 1991 - with the accidents being studies occuring earlier. Gee. You pick a time period where helmet was predominant mostly amoung "serious" cyclists and not more casual cyclists and you find that the casual cyclists make more serious mistakes and have worse crashes as a result? And you think this is surprising? And the cited text from Towner et. al. is dated 2002. But Bill - do you not understand that you have just conceded the point? Yes, the groups of helmeted and unhelmeted cyclists are different. That's what I said - that is what confounding means. For example, those from lower socio-economic groups are more likely to suffer head injury (and road traffic injury) than those from higher groups, independent of cycling. They are also less likely to wear helmets. Both injury and helmet use are strongly socio-economically stratified - and this is a confounding factor. So is the fact that risk-averse cyclists are more likely to wear helmets voluntarily. So is the fact that putting a helmet on a cyclists changes their risk-taking by some unmeasurable amount. These are all confounding factors! You say that the studies compare serious cyclists with non-serious - precisely! And to bring us back to the subject of the thread, the above is obvious from tables 3 and 4 in the 1989 Seattle study, but the original authors still cited that study *unmodified* in their 2002 Cochrane review (in which 70% of the cases reviewed were their own work - so much for independence!). If helmets prevent 85% of head injuries, as they still claimed in 2002, they also prevent 72% of broken legs, along with black skin, low income, riding on roads, riding alone rather than with families, female gender and being hit by cars. Or perhaps, as Frank and I have said all along, there are confounding factors in play. Confounding - self-selection bias to give it another common name - is inherent in observational case-control studies. Which is why Pettiti recommends the following approach to such data: o Do not turn a blind eye to contradictory evidence (such as the lack of response in head injury rates when helmet wearing increased by 40-50 percentage points following legislation) o Do not be seduced by mechanism. Even where a plausible mechanism exists, do not assume that we know everything about that mechanism and how it might interact with other factors. o Suspend belief. Of researchers defending observational studies, Pettiti says this: "Belief caused them to be unstrenuous in considering confounding as an explanation for the studies". o Maintain scepticism. See that last point? That's Frank and me, that is: we're sceptics. To a true believer a sceptic looks like an atheist, but that's the true believer's problem not the sceptics. Give us a break, Guy ... you are ranting about trivial nonsense and making a fool of yourself. So you say. Usual challenge: which text is sufficiently immoderate to be characterised as a rant? I anticipate the usual null response. It's funny, though, that you accuse me of making a fool of myself by posting quotes and citations from source data, when you are merely arm-waving and (it turns out) admitting the point while pretending to rebut it! Guy -- May contain traces of irony. Contents liable to settle after posting. http://www.chapmancentral.co.uk 85% of helmet statistics are made up, 69% of them at CHS, Puget Sound |
#216
|
|||
|
|||
"Just zis Guy, you know?" writes:
Before replying, I'll note that Guy is now trying to make a big deal about his "counfounding factors" (when it is really just a poor choice of samples) and specifically complaining about differences in helmeted versus non-helmeted cyclists. So, I'm going to give Guy a chance to prove his integrity. In a previous incarnation of this dicussion Tom Kunnich heaped accolades on a study by Paul Scuffham calling it a "watermark study" (see Message ID ). This study did not track which cyclists who were in accidents wore helmets and which did not (the data was simply not available from the sources Scuffham used). When such shortcomings were pointed out, the anti-helmet group went into overdrive denying it and calling me the usual assortment of names. Well, Guy, now is your chance. By our own argument, you should be willing to state that the praise heaped on this study was not in the least bit warranted, and that Tom Kunich, Frank Krygowski, and company were completely wrong. Will you do that? Or are you just arguing to push an agenda? Inquiring minds want to know. On Mon, 30 May 2005 19:14:16 GMT, (Bill Z.) wrote in message : Yes, Bill, I understand what "what I posted" means. Moreover, I actually understand what I posted, which evidently you do not. You may understand the sentence "what I posted" means, but what you said in "message 1" had nothing to do with your claim in "message 2" and further attempts to pretend otherwise are silly. So you say. And yet you advance no credible reason to justify your refusal to believe what is written in the reports themselves, to whit: that the groups of helmeted and unhelmeted riders differed in more ways than just their helmet wearing rates. Sigh. As I pointed out, to study helmet effectiveness, you *have* to pick two groups that differ only in their helmet-wearing rates. This is "How To Do An Experiment 101"." Yawn. Keeping all other variables constant is "how to to research 101". It is quite different from the "confounding factor" red herring you brought up by citing medical research on heart disease, where there are complex interactions between various proteins, enzymes, etc. Ah, so you /don't/ understand. Well why didn't you say? Oh but of course you don't: you flushed the posts where I explained it. So, to repeat the explanation I posted in terms you should be able to understand: Two interventions exist, I1 and I2, which are supposed to affect two health effects H1 and H2. snip Irrelevant. The term "confounding factors" (I even gave you a URL defining the term for you) applies to systematic errors. BTW, previously you anti-helmet people argued the exact opposite - that wearing helmets made people take higher risks (look up threads on "risk compensation".) Anti-helmet? Who? Where? As I said, look up the thread on risk compensation in the 1990s. And yes, the risk compensation effect is also documented. No wonder the small-scale studies have such absurd confidence intervals and fail to agree on the scale of the supposed benefit! Well, as I said, you were just arguing the opposite. Also, the date on the study you quote is 1991 - with the accidents being studies occuring earlier. Gee. You pick a time period where helmet was predominant mostly amoung "serious" cyclists and not more casual cyclists and you find that the casual cyclists make more serious mistakes and have worse crashes as a result? And you think this is surprising? And the cited text from Towner et. al. is dated 2002. But Bill - do you not understand that you have just conceded the point? No I didn't conced the point. They simply failed to pick a representative sample and found out after they had done a lot of work, and figured they'd rather try to publish than perish. At least, it served as a warning about a factor you need to control for. Yes, the groups of helmeted and unhelmeted cyclists are different. That's what I said - that is what confounding means. No it doesn't mean that. There is no causal affect that makes non-helmeted riders more accident prone than helmeted riders. If (before helmets became popular) you took a sufficiently large random sample of equally skilled cyclists, and gave a reasonable fraction of them helmets, your "difference" would go away. For example, those from lower socio-economic groups are more likely to suffer head injury (and road traffic injury) than those from higher groups, independent of cycling. They are also less likely to wear helmets. So what? You just don't compare unskilled cyclists riding on quiet residential streets wearing helmets to simlarly unskilled cyclists riding on busy urban streets in low income areas where unlicensed drivers and cars with ill maintained brakes are far more common. You say that the studies compare serious cyclists with non-serious - precisely! I said those studies are inherently unreliable. Confounding - self-selection bias to give it another common name - is inherent in observational case-control studies. Which is why Pettiti recommends the following approach to such data: I posted a definition for you and you are ignoring it. Here it is again - http://www.sysurvey.com/tips/statistics/confounding.htm. Or to quote, "In a well designed psychology experiment an investigator will randomly assign subjects to two or more groups and except for differences in the experimental procedure applied to each group, the groups will be treated exactly alike. Under these circumstances any differences between the groups that are statistically significant are attributed to differences in the treatment conditions. This, of course assumes that except for the various treatment conditions the groups were, in fact, treated exactly alike. Unfortunately, however, It is always possible that despite an experimenters best intentions there was some unsuspected systematic differences in the way the groups were treated in addition to the intended treatment conditions. Statisticians describe systematic differences of this sort as confounding factors or confounding variables." Note the first sentence. If you are not doing that, you don't have a well-designed experiment in the first place. You should expect bogus results, and those bogus results are not the result of "confounding factors" but a poorly designed experiment. rest of Guys's rant snipped -- My real name backwards: nemuaZ lliB |
#217
|
|||
|
|||
On Wed, 01 Jun 2005 05:39:39 GMT, (Bill Z.)
wrote: Before replying, I'll note that Guy is now trying to make a big deal about his "counfounding factors" (when it is really just a poor choice of samples) and specifically complaining about differences in helmeted versus non-helmeted cyclists. Who would have thought it? When discussing confounding factors, I make a big deal about confounding factors! Amazing. No, Bill, it's not "poor choice of sample", it's confounding. The profiles of the helmeted and unhelmeted riders in the studies are different: both helmet use and injury are socio-economically stratified (there is plenty of evidence for that), so the difference is not down to poor choice of samples, it is inherent in the study populations. There is poor choice of samples, too, of course. For example, the 1989 Seattle study (still the most widely-quoted study) compared a "case" group which was more likely to be poor, male, black or Hispanic, riding unaccompanied on city streets, with a negligible helmet wearing rate; with a "control" group which was predominantly white, mostly female, middle-class, riding on off-road bike trails, with a helmet wearing rate in the 20s percent - and attributed all the difference in injury profile to helmet use. You can play all sorts of games with the data from this study - you can show from the same data that the helmeted riders were many times more likely to crash, and that the protective effect against broken legs was about the same as for head injuries, or you can substitute the helmet wearing rate measured by co-author Rivara in contemporaneous street counts and see the benefit vanish. Whatever, it's all evidence of confounding factors. So, I'm going to give Guy a chance to prove his integrity. In a previous incarnation of this dicussion Tom Kunnich heaped accolades on a study by Paul Scuffham calling it a "watermark study" (see Message ID ). This study did not track which cyclists who were in accidents wore helmets and which did not (the data was simply not available from the sources Scuffham used). Keep beating, Bill, there's still a vestige of the bloody smear where that dead horse used to be! When such shortcomings were pointed out, the anti-helmet group went into overdrive denying it and calling me the usual assortment of names. What anti-helmet people? Name them. Well, Guy, now is your chance. By our own argument, you should be willing to state that the praise heaped on this study was not in the least bit warranted, and that Tom Kunich, Frank Krygowski, and company were completely wrong. Will you do that? Or are you just arguing to push an agenda? Inquiring minds want to know. Hmmmm. Tricky. Scuffham in 1997: "Results revealed that the increased helmet wearing percentages has had little association with serious head injuries to cyclists as a percentage of all serious injuries to cyclists for all three groups, with no apparent difference between bicycle only and all cycle crashes." Scuffham in 2000: "We conclude that the helmet law has been an effective road safety intervention that has lead to a 19% (90% CI: 14%, 23%) reduction in head injury to cyclists over its first three years." How to account for the difference? Aha! Figure 3 in the 2000 study shows the problem. Looking at figure 3 we see a steadily (and uniformly) declining trend over time running from 1988 to the most current data used in the study, 1997. There is a lot of noise due to the very small sample sizes used, but the trend is very clear and can be accounted for using statistical regression techniques. What happens to the 19% figure when the regression techniques are applied? It becomes zero within the limits of sampling error. The 19% figure is clearly the result of poor selection of data points. Mystery solved. So, Bill, having read both studies (and two other studies with Scuffham as co-author discussing the same subject), I fully understand what is going on, and I am not going to play your silly game. This study is, of course, independent of the discussion of confounding factors inherent in observational case-control studies, as documented in the International Journal of Epidemiology. But it does make a striking comment relevant to that discussion: it notes that the predictions from such studies are enormously higher than predicted by the case-control studies. Guess which figure it uses for the predicted reduction of brain injury? 88%! The 1989 Seattle study, by then widely criticised in press for its unrepresentative "control" group and failure to adequately account for confounding. Well, well. you advance no credible reason to justify your refusal to believe what is written in the reports themselves, to whit: that the groups of helmeted and unhelmeted riders differed in more ways than just their helmet wearing rates. Sigh. As I pointed out, to study helmet effectiveness, you *have* to pick two groups that differ only in their helmet-wearing rates. This is "How To Do An Experiment 101"." And is impossible to achieve in practice. Why do you have such difficulty accepting the study authors' own statement that there are differences between the helmeted and unhelmeted populations? Two interventions exist, I1 and I2, which are supposed to affect two health effects H1 and H2. Irrelevant. The term "confounding factors" (I even gave you a URL defining the term for you) applies to systematic errors. That is *a* definition, not *the* definition. As Pettiti says of those who believed another set of case-control studies with similar socio-economic differences between case and control groups: "belief caused them to be unstrenuous in considering confounding as an explanation for the studies". You are playing word games, Bill, and badly at that. BTW, previously you anti-helmet people argued the exact opposite - that wearing helmets made people take higher risks (look up threads on "risk compensation".) Anti-helmet? Who? Where? As I said, look up the thread on risk compensation in the 1990s. Been there, done that. Again, name names. Who do you consider to be anti-helmet? It is not obvious to me from that context. And yes, the risk compensation effect is also documented. No wonder the small-scale studies have such absurd confidence intervals and fail to agree on the scale of the supposed benefit! Well, as I said, you were just arguing the opposite. No, I said that both factors come into play and both are pretty much unmeasurable. The 2003 study by Towner et. al. discusses both phenomena. You seem to be trying to claim that only one factor can influence cyclists at a time. Also, the date on the study you quote is 1991 And the cited text from Towner et. al. is dated 2002. But Bill - do you not understand that you have just conceded the point? No I didn't conced the point. They simply failed to pick a representative sample and found out after they had done a lot of work, and figured they'd rather try to publish than perish. At least, it served as a warning about a factor you need to control for. No, Bill, what you said was: "You pick a time period where helmet was predominant mostly among "serious" cyclists and not more casual cyclists and you find that the casual cyclists make more serious mistakes and have worse crashes as a result" That is precisely what was being documented! Can you think of a case-control study which does not cover a period where helmet use was more prevalent among serious cyclists and less prevalent among leisure cyclists? It's true for all of them! Yes, the groups of helmeted and unhelmeted cyclists are different. That's what I said - that is what confounding means. No it doesn't mean that. You are wrong. This definition of confounding is used in the IJE review of observational case-control studies. There is no causal affect that makes non-helmeted riders more accident prone than helmeted riders. If (before helmets became popular) you took a sufficiently large random sample of equally skilled cyclists, and gave a reasonable fraction of them helmets, your "difference" would go away. There doesn't need to be a causal relationship, all it requires is that unhelmeted riders are less risk-averse than helmeted riders. Both head injury and helmet use are socio-economically stratified. In the one case the relationship may be causal (helmets cost money), in the other the two factors are likely common effects of a joint cause. It doesn't matter: the effect exists nonetheless. Where case-control studies have recorded the socio-economic background of riders, they have documented this. Spaite regards the differing injury profiles of helmeted and unhelmeted riders as one of his most significant findings. Any way you slice it, in a population where helmet use is voluntary, the helmeted and unhelmeted communities are likely to be different in more ways than just helmet use. And in a population where it is compulsory, we find that there is no effective difference between helmet use and nonuse in those who do not wear them voluntarily. For example, those from lower socio-economic groups are more likely to suffer head injury (and road traffic injury) than those from higher groups, independent of cycling. They are also less likely to wear helmets. So what? You just don't compare unskilled cyclists riding on quiet residential streets wearing helmets to simlarly unskilled cyclists riding on busy urban streets in low income areas where unlicensed drivers and cars with ill maintained brakes are far more common. You're right to say you /shouldn't/ compare these groups, but the most widely quoted study /does/ and so do others. You say that the studies compare serious cyclists with non-serious - precisely! I said those studies are inherently unreliable. OK, so you consider observational case-control studies to be unreliable due to underlying differences between the populations, but you don't want to call it confounding, and you don't want to say that it is the differences between populations which make them unreliable, and you're not going to let anyone say that their conclusions are wrong. That's pretty mixed up, in my view. Confounding - self-selection bias to give it another common name - is inherent in observational case-control studies. Which is why Pettiti recommends the following approach to such data: I posted a definition for you and you are ignoring it. Here it is again Yes, you posted *a* definition. It is not the only one. It is always possible that despite an experimenters best intentions there was some unsuspected systematic differences in the way the groups were treated in addition to the intended treatment conditions. Statisticians describe systematic differences of this sort as confounding factors or confounding variables." The confounding, you see, could be in the composition of the groups just as well as in the treatment of them. Simple, isn't it? The operative usage of "confounding" is the primary one in my dictionary, meaning to cause confusion. Here's another definition for you: 1. A situation in which the effects of two or more processes are not separated; the distortion of the apparent effect of an exposure on risk, brought about by the association with other factors that can influence the outcome. 2. A relationship between the effects of two or more causal factors observed in a set of data, such that it is not logically possible to separate the contribution of any single causal factor to the observed effects. Describes it perfectly. Taken from http://cancerweb.ncl.ac.uk/cgi-bin/o...ion=Search+OMD Other definitions exist too. Note the first sentence. If you are not doing that, you don't have a well-designed experiment in the first place. Correct. Oh, so very correct. And yet you persist in believing the conclusions of those poorly-designed experiments and repudiating any data which disputes them! A mystery. rest of Guys's rant snipped Usual challenge: detail the text sufficiently immoderate to be characterised as a rant. I wonder who you think you are fooling? Guy -- May contain traces of irony. Contents liable to settle after posting. http://www.chapmancentral.co.uk 88% of helmet statistics are made up, 65% of them at CHS, Puget Sound |
#218
|
|||
|
|||
On Wed, 01 Jun 2005 11:20:37 +0100, "Just zis Guy, you know?"
wrote: D'oh! Timelapse due to checking sources. it notes that the predictions from such studies are enormously higher than predicted by the case-control studies. should read: it notes that the predictions from such studies are enormously lower than those found in the sample population. But this is obvious from context, and you wouldn't be arguing the toss without your copies of the two studies in front of you, would you? Guy -- May contain traces of irony. Contents liable to settle after posting. http://www.chapmancentral.co.uk 88% of helmet statistics are made up, 65% of them at CHS, Puget Sound |
#219
|
|||
|
|||
"Just zis Guy, you know?" writes:
On Wed, 01 Jun 2005 05:39:39 GMT, (Bill Z.) wrote: Before replying, I'll note that Guy is now trying to make a big deal about his "counfounding factors" (when it is really just a poor choice of samples) and specifically complaining about differences in helmeted versus non-helmeted cyclists. Who would have thought it? When discussing confounding factors, I make a big deal about confounding factors! Amazing. No, Bill, it's not "poor choice of sample", it's confounding. No, it is a poor choice of samples. If you want to study the effects of helmets in such studies, you want the helmeted versus non-helmeted cyclists to be the same in all other respects. The profiles of the helmeted and unhelmeted riders in the studies are different: both helmet use and injury are socio-economically stratified (there is plenty of evidence for that), so the difference is not down to poor choice of samples, it is inherent in the study populations. No it isn't inherent. red herring snipped So, I'm going to give Guy a chance to prove his integrity. In a previous incarnation of this dicussion Tom Kunnich heaped accolades on a study by Paul Scuffham calling it a "watermark study" (see Message ID ). This study did not track which cyclists who were in accidents wore helmets and which did not (the data was simply not available from the sources Scuffham used). Keep beating, Bill, there's still a vestige of the bloody smear where that dead horse used to be! Uh huh. In other words, you refuse to apply the same standard to all studies. When such shortcomings were pointed out, the anti-helmet group went into overdrive denying it and calling me the usual assortment of names. Well, Guy, now is your chance. By our own argument, you should be willing to state that the praise heaped on this study was not in the least bit warranted, and that Tom Kunich, Frank Krygowski, and company were completely wrong. Will you do that? Or are you just arguing to push an agenda? Inquiring minds want to know. Hmmmm. Tricky. Scuffham in 1997: "Results revealed that the increased helmet wearing percentages has had little association with serious head injuries to cyclists as a percentage of all serious injuries to cyclists for all three groups, with no apparent difference between bicycle only and all cycle crashes." Scuffham in 2000: "We conclude that the helmet law has been an effective road safety intervention that has lead to a 19% (90% CI: 14%, 23%) reduction in head injury to cyclists over its first three years." How to account for the difference? Aha! Figure 3 in the 2000 study shows the problem. Except Tom's comments were about a study performed before 1997, much less before 2001. Now, what is your opinion of the level of praise heaped on Scuffham's earlier work? Looking at figure 3 we see a steadily (and uniformly) declining trend over time running from 1988 to the most current data used in the study, 1997. There is a lot of noise due to the very small sample sizes used, but the trend is very clear and can be accounted for using statistical regression techniques. What happens to the 19% figure when the regression techniques are applied? It becomes zero within the limits of sampling error. The 19% figure is clearly the result of poor selection of data points. Gee. When I pointed out the small sample size previously, the others whined loudly. Now's your chance to criticize Tom and Frank. Will you? Mystery solved. Nope. You are weaseling out of the question. Will you now criticize the unwarranted praise Tom, Frank, and others heaped on Scuffham's earlier study, particularly the way these anti-helmet usenet posters tried to claim this study "proved" that helmets don't work. I repeatedly pointed out that a null result due to a small sample says nothing about helmets and they howled and howled about that. you advance no credible reason to justify your refusal to believe what is written in the reports themselves, to whit: that the groups of helmeted and unhelmeted riders differed in more ways than just their helmet wearing rates. Sigh. As I pointed out, to study helmet effectiveness, you *have* to pick two groups that differ only in their helmet-wearing rates. This is "How To Do An Experiment 101"." And is impossible to achieve in practice. Why do you have such difficulty accepting the study authors' own statement that there are differences between the helmeted and unhelmeted populations? I'm not objecting to the authors saying that. I'm objecting to your spin on it. lines and lines of repetitive garbage snipped. -- My real name backwards: nemuaZ lliB |
#220
|
|||
|
|||
Bill Z. wrote: "Just zis Guy, you know?" writes: Looking at figure 3 we see a steadily (and uniformly) declining trend over time running from 1988 to the most current data used in the study, 1997. There is a lot of noise due to the very small sample sizes used, but the trend is very clear and can be accounted for using statistical regression techniques. What happens to the 19% figure when the regression techniques are applied? It becomes zero within the limits of sampling error. The 19% figure is clearly the result of poor selection of data points. Gee. When I pointed out the small sample size previously, the others whined loudly. Now's your chance to criticize Tom and Frank. Will you? Such dishonesty! You had claimed the entire population of New Zealand was too small to show any benefit for helmets. That's not at all what Guy is saying above. Guy is pointing out that there is noise, but he is OBVIOUSLY not saying that nothing can be determined, as you claimed. Note the words "the trend is very clear and can be accounted for using linear regression techniques." That means you can tell what's going on, despite the noise. Feel free to be rude, Bill - it's your nature. But don't be dishonest. When you stoop to that, you're no longer fun to read! And I'm sure you don't want to deprive others of the entertainment you generously provide! ;-) - Frank Krygowski |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Helmet propaganda debunked | [email protected] | Racing | 17 | April 27th 05 04:34 PM |
Ontario Helmet Law being pushed through | Chris B. | General | 1379 | February 9th 05 04:10 PM |
published helmet research - not troll | Frank Krygowski | Social Issues | 1716 | October 24th 04 06:39 AM |
Reports from Sweden | Garry Jones | Social Issues | 14 | October 14th 03 05:23 PM |
Helmet Advice | DDEckerslyke | Social Issues | 17 | September 2nd 03 11:10 PM |