When I moved here I chose an apartment specifically because the bus that goes directly to my workplace stops in front. The commute is about 35 minutes door to door on the bus, which gives me just enough time to read a bit, or do some useful work. Driving would probably take me about 20 minutes, but then I'd have to spend half an hour looking for a parking place. I could live closer to work, but here I have a kayak tied up across the street.
"You can't bike when its 40 below zero wind chill, or on snow and ice."
Americans are such wusses.;)
I used to bike when it was -40 before wind chill. Of course, you have to take it easy to avoid breaking bits off your bike. Unfortunately now that I've moved somewhere warmer (minimum temperature is about -20 C) I don't bike anymore because apparently bikes get stolen almost instantaneously if you park them downtown.
The AC claimed that the active ingredient in red yeast rice is "essentially the same" as statins. He was replying to a post about how statins may be dangerous.
So to answer your questions:
1) as far as the AC is concerned they are the same 2) I was replying to a post that asserted there are no real differences 3) See #1 and 2 4) No, most supplements are reasonably safe for the simple reason that they don't do anything. There's lots of research showing that. Some are dangerous, particularly for some people, because they DO have active ingredients in them, whether it's ginseng, ginkgo balboa or spinach interfering with someone on warfarin, people overdosing on fat soluble vitamins, someone working up to the LD50 on caffeine pills or, if the AC is right, statin exposure from eating red yeast rice.
Yes, that's the point. You don't look at a ten story high poster from three feet away. You also don't look at an 11x14 in your lap. Six megapixels seems to provide sufficient resolution that the angular size of a pixel is acceptably small for any reasonable viewing configuration.
People who say you need X megapixels for a Y x Z enlargement are silly. If I'm going to look at a 4x6 from an inch away with a magnifying glass I need lots of resolution too, but nobody would do that. Just like nobody would try to look at a ten story poster from three feet.
I DID see a 6 megapixel image used for a commercial, building-sized poster. It DID look fine. No, I wasn't three feet away. That would have been silly.
Experiments do establish the confidence of a causal relationship, and that confidence can be made arbitrarily close to one (or zero). Although you might argue, as does Hume, that it's philosophically impossible to actually observe causation with complete confidence, the same argument can be made for ANY observation, of anything.
"Remember, though, that randomization does not guarantee that potential confounding variables are controlled."
Proper randomization diffuses the effects of uncorrected confounders randomly through the data so any bias is eliminated. That's why it's called randomization. What you can't control, you randomize. The effect statistically is the same - you eliminate those variables as factors.
But the critical point is that in an experiment you manipulate one variable (while controlling or randomizing all the others). You MUST manipulate a variable to show a causal relationship, and experimental trials do this. Observational trials do not, and so can only show correlation. Experiments DO show causation.
Correlation isn't bad. Finding a correlation narrows things down to three possible causal relationships - A causes B, B causes A or C causes A and B, and one of those relationships is often ridiculous. People who say things like "correlation is not causation" usually don't understand this. BUT, they are correct, the only way to show a causal relationship is with an actual experiment.
I agree. It's just that the "positive publication bias" thing is a pet peeve of mine. People who talk about it are usually under the impression that "no correlation" or "not significant" is a negative result. In my experience it's not that hard to get actual negative results published, but very few people produce actual negative results, which require estimating what a meaningful effect size would be and extra stats (plus usually extra power) to show that the true effect is confidently smaller. Producing negative results is considerably harder than positive.
There IS a bias against publishing inconclusive results. I don't see a problem with that. A separate, and unrelated problem, is that people don't do stats correctly, and one of the biggest offenses is not doing multiple comparison correction.
No, you're falling into the same trap. Don't elect a party. Elect a representative. That representative may be generally associated with a party (or not), but he should vote in the interest of his constituents, not his party.
THAT is what I mean when I say that if people, including in the UK, understood that they elect a representative, not a party, they'd get a better government. It solves all the problems you mention, except the very local issue where you might have several candidates who split the vote.
At the local level, a representative elected AS a representative of his constituents (and not a voting machine for the party) will tend to represent them all, whether or not they voted for him. Yes, you can fiddle with the actual voting mechanics, such as allowing votes against a candidate or ranking, and improve things, but at the cost of complexity. But that fiddling is for naught as long as people are still voting for parties or party leaders in the first place.
Yes. It's not really a "whitepaper" at least not as we used to know them. It's a long form marketing piece with some rough technical details in it. Are you referring to something in particular?
So your position is that drugs that may be dangerous when purified and given in precise doses should instead be given in highly variable doses in poorly regulated supplements?
"they are used as evidence that a real study should be done,"
If possible. Sometimes it's not. The authors give the example of smoking - there are no randomized trials showing that smoking is bad for you. All the evidence is observational, backed up by knowledge of a bunch of likely mechanisms. In general, if you suspect that some factor is detrimental, you cannot do a randomized study because giving people something you suspect will hurt them is unethical.
Commenting on the article you linked to, in my experience it's not a failure to recognize that you have to compare the magnitude of differences directly. It's a bigger problem. People, including researchers in many fields, have the strange idea that "not significantly different" means "no difference." The problem is probably second only to the habit of not correcting for multiple comparisons.
No. When you do an experiment, i.e. purposely manipulate one variable, you establish a causal connection. Identifying and explaining the mechanism is nice, and establishes the character and directness of your causal relationship. Trials are experiments.
Correlation comes from observational studies where you do not manipulate any variables yourself, you just look for natural or preexisting variation.
A simplified example - if I look at a bunch of people who take sleeping pills and a bunch who don't, and measure how likely they are to die, I get a correlation (maybe) - dying and taking sleeping pills are correlated, but I don't know if dying causes people to take sleeping pills, whether sleeping pills tend to cause you to die, or whether some other factor (being crazy maybe) causes you to both take sleeping pills and die.
If I take a bunch of random people and give some sleeping pills and others no sleeping pills, if the ones I give the pills die significantly more often then I can conclude that sleeping pills cause death (by some mechanism I don't yet know).
"This is further biased by the fact that only positive results are reported - no one writes of all the "no correlation" results they may have found through different choices of matched sets."
If you're honest with your stats you multiply your p-value by the number of comparisons you did. Yes, some of us do this. There's nothing wrong with retrospective analyses, it's just that so many people do the stats incorrectly.
The "whitepaper" is marketing. "Oversampling" is just pixel averaging, and it has well known effects. You're better off having more pixel area and NOT averaging.
The sensor resolution does give them flexibility to do things like not include an optical zoom, etc. but the image quality will not be better than a sensor with fewer, bigger pixels.
They do get nice images, because they've put a big lens and a big sensor in a cell phone. But if you believe their claims that because of some marketing speak you're going to get results like a professional shooter with a kilo and a half, $2000 can of precision fluorite optics, you're dreaming.
"digital zooming" is just in camera cropping. If your sensor is exceeding the resolving power of your lens your digital zoom is just going to be blurry pixels. You can put a nice big number on the box though.
Silly sensor resolution is silly whether you use it for fake zooming or not.
Slashdot Mirror
When I moved here I chose an apartment specifically because the bus that goes directly to my workplace stops in front. The commute is about 35 minutes door to door on the bus, which gives me just enough time to read a bit, or do some useful work. Driving would probably take me about 20 minutes, but then I'd have to spend half an hour looking for a parking place. I could live closer to work, but here I have a kayak tied up across the street.
I want to live on a beach in the tropics and work in Canada. It's untenable. Can you get to work on fighting to make that possible for me?
Europe is pretty big too. But people there don't live so far away from where they work.
"You can't bike when its 40 below zero wind chill, or on snow and ice."
Americans are such wusses. ;)
I used to bike when it was -40 before wind chill. Of course, you have to take it easy to avoid breaking bits off your bike. Unfortunately now that I've moved somewhere warmer (minimum temperature is about -20 C) I don't bike anymore because apparently bikes get stolen almost instantaneously if you park them downtown.
The AC claimed that the active ingredient in red yeast rice is "essentially the same" as statins. He was replying to a post about how statins may be dangerous.
So to answer your questions:
1) as far as the AC is concerned they are the same
2) I was replying to a post that asserted there are no real differences
3) See #1 and 2
4) No, most supplements are reasonably safe for the simple reason that they don't do anything. There's lots of research showing that. Some are dangerous, particularly for some people, because they DO have active ingredients in them, whether it's ginseng, ginkgo balboa or spinach interfering with someone on warfarin, people overdosing on fat soluble vitamins, someone working up to the LD50 on caffeine pills or, if the AC is right, statin exposure from eating red yeast rice.
Yes, that's the point. You don't look at a ten story high poster from three feet away. You also don't look at an 11x14 in your lap. Six megapixels seems to provide sufficient resolution that the angular size of a pixel is acceptably small for any reasonable viewing configuration.
People who say you need X megapixels for a Y x Z enlargement are silly. If I'm going to look at a 4x6 from an inch away with a magnifying glass I need lots of resolution too, but nobody would do that. Just like nobody would try to look at a ten story poster from three feet.
I DID see a 6 megapixel image used for a commercial, building-sized poster. It DID look fine. No, I wasn't three feet away. That would have been silly.
Wikipedia has a reasonable section on this:
http://en.wikipedia.org/wiki/Correlation_does_not_imply_causation#Determining_causation
Experiments do establish the confidence of a causal relationship, and that confidence can be made arbitrarily close to one (or zero). Although you might argue, as does Hume, that it's philosophically impossible to actually observe causation with complete confidence, the same argument can be made for ANY observation, of anything.
"Remember, though, that randomization does not guarantee that potential confounding variables are controlled."
Proper randomization diffuses the effects of uncorrected confounders randomly through the data so any bias is eliminated. That's why it's called randomization. What you can't control, you randomize. The effect statistically is the same - you eliminate those variables as factors.
But the critical point is that in an experiment you manipulate one variable (while controlling or randomizing all the others). You MUST manipulate a variable to show a causal relationship, and experimental trials do this. Observational trials do not, and so can only show correlation. Experiments DO show causation.
Correlation isn't bad. Finding a correlation narrows things down to three possible causal relationships - A causes B, B causes A or C causes A and B, and one of those relationships is often ridiculous. People who say things like "correlation is not causation" usually don't understand this. BUT, they are correct, the only way to show a causal relationship is with an actual experiment.
Do the math.
I agree. It's just that the "positive publication bias" thing is a pet peeve of mine. People who talk about it are usually under the impression that "no correlation" or "not significant" is a negative result. In my experience it's not that hard to get actual negative results published, but very few people produce actual negative results, which require estimating what a meaningful effect size would be and extra stats (plus usually extra power) to show that the true effect is confidently smaller. Producing negative results is considerably harder than positive.
There IS a bias against publishing inconclusive results. I don't see a problem with that. A separate, and unrelated problem, is that people don't do stats correctly, and one of the biggest offenses is not doing multiple comparison correction.
No, you're falling into the same trap. Don't elect a party. Elect a representative. That representative may be generally associated with a party (or not), but he should vote in the interest of his constituents, not his party.
THAT is what I mean when I say that if people, including in the UK, understood that they elect a representative, not a party, they'd get a better government. It solves all the problems you mention, except the very local issue where you might have several candidates who split the vote.
At the local level, a representative elected AS a representative of his constituents (and not a voting machine for the party) will tend to represent them all, whether or not they voted for him. Yes, you can fiddle with the actual voting mechanics, such as allowing votes against a candidate or ranking, and improve things, but at the cost of complexity. But that fiddling is for naught as long as people are still voting for parties or party leaders in the first place.
Yes. It's not really a "whitepaper" at least not as we used to know them. It's a long form marketing piece with some rough technical details in it. Are you referring to something in particular?
You let Google Analytics right on through, hey?
"Did you ever wonder how they come up with death rates that are less than 100%?"
By making the denominator in the rate something like days, years, decades or centuries instead of multiple centuries or millennia?
Do you understand what a rate is?
So your position is that drugs that may be dangerous when purified and given in precise doses should instead be given in highly variable doses in poorly regulated supplements?
Um, so what? Confidence intervals overlap all the time. In fact, you can always make them overlap if you pick the right ones.
Besides, I'm having trouble finding which ones concern you. All the CIs I can find are on hazard ratios.
"Coumadin, Warfarin" which are the same thing (warfarin is the generic name). The common name is "rat poison."
"they are used as evidence that a real study should be done,"
If possible. Sometimes it's not. The authors give the example of smoking - there are no randomized trials showing that smoking is bad for you. All the evidence is observational, backed up by knowledge of a bunch of likely mechanisms. In general, if you suspect that some factor is detrimental, you cannot do a randomized study because giving people something you suspect will hurt them is unethical.
Commenting on the article you linked to, in my experience it's not a failure to recognize that you have to compare the magnitude of differences directly. It's a bigger problem. People, including researchers in many fields, have the strange idea that "not significantly different" means "no difference." The problem is probably second only to the habit of not correcting for multiple comparisons.
No. When you do an experiment, i.e. purposely manipulate one variable, you establish a causal connection. Identifying and explaining the mechanism is nice, and establishes the character and directness of your causal relationship. Trials are experiments.
Correlation comes from observational studies where you do not manipulate any variables yourself, you just look for natural or preexisting variation.
A simplified example - if I look at a bunch of people who take sleeping pills and a bunch who don't, and measure how likely they are to die, I get a correlation (maybe) - dying and taking sleeping pills are correlated, but I don't know if dying causes people to take sleeping pills, whether sleeping pills tend to cause you to die, or whether some other factor (being crazy maybe) causes you to both take sleeping pills and die.
If I take a bunch of random people and give some sleeping pills and others no sleeping pills, if the ones I give the pills die significantly more often then I can conclude that sleeping pills cause death (by some mechanism I don't yet know).
"This is further biased by the fact that only positive results are reported - no one writes of all the "no correlation" results they may have found through different choices of matched sets."
If you're honest with your stats you multiply your p-value by the number of comparisons you did. Yes, some of us do this. There's nothing wrong with retrospective analyses, it's just that so many people do the stats incorrectly.
The "whitepaper" is marketing. "Oversampling" is just pixel averaging, and it has well known effects. You're better off having more pixel area and NOT averaging.
The sensor resolution does give them flexibility to do things like not include an optical zoom, etc. but the image quality will not be better than a sensor with fewer, bigger pixels.
They do get nice images, because they've put a big lens and a big sensor in a cell phone. But if you believe their claims that because of some marketing speak you're going to get results like a professional shooter with a kilo and a half, $2000 can of precision fluorite optics, you're dreaming.
I'm glad you put photographer in quotes.
"digital zooming" is just in camera cropping. If your sensor is exceeding the resolving power of your lens your digital zoom is just going to be blurry pixels. You can put a nice big number on the box though.
Silly sensor resolution is silly whether you use it for fake zooming or not.
Meh, I saw a 6 MP image from an early DSLR blown up to ten story building size once. It looked fine.