Re: The False Placebo Effect
Robert J. MacG. Dawson <[EMAIL PROTECTED]> wrote: YES : Elliot Cramer wrote: :> I believe the point of the Danes :> was that a placebo should be used in :> research but that physicians should : ?"not"? :> think that they can "cure" people with :> placebos; I agree. : -Robert Dawson : = : Instructions for joining and leaving this list and remarks about : the problem of INAPPROPRIATE MESSAGES are available at : http://jse.stat.ncsu.edu/ : = = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: The False Placebo Effect
J. Williams wrote: : Correct me if I'm wrong, but as I understand it, you have the ability : to alter your diastolic reading by +/- 20 mm Hg for 3 minutes No; I said I raised it once. I doubt that it lasted long. All sorts of things raise blood pressure temporarily. I'm told that meditation lowers it and I wouldn't be surprised. BP is notoriously variable. I doubt if there : would be a statistically significant difference between a placebo : treatment and a control (no-treatment) vis a vis the diastolic reading I disagree. Of course there NEVER is NO TREATMENT; you just don't know what else is going on. A randomized placebo study controls for something one thinks is important and randomizes everything else, exactly the concept that Fisher first introduced. : As I understand your position, you maintain the diastolic readings may : be subjective as well and can be "willed" up or down even in a : controlled lab setting. Certainly up and probably down (but I havn't done a controlled experiment) I believe the point of the Danes was that a placebo should be used in research but that physicians should think that they can "cure" people with placebos; I agree. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: The False Placebo Effect
On Sun, 27 May 2001, Rich Ulrich wrote: > > I don't see how RTM can explain the average change in a prepost design > > - explanation: whole experiment is conducted on patients > who are at their *worst* because the flare-up is what sent > them to a doctor. ok > - I'm not sure what that last phrase means... "both " > 30% or so of acutely depressed patients will get quite a bit better. depressions are self-limiting; people get better unless they kill themselves > The experience of being in a research trial, by the way, seems > to produce a placebo effect, according to what people have told me. > (I think that careful scientists attribute that one to the extra time > and attention given to those subjects.) This is the historic psychological explanation. The interest in any experiment is not a comparison with what the S's were before but with whatt they would have been like absent the intervention ie with a placebo (or alternate treatment) = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: The False Placebo Effect
J. Williams wrote: My hunch : is the placebo group would not differ significantly on the diastolic : reading from the no-treatment group. Even though the placebo patients : "think" they are being treated, I wager they can't "fake" a diastolic : reading. It isn't a question of faking. A basic prnciple of experimental design going back to Fisher is to control the important variables than might affect your results. Giving someone a pill with the expectation that it will help them is such a variable. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: The False Placebo Effect
J. Williams wrote: In article <[EMAIL PROTECTED]> you wrote: :>: do you suppose a person receiving a placebo can actually :>: change his/her diastolic reading? :> :>sure; I raised mine 20 points yesterday just thinking about someone :>misusing statistics. cholesterol is another thing. :>just sitting for 3 minutes before testing will lower it : Are you sure you're not thinking about your systolic reading? No; I raised both. I've been checking it regularly and it has been averaging 131/72. I measured it 180/90. an hour later it was back down I seriously doubt if someone misusing statistics : could hike your diastolic reading by 20 mm Hg :-)) If so, get : treatment---fast--before you stroke out. I take statistics seriously. I believe that the standards you quote are for average reading over time. I doubt that I'm in danger = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: The False Placebo Effect
J. Williams wrote: : On 25 May 2001 19:39:50 GMT, Elliot Cramer <[EMAIL PROTECTED]> : wrote: : do you suppose a person receiving a placebo can actually : change his/her diastolic reading? sure; I raised mine 20 points yesterday just thinking about someone misusing statistics. cholesterol is another thing. just sitting for 3 minutes before testing will lower it = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: The False Placebo Effect
Rich Ulrich <[EMAIL PROTECTED]> wrote: : - I was a bit surprised by the newspaper coverage. I tend to : forget that most people, including scientists, do *not* blame : regression-to-the-mean, as the FIRST suspicious cause : whenever there is a pre-post design: because they have : scarce heard of it. I don't see how RTM can explain the average change in a prepost design those above the pre population mean will tend to be closer to the post population mean but this doesn't say anything about the average change. Any depression study is apt to show both a placebo AND a no treatment effect after 6 weeks = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: The False Placebo Effect
I am not impressed. I don't think much of people who compare placebo with no treatment; seems stupid to me. I would expect a "placebo" in any case in which the evaluation is a human judgement or one's expectation could reasonably be expected to affect a measured response. Thus I think you could easily get an effect in a blood pressure measurement but not cholesterol. Much ado about nothing. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Intepreting MANOVA and legitimacy of ANOVA
auda <[EMAIL PROTECTED]> wrote: : Hi, all, : In my experiment, two dependent variables were measured (say, DV1 and DV2). : I found that when analyzed sepeartely with ANOVA, independent variable (say, : IV and had two levels IV_1 and IV_2) modulated DV1 and DV2 differentially: I don't have a clue as to what you are talking about. ANOVA tests interactions, main effects, and contrasts. You have factors with levels. that's it. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: 2x2 tables in epi. Why Fisher test?
In sci.stat.math Juha Puranen <[EMAIL PROTECTED]> wrote: : Hhen N is small this is not true. Here a small example By Survo the example is irreelevaant; there are different tests of the same hypothesis eg do a t test with only the first 10 observations. Both tests are valid, the large n test is more powerful = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: additional variance explained (SPSS)
Dianne Worth <[EMAIL PROTECTED]> wrote: : I have a multiple regression y=a+b1+b2+b3+b4+b5. My Adj. R-sq is .403. you can't decompose adjusted R-sqs. The only additive decomposition (and the only decomposition that makes sense) is the stepwise composition of R-sq, adding additional variables in a specified order. This answers a well-defined question: how much does a set of variables add to a model given another set of variables. There are no other questions that can be answered by regression tests. The various SAS tests are all special cases and DO NOT test the same hypothesis for a particular effect test. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: 2x2 tables in epi. Why Fisher test?
In sci.stat.consult Juha Puranen <[EMAIL PROTECTED]> wrote: :> :> Please clarify what is meant by "the distribution does not :> involve [the fixed marginals]". I am not clear on this: :> the Fisher test statistic (hypergeometric upper tail probability) :> certainly *does* depend on the fixed marginals in this :> case -- they appear in every term in that tail sum. sorry; didn't say it right : Usual the assumptions for Fishers exact test are not true. : What you can fix are the row margins, or column margins or grand total These aren't assumptions any more than specific fixed x values are assumptions in linear regression Kendall and Stuart say (under exact test of independence 2x2 table) We may now demonstrate the remarkable result, first given by Tocher (1950) that the exact test based on the Case I probabilities actually gives UMPU tests for Cases II and III The probability statements for case I (fixed marginals) are valid conditional on the marginals for every set of marginals and do not involve the nuisance parameters for Cases II and III and thus are valid unconditionally for all three cases. This is exactly analagous to the regression model y = bx + e where you derive the t test for b conditional on the specific x values you observe, treating them as fixed. The statistic (a function of the x's) has the same t distribution regardless of what x values you observe, even if they happen to be sampled from ANY probability distribution Thus the regression test for fixed x values is valid for random x values = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: 2x2 tables in epi. Why Fisher test?
In sci.stat.consult Ronald Bloom <[EMAIL PROTECTED]> wrote: Herman as usual is absolutely correct; the validity of the Fisher test is analagous to the validity of regression tests which are derived conditional on x but, since the distribution does not involve x, are valid unconditionally even if the x's are random. Incidentally, if one randomizes to get an exact p value, the Fisher test is uniformly most powerful. Herman can tell us if this is for all three cases. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Simple ? on standardized regression coeff.
In sci.stat.consult d.u. <[EMAIL PROTECTED]> wrote: : I now think that the betas would have to be within [-1,+1]. Suppose you do a : standarized regression with response Y, and have p variables (X matrix) already in. you're wrong; 1 varb = r sy/sx = r between -1 and 1 but for 2 var everything is partialed and while partial r is betwee -1 and 1, partial sigmas are not see kendall and stuart I think this is a counter example y x1 x2 2 1 1 -1 0 -1 -1 -1 0 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: ANCOVA vs. sequential regression
Paul Swank <[EMAIL PROTECTED]> wrote: An interaction is always a test of parallel lines whether it is factoral anova, ancova, regression, or profile analysis. Not really. interaction was invented by RA Fisher for ANOVA where there are no lines. that's like saying that ANOVA is regression. It isn't and many people have screwed up by thinking it is. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: ANCOVA vs. sequential regression
William B. Ware <[EMAIL PROTECTED]> wrote: : sequential/hierarchical regression as you note below... however, ANCOVA : has at least two assumptions that your situation does not meet. First, it : assumes that assignment to treatment condition is random. Second, it : assumes that the measurement on the covariate is independent of : treatment. That is, the covariate should be measured before the treatment : is implemented. Thus, I believe that you should implement the : hierarchical regression... but I'm not certain what question you are They aren't assumptions but they do affect interpretations. either way is ANCOVA which will answer a question. Write the model comparison and you'll see that. Whether it's the question you want to answer is another = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Simple ? on standardized regression coeff.
In sci.stat.consult d.u. <[EMAIL PROTECTED]> wrote: : Hi everyone. In the case of standardized regression coefficients (beta), : do they have a range that's like a correlation coefficient's? In other : words, must they be within (-1,+1)? And why if they do? Thanks! Only for 1 x variable where it is r. In othere cases it can be anything = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: normal approx. to binomial
James Ankeny <[EMAIL PROTECTED]> wrote: : My question is, are they saying that the sampling : distribution of a binomial rv is approximately normal for large n? : It's a special case of the CLT for a binary variable with probability p, taking the sum of n observations = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Statistics teacher/professional Needed $$$$$$$$$$$$$
Marina G. Roussou <[EMAIL PROTECTED]> wrote: : A Statistics teacher/tutor/professional is needed to complete an 11 lesson : assignement paper. Each lesson comprises with approximately 5-10 questions. :what is this for??? = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: stan error of r
Elliot Cramer <[EMAIL PROTECTED]> wrote: : dennis roberts <[EMAIL PROTECTED]> wrote: : : anyone know off hand quickly ... what the formula might be for the standard : : error for r would be IF the population rho value is something OTHER than zero? correctiont: the variance is (1/n)*(1-rho^2)^2 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: stan error of r
dennis roberts <[EMAIL PROTECTED]> wrote: : anyone know off hand quickly ... what the formula might be for the standard : error for r would be IF the population rho value is something OTHER than zero? It's (1/n)*(1-rho^2)^2 = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Data Reduction
Dianne Worth <[EMAIL PROTECTED]> wrote: : Question: Should I perform Principal Components and then Factor : Analysis to determine the new constructs? I (barely) use SAS and can never probably not = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Most Common Mistake In Statistical Inference
On Fri, 23 Mar 2001, Alan McLean wrote: > The second sentence here ensures that generalisability to a population > IS an issue for statistics. And a big issue, usually overlooked. > It is not a statistical issue with a non-random sample; it is a matter of experimental judgement > For that matter, many applications of statistics do use sampling, not > random assignment (market surveys, for example) and in these > applications Dennis' observtion is spot on. I was referring to inferential statistics rather than estimating probabilities = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Most Common Mistake In Statistical Inference
given random assignment the generalizability of results to a population is not an issue for statistics. It's a question of what a plausible population is, given the procedure for obtaining subjects On Thu, 22 Mar 2001, dennis roberts wrote: > > using and interpreting inference procedures under the assumption of SRS > simple random samples ... when they just can't be > = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Most Common Mistake In Statistical Inference
W. D. Allen Sr. <[EMAIL PROTECTED]> wrote: : Either the Chi Square or S-K test, as appropriate, should be conducted to : determine normality before interpreting population percentages using : standard deviations. I don't understand why one would want to use the normal distribution for interpreting population percetages; I've never wanted to = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: misues of statistics
On Fri, 16 Mar 2001, Rich Ulrich wrote: > Elliot, > > It appears to me that Arnold Barnett is guilty > of a serious misuse of statistical argument. > > I don't think readers are apt to be misled by the media > reports; there is a very LOW rate of capital punishment > in the US, so the likelihoods are (indeed) essentially > the same as Odds Ratios. This is not a RAW difference which, I'm sure you agree, is not relevant to anything. It's the same problem as comparing University salaries between men and women. Men are much more likely to be full professors for historical reasons (40 years ago few women got PhDs). Faculty rank is relevant to salary as are a host of other variables and must be taken into account. In the death penalty situation the nature of the homicide must be taken into account (you don't get the death penalty for running over someone with a car. The odds ratio quoted is for a model that includes a host of relevant variables IN ADDITION to race. I don't think that there is ANY good statistical evidence of racial discrimination in the death penalty, but that's another issue. The point here is that there are many situations( such as the one illustrated) for which a death penalty has a high probablility which is VERY different from the odds ratio. > > When I saw mention of these data a few years ago, my first tendency > was to doubt the "what-if." P[death sentence] = 0.99? not > generally Rates of executions are low, as I said earlier. not for serial killers or for the type of homicides which lead to the death penalty > > - Now, the author is asserting that 1% versus 4% is far different > from 99% versus 96%. Statisticians should be leery of that. > NO; I think he is asserting that 20% vs 80% is far different > - the judges and journalists missed the word; they missed the math > that would have made the word important; so they ended up with the > right conclusion. > I don't think they ended up with the right conclusion at all. Heinous murderers tend to get the death penalty whether they murder blacks or whites. The point of the article is that the Supreme Court apparently understood the odd ratio to be a probability ratio. The US district court did not make this mistake and issued a devastating critique of the Baldus Study which used linear regression instead of logistic regression, amongh other things. It was VERY inadequate in dealing with nature of the crime which is the most important consideration in the death penalty. Interestingly most murder are within race; blacks murder blacks and whites murder whites. Baldus finds no discrimination based on race of the murderer, only of the victim. = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
misues of statistics
Someone had wanted a source for examples; I found this looking up Arnold Barnett in Google.com. He has other interesting examples. >From http://209.58.177.220/articles/oct94/barnett.html Arnold Barnett The Odds of Execution A powerful example of the first problem arose in 1987, when the U.S. Supreme Court issued its controversial McClesky v. Kemp ruling concerning racial discrimination in the imposition of the death penalty. The Court was presented with an extensive study of Georgia death sentencing, the main finding of which was explained by the New York Times as follows: "Other things being as equal as statisticians can make them, someone who killed a white person in Georgia was four times as likely to receive a death sentence as someone who killed a black." The Supreme Court understood the study the same way. Its majority opinion noted that "even after taking account of 39 nonracial variables, defendants charged with killing white victims were 4.3 times as likely to receive a death sentence as defendants charged with killing blacks." But the Supreme Court, the New York Times, and countless other newspapers and commentators were laboring under a major misconception. In fact, the statistical study in McClesky v. Kemp never reached the "factor of four" conclusion so widely attributed to it. What the analyst did conclude was that the odds of a death sentence in a white-victim case were 4.3 times the odds in a black-victim case. The difference between "likelihood" and "odds" (defined as the likelihood that an event will happen divided by the likelihood that it will not) might seem like a semantic quibble, but it is of major importance in understanding the results. The likelihood, or probability, of drawing a diamond from a deck of cards, for instance, is 1 in 4, or 0.25. The odds are, by definition, 0.25/0.75, or 0.33. Now consider the likelihood of drawing any red card (heart or diamond) from the deck. This probability is 0.5, which corresponds to an odds ratio of 0.5/0.5, or 1.0. In other words, a doubling of probability from 0.25 to 0.5 results in a tripling of the odds. The death penalty analysis suffered from a similar, but much more serious, distortion. Consider an extremely aggravated homicide, such as the torture and killing of a kidnapped stranger by a prison escapee. Represent as PW the probability that a guilty defendant would be sentenced to death if the victim were white, and as PB the probability that the defendant would receive the death sentence if the victim were black. Under the "4.3 times as likely" interpretation of the study, the two values would be related by the equation: If, in this extreme killing, the probability of a death sentence is very high, such that PW = 0.99 (that is, 99 percent), then it would follow that PB = 0.99/4.3 = 0.23. In other words, even the hideous murder of a black would be unlikely to evoke a death sentence. Such a disparity would rightly be considered extremely troubling. But under the "4.3 times the odds" rule that reflects the study's actual findings, the discrepancy between PW and PB would be far less alarming. This yields the equation: If PW = 0.99, the odds ratio in a white-victim case is 0.99/0.01; in other words, a death sentence is 99 times as likely as the alternative. But even after being cut by a factor of 4.3, the odds ratio in the case of a black victim would take the revised value of 99/4.3 = 23, meaning that the perpetrator would be 23 times as likely as not to be sentenced to death. That is: Work out the algebra and you find that PB = 0.96. In other words, while a death sentence is almost inevitable when the murder victim is white, it is also so when the victim is black - a result that few readers of the "four times as likely" statistic would infer. While not all Georgia killings are so aggravated that PW = 0.99, the quoted study found that the heavy majority of capital verdicts came up in circumstances when PW, and thus PB, is very high. None of this is to deny that there is some evidence of race-of-victim disparity in sentencing. The point is that the improper interchange of two apparently similar words greatly exaggerated the general understanding of the degree of disparity. Blame for the confusion should presumably be shared by the judges and the journalists who made the mistake and the researchers who did too little to prevent it. (Despite its uncritical acceptance of an overstated racial disparity, the Supreme Court's McClesky v. Kemp decision upheld Georgia's death penalty. The court concluded that a defendant must show race prejudice in his or her own case to have the death sentence countermanded as discriminatory.) = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Avoiding Linear Dependencies in Artificial Data Sets
I'm not clear on what your design is but it seems that the problem is in the between S effect not within. Note that you only have 4 df within and 4 dependent variables = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: multivariate normality
yogab <[EMAIL PROTECTED]> wrote: : in particular or comments about mutivariate testing ? or any : better way to do mutivariate normality testing ? why do you want to test it = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: algorithm cross correlation
Hanke <[EMAIL PROTECTED]> wrote: : Does anyone know a algorithm for cross-correlation between two time : series how about something like Do 10 i = 1,n-2 r1 = corr(a(1),b(i),n-i+1) 10 r2 = corr(b(1),a(i),n-i+1) where corr computes the r for n obs = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Recommend multiple regression text please
In sci.stat.edu Jim Kroger <[EMAIL PROTECTED]> wrote: I think you need a statistician rather than a book = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =
Re: Levels of measurement.
Rich Ulrich <[EMAIL PROTECTED]> wrote: : I agree, you have been thinking about it "too much." MUCH too much : I think you have to take Stevens's hierarchy of scaling more lightly. even to the point of forgetting it = Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =