Hi A directional test is based on the expectation for a particular hypothesis and does NOT state that the opposite pattern could NEVER occur even with other confounding factors, as in the therapeutic touch example. You state that in fact detection can occur under the conditions stated, which I assume means that studies have been done without the confounding and led to the rejection of no detection (null) in favor of detection (alternative). Would you want the researchers to have concluded that detection was possible based on negative results? Presumably not. And even if they had done a directional test, I suspect most researchers would minimize a difference in the unexpected direction, perhaps as a chance occurrence.
Nor do directional tests always ignore the other tail. The null for directional tests are sometimes even written to include the other end (i.e., H0: p LE .5, H1: p GT .5). In failing to reject null to discredit some theoretical expectation, as in this example, the extra sensitivity of the directional test (i.e., lower critical value) would also appear beneficial in order to lessen chance of Type II error (failing to reject false Ho). That is, it makes the case stronger. What I like about this example is the importance of readers of research being observant and thoughtful about findings and conclusions, as you were! Take care Jim James M. Clark Professor of Psychology 204-786-9757 204-774-4134 Fax [email protected] Department of Psychology University of Winnipeg Winnipeg, Manitoba R3B 2E9 CANADA >>> Paul C Bernhardt <[email protected]> 13-Jan-09 3:52 PM >>> I'll grant you that these examples appear quite reasonable to be one tailed tests. And, I'll describe another that appears, also, to be reasonably one tailed test and was published that way. Some of you may remember the study of Therapeutic Touch done by a 13 year old (at the time) young girl and her parents as part of a science fair project (initially). The study was published in the Journal of the American Medical Association (Rosa, Rosa, Sarner & Barrett, 1998). Twenty-one self-represented Therapeutic Touch practitioners had to state which of their hands the child had positioned her hand (determined randomly, and obscured by an opaque partition). If the TT practitioner was skilled at detecting the purported Human Energy Field, they should get more than 50% correct. Presumably, doing worse than chance would be meaningless. Out of the 280 trials provided by the practitioners, they guessed correctly 123 times, 44%. Obviously, this was a result in the opposite direction of the one tailed test and is not meaningful. Or is it? Turns out that 123 hits out of 280 attempts just breaks below a .05 two tailed test of significance by a binomial test. Hmmm.... Turns out I was two degrees of separation from the Rosa study, they were known by a buddy of mine. We were skeptical. That one tailed test bothered us. So, we started doing some studies of our own. Turns out without any interceding material (cloth, glass, etc) between the hands, at the distance of separation used in Rosa, et al. we should have expected better than 50% hits. People can sense the warmth of a hand even up to 5 or 6 inches away. So, how come the below chance finding of Rosa, et al? We know they used verbal instructions and (certainly unintentionally) directed their voice away from the target hand. We attempted that kind of procedure and replicated their findings pretty closely, coming in at 39%. Turns out a reasonable speculation is experimenter error, which should have been detected, and a two tailed test would have done so. You can see our study in Long, Bernhardt & Evans, 2000, a chapter in Therapeutic Touch, edited by Bela Scheiber & Carla Selby. Prometheus Books. -- Paul Bernhardt Frostburg State University Frostburg, MD, USA On 1/11/09 11:57 AM, "Wuensch, Karl L" <[email protected]> wrote: > > First, a trivial point. The F test employed in traditional > ANOVA is a one-tailed test -- regardless of the ordering of the > differences among the group means, greater differences lead to a larger > F. Accordingly, it is a one-tailed, upper-tailed, test. It could be > done as a lower-tailed test if you put the error term in the numerator, > as is done with some multivariate test statistics. > Second, a couple of examples I used in class of directional > hypotheses that seem reasonable. > > While taking a multiple choice test, where each item has four > response options, I observed student Joe Blow rolling a die once or > twice before he answers each item. I suspect that he is using the die > to chose the response option he endorses. If so, the item probability > of success (binomial p) is .25. What is p if the student is not using > the die to select responses? I dismiss the possibility that the student > knows the material and is trying to get a low score (although I can > imagine situations when this might be true, they are very unusual). If > Joe knows the material, p should be greater than .25, and that is my > "alternative" hypothesis. The "null" hypothesis is that p is less than > or equal to .25 -- Joe is using the die to select response options or > Joe knows nothing about what is being tested. > > The next one is from actual research. Richard Porter gathered > shirts worn by infants in the maternity ward. He stuffed each shirt > into a tube. He then presented two tubes to baby's Mom, one of which > contained her baby's shirt. Mom sniffed them both and then indicated > which she thought had the shirt worn by her baby. If Moms can identify > their babies by olfactory cues, what is the probability that Mom will > pick the correct tube on one trial? It is, of course, greater than .5. > We dismiss the possibility that Mom would try to mess up the research by > picking the one that is not her baby. Accordingly, the directional > hypotheses tested are "p is less than or equal to .5" and "p is greater > than .5." > > > Cheers, > > Karl W. > -----Original Message----- > From: Jim Clark [mailto:[email protected]] > Sent: Sunday, January 11, 2009 1:17 AM > To: Teaching in the Psychological Sciences (TIPS) > Subject: RE: [tips] ANOVA question (was cross-cultural) > > Hi > > I'll take Stephen's points in reverse order, starting with Abelson, in > response to my: > > --- > To make changes to your subscription contact: > > Bill Southerly ([email protected]) --- To make changes to your subscription contact: Bill Southerly ([email protected]) --- To make changes to your subscription contact: Bill Southerly ([email protected])
