tag:blogger.com,1999:blog-7890764972166411105.post6035832690032151439..comments2017-10-14T00:59:35.711+02:00Comments on Nick Brown's blog: Strange patterns in some results from the Food and Brand LabNick Brownhttp://www.blogger.com/profile/00172030184497186082noreply@blogger.comBlogger11125tag:blogger.com,1999:blog-7890764972166411105.post-65764323180538735402017-03-23T22:44:08.556+01:002017-03-23T22:44:08.556+01:00^ My question also.
^ My question also.<br />Studenthttps://www.blogger.com/profile/09738758434750487307noreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-49280282785296102842017-03-23T22:37:52.055+01:002017-03-23T22:37:52.055+01:00@Anonymous: Oops, I was forgetting that there are ...@Anonymous: Oops, I was forgetting that there are three columns and the DFs are (2,130) rather then (1,130) here. I added a note above. Thanks for spotting this.Nick Brownhttps://www.blogger.com/profile/18266307287741345798noreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-18590979769300748892017-03-23T20:38:56.953+01:002017-03-23T20:38:56.953+01:00Finally, it is unclear whether the sample of 770 m...<i> Finally, it is unclear whether the sample of 770 mentioned in this article and (in almost-identical paragraphs) in this article and this book chapter represents yet another mailing</i><br /><br />The degree of text recycling in those examples is impressive. It is as if Wansink is republishing the same chapter again and again, with different "courtesy authors" each time, who contribute nothing substantial but appreciate the log-rolling.Smut Clydehttps://www.blogger.com/profile/09409476490132867809noreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-6404873018816266222017-03-23T19:13:20.257+01:002017-03-23T19:13:20.257+01:00"because as every undergraduate knows, F(1,D)..."because as every undergraduate knows, F(1,D) is never significant at the .05 level below a value of 3.84 [actually, 3.8415] no matter how large the denominator degrees of freedom D are"<br /><br />I have taught statistics for many years. No, every undergraduate does not know this. It is not often explicitly stated in textbooks. It is evident from F tables of course but now with the availability of statistical packages many if not most textbooks omit F tables. Most professors (even those who regularly use analysis of variance) don't know this either. You probably didn't know it yourself until you looked it up in a table. A little less self-righteous hypocrisy would be good. Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-45149464560100445072017-03-23T18:19:08.124+01:002017-03-23T18:19:08.124+01:00"(For what it's worth, there is another i..."(For what it's worth, there is another incorrect significance star on the F statistic of 3.6 on the item 'In general, I am an adventurous person' here.)"<br /><br />pf(3.6, 2, 130, lower.tail = FALSE) = 0.03008158<br /><br />Do you refer to the fact that it is crowned with just one star (signifying p less than 0.10) where in fact it could've earned twice the pride (less than 0.05)? Or is there some degrees of freedom confusion or something else that I'm missing?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-88778615653103684662017-03-22T22:34:11.403+01:002017-03-22T22:34:11.403+01:00apparently, the study was rerun with a completely ...<i>apparently, the study was rerun with a completely different set of participants, and yet almost all of the measured resultsâ€”17 out of 18â€”were identical, including the decimal places.</i><br /><br />The level of replicability implies that Wansink is obtaining two or three significant digits of accuracy, from surveys of only a few hundred people, which is an accomplshment worthy of a Nobel Prize as soon as he documents the details.Smut Clydehttps://www.blogger.com/profile/09409476490132867809noreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-25699597645685777162017-03-22T22:27:22.530+01:002017-03-22T22:27:22.530+01:00@Boris: Actually I did hypothesise a priori that z...@Boris: Actually I did hypothesise a priori that zero would be the least represented digit --- that is, when deciding to copy/paste these numbers from the PDF into R to check them, I only counted the trailing zeroes --- based on page 43 of the Mosimann et al. article that is referenced in the post (and other stuff I had previously read on this topic). But I can't prove that, so feel free to apply a correction such as Bonferroni to the .00017, or just look at the chi-square p value.Nick Brownhttps://www.blogger.com/profile/18266307287741345798noreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-17701067872344404412017-03-22T22:11:16.157+01:002017-03-22T22:11:16.157+01:00Aren't you performing implicit multiple compar...Aren't you performing implicit multiple comparisons (with the binomial test) by choosing the digit that has the lowest frequency?Boris Barbournoreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-14098863289861438942017-03-22T21:23:26.756+01:002017-03-22T21:23:26.756+01:00PS: It's not completely clear to me whether th...PS: It's not completely clear to me whether the chi-square or the binomial test is the right one, so I showed both. I was quite surprised at the difference in p values between them. In another article from the same laboratory, also with a sample of 770 people (!), the same lack of randomness appears; in that case, however, the p values for the chi square and binomial test are similar (.008 and .007, respectively). Of course, it depends on just how strangely small the number of zeroes is.Nick Brownhttps://www.blogger.com/profile/18266307287741345798noreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-17568764277162955442017-03-22T21:20:47.079+01:002017-03-22T21:20:47.079+01:00@Anonymous: Benford's Law applies to the *firs...@Anonymous: Benford's Law applies to the *first* digit of numbers that correspond to real-world quantities. At issue here is the *last* digit of the means of random variables and/or the test statistics associated therewith, which ought (cf. the reference I gave in the post) to be uniformly distributed.<br /><br />Additionally, although it's not especially relevant here, Benford's Law would not be expected to apply here, because most of the quantities in question are Likert-type data that are constrained to [1..9]. In fact, assuming some sort of normaal distribution, one might expect a prevalence of numbers around the midpoint (5) for the first digits. But I did not examine the first digits, because they are not relevant to my argument about the lack of randomness in these data.Nick Brownhttps://www.blogger.com/profile/18266307287741345798noreply@blogger.comtag:blogger.com,1999:blog-7890764972166411105.post-13080347460677686822017-03-22T21:04:35.840+01:002017-03-22T21:04:35.840+01:00Are you sure the binomial test on the zeros makes ...Are you sure the binomial test on the zeros makes sense? digits are not uniformly distributed, see benford's law!Anonymousnoreply@blogger.com