SoylentNews is people
SoylentNews
The abrupt dismissal of the head of a Utah cancer center is causing backlash from its faculty—and its major philanthropic funder—in a struggle over the center's autonomy from the University of Utah in Salt Lake City. And nearly 2000 researchers have signed a petition calling on the university to reverse its decision.
For 11 years, prominent cell biologist Mary Beckerle has headed the Huntsman Cancer Institute (HCI), which is based at the university but receives its funding largely from philanthropic donations, revenue from its cancer hospital, and grants from state and from the National institutes of Health. In an email to some clinical staff on Monday, university President David Pershing and Vivian Lee, senior vice president for health sciences, announced that Beckerle would step down "effective yesterday," but would "remain on faculty as a distinguished professor in biology." Beckerle, who has not responded to Science's request for comment, told The Salt Lake Tribune that she had learned of her dismissal in an email less than an hour earlier.
Details have been scant from the university, which also did not respond to a comment request. But Beckerle's colleagues contend that the move amounts to a hostile takeover by the university aimed at capturing the cancer clinic's revenue, and other prominent scientists are rallying unquestioningly around her.
Also at Deseret News. Change.org petition. University of Utah Health press release.
(Score: 0, Troll) by Anonymous Coward on Monday April 24 2017, @01:43PM (19 children)
Save your typical statistics post for when they're on topic. Even if your premise that "science = analyzing data with statistics" were true, the post would still be off-topic.
Another suggestion: you should register (with a fake/temp email) and post a journal then a submission on the topic so you don't have to be so repetitive and it will be easier for people to learn about and debate the topic.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @01:50PM (14 children)
It is likely that the institute produces mostly NHST rather than science (usually this is about a 10^3 to 10^4 ratio), so the parent's assumption this article is about science is incorrect. Even the category used here is business. If any post was off-topic, it was that one.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @02:24PM (13 children)
usually this is about a 10^3 to 10^4 ratio
So a 1:10 ratio? Or were you indicating a range of 10³–10⁴, so 5,000 some odd units NHST to 1 unit science?
I was going to go with the latter, but then I realized I wasn't sure.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @02:29PM (12 children)
Yea, to be honest that's anecdotal though. I went to a conference a few years ago and made the time to inspect every single poster over a the few days. There were a dozen or so (out of tens of thousands) not relying on NHST to draw their conclusions. It fits well with what I've seen with the literature, so I am pretty confident the ratio is around those values.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @07:11PM (11 children)
Which field? Extrapolating from your limited anecdotal experience in a particular setting to the research done at the Utah Cancer Center and making a specific numerical conclusion seems to be a bit hypocritical for someone so concerned with proper data analysis.
Do you have a p-value less than .05?/sarcasm
(Score: 0) by Anonymous Coward on Monday April 24 2017, @07:39PM (10 children)
Biomedical research
(Score: 0) by Anonymous Coward on Monday April 24 2017, @07:58PM (9 children)
Both of the papers below are from research groups at the Utah Cancer Center. You'll notice that they do not use NHST for their data. The first is mainly concerned with medicinal chemistry SAR optimization and the second is the biochemistry of transcription factor-DNA interactions.
https://www.ncbi.nlm.nih.gov/pubmed/26182238 [nih.gov]
http://www.sciencedirect.com/science/article/pii/S002228361300747X [sciencedirect.com]
(Score: 0) by Anonymous Coward on Monday April 24 2017, @08:29PM (8 children)
I looked at the first paper, it is still NHST. For example figure 9B shows a difference between control and treatment, then they conclusions drawn is that "Our in vivo data demonstrated that the [compound] 14 treatment resulted in tumor growth inhibition compared to controls".
This is just NHST, but via "eyeballing" rather than calculating any p-value. They don't consider other explanations for these observations. For example, they don't seem to report blinding, so one explanation can be that the tech/student measuring the tumor volume was simply biased. How much can bias explain with that type of measurement, even just roughly? Another thing that comes to mind is that perhaps the treatment reduced edema within the tumor, rather than slowing tumor growth. Indeed, a quick search shows I am not the first to bring up that issue:
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0009364 [plos.org]
I'm sure I can think up more explanations if I spent the time.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @09:08PM (7 children)
This is just NHST, but via "eyeballing" rather than calculating any p-value.
No, come up with a different term if NHST doesn't fit what you mean.
perhaps the treatment reduced edema within the tumor, rather than slowing tumor growth
They report their observation and define how they measure tumor size. You are free to disagree with any conclusions that the authors present and consider other explanations.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @09:22PM (6 children)
"Eyeballing", in this case, is just another implementation of NHST along with t-tests, anovas, etc. Rather than testing their hypothesis they "test" (eyeball test) the stawman "null" hypothesis, then conclude their favorite explanation is correct. Science is about distinguishing between different explanations for what is observed, not ruling out "no difference between groups".
(Score: 0) by Anonymous Coward on Monday April 24 2017, @09:39PM (5 children)
Are you using a different definition?
NHST: Null Hypothesis Significance Testing
their favorite explanation
Do you expect them to present their least favorite explanation in their discussion/conclusion section?
You also seem to be assuming that scientific papers are supposed to be unbiased reports of data (they aren't). Can you even point to any examples of your ideal?
(Score: 0) by Anonymous Coward on Monday April 24 2017, @10:47PM (4 children)
Here is a quick example of what I expect. I collected data [grabbed the data using the R digitize package] , fit a model, and plotted it.
The solid lines are the data ( I didn't bother with the error bars) from figure 9B and the dotted are some simple models of tumor growth:
https://i.imgur.com/detFzzk.png [imgur.com]
From this simple model we can see that the increase in tumor volume for the controls could be approximately explained by an increase in division rate from 0.1 per day to 0.13 per day. But the curves don't really match up. Instead we could play with the volPerCell parameter, maybe the treatment does not affect growth rate, but does affect the volume per cell. This model does not include cell death, which no doubt happens. So we should probably add that to the model. Also the control tumors look like they are plateauing towards the end, so maybe cell death and replication rates should be functions of the number of cells.
All these parameters will be relatively unconstrained at first, so we should collect other data to constrain their values. Eventually you get a model that fits the data, then collect new data and see if the model can also fit that given the plausible range of parameter values.
(Score: 2) by aristarchus on Tuesday April 25 2017, @05:26AM (1 child)
So we should probably add that to the model.
Um, no, we should not? You are not actually a scientist, are you? You are a dear and fluffy AC! Could I interest you in an Electric Universe? One that proves that Einstein was completely wrong? Or how about some nice Homeopathy? No, nothing gay about it. Or some really good Climate Denialism? We are having a sale!
New Motto of SN"
"NO one expects the Violent Imposition of the Null Hypothesis! Those who do expect it, well, um, . . . ."
(Score: 0) by Anonymous Coward on Tuesday April 25 2017, @11:50AM
Why not?
(Score: 0) by Anonymous Coward on Tuesday April 25 2017, @02:58PM (1 child)
Are you using the same definition of NHST and do you accept that the authors did not run a significance test for the data in 9B?
I meant an example of a research article that meets your ideal of no bias.
Your analysis of the data does help me see where you are comming from, though. What you are doing seems more related to the field of computational or systems biology. In that field, scientists attempt to model raw data into equations. Because biology is such a complex (no spherical cows in a vacuum) and young science (the systems are still being characterized), these models are incredibly imprecise in their predictive power. This is probably why experimental and observational biology dominate in their ability to produce useful conclusions; however they will not be able to determine the Truth.
(Score: 0) by Anonymous Coward on Tuesday April 25 2017, @04:43PM
As mentioned, the defining feature of NHST (which is different than Fisher's original significance test, and also different from Neyman/Pearson's original hypothesis test[1]) is that you check whether the data is consistent with a "null" hypothesis (usually a so-called "nil" hypothesis of no difference between two groups). The mathematical details are not a defining feature, there are even "Bayesian significance tests"[2] that are based on an entirely different definition of probability. Just because the method used here is "look, the error bars don't touch", doesn't change that.
The ideal is not "no bias", it is to compare various explanations for the data. NB, if your study can distinguish between any two real explanations it will also always be able to rule out chance as well. This is something done as a matter of course, there is no reason to have a special step devoted to it.
Also, I've found any study designed to just check for a mere difference between two groups will have to deal with so many alternative explanations that it is pretty much impossible to be confident you are interpreting the difference correctly. Instead you need to think hard about your explanation and get a precise prediction of some sort out.
As for some papers I like off the top of my head (not that I accept all the conclusions):
http://www.sciencedirect.com/science/article/pii/S0019103516304869 [sciencedirect.com]
http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1000877 [plos.org]
http://www.pnas.org/content/101/36/13124 [pnas.org]
http://iopscience.iop.org/article/10.3847/2041-8205/816/1/L17 [iop.org]
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2007940/ [nih.gov]
I disagree, that is what they told me, then I found papers from the 1930s that had models capable of describing my system very well. Of course, once I implemented them and showed it around, no one in that area knew wtf I was talking about because they hadn't thought of these things quantitatively for decades. It is not about predicting things exactly, even nowadays they have to include a bunch of ad hoc empirically-defined adjustments to make accurate predictions of solar system dynamics.[3,4] The goal is to find "universalities" in the data that can be modeled by simple processes, not check for differences.
It seems to me this "it is too complex" idea is just a self-defeating attitude coupled with bad practices in the field (accepting very vague descriptions as "useful")[5]:
Then there is of course the issue that "reliance on significance testing retards the growth of cumulative research knowledge"[6] automatically, as if it was designed to do so. I mean not just because people aren't bothering to come up with quantitative models since it is unnecessary for career success, but that when used "correctly" you are destined to either generate conflicting conclusions or always find "significance" (depending on sample size).
[1] http://library.mpib-berlin.mpg.de/ft/gg/GG_Mindless_2004.pdf [mpib-berlin.mpg.de]
[2] http://www.tandfonline.com/doi/full/10.1080/03610926.2011.563021 [tandfonline.com]
[3] https://en.wikipedia.org/wiki/Jet_Propulsion_Laboratory_Development_Ephemeris [wikipedia.org]
[4] http://www.cv.nrao.edu/~rfisher/Ephemerides/ephem_descr.html [nrao.edu]
[5] https://www.ncbi.nlm.nih.gov/pubmed/12242150 [nih.gov]
[6] http://psycnet.apa.org/journals/met/1/2/115.pdf [apa.org]
(Score: 0) by Anonymous Coward on Monday April 24 2017, @02:26PM (3 children)
Also, this is exactly not my premise. This is, however, one (of many) faulty premises behind NHST usage.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @07:06PM (2 children)
Your premise seems to be "science = proper analysis of data with statistics" or that using improper statistical approaches to data analysis seems to invalidate experiments as being science.
Science was done before scientists used statistics regularly. NHST was better than nothing, but "good shouldn't be the enemy of great" and "perfect shouldn't be the enemy of great".
(Score: 0) by Anonymous Coward on Monday April 24 2017, @07:41PM (1 child)
No, NHST is not better than nothing. nNot learning statistics at all would be far, far superior to learning NHST.
The problem is that it reverses the logic of science. It has nothing to do with the statistics equations. Please read the Meehl paper that explains this very well.
(Score: 0) by Anonymous Coward on Monday April 24 2017, @08:14PM
The problem is that it reverses the logic of science
Possibly for psychology studies, but not necessarily for other fields. NHST is mainly used in biomedical research (which appears to be the field you studied) to show some weak evidence that the data presented is not simply the result of technical artifacts. NHST is better than showing a single "representative" biological replicate or deliberately reporting the best result (typical for %yield in synthesis papers). Even when people completely fuck-up NHST and use it on technical replicates, it is at least eliminating the data that can't be consistent enough for technical reasons.