Bit tricky on that, so let me make it easier:pic.twitter.com/HECG91HWBg
Show this thread
Prize for the first person to spot what went wrong here. (First author is an old friend so I am sure she won't mind me pointing it out. Great study! All publicity is good publicity Valentina!)pic.twitter.com/pMMhP29NgB
-
-
-
Answer in the morning from
@drgrahamcole who pointed it out to me.Show this thread -
OK I see some initial answers that are aimed at pretty high-brow analysis. You can ask F2 harrell about all that kind of fancy stuff. I am just looking at this.pic.twitter.com/Qp5DTPGErY
Show this thread -
Hmmm... maybe people are finding this harder than I expected. A further hint here.
Show this thread -
Why can we be *confident* that (a) what was described in the methods, was not was actually done and (b) that in one case, even what was done, was done incorrectly.pic.twitter.com/D79fNJL7GK
Show this thread -
Lots of sophisticated answers overnight! But we are looking for something simple. Kishore is the closest to what Graham and I had in mind. Is that a good enough hint?pic.twitter.com/5wbPErcaHD
Show this thread -
-
Step 2. Look at the top two rows of data and the bottom row.pic.twitter.com/zb3XDdIfEd
Show this thread -
Now what jumps out and slaps you in the face? Remember that we said Kishore was close to the idea we are after.
Show this thread -
Any doctor who has seen BNPs or CRPs or troponins should start laughing now. Hint. What do those three variables share, in your experience?
Show this thread -
Excellent suggestion! Not the answer I was looking for, but getting at exactly the right issue.
Show this thread -
Why do we sometimes report "median and interquartile range" rather than "mean and SD"?
Show this thread -
The *two* leading answers are correct. Let's remind ourselves about Parametric and Non-parametric STATISTICAL TESTS. (
@alexnowbar please make a note of this as a topic to cover specifically as people do find it a bit puzzling sometimes)Show this thread -
Suppose I want to describe to you the positions of ALL of the light blue pixels on this graph.pic.twitter.com/t8PO6v7qxK
Show this thread -
It's easy. I would say: "Go to point (2, -1). Every pixel that is within 3 pixels of that point, is in my region of interest. Everything outside is not." How many numerical parameters do I need to describe the position of the light blue region?
Show this thread -
Hmmm... the answer "two" seems to be inexplicably popular, winning half the votes at the 6-vote stage. I can see how Donald Trump became president.
Show this thread -
Here are the numbers I used. X coordinate of centre of circle, i.e. 2 Y coordinate of centre of circle, i.e. -1 Radius of circle, i.e. 3. How many numbers is that?
Show this thread -
It occurs to me that some people were counting (2, -1) as a single numerical parameter. Anyway I am considering 2 and -1 to be separate numbers. (I can't believe I'm having to point this out)
Show this thread -
Now look at this triangle. Think of all the ways you could describe the points of its interior. What is the smallest number of numerical parameters you can use?pic.twitter.com/tWlY5b4TDp
Show this thread -
The picture itself uses 6 parameters. With those 6 parameters, 2 coordinates for each of the 3 corners, the triangle is described unambiguously, and uniquely - that is, only that exact triangle has those particular 6 values.
Show this thread -
You could try a different combination, to try to cut down the wasteful usage of parameters. (Save the planet, etc) For example, how about: X and Y coords of point B, and then Orientation of one side (0 to 360 degrees) Orientation of other side (0 to 360 degrees) Seems to be 4?
Show this thread -
But no, if you only know the orientations, you don't know how far along the lines to put the points A and C. So you need X and Y coords of B, plus Orientation AND LENGTH of one side Orientation AND LENGTH of other side So that's 6 again. It turns out 6 is the minimum needed.
Show this thread -
Now, how many numerical parameters would you need to correctly describe all the points inside this shape?pic.twitter.com/4Mzc2sOTKb
Show this thread -
-
That's the problem we face in stats. The question we are asking in hypothesis testing is typically of this form: "These two groups of numbers, heights of men and heights of women. Could they easily be explained as one big group, with man versus woman being unrelated to height?"
Show this thread -
The statistical hypotheis test, aka the P value, is telling us: "If in reality gender has no effect on height, how likely is it that when you measure heights of men and women, you would get a result as different as this between groups?"
Show this thread -
To give you a P value, the stats test has to have a description of the data. It is really, really convenient if we can tell the stats test a simple description of the data. So suppose we have loads of data points:pic.twitter.com/LmEPzR7geD
Show this thread -
Instead of feeding into the statistical test all the INDIVIDUAL data points, we just give it a SUMMARY of the distribution of the men, and of the women. The most convenient summary is when the data look reasonably "Normal" (Gaussian) in distribution.
Show this thread -
Because then we can just say "In my sample, the men have height distributed with mean 70 inches, and standard deviation 5 inches. And the women have mean 65 inches, and standard deviation 5 inches." How many numerical PARAMETERS have we reduced our detailed graph to?
Show this thread - Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.