Hi:
WOuld it be possible for any of the following to have a value greater than 1?
accuracy, specificity, sensitivity, precision, and recall
Say, if a set of samples was taken, the Var_avg +- std would be > 1?
where Var stands for accuracy, specificity, sensitivity, precision, or recall
On Mon, 5 Aug 2019 03:12:54 -0700 (PDT), Cosine
wrote:
Hi:
WOuld it be possible for any of the following to have a value greater than 1?
accuracy, specificity, sensitivity, precision, and recall
Of course not.
Say, if a set of samples was taken, the Var_avg +- std would be > 1?
where Var stands for accuracy, specificity, sensitivity, precision, or recall
Sure, you can compute a CI that is exceeds the bounds of
(0,1) if you use simplistic formulas. But for any proportion
that is not 0.50, the more accurate statement of "error" is
never symmetric, so you would not write "+- std" if you
intended to be precise.
And - Even for 50%, the symmetrical CI will be too wide
if you use the simplisitic formulas. Consider the 99% or
99.9% (or even more extreme) CI for any small sample.
--
Rich Ulrich
Rich Ulrich? 2019?8?6???? UTC+8??1?30?30????
On Mon, 5 Aug 2019 03:12:54 -0700 (PDT), Cosine
wrote:
Hi:
WOuld it be possible for any of the following to have a value greater than 1?
accuracy, specificity, sensitivity, precision, and recall
Of course not.
Say, if a set of samples was taken, the Var_avg +- std would be > 1?
where Var stands for accuracy, specificity, sensitivity, precision, or recall
Sure, you can compute a CI that is exceeds the bounds of
(0,1) if you use simplistic formulas. But for any proportion
that is not 0.50, the more accurate statement of "error" is
never symmetric, so you would not write "+- std" if you
intended to be precise.
And - Even for 50%, the symmetrical CI will be too wide
if you use the simplisitic formulas. Consider the 99% or
99.9% (or even more extreme) CI for any small sample.
--
Rich Ulrich
Thank you for the explanation.
But how do we interpret the part exceeding [0, 1]?
Say, how do we interpret those values of accuracy being negative or greater than 1? Does it mean that those exceeding [0, 1] are meaningless?
Consider using one of the good approximate methods for computing your CIs--e.g., the Wilson score method or the Agresti-Coull method. Here's an online calculator that implements both of those:
http://epitools.ausvet.com.au/content.php?page=CIProportion
If you want SPSS code to compute these CIs, see syntax files ciprop.SPS and ciprop2.SPS on my SPSS page:
https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/spss/my-spss-page
You should be able to lift the code for the method you want to use fairly easily. E.g., if you want to use the Wilson score method, here is the basic code:
* Wilson score method (Method 3 in Newcombe, 1998) .
* Code adapted from Robert Newcombe's code posted here:
http://archive.uwcm.ac.uk/uwcm/ms/Robert2.html .
COMPUTE p = x/n.
COMPUTE q = 1-p.
COMPUTE z = probit(1-alpha/2).
COMPUTE #x1 = 2*n*p+z**2 .
COMPUTE #x2 = z*(z**2+4*n*p*(1-p))**0.5 .
COMPUTE #x3 = 2*(n+z**2) .
COMPUTE lower4 = (#x1 - #x2) / #x3 .
COMPUTE upper4 = (#x1 + #x2) / #x3 .
I used lower4 and upper4 because this was method 4 in the original syntax file. You can change it to lower and upper if you are using just one method. HTH.
Rich Ulrich於 2019年8月6日星期二 UTC+8上午1時30分30秒寫道：
On Mon, 5 Aug 2019 03:12:54 -0700 (PDT), Cosine
wrote:
Hi:
WOuld it be possible for any of the following to have a value greater than 1?
accuracy, specificity, sensitivity, precision, and recall
Of course not.
Say, if a set of samples was taken, the Var_avg +- std would be > 1?
where Var stands for accuracy, specificity, sensitivity, precision, or recall
Sure, you can compute a CI that is exceeds the bounds of
(0,1) if you use simplistic formulas. But for any proportion
that is not 0.50, the more accurate statement of "error" is
never symmetric, so you would not write "+- std" if you
intended to be precise.
And - Even for 50%, the symmetrical CI will be too wide
if you use the simplisitic formulas. Consider the 99% or
99.9% (or even more extreme) CI for any small sample.
--
Rich Ulrich
Thank you for the explanation.
But how do we interpret the part exceeding [0, 1]?
Say, how do we interpret those values of accuracy being negative or greater than 1? Does it mean that those exceeding [0, 1] are meaningless?
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 185 |
Nodes: | 16 (1 / 15) |
Uptime: | 19:33:23 |
Calls: | 3,730 |
Calls today: | 13 |
Files: | 11,159 |
Messages: | 3,452,775 |