Timeline for Normality tests for histograms
Current License: CC BY-SA 3.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 13, 2017 at 12:44 | history | edited | CommunityBot |
replaced http://stats.stackexchange.com/ with https://stats.stackexchange.com/
|
|
| May 22, 2014 at 16:23 | comment | added | whuber♦ | Yes, there are such tests. The point is that one can contemplate creating a test based on generating data consistent with the histogram, as Nick describes here, and then work out (mathematically) the distribution of the test statistic, which will necessarily depend only on the information available in the original histogram. For instance, a simple powerful test could be based on the maximum achievable value of the Kolmogorov-Smirnov statistic among all datasets consistent with the histogram (its "p-box"). That statistic is easily computed from the histogram cutpoints and heights. | |
| May 22, 2014 at 3:26 | comment | added | Nick Stauner | That I would be interested to know as well...but since the shape of a histogram is just as subject to obfuscation as the data it represents, I doubt any such method would solve the inherent problems. | |
| May 22, 2014 at 3:17 | comment | added | dainichi | Thanks for the reply. Yes, the histogram represents frequencies within binned ranges. I'm assuming that the choice of ranges is independent of the data. If there ranges were narrow enough, I guess I could reproduce the dataset by picking the middle of the ranges, distributing uniformly over the range, or something like that. But if the ranges are wider or open (as in one bin for X<k), that doesn't work. I guess my question is whether there's any test which works directly on the histogram instead of creating dummy data first. | |
| May 22, 2014 at 2:59 | history | answered | Nick Stauner | CC BY-SA 3.0 |