It was in relation to the BuMet.

I think the temperature distributions shifting to the warm side is pretty significant, statistically significant too I imagine.

Statistics are descriptors of a sample, they are not a model.

I have used the term “bell curve” in relation to the above BuMet graphic. I have not used the term “normal distribution” because that has special meaning:

Quote:

(wiki)

So a Normal Distribution is one that fits the Gaussian Probability distribution. Very symmetrical curves. If you look at the 1951–1980 maximum temperature distribution you see there is almost 2 peaks, one just to the left of the main peak. Also I talked of skew—a normal distribution is perfectly symmetrical.

If the distribution of worldwide minimum or maximum temperatures were graphed that distribution would be much closer to the normal distribution—bigger samples give more consistent results. A statistician would normally not bother working on samples of under 700 values.

That quote, about natural and social sciences, is not phrased as well as it might. A researcher would need to satisfy himself that the distribution of his sample is at least approximately normal: symmetrical, bell curve shape and the number of values one, two and three standard deviations from his mean. Mean, median and mode all close together etc. Normality is not just assumed for any old sample!

Statistic is a description of a sample: sample mean and sample standard deviation, sample skew, sample kurtosis etc. One statistic is range, the difference between the endpoints of the distribution. Useful to know, but useless in tests unless some advances have been made I don’t know about.

Where we have a population (census results) we talk of parameters! Population mean etc.

Stats are fun!

If you have two dice and toss them 100 times and graph the results you will see an approximation of a distribution called the Bernoulli distribution (look it up in wiki) Do the dice tosses 1000 times and your graph will look a lot more like the theoretical Bernoulli distribution. (The range, of course is from 2 to 12.)

Add one, two. . .etc dice and your distribution merges into—the normal distribution!

And some use these random tossing to guide their lives: the I Ching 

lee wrote on Apr 29^th, 2018 at 2:35pm:

I think I pointed out that the wiki quote wasn’t 100% accurate.

But with sufficiently large samples a distribution can generally be assumed to be approximately normal. If you want to be picky there are tests for normality.

Jovial Monk wrote on Apr 29^th, 2018 at 3:31pm:



So it is the quote that isn't 100% accurate' not the results?

Jovial Monk wrote on Apr 29^th, 2018 at 3:31pm:



Oh,oh. Approximations aren't measurements. What are the error bars on these approximations?

Jovial Monk wrote on Apr 29^th, 2018 at 3:31pm:



Of course there are and they are not 100% accurate either.

So if I ask Aussie to delete all posts after reply 499 nothing of value will go?

Oh dear.JM is stuck in a rut.

Jovial Monk wrote on Apr 29^th, 2018 at 5:18pm:



You can say what you want. Doesn't mean it has meaning.

Of course cherry picking 104 sites out of how many?