"Но сигнал прет не останавливаясь с темпом 0.2Ц/декаду"
It's not a measurable thing. It's a statistical estimate. What's the confidence interval for that?
Using the data I quoted as an example, to make it more obvious what I am asking:
Is the mean temperature 1901..1950 different from 1950..2019? Given that data, we can reject the hypothesis that the mean is the same with confidence level of 0.288. This is rubbish confidence level. Roughly speaking, we accept the means are different (temperature changed) with the probability of 28.8% that we make a mistake in doing so.
Can we estimate the temperature difference? Yes, just the difference of means. Given that data it is 0.16C. Well, it's not global temperature, so don't worry that it does not match the 0.8 per century, the clincher is the following:
What is the confidence interval for the temperature difference estimate? The confidence interval at the level 0.288 should be (roughly?) equal to the temperature difference estimate (I am not too certain how many tails for t-distribution to use here, need to check; but that's not wildly different, whichever tails you pick).
That is, you can get confidence interval smaller than the actual difference of 0.16 C only for p much larger than 0.288.
You can do the same to, say, 30-year intervals, 1901..1930 vs 1990..2019. That's better. Given that data, the 30-year means changed with confidence level of 0.09. This starts getting into interesting confidence range. The estimate of the temperature difference is 0.5C. But at 0.1 it is still +/- 0.5 C, and at confidence level of 0.25 it is +/- 0.25 C.
But then, say, 1970..2000 vs 1990..2019 - to be able to claim that it is "speeding up" - is not very good. The difference is there, yes, but at the confidence level of 0.33. The difference is 0.19 C, yes, but the confidence interval is very large.
Ok?
So when someone talks about "сигнал прет", we are talking about the estimate of difference of means going up. But what is happening to the confidence interval?
no subject
It's not a measurable thing. It's a statistical estimate. What's the confidence interval for that?
Using the data I quoted as an example, to make it more obvious what I am asking:
Is the mean temperature 1901..1950 different from 1950..2019? Given that data, we can reject the hypothesis that the mean is the same with confidence level of 0.288. This is rubbish confidence level. Roughly speaking, we accept the means are different (temperature changed) with the probability of 28.8% that we make a mistake in doing so.
Can we estimate the temperature difference? Yes, just the difference of means. Given that data it is 0.16C. Well, it's not global temperature, so don't worry that it does not match the 0.8 per century, the clincher is the following:
What is the confidence interval for the temperature difference estimate? The confidence interval at the level 0.288 should be (roughly?) equal to the temperature difference estimate (I am not too certain how many tails for t-distribution to use here, need to check; but that's not wildly different, whichever tails you pick).
That is, you can get confidence interval smaller than the actual difference of 0.16 C only for p much larger than 0.288.
You can do the same to, say, 30-year intervals, 1901..1930 vs 1990..2019. That's better. Given that data, the 30-year means changed with confidence level of 0.09. This starts getting into interesting confidence range. The estimate of the temperature difference is 0.5C. But at 0.1 it is still +/- 0.5 C, and at confidence level of 0.25 it is +/- 0.25 C.
But then, say, 1970..2000 vs 1990..2019 - to be able to claim that it is "speeding up" - is not very good. The difference is there, yes, but at the confidence level of 0.33. The difference is 0.19 C, yes, but the confidence interval is very large.
Ok?
So when someone talks about "сигнал прет", we are talking about the estimate of difference of means going up. But what is happening to the confidence interval?