tsh
Level 2 Rank
Posts: 69
|
Post by tsh on Jul 2, 2010 9:51:16 GMT
And that seems to be your only argument. Random selection is better than measuring all. Very poor. Clearly you understand nothing of which you pontificate. Random selection is just as good as exhaustive selection - the majic of statistics allows people who understand these things to determine from the random selection how likely it is to be representative of the whole.
|
|
jinki
Level 3 Rank
Posts: 123
|
Post by jinki on Jul 2, 2010 10:01:52 GMT
And that seems to be your only argument. Random selection is better than measuring all. Very poor. Clearly you understand nothing of which you pontificate. Random selection is just as good as exhaustive selection - the majic of statistics allows people who understand these things to determine from the random selection how likely it is to be representative of the whole. The word "understand" seems to be overused. Your statement is incoherent. What we are trying to achieve is whether the Sun is losing magnetic strength which supposedly will lead to the Sun not capable of producing any spots by 2015. Describe to me how a random selection process is a better method of measuring this trend against measuring every spot that makes the grade.
|
|
|
Post by ncfcadam on Jul 2, 2010 10:04:04 GMT
Strawman. Nobody has said that random selection is better than measuring everything. It's physically impossible to measure all spots anyway because we can't see those on the side of the sun facing away from us. However, random selection is better than biased selection. Your strawman is introducing farside activity. Good to see you recognize random selection is inferior. Your description of bias selection is incorrect. Measuring every spot removes any bias along with any concerns of missing information. But you're not measuring every spot. You are discarding all but the darkest in each group. That's the selection bias you are introducing.
|
|
jinki
Level 3 Rank
Posts: 123
|
Post by jinki on Jul 2, 2010 10:10:51 GMT
Your strawman is introducing farside activity. Good to see you recognize random selection is inferior. Your description of bias selection is incorrect. Measuring every spot removes any bias along with any concerns of missing information. But you're not measuring every spot. You are discarding all but the darkest in each group. That's the selection bias you are introducing. No, that is incorrect. As stated to Leif with a similar question any group that makes the 23 pixel grade is measured in total. The entire group(every pixel) is measured in total. There is no bias.
|
|
|
Post by ncfcadam on Jul 2, 2010 10:28:36 GMT
But you're not measuring every spot. You are discarding all but the darkest in each group. That's the selection bias you are introducing. No, that is incorrect. As stated to Leif with a similar question any group that makes the 23 pixel grade is measured in total. The entire group(every pixel) is measured in total. There is no bias. And recording only the highest measurement. That is the bias. And the 23 pixel limit is arbitrary: why discard smaller spots? They are still huge in reality.
|
|
jinki
Level 3 Rank
Posts: 123
|
Post by jinki on Jul 2, 2010 10:56:06 GMT
No, that is incorrect. As stated to Leif with a similar question any group that makes the 23 pixel grade is measured in total. The entire group(every pixel) is measured in total. There is no bias. And recording only the highest measurement. That is the bias. And the 23 pixel limit is arbitrary: why discard smaller spots? They are still huge in reality. The highest measurement tells us the solar potential, we could also measure the lowest measurement which might be informative but perhaps not stating what we need to know. If the gauss measurement is going to drop below 1500 we need to know the top end , not the bottom. The L&P method apart from being random tells us we have more specks, we all know that, and the need is to isolate and quarantine these specks as they will bias the record. Specks wont stop a natural rise in magnetic activity however weak it may be as we reach cycle maximum.
|
|
radun
Level 3 Rank
Posts: 152
|
Post by radun on Jul 2, 2010 11:18:38 GMT
Critical boundaries are just as solid now as they were 4-5 years ago. When an argument fails then ‘cherry picking’ is evoked.
|
|
|
Post by lsvalgaard on Jul 2, 2010 14:05:06 GMT
But you're not measuring every spot. You are discarding all but the darkest in each group. That's the selection bias you are introducing. No, that is incorrect. As stated to Leif with a similar question any group that makes the 23 pixel grade is measured in total. The entire group(every pixel) is measured in total. There is no bias. The bias is the 23 pixel limit.
|
|
|
Post by lsvalgaard on Jul 2, 2010 14:09:10 GMT
And recording only the highest measurement. That is the bias. And the 23 pixel limit is arbitrary: why discard smaller spots? They are still huge in reality. The highest measurement tells us the solar potential, we could also measure the lowest measurement which might be informative but perhaps not stating what we need to know. If the gauss measurement is going to drop below 1500 we need to know the top end , not the bottom. The L&P method apart from being random tells us we have more specks, we all know that, and the need is to isolate and quarantine these specks as they will bias the record. Specks wont stop a natural rise in magnetic activity however weak it may be as we reach cycle maximum. The specks are the record. The bottom is the important part. There is no L&P 'method', they just observe when they have telescope time and the weather permits.
|
|
jinki
Level 3 Rank
Posts: 123
|
Post by jinki on Jul 2, 2010 14:56:54 GMT
The bias has moved from measuring every spot at their maximum to not measuring speck activity that exists under 23 pixels?
|
|
|
Post by lsvalgaard on Jul 2, 2010 16:27:02 GMT
The bias has moved from measuring every spot at their maximum to not measuring speck activity that exists under 23 pixels? Perhaps a little simulation will do: Gauss=EXP(-1*(-A7+INT(A7)+0.5)^2/C7)*7000*(1-RAND())*0.5 C7=0.45-0.000064*(A7-1995) Where A7 is the year and we step 0.1 year in time. The formula generates ten values per year picked at random from a normal Gaussian distribution with a spread C7 that changes with time [this is one interpretation of the L&P effect - you can use others, this is only for illustration]: As you can see, the top values are not affected, but more and more low values will occur. Those below 1500G [red line] we cannot see. The black ragged curve [and line] is the running 12-month mean. Whenever you introduce a threshold, you introduce bias. The sun does that itself by not showing us all the spots and specks below 1500G. The L&P effect is precisely about that bias introduced by the Sun, which makes the SSN useless as solar activity indicator. Your additional bias just makes that worse [which is, of course, why you do it, to make SC24 look like SC5 or worse]
|
|
radun
Level 3 Rank
Posts: 152
|
Post by radun on Jul 2, 2010 19:45:20 GMT
|
|
jinki
Level 3 Rank
Posts: 123
|
Post by jinki on Jul 3, 2010 1:24:15 GMT
"Whenever you introduce a threshold, you introduce bias. The sun does that itself by not showing us all the spots and specks below 1500G. The L&P effect is precisely about that bias introduced by the Sun, which makes the SSN useless as solar activity indicator. Your additional bias just makes that worse [which is, of course, why you do it, to make SC24 look like SC5 or worse]"
The top values are increasing as SC24 takes hold, it will take the top values to drop under 1500 gauss for the L&P prediction to come to fruition (what was the title of the paper again?). Nothing else needs to be taken into consideration.
The 23 pixel limit in regard to comparing SC24 to SC5 is to align the method with Wolf's reconstruction as you well know. Your own research agrees with this, unless you would like to remove it from the record now?
|
|
|
Post by lsvalgaard on Jul 3, 2010 5:56:46 GMT
|
|
|
Post by lsvalgaard on Jul 3, 2010 6:03:55 GMT
"Whenever you introduce a threshold, you introduce bias. The sun does that itself by not showing us all the spots and specks below 1500G. The L&P effect is precisely about that bias introduced by the Sun, which makes the SSN useless as solar activity indicator. Your additional bias just makes that worse [which is, of course, why you do it, to make SC24 look like SC5 or worse]"The top values are increasing as SC24 takes hold, it will take the top values to drop under 1500 gauss for the L&P prediction to come to fruition (what was the title of the paper again?). Nothing else needs to be taken into consideration. The 23 pixel limit in regard to comparing SC24 to SC5 is to align the method with Wolf's reconstruction as you well know. Your own research agrees with this, unless you would like to remove it from the record now? 1: the values for SC5 are very uncertain [lack of data] by a factor of two. 2: All Wolf's [and Wolfer's and Brunner's] numbers should be increased by 20% anyway. The top values are not important as I just showed, as the sunspot number is determined by the smaller spots. For one big one in a group there can be up to a hundred small ones in the same group. A reasonable interpretation of L&P would be [just like during the Maunder Minimum] that very big spots would still be seen, but all the small ones are gone. Your threshold [as any threshold] introduces bias. Of course, some people want bias, so it is understandable why you do it.
|
|