Roger
As you have stated that odds of having 6+ quakes is 0.6 ,in window period of + or - one day.
Now here are some facts
If I take period from 1-1- 2000 to 31-12 2015 ie 16 years ie 1946 windows of 3 days
Now as per USGS
1) 6 and 6+ quakes in 16 years are 2544
2)6.1+.... Quakes are 1975
3)6.2+..............1537
4)6.3+......... 1230
5)6.4+.......966
6)6.5+..........779
Thus odds are quakes/windows are
1.3, 1.01, 0.789 ,0.632, 0.496 , 0.410
Now , as per this ,there will alwaysbe a 6 or 6 .1+ quake in 3 day window . Which never happens
The problem with this stats is it does not take in to account clusters of quakes.
On a prone date there will be several 6+ quakes in one window ,
I think ,correct picture will be arrived if we take one quake in window ,even though there are several.
Amit
As you have stated that odds of having 6+ quakes is 0.6 ,in window period of + or - one day.
Now here are some facts
If I take period from 1-1- 2000 to 31-12 2015 ie 16 years ie 1946 windows of 3 days
Now as per USGS
1) 6 and 6+ quakes in 16 years are 2544
2)6.1+.... Quakes are 1975
3)6.2+..............1537
4)6.3+......... 1230
5)6.4+.......966
6)6.5+..........779
Thus odds are quakes/windows are
1.3, 1.01, 0.789 ,0.632, 0.496 , 0.410
Now , as per this ,there will alwaysbe a 6 or 6 .1+ quake in 3 day window . Which never happens
The problem with this stats is it does not take in to account clusters of quakes.
On a prone date there will be several 6+ quakes in one window ,
I think ,correct picture will be arrived if we take one quake in window ,even though there are several.
Amit
31 comments:
Amit;
To get odds you count windows containing 1 or more quakes and divide by total number of windows.
For mag 6+ quakes in a 1 day window, odds are 0.285. A 2 day window is 0.476, a 3 day window is 0.611 and so on.
Multiple hits do not count unless predicted.
Roger
Amit;
Here's the table of odds.
Quake data from NEIS, 1973 to 2017
inclusive
Window Mag 6+ Mag 7+ Mag 8+
(Days)
1 0.285 0.039 0.003
2 0.476 0.075 0.005
3 0.611 0.113 0.008
4 0.718 0.145 0.010
5 0.783 0.179 0.013
6 0.836 0.208 0.015
7 0.873 0.245 0.019
Roger
Amit;
Here's the table of odds.
Quake data from NEIS, 1973 to 2017
inclusive
Window Mag 6+ Mag 7+ Mag 8+
(Days)
1 0.285 0.039 0.003
2 0.476 0.075 0.005
3 0.611 0.113 0.008
4 0.718 0.145 0.010
5 0.783 0.179 0.013
6 0.836 0.208 0.015
7 0.873 0.245 0.019
Roger
Roger
Can you give odds for 6.1+ ,6.2+,
6.3+,6.4+ and 6.5+ quakes
Amit
Amit;
Yes. 3 day windows?
Roger
Amit;
Same NEIC quakes, 3 day windows
6.1+ 0.537
6.2+ 0.462
6.3+ 0.397
6.4+ 0.333
6.5+ 0.283
I can use any day range or mag range desired.
6.1+ is 6.1 to 10.0 but 6.1 to 8.3 is equally possible.
Roger
Amit;
Another quake today, a miss.
You hit 6/13 quakes in 2/4 predictions.
Since the odds on a 6+ in a 3 day window are 0.611 you're doing just chance
Roger
Roger
It can not be termed as a miss
A miss is one which I have predicted but no quake occurs in window period
A date which is not predicted ,but a quake occcurs-need to be termed seperately
Amit
Amit;
It's a prediction you failed to make but that's just a comment, not part of your score.
It does affect the hit/miss ratio so instead of 6/12 quakes it's 6 of 13 quakes.
Roger
Roger
This confims that, there will be a major quakes between 7th and 8th day
Amit
Amit;
Wrong again, no major quakes those days.
I've finally completed my probability tables.
They cover all mag 6.0+ quakes from 1973 to 2017
and list window lengths from 1 to 30 days and lower limits from 6.0+ to 9.9+
Clusters are eliminated. I divide the total days into window sized blocks
and check each quake to see which block it belongs. The odds are then the number of
blocks containing quakes divided by the total number of blocks. The number in a given
block does not enter into the calculation.
This method was devised by Dr Alan Jones of New York University who called it
"observed probability" since no estimates are involved, just the actual observed quakes.
I'll send you a copy if you wish. It's 4 pages long.
Roger
Roger
Thanks
Yes pl
Mail findings to me
Obliged
Amit
Amit;
Also, you only caught 1/9 quakes so far this month.
Chance for sure.
Roger
Amit;
My mistake.
For December, 4/9 quakes are in windows, 2/4 windows contain quakes.
Still just chance.
Roger
Amit;
I just finished a total evaluation of all your predictions from 2010 thru 2018.
Since window length varies so much I chose to consider each day as a separate prediction.
The odds are then prediction days divided by total days.
A hit is a day containing one or more quakes. The number of quakes per day doesn't matter.
The expected number of hits is the number of quakes times the odds on a hit. If the observed hits are more than expected you're doing better than chance.
You had 5 years better and 5 years worse so overall you haven't demonstrated any ability.
I can send copies if you would like to check for accuracy.
Roger
Roger
Pl send copy
Also pl state which are the years where prediPredi is below average
Actually ,what I was talking was different
1)take up a meeting
2)calculate percenrage of duration covered by the windows in thatm onth
Say if three Prediction on a month
9 days/30 days time I have covered
3) find out 6+quakes ,(all 6+) in my window period
4)find total 6+ quakes in thatonth
5) take a ratio of quakes in my window decided by total quakes in that my
6)see if it is more than 9/30 or less than that
Thus logically ,I should get more 6+ toatl quakes in my window than average of that month
Amit
Some typo error .this auto spell makes more mistakes
prediPredi......prediction
1)meeting ....month
Amit;
Sent 2 years by email, one good, one bad.
I work in years. Months are too short for statistical work.
Roger
roger
Thanks ,received
1) 2016 year.....as per your evaluation less than average
74 quakes (6+) in the year
366 day sin the year
135 my prediction dates and
31 hits
now
74/366 ie 0.202 quakes per day actually happened in year 2016
and
31/135 ie 0.229 quakes per day
now this seems well above total quakes /total days
2)2018 year
135 total quakes
365 days
ie 0.369 quakes per day
where as my prediction
151 prediction days
and 66 quakes in my window period
ie 0.437 quakes per day which well above total quakes / total days
I do not think any one of these two are below average
regards
Amit
Amit;
Maybe some definitions will help.
A quake day is a day containing one or more 6+ quakes. The number doesn't matter since you don't predict the number. The mag doesn't matter provided it's 6+ or more since you don't predict an exact number.
The odds on a quake day is the number of quake days divided by the number of days in the year.
A prediction day is a day within one of your windows.
A hit is a prediction day containing one or more quakes.
The expected number of hits is the predicted days times the odds on a quake day.
Roger
Roger
But what is wrong with my calculations
A quake per day ....is new definition
Quake per day for a month is numbers of ,6+ quakes in a month devided by number of days in that month
And quake per day for prediction is number of quakes in window period devided by number of days covered by window period in that month
Amit
Amit;
Your definition does not take care of clusters. Suppose you had 30 quakes in the month.
Your method would say odds are 1.0 whereas the real odds could be anything between 1/30 and 30/30 depending on distribution.
Similarly, your hit count could be wrong if only 1 of 6 windows had all the quakes.
Roger
Roger
Pl convey the worst year of my prediction. I will calculate manually as per your method.
Quakes per day gives above average .
Amit
Amit;
That would be 2011, I'll email it.
Quakes per day is not the answer, it's hits observed versus hits expected.
Roger
Roger
Here is my manual evaluation for my worst year 2011( as per your calculations)
1) Total windows in year 122
2) 6+ quakes in windows...85
Ratio 85/122 ie 0.6967
3) Total prediction windows for 2011..38
4) 6+ hit windows in my production..27
Ratio 27/38....ie 0.7105
Which is above odds
Probably so many quakes on 11th March 2011,your software shows my prediction as below odds
Amit
Amit;
Yes, that's your mistake. The number of quakes in a prediction day does not count.
It's one hit per day if any quakes are present. You aren't predicting the number of quakes
so you can't take credit for them.
Roger
Roger
I did exactly as you have told
Details are like this
1) took a physical calender of 2011
2) mark January 1, 2, and 3 as one window...like wise mark all windows (122) of the year
3) check 6+ quakes from USGS for 2011
4)tick the first window ie ,1, 2 and 3 January 2011 if any one or more 6+ quakes are in the first window
5) Get total of such tick marked windows in 2011 ( those are 85)
6)take a list of my prediction date and mark window of that date (+ or - 1 day)
(38 windows of my prediction)
7) similarly to above tick mark my window if one or more 6+ quakes seen in my windows ( clusters and multiple quakes not counted in both cases)..
27 such hit windows
Now compare both ratios
85/122.......0.6967
27/38...... ..0.7105
Now you can compare both
Amit
Amit;
We can't seem to agree on anything except the number of 3 day windows in a year (122).
I count 89 windows with quakes, 33 prediction windows to your 38, and 18 hits to your 27.
89/122 = .729
18/33 = .545 so you lose again.
How can we resolve this?
Roger
Roger
I have done pains taking laborious work .Hence there are less chance of software mistake . I can post you each window details.
Unlike your ready CSV file I have to type all details and post.
I will find some time and will do that
Amit
Roger
As a first step here are my predicted dates for 2011
1) January...19,25
2) Feb.....3,19,21
3)March.....20
4)April....3, 9/10, 17
5) May.....7,16,29
6)June.....9,12, 15/16,30/1
7)July......1,9,31
8) August....6,13,22,30
9)Sept.....2,5,12,16,27/28
10) October....12/13 ,21,26/27
11) November...10,13,24/25
12)Decembee.....6,10,20,24
Total...38
ROGER
Received your e mail with revised calculation
It is not known why you have changed evaluation from Window based calculations to date wise calculation
As you have already told that ,whatever I have predicted will be taken , I have predicted + or - one day window
so why not window based calculations
Amit
Post a Comment