PDA

View Full Version : Testing extrapolation



PHAT
30-11-2004, 09:22 PM
ID# .Glik04sep.....1 ....2 ....3 ....4 ....5 ....6 ....7 ....8 .... 9 ....10
2016122 1792 1772 1774 1775 1776 1766 1757 1751 1747 1776 1738
2015956 1311 1295 1299 1301 1303 1263 1247 1235 1225 1310 1308
2015116 1642 1626 1625 1625 1625 1610 1596 1585 1574 1636 1624
2014932 2095 2084 2085 2086 2087 2072 2062 2055 2049 2093 2085
2013230 2334 2329 2334 2336 2339 2321 2318 2317 2322 2334 2334
2012261 1541 1528 1528 1528 1527 1519 1508 1500 1492 1536 1514
2011760 1507 1504 1506 1507 1509 1492 1494 1500 1523 1507 1507
2011056 1830 1831 1829 1827 1826 1821 1812 1805 1799 1830 1830
1291535 1163 1181 1182 1182 1183 1191 1200 1207 1212 1172 1191
1281421 1985 2025 2021 2020 2018 2034 2050 2062 2071 2018 2087
1281340 1832 1845 1845 1845 1844 1859 1864 1866 1866 1837 1844
1279615 2037 2074 2073 2073 2072 2084 2098 2107 2115 2060 2099
1276956 1650 1653 1653 1652 1652 1661 1666 1670 1672 1650 1652
1276422 1948 1981 1978 1976 1974 1998 2015 2028 2038 1964 1990
1275884 2075 2086 2089 2090 2093 2096 2110 2121 2134 2078 2095
1275405 1228 1220 1217 1215 1214 1218 1215 1212 1208 1225 1216
1274360 1656 1653 1654 1654 1654 1659 1663 1667 1670 1656 1655
1274013 1141 1112 1112 1112 1112 1091 1072 1057 1043 1126 1091
1273475 1301 1293 1293 1293 1294 1265 1258 1253 1254 1301 1300
1272716 1870 1873 1871 1870 1869 1890 1896 1900 1903 1870 1870
1272602 1295 1287 1286 1285 1284 1268 1254 1242 1232 1295 1293
1271784 1672 1652 1655 1656 1658 1636 1625 1618 1612 1664 1650
1270583 1416 1404 1406 1407 1408 1393 1386 1381 1378 1413 1405
1270325 1745 1731 1732 1733 1734 1719 1710 1705 1700 1741 1727
1194316 1560 1549 1547 1546 1545 1545 1538 1532 1526 1557 1548
1011082 1754 1725 1726 1726 1727 1717 1705 1696 1688 1734 1687
2059200 1771 1774 1772 1771 1770 1780 1777 1763 1755 1772 1771
2059185 1024 1012 1009 1008 1006 991 973 958 944 1023 1014
2058636 1432 1430 1430 1429 1429 1438 1437 1425 1422 1432 1431
2058010 1398 1389 1388 1388 1388 1382 1374 1368 1363 1396 1391
2057181 1661 1657 1655 1655 1653 1649 1644 1640 1636 1660 1658
2057096 1463 1452 1453 1454 1455 1436 1429 1423 1419 1461 1458
2055943 2400 2400 2399 2398 2398 2408 2418 2425 2431 2400 2400
2055814 1790 1771 1772 1772 1773 1762 1752 1744 1736 1778 1752
2054963 2232 2237 2237 2237 2237 2244 2248 2251 2253 2233 2235
2054145 2203 2218 2215 2214 2212 2232 2242 2249 2254 2207 2214
2054090 1532 1518 1517 1516 1515 1508 1495 1485 1476 1526 1506
2053692 1660 1642 1647 1649 1652 1620 1609 1602 1599 1657 1655
2053563 2131 2142 2140 2139 2137 2156 2161 2163 2162 2133 2137
2052992 1914 1907 1906 1905 1905 1896 1886 1879 1873 1913 1911
2052244 1644 1644 1642 1641 1640 1664 1666 1663 1656 1644 1642
2051941 1527 1519 1522 1524 1526 1497 1491 1488 1488 1527 1527
2051402 1866 1846 1845 1844 1844 1834 1824 1817 1811 1860 1847
2050842 2050 2049 2049 2049 2050 2061 2069 2073 2079 2049 2049
2050816 1528 1516 1518 1519 1520 1495 1489 1487 1487 1526 1525
2050690 2004 1997 1998 1998 1998 1966 1951 1939 1929 2004 2004
2050524 1731 1717 1715 1715 1713 1702 1687 1675 1664 1725 1710
2050491 2197 2191 2190 2189 2188 2184 2179 2175 2171 2195 2192
2049776 1468 1458 1461 1463 1464 1454 1446 1440 1435 1465 1453
2047224 1907 1913 1908 1906 1903 1938 1945 1949 1952 1907 1905
2045404 1995 1980 1979 1978 1976 1964 1949 1938 1927 1991 1977
2044052 1799 1785 1784 1784 1783 1774 1765 1758 1752 1793 1779
2041635 1642 1626 1626 1626 1627 1604 1587 1574 1563 1638 1628
2039911 1724 1714 1714 1714 1714 1700 1697 1695 1696 1723 1722
2039756 1626 1625 1625 1625 1626 1619 1616 1615 1615 1626 1626
2039616 1553 1540 1540 1539 1539 1529 1516 1507 1494 1550 1539
2039432 1421 1411 1416 1419 1423 1389 1382 1378 1379 1421 1421
2035803 1975 1966 1966 1965 1964 1949 1938 1929 1921 1974 1969
2035103 2641 2642 2641 2640 2640 2654 2660 2664 2669 2641 2641
2034042 1675 1663 1660 1659 1656 1659 1643 1629 1616 1670 1641
2034005 1410 1405 1403 1402 1402 1393 1396 1407 1424 1409 1410
2033994 1365 1341 1340 1339 1338 1318 1295 1278 1262 1361 1345
2033666 1824 1809 1805 1804 1801 1811 1794 1783 1769 1814 1779
2033106 1873 1852 1855 1856 1858 1828 1815 1807 1802 1869 1856
2032734 1440 1421 1420 1420 1420 1412 1406 1401 1398 1428 1409
2031555 1784 1767 1766 1766 1767 1756 1753 1754 1763 1779 1772
2031345 1425 1415 1415 1415 1415 1408 1401 1395 1390 1420 1403
2028512 1705 1695 1691 1690 1687 1698 1688 1680 1672 1699 1679
2026946 1913 1897 1895 1895 1895 1882 1874 1868 1863 1908 1901
2026600 1646 1628 1630 1631 1632 1608 1599 1595 1593 1642 1635
2026482 1043 1023 1022 1021 1021 1017 1016 1016 1019 1029 1003
2025701 1531 1519 1519 1520 1520 1497 1486 1478 1472 1529 1524
2025163 2073 2078 2077 2077 2077 2084 2089 2093 2096 2075 2077
2024695 1927 1911 1914 1915 1917 1890 1875 1863 1854 1924 1917
2024242 1817 1815 1812 1811 1809 1838 1843 1847 1844 1817 1816
2024065 1609 1599 1600 1601 1601 1592 1588 1585 1583 1604 1597
2022245 1390 1372 1374 1374 1375 1350 1335 1324 1314 1386 1377
2021523 1752 1756 1758 1758 1760 1740 1746 1763 1775 1754 1767
2020403 1597 1572 1571 1571 1570 1560 1542 1529 1517 1588 1553

Alan Shore
30-11-2004, 09:33 PM
......

Um.. so what's it all mean?

Could it be the Da Vinci Code? ;)

PHAT
30-11-2004, 09:40 PM
I will try to tidy the first post later.

The table above shows the Sep04 rating of 79 players and ten (10) ratings that could have been their published ratings had their performance over time been extrapolated using ten (10) different algorithms.

What I ask of The ACF ratings officer is for the performance rating of each of these 79 players during the last rating period. Using that information, I would like to determine if the Glicko2 rating for Sep2004 is a better pedictor by itself than with my own modifications.

Of the 10 algorithms, I expect only 2 or 3 to lead to a better prediction of the players' actual performance in the Sep2004 period. The other 7 or 8 are mear experiments intended to reveal mathematical patterns to the experimentor (me).

Please note that these 10 predictions for 79 players, is of their [i]actual performance, not of the Glicko2 generated Dec2004 list.

Garvinator
30-11-2004, 09:44 PM
I will try to tidy the first post later.

The table above shows the Sep04 rating of 79 players and ten (10) ratings that could have been their published ratings had their performance over time been extrapolated using ten (10) different algorithms.

What I ask of The ACF ratings officer is for the performance rating of each of these 79 players during the last rating period. Using that information, I would like to determine if the Glicko2 rating for Sep2004 is a better pedictor by itself than with my own modifications.

Of the 10 algorithms, I expect only 2 or 3 to lead to a better prediction of the players' actual performance in the Sep2004 period. The other 7 or 8 are mear experiments intended to reveal mathematical patterns to the experimentor (me).

Please note that these 10 predictions for 79 players, is of their [i]actual performance, not of the Glicko2 generated Dec2004 list.


would putting your table in code tags and disabling smilies make it easier to read? have you tried that?

Ill answer my own post as I just tried it, no it doesnt.

PHAT
30-11-2004, 09:56 PM
would putting your table in code tags and disabling smilies make it easier to read? have you tried that?

Ill answer my own post as I just tried it, no it doesnt.

I just wanted the figures "on the record" before BGs list comes out. I have it summerised on an Excel spreadsheet. Now I only need performance data from the ACF.

I have am optimistic that I will beat Glicko2 in the Prediction Stakes at the Ratings Arena tonight.

Bill Gletsos
30-11-2004, 10:12 PM
I just wanted the figures "on the record" before BGs list comes out. I have it summerised on an Excel spreadsheet. Now I only need performance data from the ACF.

I have am optimistic that I will beat Glicko2 in the Prediction Stakes at the Ratings Arena tonight.
I just picked 5 at random.

In all 5 cases all of your predictions went the opposite (wrong) way.

Also if the Glicko2 for example drops a player from 1500 to 1450 you can be certain his perfomance rating was at least as low as 1450. Therefore you can use the published rating without needing the "performance" rating.
e.g Canfell's rating is lower than all your estimates. His performance rating was lower than his published rating.

Bill Gletsos
30-11-2004, 10:18 PM
Of course this whole theory is rubbish as was explained last time you tried it.

You cannot determine a players new rating by just extrapolating based only on his past results.
If this was true, there would be no need for the players to play any games at all. You could just extrapolate them every period.

You need his results from the current period.

PHAT
30-11-2004, 10:29 PM
I just picked 5 at random.

In all 5 cases all of your predictions went the opposite (wrong) way.

Don't be silly. 5 random, for direction only, means nothing.


Also if the Glicko2 for example drops a player from 1500 to 1450 you can be certain his perfomance rating was at least as low as 1450.

Please don't tell me how to suck eggs.


Therefore you can use the published rating without needing the "performance" rating.

No I cannot, because, the magnitude - and not just direction - of the discrepency is important to determine quantitative measure of predictivity.

I trust that since you have Cranfeld's preformance rating at hand, you will be able to spend 10 minutes to retrieve the preformance rating of the other 78, some time in the next few days. Just PM them to me. Thanks.

Bill Gletsos
30-11-2004, 10:42 PM
Don't be silly. 5 random, for direction only, means nothing.
Of course it does. It shows you did not even get the direction right for 6% of yout attempts.


Please don't tell me how to suck eggs.
You apparently dont know how to.



Therefore you can use the published rating without needing the "performance" rating.

No I cannot, because, the magnitude - and not just direction - of the discrepency is important to determine quantitative measure of predictivity.
Actually you cannot use that as a true measure of predictive accuracy.
As has been explained many times on the bb the true measure of predictive accuracy is how close a players predicted score matches his actual score.


I trust that since you have Cranfeld's preformance rating at hand, you will be able to spend 10 minutes to retrieve the preformance rating of the other 78, some time in the next few days. Just PM them to me. Thanks.
Thats Canfell you idiot and no I dont know his performance rating. I know his new Dec rating. I dont have to know his performance rating to know that the rating system would not drop his Dec rating below his performance rating. Hence I know his performance rating must be equal to or below his published rating.

Lets see how many you get in the right direction first.
If you get a decent number correct, I might be inclined to determine the "performance" ratings.

I'm not going to waste my time on what is a worthless pursuit.

It is after all obvious that you cannot predict a trend.
Just because a players rating went up one period or even in successive periods does not mean it will go up in the next period. The same is true if it went down.

PHAT
30-11-2004, 10:49 PM
It is after all obvious that you cannot predict a trend.
Just because a players rating went up one period or even in successive periods does not mean it will go up in the next period. The same is true if it went down.

This statement is likely to be the mill stone that drags on you forever. :lol:

Bill Gletsos
30-11-2004, 10:53 PM
This statement is likely to be the mill stone that drags on you forever. :lol:
You know nothing about ratings just like you know nothing about other areas of chess.

PHAT
30-11-2004, 11:11 PM
You know nothing about ratings just like you know nothing about other areas of chess.

Gratuitous abuse from the President and spokesman for the NSWCA.

Not a good look for NSW chess administration.

Bill Gletsos
30-11-2004, 11:14 PM
Gratuitous abuse from the President and spokesman for the NSWCA.

Not a good look for NSW chess administration.
Its not gratuitous abuse from me at all as President of the NSWCA.

Its a statement of fact from me as the ACF Ratings Officer.

PHAT
30-11-2004, 11:16 PM
You need his results from the current period.

Only to correct the Glicko prediction for next period.

PHAT
30-11-2004, 11:25 PM
Its not gratuitous abuse from me at all as President of the NSWCA.

Its a statement of fact from me as the ACF Ratings Officer.

OK, from an official of Australia's peak chess body, the ACF.

Again, not a good look.

Bill Gletsos
30-11-2004, 11:35 PM
Only to correct the Glicko prediction for next period.
Player A rated 1500 performs at the 1600 level in the next period. In the following period he performs at 1650. What should player A's extrapolated rating be for the next period.

Player B is rated 1500 and performs at the at the 1600 level in the next period. In the following period he performs at 1650. What should player B's extrapolated rating be for the next period.

Player C rated 1500 performs at the 1600 level in the next period. In the following period he performs at 1650. What should player C's extrapolated rating be for the next period.

Player B is rated 1500 and performs at the at the 1600 level in the next period. In the following period he performs at 1650. What should player B's extrapolated rating be for the next period.

In actuality Player A's performance is 1600 whilst player B plays at 1550 and player C at 1700.

The point is there is no way to trend the ratings of players A and B and C.

Their past results do not indicate how they will perform in the next period just how well they should perform.

Just because a player does not perform to their rating does not necessarily mean their rating is incorrect, just that they no longer performed at their previous level.

Bill Gletsos
30-11-2004, 11:36 PM
OK, from an official of Australia's peak chess body, the ACF.

Again, not a good look.
Its a good look in relation to you.

PHAT
01-12-2004, 07:53 AM
Its a good look in relation to you.

Don't try to excuse yourself. Your behaviour, as an ACF official and a State President here, does chess no favors. How about you drop it - on this thread at least.

Bill Gletsos
01-12-2004, 09:44 AM
Don't try to excuse yourself. Your behaviour, as an ACF official and a State President here, does chess no favors. How about you drop it - on this thread at least.
You have abused all and sundry on this bulletin board and me in particular.
Your behaviour on this BB has certainly done chess no favours.
You have shown your complete lack of understanding of most chess related issues on numerous occasions.

The simple matter is what I said isnt abuse but just a simple statement of fact. You have no clue when it comes to ratings. This is amply demonstrated by your response to Kevin's rating question he directed to DR.

As far as I am aware no where in the world is anyone with any knowledge of ratings or ratings theory ever mentioned the idea of trending as a basis for ratings calculations. The reason why is that it has no basis in theory or fact.

Ian Rout
01-12-2004, 11:07 AM
The table above shows the Sep04 rating of 79 players and ten (10) ratings that could have been their published ratings had their performance over time been extrapolated using ten (10) different algorithms.

What I ask of The ACF ratings officer is for the performance rating of each of these 79 players during the last rating period. Using that information, I would like to determine if the Glicko2 rating for Sep2004 is a better pedictor by itself than with my own modifications.

Of the 10 algorithms, I expect only 2 or 3 to lead to a better prediction of the players' actual performance in the Sep2004 period. The other 7 or 8 are mear experiments intended to reveal mathematical patterns to the experimentor (me).

Please note that these 10 predictions for 79 players, is of their [i]actual performance, not of the Glicko2 generated Dec2004 list.

I'm pretty sure one of the ACT players on the list hasn't played this quarter so that may be a bit hard to get a trend for (or very easy, whichever way you look at it).

I notice that at least for the ACT players it doesn't look like a random sample, does it deliberately include nearly every active 40+ player? (and if so why am I not on it?)

PHAT
02-12-2004, 08:41 AM
You have abused all and sundry on this bulletin board and me in particular.
Your behaviour on this BB has certainly done chess no favours.
You have shown your complete lack of understanding of most chess related issues on numerous occasions.


None of this excuses your behaviour.

I accept the aprobrium that my behaviour might earn.

You should do likewise.

PHAT
02-12-2004, 08:50 AM
As far as I am aware no where in the world is anyone with any knowledge of ratings or ratings theory ever mentioned the idea of trending as a basis for ratings calculations. The reason why is that it has no basis in theory or fact.

Firstly, you misrepresent my positon. I do not say that trending should/could be the basis for rating calculations. I say that it can be a refinement step.

Secondly, measuring trends in the real world is done all the time in science - global warmng, fish stocks, cancer rates, bioacumulation, cognative decline! There is no theoretical reason to exclude sporting performance.

PHAT
02-12-2004, 08:53 AM
Player A rated 1500 performs at the 1600 level in the next period. In the following period he performs at 1650. What should player A's extrapolated rating be for the next period.

Player B is rated 1500 and performs at the at the 1600 level in the next period. In the following period he performs at 1650. What should player B's extrapolated rating be for the next period.

Player C rated 1500 performs at the 1600 level in the next period. In the following period he performs at 1650. What should player C's extrapolated rating be for the next period.

Player B is rated 1500 and performs at the at the 1600 level in the next period. In the following period he performs at 1650. What should player B's extrapolated rating be for the next period.

In actuality Player A's performance is 1600 whilst player B plays at 1550 and player C at 1700.

The point is there is no way to trend the ratings of players A and B and C.

Their past results do not indicate how they will perform in the next period just how well they should perform.

Just because a player does not perform to their rating does not necessarily mean their rating is incorrect, just that they no longer performed at their previous level.

Bill, could you please take another look at this quote. I think you have made a mistake in the figures you use. Players A,B and C are the same example.

PHAT
02-12-2004, 09:13 AM
I'm pretty sure one of the ACT players on the list hasn't played this quarter so that may be a bit hard to get a trend for (or very easy, whichever way you look at it).

I notice that at least for the ACT players it doesn't look like a random sample, does it deliberately include nearly every active 40+ player? (and if so why am I not on it?)

Sampling:

In order to determine if trend anaysis can work at all, I wanted to choose the active players for whom the most data was available. Therefore I chose only players whose;
1. ratings changed in no less than 6 of the last 8 rating periods - change in rating was assumed to indicate activity, and
2. were "!!" or "!" at all times, and
3. had no increases/decreases in a single rating period of >150.

The master files were pasted into Excel and the process was done manually. Starting at the top (smallest ID#) selection was made of the first 80 that met the all 3 criteria. There is sure to be some eligable players whom were missed in this cull.

Any non random attributes in the final sample are due to:
1. The selection criteria, and
2. Non-random nature of ID# to cohort.

Ian, I think that the ACT and 40yrs+ bias must have its root in the allocation of ID#.

I hope this helps

Bill Gletsos
02-12-2004, 09:55 AM
Bill, could you please take another look at this quote. I think you have made a mistake in the figures you use. Players A,B and C are the same example.
No. The example is giving three different players all with the same data.
However their performance in the next period can be totally different for all three. Hence trending is useless.

Bill Gletsos
02-12-2004, 09:58 AM
Firstly, you misrepresent my positon. I do not say that trending should/could be the basis for rating calculations. I say that it can be a refinement step.

Secondly, measuring trends in the real world is done all the time in science - global warmng, fish stocks, cancer rates, bioacumulation, cognative decline! There is no theoretical reason to exclude sporting performance.
There are too many other influences that may affect any so called trend based on factors unrelated to actual chess ability.

Bill Gletsos
02-12-2004, 10:07 AM
Of your 79 players only 73 actually payed games in the period.

Of these 73 players 34 had their ratings go in the completely opposite direction to all 10 of your estimates. As such these 34 can be seen as total failures as not one of your estimates for them went in the right direction.

Of the remaining 39 players, your estimate of their rating only went in the right direction as follows for each of the 10 estimates.


Estimate 1: 31
Estimate 2: 32
Estimate 3: 32
Estimate 4: 31
Estimate 5: 35
Estimate 6: 35
Estimate 7: 37
Estimate 8: 36
Estimate 9: 25
Estimate 10: 28

Given that your hit rate is no better than 50.68% (37/73) I have no intention of wasting my time on this to determine the performance ratings of the players.

Bill Gletsos
02-12-2004, 10:09 AM
None of this excuses your behaviour.

I accept the aprobrium that my behaviour might earn.

You should do likewise.
My behaviour is only highlighting your stupidity and lack of knowledge in chess related matters.
Thats the difference.

arosar
02-12-2004, 10:10 AM
There are too many other influences that may affect any so called trend based on factors unrelated to actual chess ability.

But you can say the same about fish stocks, global warming, cognitive ability, etc, can't you?

Look, all he wants is trend analysis based on the available data. What's the big deal? Just do it already before youse two go on for pages about ratings again.

AR

Bill Gletsos
02-12-2004, 10:11 AM
The following id numbers are in the same order as your first post.
The id's listed with a 0 rating are those that played no games in the Dec period.


2016122 1787
2015956 1324
2015116 1560
2014932 2109
2013230 2311
2012261 1442
2011760 1442
2011056 1775
1291535 1041
1281421 1984
1281340 1802
1279615 2043
1276956 1644
1276422 1931
1275884 2040
1275405 0
1274360 1661
1274013 0
1273475 1311
1272716 1819
1272602 1271
1271784 1666
1270583 0
1270325 1797
1194316 1553
1011082 1771
2059200 1710
2059185 1003
2058636 1438
2058010 1408
2057181 1683
2057096 1461
2055943 2420
2055814 1740
2054963 2257
2054145 2225
2054090 1555
2053692 1675
2053563 2118
2052992 1926
2052244 1611
2051941 1570
2051402 1879
2050842 2064
2050816 1489
2050690 1999
2050524 1741
2050491 2182
2049776 1443
2047224 1873
2045404 2016
2044052 0
2041635 1608
2039911 1710
2039756 1600
2039616 1599
2039432 1377
2035803 1958
2035103 0
2034042 1687
2034005 0
2033994 1406
2033666 1788
2033106 1865
2032734 1432
2031555 1771
2031345 1454
2028512 1708
2026946 1923
2026600 1615
2026482 1046
2025701 1530
2025163 2084
2024695 1990
2024242 1830
2024065 1609
2022245 1414
2021523 1757
2020403 1591

Bill Gletsos
02-12-2004, 10:13 AM
But you can say the same about fish stocks, global warming, cognitive ability, etc, can't you?

Look, all he wants is trend analysis based on the available data. What's the big deal? Just do it already before youse two go on for pages about ratings again.

AR
I have better things to do than waste my time AR.
After all remember I am a volunteer. ;)

Cat
02-12-2004, 11:48 PM
There are too many other influences that may affect any so called trend based on factors unrelated to actual chess ability.


No, surely not! What might these be, Bill?

Alan Shore
03-12-2004, 12:14 AM
I have better things to do than waste my time AR.
After all remember I am a volunteer. ;)

What would you do with more time Bill? Perhaps you secretly love arguing with Dave Richards and Matt Sweeney etc. ;)

Kevin Bonham
03-12-2004, 01:26 AM
Given that your hit rate is no better than 50.68% (37/73) I have no intention of wasting my time on this to determine the performance ratings of the players.

Indeed, and that was the only one over 50%, and that was just on direction alone. I'll have to test to see if his ten methods, taken together, have done significantly worse than random again.

Matt - my advice is to give up any hope of trending ever working as a general approach. Focus on seeing if you can identify ways to use magnitude of trend to isolate those individuals for whom trending is a better predictor, if any and then you might get somewhere. I'll believe it when I see it though, and withheld judgement until then.

PHAT
03-12-2004, 09:41 AM
Indeed, and that was the only one over 50%, and that was just on direction alone. I'll have to test to see if his ten methods, taken together, have done significantly worse than random again.

You had better hope that it isn't significantly worse than random, because that would mean that a simple sign reversal (+ to -) would make my predictions significantly better than random. :owned:


Matt - my advice is to give up any hope of trending ever working as a general approach.
No, not yet, if ever.


Focus on seeing if you can identify ways to use magnitude of trend to isolate those individuals for whom trending is a better predictor, if any and then you might get somewhere.

I specifically removed exactly those individuals from the sample! I wanted to look at apparently non trending players. It appears that I have identified them very well! (Recall that I removed players from the list who had any rating change >150.)

PHAT
03-12-2004, 09:56 AM
Open Request to the ACF Ratings Officer.


I understand your reservations concerning the folly of my approach to ratings calculations. However, this debate cannot be resolved one way or another without me being able to experiment on real data to which only you have access. While the published Glicko generated ratings have much to recommend them, it is not pure data, ie. it is transformed data. Therefore results that I might generate are at best, an approximation of an approximation.

I am making a request for the Performance Ratings and the number of games for rated players over, say, 8 rating periods. Only you have this data.

Thankyou in advance.

PHAT
03-12-2004, 10:46 AM
Of these 73 players 34 had their ratings go in the completely opposite direction... As such these 34 can be seen as total failures ...


You have stated on numerous occations, that there is no way of predicting if a player will go [i[up or down[/i] in the next list. As such, the up or the down must be random.

Well, it appears that the cohort that I selected (non-trending) acted in just such a manner 39/73 = 53% approx. chance.

Now I will turn my atention to the cohort the at I expect to show a non-random distribution of up/down change.


Of the remaining 39 players, your estimate of their rating only went in the right direction as follows for each of the 10 estimates.


Estimate 1: 31
Estimate 2: 32
Estimate 3: 32
Estimate 4: 31
Estimate 5: 35
Estimate 6: 35
Estimate 7: 37
Estimate 8: 36
Estimate 9: 25
Estimate 10: 28

Given that your hit rate is no better than 50.68% (37/73) I have no intention of wasting my time on this to determine the performance ratings of the players.

Unfortunately, for your statement, as I said in a previous post to Ian Rout, I only expected 2 or three of these sets of estimates be "good" because the others were experiments to look for hints for doing fine tuning of algorithms. Therefore, your critisism using hit/miss stats for most of the estimates is not valid.

Bill Gletsos
03-12-2004, 10:54 AM
Open Request to the ACF Ratings Officer.


I understand your reservations concerning the folly of my approach to ratings calculations. However, this debate cannot be resolved one way or another without me being able to experiment on real data to which only you have access. While the published Glicko generated ratings have much to recommend them, it is not pure data, ie. it is transformed data. Therefore results that I might generate are at best, an approximation of an approximation.

I am making a request for the Performance Ratings and the number of games for rated players over, say, 8 rating periods. Only you have this data.

Thankyou in advance.
Of course it can be resolved without the real data.
Your attempt gets a signifcant number of the directional movements wrong.

You dont need the performance data to get that major problem sorted out.

After all if a players published rating increased from one period to the next you can be certain their performance rating was higher than their rating, likewise if their published rating decreased from one period to the next you can be certain their performance rating was lower than their rating.

In fact any competent researcher should be able to generate test data that adequately tests their hypothesis. You would appear to have made no such attempt to do so. If you are unable to do so, either due to lack of computer skills or other issues, that is not my concern.

On top of all this there is no evidence world wide that trending of ratings has any actual use.

Therefore as I stated previously I have no intention of wasting my time on this.

Bill Gletsos
03-12-2004, 11:15 AM
You have stated on numerous occations, that there is no way of predicting if a player will go [i[up or down[/i] in the next list. As such, the up or the down must be random.
Just because you cannot predict it does not necessarily make it random.
It just means you do not have sufficient data to determine it.
e.g. a player whose rating has been fairly static decides to get some coaching. Because of this his rating improves in the next couple of periods.
Like wise a player although still playing decides to xdevote less time to chess due to other requirements (e.g. studies).
These are just two examples of many effects that you cannot measure.


Well, it appears that the cohort that I selected (non-trending) acted in just such a manner 39/73 = 53% approx. chance.
Firstly its not 39 at all. The best you got was with estimate 7 which was 37/73.

The really bad result for your trend theory appears to be that the direction of the predicted trend was correct only half the time. That obviously for you must be most disappointing to say the least. It is what one might expect of a random coin toss. Your trending theory needs more than minor first aid to recover from this near cardiac arrest.



Now I will turn my atention to the cohort the at I expect to show a non-random distribution of up/down change.

Unfortunately, for your statement, as I said in a previous post to Ian Rout, I only expected 2 or three of these sets of estimates be "good" because the others were experiments to look for hints for doing fine tuning of algorithms. Therefore, your critisism using hit/miss stats for most of the estimates is not valid.
Of course my criticism is valid.
None of your sets exceed 50.68% (37/73).

PHAT
03-12-2004, 11:31 AM
In fact any competent researcher should be able to generate test data that adequately tests their hypothesis. You would appear to have made no such attempt to do so. If you are unable to do so, either due to lack of computer skills or other issues, that is not my concern.



It is IMPOSSIBLE to generate real world data.

The data that you have already COLLECTED and collated is all that I am asking for. This task has already been completed by you. It is common practise in research, that when another researcher asks for the raw data on which conclutions were drawn, that that data is given freely.



Therefore as I stated previously I have no intention of wasting my time on this.

Over the years, you have spent a conciderable amount of real time and real effort debating this topic. It appears that you are electing to spend much more by refusing to cooperate on resolving this disagreement.

Perhaps, as the ACF Ratings Officer, you could think about your responsiblitity to the Australian chess commmunity to seek the very best of rating systems. This necessarily requires your cooperation with people who hold ideas that diverge from the orthadox.

Bill Gletsos
03-12-2004, 11:50 AM
It is IMPOSSIBLE to generate real world data.
You can generate valid test data to cover various scenarios and types of rating changes for players. This should be a first step to demonstrate that trending has any merit whatsoever.


The data that you have already COLLECTED and collated is all that I am asking for. This task has already been completed by you. It is common practise in research, that when another researcher asks for the raw data on which conclutions were drawn, that that data is given freely.
The only raw data I have is the SP files.



Over the years, you have spent a conciderable amount of real time and real effort debating this topic. It appears that you are electing to spend much more by refusing to cooperate on resolving this disagreement.
You are the one with an unproven theory.
The onus is on you to prove it has theoretical merit by generating test data to prove it.


Perhaps, as the ACF Ratings Officer, you could think about your responsiblitity to the Australian chess commmunity to seek the very best of rating systems. This necessarily requires your cooperation with people who hold ideas that diverge from the orthadox.
As I said no one involved in rating theory have shown any belief that trending has any merit.
Deciding not to waste my time on some unorthodox theory with no demonstrably theorectical benefit let alone actual is my perogative.
As a volunteer I also have the right to determine how I allocate my time to that role of ACF Ratings Officer.

PHAT
03-12-2004, 11:55 AM
Just because you cannot predict it does not necessarily make it random.
It just means you do not have sufficient data to determine it.


OR, that you don't know how to use the data that you have. Concider that for a micro second.


The really bad result for your trend theory appears to be that the direction of the predicted trend was correct only half the time. That obviously for you must be most disappointing to say the least.

Ah Grasshopper, you move to fast to truely see.

Some sets were ~50% which is what I expected would occur. Some were extraordinarity bad - or so YOU have said. eg 28/73.

But, if you turn the leaf over slowly[i] in your hand, Grasshopper, you can see the insects that you seek before they fly away.

25/73 is equivalent to 45/73 = 66%. Perhaps I have a sign change error in my algorithms that reversed the results.

I will take a look. You say, "Your trending theory needs more than minor first aid to recover from this near cardiac arrest." However, a quack can confuse indigestion with angina.

Bill Gletsos
03-12-2004, 12:04 PM
OR, that you don't know how to use the data that you have. Concider that for a micro second.
The only data you have are game results.


[Ah Grasshopper, you move to fast to truely see.

Some sets were ~50% which is what I expected would occur. Some were extraordinarity bad - or so YOU have said. eg 28/73.

But, if you turn the leaf over slowly[i] in your hand, Grasshopper, you can see the insects that you seek before they fly away.

25/73 is equivalent to 45/73 = 66%. Perhaps I have a sign change error in my algorithms that reversed the results.
Perhaps you do. However who would believe you.


[I will take a look. You say, "Your trending theory needs more than minor first aid to recover from this near cardiac arrest." However, a quack can confuse indigestion with angina.
If that is the case then the quack is you.
Or should that be goose instead of duck.

For you see the words I used:

The really bad result for your trend theory appears to be that the direction of the predicted trend was correct only half the time. That obviously for you must be most disappointing to say the least. It is what one might expect of a random coin toss. Your trending theory needs more than minor first aid to recover from this near cardiac arrest.

were but a very slight modification of the words you used to describe your trending experiment failure for the December 2003 period.
Back then you said:

The "bad" result for trending appears to be that the direction of the predicted trend was correct only half the time. That is most disapointing. It is what we might expect of a random coin toss. Trending needs more than first aid to recover from this.

PHAT
03-12-2004, 12:13 PM
You can generate valid test data to cover various scenarios and types of rating changes for players. This should be a first step to demonstrate that trending has any merit whatsoever.

Utter rubbish.



The only raw data I have is the SP files.

And you also have collated them and then run them through the Glicko algorithms. There is nothing stopping you from lifting the performance ratings out of the spread sheets and posting them to me. Nothing, stops you at all except bloody mindedness.




You are the one with an unproven theory.
The onus is on you to prove it has theoretical merit by generating test data to prove it.
I will continue to try and prove the theory when you give me the data that you have at your disposal.



As I said no one involved in rating theory have shown any belief that trending has any merit.

Tell us who, in ratings theory, has actually said that trending has no merrit.



Deciding not to waste my time on some unorthodox theory with no demonstrably theorectical benefit let alone actual is my perogative.
As a volunteer I also have the right to determine how I allocate my time to that role of ACF Ratings Officer.

http://www.umass.edu/aesop/dog/index.html

http://www.winternet.com/~mikelr/flame63.html

Rincewind
03-12-2004, 12:27 PM
I will continue to try and prove the theory when you give me the data that you have at your disposal.

I think the best scientific response would be to try to DISPROVE the theory. If you can't do that despite best endeavours, then you might be on to something. ;)

Bill Gletsos
03-12-2004, 12:29 PM
Utter rubbish.
The only rubbish being sprouted is by you.
Your obvious inability to generate meaningful test data is your problem not mine.



And you also have collated them and then run them through the Glicko algorithms. There is nothing stopping you from lifting the performance ratings out of the spread sheets and posting them to me. Nothing, stops you at all except bloody mindedness.
Performance ratings are based on players results and the ratings of their opponents. e.g. a player is rated 1500. Then his performance rating in 4 successive periods is 1600, 1620, 1650 and 1660. However if trending was in effect in some manner then this would have effected his opponents ratings after the period in which he was rated 1500. In this case his performance rating in the successive period might be 1650 or it could be 1550.

The only way to do a true comparative test if one system has superior predictive accuracy over another is to process the raw data and see which is best at prediciting the players future scores irrespective of whether that data be actual data or generated test data.


I will continue to try and prove the theory when you give me the data that you have at your disposal.
You are in for a long wait.



Tell us who, in ratings theory, has actually said that trending has no merrit.
I didnt say that. I said no one has suggested trending has any merit.
If it had any merit rating theory experts would have said so by now.

PHAT
03-12-2004, 01:08 PM
I think the best scientific response would be to try to DISPROVE the theory. If you can't do that despite best endeavours, then you might be on to something. ;)

Touche!

I have thought of that. For example:

1. What is the proportion of positive to negative changes in a period.
2. What is the mean of the positive and negative changes.
Glicko assumes
1. = 50-50
2. = equil agrigate movement in both directions.

Trending would be shown to be possible phenominum if either of these assumptions are incorrect while the other remained true. Now, since there is no inflation in the system (BG says), I could contend that #1 and #2 are both untrue assumptions but that they work in opposite directions, thus masking the actual flux (trending) of individuals in the pool. However, that is chicken and egg to say which is effecting which.

One piece of data I generated, and intially "put aside", was that in the cohort I published the mean drop in ratings points was ~7 in every period (8 periods). I thought this odd at the time since there is supposed to be no inflation/deflation. However, Ian Rout's observation that the cohort was substantially ACT players over 40 years of age, indicates that that group is experiancing a highly significant (statistically) decline in rating. The question arises: Is it due to a slow decline in ability in this group, or the dreaded "junior effect," or a bit of both?

Rincewind
03-12-2004, 03:22 PM
1. What is the proportion of positive to negative changes in a period.
2. What is the mean of the positive and negative changes.
Glicko assumes
1. = 50-50
2. = equil agrigate movement in both directions.

Glicko doesn't force 1.
Neither is Glicko zero sum (ie 2 is not forced either).

Elo doesn't even force 1 but generally would be zero sum depending if tiered k-factors were implemented.

Bill Gletsos
03-12-2004, 04:51 PM
Glicko doesn't force 1.
Neither is Glicko zero sum (ie 2 is not forced either).
The above has been pointed out numerous times on the BB.
And Matt wonders why I say he has no clue about ratings.

PHAT
03-12-2004, 10:41 PM
Glicko doesn't force 1.
Neither is Glicko zero sum (ie 2 is not forced either).

Elo doesn't even force 1 but generally would be zero sum depending if tiered k-factors were implemented.

No no. I did not say it forces. I said it assumes (that in the real world). ie Glicko assumes that the pool is in dynamic equilibrium. Further, Glicko assumes (that in the real world) there will be a zero sum gain.

Glicko2 affects a zero sum gain as evidenced by the lack of inflation. It would be extrordinary for there not to be some change in the mean chess playing ability in the population over time. That there isn't any change in the mean says something about the nature of Glicko.

For example: if the Cheap Viagra vendors in Australia started selling GetSmart pills that worked, we might all play better chess, but none of our ratings would be effected. Furthermore, the ACF ratings would lose equivalence with FIDE ratings.

PHAT
03-12-2004, 10:43 PM
The above has been pointed out numerous times on the BB.
And Matt wonders why I say he has no clue about ratings.

This is simple bullying. Please stop it.

Bill Gletsos
03-12-2004, 11:03 PM
No no. I did not say it forces. I said it assumes. ie Glicko assumes that the pool is in dynamic equilibrium. Further, Glicko assumes that their will be a zero sum gain.
Neither Glicko nor Glicko2 make any such assumption.

This is easily evidenced using Glicko in the following example:

You have a pool of 4 players A,B,C,D rated 1500, 1500, 1600 and 1600 respectively. The zaerage rating of the pool
Player A is 1500 with RD of 60. Player B is 1500 with RD of 120.
Player C is 1600 with RD of 60. Player B is 1600 with RD of 200.
A beats B and C beats D.
The ratings are now: A is 1509, B is 1463, C is 1608 and D is 1514.
The pool average is now 1523.5.
It is clear that for the games between the players it is not zero sum and this is a fundamental part of Glicko and Glicko2.
Also its is clear the average rating of the pool is not maintained. This is also a fundamental part of Glicko & Glicko2.
The drop in average rating does not represent deflation.


Glicko2 affects a zero sum gain as evidenced by the lack of inflation. It would be extrordinary for there not to be some change in the mean chess playing ability in the population over time. That there isn't any change in the mean says something about the nature of Glicko.
Again you demonstrate your complete lack of understanding.
Who said the mean does not change.
Also lack of inflation/deflation does not mean that the mean doesnt change.

Bill Gletsos
03-12-2004, 11:06 PM
This is simple bullying. Please stop it.
Its not bullying.
You keep repeating things that have previously been shown to have no basis in fact.
That is either plain stupidity or a lack of understanding on your part.
I chose the later.
Feel free to choose the former.

Rincewind
04-12-2004, 07:40 AM
No no. I did not say it forces. I said it assumes (that in the real world). ie Glicko assumes that the pool is in dynamic equilibrium. Further, Glicko assumes (that in the real world) there will be a zero sum gain.

I don't know that this is even assumed. I think the most you can say is Glicko assumes opponents ratings are accurate measures of chess play strength to within the accuracy of the RD. As players are constantly entering and leaving the system it does not make or require overall assumptions of the pool. Some pool behaviours might be deducible from this, however don't put the cart before the horse.


Glicko2 affects a zero sum gain as evidenced by the lack of inflation. It would be extrordinary for there not to be some change in the mean chess playing ability in the population over time. That there isn't any change in the mean says something about the nature of Glicko.

I think lack of inflations is not about zero sum gain, it is more about rating stability of players of constant strength. Thie rating average of the pool might shift but provided you have consistent ratings of stable players then it is reasonable to say the system is non-inflationary. This might be difficult to measure but I would say you would want to benchmark with active players who have been in the system with a reasonably static rating for a some period.

PHAT
04-12-2004, 10:00 AM
Neither Glicko nor Glicko2 make any such assumption.

This is easily evidenced using Glicko in the following example:

You have a pool of 4 players A,B,C,D rated 1500, 1500, 1600 and 1600 respectively. The zaerage rating of the pool
Player A is 1500 with RD of 60. Player B is 1500 with RD of 120.
Player C is 1600 with RD of 60. Player B is 1600 with RD of 200.
A beats B and C beats D.
The ratings are now: A is 1509, B is 1463, C is 1608 and D is 1514.
The pool average is now 1523.5.
It is clear that for the games between the players it is not zero sum and this is a fundamental part of Glicko and Glicko2.

I know this and accept it. However, it is a single example. Equily, the example could be rewritten were the average was 1576.5. The point I am making is that when all the single examples are tallied, there would appear that Glick is affecting a zero-sum gain. [perhaps ZSG is not exactly the correct term here]

Also its is clear the average rating of the pool is not maintained. This is also a fundamental part of Glicko & Glicko2.

Sorry, but that is not clear to me. The stats you have given previously given say that the average rating is neither going up or dow.



The drop in average rating does not represent deflation.

I know that. :rolleyes:




Also lack of inflation/deflation does not mean that the mean doesnt change.

I know that too. :rolleyes:


LOOK!
There are only three ways that the average rating can remain constant
1. inflation + net ability drop = 0
2. deflation + net ability rise = 0
3. no 'flation and ability constant.


#1 and #2 cannot be true, as you have said so often and loudly that there is no 'flation in the system.

#3 can only be true if, along with no 'flation, the net ability of Australian players is constant.

It would be a brave man who would assert that the net ability of Australian players is "constant," given the current large influx of improving juniors into the active pool.

In conclussion, I assert that Glicko imposes, by it's design, a zero sum gain for the population as a whole.

PHAT
04-12-2004, 10:25 AM
I think the most you can say is Glicko assumes opponents ratings are accurate measures of chess play strength to within the accuracy of the RD.

Correct. But Glicko is wrong to do so. The RD functions bi-directionally. It thus assumes that the rise or a fall in performance is equily likely for all players at all times. Therefore, for improving players (~10%) and falling players (~10%) the RD is not high enough to allow them to rise/fall at a fast enough rate. For these players, Glicko is always playing catch-up - thus (~20%) of players are chronically under/over rated. [Note the correct use of "chronic"]



I think lack of inflations is not about zero sum gain, it is more about rating stability of players of constant strength. The rating average of the pool might shift but provided you have consistent ratings of stable players then it is reasonable to say the system is non-inflationary. This might be difficult to measure but I would say you would want to benchmark with active players who have been in the system with a reasonably static rating for a some period.

Yes, quite true. However, as you say, there is a difficulty in identifying benchmark players. I would say, given the nature of natural systems, like the human brain and its environment, that flesh and blood should not be used. Perhaps, once every few years having a large number of rated players play against a standard chess program set at say 1500 would provide stable benchmark.

PHAT
04-12-2004, 10:29 AM
Its not bullying.
You keep repeating things that have previously been shown to have no basis in fact.
That is either plain stupidity or a lack of understanding on your part.
I chose the later.
Feel free to choose the former.

I am being civil with you, please return my civility.

Rincewind
04-12-2004, 11:21 AM
Correct. But Glicko is wrong to do so. The RD functions bi-directionally. It thus assumes that the rise or a fall in performance is equily likely for all players at all times. Therefore, for improving players (~10%) and falling players (~10%) the RD is not high enough to allow them to rise/fall at a fast enough rate. For these players, Glicko is always playing catch-up - thus (~20%) of players are chronically under/over rated. [Note the correct use of "chronic"]

The RD is not a measure of the likelyhood of a rating movement, but rather a region within which the true rating is believed to be contained. It does, I think, assume a symmetric distribution in that region and if by trending you mean to introduce a measure of skewness to this region then maybe that is worth looking at how this affects Glickman's formulae.


Yes, quite true. However, as you say, there is a difficulty in identifying benchmark players. I would say, given the nature of natural systems, like the human brain and its environment, that flesh and blood should not be used. Perhaps, once every few years having a large number of rated players play against a standard chess program set at say 1500 would provide stable benchmark.

That is an interesting idea however logistically it might prove expensive to implement. You also introduce noise nito the system as some people might perform better or worse against silicon-based (than they do against the regular carbon-based) opposition.

Bill Gletsos
04-12-2004, 01:35 PM
I know this and accept it. However, it is a single example. Equily, the example could be rewritten were the average was 1576.5.
So what.
The fundamental part of Glkicko is that an individual game can result in the gain by one player not equaling the loss by the other.


The point I am making is that when all the single examples are tallied, there would appear that Glick is affecting a zero-sum gain. [perhaps ZSG is not exactly the correct term here]
This is rubbish.
There is absolutely no reason for you to be assuming the the summing of the single examples results in a zero sum.


Sorry, but that is not clear to me.
Your lack of understanding is not my problem.


The stats you have given previously given say that the average rating is neither going up or dow.
The stats I have previously given have never stated that.
I have spoken about inflation/deflation of the pool not the average rating of the pool.



I know that. :rolleyes:

I know that too. :rolleyes:
Your comments would seem to indicate you do not.



LOOK!
There are only three ways that the average rating can remain constant
1. inflation + net ability drop = 0
2. deflation + net ability rise = 0
3. no 'flation and ability constant.
No one has stated the average rating of the pool remains constant.




#1 and #2 cannot be true, as you have said so often and loudly that there is no 'flation in the system.

#3 can only be true if, along with no 'flation, the net ability of Australian players is constant.
You are assuming the average rating of the pool remains constant.
I havent said this.


It would be a brave man who would assert that the net ability of Australian players is "constant," given the current large influx of improving juniors into the active pool.
I havent said this.


In conclussion, I assert that Glicko imposes, by it's design, a zero sum gain for the population as a whole.
You have no clue as usal what you are talking about.

Bill Gletsos
04-12-2004, 01:38 PM
I am being civil with you, please return my civility.
Considering you are the most uncivil individual on this BB this is a bit rich.
Of course being polite at this time is simply because it suits your purpose.

However civility has nothing to do with it.
You keep making unfounded and unjustified statements.
That gives rise to the two possabilities that I previously mentioned of why that is the case.

PHAT
04-12-2004, 02:43 PM
The RD is not a measure of the likelyhood of a rating movement, but rather a region within which the true rating is believed to be contained.

I fail to see the practical difference.

If the "region within which the true rating is believed to be contained," gets bigger, then the "likelyhood of a rating movement," also increases. Furthermore I suggest the the relationship is linear :cool:

Rincewind
04-12-2004, 02:51 PM
I fail to see the practical difference.

If the "region within which the true rating is believed to be contained," gets bigger, then the "likelyhood of a rating movement," also increases. Furthermore I suggest the the relationship is linear :cool:

I think they are completely different things. Especially when you are looking at trending.

One sort of rating "movement" is the strength of a player changing. This is the sort of thing I believe you are trying to capture in your trending model.

The other is not the strength of the player changing, just the system determining more accurately what the actual rating should be. I would expect these sort of movements to be practially random and therefore untrendable.

Bill Gletsos
04-12-2004, 03:24 PM
I fail to see the practical difference.
Why doesnt that surprise me.


If the "region within which the true rating is believed to be contained," gets bigger, then the "likelyhood of a rating movement," also increases. Furthermore I suggest the the relationship is linear :cool:
Your continual ability to pull theories out of thin air with no basis in fact is not surprising.

ursogr8
04-12-2004, 08:06 PM
Your continual ability to pull theories out of thin air with no basis in fact is not surprising.

I am glad this disfunction is not contagious; otherwise the theories I derive from metrics would be at risk. :uhoh:

starter

PHAT
04-12-2004, 10:19 PM
Your continual ability to pull theories out of thin air with no basis in fact is not surprising.

It is called imagination, Bill. So please, refrain from denigrating me, I don't think it is all that necessary.

Bill Gletsos
04-12-2004, 10:22 PM
It is called imagination, Bill. So please, refrain from denigrating me, I don't think it is all that necessary.
Rubbish.
Its not called imagination.
Its called not having a clue.

ursogr8
04-12-2004, 10:32 PM
Rubbish.
Its not called imagination.
Its called not having a clue.


hey Bill

I am keeping some metrics on denigration/thread.
Should I count 'Its called not having a clue.' as a denigration, or are you just classing this as a descriptor?

starter

Bill Gletsos
04-12-2004, 10:35 PM
I am glad this disfunction is not contagious; otherwise the theories I derive from metrics would be at risk. :uhoh:

starter
I'm sure some people here would suggest they are.

Bill Gletsos
04-12-2004, 10:37 PM
hey Bill

I am keeping some metrics on denigration/thread.
Should I count 'Its called not having a clue.' as a denigration, or are you just classing this as a descriptor?

starter
I see it as a statement of fact.
What you classify it as is up to you.

ursogr8
05-12-2004, 08:34 AM
I see it as a statement of fact.
What you classify it as is up to you.

Bill
Thought you might have intended it in that way.
As a denigrator it would have been humourous.
starter

PHAT
05-12-2004, 10:30 PM
Bill
As a denigrator it would have been humourous.


Yes, a flame without flare illuminates naught.