04-07-2008, 03:30 PM
Imagine an amazing intro...
I should been studying for another things but i like to play with spss. That said. I will upload images after my school the load has lightened. I have drafts of the result images that can be seen at google docs; just pm me or state that you want to see them (there's no personal info).
I would love to see challenged my method, this way i would know if i did it right. I made a proxy boost to deal with the problem of skipping values (the chart is available). Also i just took the score and the number of winnings against Riilan and Grundobot because those victories are hard to achieve with the overpowered sparkshooter, in an attempt to minimize the weapon effect. Now the findings.
There's a strong correlation between HP with Scores and Victories (~0.81) and STR and DEF <proxy> has an almost strong association with the others ; from stronger to weaker: Riilan (~0.71), Score (~0.7) and Grundobot (~0.59), being ST the stronger.
When the data was graphed. One can believe that the proxies has a linear relationship but the HP can bee seen differently. Aside the atypical/extreme data. It follows a cubic (is the correct word?), X^3 , pattern; at the begging rising as you stats grows then a plateau and ending with the rising pattern.
Now the linear regression says that you can predict ~71% of the victories/score with HP+proxy DEF+proxy STR. The main predictor is HP and then proxy DEF both with good levels of significance; that means that they work and the probability to be wrong is almost zero. But STR (proxy, boost and "number") is not significant and if it were it would say that it works against (more strength means less score & victories).
I also made a table with the Guilds rankings. If we take the people with no guild label they end 9th and if we put together the guilds with only one o two players they get the 6th position.
I hope this semester regression won't be a short class. After it i perhaps will make a new set of test.
Have fun, lots of fun
EDIT: my R key is not working properly so i may loose some "r"
I should been studying for another things but i like to play with spss. That said. I will upload images after my school the load has lightened. I have drafts of the result images that can be seen at google docs; just pm me or state that you want to see them (there's no personal info).
I would love to see challenged my method, this way i would know if i did it right. I made a proxy boost to deal with the problem of skipping values (the chart is available). Also i just took the score and the number of winnings against Riilan and Grundobot because those victories are hard to achieve with the overpowered sparkshooter, in an attempt to minimize the weapon effect. Now the findings.
There's a strong correlation between HP with Scores and Victories (~0.81) and STR and DEF <proxy> has an almost strong association with the others ; from stronger to weaker: Riilan (~0.71), Score (~0.7) and Grundobot (~0.59), being ST the stronger.
When the data was graphed. One can believe that the proxies has a linear relationship but the HP can bee seen differently. Aside the atypical/extreme data. It follows a cubic (is the correct word?), X^3 , pattern; at the begging rising as you stats grows then a plateau and ending with the rising pattern.
Now the linear regression says that you can predict ~71% of the victories/score with HP+proxy DEF+proxy STR. The main predictor is HP and then proxy DEF both with good levels of significance; that means that they work and the probability to be wrong is almost zero. But STR (proxy, boost and "number") is not significant and if it were it would say that it works against (more strength means less score & victories).
I also made a table with the Guilds rankings. If we take the people with no guild label they end 9th and if we put together the guilds with only one o two players they get the 6th position.
I hope this semester regression won't be a short class. After it i perhaps will make a new set of test.
Have fun, lots of fun
EDIT: my R key is not working properly so i may loose some "r"