SK Race Choice

[bcbrown]

Originally Posted by Snaggles [You must be logged in to view images. Log in or Register.]

My issue is when people hold their anecdotal “evidence” too proudly. In all my EQ years I’ve only seen a few examples of what appears a professional statistician fixating on a video game and doing research. My own testing is useful (for me) but less reliable, I NEVER indicate it’s the law due to short sampling and less the perfect testing. I probably have 500+ Vindi parses currently in my logs. I can draw some conclusions but I haven’t confined variables and ran a parse on him for like 1,000 hours. My data curve is still wobbly.

By no means am I a professional statistician, but I have worked with plenty. High quality statistical testing requires, at a minimum:

An explicit mental model of the world, with all assumptions stated clearly. You need to be able to articulate your understanding of the world before you can know whether your results indicate an improvement of that understanding.

An explicit hypothesis. You need to know what you are looking for in your experiment. Ad-hoc exploratory data gathering can be useful in trying to formulate a hypthothesis, but that exploratory data will not be useful in determining whether a hypothesis is confirmed or rejected.

An experiment design. You need to know ahead of time what data you wish to gather, how to gather it, and when to stop the experiment. An example of a flawed experiment would be trying to show that a certain gear combo causes a certain DPS increase, and then stopping your parse as soon as you show that DPS increase.

Sanity-checking the resulting data to confirm your assumptions have been met. If not, then your understanding of the world is flawed and your data unusable. You need to first run a different experiment to find and fix the flaws in your assumptions.

Run a well-defined, repeatable analysis. You need to know ahead of time what metrics you wish to calculate. You should also do some sort of calculation of statistical confidence, whether frequentist or Bayesian.

Scientific integrity. You need to publish your results whether or not they support your hypothesis. If the results violate some of your assumptions, you cannot rely on the results of any data analysis.

Originally Posted by Troxx [You must be logged in to view images. Log in or Register.]

I once completed a successful high end group without my pants equipped on my warrior. This is a game and games are based on rules and math. My pants were not needed to successfully win these encounters.

I therefore assert that pants are not needed and thus do not need to be worn.

Time for us all to play commando guys!

This is a lovely comment when combined with the image in your signature. Horseriding while commando, though, whew. I pity the saddle.