I started this out over on Doc's forum and want some discussion or even ideas. I just had these realizations - basically that most paintball "testing results" are based upon general observations of differences, as opposed to statistically testing the data to find significant results.
Here was my first post with ideas:
I'm thinking of some interesting studies to do related to probability, prediction, association, and relationship.
No particular direction with this yet, not for a class or anything. I crunch statistics and do studies at a university as a big part of my day job and my mind just winds this way.
See if there is a relationship between the amount of paint shot by a team in a tournament game and winning or losing. This would be easy to do throughout an entire season through the NPPL or such by adding one column on the judges scoresheet for each game - where each team reports an approximate amount of paint they shot during the game. Could be an interesting industry statistic (especially if the results showed a direct positive relationship between the paint shot and scores). Would even be more interesting to teams if no relationship was shown - then they would have to figure out why teams win.
Could study based upon certain gun types, barrel types, air-systems, regulators, etc. Again, same type of study, to see if a relationship exists between using equipment type and scores in tournaments. If a relationship is found - wow manufacturers would surely jump upon that. But, the factor may just be belief or confidence in winning when using a certain type of equipment - rather than the actual benefit of using the said type of equipment.
Other studies might include:
Relationship between time when first competitor is eliminated and scores or winning.
Relationship between times of each elimination of a competitor/teammate and scores or winning.
ad. nauseum...
Anyway, what are some studies you would like to see conducted...
Thanks,
Nun
One guy asked me if I was an actuary or bookie...Here is my reply...
I just do stats and assessment as part of my job. With the right data, massaged just so and set up right from the beginning, I can do some pretty cool testing about probabilities and relationships.
It's just all statistical formulas - now actuaries - they can figure some pretty trippy stuff - my brand is a bit more mundane - Spearman rho correlations (determines relationship and the direction of the relationship), Pearson r correlations (relationships again), linear regression (variable prediction from another variable), multiple linear regression (prediction of a variable from multiple variables), factorial ANOVAs (comparing means of each of several components to determine if differences are significant or due to error), etc.
With some willing subjects - or willing tourney promoters - some pretty cool testing could be done. I am sure - the probability is in our favor - that there are some player controllable variables that lead to more consistent winning or losing - most easily tested in tournament type situations.
Give me enough good data (normally distributed with low standard deviations)about paintballs and barrels and I can develop some probabilities and relationships or any number of other statitistics.
One problem with past testing may have been in simple sampling error (or maybe not), but I bet some of the most venerated tests haven't been statistically tested for reliability. While the observed differences of this or that barrel test may have shown x or y result, I bet money that nobody did reliability testing on that data to see if that was actually true (within a certain margin of error) or if it was just due to the that particular sample and batch of balls. Running statistical tests can show, for example, if accuracy does indeed go up or down with length - all to within percentage of certainty and predictability (usually within 95-99%)- that will be predictable in most circumstances. There are so many, many other kinds of testing that can be done. However, most paintball testing I've seen just makes observations from a pile of raw data (or maybe some simple descriptive statics). The data has to be statistically tested before conclusiions can be drawn. Otherwise, error in interpretation is a likely result.
Anyway, I would be interested in exploring the possibilities, if people have data they want tested, etc.
Nun
A bit more...
One example in particular...Tom's data on barrel length, velocity, accuracy, etc. - what about statitistically testing that data? Were the results Statistically significant or just from observed differences? If the data was never statitistically tested - then erronious conclusions have probably been drawn. For example, data can be tested for difference, relationship, and predictability - all to within a certain probability - when probability get's to within 97 - 99% - then statisticaly significant results can be drawn. The deal is, without proper testing of the data itself, those kind of conclusions can't be made.
I've talked in circles and repeated myself enough.
Respond if you desire...
Nun
Here was my first post with ideas:
I'm thinking of some interesting studies to do related to probability, prediction, association, and relationship.
No particular direction with this yet, not for a class or anything. I crunch statistics and do studies at a university as a big part of my day job and my mind just winds this way.
See if there is a relationship between the amount of paint shot by a team in a tournament game and winning or losing. This would be easy to do throughout an entire season through the NPPL or such by adding one column on the judges scoresheet for each game - where each team reports an approximate amount of paint they shot during the game. Could be an interesting industry statistic (especially if the results showed a direct positive relationship between the paint shot and scores). Would even be more interesting to teams if no relationship was shown - then they would have to figure out why teams win.
Could study based upon certain gun types, barrel types, air-systems, regulators, etc. Again, same type of study, to see if a relationship exists between using equipment type and scores in tournaments. If a relationship is found - wow manufacturers would surely jump upon that. But, the factor may just be belief or confidence in winning when using a certain type of equipment - rather than the actual benefit of using the said type of equipment.
Other studies might include:
Relationship between time when first competitor is eliminated and scores or winning.
Relationship between times of each elimination of a competitor/teammate and scores or winning.
ad. nauseum...
Anyway, what are some studies you would like to see conducted...
Thanks,
Nun
One guy asked me if I was an actuary or bookie...Here is my reply...
I just do stats and assessment as part of my job. With the right data, massaged just so and set up right from the beginning, I can do some pretty cool testing about probabilities and relationships.
It's just all statistical formulas - now actuaries - they can figure some pretty trippy stuff - my brand is a bit more mundane - Spearman rho correlations (determines relationship and the direction of the relationship), Pearson r correlations (relationships again), linear regression (variable prediction from another variable), multiple linear regression (prediction of a variable from multiple variables), factorial ANOVAs (comparing means of each of several components to determine if differences are significant or due to error), etc.
With some willing subjects - or willing tourney promoters - some pretty cool testing could be done. I am sure - the probability is in our favor - that there are some player controllable variables that lead to more consistent winning or losing - most easily tested in tournament type situations.
Give me enough good data (normally distributed with low standard deviations)about paintballs and barrels and I can develop some probabilities and relationships or any number of other statitistics.
One problem with past testing may have been in simple sampling error (or maybe not), but I bet some of the most venerated tests haven't been statistically tested for reliability. While the observed differences of this or that barrel test may have shown x or y result, I bet money that nobody did reliability testing on that data to see if that was actually true (within a certain margin of error) or if it was just due to the that particular sample and batch of balls. Running statistical tests can show, for example, if accuracy does indeed go up or down with length - all to within percentage of certainty and predictability (usually within 95-99%)- that will be predictable in most circumstances. There are so many, many other kinds of testing that can be done. However, most paintball testing I've seen just makes observations from a pile of raw data (or maybe some simple descriptive statics). The data has to be statistically tested before conclusiions can be drawn. Otherwise, error in interpretation is a likely result.
Anyway, I would be interested in exploring the possibilities, if people have data they want tested, etc.
Nun
A bit more...
One example in particular...Tom's data on barrel length, velocity, accuracy, etc. - what about statitistically testing that data? Were the results Statistically significant or just from observed differences? If the data was never statitistically tested - then erronious conclusions have probably been drawn. For example, data can be tested for difference, relationship, and predictability - all to within a certain probability - when probability get's to within 97 - 99% - then statisticaly significant results can be drawn. The deal is, without proper testing of the data itself, those kind of conclusions can't be made.
I've talked in circles and repeated myself enough.
Respond if you desire...
Nun