Upgrade your account to a Community Supporter account and remove most of the site ads.
Rocket your D&D 5E and Level Up: Advanced 5E games into space! Alpha Star Magazine Is Launching... Right Now!

Reply to thread

Message: <blockquote data-quote="Olgar Shiverstone" data-source="post: 994266" data-attributes="member: 5868">OK, given that you're already dropping outliers and using a minimum cutoff score, at a minimum what I'd recommend you do next is normalize all the scores by individual voter.  Here's how:1.  Take the average of all of an individual voters scores (Sum of scores/# of scores -- ignore "not familiars").  2.  Calculate the sample standard deviation for that individual's scores (easiest way to do this is with a function like STDEV in Excel, or I'm sure there's a similar capability in whatever tools Blacksway is programming in.  There is a brute force method I can provide if need be).3.  Correct each individual score, using the formula:(X - Mean)/STDEV, whereX- scoreMean -- average of that individual's scoresSTDEV -- the standard deviation calculated above.At this point you'll have converted all the scores into a range that will generally fall between -3 and -3, with the exact mean at zero.  The number represents the number of standard deviations away from average the product falls on a standard normal (bell) curve.  Example (hopefully the formatting comes through)Score       Norm        Percentile4    -0.2397    41%3    -0.6391    26%9    1.7577    96%2    -1.0386    15%3    -0.6391    26%4    -0.2397    41%4    -0.2397    41%8    1.3582    91%7    0.9587    83%2    -1.0386    15%Mean: 4.6        STD DEV: 2.503331114        There's a choice here -- you can work directly with these scores, or convert them into percentile ranks (using a Z-table function, like NORMDIST in Excel, which I've used above).  What you do depends upon how you want to make the final decision:For example, the winner could be the one with the highest average score.  In that case, sum all the normed scores for that particular product, and divide by the number of votes the product received (you could convert the to percentages before summing or after dividing, if you want a score between 0 and 100).  This gives the win to the highest "quality" grade of the product, but the actual number of votes doesn't matter, so long as there's enough to meet the minimum requirement (more votes will mean a more accurate assessment of product quality, though).Or, you could use the highest total score.  This rewards a large number of votes, but takes into account the quality of a product In this case, sum the normed scores -- this is important, because the negative scores actually reduce the total, so it isn't possible for a product with all below-average scores to get a positive score, regardless of the number of votes it receives.  It guarantees that the highest positive score had a majority of votes with at least "above average" quality.  It won't actually select the "best quality" product based on ratings -- a product with a lot of "8's", for example, can beat a product with a smaller number of "10" votes.  It does provide a balance of quality and quantity, though, and the normed scores won't reward below-average product ratings.  OF course, the final number won't mean much to most people.A third option would be to provide use a weighted decision criterion between the first method I mentioned and the second.  Not knowing exactly how you want the awards assigned, you'll have to consider the options.  Overall I think the second method provides a reasonable balance that rewards both the popularity of a product (number of votes received) and the quality of product (level of scores).As to running an ANOVA at the end, you really can get away without doing this just by looking at the scores and applying a little common sense.  I suspect in the end that there will be fairly clear winners in each catagory; the ANOVA comes into play where one product wins out numerically by a second or third is very close behind.  In any case, the ANOVA would require a dedicated stats package to run -- the full procedure is too long to detail here.Food for thought.</blockquote>

Verification