Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="Conaill" data-source="post: 1725115" data-attributes="member: 1264"><Another post with nothing but voting systems wonkiness. Feel free to skip if this doesn't interest you...>1) Regarding IMDB's voting scheme. They do *not* currently reveal the exact formula they use to calculate the score. Here's some comments from <a href="http://www.imdb.com/help/show_leaf?votes" target="_blank">their page on this issue</a>:After some digging around, I found out that IMBD originally did mention they were using a Bayesian estimate. They seem to have removed the formula from most of their web pages since, but you can still find it at the bottom of their <a href="http://www.imdb.com/chart/animation" target="_blank">Top Rated "Animation" Titles</a> page:It's important to point out that a Bayesian estimate doesn't help you get a more "fair" result in any way, it just helps you deal with the uncertainty caused by having a small number of votes (essentially, it adds a number of votes with average scores to get a more robust estimate of the population average). The Bayesian estimation is a pretty simple trick which should probably be used anyway, regardless of what other manipulations are done to the score (assuming, in our case, that we do have some choices with a small number of votes, say less than 100). I assume imdb has since built in a lot more additional tricks to avoid vote stuffing. From their explanation, they probably throw away a few of the top and bottom % votes, plus it sounds like they may have a sophisticated way to estimate the likelihood that someone is manipulating the vote, downweighting those votes accordingly.Do keep in mind that IMDB's goal with all of this is to get a score *number* for each "candidate" (i.e., movie), whereas the Ennies' goal is to pick a *winner*. Getting an actual score is a much harder issue, and we should be able to avoid an awful lot of the problems IMDB is forced to deal with. In particular, I strongly believe we should NOT let the voters assign a score, because that is where most of the abusability comes into play. Only allowing the voters to check of one (as in One-person-one-vote) or several choices (as in approval voting), or ranking the choices (as in instant runoff or other systems) gets rid of most of the power malicious voters have to manipulate the system.2) <a href="http://bcn.boulder.co.us/government/approvalvote/center.html" target="_blank">Approval voting</a>:Actually, people who vote for only one product have exactly the same amount of power as people who vote for all but one product, so "gaming the system" is not nearly as simple as casting more or less votes. In fact, the optimal "gamed" vote using approval voting relies on having an accurate estimate of the number of votes each candidate will get, and picking an optimal cutoff for the number of your favorites you vote for.3) Instant Runoff VotingThis method is called Instant Runoff Voting (or IRV), because it's essentially equivalent to a bunch of runoff elections, each time throwing out the candidate with the lowest number of votes until someone gets a majority. It's coincidentally also the method the US Green Party has been proposing for years for local and state elections, because it tends to avoid the "spoiler" scenario. (Noet however that unlike Approval Voting, IRV still has a problem with "spoilers" as soon as there are more than two viable candidate).You may have noticed I recommended ranking choices in my first post. IRV is one such system which uses ranked votes, but there are others, such as Aaron's suggestion of assigning point values to ranks (formally called "<a href="http://en.wikipedia.org/wiki/Borda_count" target="_blank">Borda Count</a>"). Personally, I greatly prefer IRV over any point-based system, but I was a little worried about scaring off people with its complexity <img src="https://cdn.jsdelivr.net/joypixels/assets/8.0/png/unicode/64/1f609.png" class="smilie smilie--emoji" loading="lazy" width="64" height="64" alt=";)" title="Wink    ;)"  data-smilie="2"data-shortname=";)" />. Note that Borda Count is still subject to strategic voting as well, although it's less of a problem than when assigning poiunts freely. IRV also deals with missing votes (i.e. unfamiliar products in the Ennies) in a more natural way than Borda Count. Another ranked method, which is supposed to be even more fair than both IRV and Borda is the <a href="http://electionmethods.org/Condorcet.htm" target="_blank">Condorcet method</a>. Unfortunately, it's also a little more complicated...In theory, how much a system can be exploited is directly related to how many different choices you leave to the voter. So in principle, a ranking scheme should be more "gameable" than one where you only pick candidates. For all of these reasons, and because of its simplicity, I generally prefer Approval Voting over IRV and other ranked methods. If it's good enough for the the Mathematical Association of America, and the American Statistical Association, it's good enough for me! <img src="https://cdn.jsdelivr.net/joypixels/assets/8.0/png/unicode/64/1f600.png" class="smilie smilie--emoji" loading="lazy" width="64" height="64" alt=":D" title="Big grin    :D"  data-smilie="8"data-shortname=":D" />For some more info on Approval Voting, you can check <a href="http://bcn.boulder.co.us/government/approvalvote/center.html" target="_blank">here</a>, <a href="http://www.kellogg.nwu.edu/faculty/weber/papers/approval.htm" target="_blank">here</a>, or <a href="http://electionmethods.org/approved.htm" target="_blank">here</a> These pages will also lead you to links to various other pages on voting systems...</blockquote>

Verification