I notice there are several explicit approval cutoff methods being discussed on the list. Kevin Venzke has suggested a number of rank-flattening proposals, with Chris Benham proposing a simplified version of one of them: Condorcet-using-full-preferences // Condorcet-using-approved-preferences // Approval. The latter can be used to handle Forest Simmon's 3 approval cutoff examples. I like this method due to its relative simplicity and self consistency. I propose calling this it Instant Round Robin Rank Flattened Fallback, or IR3F2. Whichever approval cutoff method one uses, there is some latitude in how one could implement an explicit approval cutoff. I've been thinking about 6, 7 or 10 slot methods, with an explicit Disapproval Level added as a candidate (any candidate rated at DL or below is disapproved). If the default rate is lowest rate, then it makes sense for DL to be an extra "candidate", with the same default. Then approval is any rate above that given to DL. Using this form, a voter who doesn't exercise their option to reset the Disapproval Level is basically using implicit approval cutoff at rate 0. Having an extra DL candidate also enables the opportunity for a voter to disapprove every candidate by voting DL at top rate, which I could see people doing as some form of None-of-the-above protest. I dimly recall something of this sort being discussed a few years back. When I am able to chat about this with my 18yo son as a captive audience (e.g. last night while building IKEA furniture together), he seems to like ranks 1 through 10, and can follow the rank flattening logic just fine, but thinks that it would be a challenge for most voters to understand the Condorcet method itself, let alone the rank flattening. Granted, this is a sample size of one :-). One issue with doing pairwise comparisons with flattening is the requirement to store 2 different arrays. Is there any way to use the extra DL candidate's pairwise scores to infer the amount that needs to be modified in the normal full-preference pairwise array? If not, what if there were a minimum-approval candidate instead?

Some more thoughts on approval cutoff ballots, for use in Smith//Approval, Condorcet-full//Condoret-approved//Approval, Approval Sorted Margins, or other methods. When setting up an explicit cutoff, the hardest part is making it apparent and natural. With the Condorcet variants above, it is not as crucial that the number of available ranks be even, or even that the number of approved ranks should equal the number of disapproved ranks. Therefore, I think that a positive/negative scale has benefit. So, for example, 3 approved, 4 disapproved ranks: Highest approved = 3, Most disapproved = -3, cutoff = 0 (and is therefore disapproved instead of neutral as some might suppose). Has some advantages if ranks/rates are given names as in Majority Judgment. 4 approved, 5 disapproved ranks (or ratings): Highest approved = 4, Most disapproved = -4, cutoff = 0 [I think this might be the most practical as one can store the rates as zero through 8 internally, by adding 4 to the voted values]. Four approved ranks are analogous to A, B, C, D grading. 5 approved, 6 disapproved ranks / ratings: Highest approved = 5, most disapproved = -5, cutoff = 0. Analogous to 0 to 10 star rating. Disadvantage in that recalibrated rating can't be stored in single digit format. In all of these scenarios, it is clear to the voter that a positive vote is approval, while a zero or negative vote implies disapproval. My preference for the default vote for blanks should be the most negative score, but I believe that the three explicit approval methods I named above should do reasonably well even if the default is zero. On Mon, Jul 8, 2019 at 9:28 AM Ted Stern <dodecatheon@gmail.com> wrote: > I notice there are several explicit approval cutoff methods being > discussed on the list. > > Kevin Venzke has suggested a number of rank-flattening proposals, with > Chris Benham proposing a simplified version of one of them: > Condorcet-using-full-preferences // Condorcet-using-approved-preferences // > Approval. The latter can be used to handle Forest Simmon's 3 approval > cutoff examples. I like this method due to its relative simplicity and > self consistency. I propose calling this it Instant Round Robin Rank > Flattened Fallback, or IR3F2. > > Whichever approval cutoff method one uses, there is some latitude in how > one could implement an explicit approval cutoff. > > I've been thinking about 6, 7 or 10 slot methods, with an explicit > Disapproval Level added as a candidate (any candidate rated at DL or below > is disapproved). If the default rate is lowest rate, then it makes sense > for DL to be an extra "candidate", with the same default. Then approval is > any rate above that given to DL. Using this form, a voter who doesn't > exercise their option to reset the Disapproval Level is basically using > implicit approval cutoff at rate 0. > > Having an extra DL candidate also enables the opportunity for a voter to > disapprove every candidate by voting DL at top rate, which I could see > people doing as some form of None-of-the-above protest. I dimly recall > something of this sort being discussed a few years back. > > When I am able to chat about this with my 18yo son as a captive audience > (e.g. last night while building IKEA furniture together), he seems to like > ranks 1 through 10, and can follow the rank flattening logic just fine, but > thinks that it would be a challenge for > most voters to understand the Condorcet method itself, let alone the rank > flattening. Granted, this is a sample size of one :-). > > One issue with doing pairwise comparisons with flattening is the > requirement to store 2 different arrays. Is there any way to use the extra > DL candidate's pairwise scores to infer the amount that needs to be > modified in the normal full-preference pairwise array? If not, what if > there were a minimum-approval candidate instead? > >

After receiving some private comments, and discussion with other friend groups, the 6 slot method appears to be the most popular. Proposal: A B C = approved; A = most-approved, B = moderately approved, C = slightly approved D E F = not approved; D = slightly not-approved, E = moderately not-approved, F = most not-approved Should there be a Neutral or No preference option? If so, and a candidate is given such a selection on a ballot, it seems like the appropriate way to count that would be to not include that candidate in pairwise or approval counts for that ballot. One friend suggested the following emoji method of denoting the rating/ranking. In case it doesn't show in your browser, the grades A through F are displayed as Very happy to Very sad face emojis. 😍 II😀🙂🤨☹️🤮 On Thu, Jul 25, 2019 at 10:35 AM Ted Stern <dodecatheon@gmail.com> wrote: > Some more thoughts on approval cutoff ballots, for use in Smith//Approval, > Condorcet-full//Condoret-approved//Approval, Approval Sorted Margins, or > other methods. > > When setting up an explicit cutoff, the hardest part is making it apparent > and natural. > > With the Condorcet variants above, it is not as crucial that the number of > available ranks be even, or even that the number of approved ranks should > equal the number of disapproved ranks. > > Therefore, I think that a positive/negative scale has benefit. So, for > example, > > 3 approved, 4 disapproved ranks: Highest approved = 3, Most > disapproved = -3, cutoff = 0 (and is therefore disapproved instead of > neutral as some might suppose). Has some advantages if ranks/rates are > given names as in Majority Judgment. > > 4 approved, 5 disapproved ranks (or ratings): Highest approved = 4, Most > disapproved = -4, cutoff = 0 > [I think this might be the most practical as one can store the rates as > zero through 8 internally, by adding 4 to the voted values]. Four approved > ranks are analogous to A, B, C, D grading. > > 5 approved, 6 disapproved ranks / ratings: Highest approved = 5, most > disapproved = -5, cutoff = 0. Analogous to 0 to 10 star rating. > Disadvantage in that recalibrated rating can't be stored in single digit > format. > > In all of these scenarios, it is clear to the voter that a positive vote > is approval, while a zero or negative vote implies disapproval. > > My preference for the default vote for blanks should be the most negative > score, but I believe that the three explicit approval methods I named above > should do reasonably well even if the default is zero. > > On Mon, Jul 8, 2019 at 9:28 AM Ted Stern <dodecatheon@gmail.com> wrote: > >> I notice there are several explicit approval cutoff methods being >> discussed on the list. >> >> Kevin Venzke has suggested a number of rank-flattening proposals, with >> Chris Benham proposing a simplified version of one of them: >> Condorcet-using-full-preferences // Condorcet-using-approved-preferences // >> Approval. The latter can be used to handle Forest Simmon's 3 approval >> cutoff examples. I like this method due to its relative simplicity and >> self consistency. I propose calling this it Instant Round Robin Rank >> Flattened Fallback, or IR3F2. >> >> Whichever approval cutoff method one uses, there is some latitude in how >> one could implement an explicit approval cutoff. >> >> I've been thinking about 6, 7 or 10 slot methods, with an explicit >> Disapproval Level added as a candidate (any candidate rated at DL or below >> is disapproved). If the default rate is lowest rate, then it makes sense >> for DL to be an extra "candidate", with the same default. Then approval is >> any rate above that given to DL. Using this form, a voter who doesn't >> exercise their option to reset the Disapproval Level is basically using >> implicit approval cutoff at rate 0. >> >> Having an extra DL candidate also enables the opportunity for a voter to >> disapprove every candidate by voting DL at top rate, which I could see >> people doing as some form of None-of-the-above protest. I dimly recall >> something of this sort being discussed a few years back. >> >> When I am able to chat about this with my 18yo son as a captive audience >> (e.g. last night while building IKEA furniture together), he seems to like >> ranks 1 through 10, and can follow the rank flattening logic just fine, but >> thinks that it would be a challenge for >> most voters to understand the Condorcet method itself, let alone the >> rank flattening. Granted, this is a sample size of one :-). >> >> One issue with doing pairwise comparisons with flattening is the >> requirement to store 2 different arrays. Is there any way to use the extra >> DL candidate's pairwise scores to infer the amount that needs to be >> modified in the normal full-preference pairwise array? If not, what if >> there were a minimum-approval candidate instead? >> >>