[EM] the "meaning" of a vote (or lack thereof)

[Juho Laatu]

On 27.8.2011, at 17.38, Jonathan Lundell wrote:

> On Aug 27, 2011, at 12:25 AM, Juho Laatu wrote:
> 
>> On 27.8.2011, at 2.13, Jonathan Lundell wrote:
>> 
>>> On Aug 26, 2011, at 1:17 PM, Juho Laatu wrote:
>>> 
>>>> On 24.8.2011, at 2.07, fsimmons at pcc.edu wrote:
>>>> 
>>>>> But back to a possible generic meaning of a score or cardinal rating:  if you think that candidate X would 
>>>>> vote like you on a random issue with probability p percent, then you could give candidate X a score that 
>>>>> is p percent of the way between the lowest and highest possible range values.
>>>>> 
>>>>> Note that this meaning is commensurable across the electorate.
>>>> 
>>>> This is the best proposal so far since this takes us as far as offering commensurable ratings. Maybe we should add also voter specific weights to the different issues.
>>>> 
>>>> Voters could start from the set of issues that the representative body or single representative covered during the last term. They could adjust those issues a bit to get a list of issues that are likely to emerge during the next term. That makes a list that is the same to all (and that makes the opinions therefore commensurable). Weighting makes the results more meaningful since to some voters some questions might be critical and others might be irrelevant. Without the weights the ratings might not reflect the preference order since we might have misbalance due to too many questions of one kind or due to questions of varying importance.
>>>> 
>>>> In principle one could collect the opinions also indirectly by generating an explicit list of issues and asking voters to mark their opinion an weight on each issue. That list could be structured or allow voters to indicate the importance of each group of questions. It is however not obvious how the questions should be grouped. Grouping could also influence the results. It would be also difficult to the voter to estimate the level of overlap between different issues. In practice one may get equally good results by simply asking "how much do you think you will agree with this candidate (from 100% to 0%)".
>>> 
>>> I'm repeating myself here, sorry, but...
>>> 
>>> 1. Why isn't this replacing one ineffable candidate utility with n ineffable issue-agreement utilities (where each issue utility is the (signed) issue weight)? 
>> 
>> Maybe because the voter answers question "how often do you agree" instead of "how strongly do you agree". Time and number of occurrences are commensurable but voters' interpretations of the chemical and physical reactions in their brain and heart are not (maybe one approach would be to use some instruments to measure brain and heart activity with some external device :-) ). With weights added the question continues "... and estimate the importance of those agreements". This is based purely on personal feelings as taken from the brain and heart, but that should not destroy commensurability since all the voters are still on the commensurable scale from 100% agreement to 0% agreement, and the voters are still supposed to answer question "how often, if all issues would get the time that they deserve".
> 
> Set aside the question of the meaningfulness or commensurability of utilities. My point is that such a scheme merely changes the need for a voter to determine one utility (for the candidate) to determining n utilities (for n issues). And the issues we care about tend not to be simple.

I attempted to create a scenario where we do not try to measure utilities (= strength of personal feelings) but use some other units that can be measured (= same scale for all). In this case the unit of measure was the number of agreements of some given set of issues (taken from fsimmons' mail).

If we use fsimmons' original scenario to compare voter opinions and candidate opinions using a fixed set of binary decisions, then the strength of feelings plays no role. We measure only if the voter agrees with some candidate. That should be commensurable.

If we add weights, and consider also overlaps (/ non-orthogonality / grouping) of the issues, and if we have also other than binary decisions, we have to be careful not to include any "strength of preference" style measurements into the ballots. I hope my explanation managed to stay on the non-utility side also here.

I tried to cover the problem of dividing one question to n smaller questions (whose answers might contain utility strength information) in the paragraph below. I hope the answers to the n smaller issues were not utility based, nor the way they are summed up (using weights and overlap estimates).

My claim was thus that although I used weights that are based on personal feelings, the end result (= ratings of the ballots) would still measure the number of agreements rather than the strength of personal preferences.

Let's take one of the small decisions. It could be a binary question on if we should have a new law L. Voters and candidates either agree or not. Every candidate gets either a plus or minus point from the voter. This does not depend on how strongly the voter feels about this issue.

Then the voter adds weights to all the issues (taking also overlap into account). Although the weights depend on personal feelings, the sum of all small weighted issue agreements in commensurable since those values are still on the 100% - 0% commensurable scale. It does not matter if different voters use different weights. (I don't have any better short description available right now. Please tell if you think that this assumption is wrong.)

I don't know if I added much new information here, but hopefully I managed to explain the idea of avoiding utility style measurements better than in the previous mail.

Juho


P.S. One alternative way to construct this scenario is to allow every voter to list 20 questions that cover areas that the voter feels important. Then every candidate is compared against this list. In this approach there is no weighting.


> 
> 
>> 
>> The n issues could be all binary decisions, "agree" or "disagree". In that case they are commensurable. If they are more complex, e.g. numeric decisions, then the voter must estimate the level of agreement somehow. Maybe the voter should decide on some hard limits to what is agreeable and then decide which candidates agree with him and which ones do not. Also numeric differences would do. This way we can (at least in principle) escape the non-commensurable "strength of agreement" questions.
>> 
>>> 
>>> 2. One doesn't vote for a candidate strictly on predetermined issues. You don't know which issues will arise in the next 2-4-6-whatever years, and the work of an elected official (a president in particular, but also other offices) consists of more than voting on issues.
>> 
>> Yes, but the set-up is the same for all voters. Voters will make wrong guesses on what will happen during the next term, but in principle they will all answer the same commensurable question and their answers will approximate this ideal.
> 
> 
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info