To make a good decision, collect accurate and relevant data and then analyze it intelligently.
As sensible as this advice may seem, we often settle for merely adopting some procedure to standardize how the decision is made and call that approach fair. Eventually, we come to feel that the familiar process is a good one, possibly even the best possible. So, we mostly use plurality voting, a system no doubt handed down to us by stone age ancestors. But surely for important decisions we should at least re-think these traditions; we might find a way to improve on that ancient tradition.
Experience has taught us that it is wise to support our decisions with data. But surely that data must be relevant and reasonably accurate for that to work. In the previous article of this series, we focused on the accuracy of the data collected for each of four different voting systems; plurality voting and ranked-choice voting were judged as unsatisfactory systems merely because of the poor reliability of the data that they collect; decisions guided by faulty data tend not to be good decisions. But accuracy of the data is only part of the story; even if the data is accurate (and sufficient for the purpose) the method used to evaluate that data must also be sound. Even if the data itself is relevant and sufficient, an erroneous analysis of the data can produce poor decisions.
For an electoral system, testing the soundness of the analysis is challenging because generally we do not know what the right outcome should be. A good test might be possible if we could find an exceptional election where the correct outcome is apparent, that might let us detect such a fault with a voting system. Multiple examples would seem better, but even just a single example is surely enough to raise serious doubts about a voting system. If a voting system fails the only test we know about, surely that voting system should be avoided, at least for making any important decisions.
Might we find an election where the data from voters' is accurate and where the right winner is known? If we discover an election which meets these two conditions, that would be an ideal test case for evaluating the dependability of a voting system. For such an election, failure to elect the correct candidate would demonstrate that the voting system itself is faulty and not to be trusted.
An earlier article in this series, comes to mind because it described an election where there seemed little doubt about which candidate should win. But in that ranked-choice election, a different candidate was elected. That election involved eleven candidates. Although that election would serve our purposes here, a very similar but simpler example, with only five candidates will work here just as well.
The election in question is for a small town's mayor. 4005 voters cast ballots. The town is closely divided regarding two different issues, with half of the town concerned primarily with issue A and the other half are most concerned about issue B. All proposals for resolving these issues are expensive and it does not seem practical to address both issues at once. Nevertheless, all 4005 voters do have opinions about solving each issue. Two of the candidates, A1 and A2, have campaigned on their own, very different, proposals concerning issue A and voters split close to evenly in their support for each of those two candidates. The same can be said about issue B and the two candidates, B1 and B2, who have urged voters to support quite different approaches.
But in addition, there is one more candidate, C. Observing that there currently is no clear consensus about either of these two issues, C promises that, if elected, she will delay taking any action, thereby allowing time for further study and discussion of both issues. Voters are united in thinking that (unless the one proposal they most favor is adopted) that C has the right approach. While candidate C is not the first choice of even one voter, she is the second choice every voter.
Which candidate should win the election? If C wins then all voters will feel satisfied, though probably not ecstatic; although they would have preferred their first choice, their second choice candidate did win. However, if any of the other candidates wins, more than half of the voters will be greatly disappointed, not getting their first or second choice and in many cases not their third or fourth choice either. Surely a voting system that fails to elect C in this election is faulty.
Table 1 summarizes the voting for both plurality and ranked-choice voting. In an election using plurality voting, B1 wins with 1003 votes, to the disappointment of 3002 voters. But an interesting consideration is that with plurality voting, the vote might be different from what is shown in Table 1. If C is widely considered the most electable candidate, then many voters will vote strategically, primarily taking electability into account. They may lie about their true preference by marking C as their first choice. In effect, the inaccuracy that lying creates within the ballot data would override the deficiencies inherent in the voting system to make the rightful winner, C, the actual winner. This is an example of good strategic voting working as intended. In this case voters deserve to be congratulated for having found a way, by voting strategically, to draw the best outcome. It would seem to be a mistake to expect such serendipity in other elections, however. A more sensible response would be to adopt an evaluative voting system (such as BAV) that would make voters feel more comfortable about voting truthfully.
If we were to conduct the entire, four-stage ranked choice vote tally we would find B1 to win the ranked-choice election (in the fourth and final round by a margin of 501 votes). But with considerably less effort we can see that C is eliminated in the first round of the tally making it certain that one of the other candidates will win the election. Whichever of the four remaining candidates would win, the count of disappointed voters would change very little - it would be around 75% of the voters. An easy examination of table 1 easily reveals that the ballot data to quite accurately reflect voter opinion so in this election, it is either the analysis of that data (or perhaps relevance of the data collected) that is at fault.
For the evaluative systems, BAV and especially for its unbalanced relative, approval voting, Table 1 is insufficient for determining the voting. However, voter attitudes, as described, suggest that many, let us say half, of the voters will support both C and their first-choice candidate. And it seems unlikely that any voter will oppose C. This makes it likely for C to win the election with a net vote of approximately 2000. Using BAV, the other candidates would likely each have a net vote closer to 500. For approval voting, these net votes would be higher since voters will lack any option for effectively expressing opposition; however, it would still seem likely that these other candidates would have net votes of 1000 or less and consequently that C would win the election.
Both evaluative systems seem to pass this test, but there might be other tests that are yet to be discovered, perhaps forcing us to reconsider our favorable opinions about those conclusions. On the other hand, a single example of failure seems definitive about unreliability. At present, Table 2 would seem to summarize judgements based on this article together with the article that immediately precedes it.
Post Article Comment and Rate This Article
