While analyzing some approval voting election results, I came across a problem that I attributed at least partially to “junk candidates”: candidates who, whether due to their radical political views, dearth of charisma, or general weirdness, can’t realistically achieve substantial support. When approval elections featured lots of candidates, voters didn’t vote for as many candidates as you would expect based on their behavior in elections with smaller numbers of candidates. Though this is likely attributable to several factors along with the appearance of junk candidates (voter fatigue and limited voter knowledge), it got me thinking about the many ways that these junk candidates can mess with attempts to mathematically model elections.
For a lot of purposes, these “junk candidates” don’t matter much; they aren’t going to win elections unless something extremely strange (or several extremely strange things) happen. The main problem with these “junk” candidates, from the perspective of someone who is trying to model electorates, is that, while they are rarely threats to win an election themselves, they can affect the election in all sorts of (unpredictable) ways. They can act as spoiler candidates, by taking votes that would otherwise go to more viable candidates. They can impact other measures of elections, such as how many candidates we should expect voters to include on ballots in voting systems like Approval or Ranked Choice where voters have the freedom to include or exclude as many candidates as they want. They can mess with models of voter decision making, which can capture how voters behave regarding “normal” non-junk candidates but break down for junk candidates.
I’m somewhat of an advocate of spatial models of voting, where voters and candidates are plotted based on their ideological positions, and voters prefer candidates based on their ideological positions. This kind of modeling can capture a lot of the dynamics we see in real elections, from the obvious “voters like candidates similar to them” to the more nuanced “voters prefer candidates who push policies in their preferred direction as much as possible”.
But these spatial models can’t always sniff out a junk candidate. They’ll inadvertently treat some kinds of junk candidates like normal, viable candidates, and will have no idea how to even start modeling others. Getting an idea of what kinds of junk candidates there are and how models of elections can handle them is the first step to building models that can better handle these junk candidates.
Four categories of “junk” candidates
Junk candidates come in some wildly different forms.
Fringe ideologues
These are candidates with political positions that are so far out of the mainstream that they are unable to attract any significant base of support. Imagine a communist running for Governor of Kansas; they’re so far away from the political positions of the vast majority of voters that they have a very low ceiling in all but the most bizarre of elections.
Spatial models of voters have the tools to handle fringe candidates, since they naturally capture that these candidates have very different political positions than the vast majority of voters. A purely directional model, where voters want candidates who push politics in their preferred direction as much as possible, will unrealistically favor fringe candidates. Instead, voters should, at least to some degree, prefer candidates based on how ideologically similar they are to themselves.
The unviable “normal” candidate
This is a candidate who looks like a perfectly viable candidate when you focus on their political positions and background, but for whatever reason, lacks any notable base of support. Perhaps they got in the race far too late, like Deval Patrick in the 2020 Democratic presidential primary, they lack name ID, campaign funding, or charisma, or voters perceive them as being redundant to another, better, candidate. Look to party primaries to find countless examples of candidates whose views suggest that they would be viable candidates but whose electoral performance disagrees.
Spatial models really don’t naturally handle these candidates well. If a candidate’s ideological position makes them look viable, the spatial model will treat them as such.
This is different than candidates getting squeezed out because their ideological niche is flanked by other candidates. In a spatial model, a candidate can get squeezed out of contention if there are other candidates around him. For instance, if a candidate has a competitor just barely to his right and another competitor just barely to his left, then virtually all voters will prefer one of his opponents over him, rendering him unviable in many voting systems due to a phenomenon known as “center squeeze“.
These unviable normies aren’t necessarily getting ideologically squeezed out. They might even be the candidates who should be squeezing out the other candidates, according to spatial models. Unviable “normal” candidates aren’t necessarily extreme examples of center squeeze, they’re candidates who get lost in the mix, despite their ideological positions.
We could add an additional value — name recognition or something similar — for each candidate that represents an ideologically independent measure of their viability. For the 2020 Democratic primary, Joe Biden would score high on this metric while Patrick would score very low.
Single-issue candidates
These candidates are so focused on one issue that their positions on other issues either aren’t known or are completely overshadowed. They’re particularly “junky” if their chosen issue isn’t high priority for most voters: think Lincoln Chafee running for President with the platform of making the US adopt the metric system. These candidates may have extremely narrow appeal, but don’t have the means or perhaps even the desire to expand their base beyond the small number of supporters who are also hyper-focused on one particular issue.
Candidates that are hyper-focused aren’t well-modeled by ideological spaces, since each dimension is meant to capture multiple issues, and the single-issue candidate’s positions on most issues either are overshadowed or aren’t even known. Plopping a single-issue candidate down in ideological space and modeling them like any other candidate won’t work. They might as well not exist in the same ideological space as other candidates.
We could try adding dimensions to our ideological space to capture more nuance in political preferences, but this launches us into a host of conceptual issues. How many dimensions should we add? Is every dimension equally important? Surely a presidential candidates’ position on the adoption of the metric system has virtually no significance compared to how left- or right-leaning the candidate is on economic issues. What makes an issue worthy of a dimension? We might be better off treating these candidates like fringe candidates or unviable normies.
Weirdos and joke candidates
These are the folks who look at the political landscape and boldly proclaim, “There is not enough discussion about alien abduction.” Maybe they’re running as a joke, or maybe they truly believe that stopping the government from controlling citizens’ minds with fluoride should be a top priority, but they aren’t prioritizing traditional political issues.
There may be some conceptual overlap with single-issue and fringe candidates, but what set weirdos apart is that the issues they care about may not even be what voters even consider relevant political issues. These candidates really don’t fit anywhere in ideological space, because the issues they care about only incidentally overlap with mainstream politics, and those overlaps are barely noticeable compared to the candidate’s other… qualities.
They can act as protest candidates (either knowingly or unknowingly) and capture limited support from voters who don’t want to vote for any of the viable candidates, but their ceiling of support is (typically) very low. Spatial models can chuck them out on the fringes as a shortcut, but which fringe they belong to can be arbitrary. These candidates tend to be agents of chaos, which is not something that most models like. Thankfully, the amount of electoral chaos they create tends to be miniscule.
Why does understanding “junk” candidates matter?
For a lot of purposes, we can treat junk candidates as the name suggests — toss them out and forget about them. But having some ideas for how to think about them can help us understand when they can throw dirt on our tidy election models. There’s certainly value in knowing when your model will handle something well and when you need to adjust your model for a particular kind of case.
If we only care about finding out which candidate will win an election, we can often ignore junk candidates. But there are a lot of other important features of elections that are crucial for understanding how different voting systems perform, like ballot exhaustion in ranked choice systems or number of approvals in approval voting, which are heavily influenced by the presence of junk candidates. Part of understanding these phenomena and what they tell us about how well voting systems will work in real elections is understanding the effects of junk candidates on the data we get.