2357: Polls vs the Street

Explain xkcd: It's 'cause you're dumb.
Jump to: navigation, search
Polls vs the Street
Other pollsters complain about declining response rates, but our poll showed that 96% of respondents would be 'somewhat likely' or 'very likely' to agree to answer a series of questions for a survey.
Title text: Other pollsters complain about declining response rates, but our poll showed that 96% of respondents would be 'somewhat likely' or 'very likely' to agree to answer a series of questions for a survey.


This comic discusses getting data or opinions through a study (polls) or by getting them anecdotally (on the street). The phrase "voice on the street" is commonly used by news reporters who get opinions on issues by literally asking people walking by what they think, and has been previously mentioned (and derided) in 756: Public Opinion.

Many news organizations, and other data-driven institutions, conduct or commission polls to assess the opinions of the general public. These polls generally rely on asking a randomly selected and anonymous set of people a set of consistent, prepared and deliberately crafted questions about their opinions, experiences, and intents. The results of these polls are traditionally held to reflect the views of the public as a whole, within certain margins for error. Many news shows also conduct "man-on-the-street" interviews (more formally known as vox populi, "voice of the people"), to provide a human face of "the public" and engage viewers more. Many pollsters, pundits, and politicians worry that polling data may not accurately reflect the true trends in public opinion, as in the infamous "Dewey Defeats Truman" newspaper headline, and so White Hat is here extolling the virtues of interviewing "real people" to get at that ground truth.

White Hat suggests that, while polls suggest "candidate X" is more favored, the people on the street that White Hat interviews are more supportive of "candidate Y". He implies that his experiences reflect reality better than the polls. There are a number of reasons why polls may not be entirely representative. The sampling method might not be genuinely random, some groups might be less likely than others to respond to a poll, and it's argued that some people express views that they consider to be more socially acceptable, even in anonymous polls, but vote differently in actual elections (examples include the "Bradley effect" and the "shy Tory factor"). Despite these concerns, there is little evidence that individual conversations do a better job at determining public opinion than polling.

This comic is very likely a reference to the 2020 United States presidential election, which occurred on November 3, 2020 (about 2 months from the time of the comic's publication), which Democrat Joe Biden democratically won. Most polls showed Biden polling ahead of incumbent Donald Trump, but Trump and his supporters frequently argued that the polls are inaccurate, often arguing that they personally knew or talked to many Trump supporters, and few Biden supporters. At the same time, the fact that Trump won the 2016 election astonished many (including Randall) who had seldom met Trump supporters in their own lives and within their own social circles. This kind of anecdotal evidence is generally a poor basis for gauging public support, for multiple reasons. Politics in the US are frequently regional, so sampling in a single area is unlikely to be representative of the whole country, or even a whole state. It's not uncommon for gathering places (both physical and virtual) to attract people from one political group more than another, producing a skewed sample. If someone uses their own perception, rather than rigorous analysis, confirmation bias is likely to have a major impact (a person might pay more attention to supporters of their preferred candidate, and ignore political opponents).

This strip lampoons such thinking, as it quickly becomes clear that White Hat's methodology is heavily driven by selection bias. He's apparently talking only to the residents of his town, and extrapolating those results to the whole country. By that logic, he would conclude that everyone has visited his town, and most people live there. It is true that he's getting "ground truth", but it's also true that he's only sampling a very small (and highly idiosyncratic) part of the whole population.

The punchline in the final panel is a joke about the phrase "on the street". Usually this phrase means "anywhere out in public where the interviewer can openly approach people" (often a sidewalk near the studio), but White Hat is presumably taking the phrase literally and interviewing people he meets on the roadway. In the US, roads are generally reserved for vehicles (cars, trucks, motorcycles and in most areas bicycles), and walking or standing in the roadway for long periods is dangerous and usually illegal. White Hat's sample population thus consists only of the people who can be found on the roadway outside of designated pedestrian zones, who are generally from the small fraction of the population who have no qualms about the risks of being struck by moving vehicles or causing accidents when drivers swerve to avoid them.

The title text is a joke about selection bias and tautology. People who don't feel like taking surveys wouldn't get as far as answering a survey question about survey questions. However, it does touch on an issue raised by FiveThirtyEight after the election: that polls only measure people who are interested in answering polls, and that population may not be politically representative of the entire country.


White Hat: Polls are just numbers.
White Hat: You have to talk to people on the street.
White Hat: Polls say most people support <Candidate X>.
White Hat: But the people I talk to on the street support <Candidate Y>.
White Hat: Polls claim most people don't live in my town and have never been here.
White Hat: But the people I meet on the street tell a very different story.
White Hat: According to polls, most people don't like playing in traffic.
White Hat: So why do I never seem to meet these people on the street?

comment.png add a comment! ⋅ comment.png add a topic (use sparingly)! ⋅ Icons-mini-action refresh blue.gif refresh comments!


Love the title text! Fwacer (talk) 23:56, 9 September 2020 (UTC)

I Edit Conflicted with someone (2 minutes too late, after quite a bit of typing, then half a dozen Captcha submissions - just two to put this text in). If anyone wants to review my attempt, I'm HTML-commenting it in this gap... ...I already knew I'd have to Wikilink some bits, and can see at least one typo. Maybe I'll integrate some into what's there now, myself but probably not tonight. 00:39, 10 September 2020 (UTC)

(Self-reply) Looking into it, I must have been editing for a whole hour, actually. Didn't feel like it, but given there wasn't even a transcript when I started (but the BOT had been replaced) I must have been. And I want paying for all the Captcha responses I'm asked for. It seems I'm either being 'a useful idiot' for slavishly helping the Algorithm, or I am far better(/worse?) at identifying traffic lights, crosswalks, motorbikes, traffic and bicycles than "the man on the street" that pre-populated the Captcha knowledgebase thresholds... 00:54, 10 September 2020 (UTC)
I swear half of the answers were wrong in the first place. One time I took a capatcha and it misidentified a mailbox as a parking meter and I had to answer wrong on purpose to get through.
One that I didn't get a 'by' on (I'm convinced that being correct just gives you another test in order to generate that result with more authority) was a request for "bicycle" that featured a surface-painted bike-lane symhol. I said "skip" as there was no actual bike (like features that clearly look cross-walkish but aren't even on the road, or horizontal). So unless most other people had a more generous/playful citeria, I should have been correct. And I'm a whiz at identifying hydrants (in my mind?) Of all kinds of colours, but I think there must be far too many opposing opinions. (Traffic Lights: Do you just highlight the lit bits? The composite frames upon which all the lights sit? The whole lot including the support poles/cables? And do you choose frames that have a sneaky tiny bit of overlap of your chosen feature, but are otherwise mostly empty? Or exclude squares that have a slice that wouldn't be recognised as such if given in isolation?) 13:04, 10 September 2020 (UTC)
Select everything containing a part you would expect to receive if you ordered it fully assembled from amazon. Full frame, but cables not included. All of the car. All the nuts on the hydrants. All the lines on the crosswalk. Mailbox includes the stick in the ground, but not the wall of the house when mounted on one. If it takes a second glance to tell it isn't what it asked for, then select it anyways. normies don't have time to double check their answer when trying to post their lols on cat videos. 13:59, 11 September 2020 (UTC)

Well, sounds like what I do with edge-cases (except I do check carefully, so that I'm right, no matter what), but if other people are being sloppy, I'll have to he careful to be sloppy, eh? ;) 01:10, 12 September 2020 (UTC)

Reminiscent of this beautifully snarkastic tweet. https://twitter.com/DavidLJarman/status/1302719537234599936 03:27, 10 September 2020 (UTC)

It seems to me that it is morally indefensible to tell the truth to pollsters. And also to people doing a Vox Populi. If you believe one party is better for the people in your country, then you have a moral duty not to sabotage them. But if everyone tells the pollsters honestly that they are definitely going to vote for party X, you could have everyone believing that it's a foregone conclusion that party X will win, so they don't bother to go out in the rain to vote, and party Y gets in - because they thought their party needed their vote. 21:45, 12 October 2020 (UTC)

In England the word on the Street is most often SLOW (in Wales it's is ARAF SLOW) :-D RIIW - Ponder it (talk) 18:56, 11 September 2020 (UTC)

Or BUS / BWS STOP 09:53, 13 September 2020 (UTC)