A postmortem on the polls
Here’s an interesting article by Michael Barone that tries to say why the pollsters got it wrong.
First of all, I’d like to add that they didn’t get it quite as wrong as some people think. Many of the polls in the last week or so were well within the margin of error, and the election also ended up being close. Not as close as in 2000, but close enough that many states were won by small margins, and the loser won the popular vote, as in 2000. Here’s a post by Scott Alexander that says the same thing about the closeness of the polls—only he wrote it before the election.
But a win is a win is a win, and this particular win by Trump was predicted consistently only by the LA Times polling, which kept showing Trump ahead. But there’s no doubt that in general the poll averages (which have often been accurate in the past) got it wrong.
Barone mentions five elements that may have led to the error; please read the whole thing.
I’ll add that I’m very aware that some people think that polls are purposeful and strategic lies. I do not ascribe to that notion. As I’ve written many times before, however, polls are inherently flawed. There are too many things they must predict in order to predict the winner—such as, for example, turnout, which is one of the most changeable, difficult, and important to get right. However, pollsters would actually like to get it right, particularly in the weeks close to the election, which are the polls that matter most. Pollsters’ reputations are damaged by getting it so wrong, and they are running a business.
Turnout is always hard to predict, but it was especially hard to predict this year because the candidates (both of them) were so unusual. The Shy Tory effect may indeed have been in effect, as well. But purposeful skewing of polls in Hillary’s favor would have been a dangerous practice for the left anyway, as likely to depress Democratic turnout through cockiness as it would be to depress Republican turnout through despair.
So the polls were close enough at the end that, although they consistently predicted a Hillary win, there was no way to say that a Trump loss was an absolute certainty, and I don’t think many people were saying that. A Trump win was always a possibility, one I seem to recall as about 1 in 3 towards the end, and 1 in 3 aren’t such bad odds.
Here is a particularly interesting point by Barone about the geographic breakdown of voters these days:
The fact that Democratic voters are so clustered, concentrated heavily in central cities, sympathetic suburbs and university towns, has helped its presidential candidates in past elections (this one, not so much) but hurt them in congressional and legislative elections conducted in equal population districts.
Clustering, to oversimplify a bit, means that Democrats tend to be concentrated in a few 80 percent Democratic areas, while Republicans are spread more evenly around the country in districts that average somewhere between 55 and 60 percent Republican. To see the effect, consider the number of Republicans and Democrats elected to the lower houses of state legislatures in what turned out to be the 2016 target states. I cite the lower house, because some such states elected only half or none of their state senators this year; Virginia elected no state legislators except to a couple of vacant seats.
The states are ranked in the order of their percentage margins, starting with the most Republican:
Iowa 59-41
Texas 94-55
Ohio 66-33
Georgia 116-62
Arizona 32-26
North Carolina 72-45
Florida 78-39
Pennsylvania 118-81
Wisconsin 64-35
Michigan 63-38
New Hampshire 201-161
Minnesota 75-57
Maine 72-76
Colorado 28-36
New Mexico 31-37There was lots of straight-ticket voting this year, which accounts for the Democratic edges in the more Democratic states at the bottom of the list. But the bottom line is pretty astonishing, particularly considering that these are states which were, like the nation as a whole, roughly equally divided between the parties at the presidential level.
It does seem as though polls have been less accurate in recent years, but one reason may be that the electorate has grown more evenly divided. Close races are harder to call, because small variations matter more. Pollsters also have had to scramble to keep up with changes such as the proliferation of cell phones. But the poll response rate still goes down (I never answer polls myself, for example) and that’s hard to correct for. And even exit polls—which after all, don’t have to predict anything, because everyone in an exit poll actually voted—still have the problem of low response rate as well as the complication of the increasing prevalence of early voters, a whole bloc that is missed by exit polls.
I live in an area in which I have the luxury of being able to vote on Election Day without having to wait or stand in line. I’ve wondered how much the inconvenience for those in more densely populated areas affects turnout. Whether it affects Democrats more than Republicans, for example. Or how much advantage it gives to the candidate with the more enthusiastic (or angrier) supporters. And whether those factors are in any way accounted for in polling.
In terms of influencing a future vote, a poll berween two candidates either helps the leader, hurts the leader, or makes no difference. Common theory is that a poll helps the leader. So called positive feedback. Therefore the trailing candidate faces a headwind.
The averages of the national and state polls only accurately predicted the outcome if you apply the “margin of error” of the individual polls themselves. However, one of the reasons you average polls is to get the input of a much larger sample size than is contained in the individual polls by themselves, which is supposed to narrow the error bars for confidence levels. I keep seeing a bunch of individual polls each with 3-4% 95% confidence levels averaged together to get a number with the same 3-4% 95% confidence levels, at least post hoc.
The polls were badly wrong, especially at the state levels, and only slightly less wrong at the national level.
I still think, though, the major factor for the mistakes are modelling errors that assumed the electorate would conform to 2012 and 2008. I often wonder whether or not this was intentionally done to make Clinton look stronger than she was, or whether it was a natural over-response to the miss in 2012 in the other direction.
2012 is interesting because the polls showed Romney stronger than he actually ran. I guess the real question I have about this as compared to 2016 is why that was the case. The biggest difference in the two elections is actually that Trump had a great part of the Republican establishment against him, including traditionally Republican supporting newspapers, who, by the way, also poll their states.
The error was all in the turnout models they used.
I don’t know that the polls have a rigorous set of questions to test for that. If not, and assuming patterns hold from prior elections, is going to skew the results.
Having two historically net negative candidates made this issue a critical point.
How much of the margin of error covers for a measurement error, due to turnout assumptions? IDK.
BigMaq,
A lot of mischievousness can be hidden in sampling method. Turnout models have to be based on something, and I think an uncritical pollster is making a big mistake in assuming Hillary! was Obama.
Yes, you can get at likelihood of turnout by proper questioning, if you want to. My suspicion is that, in this election, a lot of people wanted to be fooled.
And I will point out that in 2012, a lot of the people who wanted to be fooled were Romney supporters, i.e. mainline Republican establishment figures. In that example, I think the problem was them assuming, and wanting to, Obama 2012 wasn’t Obama 2008. I wonder what happens if one takes the samples of 2016’s polling and model it on the election of 2004? I suspect, in that case, one would have realized Trump was 50/50 at worse.
The impact of the voting rights act, champion by Democrats, has devastated the Democrats. By requiring majority minority districts, it has caused the Democrats to have leftists re-elected in safe districts, while the other democrats are forced into districts with a 10% Republican advantage. Over a few decades due to the seniority system choosing house and senate leaders, the party has lurched even further left. So much that you have a 911 truther radical Muslim getting serious attention as the face of the party.
“Yes, you can get at likelihood of turnout by proper questioning, if you want to. My suspicion is that, in this election, a lot of people wanted to be fooled.” – Yancey
And lose their credibility (and business) in the meantime?
How many of these pollsters (companies and their top level staff) rely exclusively on political polls for their income?
Even if they were trying to curry favor with the dems, how does their performance in this election cycle help them?
Seems to me many have lost a fair bit of credibility. That HAS to hit their bottom line with every type of client they might have.
BM
They are simply not able to jump away from their own hopes and biases.
Or, as Scott Adams would term them, they are moist robots.
The 0-care premium bumps were decisive come November 8th.
I posted time and again that the race was as tight as a drum.
It was — until the last week when the undecideds broke 8 to 2 for change — for Trump.
Minnesota is interesting. It went for Hillary. Yet, at the county level Trump swept the state. He just lost the twin cities. That one leftist bastion pulls the entire state from red to blue.
The same thing can be said for Illinois, Michigan, Wisconsin.
Hillary swept the gibsmedat vote. She was blown out with taxpayers.
The decisive demographic was taxpayers.
‘It does seem as though polls have been less accurate in recent years, but one reason may be that the electorate has grown more evenly divided.’
Actually, the polls have grown more accurate in recent years.
Well, pollsters have discussed some of their problems with getting responses. I don’t know how they account for people who have no land line phone, or if they do. And when they connect with someone, I am sure that many people, like Neo and me, will not answer their questions.
Older and Wheezier, I think the Democrats have done a pretty good job of taking the inconvenience out of voting–and some of the legitimacy as well. I have not gone to the polls since I moved to California; and none of my extended family go either. We all vote by mail. (I may have mentioned before that my wife was registered in California by a man sitting outside of the grocery store collecting petition signatures for something. He was wearing a “Retired Navy” ball cap, and we struck up a conversation. Because of the connection, he just went ahead and registered my wife to vote with no ID. I have no idea how he was authorized to register voters. I had already registered at DMV. Does that say anything about the legitimacy of the vote?)
@dilblert – if you continue to reference Scott Adams, I will continue to ignore you, since you are obviously a “meat puppet”.
/s
We quit answering our land line in late September.
Don’t forget the Bradley Effect. The Trump voters had been slagged to a staggering effect. Some of them may not have wanted to identify. Various kinds of retaliation from supposedly confidential sources have been rumored.
Are any pollsters in the business of trying, by exaggerating a lead or a deficit, to reduce turnout by one side or another?
The problem with the argument of the winning happening within the margin of error overlooks that the errors all fell on one side of the spread: they were all Hillary-weighted.
That’s bias. There is still error, but so many polls skewing in the same direction is a demonstration of significant bias.
Whether the bias was deliberate or not is the real question. I’ve read that it’s not deliberate, but it has certainly persisted through several election cycles. And that it drives the pollsters up the wall ’cause they cannot explain it, nor therefore can they eliminate it.
Skewed polls only serve hucksters. The professional polling organizations’ bread-and-butter depends upon accuracy, which accuracy currently escapes them.
Steve S:
No, that is not an indication of bias.
First of all, one poll made an error in the opposite direction, and did so consistently. That was the LA Times poll, as I said. It had Trump winning by too big a margin. And the LA Times is definitely not a pro-Trump paper.
The other polls, however, did consistently err in the direction of favoring Hillary. But that gives us absolutely zero information as to whether the error was purposeful or accidental. If they were all using a flawed assumption about turnout, for example, and that assumption was similar for all of them (which is very plausible—for example, basing turnout on a previous election) then you would expect the errors to all point in the same direction.
I am not inclined to participate in polls, generally speaking. This year I was even less inclined than usual because of my very high level of frustration with the trajectory and outcome of the Republican primary, and later because of the sorry two candidates (IMO) we ended up with in the general election. In the weeks leading up to Nov. 8, I got literally dozens of phone calls from unknown numbers on my landline (here in the heart of western Pennsylvania’s Rust Belt). I ignored them all. I did likewise with unknown callers on my cell phone. I knew I wouldn’t be voting for Hillary, but I didn’t definitively decide to vote Trump until I was driving to the voting booth, and even then did so very reluctantly because the polls were tightening up in my state and my vote might actually make a difference in defeating Clinton. Afterward, I kept my vote under wraps with the majority of my friends and family (mostly Dems) and my colleagues in higher education.
Whether I’m a Shy Tory or not, bottom line, my opinions and vote were not reflected in any poll, including exit polls. I suspect I am not alone.
neo
I would expect it would be the business of pollsters to use correct models, which includes brainstorming the question as to whether their old turn out model is relevant this time.
Should we not, would say somebody whose pension is fully vested, ask ourselves how we know this. Remember Tom Bradley. Brexit. Okay, I’ve already cleaned out my desk.
If it is true they all used the same/wrong turnout models, then they’re more collegial than competitive, and not worth worrying about.
Related, after the Brexit vote, a British writer argued, “if you want more accurate polls, stop blaming Shy Tories”:
https://www.theguardian.com/commentisfree/2015/may/12/polls-shy-tories-left-conservatives
It shouldn’t be lost on us that the guy elected put a lot of credence in the “beautiful” polls early on, and then later to claim that they are “biased” against him.
The value and credibility of the polls seems to then be dependent on what outcome one desires – a rather poor standard to judge the quality of polls from.
.
If rolling the dice 10 times and having a number less than seven shows up eight of those times, is the person rolling the dice biasing the rolls?
If 10 people roll a pair of dice and each have more than five instances of less than seven, are they all biased?
Or, could the results individually and collectively be within the “margin of error”?
If results that fall within the margin of error are “obviously biased”, then one doesn’t understand the mathematical relationships behind the notion of margin of error.
.
As for “Bradley Effect” or “shy Tory”:
It is completely at odds with the enthusiastic and sizable crowds that trump and his supporters bragged about.
It is completely at odds with the relatively small margins trump won by over clinton, and the lower popular vote total.
It is completely at odds with the smaller share of the eligible voters than what Romney (and possibly McCain?) received.
.
clinton had a catastrophic loss of votes relative to the last two elections.
trump won a segment of traditional dem voters. This put trump over the edge in the rust belt with enough electoral votes to swing the election.
There were ~10M new eligible voters from 2012. Neither candidate picked up their ~5M share of those additional votes, as the total votes cast were less than 2012’s.
If trump picked up dem voters, that must mean that a very large number of would be GOP voters stayed home – hardly a shy tory effect.
It is also in a way, at odds with how many of the GOP Congressional wins were by greater margins than trump’s within their geography.
.
Let’s put to bed the notion that there was some “silent majority” who turned up and gave trump a “landslide” (it was not) victory.
The GOP gained decisive power. It was not a resounding endorsement of trump or his policies. It was a devastating defeat of clinton and the dems.
Many good things can (and should) be done with that power, BUT a “mandate” it was not.
The election was more reflective of having two candidates with historically negative approval ratings.
“The issue is self-interest. You have permission in this country to defend your own interests, and to vote accordingly, only if you’re poor, or otherwise disadvantaged. For the prosperous — according to the UK tax code, anyone who makes more than about £42k per year — voting in accordance with self-interest is unseemly, ungenerous, greedy, and mean.“ – CV’s linked article
When I look at the celebrity set, it seems this might be true, as many became very wealthy off of something that they might not think of as “legitimately hard work”.
https://www.youtube.com/watch?v=L7Oc2kEbn3Q
It seems that there might be some truth to the idea that the “easy” wealth of those folks creates an uneasiness with how they are so blessed, thus, feel the need to “help” those less fortunate.
Only, they don’t seem to want to do so merely by acts of their own, but they want others to “share the burden”, hence, their support of dem redistributive (and other “equalizing”) policies.
It is rather relatively easy to lend their name to support such, knowing that even if they lost 50% of what they have, they’d still have a very comfortable life.
Besides, it helps drive their popularity at the same time.
Who says populism isn’t good for business?
One possible, but unprovable, and, perhaps, cynical, explanation, anyway.
I don’t understand why Kentucky is consistently left off of the list of states where the Republicans won in the lower legislative house.
This year, the Republicans won a majority in the Kentucky house, and they now control both houses of the legislature and the governorship. This, for the first time since 1921.
Not only that, but when Grady Stumbo, the Democrat speaker of the house, lost his seat to a Republican, the governor when asked to comment, didn’t make nice about it. Right there in front of G_d and all hands, he said: “Good riddance.”
“But the poll response rate still goes down (I never answer polls myself, for example) and that’s hard to correct for.”
And it will continue as more and more people realize that “anonymous” polls are impossible to guarantee, combined with the simple fact that we finally have a major political party, ON CAMERA, admitting that they purposely hired personnel to go to the opposing party’s rallies and incite violence.
Let’s say a pollster (robotic or actual) calls your phone (land or cell; I actually got both). You are absolutely sure of these things:
1. They have happened on / used something that can be personally tracked back to either you or your household.
2. They have to have made a record that the number was called, and any responses, if for no other reason than to tally results and make sure they don’t call that number twice.
3. The technology exists to datamine that info; commercial companies (such as Amazon, owned by Democrat Jeff Bezos) do it routinely.
4. Again, we have clear evidence that Democrats are paying for violence…. and have made it a point that “the personal is political” by sending protesters to people’s homes in the past.
Finally, for those who want to scoff by citing rally attendance: With the proliferation of street cameras and facial recognition software (controlled by Democrat union bureaucrats), it will be only a matter of time before people start recognizing that the anonymity of crowds is a thing of the past.
Trump’s big mouth regarding the polls, or anything else, notwithstanding, the fact remains that more people turned out for him than projected.
Hillary failed to light up the expected demographic groups, and was about, iirc, 3 million votes shy of Obama’s last effort. This leads to an interesting speculation; were some of the people polled who said they’d vote for Hillary afraid to say they planned not to vote at all?
Richard Aubrey:
Of course they are always trying to tweak the turnout models and get it right. Trouble is that this is inherently difficult to do. They are probably all making similar assumptions when they get it wrong, not because they’re in cahoots but because these are logical (although incorrect, it turns out) assumptions to make. This year was exceedingly different, so it makes perfect sense that they had trouble predicting turnout and that they may have all erred in a similar direction (except the LA Times, which erred in the opposite direction).
I wrote about this problem here and especially this.
I should make up another maxim about all errors going in a similar direction. I could use the global warming models for an example.
Something like once is happenstance, twice is coincidence and thrice is enemy action.
Is there any reason to think pollsters are less bubbled than the MSM?
I suppose that could cover incompetence and collegiality equally.
Richard Aubrey:
But if you did that, you would be wrong.
Not only did all the polls not go in the same direction (LA Times, which is NOT a conservative group, consistently predicted a Trump victory by too high a margin, and quite a few polls sometimes showed a very close race, which turned out to be correct), but in 2012 the polls were fairly consistently wrong in the opposite direction, pro-Romney (see this).
Perhaps you think that in 2012 the pollsters were cooking the polls for the GOP, and then in 2016 they decided to do the same for Clinton?
It makes no sense. And it makes no sense for the polls to have favored either of them purposely, because (as I’ve said many many times, and I’m getting tired of repeating myself) each false prediction could easily have had the opposite effect from the one supposedly intended. It could discourage the side that was behind OR it could charge them up, and it could discourage the side that was ahead because that side got too cocky and felt safe or it could encourage them because they felt good.
Not to mention the fact that pollsters don’t like to look like fools and have their validity and competence questioned.
Let’s say for the sake of argument that AGW scientists are lying purposely (I’m not at all sure they are, by the way, but let’s say they are). That’s quite different than polling. First of all, if they are believed, it leads to a more likely passage of certain policies. So the chain of causation is much clearer than it would by with lying polls, which have a much more unpredictable result. In addition, if pollsters lie about polls for an election, they are found out to have been incorrect or lying soon enough, because an election is a single event with a definite and certain outcome, an event that takes place on a single day. AGW is not. In other words, the errors or the lies of pollsters are discovered quite soon, and it discredits them. AGW scientists would have more incentive to lie because it is much harder to discredit predictions that are so changeable, hard to measure, and take place over a longer period of time (actually, an almost indeterminate period of time).
So your analogy is a poor one, as well.
Neo. What happens to a pollster who is wrong? The fallout is considerably different than what happens to , for example, a restaurant responsible for a blast of food poisoning.
What happens to a pollster is…nada.
That being the case, the indefinite time frame for one’s errors to bite is the same as the AGW guys.
Whether the error facilitates the goal or, as you suggest is possible, reverses it is relevant. But, as my sainted mother used to say, “It’s thought that counts, dear.” And it’s the thought that wins the next contract.
Perhaps next time we’ll see half right and half wrong.
Richard Aubrey:
No, it’s not nada. But it depends on the type of error, the degree of the error, and the consistency of the error. People also don’t understand that polling is inherently difficult to get right.
If one pollster keeps getting it wrong when all the other pollsters gets it right, that pollster isn’t going to last long or if they last for a while they don’t have a good reputation (their greater money-making activities—internal polling and market research for businesses—depend on maintaining their reputation for accuracy).
This was written shortly after the 2016 election:
So we have 2 pollsters that come out smelling like roses, and their reputation is enhanced and they will be used more next time. If a pollster is consistently wrong—and particularly if that wrong pollster is out of line with the others—that pollster will be doing less business. The are also organizations that rank pollsters for accuracy over time, so among people who hire pollsters, the pollsters’ reputations are known.
I have yet to understand why everybody seems to think the polls were so very, outlandishly wrong this year, because they were not (except at the state level in a couple of states, particularly Wisconsin; but state polls are known to be much more difficult to do in terms of sampling, and are often somewhat outdated by the time the election rolls around).
Why do I say the national polls were not so far off? Take a look at what the polls were actually saying the day before the election, rather than what you think you remember about what they were saying.
Remember also, when you look at this, that Clinton actually (so far, anyway) has won the popular vote by 1.25%:
So, the average of polls was only off by 1.25%. That’s really pretty close. And the article goes on to state some of those factors that are not reflected in the polls that affect the final outcome—such as early voters, turnout, margin of error, and vote distribution and the Electoral College (this turned out to be a very large factor). The article also mentions that there is an unusual amount of uncertainty for this particular election. It doesn’t mention the undecideds, but we have learned since the election that they broke disproportionately for Trump.
So I actually don’t see that the polls were especially inaccurate. Polls of course are never totally accurate, they are merely indications based on a whole bunch of assumptions that can be wrong.
Take a look at what Nate Silver was saying the day before the election:
Predicting human behaviour is always going to be a challenge. In terms of creating a poll, we probably can’t do much better. We’ve seen a decline in polling participation, but it’s inconclusive whether that has affected their accuracy. The place where there’s still a high marginal return is in interpreting polls: both in proper weighting of the results and in explaining what they mean.
Weighting a poll has got to become more of a science. It’s (apparently) fairly easy to get answers about voter preferences, but fairly difficult to get answers about voter enthusiasm. For the past several cycles, the debate has been, “D+4? D+8? Oh, it turned out to be D+6.” That’s not scientific. Where’s the innovation going to come from? I don’t know. For the time being, though, they’re using old trends and anecdotes.
And I guarantee you the industry would like to solve this puzzle. Campaigns care about turnout just as much as they care about preference.
Neo,
It would be more acceptable if the error in the polls were scattered to both sides. However, if the polls, save one, are all off in the same direction, there is a strong indication of bias in the polling models. This time around it was the LA Times poll; in 2012 it was the gang at 538. Furthermore, the accuracy vs. the results were only validated in the polls conducted the final week before the election. Prior to that, only the LA Times was coming up on the other side of the results – and spent most of the election season being tagged and discarded as an outlier.
I seriously doubt that any bias is/was deliberate. As you point out, the polling companies need to produce a reliable product. I really wish I could find that article I mentioned on how pollsters are at their wits end because a persistent, but inexplicable, bias keeps showing up; that they don’t want it there, but cannot identify it sufficiently to correct for it: there is no consistent theory that explains it.
“Something like once is happenstance, twice is coincidence and thrice is enemy action.
…
What happens to a pollster is…nada.” – Richard A
Your argument, in essence, refutes the existence of market forces.
It posits that pollsters are essentially being paid to provide propaganda.
To believe that would be to believe that there would then be no room for any enterprising pollster to provide “accurate” polls.
To believe that would be to believe that nobody cares about “accurate” polls and wouldn’t pay for them, nor want to promote them as an alternative to the “propaganda” ones.
To believe that would be to believe that “conservative” media would be happy with the status quo of the “propaganda” polls having their way.
To believe that all is to believe in conspiracy theories in all their folly, and that the market cannot address such vacuums of quality.
.
Thus, we are back to the standard of “valid” polls are the ones that fit my view, and “invalid” ones are the ones that I disagree with – a rather poor standard from which to judge the quality of polls from.
The end result of that conspiracy theory, in the extreme, is that one has to believe that no polls can be trusted, as they all have to be biased in some way based on the leanings of the pollster.
.
Could a better understanding of “margin of error” help in this case? IDK. If one is willing to throw out or deny marketplace operating principles, is there any room left?
Big Maq Says:
November 21st, 2016 at 12:27 pm
It posits that pollsters are essentially being paid to provide propaganda.
To believe that would be to believe that there would then be no room for any enterprising pollster to provide “accurate” polls.
Follow the money.
Who is paying for public polls? You? Me?
Media companies are paying for the polling.
They have no real interest in generating accurate polling. Whether a candidate is up or down, a story can be generated that validates the view of whatever group the media is interested in selling to.
Yep.
I developed a tic in my tendency to comment on threads – i.e., deleting my comment entirely upon reflection instead of hitting the Submit button – several years back. For pretty much the same basket of reasons you suggest.
I am no longer able to be sanguine about the considerations involved.
Christopher B, and just about everyone else:
I have a new post up that’s relevant to this subject. Take a look.
Plus, if you want to know who pays for political polls: they are not the moneymakers for pollsters. The moneymakers are two things, mainly: internal polling for candidates small and large, and in particular market research for businesses. The pollsters have a strong motivation to be accurate in national polls, though, because that is how they build their reputations. Pollsters are rated for accuracy by poll-rating companies/organizations, and good reputations attract their lucrative clientele.
@Christofer B – you failed to address where the “conservative” media is in all of this.
Not like they don’t exist in this world and don’t have the means to finance an alternative.
Why the h*ll would they sit on the sidelines?
Your position doesn’t even hold minimal logic, you so want to believe everything and everyone you disagree with is involved in one grand conspiracy.
Did you happen to drive a taxi, previously 😉
https://www.youtube.com/watch?v=snqFHjpDpes