Tip jar

If you like CaB and wish to support it, you can use PayPal or KoFi. Thank you, and I hope you continue to enjoy the site - Neil.

Buy Me a Coffee at ko-fi.com

Support CaB

Recent

Welcome to Cook'd and Bomb'd. Please login or sign up.

March 28, 2024, 08:11:45 PM

Login with username, password and session length

Election Polling: A How-to Guide

Started by New folder, November 24, 2019, 05:45:28 PM

Previous topic - Next topic

New folder



I know that there are a lot of strong feelings about the current polling results, and even some people casting *gasp* doubt on their validity. So, after some research, I was able to compile a quick handy guide for any potential pollster, in order to get the most accurate and honest results -- and more, importantly, predict the election correctly! (see above)


So let's assume you've got a nice representative sample of elderly people you called over the phone. Here's how you interpret these results:


STEP 1: Answer the question for anybody undecided



If somebody doesn't know who they will vote for, it is very important that you answer the question on their behalf. This ICM/Guardian polling guide tells you exactly how to extract an answer from these annoyingly uncertain people. Simply take 75% of the Don't Knows, and assume they didn't change their mind since the last election. Why 75%, you ask? Because we decided it. Why didn't they change their minds? Because we decided they didn't.

This process allowed ICM to precisely predict that the Conservatives would win the 2017 election by 12%. Which of course, they did.


STEP 2: Add votes to the Conservatives



In the next phase of the adjustment procedure, you add votes to the Conservatives. When people didn't vote in the previous election, it becomes difficult to answer the question for them. So, you simply must assume that more of them are Conservatives. Problem solved.


STEP 3: Remove anybody not engaged in politics



Observe the following criteria when deciding which respondents to remove from your survey:



If somebody does not meet these criteria, be sure to consider their opinion to be less valuable than others. This technique is even more effective when historical data is used, because the political engagement of groups of people is completely static and will never change.


STEP 4: Only consider those most certain to vote



Uncertain people are bad. We must eliminate them from our survey, in the hope that they will just disappear altogether. If somebody is not 10 out of 10 certain they will vote, why even bother listening to them? Even an 8 out of 10 person can be considered to have basically no voting rights (as they should). This is a crucial trick used by all reputable pollsters to get the results as true to reality as possible.


CASE STUDIES

Now let's see these techniques put into action! Look at this Opinium Poll, taken the day before the 2017 election:



As you can see, they employ all of the tricks we have outlined above (although personally, I think that Opinium are too generous, allowing those ghastly 90% certain voters in).
This can be compared favourably to the raw data they obtained:



Clearly, given the 2017 result a day later, this is completely wrong. Luckily they made all these adjustments, so that people's actual opinions were more clearly represented.

Onto YouGov, for another example of these techniques leading to a more accurate result. First, look at the raw polling data from the last election (Don't Knows included):



A 2% gap between Labour and the Conservatives!?!? Imagine if that were to transpire a day later on election night! Well, people clearly just have the wrong opinion. Luckily, YouGov was able to put things right with its infinite wisdom:



Ah, much better. A 7 point lead for the Conservatives -- this reflects the consensus of the blue-tick commentariat, so it must be more correct than people's actual, unmanipulated opinions.


HONOURABLE MENTIONS

Let's take a trip back in time through the Polling Hall of Fame. Since it's 3 weeks from election day, we'll look at polls taken at a similar point in the campaign in 2010.



Personally, I loved Nick Clegg's term as Prime Minister. The Lib Dem majority that the polls predicted 3 weeks before the election was almost spookily prophetic, wasn't it? I heard that they didn't even need to watch the results on election night, so clairvoyant they were with their predictions.


CONCLUSIONS

Heavily statistically manipulated data based on spurious assumptions is an accurate reflection of future election results. When looking at current polls for the 2019 election, please listen to the weighted headline data that has been 100% correct for the past 3 general elections. Polling companies are some of the most honest institutions and should be trusted above all else. Any attempt to imply that they have fabricated their data is a conspiracy theory. Please ignore any people who claim this, and continue to be demoralised.


REFERENCES

https://www.icmunlimited.com/wp-content/uploads/2017/05/2017_guardian_campaign_poll6.pdf
https://d25d2506sfb94s.cloudfront.net/cumulus_uploads/document/d8zsb99eyd/TimesResults_FINAL%20CALL_GB_June2017_W.pdf
https://en.wikipedia.org/wiki/Opinion_polling_for_the_2010_United_Kingdom_general_election
https://www.drg.global/wp-content/uploads/2018/01/W10470w7tablesforpublication.pdf
https://dera.ioe.ac.uk/29007/1/CBP-7501.pdf
https://www.icmunlimited.com/wp-content/uploads/2017/06/2017_guardian_prediction_PRELIM_1500.pdf
https://www.ipsos.com/sites/default/files/2017-06/pm-election-2017-final-tables.pdf

idunnosomename

Why not simply be patient and wait till the general election

greencalx

Thanks for this. Do we know what the unweighted polls look like?

New folder

Curiously, the raw data of most polls is a bit harder to find this time around. Although there was a Kantar poll last week which gave Con 32% and Lab 31%, before adjustments (https://www.dropbox.com/s/0kcvu5ahy5am9t9/kantar-11-nov.xlsx?dl=0).

But regardless, the point of the post was to illustrate how any polling data fundamentally cannot be trusted. It also appears that "curbstoning" (fabricating the results of a survey to save money or time) is a known problem in the polling world -- see https://content.iospress.com/articles/statistical-journal-of-the-iaos/sji917

Another thing I've noticed lately, is that after an election (most significantly after the 2017 General and 2019 European ones), the polls somehow magically align with the real results, despite being very poor predictors before the vote. This type of pattern does make the polls appear to be reflecting, rather than predicting. They seem to follow what they believe is the current political narrative, and are probably adjusted, weighted or even fabricated to suit this. I mean, look at the results around this year's EU elections:



Obviously, the Lib Dems and the Brexit Party won the EU elections because nobody else gives a flying fuck about EU elections. But for some reason, this victory bled through to public opinion so strongly and so immediately, that the Lib Dems were beating both the Tories and Labour the very next week, even though they claimed Labour had a 10% lead two days before the election. Next, the Brexit party took the lead, because they also performed well in the EU elections. It's hard to imagine that such a seismic shift in mainstream opinion occurred as a result of some MEPs being elected. This makes it very clear that the polls are shutting the barn door after the horse is already a glue stick.

New folder

Quote from: idunnosomename on November 24, 2019, 06:00:01 PM
Why not simply be patient and wait till the general election

I fucking wish. I hate polls. But they are here, and lots of people believe them, and become discouraged or demoralised by them. Which is likely their primary purpose. So I thought it would be useful to give direct evidence of them being total horseshit, so that maybe people will find it easier to ignore them.

Jerzy Bondov

Thanks for this very interesting thread!

Absorb the anus burn