explain xkcd - User contributions [en]

Talk:2440: Epistemic Uncertainty

2021-03-23T06:46:37Z

172.69.135.234:

I definitely thought "adulterer" referred to someone who commits adultery, as in cheating on one's spouse. I thought it was a secondary joke, introducing another person referred to as "[name] the [undesirable action]er".
[[Special:Contributions/172.69.170.56|172.69.170.56]] 02:03, 23 March 2021 (UTC)
:"Adulterer" and "adulterator" have different definitions - to "adulterate" a substance is to mix it with an unintended additive. [[Special:Contributions/172.69.135.234|172.69.135.234]] 06:46, 23 March 2021 (UTC)
Is the "George" referred to here possibly the name of black hat?
:I doubt it. The hat silhouette is not the same pork pie hat as Black Hat [[Special:Contributions/172.68.86.20|172.68.86.20]] 04:34, 23 March 2021 (UTC)

The name "Evangeline" could be a reference to how "Eve" is usually the name of a hypothetical hacker used when teaching people about computer science. You know, that whole "Alice sends Bob a private message but Eve wants to read it" thing. [[Special:Contributions/108.162.245.122|108.162.245.122]] 05:22, 23 March 2021 (UTC)

2440: Epistemic Uncertainty

2021-03-23T06:44:17Z

172.69.135.234: /* Explanation */ Long explanation of confidence intervals, link between epistemiology and knowledge in this study

{{comic
| number = 2440
| date = March 22, 2021
| title = Epistemic Uncertainty
| image = epistemic_uncertainty.png
| titletext = Luckily, unlike in our previous study, we have no reason to believe Evangeline the Adulterator gained access to our stored doses.
}}

==Explanation==
{{incomplete|Created by GEORGE THE DATA TAMPERER. Please mention here why this explanation isn't complete. Do NOT tamper this tag too soon.}}

This comic is a comparison of two different "research studies", one which shows "regular uncertainty" and one which shows "epistemic uncertainty." In both panels, the core data is the same - the drug in question is 74% effective - but the uncertainty qualities are different. The first is straightforward; the confidence interval (the error bars on the chart) is from 63 to 81%. The second panel includes the additional wrinkle of "George the Data Tamperer, whose whims are unpredictable."

In statistics, a {{w|confidence interval}} is an estimate which provides a range of values, based on the statistical probability that the data collected represents a certain result. It is a reflection on the uncertainty imposed by the limits of study sample sizes. Since no study will ever have an infinite data set, it is possible for a small sample to give a skewed result, but the small skews are more probable than large ones. For example, if a drug was 80% effective it would be possible for a study with a sample size of 100 to randomly end up with 74 positive and 36 negative results. If the drug was 99% effective it would still be possible to randomly end up with the same data, but it would be highly unlikely. This gives us a spread of "likely" results, with results outside a certain interval being considered too unlikely to be realistic.

{{w|Epistemology}} – unlike {{w|epidemiology}} – is the branch of philosophy related to knowledge. It seems that the "epistemic uncertainty" data has a 25% chance of data {{w|tampering}}, by an individual called "George the Data Tamperer". In contrast to the previous study, where the data is known but its reflection of the general case is uncertain to an extent, in this study even the knowledge of "whether any single data point is correct" is uncertain. Thus, their data has a 25% chance of being incorrect with no possible statement about <i>how</i> incorrect it may be.

The title text mentions an individual called "Evangeline the Adulterator", who [https://www.dictionary.com/browse/adulterate adulterates] their drug doses. If this happened, the researchers would not even be sure the patients received the dosages as prescribed, and the study methodology itself would be in doubt.

==Transcript==
{{incomplete transcript|Do NOT delete this tag too soon.}}

In two near-identical panels, labeled "uncertainty" and "epistemic uncertainty," Megan stands in front of a data presentation.
:[Panel titled 'Regular Uncertainty'. Meagan standing in front of a presentation of data with error bars.]
:Meagan: Our study found the drug was 74% effective with a confidence interval from 63% to 81%

:[Panel titled 'Epistemic Uncertainty'. Meagan standing in front of a presentation of data with a silhouette labelled with a question mark. ]
:Meagan: Our study found the drug was 74% effective. However, there is a 1 in 4 chance that our study was modified by George the Data Tamperer, whose whims are unpredictable.

{{comic discussion}}
[[Category:Comics featuring Megan]]
[[Category:Statistics]]
[[Category:Research Papers]]
[[Category:Biology]]

2370: Prediction

2020-10-10T14:13:45Z

172.69.135.234: /* Explanation */

{{comic
| number = 2370
| date = October 9, 2020
| title = Prediction
| image = prediction.png
| titletext = You'd think it'd be easy to just bet money against these people, but you have to consider the probability of them paying up.
}}

==Explanation==
{{incomplete|There is definitely not a 50/50 chance this was created by a BOT. Please mention here why this explanation isn't complete. Do NOT delete this tag too soon.}}

This comic is about misunderstanding {{w|probability}}. Sometimes people will incorrectly assume that if one event is likelier than another to occur, then that event WILL occur, or that if one names two or more outcomes, they are equally likely to occur when in fact they might have different probabilities.

Saying that an event is more likely to happen than not to happen is not the same as saying that the event is definitely going to happen. At the same time, even if the event not happening is possible, it's not 50/50 odds that the event will happen. People have difficulties understanding statements like "event A has a 70% probability to happen" and internally understanding it to be one of the two misconceptions above.

Some don't like probability statements because they are not definite and therefore cannot be proven wrong. For example, if a probability statement says "event A has a 1% probability to happen" and event A actually happens, that does not prove the statement wrong, because the statement admits of the possibility of event A happening.

For example, FiveThirtyEight famously gave Trump a higher odds of winning the 2016 U.S. presidential election than most other models did just before the election, but still not more likely ([https://projects.fivethirtyeight.com/2016-election-forecast/ 28.6%]) than his opponent. However, many readers at the time interpreted that as "Trump is definitely going to lose", and after he won that election, blasted FiveThirtyEight for getting its prediction "wrong". However, that interpretation is mistaken. 28.6% means Trump had a real chance at winning, between throwing a coin twice and both time landing on heads (1/4 or 25%) and throwing a normal 6-sided die and getting a 1 or 2 (1/3 33.333...%), both of which events are intuitively possible.

The correct interpretation of a probability statement like "event A has a 70% probability to happen" is that in the long run, of those events you give 70% probabilities to, about 70% of them end up happening. If, for example, 99% of those events ended up happening, the 70% probabilities you gave those events may likely be wrong (you should've given probabilities closer to 1), even though you "called" almost all events correctly (in the sense that 70% means the events are more likely to happen than not to happen, and almost all of them happened). Looking back at your predictions and seeing if the results are what you should expect is called {{w|Calibration (statistics)|calibration}} ([https://projects.fivethirtyeight.com/checking-our-work/ example]).

In the last panel, it is shown that [[Cueball]] anticipated this lack of understanding, so he plays pre-recorded audio of his prediction for the conversation.

The title text says that these people are gullible enough to the point that they would accept a disadvantageous bet. However, it also says that they might not actually go through with paying the bet if they lose.

At the time of writing, the 2020 United States presidential and congressional elections were less than a month away. This is a time when polls showing one or the other candidate leading are common, and may be misinterpreted to mean that the candidate is certain to win. Additionally, after the 2016 election saw Donald Trump, the trailing candidate in the polls, winning, many also interpreted this to mean that the polls were useless and/or wrong, or even go beyond this and take an adverse poll prediction as a perversely authoritative indication that the exact opposite result (which they would favour) is now a certainty. Cueball has previously shown an interest in U.S. election polling, for example in [[500: Election]].

==Transcript==
:[White Hat and Cueball standing next to each other. Cueball has his palm out.]
:Cueball: Event A is more likely than Event B.

:[White Hat touches chin thoughtfully]
:White Hat: So you're saying that Event A will happen.
:Cueball: No, Event B could also happen.

:[A frameless panel]
:White Hat: So you're saying it's 50/50.
:Cueball: No, it's definitely not 50/50.

:[Cueball produces a phone]
:White Hat: Sounds like you have no idea what will happen.
:Cueball: And yet I knew exactly how this conversation would go. Here, listen:
:<nowiki>*Click*</nowiki>
:Phone: ''Then you'll say, "So it's 50/50"''

{{comic discussion}}

[[Category:Comics featuring Cueball]]
[[Category:Comics featuring White Hat]]
[[Category:Statistics]]