Jump to content

Talk:Berkson's paradox

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia


Wikipedia is not for statisticians

[edit]

From the article:

The result is that two independent events become conditionally dependent (negatively dependent) given that at least one of them occurs. Symbolically:
if 0 < P(A) < 1 and 0 < P(B) < 1,
and P(A|B) = P(A), i.e. they are independent,
then P(A|B,C) < P(A|C) where C = AB (i.e. A or B).
In words, given two independent events, if you only consider outcomes where at least one occurs, then they become negatively dependent.

You don't define "event", "conditionally dependent", P(X) as the function "Probability of landing in X", the "|" as "given" (and even then, you would have to explain what "given" means), the comma in P(A|B,C), and the "union" symbol. --68.161.181.109 (talk) 10:22, 15 January 2009 (UTC)[reply]

There is a link in the first sentence to conditional probability. Would you expect an article about Richard Nixon to define what is meant by "President of USA" or even by "USA". After all Wikipedia isn't just for Americans just as it isn't just for statisticians, mathematicians or probabilists. Melcombe (talk) 11:40, 15 January 2009 (UTC)[reply]
I would expect it to wikilink "USA". I would expect it to wikilink or explain any uncommon terms such as "POTUSA" and "Air Force One". At the very least, I would expect some clear indication to where prerequisite knowledge would be, rather than just pointing out after the fact that the article mentioned that this was part of some field of math that I had no business learning about. Besides, if you had checked, nowhere in the conditional probability article was there a mention of the terms "negative dependency" or "conditional dependency". --68.161.181.109 (talk) 15:52, 22 January 2009 (UTC)[reply]

Please restate the definition in plain English.

[edit]

I agree entirely with User:68.161.181.109. The article on “conditional independence” is even harder to understand than this one, as is much of the article on “conditional probability” (significantly harder, in fact. At least I could get the gist of this article, even if I could not follow the mathematical notation). The mathematical notation is not common knowledge, whereas the meanings of “USA” or “president” are common knowledge in English-speaking countries. One does not need to be from the UK to understand what “David Cameron is the prime minister of the U.K.” means. At the very least, the section entitled “statement” should be accompanied by a restatement of the definition in plain English. I can translate the first two parts of the definition as:

Informally,

  1. If the probability of event ‘A’ is greater than 0 but less than 1, that is, it might or might not happen, and if the probability of event ‘B’ is also greater than 0 but less than 1 (where 0 means that the event is guaranteed not to happen, i.e. a 0% chance, and 1 means that the event will definitely happen, i.e. a 100% chance),
  2. and if events ‘A’ and ‘B’ are independent, that is, the probability of event ‘A’ occurring given that event ‘B’ actually occurs is the same as the probability of ‘A’ occurring regardless of whether ‘B’ actually occurs, then...

but the left side of the third part of the definition (the conclusion) gets me lost. And it is not because the conclusion is counterintuitive, but because I do not really understand what the notation is saying. 71.178.164.165 (talk) 19:44, 16 January 2014 (UTC)[reply]

Selection bias

[edit]

This is an example of selection bias. One could simplify this page by linking to that. — Preceding unsigned comment added by Bakerstmd (talkcontribs) 21:01, 17 December 2013 (UTC)[reply]

Doesn't address the actual fallacy Berkson described

[edit]

The article only states (in a convoluted way) the rather obvious fact that P(A|B) < P(A|A∪B). It doesn't address the fallacy as described by Berkson.

Despite giving the paper of Berkson as (only) reference, it writes:
For example, a hospital patient without diabetes is more likely to have cholecystitis, since the patient must have had some non-diabetes reason to enter the hospital in the first place. (also mentioned in Sampling bias, which states: "This can result in a spurious negative correlation between diseases".)
In fact, his (hypothetical) example showed a spurious positive correlation between diabetes and cholecystitis when compared to a control group of patients with refractive errors. Prevalence 15:16, 10 May 2016 (UTC)[reply]
Rephrased, For example, a person in a casino who doesn't gamble is more likely to be a prostitute or pickpocket, since they must have had some non-gambling reason to enter the casino in the first place. This is purely statistics, equally applicable in many fields of study. The article doesn't (as suggested elsewhere) belong in WikiProject Medicine, nor WikiProject Sexuality, nor WikiProject Criminology for that matter. LeadSongDog come howl! 21:56, 10 May 2016 (UTC)[reply]
Yes, I can think of many examples of P(A|B) < P(A|A∪B). I am saying that is not what Berkson's paper was about. He compared the prevalence of disease A in people with disease B to the prevalence of disease A in people with disease C. Or in your example: the percentage of hookers among the blackjack players compared to the percentage of hookers among the roulette players. Prevalence 01:00, 11 May 2016 (UTC)[reply]
The problem is: what is the correct definition of "Berkson's fallacy"? Is it the original example given by Berkson, and repeated in sources about epidemeology, or is it "Two independent events become conditionally dependent if it is given that at least one of them occurs or does not occur."? That definition is from "A Dictionary of Statistics (Oxford)", written by Graham Upton, head of the Mathematical department at University of Essex. The definition given in "Medical Statistics from A to Z: A Guide for Clinicians and Medical Students" is: "the existence of artefactual associations between two medical conditions, or between a disease and a risk factor, arising from the interplay of differential admission rates with respect to the suspected causal factor." "The Cambridge Dictionary of Statistics" gives the same definition, both are are written by Brian S Everitt, former head of the Biostatistics and Computing Department at the University of London.
If it has one meaning in statistics and another meaning in epidemiology, it seems reasonable to include both of them. Prevalence 01:39, 11 May 2016 (UTC)[reply]
The math is the math. Its applications include epidemiology, sure, but that boils down to what labels one writes on the Venn diagram. It might call for a section on applications, but it is certainly not definitive of the "paradox". LeadSongDog come howl! 16:36, 11 May 2016 (UTC)[reply]
Sure, and ex is the same as cos(x), it simply boils down to what coefficients one uses in the Taylor series...
I have not found any RS for the content of the article as it currently stands. "A Dictionary of Statistics" doesn't qualify as RS, it's a tertiary source, and "The Oxford Dictionary of Statistical Terms", by the same publishers, claims that Berkson's fallacy is just another name for the Yule-Simpson paradox. At the moment, nothing in the article is supported by the reference given. Prevalence 21:31, 11 May 2016 (UTC)[reply]

Is there a barnstar for worse wikipedia page?

[edit]

there should be 208.53.114.212 (talk) 22:06, 7 July 2017 (UTC)[reply]