Talk:Observational interpretation fallacy

Update with Questionable Cause

User:Paradoctor (again correctly) pointed out that I had put the question in the wrong forum, so I'll move it here.

Looking through the source material on the Observational interpretation fallacy page it is clear that this is either a form of Questionable Cause fallacy or special case (when the fallacy is conducted via an observational study.

I'd like to propose making this edit to the Observational interpretation page, but first wanted to get both of your inputs.

In observational research, distinguishing between association and causation is critical for avoiding spurious causation and in drawing accurate conclusions. Association is a statistical relationship between two variables, whereas causation implies that one variable directly affects the other.

I think this change is warranted as the next section of the article (along with the sources) go on to describe confounding, which is simply the mechanism for the questionable cause fallacy (per the parent article). The first source on the page for example (Hammer et al) points out all the sub-categories of questionable cause as variants of the Observational interpretation fallacy and even titles the article the informal name for questionable cause.

Squatch347 (talk) 18:19, 21 January 2025 (UTC)

You propose the following change:

−

In medical research, distinguishing between association and causation is critical for drawing accurate conclusions and making informed decisions. Association is a statistical relationship between two variables, whereas causation implies that one variable directly affects the other.

+

In observational research, distinguishing between association and causation is critical for avoiding [[Questionable cause|spurious causation]] and in drawing accurate conclusions. Association is a statistical relationship between two variables, whereas causation implies that one variable directly affects the other.

Empirical research is better than "observational research".

Classifying the fallacy as a form of questionable cause needs a citation. The wording could be improved, too. Paradoctor (talk) 16:17, 22 January 2025 (UTC)

I would prefer it as well, but the current citations are pretty universal in their usage of the terminology of "observational research" so I wanted to be cautious on using what I assumed was the broader, correct term. Happy to go with empirical if we are all confident.

For citation I would propose we use https://pubmed.ncbi.nlm.nih.gov/39733264/ and https://pmc.ncbi.nlm.nih.gov/articles/PMC2780010/ The first is already in the article, the second I had thought was in there, but apparently wasn't. Both describe this fallacy as a collection of cognitive and design errors centered around confounding. Per the wiki article on confounding (which links to this article), confounders are the quantitative explanation of why correlation does not imply causation, which links to the questionable cause article.

Squatch347 (talk) 19:25, 22 January 2025 (UTC)

If the sources say "observational" then we follow the sources. Please don't link it, I just learned that observational research also means field research, as opposed to lab work. Sloppy editing in the sources. 🤷

For "questionable cause", we need a source explicitly stating the connection. The chain of reasoning you presented is at least WP:SYN, at worst using unreliable sources, so we can't use that. WP:BURDEN: A source "directly supports" a given piece of material if the information is present explicitly in the source, so that using this source to support the material is not a violation of Wikipedia:No original research.

The second source does not contain "questionable" at all. The first was paywalled, maybe you can provide a relevant quote from it? Paradoctor (talk) 20:20, 22 January 2025 (UTC)

I don't think this qualifies as synth because we are referencing terms for the same thing not merging two separate concepts. The problem with this is that we are using terms from largely different fields to talk about the same concept. Philosophers and statisticians wouldn't use this term because they already have language for the error being described.

It would be like having a source say "The US provided M777 howitzers to Ukraine" and the wiki article saying the word artillery or cannon. Those are two different ways of saying the same thing from two different fields. The sources both refer to confounding, which is the questionable cause fallacy. The sources are saying that this is the fallacy construction of confounding error. IE the two are equivalent and confounding error is part of the questionable cause fallacy.

Correlation does not imply causation § Third factor C (the common-causal variable) causes both A and B

Confounding

If we need to establish that connection beyond the parent articles for those topics we can use the sources from those parent articles that establish that relationship as well, it just seems a bit like over sourcing given we are linking to the more indepth explanation of the topic already.

Squatch347 (talk) 20:50, 22 January 2025 (UTC)

The point is that any claim must be sourced where it is made. Thems the rules. Paradoctor (talk) 21:05, 22 January 2025 (UTC)

Ok, I did a bit more digging on this to see if I could get a source for the statement:

"In observational research, distinguishing between association and causation is critical for avoiding spurious causation and in drawing accurate conclusions."

It still seems relatively clear to me that this is the same thing given the two definitions:

"The differentiation between association and causation is...complicated by cognitive biases that erroneously interpret coincidental observational data as indicative of causality" and "Concluding that one thing caused another, simply because they are regularly associated." https://www.logicallyfallacious.com/logicalfallacies/Questionable-Cause

Or, from our source on the questionable cause article: "This fallacy is committed when it is concluded that one thing causes another simply because they are regularly associated. More formally, this fallacy is committed when it is concluded that A is the cause of B simply because A and B are regularly connected. Further, the causal conclusion is drawn without considering the possibility that a third factor might be the cause of both A and B." https://web.archive.org/web/20090522103015/http://www.opifexphoenix.com/reasoning/fallacies/ignorecc.htm

But, digging a bit more I found the following sources discussing the issue of incorrect inference of causality and ignoring a third cause in observational research:

• "The major challenge using observational data for causal inference is confounding variables: variables affecting both a causal variable and response of interest. Unmeasured confounders—known or unknown—lead to statistical bias, creating spurious correlations and masking true causal relationships."

https://onlinelibrary.wiley.com/doi/full/10.1111/ele.70023

• "This type of variable is known as a confounding variable and it can confound the results of a study and make it appear that there exists some type of cause-and-effect relationship between two variables that doesn’t actually exist."

https://www.statology.org/confounding-variable/

• "For example, an observational study might show an association (correlation) between a predictor variable (independent variable) and an outcome (dependent variable); however, the results may not represent cause and effect (Hulley, 2013). Causal inference implies an intervention, e.g., treatment or behavior was the ‘cause’ of the effect (or outcome). Understanding causal inferences between predictor(s) and outcome(s) can provide insights to understanding the etiology of a disease, identify methods to prevent or reduce disease (or occurrence) and potentially initiate the development of treatments (Hulley, 2013). For example, are eating carrots associated with improved eye health?

It is important to note that some associations found in an observational study do not represent cause and effect. However, there are well-recognized explanations for associations between a predictor and outcome in such studies (Hulley, 2013). These occur by chance (random error), bias (systematic error), and confounding variables." https://pmc.ncbi.nlm.nih.gov/articles/PMC10036082/

• "Confounding is often referred to as a “mixing of effects” wherein the effects of the exposure under study on a given outcome are mixed in with the effects of an additional factor (or set of factors) resulting in a distortion of the true relationship. In a clinical trial, this can happen when the distribution of a known prognostic factor differs between groups being compared.

Confounding factors may mask an actual association or, more commonly, falsely demonstrate an apparent association between the treatment and outcome when no real association between them exists. The existence of confounding variables in studies make it difficult to establish a clear causal link between treatment and outcome unless appropriate methods are used to adjust for the effect of the confounders (more on this below). Confounding variables are those that may compete with the exposure of interest (eg, treatment) in explaining the outcome of a study. The amount of association “above and beyond” that which can be explained by confounding factors provides a more appropriate estimate of the true association which is due to the exposure." https://pmc.ncbi.nlm.nih.gov/articles/PMC3503514/

I think any of these would be fine as a source, though I suggest either the logicallyfallacious source and the first two new links are probably the best to make this connection. Squatch347 (talk) 15:46, 23 January 2025 (UTC)

Re "logicallyfallacious": Hell no. This is at best a WP:SPS lacking the requisite credentials, and at worst refspam.

But the real problem is that none of these even mention either "questionable cause", "spurious causation", or "accurate conclusions". Without that, none of them supports the proposed formulation. Paradoctor (talk) 19:02, 23 January 2025 (UTC)

Hmm, we have sources saying the following:

• creating spurious correlations and masking true causal relationships.
• make it appear that there exists some type of cause-and-effect relationship between two variables that doesn’t actually exist.
• For example, an observational study might show an association (correlation) between a predictor variable (independent variable) and an outcome (dependent variable); however, the results may not represent cause and effect
• falsely demonstrate an apparent association between the treatment and outcome when no real association between them exists.

These are all exactly what is described in the wiki article. To wit;

• The questionable cause—also known as causal fallacy, false cause, or non causa pro causa ("non-cause for cause" in Latin)—is a category of informal fallacies in which the cause or causes is/are incorrectly identified. In other words, it is a fallacy of reaching a conclusion that one thing caused another, simply because they are regularly associated.

If our source said, "Steve was convicted in California of the unlawful killing of a human being with malice aforethought" we wouldn't balk at calling it murder. It is literally what that word means.

What's more we do have (including the definitional source for this article) direct references to this fallacy. The title of the first source in this article is directly referencing a name of this fallacy (Correlation implies causation v "Association Does Not Mean Causation...")

Squatch347 (talk) 20:22, 23 January 2025 (UTC)

"Spurious correlation" and "spurious causation" are different things.

Ok, I think I see the problem. You conflate "correlation implies causation" with "questionable cause". They are not the same. The former is a form of the latter, but questionable cause covers far more than "correlation implies causation". If you link to the former instead of the latter, we're good on that.

Leaves only "spurious causation" and "drawing accurate conclusions". Tally-ho! Paradoctor (talk) 21:04, 23 January 2025 (UTC)

Ok, let's go with that. I'm not sure I fully agree with that hierarchy, but that is a broader discussion than is warranted here.

In observational research, distinguishing between association and causation is critical for avoiding spurious causation and in drawing accurate conclusions. Association is a statistical relationship between two variables, whereas causation implies that one variable directly affects the other.

References

Squatch347 (talk) 21:27, 23 January 2025 (UTC)

Um, sources? Paradoctor (talk) 21:34, 23 January 2025 (UTC)

Whoops, copy/pasted wrong text. Happy to add more, but thought these two should cover it. How about now? Squatch347 (talk) 21:44, 23 January 2025 (UTC)

The second source talks of "spurious correlation", not "spurious causation". It also never talks about "accurate conclusions", so it doesn't get the job done.

The first source is paywalled, so please provide relevant quote(s) from it. Paradoctor (talk) 22:14, 23 January 2025 (UTC)

I would rely a bit more heavily on its reference to masking true causal relationships. The full quote from that abstract that is relevant is: "The major challenge using 'observational data for causal inference' is confounding variables: variables affecting both a causal variable and response of interest. Unmeasured confounders—known or unknown—lead to statistical bias, creating spurious correlations and 'masking true causal relationships.'"

If you would prefer, this source also makes that connection: "For example, an observational study might show an association (correlation) between a predictor variable (independent variable) and an outcome (dependent variable); however, the results may not represent cause and effect." https://pmc.ncbi.nlm.nih.gov/articles/PMC10036082/

For the first source, it is both the title of the article and in the background section of the paper: "The differentiation between association and causation is a significant challenge in medical research, often further complicated by cognitive biases that erroneously interpret coincidental observational data as indicative of causality."

Squatch347 (talk) 22:25, 23 January 2025 (UTC)

Same issue as with the second source.

You've ignored this several times now: the exact terms must be in the source. They are not. You're wasting your time and mine unless you propose sources that pass this minimal requirement.

WP:BURDEN: The burden to demonstrate verifiability lies with the editor who adds or restores material, and it is satisfied by providing an inline citation to a reliable source that directly supports the contribution. (added emphasis mine)

The operative word is "directly", and I already quoted its definition above. This is core content policy, and you won't be able to argue your way around it. This is the way. Paradoctor (talk) 23:10, 23 January 2025 (UTC)

So then, this should be acceptable then?

In observational research, distinguishing between association and causation is critical for avoiding spurious correlation that could imply a cause-and-effect relationship exists where none actually does exist and in drawing accurate conclusion as to the magnitude of the effect. Association is a statistical relationship between two variables, whereas causation implies that one variable directly affects the other.

References

I would say that the interpretation of explicit you are using here isn't appropriate. Explicit does not mean verbatim.[1] Wikipedia is not just a series of quotes. Explicit means the phrase or concept is clearly being referenced in the source and that it doesn't require interpretation to get from the wiki text to the source text. That is undoubtedly the case here with the original phrasing. When a source says that you should be warry that data could "make it appear that there exists some type of cause-and-effect relationship between two variables that doesn’t actually exist" they are, quite explicitly, saying that you should be careful of implying causation (a cause-and-effect relationship) that is spurious (doesn't actually exist). This isn't OR or Synth or interpretation, its just normal writing.

I might be a bit more sympathetic if spurious causation was a term of art that might have some additional context or subtle meaning lost on us, but it is just the word "spurious" (not what it appears to be) and "causation".

Squatch347 (talk) 14:41, 24 January 2025 (UTC)

Ok, we two clearly have arrived at WP:NOCONSENSUS, meaning no change for now.

If you don't like WP:HORSEMEAT, your next stop is WP:CONTENTDISPUTE. Paradoctor (talk) 15:03, 24 January 2025 (UTC)

I would certainly agree that we don't have a consensus on the interpretation WP:Burden, but that isn't really key to the edit. I updated the recommended edit to your standard of verbatim text from the source, so it should be ok now. Is there a different objection I missed? Squatch347 (talk) 15:15, 24 January 2025 (UTC)

You've been missing my objections all the time, that's the problem. Frankly, I'm exhausted having tried to work with that.

I'm not going to reply to you on this matter anymore unless I see you conducting productive WP:CONTENTDISPUTE resolution. Paradoctor (talk) 15:27, 24 January 2025 (UTC)

Fair enough, before I post this for a third opinion do you want to make an edit to the summary text? Just want to make sure it is neutral and accurate.

The initial discussion has ground to a standstill primarily over whether or not the sources attached to a proposed edit supports the content. To wit, does the source's language of "The major challenge using observational data for causal inference is...statistical bias, creating spurious correlations and masking true causal relationships" support this content "In observational research, distinguishing between association and causation is critical for avoiding spurious correlation that could imply a cause-and-effect relationship exists where none actually does exist"

Squatch347 (talk) 15:42, 24 January 2025 (UTC)

(Here from Third Opinion): We'll have to split a few hairs here to answer this question, but technically the excerpt is talking about the disutility that observational data itself can have when trying to make causal inferences, whereas the proposed claim by Squatch347 seems to connote that there is this special research skill that one must possess while conducting observational research: the ability of "distinguishing between association and causation", which is misleading. If I were to paraphrase that excerpt and try to hew closely to the terms in your claim, it would be:

"Observational research has limited value in establishing causation because it cannot reliably distinguish between causation and mere association. Failing to recognize this important limitation can lead to spurious correlations, where a cause-and-effect relationship is mistakenly assumed, even though no such relationship exists. Manuductive (talk) 15:46, 25 January 2025 (UTC)

I think your critique is correct, I was trying to shoe-horn in some language from the source as part of the discussion above that doesn't quite fit. The proposed addition, is more a comment about the possibility of committing a fallacy when relying on observational data (which is what I read the source as saying) rather than about the need for any particular research design or statistical skill.

Perhaps the issue is that my rewrite wasn't bold enough. Perhaps is should be something more akin to:

Using observational data in research can present significant obstacles to accurately establishing causation including, primarily, spurious correlation where a causal relationship is established rather than just a broader association between variables. (Sources)

Squatch347 (talk) 19:10, 25 January 2025 (UTC)

What source is that excerpt from? Please cite the source when you quote an excerpt. Manuductive (talk) 16:58, 26 January 2025 (UTC)

The section on the bottom was my recommended addition to the article, not a quote. The source is the two sources I referenced on the previous suggested text. Specifically, [2] and [3] Squatch347 (talk) 20:18, 26 January 2025 (UTC)

What I mean is, which source contains the excerpt you provided in your query: "The major challenge using observational data for causal inference is...statistical bias, creating spurious correlations and masking true causal relationships" Manuductive (talk) 21:58, 26 January 2025 (UTC)

Ahh, I see. The reference for that statement is on here [4] as "3" which is linked above. The specific quote is in the abstract "The major challenge using observational data for causal inference is confounding variables: variables affecting both a causal variable and response of interest. Unmeasured confounders—known or unknown—lead to statistical bias, creating spurious correlations and masking true causal relationships."

But I'll also say if you scroll up a bit in the conversation there are a good half dozen other sources and quotes in case those might offer a bit more support for the edit.

Squatch347 (talk) 23:33, 26 January 2025 (UTC)

Your claim is too far off from that source to be verified, but you could say something like:

"Researchers aiming to use observational data to accurately establish causation must control for confounding variables, as failing to do so can lead to spurious correlations—where a causal relationship is mistakenly inferred from a mere general association between variables." Manuductive (talk) 00:25, 27 January 2025 (UTC)

OK, I think that that is pretty close. I might suggest this wording:

• Researchers aiming to use observational data to accurately establish causation must control for chance, (random error), bias (systematic error), and confounding variables, as failing to do so can lead to spurious correlations—where a causal relationship is mistakenly inferred from a mere general association between variables.

I'd add this source from above [5] based on this quote, "It is important to note that some associations found in an observational study do not represent cause and effect. However, there are well-recognized explanations for associations between a predictor and outcome in such studies (Hulley, 2013). These occur by chance (random error), bias (systematic error), and confounding variables."

Squatch347 (talk) 00:39, 27 January 2025 (UTC)

Where in either of these texts does it say that failing to control for chance (random error) and bias (systematic error) can lead to spurious correlations? Manuductive (talk) 08:47, 27 January 2025 (UTC)

From the source listed as "5" above we have this:

This paper focuses on understanding causal inferences and methods to improve them for observational studies. For example, an observational study might show an association (correlation) between a predictor variable (independent variable) and an outcome (dependent variable); however, the results may not represent cause and effect (Hulley, 2013). Causal inference implies an intervention, e.g., treatment or behavior was the ‘cause’ of the effect (or outcome)...It is important to note that some associations found in an observational study do not represent cause and effect. However, there are well-recognized explanations for associations between a predictor and outcome in such studies (Hulley, 2013). These occur by chance (random error), bias (systematic error), and confounding variables...Chance or random error creates associations between a predictor and an outcome from an observational study sample which is erroneous...Systematic bias can result in an erroneous association where no relationship exists or masks a true association between variables...[6]

The main thrust of this paper is that we can see correlations (associations) in data sets that are spurious and do not represent cause an effect. The source of that correlation in cases where there is no cause and effect relationship can be of three types: Chance, Bias, Confounding.

Squatch347 (talk) 14:29, 27 January 2025 (UTC)

No, I think erroneous association and spurious correlation are different enough concepts that you're going to have to re-word that claim in order to get it verified. Manuductive (talk) 16:46, 27 January 2025 (UTC)

I'm open to any suggestions you might have for phrasing that would work better. Or maybe could you expand on how they are different concepts?

The source equates association and correlation in the abstract as does the wiki article on Spurious relationship and correlation which is the major term that might have subject matter distinction.

The only question it would seem is whether or not erroneous and spurious mean the same thing in this context. The sources don't use erroneous as a term of art, but exclusively to mean mistaken or incorrect which is exactly what is meant by spurious correlation.

From the source above:

"Chance or random error creates associations between a predictor and an outcome from an observational study sample which is erroneous (Hulley, 2013). In other words, the association does not exist in the population"

and

"Systematic bias can result in an erroneous association where no relationship exists"

Our primary source on the article [7] defines this fallacy as "The differentiation between association and causation is a significant challenge in medical research, often further complicated by cognitive biases that erroneously interpret coincidental observational data as indicative of causality."

And, from our source on the definition of Spurious Correlation:

"A situation in which measures of two or more variables are statistically related (they cover) but are not in fact causally linked."[8] (Our source for the definition on the wiki page)

The sources use the term erroneous to mean assuming a causal association (correlation) between variables that isn't there, which is what is meant by the term "Spurious Correlation"

Squatch347 (talk) 18:56, 27 January 2025 (UTC)

I think you make a good case for your position. I'd be interested to hear what @Paradoctor has to say about this and whether you guys can come to a consensus on that. Manuductive (talk) 19:03, 27 January 2025 (UTC)

All I'm really interested in here is that any claims added are verifiable from sources without having to do interpretive dance routines to make the connection. I'm flatfooted. 🤷 Paradoctor (talk) 19:52, 27 January 2025 (UTC)

I think there might be a bit of creative interpretation going on here, but it could just be that Squatch is a pretty competent editor and good at weaving together the different sources into one claim. Really, the only issue I have is whether the words "spurious correlation" in Byrnes 2025 refers literally to the exact same concept as "erroneous association" in Capili 2023. Capili 2023 does say "association (correlation)" which you could say means that they are the same thing. And then Squatch says the word "spurious" and "erroneous" are synonyms. If you don't buy it, then we should probably not go with it and perhaps re-write the claim as:

Researchers aiming to use observational data to accurately establish causation must control for confounding variables, as failing to do so can lead to spurious correlations—where a causal relationship is mistakenly inferred from a mere general association between variables. (Byrnes 2025) Associations in observational studies may not indicate causation and can arise due to random error (chance), systematic error (bias), or confounding variables influencing both the predictor and outcome. (Capili 2023)

Manuductive (talk) 20:34, 27 January 2025 (UTC)

Overall, I think this is actually a great edit. If anything, I like the edit to the last sentence more than mine above. I can draft it up and add if there aren't objections. Squatch347 (talk) 20:41, 27 January 2025 (UTC)

Please make sure to use the |quote= parameter of the citation template. Generally a good habit, but here I consider it essential. Paradoctor (talk) 21:01, 27 January 2025 (UTC)

Proposed final wording

Researchers aiming to use observational data to accurately establish causation must control for confounding variables, as failing to do so can lead to spurious correlations—where a causal relationship is mistakenly inferred from a mere general association between variables. Associations in observational studies may not indicate causation and can arise due to random error (chance), systematic error (bias), or confounding variables influencing both the predictor and outcome. — Preceding unsigned comment added by Squatch347 (talk • contribs) 2025-01-27T21:39:53 (UTC)

References

Couple of changes

"Accurately establish" is not in the source, inference is. Copyedit final clause, correlation is not inferred, causation is.

−

Researchers aiming to use observational data to accurately establish causation must control for confounding variables, as failing to do so can lead to spurious correlations—where a causal relationship is mistakenly inferred from a mere general association between variables.

+

Researchers aiming to use observational data to infer causation must control for confounding variables, as failing to do so can lead to spurious correlations, which then lead to mistakenly inferring causal relationships from mere associations between variables.

The last sentence is not in the paper.

−

The major challenge using 'observational data for causal inference' is confounding variables: variables affecting both a causal variable and response of interest. Unmeasured confounders—known or unknown—lead to statistical bias, creating spurious correlations and 'masking true causal relationships...For example, an observational study might show an association (correlation) between a predictor variable (independent variable) and an outcome (dependent variable); however, the results may not represent cause and effect.

+

The major challenge using 'observational data for causal inference' is confounding variables: variables affecting both a causal variable and response of interest. Unmeasured confounders—known or unknown—lead to statistical bias, creating spurious correlations and 'masking true causal relationships

Pruned to reveal the sexy bits.

−

This paper focuses on understanding causal inferences and methods to improve them for observational studies. For example, an observational study might show an association (correlation) between a predictor variable (independent variable) and an outcome (dependent variable); however, the results may not represent cause and effect (Hulley, 2013). Causal inference implies an intervention, e.g., treatment or behavior was the ‘cause’ of the effect (or outcome)...It is important to note that some associations found in an observational study do not represent cause and effect. However, there are well-recognized explanations for associations between a predictor and outcome in such studies (Hulley, 2013). These occur by chance (random error), bias (systematic error), and confounding variables...Chance or random error creates associations between a predictor and an outcome from an observational study sample which is erroneous...Systematic bias can result in an erroneous association where no relationship exists or masks a true association between variables...

+

It is important to note that some associations found in an observational study do not represent cause and effect. However, there are well-recognized explanations for associations between a predictor and outcome in such studies (Hulley, 2013). These occur by chance (random error), bias (systematic error), and confounding variables

With these changes, the proposal has my WP:!VOTE. Paradoctor (talk) 23:14, 27 January 2025 (UTC)

First change has my support.

Second change is a great catch, thank you. When I was compiling the quotes I mixed the Capeli and Byrnes quotes. I'll transfer that one over to the correct citation.

Third change makes a lot of sense as well.

I concur on Paradoctor's updated text.

Squatch347 (talk) 23:36, 27 January 2025 (UTC)

I'm proud of you guys. Manuductive (talk) 12:52, 28 January 2025 (UTC)

Ok, I added to page with diff. I reviewed it a couple of times to make sure I got it right, but let me know if I missed something. Squatch347 (talk) 16:28, 28 January 2025 (UTC)