Skip to content

I’m with Errol: On flypaper, photography, science, and storytelling

[image of a cat going after an insect]

I’ve been reading this amazing book, Believing is Seeing: Observations on the Mysteries of Photography, by Errol Morris, who, like John Waters, is a pathbreaking filmmaker who is also an excellent writer.

I recommend this book, but what I want to talk about here is one particular aspect of Morris’s work which brings together a bunch of things I’ve been thinking about in the past few years, regarding how we understand and communicate about the world.

While reading Morris’s book I came across this quote:

Photographs attract false beliefs the way flypaper attracts flies.

But it’s more than that. Flypaper doesn’t just attract flies, it also traps them. Similarly, photographs, by having many appealing hooks, attract false beliefs, and photographs also can sustain false beliefs by appearing to supply evidence to confirm all sorts of erroneous notions.

This connects to two of our recurring themes:

1. The immutability of data and stories. Recall that Thomas Basbøll and I argued that we learn from stories because of their anomalousness and their immutability: a story is anomalous when it refutes some existing theory we had about the world, and it’s immutable to the extent that it has an existence independent of these theories.

In my terminology, stories, which have immutable, telling details, are different from parables, whose details can be adapted to fit the message they are sending. In our other paper, Basbøll and I criticized plagiarist Karl Weick in part because, by changing the specifics of the story he was plagiarizing, he was allowing the story to convey a message that was in many ways opposite to that of the original telling. Weick was turning the story into a parable, and his plagiarism facilitated that transformation by making it harder for readers to learn from the original source.

2. A central message of Morris in his book on photography, and in his work more generally, is that images (or, more generally, statements and other data) can mislead, but if we look at them carefully and with skepticism, we can discover things we never would’ve learned without careful inspection and interrogation.

This is related to our emphasis on the complementarity of two statistical methods that are often taken as competing: exploratory data visualization, and complex Bayesian modeling.

We learn by looking hard at an image (or transcript, or other form of data), but we can learn more in the context of a model—and we can try out model after model to see how they fit the data. This was how Morris’s breakout movie, The Thin Blue Line, worked. The notorious reconstructions in that movie were, essentially, posterior predictive checks. And this in turn connects to our idea of storytelling as predictive model checking: storytelling is a creative exploration of possibilities and in that sense can be considered as a form of deductive reasoning, or working out of consequences.

3. Finally, let’s return to the subject of false beliefs that are inspired by, and appear to be supported by, data. This has come up a lot recently with debunkings of famous studies that had purported to show that being primed with elderly-related words makes you walk more slowly, or that elections are determined by shark attacks, or that subliminal smiley faces can have big effects on political attitudes on immigration, or that 20% of women change their vote preferences based on the time of the month, or that beautiful parents are more likely to have girls, and a million other things. What was striking about these cases is that, once people had what appeared to be convincing evidence of their theories, they refused to back down—even after careful investigation of the evidence made it clear that these theories were not supported at all.

A theory is created out of misleading evidence, the evidence is knocked down, the theory remains. Again, this is a theme that should be familiar to Errol Morris fans, along with the followup, which is when defenders of the unsupported theory start to argue that the details of the evidence don’t really matter.


  1. I am reminded of a passage from G. K. Chesterton’s essay “The Fallacy of Success.” He quotes the author of an article titled “The Instinct That Makes People Rich” (which praises Cornelius Vanderbilt to the outer spheres):

    “In olden days [this instinct’s] existence was fully understood. The Greeks enshrined it in the story of Midas, of the ‘Golden Touch.’ Here was a man who turned everything he laid his hands upon into gold. His life was a progress amidst riches. Out of everything that came in his way he created the precious metal. ‘A foolish legend,’ said the wiseacres of the Victorian age. ‘A truth,’ say we of to-day. We all know of such men. We are ever meeting or reading about such persons who turn everything they touch into gold. Success dogs their very footsteps. Their life’s pathway leads unerringly upwards. They cannot fail.”

    Chesterton responds:

    “Unfortunately, however, Midas could fail; he did. His path did not lead unerringly upward. He starved because whenever he touched a biscuit or a ham sandwich it turned to gold. That was the whole point of the story, though the writer has to suppress it delicately, writing so near to a portrait of Lord Rothschild. The old fables of mankind are, indeed, unfathomably wise; but we must not have them expurgated in the interests of Mr. Vanderbilt. We must not have King Midas represented as an example of success; he was a failure of an unusually painful kind. Also, he had the ears of an ass. Also (like most other prominent and wealthy persons) he endeavoured to conceal the fact. It was his barber (if I remember right) who had to be treated on a confidential footing with regard to this peculiarity; and his barber, instead of behaving like a go-ahead person of the Succeed-at-all-costs school and trying to blackmail King Midas, went away and whispered this splendid piece of society scandal to the reeds, who enjoyed it enormously. It is said that they also whispered it as the winds swayed them to and fro. I look reverently at the portrait of Lord Rothschild; I read reverently about the exploits of Mr. Vanderbilt. I know that I cannot turn everything I touch to gold; but then I also know that I have never tried, having a preference for other substances, such as grass, and good wine. I know that these people have certainly succeeded in something; that they have certainly overcome somebody; I know that they are kings in a sense that no men were ever kings before; that they create markets and bestride continents. Yet it always seems to me that there is some small domestic fact that they are hiding, and I have sometimes thought I heard upon the wind the laughter and whisper of the reeds.”

  2. Paul Alper says:


    “Finally, let’s return to the subject of false beliefs that are inspired by, and appear to be supported by, data.”

    From Susan Perry,

    commenting on

    “The use of calcium and vitamin D supplements does not appear to protect older adults from breaking their hips or other bones, according to a major new study published last week in the Journal of the American Medical Association (JAMA).”

    Further, “The meta-analysis also found that people with at least 20 nanograms per milliliter of vitamin D in their blood — an amount considered healthy — were at an increased [!!] risk of hip fracture if they subsequently started taking vitamin D supplements. The reason for this association was unclear, the study’s authors write.”

    To make this information perhaps even more timely, Orrin Hatch has just finally decided to leave the U.S. Senate. According to

    “[in 1994] Senator Orrin G. Hatch of Utah midwifed through Congress a new industry protected from all but minimal regulation. It is also an industry that would make many of his closest associates and family members rich. In turn, they’ve rewarded him with sizable campaign contributions.”

    Further information about Hatch can be found at

    Back to vitamin D but this time with Alex Jones who is not retiring but is selling a product known as “Winter Sun”:

    “You could grab a bottle for around $10 and skip the 2X+ price markup from Infowars.”

  3. Z says:

    This is a timely post as I just finished watching Wormwood, Errol Morris’s new documentary miniseries on Netflix. Wormwood also provoked some vaguely Bayesian thoughts about thinking. [SEMI-SPOILERS AHEAD] It convinced me that the CIA has routinely committed all sorts of egregious crimes like assassinating American citizens to cover up other misdeeds. Before watching the movie, I’d have said, “Of course the CIA does super nefarious things, it would be totally naive to believe otherwise!” Yet if you believed that the CIA was behind any given specific incident I’d think you were a crackpot conspiracy theorist.

    Before Wormwood, due to general ignorance, I hadn’t been aware of any specific CIA murder. I thought that while the CIA almost certainly did crazy bad stuff from time to time, any given suspicious incident was a priori highly unlikely to be their doing. People in positions of power working on highly classified projects do sometimes get into car crashes or commit suicide just like everyone else, after all. After Wormwood showed me convincing evidence that the CIA were behind one particular murder (and that they had a playbook for similar murders), though, I find myself much more willing to indulge the possibility that the CIA is behind other specific suspicious incidents. This doesn’t seem rational, since I already “knew” that they did bad stuff before the movie. Does Kahneman have a name for this type of thought error (if it is one)? Maybe I was just assigning a higher probability than I realized to the possibility that the CIA didn’t do stuff like this? Maybe it’s difficult to assign as high a prior probability to something in your heart of hearts as you intellectually believe you should if you haven’t seen a concrete example? Or maybe your prior was correct in both heart and mind, but seeing an example makes you overadjust?

  4. Wittgenstein suggested that a proposition is a “picture of a fact”. That’s importantly different from a picture of, say, a scene or a landscape or a platter of fruit. A picture in the ordinary sense doesn’t claim that anything in particular is true. It just looks like something that could be true.

    Perhaps we can think of models as pictures, not of facts, but of probability spaces for facts, or just probable facts. While a proposition says that one or another thing is the case, i.e., makes a truth claim, the model only tells us how likely various states of affairs are.

    Stories, we might now say, are pictures, not of fact or probable facts, but (as Heminway put it) “sequences of motion and fact”. They are, perhaps, “motion pictures of facts” and, since they must always simplify the sequences in some way, are ultimately like models, albeit governed by plausibility rather than probability.

    There’s a point at which a story becomes so implausible that you may as well be recounting a dream. I think that’s a good analogy for what happens to models when, as Andrew has been arguing lately, they are detached from theory. Or the forking paths problem is like that old image of being awake vs. dreaming as like the difference between reading a book from start to finish and just flipping randomly through it.

    Some people tell stories that don’t “fork” when you ask questions. They just go into another level of detail that is plausible in the context of the broader strokes you’ve already heard. You can tell someone is making things up when a new detail doesn’t seem to care about previous motions and facts in the narrative. At some point you’re left with a pile of images that stand in no logical relationship to each other.

    And the individual photograph of course is always such an image. It’s not even a picture of a fact. It’s just a picture.

    • Andrew says:


      I’m reminded of the version of 20 Questions where, instead of having a particular object in mind, you’re allowed to keep answering the questions however you’d like, as long as you can think of at least one object that’s consistent with all your answers so far. It can be surprisingly difficult to do this.

      • Yes, it’s surprising how much “reality” actual constrains us if we respect it. Like how difficult lying is when someone actual gets interested in the story. Or how surprisingly short a string of words can be before it is almost certainly plagiarism if it is identical with a string someone else has already written.

    • Keith O'Rourke says:

      Alternatively, Peirce argued photographs are signs that represent indexically – the same as a footprint being left in the clay – actually being physically connected to brute force reality (there are other ways signs represent). To Peirce “A sign not only stands for something; it also stands to someone or to something, by creating in that person or in that process (natural or artificial) another sign, the interpretant, that puts the sign into relation with its object”.

      So perhaps for this post – a photograph has a true representation (being an index) but can create many other signs, interpretants, that are not. On the other hand the true representation always holds the promise of creating an interpretant that is a true representation.

      Setting aside fading and digital photography.

      • Where does Peirce say that? It’s uncannily close to something I wrote recently about the possibility of a “natural” form of writing, like a camera obscura produces natural pictures or a shadow is a kind of natural picture of an object. “Perhaps the individual footprint is a picture, but a series of footprints tells a story.”

        I don’t agree that a single photograph or footprint “represents” anything. But an arrangement of such pictures, along with some assumptions, may indeed do so. Maybe with enough assumptions a single footprint can represent something quite vague (“someone walked here once”) but the assumptions are going to be doing a lot of work there.

        • Dzhaughn says:

          I don’t think the presence of a series matters so much. The switch from photograph to story is automatic in the human mind; the mind will imaginatively create the series. (Even more elementary, it will organize a set of photographs into a series according.)

          I’m reminded of the introductory Bayesian exercise “You see cable car number 73. Estimate the number of cable cars in the city.” And you can. When you subsequently see number 37, your estimate gets a lot narrower. (When you see another numbered 73, that’s different.)

          • If I understand you correctly, this point is captured by the assumptions I mentioned. Yes, the human mind almost automatically constructs a story from even a single photograph (or even a single footprint). But we usually understand that the photograph is providing us with very little evidence to support that story. The rest comes from background knowledge (perhaps about the people in the photo.)

            If we have a series of photographs but don’t know the right order, we can have interesting discussions about what the most likely order is by telling more or less plausible stories about what could have “happened”. Suppose in half the pictures Bob’s glasses are on and in half the pictures they are off. At first pass, it’s less plausible to alternate the pictures with them on and off than to imagine that, as some point in the “story”, Bob took them off or put them on. Other details in the photographs will suggest when/why that happened.

        • Keith O'Rourke says:

          > Where does Peirce say that?
          This might be a good start

          >I don’t agree that a single photograph or footprint “represents” anything.
          From a reference on a post for later today “Peirce notably argued, against Descartes, that no idea can be conceived as clear and distinct by being considered alone. It is rather meaningful and determinate only in relation with other ideas that interpret it, and with the conduct it contributes to produce. Meaning is thus never located, for Peirce, in a single thought: it is rather found in the process through which a thought refers to some object by virtue of the thought that precedes it and by giving rise to a new thought that interprets it, itself referring to some object by virtue of the production of a new thought and of its relation to the thought it interprets, and so on ad libitum;”

          > shadow is a kind of natural picture of an object
          I used shadow(s) as a metaphor for sample(s) that statisticians use to guess at what cast the sample(s) “Nature casts shadows, statisticians wisely guess – fallibly”. I should do a post on that.

    • Dzhaughn says:

      It’s interesting how you seem to take picture to mean photograph. (I don’t know if that’s what Wittgenstein intended or not.)

      Is it different if he said it was a “painting of a fact?” Why?

      How about if he said “graph of a fact?”

      • “Picture” translates the German “bild”, which could also be rendered “image”. (Imagination is “einBILDungskraft” in German.) By “picture”, I don’t mean photograph but what (figurative) paintings, drawings, and photographs (as well as shadows and footprints) have in common. Both “painting” and “photograph” are far too specific about process and materials to capture Wittgenstein’s meaning.

        A graph, perhaps, is a picture of the data and, in so far as a distribution is a fact, a picture of a large, complex fact.

  5. Steve Wang says:

    I’m a big fan of Errol Morris’s movies. I love The Thin Blue Line, and Fast Cheap and Out of Control is one of my all-time favorites. I would not have expected to read about his work on a statistics blog. Great post.

  6. Nice post. Maybe, as a ESL learner, I missunderstand some subtetlities in the post, particularly the differences between “stories” and “parables”.
    Therefore, I am not able to translate those differences to exploratory data analysis and Bayesian modelling. For me, data analysis is a tool to extract information from observed data. While, bayesian modelling is a complete framework to interpret the world producind those observations.
    Exploratory data analysis would be like developing a 35 mm film roll in the dark room (I prefer this image instead of digitally developing photos with software). Bayesian modelling seems Andy Warhol composing his Campbell’s soup pictures.

    • Keith O'Rourke says:

      > particularly the differences between “stories” and “parables”.
      Perhaps think of an un-altered photograph as a story and an altered photograph (e.g. photo-shopped) as a parable.

Leave a Reply