We already know that men are pigs, but now we also know that men are the kind of pigs who corrupt the morality of dogs. And how do we know this? Well, scientists have discovered it, and the media are duly reporting it. “Dogs walked by men are four times more likely to bite other dogs,” the Daily Mail tells us. “Dogs walked by men are more aggressive,” says Discovery News. “Study Finds Dogs are More Aggressive When Walked by a Man,” reports AllPetNews.com, who I assume must know what they are talking about.

These stories relate to a new study published in the latest issue of the academic journal Applied Animal Behaviour Science. The scientists, from Mendel University in the Czech Republic, had their research assistants stand in local parks for weeks on end, watching passers-by walking their dogs; whenever two dogs (and their owners) walked past each other, the research assistants recorded a set of specific observations on a checklist. Now while all that might seem pretty simple, this study was marked by its very large sample size and the strictness of its inclusion criteria. The assistants recorded observations at over 30 locations and amassed data on 935 dog-vs.-dog encounters (i.e., interactions involving 1,870 dogs). Furthermore, they restricted themselves to purely single-dog/single-owner dyads, specifically excluding cases where an owner was seen walking multiple dogs or where single dogs were being walked by multiple owners. (Those research assistants seriously earned their measly stipends.)

Yet again, the subtext here is that the findings must be true because they were discovered by scientists. However, while society might teach us that men are pigs, even when it comes to dogs, I for one smell a rat.

Here is a list of some of the research findings highlighted in the original article. See if you can spot any — any — that surprise you:

1. Male dogs “sniffed” female dogs more than female dogs sniffed males (such “sniffing” marks a dog’s attempt to initiate mating)
2. Male dogs “sniffed” female dogs a helluva lot (technical term) more than they sniffed other males
3. Puppies played with other puppies more than they played with adult dogs, and a helluva lot more than they played with geriatric dogs
4. Small dogs played with other small dogs, medium-sized dogs with other medium-sized dogs, and large dogs with other large dogs
5. Dogs behaved more boisterously when on their leashes than when off their leashes
6. Male dogs appeared more threatened by other male dogs, and female dogs by other females

Now, in these findings all I see is corroboration of what I already know about dogs in particular, and about sexual behaviour in biological organisms and ethology more generally. In other words, as with other mammals with relatively long gestation periods, males are primed to initiate more sexual interactions (or at least to try to do so) and to feel more competition from other males. And dogs, being not stupid, exhibit a certain amount of bravado when attached to leash but are a little more coy when they might be expected to follow through on any posturing.

But here is the finding that led this fairly routine study to pique the interest of the Discovery Channel, and others, these last few days:

• Dogs of either gender bit other dogs four times more often (which sounds like a helluva lot) when both owners were men than when both owners were women.

So there you have it, kind of. Male owners, presumably because of their maleness (eugh!), make their dogs go bad. Morally warped by their testosterone-addled brains, they contaminate the innate nobility of their animal companions and force them to manifest human-like levels of depravity and violence. This is nothing other than man’s (sic) inhumanity to canines.

However, let’s not forget that rat I sniffed — sorry, smelled — earlier. You see, this study doesn’t exactly reveal what it is reported to reveal. For one thing, it is a cross-sectional observational study. This means that we can’t use the data to establish cause-and-effect relationships because we don’t know how things are interconnected, or what happened before or after what. It’s like taking a photograph of a crowded shopping mall with the aim of figuring out why it is all those people decided to go shopping today.

The result is we cannot know why these particular dogs were biting each other. Yes, they were being walked by men, but then maybe these dogs were being walked by men precisely because they were more aggressive than other dogs. In other words, maybe there was a tendency across dog-owning families to ask their menfolk to walk the more troublesome canines. This practice doesn’t have to be widespread for a result to emerge; even a small trend would lead to a statistical finding.

If the researchers wanted to rule out such a possibility, they could have simply asked the men why they were walking those particular dogs. Or they could have handed out brief questionnaires as the walkers passed by. However, as the study was purely observational, they didn’t do any of this and instead invested their resources in maximizing the sample size. But unfortunately, even a large sample size can’t deal with the limitations of observational research.

As an alternative, the researchers could have included things like breed-of-dog as control variables in their statistical analyses. After all, they did report that men were more likely to be seen walking large dogs than small dogs. However, the researchers didn’t use any multivariate statistical tests that would have allowed them to take such known information into account. Instead they looked at plain frequency data and tested statistical significance using rudimentary procedures like the chi-squared test. I have to say that this is all pretty simple stuff. The big problem is that using simplistic statistical procedures greatly increases the risk of generating unreliable conclusions, precisely because of the inability to “control out” alternative explanations.

A related problem is that the investigators tested a huge number of separate research questions but did not use any statistical method to minimize the resulting likelihood of chance findings (in technical language, they failed to “control for multiple testing“). When researchers conduct lots of statistical analyses, the probability-based nature of such tests guarantees that sooner or later a result will emerge that looks like a finding but is not really a “true” finding. In other words, it will be a “false positive“. Given the very large number of separate statistical tests reported in this paper (there were several dozen), it is very likely that some of its results are indeed false positives.

And finally, while the overall research project looked at over 1,870 dogs, the analyses related only to one dog from each dyad (i.e., to 935 dogs), of which, only 3% were seen biting. Therefore, when it came to looking at the impact of owner’s gender on dog aggression — which amounted to a sort of “study-within-a-study” — the researchers were working with a sample of just 28 dogs.

When the media reported their men-transmit-their-intrinsic-aggression-to-their-dogs headlines, virtually all writers insisted on describing the evidence for this as being based on “nearly 2,000 dogs“. Unfortunately for science journalism they were wrong. By a factor of about 100.

So why didn’t they notice? Well, in all likelihood, this is an example of a reasoning problem known as confirmation bias. In a sense, the “finding” here was so much in line with prevailing social prejudices about men and masculinity, that those who chose to report the result as “news” failed to spot the weakness of its foundations. In other words, they allowed their beliefs to shape their interpretation of the data, rather than the other way around.

Confirmation bias is more typically a feature of pseudoscience than of science. Commentators who wish their views to be supported by evidence tend to interpret ambiguous evidence in a way that suits their argument. Such a reasoning error shows us how people’s preconceptions can influence their interpretation of new information.

It also helps us understand another common observation: why it’s hard to teach an old dog new tricks…

Brian Hughes

Brian Hughes is an academic psychologist and university professor in Galway, Ireland, specialising in stress, health, and the application of psychology to social issues. He writes widely on the psychology of empiricism and of empirically disputable claims, especially as they pertain to science, health, medicine, and politics.