Regression to the mean, publication bias and confirmation bias

You’ve had a cold, and someone suggested drinking an extract from a new wonder plant. The cold got better. You’ve Googled it and lots of people have noticed the same thing, so you’re telling all your friends about it. Well, you’ve probably just experienced regression to the mean and publication bias. There might even be some confirmation bias growing in there as well.

We all do it sometimes. It’s an easy trap to fall into as we seem to be hard-wired in some ways to deceive ourselves. But if you keep in mind what these things are you can avoid becoming a victim of regression to the mean, publication bias and confirmation bias.

And it can save you money!

Regression to the mean

Many things in life follow a pattern: they oscillate around a baseline or average. They get better one day and worse another or bigger sometimes and smaller at other times.

Health is like that. Take a common cold. One day we feel OK. This is the mean. Then we feel unwell with typical cold symptoms. The symptoms get worse for a few days, then they peak, and then after another fews days we feel well again. This is regression to the mean condition of feeling well.

So using this example, when are you most likely to take some action to treat the cold? For most people it’s when the cold is at its peak. You put up with it for a few days and then you take action. Perhaps you visit the chemist and buy some medication, and then a few days later you’re better.

So what cured you? Was it the medication, or was is just regression to the mean? Well the evidence is that it was just regression to the mean. You would have recovered in the same time without medication.

Here’s another example for those of you who follow cricket. Steve Smith, one of the best batsmen around at the moment, is averaging just over 56 runs per test match innings. Sometimes he seems to be scoring 100+ runs every innings, and the bowlers have no way of getting him out. But over time he’ll also have some bad patches where he scores some ducks (0 runs) and some scores in the 10s and 20s. Overall, whether he’s going through a purple patch or going through the horrors, his average will still be close to 56 runs per innings. This is regression to the mean. Half his innings will be below average and half above average.

And another example: parents with high IQ are more likely to have children with lower IQ than them. Similarly, parents with low IQ are likely to have children with higher IQ than them.

It’s simply regression to the mean. Read more about it here:
https://en.wikipedia.org/wiki/Regression_toward_the_mean

Publication bias

Good results from trials are more likely to get the attentions of the media. This is true of both traditional media and social media. For example, a clinical trial of a new drug for dementia is more likely to appear in the nightly news if the results are encouraging. Your favourite TV channel is unlikely to report on a later trial of the same drug that suggested it was no more effective than others already on the market.

Another example: you friend tries a home-remedy for pain relief and the pain goes away. The friend is more likely to post it in FaceBook than another friend who did not get any pain relief. The positive post is likely to get comments and Likes from others who believe they have experienced the same thing. Those who did not experience the same thing are less likely to comment or Like the post.

This is publication bias. Read more about it here:
https://en.wikipedia.org/wiki/Publication_bias

Confirmation bias

Do you have a strong opinion on particular subjects? As a hypothetical, let’s say I believe cold weather makes the pain of arthritis worse. Whenever I have pain in my joints and it’s cold I will tell anyone willing to listen that winter is a bad time for arthritis sufferers. My belief is reinforced by these events. Some will answer that theirs too is worse in winter, further reinforcing my belief and theirs.  I probably won’t notice those cold periods where my arthritis is about the same as normal. In summer if my arthritis plays up then it’s just because I have arthritis. If I see a report that there is little evidence to support seasonal arthritic cycles then I will probably not read it. On the other hand, if I see a report that suggests there is a strong link  between arthritis and cold miserable weather then I will certainly read it and share it with my social media friends.

This is confirmation bias, or selective interpretation of events, that reinforce an already held belief.

Weather is another common example of confirmation bias. Those of your friends who believe in climate change will probably remember and comment more on the unusually warm days whereas those who are climate change deniers will probably notice and remember the colder days, and claim this supports their belief. Of course they’re all talking about weather, not climate, but it does illustrate confirmation bias.

Diet is another area where confirmation bias features strongly. The followers of fad diets swear by them, and attribute any positive change to the diet. (“I haven’t had the flu since I started this diet, but last year I had it twice.”) Negative changes are unlikely to be linked to the diet. (“Can’t understand why I’m getting a lot of gas lately.”)

Read more on confirmation bias here:
https://en.wikipedia.org/wiki/Confirmation_bias