Making better decisions is about the highest leverage area any business can address. This fact is being recognized more often now than in the past — as an example, the decision-related topics on business sites are some of the most popular.

To improve decision making, it is important to understand the basis for your decisions as well as the decision processes of others. One neat little assessment tool you can try on yourself and others is nothing more complicated than the humble coin toss. And you really don’t even have to do the actual flipping in conducting this parlor game — you can just pretend you have flipped the coin.

Anyway, the test goes like this. Assume I take a coin out of my pocket, and I flip it 100 times. Let’s say it comes up heads 64 times. The question is: what do you assign as the probability of the 101st flip being a head? Don’t worry – there is no right answer to this – but any answer you give will reveal a bit about the way you process information and deal with uncertainty, and how you make decisions.

Typically, the most popular answer given is 50%. If you gave that answer, then your implicit assumption is that you know a lot about coins, and you had a prior hypothesis that a fair coin that is flipped a large number of times has on average the head side of the coin face-up half the time. Because 64 heads is likely within the range of outcomes you would expect given this hypothesis, the “experiment” of the 100 flips did not disconfirm your hypothesis. So you assigned 50% as the probability of a head on the next flip. Note that in a sense you used the information provided by 100 coin flips only to reject or accept your previous hypothesis.

Another answer you might give is 64%. If you gave that answer, then your implicit assumption is that all you really know about the coin I flipped is the results of the experiment of flipping it 100 times. And since it came up heads 64 times, you should therefore assign the probability of it coming up heads on the 101st flip as 64%. This seems pretty intuitive, but this approach has a fancy mathematical name – maximum likelihood estimation. Note that in this case, your assumption was that the ONLY information you had was the outcomes of the 100 flips – you implicitly assumed that you had no useful prior information about coin flipping.

Finally, you might give an answer that is somewhere between 50% and 64%; maybe just a little above 50%. If you gave that answer, then you probably feel you know quite a bit about flipping coins, and generally you would expect 50% heads. But you feel that the experiment of the 100 flips should also have some weight in the manner, although probably a lot less than your prior experience. So you shaded your expectation a little above 50%. This approach also has a fancy mathematical name: it’s a Bayesian approach. Note that in this case you used information from BOTH your prior experience and the current experiment.

Ok, so what’s this got do with decision making? Well, hypothesis testers tend to stick with the status quo, unless the proof for change is definitive. Individuals with scientific training are often in this category – hypothesis testing underpins a lot of scientific decision making. Hypothesis testing is often a good approach when change is incremental, but it can be very late in recognizing “paradigm shifts”.

Maximum likelihood estimators can disregard experience, and be too prone to the recency effect – sometimes leading to reckless decisions.

You can probably guess I favor the Bayesian approach – balancing experience with new information as it becomes available. The trick though is to get the proper balancing, and that is generally more art than science.

But try some coin flips with your friends and colleagues – you’ll all learn something — if you aren’t strict hypothesis testers, that is 😉


(Guest post by Steve Flinn)