On Predictions

There are three basic forms of predictions that I make, which given this recent podcast I thought I'd better make clear.

First, there are predictions I make regarding the overall process of evolution and the cycle of change. These are based upon causation, correlation and past testing and I'm confident enough in this to describe the model as a weak hypothesis.

Second, there are predictions I make regarding specific industry actors. These are based upon my opinion and naturally any outcomes depends upon the strategy of the actor, the impact of others and economic events. These predictions are therefore inherently uncertain. I'm not confident in these at all as they are purely conjecture that I consider likely.

Lastly, there are predictions I make regarding my on going exercise of prediction testing. Here I'm deliberately aiming for a 50% prediction rate of the headline predictions which are based upon a hundred plus components. This is purely an experiment I'm running to see if I can improve some general models I have. Overall, this experiment is simply an idea at the moment which I'm refining.

For reference, the general scale I use is as follows** -

Idea = anything you think of. It's an idea. No Confidence.

Conjecture = idea + some initial supporting data (upto a hundred or so data points) demonstrating merit. Extremely Low Confidence

Reasoned Hypothesis = idea + initial data + correlation (through a couple of experimental observations) + initial models of causation. Low Confidence

Weak Hypothesis = idea + data (a few hundred to a thousand or so data points) + correlation (supported by a couple of experimental observations) + causation + verification (either independent confirmations or a couple of prediction tests). Weak Confidence

Strong Hypothesis = Weak Hypothesis + large volume of confirming experimental data (tens of thousands of data points) + correlation supported by significant independent experimental observation (in the order of hundreds of) + verification (peer review and and hundreds of prediction tests).  Reasonable Confidence

Theory = Strong Hypothesis + significant volume of data (hundreds of thousands of data points) + independent verification of experiments (in the order of many thousands of) + widespread acceptance by scientific community + large volume of prediction tests (many thousands of). High Confidence

Universally Accepted Theory (law* in all but name) = Theory + so much data (many millions of data points) + so much independent verification (hundreds of thousands of independent experimental observations and predictions tests) that it has become universally accepted within the community, we've almost stopped asking the question and will be genuinely surprised if someone shows it to be wrong. Extremely High Confidence

* In the normal sense a scientific law is simply a description of observed phenomenon whereas theory is the explanation of those observed phenomenon e.g. Newton's Law of Gravity is not the same as the Theory of Gravity. However, I don't know an appropriate word to describe a universally accepted theory, the temptation is to call it a 'law' except that creates all sorts of other confusion. So, I'll simply note that there are categories of theory which have become universally accepted. 

** I refined this list on the 17 August 2013 to give some examples of volume of data that I'd expect to see.