"Causal Analytics" is ...
... the study of cause and effect relationships found in data through inductive reasoning, the process of deriving general principles from particular facts or instances.  Causal Analytics seeks to answer three important questions:

The Importance
Causal Analytics is the essential reason most people use mathematics, or perform data mining because with it, we can:

• Understand which variables have the greatest bearing on performance or events of interest

• Predict what is going to happen (if we see a cause occur, we can anticipate the effect and its timing)

• Change what is going to happen (if we can influence a cause, we can achieve an effect)

• Diagnose why something happened (if we see the effect, we can look to the causes as to why)

• See shifting causes in non-stationary processes and determine when we are in and out of control and why

Thus, Causal Analytics is important to understanding, modeling, optimization and diagnosis.  Through Causal Analytics, we can more accurately & confidently predict performance, foresee problems, have realistic expectations about what we can do, understand how controllable and consistent our process are, see what factors are the root cause of problems and most importantly, take appropriate actions to achieve higher performance.

Time is a Key Factor
Cause and effect relationships are manifested in TIME.  We theorize that there is always a lapse in time, regardless of how large or small, between a change in a cause and the appearance of the effect.  If A drives B, and A changes, then B will change after some amount of time elapses, whether that is picoseconds or eons.  Time tells us that If B occurs before A, then A cannot be a cause of B.  Accordingly, the analysis of time gives us important clues to cause and effect.  To support this, it is helpful (but not essential) to sample data as fast as the causal time delay to see these effects.  If you know A drives B and in your data it appears that A instantly causes B, then it is likely that the causal propagation is occurring between sampling periods.  This is common for "batch" data, where various measures of each batch of product are made and collected and most effects are within the making of the batch, so most causal effects appear with a zero delay (seemingly instant).  The data is still valuable, its just that we don't know the actual timings of the relationships in those cases.

Causal timing can be a bit more complex because of cause and effect feedback loops.  A feedback loop is where A drives B which in turn drives A again.  An example might be when you increase the thermostat in your house (A), the furnace responds and your house warms (B), but then you feel too warm so you reduce the thermostat a bit (A again).  Thus A causes B and B causes A. However, there is latent time information that will indicate that the relationship is A => B => A ...

There are a variety of "causal relationships".  Next we discuss the four most common.

 Next page >>