The Science Of Whitewash



A complaint often leveled against climate scientists is that they fake their graphs and that investigations into such fakery which fail to show fakery must accordingly be a whitewash.

I have news for these people.

All graphs are fake.

An argument often used in philosophy is that nothing can be known as 100% truth.  The same can more properly be argued from the findings of science.  Every bit of knowledge comes from human experience and judgment.  The human brain is easily fooled: illusions, hallucinations and mirages demonstrate this. 

We, as human observers, can never be sure that our observations of other humans and of their observations are valid.  No fallible human can ever discover a universal and immutable truth.  Everything we claim to know can be shown by an inexorable chain of logic to be derived from fundamental observations by fallible humans.  And even that chain of 'inexorable' logic falls within the scope of human fallibility.

The way in which we cope with this fallibility as humans and as scientists is this: we polate.  Polation is the filling in of gaps in data streams.  It comes from a Latin word meaning 'to polish, smooth or whitewash'.  The validity of the method is assumed from innumerable instances of observed chains of events.

Cyclic series are easiest to 'whitewash'.  If a calender is printed with all the Sundays missing, almost anybody could figure out to fill in the blanks accurately.  No rational person would accuse the 'data smoother' of cheating.  Almost every observation we make about the real world involves some degree of 'filling in the blanks'.  Humans are adept at interpolation and extrapolation.

Throw the series 'Sunday Monday Tuesday' at anyone and they will guess Wednesday as the next in the series.  That is extrapolation.  Ask someone to insert the missing term in any simple series and they will do it.  That is interpolation.

An experiment: generate a set of numbers using random throws of 1, 2, 3, 4 ... N dice.  Have someone plot the results on a graph, but don't tell them where the numbers came from.  Most people will see a trend in the graph.  Seeing a trend is a form of extrapolation.  You could be a total sadist and set your victim the task of determining the derivation of that trend.  If anyone figures it comes from dice, remember never to gamble money with them.

Interpolation and extrapolation are a part of ordinary, everyday experience.  The fact that we correctly use expressions such as 'A to Z', 'Monday to Friday' in place of lists demonstrates this.  On a different level, we can see an approaching rain cloud and determine ahead of time that we will experience a sequence of dry - wet - dry.  That is extrapolation.  Our expectation of tomorrow's sunrise is extrapolation.

Small children enjoy 'join the dots' pictures.  Joining the dots is an exercise in interpolation, whether done by a four-year-old or a Nobel Prizewinner.  The difference lies, not in the method, but in the closeness of the dots.  A graph of data points becomes ever more accurate with ever more data.  But ultimately, we can never get enough data because the graph is analog and the data is digital.

If you were to log the temperature of a specific location every millisecond, you would get a very accurate temperature log over a decade, but at an enormous cost in hardware.  If the purpose is to determine maximum daily temperatures, then one logged measurement per day is enough.

A graph of maximum daily temperatures for a single location is very easy to draw.  But whatever the resolution of the time axis, lines linking data points must of necessity pass through the area between data points where there is by definition no data.

If you zoom in to pixel level on any published graph of any data you will find many pixels which lie between measured data points.  Every one of those pixels is there solely to smooth the visual impact.


It is inherent in the conversion of lists of numbers into a graphic that many pixels will have more of art and less of science.  Interpolation is unavoidable.  But if you dislike that the graph resembles a golf club or your granny's false teeth - go pick holes in the raw data.