In a recent posting on VKLOGGER, I posted the graph below of licence trends to Jun 2013.
VK3HZ commented at Re: Amateur population trends – 2013:
I wonder who it really was who first said “Lies, damned lies and statistics”?
I’m surprised any conclusions can be reached based on the data presented:
– the last point on the graph totally influences the outcome. Take that point away and there’s quite a different (rising) trend line.
– Is this final point influenced by some other effects (e.g. changes or delays in license processing …)?
– why should a second-order polonomial be used to model number of licenses vs. time? Why does it have to have a turning point?
– the vertical axis is greatly expanded, over-emphasising visually any noise on the data
– if the same analysis had been done after the 2007 data had been released, the trend would be even more disastrous. However, reality was much different.
So, I don’t (yet) believe the sky is falling.
I will respond here as I now refrain from posting on VKLOGGER.
To respond to his points:
1. The nature of the curve fit is that it is a least squares fit to an order two polynomial. Only the statistically uninformed could make the statement
the last point on the graph totally influences the outcome;
2. Not known, all the data is influenced by external factors and time will tell whether that point is an outlier. The June 2014 suggests it is not an outlier.
3. A second order polynomial allows the line to have a turning point, that is to change from an upward slope to a downward slope. To force a first order (or linear) fit would be to deny that the data could reach a maximum and trend downwards… it is the question one might expect from deniers.
4. The Y axis is clearly shown as expanded, and it is done to show better the detail of the model fit to the data. To use an axis scaled from zero would make that fit less clear, but interpretation of graphs with expanded scales is a challenge for some people. Deniers may seize on it to claim dishonesty in presentation, but in this case, the scale is clearly shown and it is not dishonest but more informative.
5. The statement
if the same analysis had been done after the 2007 data had been release, the trend would be even more disastrous has been made without due consideration of the facts.
Above is a plot of the model applied to the data prior to 2007. It is a weak model since it depends on only three data points, but it certainly does no show the outcome stated by VK3HZ.
Above, the chart includes the 2007 data points and although the last point is downwards, the model still shows an upward trend… which again shows that the comment in point 1 is uninformed.
Forecasting is always a risky business, and to some, avoiding the information from forecasts of this type is just a form of denial.
Above, the chart update with the June 2014 data, which is fairly consistent with the observation that the trend would peak and turn downwards. The writing was on the wall years ago, and it has come to fruition.
To VK3HZ’s closing comment
so, I don’t (yet) believe the sky is falling.
Quite. I have consistently held the view that the number of licenced amateurs is not a very important statistic, I hold the view that ham radio was a better fit for the ITU definition of the nature of amateur radio when the amateur population was half its current value.
Nevertheless, there will be hand wringing in some quarters, and the actual downwards trend is likely to be cited as reason to dumb the hobby down in the LCD to be remade shortly.