10. The graphs were then redrawn with the best fit line represented as well as the data points, with rogue points processed out.
As per above, such graphs with less than 5 data points and/or with a slope (b) of less than 0.5 were almost always unsatisfactory.
11. When the suitability of the price vs carcase weight graphs was compared with the suitability of the price vs edible weight graphs, the results were an almost imperceptible difference in r values but a significant improvement in SD values when using edible weight on the x axis - i.e. using best fits on log(price) vs. log(edible weight) appears to produce the closest correlation with recorded data.
This table clearly also shows that liveweight is an unsuitable standard upon which to base a log-log best-fit method (entirely to be expected).
12. Incidentally, this SD measure (rather than the r measure) could enable us to compare these best fit log-log techniques with more primitive averaging techniques (e.g. average cost per live kg, average cost per carcase kg, average cost per edible kg).
13. Our conclusion from the above is that best-fit straight line techniques on log(price) vs. log(edible weight), for each market-day-sex combination, gives us an accurate technique to eliminate rogue data and to average out the remaining non-rogue data. It also gives us a measure of our accuracy (with the SD parameter).
The task was then to incorporate this data smoothing technique into the computer system.