|Metrix » Tuning|
The only reason Metrix would deselect a point in a regression (besides a point not being in the selected tuning period interval) would be if the degree days per day were below the degree days per day minimum or if there was missing weather data for given billing period.
Correlations between energy usage and degree days may be weaker in billing periods with few degree days. This occurs during the crossover months where other factors often have more of an influence on heating or cooling energy than weather--such as when the building engineers switch 2 pipe systems from summer to winter, some control strategies, etc.
There is another way to look at this. If you look at a kWh graph from the tuning screen where CDD was the x-axis, then look to the left (along the y-axis), there is probably a cluster of points there. These are the points that represent the baseload kWh (when no cooling was necessary). As shown below, some bills are higher, and some are lower than the regression's y-intercept due to fluctuations in baseload usage
Excluding these points in your regression is that these points will may improve your R2 value. Not only that, these data points will (unfairly) influence your selection of balance point temperature.
If you play with the min degree days per day values, you should notice that the largest effect takes place around zero. If you enter zero, then you are including all of those baseload days and their non-weather affected fluctuations. It is best to select a number greater than zero, for example 1. At present, no studies exist which show what number is best. The Metrix default of 2.0 was chosen from our experience, however, you should feel free to select other numbers.
If your concern is to keep as many points in the regression as possible, but still have a good regression, we suggest you pick 1 for min degree day per day. However, selecting more points close to the y-intercept doesn't really enhance the statistical accuracy of the correlation much. What is important isn't so much number of points in the regression, but number of points in the higher DD regions.
Three points on a line can give you an R2 of 1.0, however if all three points were close to the origin, you cannot be sure that the correlation will be a good predictor of energy usage far from the origin (in the high DD region). If you have a regression with 12 points and only 1 or 2 are in the higher DD region and the rest are clustered by the origin, then you should wonder about how well your correlation will predict future usage.
Otherwise feel free to contact our Tech Support staff at (805) 329-6565, or via email at firstname.lastname@example.org.