**Q=** When estimating the total yearly absolute precision with a regression model, we need to consider both the modelling uncertainty and the meter/measurement uncertainty is that correct? Or the modelling uncertainty is sufficient especially if a class 0.5s meter is used for measuring electricity used?

Now, if we need to consider both uncertainties, how exactly do we calculate the total meter uncertainty/precision of the energy reading?

The IPMVP Concepts and Options for Determining Energy and Water Savings, Volume I, January 2012 seems to suggest that we need to take the square root of the sum of the square of precision/standard error of each data interval to get the total precision for say the year long baseline period

However, based on my understanding, since the same meter is used to measure the energy for each interval, then this is considered as dependent uncertainty and therefore to estimate the total precision by the meter we need to directly add the precision of each interval rather than the square root of the sum of the squares (i.e. SE_tot=SE1+SE2 +...SEn rather than sqrt(SE1^2+ SE2^2+...SEn^2)

**A=** Savings uncertainty for a project or a measure can arise from different sources including measurement, sampling, modeling. Typically all these different uncertainties are independent of each other, hence the total uncertainty of the project/measure is calculated as the square root of the sum of the squares of these different errors. In most cases, we more or less assume the measurement error is very small compared to the modeling error for Option B and C, especially if you are relying on a revenue-grade meter, and can be ignored. Sampling error is mostly applicable for Option A.

Uncertainties can be thought of as standard deviations in the data set, and hence can not be simply added as +ve deviations can offset -ve ones, hence masking the true variability in data. Only variances the sum of squared distances from the mean, can be added for a given data. The square root of this variance is the standard deviation which provides an assessment of the variability in the data in its native units.

In addition, it is probably a good practice to compare meter data to utility (if whole building) just to make sure they look similar. Or another reference such as a portable power meter. CT error is usually greater than the meter, especially if split core or Rogowski, so that will drive system error.

Also basic things like checking line voltage, orientation of CT’s and wiring. Otherwise, overall error of revenue grade meter and CT’s typically 1-2% or less.

Please also refer to the IPMVP Application Guide "Uncertainty Assessment for IPMVP, July 2019"