that were chosen. The mean area of the sample is 17,875m2 with
a standard deviation of 14,113m2. The minimum is 1748m2 and
maximum 63,585m2. Since the logarithm of area was also found
to be significant indicating that scale differences affect energy efficiency,
this variable was also included to normalise EPI. Regression
coefficients were obtained for each of these three standardised variables
with EPI as the dependent variable. An adjusted R-squared of
0.59 was obtained for these three variables and all three variables
are significant at the 0.01 level. Residuals were tested for normality
and both Kolmogorov–Smirnov and Shapiro–Wilk tests were significant
(p > 0.15). A bootstrapping technique was used to arrive
at the final parameter estimates since the sample size is small.
The bootstrapped parameter estimates were then used in Eq. (1)
to determine normalised EPI for each facility. Three normalised
benchmark ranges were determined using the lower third, middle
third and top third groups from the sample to represent low, moderate
and high energy efficiency ranges. The benchmark percentiles
were also estimated using bootstrapping.