CBECS 2018 Misclassification: Issue and Solutions
Frost Hubbard
Westat
James Green
Westat
David Morganstein
Westat
Katie Lewis
US Energy Information Administration
Michelle Amsbary
Westat
The 2018 Commercial Buildings Energy Consumption Survey (CBECS) aimed to efficiently oversample buildings greater than 200,000 square feet due to their greater energy consumption. Previous CBECS cycles found building size was correlated with key statistics of interest such as total annual fuel use. In 2018 and previous cycles, CBECS used a multi-frame design (i.e., multiple existing building lists combined with a multi-stage area probability frame) to meet this objective. Thus, the sample design used a building-size categorical variable to help create the relatively homogenous sampling strata for the building-level selection stage. Not unexpectedly, the categorical frame variables used to create the sampling strata had significant classification error. This research explores how misclassification rates can vary by frame in an establishment survey context and the negative effects on precision of the sample. Specifically, we compare the building-size category assigned on the frame to the building-size category reported by the respondent. Based on these results, we suggest two competing design modifications that may generate a more efficient design in future CBECS cycles.