Estimates were based on the principle that each sample building represents a certain number of buildings in the target population as defined in section A2. Therefore, each respondent building was given a survey weight that indicates how many buildings of the population are represented by this building. For buildings sampled from the area frame, this survey weight is the product of three factors: the first-stage sampling weight, the second-stage sampling weight and an adjustment factor for non-response. For the sampled buildings from the lists, this survey weight is the product of the sampling weight and adjustment factor only.
For buildings sampled from the area frame, the sampling weight is the product of the first-stage sampling weight and the second-stage sampling weight. The first-stage sampling weight (for a building in stratum i and EA j) is
w1ij | = | total number of buildings in stratum i |
_______________________________________________________ | ||
total number of buildings in EA j x number of EAs sampled in stratum i |
And the second-stage weight (for a stratum k of EA j of stratum i) is given as follows
w2ijk | = | total number of buildings in stratum k of EA j of stratum i |
_______________________________________________________ | ||
total number of buildings sampled in stratum k of EA j of stratum i |
The sampling weight is ultimately
wijk = w1ij x w2ijk |
There was an EA census for certain second-stage strata. For buildings in these strata, the sampling plan therefore became a simple stratified sampling plan. Details are provided in section A6.1.2
For buildings taken from lists and for buildings from the area frame for which the first-stage strata were "take-all" units, the sampling weight was given as follows:
wk | = | total number of buildings in stratum k |
_______________________________________________________ | ||
total number of buildings sampled in stratum k |
When the respondent failed to answer certain questions only, the missing data were imputed in accordance with the methods described in section A5.1. In other words, the weight was adjusted to offset total non-response when contact was impossible, when a respondent refused to answer or when the data provided could not be used (see section A5.2 for exceptions).
Weight adjustment for non-response is based on the principle that respondent buildings can be used to represent all buildings: respondents and non-respondents. The sample is first subdivided into distinct non-response classes to increase the chances that respondents and non-respondents have similar characteristics. In this survey, non-response classes correspond with second-stage strata for buildings from the area frame and with first-stage strata for buildings from lists.
Before making the weight adjustment for non-response, all strata in which there were no respondents, or at least one non-respondent, were identified and paired with strata having at least one respondent. Five trials were held to find one stratum similar to the problem strata, as shown in Table A10.
TABLE A10. TRIALS AND STRATA CHARACTERISTICS
Strata characteristics | |
Trial | |
1 | Same EA, same type of building, same size group |
2 | Same EA, same type of building, same age group, next larger size group |
3 | Same EA, same type of building, same age group, next smaller size group |
4 | Same EA, same type of building, next larger size group |
5 | Same EA, same type of building, next smaller size group |
If, after these trials, certain problem strata were still not paired, all non-respondents from these strata were mass imputed (see section A5.2 for details).