CCRI Sampling - Sample Weighting
In each census, all dwellings with 30 or fewer residents (“regular-sized” dwellings) were selected with equal probability, 5 percent of all dwellings in 1911, 4 percent in 1921 and 3 percent in 1931, 1941, and 1951. In a selected dwelling every person was sampled.
“Large" dwellings, with 31 or more residents, are sampled in two different ways:
In dwellings in which 50 percent or more of residents were deemed to be unrelated persons, one in every 10 respondents was selected (after a random start)
In multi-unit dwellings, those in which 50 percent or more of the residents resided in identifiable household units,
- In 1911, 1 in 4 units was selected at random and all residents of a selected unit were sampled;
- In 1921-1951, 1 in 5 units was selected at random and all residents sampled.
PERSON Weights, for population estimates.
Type of Dwelling
Year “Regular-sized” “Large Single-Unit” Units in “Large Multi-Unit”
1911 20 10 4
1921 25 10 5
1931 33.3333 10 5
DWELLING weights, for the analysis of dwelling characteristics (e.g. numbers of residents, building materials etc.).
For “regular-sized” dwellings, the weights are the same as above; for large dwellings the weights are always 1 (one), since all large dwellings are represented in the sample.
Note on precision in weighted data:
The most precise population estimates that could be made for the large dwelling oversamples would use exact information on the number of residents or units in each dwelling and the exact number of person records or units sampled, depending on the type of large dwelling considered.
For “large single unit” dwellings the ratio would be: Number of residents in selected dwelling/number of persons selected for the sample.
For “large multi-unit” dwellings, the ratio would be: Number of household or family units in the selected dwelling/number of selected units.
Such weights are a form of “post-stratification.” Large dwellings may differ in some systematic ways and such precise weights would make the contribution of the data from each dwelling exactly proportional to its size. Our sampling procedures were sufficiently complex and costly to prohibit the collection of detailed information on the size of each single-unit large dwelling or on the number of units in each multi-unit large dwelling. Normally, such weights only differ slightly from those given above and increase precision very slightly. It is unlikely that the difference in weights will much affect substantive analysis of the weighted large dwelling data.
