Rwanda - Rwanda Labour Force Survery 2021

The sample design of the LFS is a two-stage stratified design according to which at the first stage of sampling, a stratified sample of enumeration areas from the latest population census is drawn with
probabilities proportional to size measured in terms of the census number of households or census number of household members, and at the second stage of sampling, a fixed number of sample of
households is selected with equal probability within each sample enumeration areas. Finally, all household members in the sample households are selected for survey interviewing.
The scope of the survey is all persons living in private households. It excludes the institutional population permanently residing in houses such as hostels; health resorts; correctional establishments etc., as well as persons living in seasonal dwellings not covered in the survey. It also excludes workers living at their work-sites. A household is a group of persons who live together and make common provision for food and other essentials for living. The people in the group may be
related or unrelated or a combination of both. A household may consist of only one person or several persons.

Sample size determination in most household-based surveys with multi-stage stratified design is based on the principle of first calculating the required sample size for a single «domain» assuming a
simple random sample design and no non-response. A domain is a well-defined population group for which estimates with pre-determined accuracy are sought. The results are then extended to allow
for non-response and deviation from simple random sampling. The application of this principle with the choice of parameters described below leads to a sample size requirement of 18,691 households for measuring annual unemployment with margin of errors of +/-0.3% at 95% confidence level. In these calculations, the main indicator is the ratio of unemployment to the working age population, set at r=0.024 according to the 2012 population census of Rwanda.
The design effect (deff) is set at 3, within the range of values (3 to 4) generally used for labour force surveys. The margin of errors (ME = 0.0026) is twice the standard error of the estimate.17Similar
calculations for employment gave a sample size requirement of 18,792 households. The Labour Force survey programme begun with a pilot survey conducted in February 2016.
The first round of the survey was conducted in August 2016, and continued on bi-annual basis until August 2018. Since the year 2019, the survey was re-designed to provide estimates of the labour
market aggregates on quarterly basis. The sample size in each quarter is about 4608 households, composed of three rotation groups marked with the symbol r in the table. The proposed rotation design is 1-1-1, that is each sample household is interviewed three times, once every two quarters. Accordingly, a sample household is in the sample in one quarter, leaves the sample in the next quarter, returns in the sample in the following quarter, leaves again the sample in the subsequent quarter, and finally returns in the sample for the third time before leaving the sample altogether

Sample rotation

The main purpose of sample rotation is to improve the precision of the trend estimates. It also allows obtaining data on labour force flows by matching sample individuals common in different survey
rounds. The sample rotation scheme adopted for the Rwanda LFS programme builds on the possibility of transiting to quarterly sub-rounds after a few years of half-annual sub-rounds.An implication of these modifications is a slight reduction of the sample size in August 2018 to ensure smooth transition to the sample size requirement of the quarterly design which must be multiple of 48 (=3x16), due to the transition from four to three visits per household and the
maintenance of the sample-take of 16 households per PSU. The sample of the August 2018 survey has been supplemented by considering a sample-take of 24 households per PSU in one-third of the rotation groups 8 and 9 of the previous bi-annual design

A primary sampling unit is an enumeration area of the Population and Housing Census 2012. There are altogether 14,784 enumeration areas in the sampling frame constructed by NISR based on the population and housing census.
To ensure adequate geographical distribution of the sample over the different parts of the population, the sample is allocated among the 30 districts of the country. Prior to sample selection, the sampling frame is sorted according to urban and rural areas within districts. This provides an implicit stratification of the population by urban and rural areas. The resulting distribution of the sample is obtained on the basis of the square-root allocation.It should be mentioned that the square-root allocation rule allocates the sample in proportion to the square-root value of the size of the district. It is a compromise between equal and proportional allocation. Equal allocation gives the same allocation to each district regardless of size. It may be appropriate if each district is to be treated as separate reporting unit (domain) with the same precision requirement as the others. Proportional allocation distributes the sample in proportion to the size of the districts. With proportional allocation, the geographical composition of the population is preserved, but it may lead to very small sample sizes for certain districts. Selection of sample enumeration areas

The next step in sample design was the selection of the 288 sample enumeration areas in each round. The sample was drawn in each district by probability proportional to size (pps) from the sampling frame. The sample selection of PSUs has been carried out on the basis of the PSUs already selected under the bi-annual survey design. Thus, a sample of 144 PSUs have been selected from the 146 PSUs already designated for the August 2018 survey and what would have been the bi-annual February 2019 survey (old rotation groups 8 and 9). These PSUs have been randomly divided into three equal parts of each 46 PSUs, to serve for the new sample PSUs for introduction in the three consecutive
quarters, August 2018, February 2019, and May 2019. Similarly, another sample of 144 PSUs have been selected from the 146 PSUs already designated for what would have been the bi-annual August
2019 and February 2020 survey (old rotation groups 10 and 11). These PSUs have then been randomly divided into three equal parts of each 46 PSUs, to serve for the new sample PSUs for
introduction in the three consecutive quarters, August 2019, November 2019 and February 2020.

The sample enumeration areas were freshly listed prior to selection of the final sample of households. For each quarter, 16 sample households were selected from the list of households in each sample enumeration area by systematic sampling with equal probability. If the list contained 16 households or less all households in the sample enumeration area were drawn in the sample. If the list contained more than 16 households, a sample of fixed size (16 households) was drawn from the list by systematic random sampling. In order to keep almost the same number of households,. The sample size in each quarter was about 4,668 households. The purpose of listing was to ensure that every household currently residing in the sample locality has a non-zero probability of selection. Listing permits to update the sampling frame and account population movements and new household formations that have occurred since the last preparation
of the sampling frame. It aims at covering the newly constructed buildings with living quarters and taking into account demolished or vacant buildings, or transformed dwellings no longer used as living quarters, such as dwelling addresses turned to stores or workshops, or living quarters used as secondary housing units or for holidays. The listing form was developed to capture the information categorized in four columns: (1) A threedigit serial number for listing buildings; (2) Another three-digit serial number for listing dwellings within buildings; (3) The name of the head of household residing in the dwelling; and (4) The street address of the household.



Three steps were involved in the calculation of the sample weights: Calculation of the design weight, Adjustment for non-response; and Calibration to known population projections.
The design weight of a given sample household is the inverse of the probability of selection of a sample household. This probability is calculated as the product of two probabilities.