# Rwanda - Integrated Household Living Conditions Survey (EICV5), 2016-2017, Cross-Sectional Sample

Reference ID | RWA-NISR-EICV5-CS-2016-2017-V0.1 |

Year | 2016 - 2017 |

Country | Rwanda |

Producer(s) | National Institute of Statistics Rwanda (NISR) - Ministry of Finance and Economic Planning |

Sponsor(s) | Government of Rwanda - GoR - Financial Partner World Bank - WB - Financial Partner UKaid - Ukaid - Financial Partner European Union - EU - Financial Partner One UN - One UN - Financial Partner |

Metadata | Documentation in PDF |

Created on

Dec 11, 2018

Last modified

Dec 16, 2018

Page views

479348

Sampling

Sampling Procedure

The sampling frame for the EICV5 cross-sectional survey is based on the NISR master Sample data. More recently the NISR used the 2012 Census frame to select a large Master Sample of villages 3,960 that can be used for the different national household surveys in Rwanda. The primary sampling units (PSUs) for the Master Sample are individual villages, or a combination of small villages, with the number of households tabulated from the 2012 Census data. A new listing of households was conducted in order to update the frame for the EICV5 cross-sectional survey. The sample households in the EICV5 sample villages were selected from the new listing.

1) The EICV5 Cross-sectional survey sample size

The sample size for the EICV5 cross-sectional survey depends on the level of precision that is required for key indicators at the district level, as well as on resource constraints and logistical considerations. It is very important to ensure good quality control in order to minimize the nonsampling errors. The estimates of the sampling errors for the poverty rate by district from the EICV4 data were examined in order to determine whether it would be necessary to adjust the sample size.

For EIVC4 the number of households selected per cluster was 9 for Kigali Province, which is mostly urban, and 12 for the remaining provinces, which are mostly rural. This sampling strategy has been consistent for all the EICV surveys because it is statistically efficient and is also effective for the EICV logistics of the fieldwork and the workload of the team of enumerators each cycle. The urban areas generally have a higher intraclass correlation for socioeconomic characteristics between households within a cluster compared to rural areas. There is also a different interviewing schedule for the sample households in Kigali Province, so only 9 households are interviewed in each cluster. In terms of the number of sample clusters allocated to each district, it should be a multiple of 10 so that the sample can be evenly distributed to the 10 cycles. In the case of EICV4 the districts in Kigali Province were assigned 5 sample clusters each month, and in the other provinces each district was assigned 4 sample clusters each month.

In EICV5 the sample was increased for the districts in Kigali Province because the estimates of the poverty rate for those districts had higher coefficients of variation (CVs) or relative standard errors (RSEs) compared to the other districts. However, one reason why the RSEs for the districts of Kigali Province were higher is that the value of the poverty rate is lower for these districts. It was pointed out that in the case of estimates of percentages or proportions, it is more effective to use the margin of error to study the sample size. The margin of error is equal to half of the width of the 95% confidence interval, or 1.96 times the standard error. Therefore the margins of error for the estimates of the poverty rate by district were also examined. In this case the margins of error were also higher for the districts of Kigali Province, given the relatively higher design effects (especially for Gasabo District), and considering that the number of sample households for these districts in EICV4 was only 450, compared to 480 sample households in the districts of the other provinces. For these reasons, it was decided to increase the number of sample PSUs for each district in Kigali Province from 50 to 60, for a total increase of 30 sample clusters and 270 sample households. For the districts in the other provinces it was decided to have the same sample size of 40 clusters and 480 households each cycle, since the level of precision of the EICV4 results for these districts was considered satisfactory.

The sample PSUs in each district were allocated to the urban and rural strata proportionately to the number of households in the 2012 Census frame. In the case of districts where the proportional number of sample PSUs was only 1 for the urban stratum, the number of sample PSUs was increased to 2. For the selection of sample villages for EICV5, it was assumed that the Master Sample villages for each district were explicitly stratified by urban and rural areas. A separate subsample of villages was selected within each stratum from the Master Sample.

At the national level there are 1,260 sample villages and 14,580 sample households. In the urban strata there are 245 sample villages and 2,526 sample households, and in the rural strata there are 1,015 sample villages and 12,054 sample households. The sample size for the EICV5 cross-sectional survey has 30 more sample PSUs and 270 more sample households than the corresponding sample for EICV4.

In the case of EICV4 the national sample of 177 villages selected from EICV3 for the Panel Survey were also used as part of the EICV4 cross-sectional survey. However, for EICV5 it was decided to select a completely separate sample of villages for the cross-sectional survey.

2) Assignment of sample villages to cycles and sub-cycles

Similar to the EICV4 methodology, a nationally-representative sample of clusters will be assigned for the EICV5 data collection each cycle, so that the sample is geographically representative over time. A subsample serial number from 1 to 10 can be assigned systematically to the geographically ordered list of all sample clusters in each district. In order to assign the cycles to the EICV5 cross-sectional sample villages, random cycle numbers from 1 to 10 were generated to identify the selection sequence. For the 27 districts outside of Kigali Province, the sub-cycle numbers of 1 or 2 were assigned systematically with a random start. This process ensured that the final distribution of the sample clusters to cycles and sub-cycles was geographically representative within each district.

Response Rate

The response rate for EICV5 (cross-sectional) is 100%. All households sampled(14,580) were interviewed with no refusal.

Weighting

The EICV5 Cross-Sectional Survey is designed to represent the current household-based population in each of the 30 districts of Rwanda. A stratified multi-stage sample was selected for the EICV5 Cross-Sectional Survey, based on the NISR Master Sample selected from the 2012 Rwanda Census frame. For the data collection a reserve random sample of 3 households per cluster (in Kigali Province) or 4 households per cluster (in the other provinces) was selected for possible replacements. It was possible to replace all non-interview households to complete the target number of 9 household interviews in each sample cluster of Kigali Province, and 12 household interviews for each sample cluster in the other provinces. The weights for the EICV5 Cross-Sectional Survey were calculated based on the probabilities of selection from each sampling stage. This will ensure that the data will be weighted up to represent the total household-based population of Rwanda. Generally it is necessary to adjust the weights to take into account the non-interview households in each sample village. However, if all of the non-interview households are replaced during the EICV5 Cross-Sectional Survey data collection in each sample village, the final number of completed households will be exactly 9 for each sample village in Kigali Province, and 12 for each sample village in the remaining provinces. In this case there is no need to adjust the EICV5 Cross-Sectional Survey weights for non-response.