National Household Forest Survey 2018-2019

Given the focus of the NHFS on the population living in close proximity to forests4, a first step was to clearly define forest for the purposes of the survey. Building on the national definition of forest used in Liberia, and modifying it in order to minimize the impact of small urban forests and facilitate survey operations, the NHFS employed the following definition:

Forest = area with at least 30 percent tree canopy cover, with trees higher than 5 meters and at least 50 hectares in size

The forest cover was determined using high-resolution forest cover data produced in 2019 based on satellite information on forest cover in Liberia for 2015.6 All EAs within 2.5 kilometers of forests identified with this definition were deemed eligible for inclusion in the NHFS.7 EAs from the Montserrado county (part of Greater Monrovia) were excluded from the sample universe due to the high rate of urbanization. However, rural parts of Montserrado county were included in the sample universe.

Based on the forest definition defined above, the distance from each EA in the country (except urban Montserrado) to the nearest forest was computed. That distance was subsequently used to assign each EA to one of the following strata: S1 (less than 2km from forest); S2 (two to 7 km from forest); S3 (7 to 15 km from forest).

Following strata classification, a total of 250 EAs were selected through a Probability Proportional to Size (PPS) sampling approach within each stratum, with the following purposeful allocation across strata: 90 EAs in S1; 90 EAs in S2; 70 EAs in S3.8 The measure of size for each EA was based on the total number of households listed in the 2008 PHC.

Following the selection of the 250 sample EAs, a listing of households was conducted in each sample EA to provide the sampling frame for the second stage selection of households. Random sampling was used to select 12 households from the household listing for each sample EA.

The original sample design provided a total household sample size of 3,000 (250 EAs with 12 households sampled per EA), data from 14 households are missing or unusable, representing 0.05 percent of the sample and resulting in a final sample of 2,986 households. Similarly, data from 5 of the community questionnaires were missing or unusable, resulting in a total sample of 245 community questionnaires. The final sample of 2,986 households is distributed across counties.

Upon post-data collection analysis, it was discovered that the initial variable that was used to stratify EAs by distance to forest was incorrectly computed. Despite thorough attempts to understand the nature and source of the error, it was determined that a mechanical error must have occurred during the process of the distance calculations. This error rendered the stratification incorrect. Therefore, the stratification by distance to forest has been abandoned and the sample weighted to reflect only geographic clusters, not distance to forest. This was determined to be the most appropriate way forward following consultation with sampling experts.

The resulting sample, therefore, is weighted to reflect all EAs in Liberia (with the exception of urban Montserrado) that fall within 2.5 km of the nearest forest, which was the upper bound of the distances for the selected EAs.