ÐÓ°ÉÂÛ̳

What

The "Listening to" is a global initiative. An approach involves monthly, nationally representative surveys to monitor dynamic situations, emphasizing wellbeing and public perception. ÐÓ°ÉÂÛ̳ has launched an initiative in the South Caucasus region recently in Armenia and Georgia. The fundamental structure involves conducting an extensive face-to-face baseline survey, followed by monthly phone interviews with a randomly chosen subset of participants from the baseline survey.

Objective

This survey collects detailed information beyond typical household surveys with the aim to cover existing data gaps on health services, migration, employment, social protection, energy consumption and disruption issues, and public perceptions on basic and digital services. This allows for precise analyses and policy studies, measuring not just poverty indicators but also multidimensional poverty, social exclusion, and other indicators critical for assessing current issues in developing countries, particularly in conflict-prone regions. "Listening to" surveys evaluate public perceptions of reforms, monitor policy distributional effects, identify those adversely affected, and help design mitigation measures. In the South Caucasus, the initiative aims to monitor citizen attitudes during ongoing structural reforms, assess the effects on households, and provide early warnings about potential roadblocks or unintended consequences of reform efforts. 

How

"Listening to Armenia" started with a face-to-face nationally representative baseline survey of more than 3,000 households. A two-stage stratified sampling design is employed to guarantee representation across regions, including both urban and rural areas. The first even digital sampling frame in Armenia was used to conduct the survey.    Subsequently around 1,060 respondents are being surveyed monthly starting from June 2025.  The baseline survey focuses on core socio-economic characteristics that change slowly, while monthly phone interviews address variable topics like employment, income, migration, life satisfaction, and public utility services. The surveys include standard questions asked monthly, with flexible modules to accommodate evolving circumstances or specific needs. This approach allows for comprehensive exploration of seasonal changes and in-depth examination of agricultural and migration activities.

Timeline

The baseline face-to-face survey was implemented in October 2024 -January 2025. Monthly phone interviews began in June 2025.

Who

Listening to Armenia is led by the Poverty and Equity South Caucasus team of the ÐÓ°ÉÂÛ̳ in close collaboration with other partners.

Listening to South Caucasus - Baseline Survey in Armenia

(Sampling Design)

Background

Listening to South Caucasus (L2SC) is an expansion of a collaborative effort conducted across Europe and Central Asia. This initiative aims to comprehensively monitor the views and well-being of a representative group of people as the country introduces social and economic reforms, reaching every business and affecting every citizen. By reflecting on the experience of this group over a year, the study provides an up-to-date understanding of how policies reflect on people¡¯s daily lives.

The study comprises a nationally representative baseline survey of households and a high-frequency panel survey of a subset of the baseline survey. The information collected in the L2SC initiative informs reform efforts directly by (1) raising the profile of citizens¡¯ views and (2) enabling in-depth economic analysis.

After completion of the face-to-face baseline survey, interviewers began regularly calling a randomly selected panel of households over the phone to conduct short interviews, following a monthly schedule agreed upon by the participating households. The questionnaire for these phone interviews was designed to monitor trends in income and savings, subjective well-being, public perception, employment, migration, and indicators related to shocks. Phone-based interviews are conducted monthly for a year.

Sample Design

This note briefly describes the sampling design of the baseline survey of L2SC conducted in Armenia, or Listening to Armenia (L2Arm). The sampling design optimizes the spatial allocation of the household sample to provide valid representativeness for the national level, urban and rural areas, and each administrative region.

A two-stage stratified sampling design is employed to select participating households to ensure a balanced sample distribution across regions and account for differences between urban and rural areas, survey budgets, and discrepancies in population estimates. The design closely followed the protocols applied to Living Standards Measurement Study (LSMS) type surveys. In the first stage, a certain number of primary sampling units (PSUs) will be selected in each urban and rural stratum (urban and rural areas within each administrative region or marz). Then, the ultimate sampling units or the secondary sampling units (SSUs), i.e., households in the case of L2Arm, will be randomly selected within each PSU in the second stage. The survey is then implemented among the selected households.

Sampling Frame

The L2Arm employs the national sampling frame based on pre-census enumeration areas (pre-EAs) as the smallest geographic unit. It provides up-to-date information on the geographic distribution of the population across Armenia and allows for a precise formulation of an optimal sample design. In this sampling frame, administrative boundaries based on the 2011 Census for Armenia are obtained from the Humanitarian Data Exchange (HDX) and are strictly respected for constructing the pre-EA boundaries.[1]

[1] The national sampling frame employed for the L2Arm survey was constructed by Ismailakhunova, Purevjav, Byambasuren, and Qader (2025) ¡°¡±, who used an innovative semi-automatic spatial approach to create the national sampling frame based on population estimates from the WorldPop.

Sample Allocation

The objective of any sampling design is to extract the most precision in indicators of interest given survey parameters. Thus, the sampling design seeks to find the most optimal design that efficiently allocates the given PSUs across strata and given households across PSUs. For L2Arm, 400 PSUs are randomly selected across strata¡ªurban and rural areas within each marz¡ªin the first stage of the two-stage sampling design via systematic random sampling with probability proportional to size (PPS, i.e., implicit allocation), with some adjustments. The PPS method systematically assigns each PSU¡¯s likelihood of selection to the percentage of the PSU¡¯s size in the stratum. For L2Arm, the population is used to measure the size due to a lack of data on the number of households at the PSU level. Thus, the PSU¡¯s likelihood of selection is equal to the percentage of the stratum population residing in the PSU. Figure 1 illustrates the distribution of selected or targeted (panel a) and surveyed (panel b) PSUs. While the randomly selected 400 PSUs have been targeted, the survey has been collected from 361 PSUs. Most of the remaining PSUs (particularly, 26 of the 39 PSUs) are from the capital of Yerevan and were uninhabited on the ground.[2]

[2] It is worth noting that the 39 PSUs selected but not covered in the sample are fairly randomly distributed across space and thus likely do not introduce any systematic bias in the collected data.

Figure 1: Spatial distribution of PSUs in Armenia

(a) Target

(b) Actual

Notes: Panel (a) plots the distribution of randomly selected 400 PSUs in Armenia (147 in Yerevan), while panel (b) displays the distribution of 361 PSUs (121 in Yerevan) covered by the survey.

In the second stage, 10 households are randomly selected with equal probability from each chosen PSU. Due to the absence of detailed administrative data and a complete list of households or dwellings for each PSU, a ¡°random walk¡± approach has been used to select the households within each PSU. Table 1 presents the targeted sample allocation in the first and second stages.

Table 1: Baseline sample design based on proportional sample allocation (Target)

To produce an efficient sample allocation, it is necessary to understand the target populations by calculating the intra-cluster correlation and the relative standard error (RSE) for selected indicator(s) of interest, such as per capita consumption. The indicator could be any proxy of a welfare measure. The intra-cluster correlation is a measure of the degree of homogeneity for units of analysis within a given area, i.e., households in a cluster. The level of variation between units of analysis is a key factor in the efficiency and precision of a sample. In general, the more similar the units of analysis are within a given area, the higher the design effects and the higher the error. Increasing the number of units of analysis that are relatively homogeneous in a given area further amplifies the loss of precision. The intra-cluster correlation of an indicator of interest x for the unit of analysis j in cluster c is calculated as:

where ¦Ñ is the intra-cluster correlation coefficient, C is the total number of clusters in the region, m is the total number of units per cluster, and s is the standard deviation of x from the cluster average. The intra-cluster correlation coefficient can subsequently be used as an input to determine the estimated standard error from a given sample design. We can determine the error of a two-stage sample design as:

where (1+¦Ñ(³¾-1)) is the design effects in a clustered sampling design, which include both the cluster size  and intra-cluster correlation . The existing data, ideally a survey sample with a similar design that employs the same national sampling frame, must be used to reveal the nature of the intra-cluster correlation and analyze the design effect and RSE. Since it is the first time the sampling frame employed for the L2Arm is being used in Armenia, no existing survey data is leveraging the sampling frame. So, we could not perform the analysis before conducting the L2Arm baseline survey. However, given the successful completion of the L2Arm baseline survey, we check the sample representativeness and examine the sample allocation using the collected baseline data.

Since the sample has already been allocated across the 21 strata for the L2Arm baseline, we calculate the RSE for the chosen sample allocation to show the representativeness. Table 2 presents the representativeness of the L2Arm baseline sample in terms of RSE at the national, urban/rural, marz, and stratum levels. The baseline sample size is 3024 households, smaller than the targeted size of 4000 households, primarily due to uninhabited PSUs and high non-response rates in urban areas. For the inhabited areas, despite the random walk procedure in the second stage, households in the selected PSUs are exhausted before reaching the target of 10 households in the PSU because of the non-responses.

The design effect (DEFF) is typically 1 for simple random sampling, indicating that sampling with a unit design effect is close to random sampling. Given that the calculated DEFF is approximately 1 at all levels, the L2Arm baseline sample is almost as good as random. The sample allocation is evaluated based on the resulting RSE, which is the standard error of the estimate relative to the mean and allows for the comparison of precision. Generally, the lower the RSE, the better the precision, and a 10 percent maximum is commonly recognized as the maximum acceptable RSE. The L2Arm baseline sample is thus representative at the national, urban/rural, and regional levels and is not representative at the stratum level.

Table 2: Actual sample allocation, design effect, and representativeness

Weights

Sampling weights are necessary when computing representative statistics, as the weights account for the fact that different population members have different probabilities of being selected for interviews. Sampling weights are also adjusted, accounting for non-response rates given the survey design, if necessary.

There will be two sets of weights in the dataset:

1. Household weights
2. Individual weights

Household weights are the inverse probability of selection of households and are calculated from the following two components in our two-stage sampling design:

1. Sampling weight (inverse probability of selection) of the PSU within the stratum,
2. Sampling weight (inverse probability of selection) of the selection of a household within the PSU.

For calculating individual weights, we add the following third component to the calculation of the household weights:

       3. Sampling weights (inverse probability of selection) of individuals within the household.

Each of the components is calculated as follows:

Component 1. The inverse probability of selection of PSU within the stratum by using PPS is calculated as:

where

W_psu is the sampling weight of PSU within the stratum,

P_psu is the probability of selection of PSU within the stratum,

n_psu in the number of selected PSUs within the stratum,

N_psu is the size of the selected PSU,

N_stratum is the size of the stratum.

Component 2. The inverse probability of selection of a household within PSU is calculated as:

W_(hh_psu ) is the sampling weight of the household within PSU,

P_(hh_psu ) is the probability of selection of household within PSU,

n_(hh_psu ) is the number of sampled (interviewed) households within PSU,

N_(hh_psu ) is the total number of households within PSU.

Component 3. The inverse probability of selection of individual within the household is calculated by:

W_(ind_hh ) is the sampling weight of the individual within the household,

P_(ind_hh ) is the probability of selection of individual within the household,

n_(ind_hh ) is the number of sampled (interviewed) individuals within the household, equal to 1, as only one individual was allowed to be interviewed from each household,

N_(ind_hh ) is the size of the household surveyed (asked and recorded during the interview).

Based on these components, household and individual weights are calculated as:

Conclusion

This note briefly describes the sampling method used for the baseline survey of the Listening to South Caucasus (L2SC) in Armenia. The design optimally allocates the sample to satisfy an acceptable precision and statistical efficiency. Using the proposed stratified two-stage sampling design stands to provide valid survey estimates. It depends on the accuracy of the sample design and the selection of PSUs and households at every phase. Identification of the sample design, including stratum identification, PSU identification, household identification, and survey weights, is included in the final datasets.

Listening to Armenia:

Listening to Armenia:

Listening to Armenia:

Listening to Armenia:

Listening to Armenia: