Multi-Tier Framework Survey for Measuring Energy Access 2017-2018

A. SAMPLE SIZE CALCULATION PARAMETERS

The sample size proposed for Zambia is designed to get sufficiently precise estimates of each tier at national as well as urban and rural level. A much smaller sample size would have been adequate to produce precise estimates at the national level within those domains. This section discusses the factors that should be taken into consideration in the determination of sample size calculation and provides a justification for the proposed sample size for each country. The major issues considered in determining the appropriate sample size for a survey are:

1. The precision of the survey estimates

The concept of the precision of a sample survey estimate is crucial in determining the sample size. By definition, a sample from a population is not a complete picture of the population. However, an appropriately drawn random sample of reasonable size can provide a clear picture of the characteristics of that population, certainly sufficient for policy implication or decision-making purposes. From a sample of households, one can collect data and generate a sample (or survey) estimate of a population parameter. The population parameter value of a characteristics of interest is generally unknown.

2. The quality of the data (Non-sampling error)

Besides sampling errors, data from a household survey are vulnerable to other inaccuracies from causes as diverse as refusals, respondent fatigue, measurement errors, interviewer errors, or the lack of an adequate sample frame. These are collectively known as non-sampling errors. Non-sampling errors are harder to predict and quantify than sampling errors, but it is well accepted that good planning, management, and supervision of field operations are the most effective ways to keep them under control. Moreover, it is likely that management and supervision will be more difficult for larger samples than for smaller ones (Grosh and Muñoz 1996, p. 56). Thus, one would expect non-sampling error to increase with sample size and we would like to limit the sample size to less than 5,000.

3. The cost of data collection, processing, and dissemination.

The sample size can affect the cost of the survey implementation dramatically. It will also affect the time in which the data can be collected, processed and made available for analysis. The availability of survey firm and cost for each country would affect the total cost of survey implementation, too. Thus, the cost of data collection, processing, and dissemination should be considered in determining the sample size for each country.

B. SAMPLING APPROACH

In this study, a stratified random sampling technique is used. The first stratification involves stratifying into urban and rural strata. The second stratification is based on the electrification status of the enumeration areas (EAs) in the study population.

• Urban and Rural stratification

The primary sampling units (PSUs) in this study are EAs, selected randomly from the list of EAs in Zambia obtained from CSO Zambia. The EAs were stratified into rural and urban strata. For each stratum, random numbers were allocated to each EA and these EAs were arranged in ascending order. The first EAs to satisfy the sample quota of each province were picked. The number of EAs picked in each province for either rural or urban stratum were dependent on the sample size of each province. A total of 14 households were sampled in each EA, so the sample size of each of the province was divided by 14 to get the total number of EAs to be sampled. An equal split of the sample between rural and urban stratum was done at the national level.

• Electrified or non-electrified stratification

Listing was conducted only in the sampled EAs to determine whether to classify an EA into either electrified or non-electrified stratum. EAs with at least 3% of households that were connected to the national grid were classified as electrified while those with less than 3% of households connected to the national grid were classified as non-electrified. A 50-50 ratio of distribution of sample between grid and non-grid users was achieved.

• Household selection

During the listing process, information on electricity connection (the number of households with or without electricity in a sampled EA) was collected. Random numbers were allocated to each household and arranged in ascending order for each stratum.

Of the original sample size of 3,668 targeted households in 262 EAs (130 EAs in urban and 132 EAs in rural areas), 3,612 households in 260 EAs were contacted, and 3,537 in 260 EAs were effectively interviewed. The response rate is thus 96%, which is the difference between the sample of households originally targeted and those finally interviewed. As explained in paragraph 4, the non-response was mainly due to movement out of the dwelling of respondents (43 households) and unwillingness to participate in the survey.