High Frequency Phone Survey, 2020

The sampling weights were developed for round one of the Solomon Islands high frequency phone survey in a series of steps. As the main shortcoming of using random digit dialing is that the resulting data is representative of the population of mobile phone owners, and according to the most recent data available for mobile phone penetration estimates usage as 74 percent of the population, coverage is concentrated in population centers and better off households and individuals are more likely to have a mobile phone which is charged and turned on. Therefore, the pool of respondents is very different from a representative sample of the Solomon Islands population.

Auxiliary data to serve as inputs to the weights is severely limited as there are few recent nationally representative sources. The results from the recently completed census are not yet available and the last Household Income and Expenditure Survey (HIES) was from 2012/2013. The most recent nationally representative dataset including a measure of welfare was the 2015 Demographic and Health Survey (DHS) and therefore this survey is used as the base for the re-weighting.

Sampling was conducted using random digit dialing with a target sample size of 2,650 respondents. The mobile phone survey sample was designed to mimic the proportions of the 2015 DHS but for a smaller total overall sample. The achieved sample heavily overrepresented the population on Honiara, with a total sample size of 921 for a target of 365, and slightly oversampled Rennell-Bellona, with a total sample of 18 compared to a target of 13. The oversampling in Honiara is most likely attributable to households in Honiara being more likely to have mobile phones that were switched on at the time of the call.

The other provinces were under-sampled to varying degrees, with ratios of achieved-to-targeted samples varying from 40.9 percent in Makira-Ulawa to 87.7 percent in Malaita. Additionally, it was not possible to target between urban and rural areas as that information is not available in a Random Digit Dialing design. Due to the limited sample sizes outside of Honiara, most results are disaggregated into only three geographic regions: Honiara, other urban areas, and rural areas.

Weights are required for unbiased estimation. In addition to the geographic oversampling above, because the survey was administered by mobile phones, the respondents were a representative sample of mobile phone holders, not the population overall, and non-random non-response can exacerbate these differences. Previous literature has shown that mobile phone holders are more likely to be male, urban, wealthier, and more highly educated. To make inferences at the level of the population instead of mobile phone holders, it was necessary to reweight the survey data.

Though it is possible to reweight data to yield unbiased estimates, it is not possible to create additional observations for populations of interest using standard statistical approaches.

Definitionally, the DHS deciles each contain 10 percent of the sample. Using the maximum and minimum threshold values for the DHS deciles to map the mobile phone survey results, it is clear there is a strong bias toward the upper deciles (wealthier) households in the distribution. While weighting can adjust for the bias, there are only 2 and 9 observations in the bottom two deciles of the distribution, respectively. These sample sizes are too small to yield estimates of adequate precision to report results.

Therefore, direct analysis is limited to the bottom four deciles (bottom 40 percent), and then the middle two deciles (middle quintile) and top four deciles (top 40 percent). In addition, each statistic is reported with its confidence interval and all econometric findings are statistically significant, unless otherwise stated.

For more information on weighting, please refer to the "Weighting" section (p.42) of the report provided in the External Resources.

The "weight" variable in the dataset is called "weight".