The aim of the WMS 2013 was to interview the same ‘well-informed respondent’ in each household who had participated in the 2009 and 2011 surveys. The longitudinal dataset enables analysis of changes in household and personal circumstances over a four-year period. Fieldwork began on 23 July 2013 and finished on 15 August 2013. It was carried out by 122 interviewers, with regional supervisors, all across Georgia.

Sampling

The sampling strategy targeted the 4,808 households [1], in which face-to-face interviews had been completed in 2009. Successful interviews were held with respondents from 4,147 households, an 86 per cent response rate in 2011. In the third round, 3,726 questionnaires were completed, constituting an 89.8 per cent response rate of 2011 sample.

Table 2.1: Survey response rates by region in 2013.

Of the 421 non-responses, only 48 were due to refusals to participate. But in the majority of cases, respondents had moved away, either temporarily or permanently.

Data weighting

A household weighting variable was provided by the survey research company for use with the 4,147 households in the sample in 2011. This was based on the weights used in the 2009 survey, as stratified by region and type of location. Table 2.2 shows that the geographical distribution of the population of Georgia changed little between 2009 and 2013, so the new 2013 weights only adjust for changes in sample size due to non-response. Separately individual weights are used on questions addressing individual household members. 

Table 2.2: Distribution of population by region at the beginning of 2009, 2011 and 2013.

Source: http://www.geostat.ge/.

Comparability with previous study

The achieved survey sample spanning 2009, 2011 and 2013 was 3,726 households. Examination of the 2013 survey data shows that 3,586 households completed a full questionnaire in all three survey years (Table 2.3).

Table 2.3: Number of surveyed households in 2013.

Seventy-five households in the 2013 sample were interviewed in 2009, but were not correctly identified in 2011. Sixty-five households were not linked with either 2009 or 2011 surveys, thus were not included in the panel. These 140 households are still included in the calculations of 2013 data. The separate weighting is used for 3,586 true panel households for the analyses tracing the trajectories of individual households across time.

Adjusting for inflation

A sustained increase in the general price level is measured by the consumer price index (CPI), based on the cost of a typical basket of consumer goods and services in a particular year. The National Statistics Office of Georgia on the website of the National Bank of Georgia (http://www.nbg.ge/index.php?m=306#monetarystatistics) gives a CPI for the months in which fieldwork was completed: 129.6 for July 2009, 152.1 for August 2011 and 151.1 for August 2013 [2]. For comparing changes between the three pillars, monetary data for 2013 are converted to 2009 prices by dividing by 151.1 and multiplying by 129.6.

Income and expenditure per adult equivalent

As in the 2011 analysis, we use measures of income and expenditure to compare households of different sizes and compositions. Where relevant, some statistics are adjusted to GEL per equivalent adult (PAE), according to methods used by the Georgia Department of Statistics. First, household members are classified by age and gender, and assigned an equivalent adult coefficient (Table 2.4). The sum of these coefficients represents the number of equivalent adults in the household. To correct for economies of scale in larger households, the number of equivalent adults is then raised to the power a, where a=1 for a single person household, and a=0.8 where household size is greater than one.

Table 2.4: The scale used to calculate number of equivalent adults in a household.

Statistical significance

In sample surveys, we do not have information concerning the whole population. Instead, we make inferences about the population from statistical tests carried out on the sample. p represents the probability that the relationships we observe in the sample only occur by chance (rather than from real effects within the whole population). The smaller the value of p, the greater our confidence that patterns observed in the sample hold true for the larger population. Probabilities of less than 0.05 mean we can be at least 95 per cent confident. This level is conventionally used to indicate statistical significance.

In the tables that follow, the conventions used to represent probability values are:

* means p < 0.05 (95 per cent confidence)

** means p < 0.01 (99 per cent confidence)

*** means p < 0.001 (99.9 per cent confidence)

ns means Not statistically significant (less than 95 per cent confidence)

[1] - The 2009 survey included a separate diary to collect consumption information. This was completed by only 4,646 of 4,808 respondents. In 2011, the collection of consumption data was incorporated in the main questionnaire completed by all respondents.

[2] - CPI is indexed to 2005, so 2005 = 100.0.