Caucasus Barometer, 2009

Data Archive > International Surveys and Data > Multiple Nation Surveys > Summary

The Caucasus Barometer is an annual nationwide survey conducted by the Caucasus Research Resource Centers (CRRC) in Georgia, Armenia, and Azerbaijan. The target population for the 2009 Caucasus Barometer was all non-foreign adults residing in Armenia, Azerbaijan, and Georgia outside of occupied territories and the Nakhchivan Autonomous Republic of Azerbaijan during the period of September-October 2009. The Caucasus Barometer was designed in 2003 in order to collect reliable representative data on a wide range of social, political, and economic attitudes of the population of the South Caucasus, as well as information on household composition and household economic behavior. From the very beginning, the data collected by CRRC was meant to be open to all interested researchers and/or policymakers both from the region and from other parts of the world. For more information, visit the CRRC website.

Religion variables include religious preference, the extent to which religious beliefs help in decision-making, frequency of fasting, and frequency of attendance at religious services.

Data File
Cases: 5,761
Variables: 438
Weight Variable: WWIND, WWHH
Data Collection
Date Collected: September-October 2009
Original Survey (Instrument)
CRRC Website
Funded By
CRRC is a program of the Eurasia Partnership Foundation and is an network of resource, research, and training centers established in 2003 in the South Caucasus with the goal of strengthening social science research and public policy analysis in the region. The program is supported by Carnegie Corporation of New York.
Collection Procedures
In Georgia and Armenia, sampling frames of all households within sampled PSUs were constructed by surveyors (“blocklisting”), and households were selected from those sampling frames via SRS. In Azerbaijan, due to budgetary constraints, households were selected via random route sampling. Random walks were begun at precinct polling stations and conducted with step sizes of nine outside of buildings of less than five stories and 14 within buildings of five or more stories. Within selected households, the adult household member targeted for interview was randomly selected using the Kish table (Kish 1949). In the event that contact was not made with the household or that the selected household member was not available, a total of three contact attempts were made before classifying the attempted interview as non-response.
Sampling Procedures
Primary sampling units (PSUs) were electoral precincts and the sampling frame was divided into three “macro-strata”: capital, urban, and rural. Necessary sample size calculations were made in two stages. First, the design effect (DEFF)—the factor by which the width of the sampling distribution is inflated due to the structure of the sample—was calculated based on data from questions on sex, age, education, employment status, and health in the 2008 Caucasus Barometer (CRRC 2008) which had the same survey structure as the 2009 Caucasus Barometer. Then, a base sample of 250 respondents in each macro-stratum was multiplied by the corresponding DEFF in order to compute the sample sizes equivalent to simple random samples (SRSs) of 250 respondents.

The urban and rural macro-strata were each further divided into geographic sub-strata: northeast, northwest, southeast, and southwest. The capitals were not sub-divided in any way, resulting in a total of nine strata per country. The number of PSUs to be sampled in each urban and rural macro-stratum was allocated amongst the four geographic strata in proportion to the total population size of each stratum.
Principal Investigators
Tinatin Zurabishvili
Citing the Data
Please use the following citation when citing data from the Caucasus Barometer:

Caucasus Research Resource Centers. (dataset year) "Caucasus Barometer". [dataset] Retrieved from on {date of accessing the database}.

CRRC requests that those who cite CRRC data to notify them by completing this form.
Interview Disposition Codes

For all fully and partially completed interviews, the proportion of applicable questions that the respondent answered was calculated. Disposition codes were assigned according to AAPOR standards, so that interviews wherein the respondent answered fewer than 50% of the applicable questions were assigned a disposition code of zero, those wherein the respondent answered at least 50% but fewer than 80% of the applicable questions were assigned a disposition code of 0.5, and those wherein the respondent answered at least 80% of the questions applicable to him or her were assigned a disposition code of one. Interviews with a disposition code of zero were classified as non-response and were not assigned sampling weights.

Household Weights

Population weights for households were calculated as the inverse of selection probability, so that each household’s weight is equivalent to the number of households that it represents in the entire population of households in the country. Although precincts were selected without replacement, selection probabilities were calculated as though they had been selected with replacement. Because each precinct comprises such a small proportion of the total population of each stratum, the resulting selection probabilities are not different in any meaningful way, and the process of computing weights is much more computationally efficient.
Description of Flag Variables
FLAGM1M2: In variables INFSOU1/INFSOU2, newspapers were mentioned as the first of the second source of information, while in READPAP the respondent doesn't read newspapers at least once a week.

FLAGM1M3: In variables INFSOU1/INFSOU2, TV was mentioned as the first source of information, while in WATCHTV the respondent doesn't watch at least one news program per day.

FLAGM1M4: In variables INFSOU1/INFSOU2, TV was mentioned as the first source of information, while in RECEPTV the respondent has reception of 0 channels.

FLAGM1M5: In variables INFSOU1/INFSOU2, TV was mentioned as the first source of information, while in NUMCHAN the respondent prefers watching 0 channels for news.

FLAGC12: No source of income was mentioned by the respondent.

FLAGC4C8: In GETARND, a personal car is mentioned as a mean of daily transportation, while in OWNCARS the respondent doesn't own a car.

FLAGN12: In FAMCLAB and CLFRDAB, the respondent has no friend or relative living abroad while in INCRARM one of the sources of income is money transfers from relatives living abroad.

FLAGD3A6: Years of education in EDUYRS doesn't match with the respondent's education level in RESPEDU.

FLAGC84C: Automatic washing machine was purchased before 1990.

FLAGC87C: Mobile phone was purchased in 1990.

FLAG_A8: The respondent has a current job longer than he's been working since 1992, has been unemployed since 1992, could not work since 1992, has been unemployed since after 1992, or could only work for specific years since 1992.