Abt SRBI conducted the survey of likely voters in Georgia on behalf of the Atlanta Journal Constitution. The survey included telephone interviews with a representative sample of 2,003 Georgia residents. Telephone interviews were conducted by landline (n=1,201) and cell phone (n=802, including 481 without a landline phone or “cell-only”). The final sample produced n=1,170 likely voters statewide. Interviewing was conducted from October 16 to October 23, 2014.
Sampling
The sample design was a random digit dialed sample of cell phone numbers and landline numbers with a Georgia telephone exchange. This sample design is referred to as a “dual-frame” because it includes cell phones and landlines.
The landline frame is constructed by compiling all Georgia telephone exchanges that are classified as providing regular telephone service. The frame is referred to as “list-assisted” because a complete file of directory-listed residential numbers is used to remove 100-banks from the frame if they contain zero residential listings. The remaining 100-banks are “working” and used to enumerate all the telephone numbers within the bank from which a sample is drawn. All landline numbers (directory-listed and unlisted) in the working banks are eligible to be randomly dialed. Telephone numbers known to belong to businesses are removed.
The cellular telephone frame begins with 1,000-blocks constructed from exchanges that provide cellular telephone service. The frame of 1,000-blocks is then expanded to the 100-block level to identify and remove "mixed use" 100-blocks, or those that include landline numbers. The result is a sampling of cellular 100-blocks that is mutually exclusive of the list-assisted RDD sampling frame described above. After the initial cellular sample was drawn, numbers flagged by the sample vendor as "inactive" using Cell-WINS[1] were removed and not dialed.
For the landline sample, interviewers were asked to speak with the youngest adult male or female currently at home based on a random rotation. If no male/female was available, interviewers asked to speak with the youngest adult of the other gender. For the cell sample, interviews were conducted with the person who answered the phone. Interviewers verified that the person was an adult and in a safe place before administering the survey.
All cooperating respondents from both samples were asked about their voter registration status. Registered voters were further asked a series of likely voter questions including intention to vote in the November general election, attention to the race, and history of voting. People who said they were not registered to vote or were registered but determined to be unlikely to vote in November election were only asked demographic questions for weighting purposes.
Weighting
The final weights produced for this survey accounted for the dual-frame sample design and aligned the sample to match the population parameters of the adult population in Georgia. To construct the weights, we used the full sample of 2,003 Georgia residents. The full sample was post-stratified (raked) to benchmark demographic distributions for the Georgia adult population, as described below. The benefit of this approach is that statewide benchmarks for all adults are available from the Census Bureau and highly accurate and reliable.[2]
The first stage of weighting corrected for different probabilities of selection associated with the number of adults in the household and the respondent’s telephone usage (landline only, cell phone only or has both kinds of phones). This weighting also adjusts for the overlapping landline and cell sample frames and the relative sizes of each frame and each sample.
The second stage of weighting balanced sample demographics to estimated adult population parameters for the state of Georgia. The sample was balanced to match population parameters for sex, age, education level, race/Hispanic ethnicity, region (North, Atlanta Metro, Atlanta Exurbs, Southeast, Southwest), and telephone usage (cell-only, dual-user, landline-only). The demographic population parameters were computed from the 2013 American Community Survey (ACS). The population parameter for region of state was obtained from the 2013 Census Population Estimates. The telephone usage population estimates were constructed from the model-based estimates for Georgia released by the National Center for Health Statistics for the year 2012.[3] Since the cell phone-only adult population has increased every year since 2012, the state-level estimate was updated to reflect national trends according to the 2014 NCHS report.[4]
The second-stage weighting was conducted using an operation known as raking ratio estimation, or “raking”. Raking is used to reduce the risk of biases due to nonresponse and non-coverage in sample surveys. The raking procedure uses an iterative technique that simultaneously calibrates the sample to population distributions defined by socio-demographic parameters. After the raked weights were generated, we examined the distribution of values.
Margin of Error
The margin of error for an estimate is a measure of uncertainty that reflects the fact that the estimate is derived from a sample drawn from the population. If one were to draw a second sample in the exact same manner, the estimate would be different from the first simply due to the fact that the sample contains different members of the population. A third sample would be different from the first two, and so on. The margin of error measures how different estimates could be based on drawing different samples from the same population.
The error margin for the sample of 1,170 likely voters is +/-3.6 percentage points. This includes a “design effect” of 1.59. The design effect is the amount of variability introduced by the sample design, such as the dual-frame sample and weighting.
[1] A description of this service is available here:
http://www.m-s-g.com/CMS/ServerGallery/MSGWebNew/Documents/GENESYS/whitepapers/Cell-WINS.pdf
[2] Polls that take the alternate approach of weighting just the weighting the likely voter sample to previous turnout numbers run the risk of using benchmarks that are inaccurate as the demographic profile of who turns out to vote varies from election to election.
[3] Blumberg SJ, Ganesh N, Luke JV, Gonzales G. Wireless substitution: State-level estimates from the National Health Interview Survey, 2012. National health statistics reports; no 70. Hyattsville, MD: National Center for Health Statistics. 2013.
[4] Blumberg SJ, Luke JV. Wireless substitution: Early release of estimates from the National Health Interview Survey, July-December 2013. National Center for Health Statistics. July 2014.
