Sampling Form

National samples for face-to-face surveys are typically drawn by an area probability method, which relies on US Census figures and maps of the country. For example, the 2016 ANES used a multi-stage process to draw its national sample. This process is called a multistage stratified cluster sample, and has a long history of use in face-to-face surveys. The stages were:

  1. Selecting primary sampling units (PSUs) from the 48 contiguous states and Washington, DC. (For the face-to-face survey, cost considerations precluded interviewing respondents in Alaska and Hawaii. Respondents in those states were, however, included in the Internet sample.) The PSU’s were counties or portions of counties (Los Angeles County was divided into two PSU’s because of its large area). The PSU’s were then listed by what Census region they were in, the prevalence of poverty in the county, size of minority population, and population size. Sixty PSU’s were then chosen from this list; the probability of being chosen was based on the number of adult citizens in the PSU, so larger population PSU’s had a higher chance of being selected and smaller ones had a lower chance. Cook County, IL (Chicago area), Harris County, TX (Houston area), Maricopa County, AZ (Phoenix area) and both halves of the Los Angeles County PSU were included in the sample of PSU’s.
  2. Four secondary sampling units were then identified in each PSU.
  3. Within each secondary sampling unit, individual households were randomly chosen using the US Post Office’s computerized delivery sequence file (DSF). Each of the selected household received a study brochure and $5 in cash.
  4. An eligible person to be interviewed in the each household was then randomly identified by an interviewer who visited the household.

Interviewees received either $25 or $50 as incentives to complete the interview. As the study went on, this was increased to $100 for non-respondents.

Sampling for telephone surveys is fairly simple as national lists of telephone area codes, exchanges, and even individual numbers exist. One commonly used method begins with all the area codes in the US and then identifies all the exchanges within each area code. After this is done, a computer is programmed to dial a four-digit random sequence of numbers added to each combination of area code and telephone exchange. The actual number of respondents in any code is determined by the actual number of telephone numbers assigned in the geographic area for which the area code is used. This technique is called random digit dialing (RDD). It has advantages over using a telephone book to identify a sample—people with unlisted telephone numbers might be contacted, and there is no telephone book listing of cell phone numbers. A disadvantage is, however, that large numbers of telephone numbers in any area code/exchange combination might be unassigned or are business numbers rather than home numbers. Often five calls must be made to get one working residential number.

RDD can be used to include cell phones as well as land lines in the sample, however it is illegal to use automatic dialers to contact cell phone users without their express permission. Many survey houses do not call cell phones for a variety of reasons—many people do not want cell phone calls from unidentified callers; the cell phone owner may be driving, and so on. Cell phone-only households (pdf) are, however, on the rise in the US, so the problem of contacting people for surveys over the phone is growing. As of 2016, more than half of US households (50.8 percent) were cell phone-only. Much higher percentages of people under the age of 30 and poorer people lived in this type of household. Some 62 percent of those aged 18 to 24 and 73 percent of those aged 25 to 29 lived in cell phone only households; the percentage of cell phone only households then drops significantly for older householders. Nearly two-thirds (66.3 percent) of people living in poverty live in cell phone only households as compared to 48.5 percent of non-poverty households. Cell phone-only households also varied widely by state, from 21 percent in New Jersey to nearly 54 percent in Arkansas (pdf).

Most survey organizations modify the pure random digit dialing technique to reduce the number of useless phone calls. One widely used method that dramatically increases the odds of getting a working residential phone number is the Waksberg method.

For many years, sampling for Internet-based surveys involved identifying a population of Internet users who would volunteer (often they are provided incentives for completing surveys) and then taking a sample of this population. More recently, however, the percentage of the U.S. population who use the Internet has grown to such a level that survey organizations can skip the step of identifying Internet users. So in 2016, ANES worked with Marketing Systems Group to develop the Internet survey. This was done by sending a series of letters to randomly selected addresses drawn from the Post Office’s Delivery Sequence File, inviting one household member to participate in the Internet survey. The invitation included $20 in cash; an additional $40 in cash was promised for completing the online survey. This was increased to $80 for those who initially did not respond.