About This Winthrop/ETV Poll

This Winthrop/ETV Poll reports responses from 722 randomly sampled South Carolinians over the age of 18.  This results in a margin of error of +/-3.65 percent. 

This survey was in the field from Feb. 2 to Feb. 17, 2008, however no calls were placed on Sunday, Feb. 3 (Super Bowl Sunday).

 

Survey Methodology FAQs

Q: How do you select phone numbers to call?

A: That depends on the population we wish to survey. For example, if the population under study is “Registered Voters in South Carolina,” we purchase the names and numbers of registered voters and randomly sample from that list. Of all the sampling procedures we might use, this is the most straightforward. Of course, we could get very specific and only sample registered voters who voted in the last election or some particular primary, but in general, the sample is drawn from the full population of registered voters.  An exception is if we are screening for likely voters for some upcoming election.”

If we are sampling some specific subgroup, such as a particular ethnicity, the method is somewhat more complex. We purchase a sample from a company that uses several methods for determining an ethnic, or race-based sample. Relying on extensive research, a list of probable matches based on first name and surname match is created. This list is supplemented with other external data indicating a particular race within the household. This data may be drawn from direct marketing sources, previous survey responses and/or subscriber data. This supplemental external data is provided by a third party vendor. Some of their methodologies are considered proprietary. However, in order to get as broad a range of respondents as possible, we complement the match sample (previously described) with an enhanced RDD methodology (a full explanation of Random Digit Dialed [RDD] methodology may be found below). Using various sources, including, but not limited to, census data, neighborhoods or communities are targeted based on whether they meet a certain threshold for minority population. We typically use a threshold of forty percent. If a neighborhood or community is known to have at least forty percent of our target population (e.g. African American) then a RDD sample is created for that neighborhood. We believe the matching methodology and the targeted RDD method complement each other well, ensuring that we are getting an accurate glimpse of our target population.

If the population of interest is the general population of adults in South Carolina, the sampling methodology is a bit more complex still. We can’t simply pick numbers from the phone directory because as many as 35 percent of telephone numbers are unpublished. In general, numbers can be unpublished because of mobility or choice. Numbers unpublished because of mobility are the result of a recent move where a number has been assigned, but a new phone book has not been published. People in households with these types of unlisted numbers tend to be younger and have lower incomes. People in households who choose to maintain unlisted phone numbers are, on average, a bit older and have a higher household income. If we simply picked numbers out of a phone book, these folks would be excluded and our results would be biased.

To remedy this, we purchase a sample that employs Random Digit Dialed (RDD) methodology. American phone numbers are broken down into several sub-parts. First is area code. In the following number, (999) 555-3456, “999” is the area code. This is followed by the exchange. In this example, “555” is the exchange. A phone number is further broken down into active blocks. In the number above, “34” is the active block and there are 100 numbers (00 – 99) in that active block. Some exchanges, as well as some active blocks, are put aside for business numbers. If at least one working residential number cannot be found in an active block, that block is excluded.

Exchanges and blocks are weighted prior to sample selection to reflect the known distribution of numbers. For example, in your hometown, there are probably some very old exchanges with lots of assigned numbers and some newer ones that only started being used as your town grew and have fewer assigned numbers. If we selected an equal proportion of numbers from each exchange we would be over-sampling the newer exchange and under-sampling the older exchange.

In all, employing RDD methodology is less efficient and more cumbersome, but significantly more accurate. For any given RDD sample 30-35 percent are likely to be non-working numbers, another notable percentage will be businesses that made it through the business screening procedure, and another percentage will be government numbers and fax machines. This translates into more numbers dialed to get a live respondent than for a registered voter poll (where we have a known working residential number), but there is simply no better way to accurately sample the general population.

There is more to good sampling, however, than just how you generate phone numbers.

Q: What are other issues related to good sampling procedures?

A:One of the most important issues is when you call. When doing a registered voter poll, targeted subgroup poll, or general population survey, the Social & Behavioral Lab only calls between 4 – 9 p.m. Monday through Friday, 9 a.m. – 9 p.m. Saturday, and 1– 8 p.m. Sunday.

Q: What difference does it make when you do your calling?

A: If we called during the day, our sample would be too heavily weighted toward retirees and individuals who stay at home during the day. By calling on weekends and only in the evenings during the weekdays, we ensure that everyone has a chance to be selected whether they stay at home during weekdays, work during weekdays, or work swing shifts during the week. Again, it’s an issue of accurate sampling.

Q: Why haven’t I ever been called to take your poll?

A: Every person with a working phone has an equal probability of being selected to participate in a poll. However, just like flipping a coin, the probability “resets” each time. For example, there are two sides to a coin. This doesn’t mean that if you flip it twice, you are guaranteed to get “heads.” Rather, each time you flip the coin, you have a one out of two, or 50%, chance of getting heads. Similarly, every time we go into the field with a survey, you have a one out of “x” chance of being called where “x” represents the number of households in South Carolina with a working phone.

Q: How can you really know what so many people are thinking by only talking to several hundred?

A: It all hinges on the sampling. If we have accurately sampled the population, then every element of that population is accurately represented in our sample. When you have blood tests done at the physician’s office, they don’t take all of your blood; they only take a few drops. That is because every element within your blood is accurately represented in that sample. The same is true of a well-sampled population.