The Times of Israel poll methodology

The fieldwork for The Times of Israel pre-election survey was conducted by Rotem TRI – Strategic Research and overseen by Stephan Miller of (202) Strategies.

Dr. Arie Rotem, head of Rotem TRI, notes the following regarding the survey’s methodology:

“The sample of the Times of Israel pre-election survey was based on a representative sample of 803 respondents; 10% of them were Arabs and were surveyed in Arabic. The sampling procedure was simple random sampling and the sample was randomly drawn by software that includes the list of all published (public) phone numbers of households in Israel (estimated to be at least 85%).

“Data were collected via telephone interviews in which 90% of the calls were directed to landline home phones and 10% were directed to mobile phones. The rationale behind calling mobile phones is to overcome the obstacle for getting a representative sample that includes the younger population of 18- to 34-year olds. It is well known and documented that a high percentage of the younger population in Israel does not have regular landline phones and therefore when data collection takes place they are often under-sampled and underrepresented in the current election polls (and in order to solve this problem pollsters tend to weigh their results based on their actual size from the CBS, resulting in less accurate data). In order to overcome this obstacle we conducted 82 calls to mobile numbers, interviewing mostly younger respondents.

“Another challenge with the younger populations is that a relatively high percentage of them tend to be unavailable in the evenings and again are under-sampled in polls conducted via home phones in the evening. In order to overcome this challenge, pollsters tend to run a few callbacks (calling the same phone numbers who are not at home). Due to the fact that some of the election polls are conducted in one day, it is hard to conduct the callback procedure. Our data was collected over several days between December 25 and January 2, and each of the ‘not at home’ phone numbers was randomly dialed 3 more times.

“The interview included a screening phase in which respondents were asked if they intended to vote in the coming election. Respondents who indicated that they were not going to vote or that there was a very low chance that they would vote were screened out and the survey did not continue. Of all respondents to the telephone surveys, 10.2% were screened out following this procedure, creating a more accurate representative sample of likely voters.

“In order to get a more accurate sample of the voting population, respondents were asked to indicate who they had voted for in the 2009 election. Based on these results and the actual results of the 2009 national elections, we applied minor weights to the results in way that gave a more accurate prediction of the coming elections. It is extremely important to indicate that there was a very minor change between the weighted and the non-weighted results.

“Maximum sampling error for the conducted samples size is +/-3.5% in 95% confidence interval.”