Survey Research Unit -: PSR Survey Methodology
|Survey Research||Index of Polls||PSR Home|
PSR Survey Research Methodology
Palestine is divided into several strata with each representing the towns, cities, villages, and refugee camps in the 16 governorates (muhafazat). Palestine is also divided into "counting areas," or clusters, with each containing a number of families (ranging from 80 to 160 families in each cluster). The number of families in each cluster designates the size of that cluster. The 1997 census provides detailed data on the families as well as detailed maps showing every house in each cluster. The total number of clusters in Palestine is 3200.
PSR sampling process goes trough three stages (1) randomly selecting population locations (clusters or blocs) using probability proportionate to size; (2) randomly selecting households from the population locations using updated maps; (3) selecting a person who is 18 years or older from among the persons in the house using Kiesh tables' method. The sample should be self-weighting, but we do make sure that the age groups we obtain are similar to those in the society using data from the Palestinian Central Bureau of Statistics. Reweighing is done if necessary.
A sample of 127 clusters is randomly selected using probability proportionate to size. Clusters are organized according to size (number of families) and geographic location (West Bank-Gaza Strip) in order to insure representation of all strata and clusters of all sizes. After selecting the cluster samples in the West Bank and the Gaza Strip, 10 homes are selected in each cluster using systemic sampling. Total size of the sample is 1270 adults. The third stage in the sampling process occurs inside the house. Using Kish table, PSR fieldworkers select an adult (over 18 years of age) from among the adults in the house for the interview. Interviewees are assured of complete confidentiality before starting the interview.
Since the sample is a multistage one, two components constitute the variance in the estimates: the within-cluster variation and variation among clusters. We reduce the within-cluster variation by increasing the sample size selected from each cluster or bloc. By increasing the number of clusters selected, the error resulting due to variation among clusters is reduced. Among-cluster variation constitutes the biggest source of sampling error, while the error resulting from the within-cluster variation is negligible relative to the one among clusters. Hence, in this case the margin of error is dependent on the number of clusters considered in the survey. The number of clusters (127) and the number of households in each cell (10) ensure a maximum 3% sampling error.
Our non-response rate ranges between 2% to 9%. The non-response rate is calculated based on the number of household rejections and the number of persons not willing to complete the questionnaire relative to the total sample. In order to prevent errors caused by non-response, we have used over the years three methods: (1) rigorous training of fieldworkers; (2) testing the questionnaire before going to the field; and (3) quality control measures to test the reliability and suitability of fieldworkers.
In order to encourage respondents to talk freely, we assure them of complete anonymity. Respondents do not give their names and filled questionnaires are placed in a folder where they are mixed with all others; it is not possible to trace a specific questionnaire to a certain respondent.
In order to maximize the chances to enter all homes in the sample, two fieldworkers, a male and a female or two females, conduct every interview. By this method we almost double the cost of fieldwork, but we also overcome social difficulties that may prevent a male/female from entering a home that does not have males/females at the time of interview.