Tuesday, March 29, 2011

Depicting Division, Sample Design

As the results of my survey results have gained more attention, some have wondered about the selection of those who were interviewed--i.e., have questioned whether the sample I used is indeed a nationally-representative one. Without going too far into the specifics of survey methodology, I will say that, yes, in fact, the 500-household sample is quite representative of Bahrain.

For anyone familiar with Bahrain's administrative blocks ("block numbers"), you can infer the proportion of respondents from each general location in Bahrain by looking at the following histogram of block numbers comprising my sample.

(For those unfamiliar: block numbers beginning with 100 correspond to the area of al-Hidd; the 200s to Muharraq; 300s to Manama and the island of Nabih Salih; 400s to Jidd Hafs and several Shi‘a villages; 500s to the “Northern Region” dominated by Shi‘a villages; 600s to the Shi‘a stronghold of Sitra; 700s to a Shi‘a “Central Region”; 800s to Madinat ‘Isa (Isa Town); 900s to Al Khalifa tribal ally-dominated al-Rifa‘, and the sparsely-populated, militarized southern two-thirds of the island; 1000s to a “Western Region” inhabited only in a few coastal villages; the 1100s to the Hawar Islands; and the 1200s to the ethnically-mixed Madinat Hamad (Hamad Town), the country’s newest urban development and home to many newly-naturalized Sunnis.)

From the figure above, then, one can easily judge the representativeness of the sample by comparing the frequency of specific block numbers to the known populations of the districts to which they correspond. The 100 and 200 blocks, for example, comprise the Governorate of Muharraq, whose citizen population was officially reported in 2007 as being 94,558, or about 17.9% of Bahrain’s total 527,433. A much more recent figure based on the number of voters registered for the 2010 parliamentary elections puts this proportion at 57,233 of a total 318,668, or 18.0%. Computing the proportion of 100 and 200 blocks to the entire sample, then, we see that Muharraq households comprise 92 of the 500 total, or exactly 18.4%. When we repeat these calculations for the remaining four governorates we find that the rest of the sample contains 83 or 16.6% Capital Governorate households, 145 or 29.0% in the Central Governorate, 150 or 30.0% in the Northern, and 30 or 6.0% in the Southern. By comparison, the respective 2007 census figures are: 13.3%, 29.7%, 33.1%, and 5.9%.

In sum: for those of you worrying about the sample: don't.

Go to the Results (اذهب إلى النتائج) —>


  1. I hope you're not relying heavily on official CBS figures, unfortunately official figures are cooked to prevent people from deducing anything substantial.

    There is a wikileaks article on this. Also, it is well known that when Bahrain tried to publish some per capita statistics a few years ago, the numbers were shockingly lower than early 2000s mainly because, wealth indicators did not grow as fast as the population (read political naturlisation). What did the official source do? They computed per capita on pre-naturlisation population figures.

    I think overall this will not impact your research as the effect prob. cancels out, but just a word of caution for you - although I think you already know all this.

  2. Don't worry, it's individual-level data and I have very, very good reason to believe that it had not been tampered with.