The sample used by Osman (1995), described earlier, also illustrates self-weighting sampling. His sample was scattered over a number of villages. To avoid a weighting problem, he selected samples in each village proportional to the size of the number of households in each village. Larger samples were selected from villages with more households and smaller samples from villages with fewer households. As a result, the sample became self weighted. Data from each household were combined for the total sample without having to make complicated mathematical adjustments.

We also want to point out how Osman handled a practical problem in selecting his sample. He did not know the exact population of each village. However, he was able to get a list of households for each village and used these lists for drawing his sample. Households, in effect, became an indicator for population. This approach was based on the assumption that the average household size in each village was about the same, which probably was a safe assumption to make.

Problems to watch for in sampling

In selecting a probability sample, researchers strive to avoid or at least minimize problems that can bias the sample.   Three kinds of problems often occur:

The target population is not clearly defined. This error occurs most often when the population is left at a general or abstract level instead of being defined in concrete, operational terms. An error like this led to an international issue in February, 2002, involving high government officials in Kuwait and the United States. Stone (2002), an American reporter, wrote a story based on surveys in Kuwait and eight other Muslim countries. The Gallup Poll, a highly respected survey organization, conducted the surveys. The results showed that 36% of the Kuwaiti respondents said the attacks on the World Trade Centers in New York City on September 11, 2001, were morally justified. This was the highest percent found in any of the countries included in the survey. Further, only 17% of the Kuwaiti respondents approved of the activities of the United States in Afghanistan. Americans were outraged that the citizens of Kuwait, who United States had rescued in the Gulf War, had such negative attitudes toward the American government. The Kuwaiti ambassador to the United States, however, correctly pointed out that the responses did not represent the views of Kuwaiti citizens.

The Gallup Poll, the organization that conducted the surveys, had selected a sample of persons who lived in Kuwait and not a sample of Kuwaiti citizens. Workers from other countries make up 60% of the population of Kuwait, with many drawn from Pakistan, Egypt, and other Arab countries. The Gallup organization failed to distinguish between Kuwaiti citizens and persons who lived in Kuwait. They did not define the target population and limit their sampling to Kuwaiti citizens. The lesson of this episode is clear: Base your sample on a clearly defined target population and make sure this is the population you want to describe something about.  

Poor sample frame. Bias at this step can be eliminated by making sure the frame is up-to-date, complete, does not contain duplications, and in all other ways matches the target population or comes reasonably close to doing so.

Mistakes are made in sample selection. Mistakes can occur, even when one is careful.   Errors can occur in numbering elements, in using a table of random numbers, in copying the numbers selected, and in other ways.   The only way to avoid such errors is to check and double check each step in sampling and to correct each error that is found.

Low response rate. This problem arises because data are not obtained from some of the persons or other units selected as members of the sample. Some sample members cannot be located, some are never at home when an interviewer tries to contact them, and some refuse to be interviewed. Whatever the reason, the actual number of persons interviewed generally is less than the number selected to form the sample. As the response rate declines, the resulting sample becomes less representative of the target population. This reduces the value of the results for generalizing to that population.

At this point, we suggest you look at other discussions of probability sampling. We recommend any of the following sites for further discussion on probability sampling: