7Technically, because of the square root in the sampling error formula,pffisffi=ffiffiffinffiffi; reducing sampling error by half meansthat the sample size has to be quadrupled.8A sample of 120 respondents describes a population of 12,000or 12,000,000 with virtually the same accuracy, assuming that all other aspects of the sampling design were the same (Fowler, 1984; Mangione, 1995). But, a sample of 120 obtained as a 10% response in a randomly selected initial sample of 1,200 is more likely to be biased, and thus, less likely to accurately describe the population than a sample of 120 obtained as an 80% response in a randomly selected initial sample of 150. 9Actually, smaller samples are biased against finding any statistically significant differences, that is, they are likely to rule out relatively small differences that might really exist in the population as a chance product of sampling. In other words, small samples are more likely to detect the larger differences only, that is, differences that are likely to be substantively significant (Sapsford, 1999).
đang được dịch, vui lòng đợi..