Читать книгу Interpreting and Using Statistics in Psychological Research - Andrew N. Christopher - Страница 29

Of course, samples are not always (and hardly ever are) perfect representations of populations. The extent to which a sample doesn’t reflect the population is called sampling error. Sampling error occurs when there is a difference between the characteristics of the population and the characteristics of the sample. Let’s consider an example of sampling error. Let’s again consider Mueller and Oppenheimer’s (2014) research. They drew their sample of college students from two schools, Princeton and UCLA. From this sample, they wanted to learn about the population of college students. Let’s list some parameters for the population of college students in the United States (National Center for Education Statistics, 2014). These parameters for three variables appear in the top portion of Figure 1.5. How well does the sample that Mueller and Oppenheimer (2014) used map onto the population? The statistics for the same variables appear in the lower portion of Figure 1.5.

Sampling error: discrepancy between characteristics of the population and characteristics of the sample.

As you can see, there are some discrepancies between the population parameters and the sample statistics. For example, the sample did not contain any students from two-year schools. Does such sample error make Mueller and Oppenheimer’s (2014) work pointless? Absolutely not. We just need to remember the sample characteristics when drawing conclusions from this research. For instance, it might be a worthwhile idea to conduct this study again using a sample of students from two-year colleges, as no such students were included in this sample. Indeed, learning about the world around us through research is a process that can never be accomplished in a single research study. Indeed, one could argue correctly that the population in this research was in fact college students at four-year universities. Would the results of this research apply to my school, which is a four-year liberal arts college that serves only undergraduate students? Would the results apply to your school? Without conducting the research with a sample of students from my school and a sample of students from your school, we cannot know.

Figure 1.5 Population Parameter and Sample Statistics in Mueller and Oppenheimer’s (2014) Research

To take another example of sampling error, suppose the average SAT score at your college or university is 1700 (if your college or university required the ACT, suppose that the average ACT score is 22). The population in this instance is students at your school. Now, if you take two samples of 10 students each, do you think each sample will have an average SAT of 1700 (or average ACT of 22)? Probably not; in fact, I bet neither sample will have precisely that average. One sample may have an average SAT of 1684 (20 ACT), and the second sample may have an average SAT of 1733 (23 ACT). The discrepancy is the sampling error.

You may often hear on the news about public opinion surveys that various polling groups (e.g., the Wall Street Journal or CNN) conduct on different topics (e.g., feelings about the economy or Congressional approval ratings). To conduct such surveys, the polling agencies do not (and realistically cannot) ask every member of the population for their input. So, they sample the population of interest. As you know, we now have to consider sampling error. That is what is meant when you hear about this “margin of error” the news is (or should be) reporting. If the president has a 54% approval rating, there should be a margin of error reported. That 54% is based on the sample, so there will be some variability around that number in the larger population. A 54% approval rating with a “plus or minus 3% margin of error” means that the presidential approval rating in the population is between 51% and 57%.